Module Gnumed.business.gmEMRStructItems

GNUmed health related business object.

license: GPL v2 or later

Expand source code
# -*- coding: utf-8 -*-
"""GNUmed health related business object.

license: GPL v2 or later
"""
#============================================================
__author__ = "Carlos Moro <cfmoro1976@yahoo.es>, <karsten.hilbert@gmx.net>"

import sys
import datetime
import logging
import inspect
import os


if __name__ == '__main__':
        sys.path.insert(0, '../../')
        _ = lambda x:x
from Gnumed.pycommon import gmPG2
from Gnumed.pycommon import gmI18N
from Gnumed.pycommon import gmTools
from Gnumed.pycommon import gmDateTime
from Gnumed.pycommon import gmBusinessDBObject
from Gnumed.pycommon import gmExceptions
from Gnumed.pycommon import gmMatchProvider

from Gnumed.business import gmClinNarrative
from Gnumed.business import gmSoapDefs
from Gnumed.business import gmCoding
from Gnumed.business import gmPraxis
from Gnumed.business import gmOrganization
from Gnumed.business import gmExternalCare
from Gnumed.business import gmDocuments


_log = logging.getLogger('gm.emr')

#============================================================
# diagnostic certainty classification
#============================================================
__diagnostic_certainty_classification_map = None

def diagnostic_certainty_classification2str(classification):

        global __diagnostic_certainty_classification_map

        if __diagnostic_certainty_classification_map is None:
                __diagnostic_certainty_classification_map = {
                        None: '',
                        'A': _('A: Sign'),
                        'B': _('B: Cluster of signs'),
                        'C': _('C: Syndromic diagnosis'),
                        'D': _('D: Scientific diagnosis')
                }

        try:
                return __diagnostic_certainty_classification_map[classification]
        except KeyError:
                return _('<%s>: unknown diagnostic certainty classification') % classification

#============================================================
# Health Issues API
#============================================================
laterality2str = {
        None: '?',
        'na': '',
        'sd': _('bilateral'),
        'ds': _('bilateral'),
        's': _('left'),
        'd': _('right')
}

#============================================================
class cHealthIssue(gmBusinessDBObject.cBusinessDBObject):
        """Represents one health issue."""

        #_cmd_fetch_payload = u"select *, xmin_health_issue from clin.v_health_issues where pk_health_issue=%s"
        _cmd_fetch_payload = "select * from clin.v_health_issues where pk_health_issue = %s"
        _cmds_store_payload = [
                """update clin.health_issue set
                                description = %(description)s,
                                summary = gm.nullify_empty_string(%(summary)s),
                                age_noted = %(age_noted)s,
                                laterality = gm.nullify_empty_string(%(laterality)s),
                                grouping = gm.nullify_empty_string(%(grouping)s),
                                diagnostic_certainty_classification = gm.nullify_empty_string(%(diagnostic_certainty_classification)s),
                                is_active = %(is_active)s,
                                clinically_relevant = %(clinically_relevant)s,
                                is_confidential = %(is_confidential)s,
                                is_cause_of_death = %(is_cause_of_death)s
                        WHERE
                                pk = %(pk_health_issue)s
                                        AND
                                xmin = %(xmin_health_issue)s""",
                "select xmin as xmin_health_issue from clin.health_issue where pk = %(pk_health_issue)s"
        ]
        _updatable_fields = [
                'description',
                'summary',
                'grouping',
                'age_noted',
                'laterality',
                'is_active',
                'clinically_relevant',
                'is_confidential',
                'is_cause_of_death',
                'diagnostic_certainty_classification'
        ]

        #--------------------------------------------------------
        def __init__(self, aPK_obj=None, encounter=None, name='xxxDEFAULTxxx', patient=None, row=None):
                pk = aPK_obj

                if (pk is not None) or (row is not None):
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj=pk, row=row)
                        return

                if patient is None:
                        cmd = """select *, xmin_health_issue from clin.v_health_issues
                                        where
                                                description = %(desc)s
                                                        and
                                                pk_patient = (select fk_patient from clin.encounter where pk = %(enc)s)"""
                else:
                        cmd = """select *, xmin_health_issue from clin.v_health_issues
                                        where
                                                description = %(desc)s
                                                        and
                                                pk_patient = %(pat)s"""

                queries = [{'cmd': cmd, 'args': {'enc': encounter, 'desc': name, 'pat': patient}}]
                rows = gmPG2.run_ro_queries(queries = queries)

                if len(rows) == 0:
                        raise gmExceptions.NoSuchBusinessObjectError('no health issue for [enc:%s::desc:%s::pat:%s]' % (encounter, name, patient))

                pk = rows[0][0]
                r = {'data': rows[0], 'pk_field': 'pk_health_issue'}

                gmBusinessDBObject.cBusinessDBObject.__init__(self, row=r)

        #--------------------------------------------------------
        # external API
        #--------------------------------------------------------
        @classmethod
        def from_problem(cls, problem:'cProblem') -> 'cHealthIssue':
                """Initialize health issue from issue-type problem."""
                if isinstance(problem, cHealthIssue):
                        return problem

                assert isinstance(problem, cProblem), 'cannot convert [%s] to health issue' % problem
                assert problem['type'] == 'issue', 'cannot convert [%s] to health issue' % problem
                return cls(aPK_obj = problem['pk_health_issue'])

        #--------------------------------------------------------
        def rename(self, description:str=None) -> bool:
                """Rename health issue.

                Args:
                        description: the new descriptive name for the issue
                """
                if description.strip() == '':
                        return False

                # update the issue description
                old_description = self._payload['description']
                self._payload['description'] = description.strip()
                self._is_modified = True
                successful, data = self.save_payload()
                if not successful:
                        _log.error('cannot rename health issue [%s] with [%s]' % (self, description))
                        self._payload['description'] = old_description
                        return False

                return True

        #--------------------------------------------------------
        def get_episodes(self) -> list['cEpisode']:
                """The episodes linked to this health issue."""
                cmd = "SELECT * FROM clin.v_pat_episodes WHERE pk_health_issue = %(pk)s"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]

        episodes = property(get_episodes)

        #--------------------------------------------------------
        def close_expired_episode(self, ttl:int=180) -> bool:
                """Close open episode if "older" than the Time To Live.

                Args:
                        ttl: maximum "age" of episode in days
                """
                open_episode = self.open_episode
                if open_episode is None:
                        return True

                #clinical_end = open_episode.best_guess_clinical_end_date
                clinical_end = open_episode.latest_access_date          # :-/
                ttl = datetime.timedelta(ttl)
                now = datetime.datetime.now(tz = clinical_end.tzinfo)
                if (clinical_end + ttl) > now:
                        return False

                open_episode['episode_open'] = False
                success, data = open_episode.save_payload()
                if success:
                        return True

                return False

        #--------------------------------------------------------
        def close_episode(self) -> bool:
                """Unconditionally close the open episode, if any."""
                open_episode = self.get_open_episode()
                open_episode['episode_open'] = False
                success_state, data = open_episode.save_payload()
                return success_state

        #--------------------------------------------------------
        def get_open_episode(self) -> 'cEpisode':
                SQL = "select pk FROM clin.episode WHERE fk_health_issue = %(pk_issue)s AND is_open IS True LIMIT 1"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': SQL, 'args': {'pk_issue': self.pk_obj}}])
                if rows:
                        return cEpisode(aPK_obj = rows[0][0])

                return None

        open_episode = property(get_open_episode)

        #--------------------------------------------------------
        def age_noted_human_readable(self):
                if self._payload['age_noted'] is None:
                        return '<???>'

                # since we've already got an interval we are bound to use it,
                # further transformation will only introduce more errors,
                # later we can improve this deeper inside
                return gmDateTime.format_interval_medically(self._payload['age_noted'])

        #--------------------------------------------------------
        def add_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "INSERT INTO clin.lnk_code2h_issue (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
                args = {
                        'item': self._payload['pk_health_issue'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def remove_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "DELETE FROM clin.lnk_code2h_issue WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                args = {
                        'item': self._payload['pk_health_issue'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a'):
                rows = gmClinNarrative.get_as_journal (
                        issues = [self.pk_obj],
                        order_by = 'pk_episode, pk_encounter, clin_when, scr, src_table'
                )

                if len(rows) == 0:
                        return ''

                left_margin = ' ' * left_margin

                lines = []
                lines.append(_('Clinical data generated during encounters under this health issue:'))

                prev_epi = None
                for row in rows:
                        if row['pk_episode'] != prev_epi:
                                lines.append('')
                                prev_epi = row['pk_episode']

                        when = gmDateTime.pydt_strftime(row['clin_when'], date_format)
                        top_row = '%s%s %s (%s) %s' % (
                                gmTools.u_box_top_left_arc,
                                gmTools.u_box_horiz_single,
                                gmSoapDefs.soap_cat2l10n_str[row['real_soap_cat']],
                                when,
                                gmTools.u_box_horiz_single * 5
                        )
                        soap = gmTools.wrap (
                                text = row['narrative'],
                                width = 60,
                                initial_indent = '  ',
                                subsequent_indent = '  ' + left_margin
                        )
                        row_ver = ''
                        if row['row_version'] > 0:
                                row_ver = 'v%s: ' % row['row_version']
                        bottom_row = '%s%s %s, %s%s %s' % (
                                ' ' * 40,
                                gmTools.u_box_horiz_light_heavy,
                                row['modified_by'],
                                row_ver,
                                gmDateTime.pydt_strftime(row['modified_when'], date_format),
                                gmTools.u_box_horiz_heavy_light
                        )

                        lines.append(top_row)
                        lines.append(soap)
                        lines.append(bottom_row)

                eol_w_margin = '\n%s' % left_margin
                return left_margin + eol_w_margin.join(lines) + '\n'

        #--------------------------------------------------------
        def format (self, left_margin=0, patient=None,
                with_summary=True,
                with_codes=True,
                with_episodes=True,
                with_encounters=True,
                with_medications=True,
                with_hospital_stays=True,
                with_procedures=True,
                with_family_history=True,
                with_documents=True,
                with_tests=True,
                with_vaccinations=True,
                with_external_care=True
        ):

                lines = []

                lines.append(_('Health Issue %s%s%s%s   [#%s]') % (
                        '\u00BB',
                        self._payload['description'],
                        '\u00AB',
                        gmTools.coalesce (
                                value2test = self.laterality_description,
                                return_instead = '',
                                template4value = ' (%s)',
                                none_equivalents = [None, '', '?']
                        ),
                        self._payload['pk_health_issue']
                ))

                if self._payload['is_confidential']:
                        lines.append('')
                        lines.append(_(' ***** CONFIDENTIAL *****'))
                        lines.append('')

                if self._payload['is_cause_of_death']:
                        lines.append('')
                        lines.append(_(' contributed to death of patient'))
                        lines.append('')

                enc = cEncounter(aPK_obj = self._payload['pk_encounter'])
                lines.append (_(' Created during encounter: %s (%s - %s)   [#%s]') % (
                        enc['l10n_type'],
                        enc['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                        enc['last_affirmed_original_tz'].strftime('%H:%M'),
                        self._payload['pk_encounter']
                ))

                if self._payload['age_noted'] is not None:
                        lines.append(_(' Noted at age: %s') % self.age_noted_human_readable())

                lines.append(' ' + _('Status') + ': %s, %s%s' % (
                        gmTools.bool2subst(self._payload['is_active'], _('active'), _('inactive')),
                        gmTools.bool2subst(self._payload['clinically_relevant'], _('clinically relevant'), _('not clinically relevant')),
                        gmTools.coalesce (
                                value2test = diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification']),
                                return_instead = '',
                                template4value = ', %s',
                                none_equivalents = [None, '']
                        )
                ))

                if with_summary:
                        if self._payload['summary'] is not None:
                                lines.append(' %s:' % _('Synopsis'))
                                lines.append(gmTools.wrap (
                                        text = self._payload['summary'],
                                        width = 60,
                                        initial_indent = '  ',
                                        subsequent_indent = '  '
                                ))

                # codes ?
                if with_codes:
                        codes = self.generic_codes
                        if len(codes) > 0:
                                lines.append('')
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        del codes

                lines.append('')

                # patient/emr dependant
                if patient is not None:
                        if patient.ID != self._payload['pk_patient']:
                                msg = '<patient>.ID = %s but health issue %s belongs to patient %s' % (
                                        patient.ID,
                                        self._payload['pk_health_issue'],
                                        self._payload['pk_patient']
                                )
                                raise ValueError(msg)
                        emr = patient.emr

                        # episodes
                        if with_episodes:
                                epis = self.get_episodes()
                                if epis is None:
                                        lines.append(_('Error retrieving episodes for this health issue.'))
                                elif len(epis) == 0:
                                        lines.append(_('There are no episodes for this health issue.'))
                                else:
                                        lines.append (
                                                _('Episodes: %s (most recent: %s%s%s)') % (
                                                        len(epis),
                                                        gmTools.u_left_double_angle_quote,
                                                        emr.get_most_recent_episode(issue = self._payload['pk_health_issue'])['description'],
                                                        gmTools.u_right_double_angle_quote
                                                )
                                        )
                                        for epi in epis:
                                                lines.append(' \u00BB%s\u00AB (%s)' % (
                                                        epi['description'],
                                                        gmTools.bool2subst(epi['episode_open'], _('ongoing'), _('closed'))
                                                ))
                                lines.append('')

                        # encounters
                        if with_encounters:
                                first_encounter = emr.get_first_encounter(issue_id = self._payload['pk_health_issue'])
                                last_encounter = emr.get_last_encounter(issue_id = self._payload['pk_health_issue'])

                                if first_encounter is None or last_encounter is None:
                                        lines.append(_('No encounters found for this health issue.'))
                                else:
                                        encs = emr.get_encounters(issues = [self._payload['pk_health_issue']])
                                        lines.append(_('Encounters: %s (%s - %s):') % (
                                                len(encs),
                                                first_encounter['started_original_tz'].strftime('%m/%Y'),
                                                last_encounter['last_affirmed_original_tz'].strftime('%m/%Y')
                                        ))
                                        lines.append(_(' Most recent: %s - %s') % (
                                                last_encounter['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                                                last_encounter['last_affirmed_original_tz'].strftime('%H:%M')
                                        ))

                        # medications
                        if with_medications:
                                meds = emr.get_current_medications (
                                        issues = [ self._payload['pk_health_issue'] ],
                                        order_by = 'is_currently_active DESC, started, substance'
                                )
                                if len(meds) > 0:
                                        lines.append('')
                                        lines.append(_('Medications and Substances'))
                                for m in meds:
                                        lines.append(m.format(left_margin = (left_margin + 1)))
                                del meds

                        # hospitalizations
                        if with_hospital_stays:
                                stays = emr.get_hospital_stays (
                                        issues = [ self._payload['pk_health_issue'] ]
                                )
                                if len(stays) > 0:
                                        lines.append('')
                                        lines.append(_('Hospitalizations: %s') % len(stays))
                                for s in stays:
                                        lines.append(s.format(left_margin = (left_margin + 1)))
                                del stays

                        # procedures
                        if with_procedures:
                                procs = emr.get_performed_procedures (
                                        issues = [ self._payload['pk_health_issue'] ]
                                )
                                if len(procs) > 0:
                                        lines.append('')
                                        lines.append(_('Procedures performed: %s') % len(procs))
                                for p in procs:
                                        lines.append(p.format(left_margin = (left_margin + 1)))
                                del procs

                        # family history
                        if with_family_history:
                                fhx = emr.get_family_history(issues = [ self._payload['pk_health_issue'] ])
                                if len(fhx) > 0:
                                        lines.append('')
                                        lines.append(_('Family History: %s') % len(fhx))
                                for f in fhx:
                                        lines.append(f.format (
                                                left_margin = (left_margin + 1),
                                                include_episode = True,
                                                include_comment = True,
                                                include_codes = False
                                        ))
                                del fhx

                        epis = self.get_episodes()
                        if len(epis) > 0:
                                epi_pks = [ e['pk_episode'] for e in epis ]

                                # documents
                                if with_documents:
                                        doc_folder = patient.get_document_folder()
                                        docs = doc_folder.get_documents(pk_episodes = epi_pks)
                                        if len(docs) > 0:
                                                lines.append('')
                                                lines.append(_('Documents: %s') % len(docs))
                                        del docs

                                # test results
                                if with_tests:
                                        tests = emr.get_test_results_by_date(episodes = epi_pks)
                                        if len(tests) > 0:
                                                lines.append('')
                                                lines.append(_('Measurements and Results: %s') % len(tests))
                                        del tests

                                # vaccinations
                                if with_vaccinations:
                                        vaccs = emr.get_vaccinations(episodes = epi_pks, order_by = 'date_given, vaccine')
                                        if len(vaccs) > 0:
                                                lines.append('')
                                                lines.append(_('Vaccinations:'))
                                        for vacc in vaccs:
                                                lines.extend(vacc.format(with_reaction = True))
                                        del vaccs

                        del epis

                if with_external_care:
                        care = self._get_external_care(order_by = 'organization, unit, provider')
                        if len(care) > 0:
                                lines.append('')
                                lines.append(_('External care:'))
                        for item in care:
                                lines.append(' %s%s@%s%s' % (
                                        gmTools.coalesce(item['provider'], '', '%s: '),
                                        item['unit'],
                                        item['organization'],
                                        gmTools.coalesce(item['comment'], '', ' (%s)')
                                ))

                left_margin = ' ' * left_margin
                eol_w_margin = '\n%s' % left_margin
                lines = gmTools.strip_trailing_empty_lines(lines = lines, eol = '\n')
                return left_margin + eol_w_margin.join(lines) + '\n'
        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_external_care(self, order_by=None):
                return gmExternalCare.get_external_care_items(pk_health_issue = self.pk_obj, order_by = order_by)

        external_care = property(_get_external_care)

        #--------------------------------------------------------
        def _get_first_episode(self):

                args = {'pk_issue': self.pk_obj}

                cmd = """SELECT
                        earliest, pk_episode
                FROM (
                                -- .modified_when of all episodes of this issue,
                                -- earliest-possible thereof = when created,
                                -- should actually go all the way back into audit.log_episode
                                (SELECT
                                        c_epi.modified_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM clin.episode c_epi
                                 WHERE c_epi.fk_health_issue = %(pk_issue)s
                                )
                        UNION ALL

                                -- last modification of encounter in which episodes of this issue were created,
                                -- earliest-possible thereof = initial creation of that encounter
                                (SELECT
                                        c_enc.modified_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.episode c_epi
                                                INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                                INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- start of encounter in which episodes of this issue were created,
                                -- earliest-possible thereof = set by user
                                (SELECT
                                        c_enc.started AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.episode c_epi
                                                INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                                INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- start of encounters of clinical items linked to episodes of this issue,
                                -- earliest-possible thereof = explicitly set by user
                                (SELECT
                                        c_enc.started AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.encounter c_enc ON (c_cri.fk_encounter = c_enc.pk)
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- .clin_when of clinical items linked to episodes of this issue,
                                -- earliest-possible thereof = explicitly set by user
                                (SELECT
                                        c_cri.clin_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- earliest modification time of clinical items linked to episodes of this issue
                                -- this CAN be used since if an item is linked to an episode it can be
                                -- assumed the episode (should have) existed at the time of creation
                                (SELECT
                                        c_cri.modified_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- there may not be items, but there may still be documents ...
                                (SELECT
                                        b_dm.clin_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        blobs.doc_med b_dm
                                                INNER JOIN clin.episode c_epi ON (b_dm.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                  WHERE c_hi.pk = %(pk_issue)s
                                )
                ) AS candidates
                ORDER BY earliest NULLS LAST
                LIMIT 1"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                if len(rows) == 0:
                        return None
                return cEpisode(aPK_obj = rows[0]['pk_episode'])

        first_episode = property(_get_first_episode)

        #--------------------------------------------------------
        def _get_latest_episode(self):

                # explicit always wins:
                if self._payload['has_open_episode']:
                        return self.open_episode

                args = {'pk_issue': self.pk_obj}

                # cheap query first: any episodes at all ?
                cmd = "SELECT 1 FROM clin.episode WHERE fk_health_issue = %(pk_issue)s"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                if len(rows) == 0:
                        return None

                cmd = """SELECT
                        latest, pk_episode
                FROM (
                                -- .clin_when of clinical items linked to episodes of this issue,
                                -- latest-possible thereof = explicitly set by user
                                (SELECT
                                        c_cri.clin_when AS latest,
                                        c_epi.pk AS pk_episode,
                                        1 AS rank
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- .clin_when of documents linked to episodes of this issue
                                (SELECT
                                        b_dm.clin_when AS latest,
                                        c_epi.pk AS pk_episode,
                                        1 AS rank
                                 FROM
                                        blobs.doc_med b_dm
                                                INNER JOIN clin.episode c_epi ON (b_dm.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- last_affirmed of encounter in which episodes of this issue were created,
                                -- earliest-possible thereof = set by user
                                (SELECT
                                        c_enc.last_affirmed AS latest,
                                        c_epi.pk AS pk_episode,
                                        2 AS rank
                                 FROM
                                        clin.episode c_epi
                                                INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                                INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )

                ) AS candidates
                WHERE
                        -- weed out NULL rows due to episodes w/o clinical items and w/o documents
                        latest IS NOT NULL
                ORDER BY
                        rank,
                        latest DESC
                LIMIT 1
                """
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                if len(rows) == 0:
                        # there were no episodes for this issue
                        return None
                return cEpisode(aPK_obj = rows[0]['pk_episode'])

        latest_episode = property(_get_latest_episode)

        #--------------------------------------------------------
        # Steffi suggested we divide into safe and assumed (= possible) start dates
        def _get_safe_start_date(self):
                """This returns the date when we can assume to safely KNOW
                   the health issue existed (because the provider said so)."""

                args = {
                        'enc': self._payload['pk_encounter'],
                        'pk': self._payload['pk_health_issue']
                }
                cmd = """SELECT COALESCE (
                        -- this one must override all:
                        -- .age_noted if not null and DOB is known
                        (CASE
                                WHEN c_hi.age_noted IS NULL
                                THEN NULL::timestamp with time zone
                                WHEN
                                        (SELECT d_i.dob FROM dem.identity d_i WHERE d_i.pk = (
                                                SELECT c_enc.fk_patient FROM clin.encounter c_enc WHERE c_enc.pk = %(enc)s
                                        )) IS NULL
                                THEN NULL::timestamp with time zone
                                ELSE
                                        c_hi.age_noted + (
                                                SELECT d_i.dob FROM dem.identity d_i WHERE d_i.pk = (
                                                        SELECT c_enc.fk_patient FROM clin.encounter c_enc WHERE c_enc.pk = %(enc)s
                                                )
                                        )
                        END),

                        -- look at best_guess_clinical_start_date of all linked episodes

                        -- start of encounter in which created, earliest = explicitly set
                        (SELECT c_enc.started AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = c_hi.fk_encounter)
                )
                FROM clin.health_issue c_hi
                WHERE c_hi.pk = %(pk)s"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                start = rows[0][0]
                # leads to a loop:
                #end = self.clinical_end_date
                #if start > end:
                #       return end
                return start

        safe_start_date = property(_get_safe_start_date)

        #--------------------------------------------------------
        def _get_possible_start_date(self):
                args = {'pk': self._payload['pk_health_issue']}
                cmd = """
SELECT MIN(earliest) FROM (
        -- last modification, earliest = when created in/changed to the current state
        (SELECT modified_when AS earliest FROM clin.health_issue WHERE pk = %(pk)s)

UNION ALL
        -- last modification of encounter in which created, earliest = initial creation of that encounter
        (SELECT c_enc.modified_when AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = (
                SELECT c_hi.fk_encounter FROM clin.health_issue c_hi WHERE c_hi.pk = %(pk)s
        ))

UNION ALL
        -- earliest explicit .clin_when of clinical items linked to this health_issue
        (SELECT MIN(c_vpi.clin_when) AS earliest FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s)

UNION ALL
        -- earliest modification time of clinical items linked to this health issue
        -- this CAN be used since if an item is linked to a health issue it can be
        -- assumed the health issue (should have) existed at the time of creation
        (SELECT MIN(c_vpi.modified_when) AS earliest FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s)

UNION ALL
        -- earliest start of encounters of clinical items linked to this episode
        (SELECT MIN(c_enc.started) AS earliest FROM clin.encounter c_enc WHERE c_enc.pk IN (
                SELECT c_vpi.pk_encounter FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s
        ))

-- here we should be looking at
-- .best_guess_clinical_start_date of all episodes linked to this encounter

) AS candidates"""

                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return rows[0][0]

        possible_start_date = property(_get_possible_start_date)

        #--------------------------------------------------------
        def _get_clinical_end_date(self):
                if self._payload['is_active']:
                        return None
                if self._payload['has_open_episode']:
                        return None
                latest_episode = self.latest_episode
                if latest_episode is not None:
                        return latest_episode.best_guess_clinical_end_date
                # apparently, there are no episodes for this issue
                # and the issue is not active either
                # so, we simply do not know, the safest assumption is:
                return self.safe_start_date

        clinical_end_date = property(_get_clinical_end_date)

        #--------------------------------------------------------
        def _get_latest_access_date(self):
                cmd = """
SELECT
        MAX(latest)
FROM (
        -- last modification, latest = when last changed to the current state
        -- DO NOT USE: database upgrades may change this field
        (SELECT modified_when AS latest FROM clin.health_issue WHERE pk = %(pk)s)

        --UNION ALL
        -- last modification of encounter in which created, latest = initial creation of that encounter
        -- DO NOT USE: just because one corrects a typo does not mean the issue took any longer
        --(SELECT c_enc.modified_when AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
        -- )
        --)

        --UNION ALL
        -- end of encounter in which created, latest = explicitly set
        -- DO NOT USE: we can retrospectively create issues which
        -- DO NOT USE: are long since finished
        --(SELECT c_enc.last_affirmed AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
        -- )
        --)

        UNION ALL
        -- latest end of encounters of clinical items linked to this issue
        (SELECT
                MAX(last_affirmed) AS latest
         FROM clin.encounter
         WHERE pk IN (
                SELECT pk_encounter FROM clin.v_pat_items WHERE pk_health_issue = %(pk)s
         )
        )

        UNION ALL
        -- latest explicit .clin_when of clinical items linked to this issue
        (SELECT
                MAX(clin_when) AS latest
         FROM clin.v_pat_items
         WHERE pk_health_issue = %(pk)s
        )

        -- latest modification time of clinical items linked to this issue
        -- this CAN be used since if an item is linked to an issue it can be
        -- assumed the issue (should have) existed at the time of modification
        -- DO NOT USE, because typo fixes should not extend the issue
        --(SELECT MIN(modified_when) AS latest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

) AS candidates"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return rows[0][0]

        latest_access_date = property(_get_latest_access_date)

        #--------------------------------------------------------
        def _get_laterality_description(self):
                try:
                        return laterality2str[self._payload['laterality']]
                except KeyError:
                        return '<?>'

        laterality_description = property(_get_laterality_description)

        #--------------------------------------------------------
        def _get_diagnostic_certainty_description(self):
                return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])

        diagnostic_certainty_description = property(_get_diagnostic_certainty_description)

        #--------------------------------------------------------
        def _get_formatted_revision_history(self):
                cmd = """SELECT
                                '<N/A>'::TEXT as audit__action_applied,
                                NULL AS audit__action_when,
                                '<N/A>'::TEXT AS audit__action_by,
                                pk_audit,
                                row_version,
                                modified_when,
                                modified_by,
                                pk,
                                description,
                                laterality,
                                age_noted,
                                is_active,
                                clinically_relevant,
                                is_confidential,
                                is_cause_of_death,
                                fk_encounter,
                                grouping,
                                diagnostic_certainty_classification,
                                summary
                        FROM clin.health_issue
                        WHERE pk = %(pk_health_issue)s
                UNION ALL (
                        SELECT
                                audit_action as audit__action_applied,
                                audit_when as audit__action_when,
                                audit_by as audit__action_by,
                                pk_audit,
                                orig_version as row_version,
                                orig_when as modified_when,
                                orig_by as modified_by,
                                pk,
                                description,
                                laterality,
                                age_noted,
                                is_active,
                                clinically_relevant,
                                is_confidential,
                                is_cause_of_death,
                                fk_encounter,
                                grouping,
                                diagnostic_certainty_classification,
                                summary
                        FROM audit.log_health_issue
                        WHERE pk = %(pk_health_issue)s
                )
                ORDER BY row_version DESC
                """
                args = {'pk_health_issue': self.pk_obj}
                title = _('Health issue: %s%s%s') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['description'],
                        gmTools.u_right_double_angle_quote
                )
                return '\n'.join(self._get_revision_history(cmd, args, title))

        formatted_revision_history = property(_get_formatted_revision_history)
        #--------------------------------------------------------
        def _get_generic_codes(self):
                if len(self._payload['pk_generic_codes']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2h_issue WHERE fk_item = %(issue)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'issue': self._payload['pk_health_issue'],
                                        'codes': self._payload['pk_generic_codes']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2h_issue (fk_item, fk_generic_code) VALUES (%(issue)s, %(pk_code)s)',
                                'args': {
                                        'issue': self._payload['pk_health_issue'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                return

        generic_codes = property(_get_generic_codes, _set_generic_codes)

#============================================================
def create_health_issue(description=None, encounter=None, patient=None, link_obj=None):
        """Creates a new health issue for a given patient.

        description - health issue name
        """
        try:
                h_issue = cHealthIssue(name = description, encounter = encounter, patient = patient)
                return h_issue

        except gmExceptions.NoSuchBusinessObjectError:
                pass

        queries = []
        cmd = "insert into clin.health_issue (description, fk_encounter) values (%(desc)s, %(enc)s)"
        queries.append({'cmd': cmd, 'args': {'desc': description, 'enc': encounter}})
        cmd = "select currval('clin.health_issue_pk_seq')"
        queries.append({'cmd': cmd})
        rows = gmPG2.run_rw_queries(queries = queries, return_data = True, link_obj = link_obj)
        h_issue = cHealthIssue(aPK_obj = rows[0][0])
        return h_issue

#-----------------------------------------------------------
def delete_health_issue(health_issue=None):
        if isinstance(health_issue, cHealthIssue):
                args = {'pk': health_issue['pk_health_issue']}
        else:
                args = {'pk': int(health_issue)}
        try:
                gmPG2.run_rw_queries(queries = [{'cmd': 'DELETE FROM clin.health_issue WHERE pk = %(pk)s', 'args': args}])
        except gmPG2.dbapi.IntegrityError:
                # should be parsing pgcode/and or error message
                _log.exception('cannot delete health issue')
                return False

        return True

#------------------------------------------------------------
# use as dummy for unassociated episodes
def get_dummy_health_issue():
        issue = {
                'pk_health_issue': None,
                'description': _('Unattributed episodes'),
                'age_noted': None,
                'laterality': 'na',
                'is_active': True,
                'clinically_relevant': True,
                'is_confidential': None,
                'is_cause_of_death': False,
                'is_dummy': True,
                'grouping': None
        }
        return issue

#-----------------------------------------------------------
def health_issue2problem(health_issue=None, allow_irrelevant=False):
        return cProblem (
                aPK_obj = {
                        'pk_patient': health_issue['pk_patient'],
                        'pk_health_issue': health_issue['pk_health_issue'],
                        'pk_episode': None
                },
                try_potential_problems = allow_irrelevant
        )

#============================================================
# episodes API
#============================================================
class cEpisode(gmBusinessDBObject.cBusinessDBObject):
        """Represents one clinical episode.
        """
        _cmd_fetch_payload = "select * from clin.v_pat_episodes where pk_episode=%s"
        _cmds_store_payload = [
                """update clin.episode set
                                fk_health_issue = %(pk_health_issue)s,
                                is_open = %(episode_open)s::boolean,
                                description = %(description)s,
                                summary = gm.nullify_empty_string(%(summary)s),
                                diagnostic_certainty_classification = gm.nullify_empty_string(%(diagnostic_certainty_classification)s)
                        where
                                pk = %(pk_episode)s and
                                xmin = %(xmin_episode)s""",
                """select xmin_episode from clin.v_pat_episodes where pk_episode = %(pk_episode)s"""
        ]
        _updatable_fields = [
                'pk_health_issue',
                'episode_open',
                'description',
                'summary',
                'diagnostic_certainty_classification'
        ]
        #--------------------------------------------------------
        def __init__(self, aPK_obj=None, id_patient=None, name='xxxDEFAULTxxx', health_issue=None, row=None, encounter=None, link_obj=None):
                pk = aPK_obj
                if pk is None and row is None:

                        where_parts = ['description = %(desc)s']

                        if id_patient is not None:
                                where_parts.append('pk_patient = %(pat)s')

                        if health_issue is not None:
                                where_parts.append('pk_health_issue = %(issue)s')

                        if encounter is not None:
                                where_parts.append('pk_patient = (SELECT fk_patient FROM clin.encounter WHERE pk = %(enc)s)')

                        args = {
                                'pat': id_patient,
                                'issue': health_issue,
                                'enc': encounter,
                                'desc': name
                        }

                        cmd = 'SELECT * FROM clin.v_pat_episodes WHERE %s' % ' AND '.join(where_parts)

                        rows = gmPG2.run_ro_queries (
                                link_obj = link_obj,
                                queries = [{'cmd': cmd, 'args': args}],
                        )

                        if len(rows) == 0:
                                raise gmExceptions.NoSuchBusinessObjectError('no episode for [%s:%s:%s:%s]' % (id_patient, name, health_issue, encounter))

                        r = {'data': rows[0], 'pk_field': 'pk_episode'}
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, row=r)

                else:
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj=pk, row=row, link_obj = link_obj)

        #--------------------------------------------------------
        # external API
        #--------------------------------------------------------
        @classmethod
        def from_problem(cls, problem:'cProblem') -> 'cEpisode':
                """Initialize episode from episody-type problem."""
                if isinstance(problem, cEpisode):
                        return problem

                assert isinstance(problem, cProblem), 'cannot convert [%s] to episode' % problem
                assert problem['type'] == 'episode', 'cannot convert [%s] to episode' % problem
                return cls(aPK_obj = problem['pk_episode'])

        #--------------------------------------------------------
        def get_patient(self):
                return self._payload['pk_patient']

        #--------------------------------------------------------
        def get_narrative(self, soap_cats=None, encounters=None, order_by = None):
                return gmClinNarrative.get_narrative (
                        soap_cats = soap_cats,
                        encounters = encounters,
                        episodes = [self.pk_obj],
                        order_by = order_by
                )

        #--------------------------------------------------------
        def rename(self, description=None):
                """Method for episode editing, that is, episode renaming.

                @param description
                        - the new descriptive name for the encounter
                @type description
                        - a string instance
                """
                # sanity check
                if description.strip() == '':
                        _log.error('<description> must be a non-empty string instance')
                        return False
                # update the episode description
                old_description = self._payload['description']
                self._payload['description'] = description.strip()
                self._is_modified = True
                successful, data = self.save_payload()
                if not successful:
                        _log.error('cannot rename episode [%s] to [%s]' % (self, description))
                        self._payload['description'] = old_description
                        return False
                return True

        #--------------------------------------------------------
        def add_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""

                if pk_code in self._payload['pk_generic_codes']:
                        return

                cmd = """
                        INSERT INTO clin.lnk_code2episode
                                (fk_item, fk_generic_code)
                        SELECT
                                %(item)s,
                                %(code)s
                        WHERE NOT EXISTS (
                                SELECT 1 FROM clin.lnk_code2episode
                                WHERE
                                        fk_item = %(item)s
                                                AND
                                        fk_generic_code = %(code)s
                        )"""
                args = {
                        'item': self._payload['pk_episode'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return

        #--------------------------------------------------------
        def remove_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "DELETE FROM clin.lnk_code2episode WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                args = {
                        'item': self._payload['pk_episode'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a'):
                rows = gmClinNarrative.get_as_journal (
                        episodes = [self.pk_obj],
                        order_by = 'pk_encounter, clin_when, scr, src_table'
                        #order_by = u'pk_encounter, scr, clin_when, src_table'
                )

                if len(rows) == 0:
                        return ''

                lines = []

                lines.append(_('Clinical data generated during encounters within this episode:'))

                left_margin = ' ' * left_margin

                prev_enc = None
                for row in rows:
                        if row['pk_encounter'] != prev_enc:
                                lines.append('')
                                prev_enc = row['pk_encounter']

                        when = row['clin_when'].strftime(date_format)
                        top_row = '%s%s %s (%s) %s' % (
                                gmTools.u_box_top_left_arc,
                                gmTools.u_box_horiz_single,
                                gmSoapDefs.soap_cat2l10n_str[row['real_soap_cat']],
                                when,
                                gmTools.u_box_horiz_single * 5
                        )
                        soap = gmTools.wrap (
                                text = row['narrative'],
                                width = 60,
                                initial_indent = '  ',
                                subsequent_indent = '  ' + left_margin
                        )
                        row_ver = ''
                        if row['row_version'] > 0:
                                row_ver = 'v%s: ' % row['row_version']
                        bottom_row = '%s%s %s, %s%s %s' % (
                                ' ' * 40,
                                gmTools.u_box_horiz_light_heavy,
                                row['modified_by'],
                                row_ver,
                                gmDateTime.pydt_strftime(row['modified_when'], date_format),
                                gmTools.u_box_horiz_heavy_light
                        )

                        lines.append(top_row)
                        lines.append(soap)
                        lines.append(bottom_row)

                eol_w_margin = '\n%s' % left_margin
                return left_margin + eol_w_margin.join(lines) + '\n'

        #--------------------------------------------------------
        def format_maximum_information(self, patient=None):
                if patient is None:
                        from Gnumed.business.gmPerson import gmCurrentPatient, cPerson
                        if self._payload['pk_patient'] == gmCurrentPatient().ID:
                                patient = gmCurrentPatient()
                        else:
                                patient = cPerson(self._payload['pk_patient'])

                return self.format (
                        patient = patient,
                        with_summary = True,
                        with_codes = True,
                        with_encounters = True,
                        with_documents = True,
                        with_hospital_stays = True,
                        with_procedures = True,
                        with_family_history = True,
                        with_tests = False,             # does not inform on the episode itself
                        with_vaccinations = True,
                        with_health_issue = True,
                        return_list = True
                )

        #--------------------------------------------------------
        def format(self, left_margin=0, patient=None,
                with_summary=True,
                with_codes=True,
                with_encounters=True,
                with_documents=True,
                with_hospital_stays=True,
                with_procedures=True,
                with_family_history=True,
                with_tests=True,
                with_vaccinations=True,
                with_health_issue=False,
                return_list=False
        ):

                if patient is not None:
                        if patient.ID != self._payload['pk_patient']:
                                msg = '<patient>.ID = %s but episode %s belongs to patient %s' % (
                                        patient.ID,
                                        self._payload['pk_episode'],
                                        self._payload['pk_patient']
                                )
                                raise ValueError(msg)
                        emr = patient.emr
                else:
                        with_encounters = False
                        with_documents = False
                        with_hospital_stays = False
                        with_procedures = False
                        with_family_history = False
                        with_tests = False
                        with_vaccinations = False

                lines = []

                # episode details
                lines.append (_('Episode %s%s%s   [#%s]') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['description'],
                        gmTools.u_right_double_angle_quote,
                        self._payload['pk_episode']
                ))

                enc = cEncounter(aPK_obj = self._payload['pk_encounter'])
                lines.append (' ' + _('Created during encounter: %s (%s - %s)   [#%s]') % (
                        enc['l10n_type'],
                        enc['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                        enc['last_affirmed_original_tz'].strftime('%H:%M'),
                        self._payload['pk_encounter']
                ))

                if patient is not None:
                        range_str, range_str_verb, duration_str = self.formatted_clinical_duration
                        lines.append(_(' Duration: %s (%s)') % (duration_str, range_str_verb))

                lines.append(' ' + _('Status') + ': %s%s' % (
                        gmTools.bool2subst(self._payload['episode_open'], _('active'), _('finished')),
                        gmTools.coalesce (
                                value2test = diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification']),
                                return_instead = '',
                                template4value = ', %s',
                                none_equivalents = [None, '']
                        )
                ))

                if with_health_issue:
                        lines.append(' ' + _('Health issue') + ': %s' % gmTools.coalesce (
                                self._payload['health_issue'],
                                _('none associated')
                        ))

                if with_summary:
                        if self._payload['summary'] is not None:
                                lines.append(' %s:' % _('Synopsis'))
                                lines.append(gmTools.wrap (
                                                text = self._payload['summary'],
                                                width = 60,
                                                initial_indent = '  ',
                                                subsequent_indent = '  '
                                        )
                                )

                # codes
                if with_codes:
                        codes = self.generic_codes
                        if len(codes) > 0:
                                lines.append('')
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        del codes

                lines.append('')

                # encounters
                if with_encounters:
                        encs = emr.get_encounters(episodes = [self._payload['pk_episode']])
                        if encs is None:
                                lines.append(_('Error retrieving encounters for this episode.'))
                        elif len(encs) == 0:
                                #lines.append(_('There are no encounters for this issue.'))
                                pass
                        else:
                                first_encounter = emr.get_first_encounter(episode_id = self._payload['pk_episode'])
                                last_encounter = emr.get_last_encounter(episode_id = self._payload['pk_episode'])
                                lines.append(_('Last worked on: %s\n') % last_encounter['last_affirmed_original_tz'].strftime('%Y-%m-%d %H:%M'))
                                if len(encs) < 4:
                                        line = _('%s encounter(s) (%s - %s):')
                                else:
                                        line = _('1st and (up to 3) most recent (of %s) encounters (%s - %s):')
                                lines.append(line % (
                                        len(encs),
                                        first_encounter['started'].strftime('%m/%Y'),
                                        last_encounter['last_affirmed'].strftime('%m/%Y')
                                ))
                                lines.append(' %s - %s (%s):%s' % (
                                        first_encounter['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                                        first_encounter['last_affirmed_original_tz'].strftime('%H:%M'),
                                        first_encounter['l10n_type'],
                                        gmTools.coalesce (
                                                first_encounter['assessment_of_encounter'],
                                                gmTools.coalesce (
                                                        first_encounter['reason_for_encounter'],
                                                        '',
                                                        ' \u00BB%s\u00AB' + (' (%s)' % _('RFE'))
                                                ),
                                                ' \u00BB%s\u00AB' + (' (%s)' % _('AOE'))
                                        )
                                ))
                                if len(encs) > 4:
                                        lines.append(_(' %s %s skipped %s') % (
                                                gmTools.u_ellipsis,
                                                (len(encs) - 4),
                                                gmTools.u_ellipsis
                                        ))
                                for enc in encs[1:][-3:]:
                                        lines.append(' %s - %s (%s):%s' % (
                                                enc['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                                                enc['last_affirmed_original_tz'].strftime('%H:%M'),
                                                enc['l10n_type'],
                                                gmTools.coalesce (
                                                        enc['assessment_of_encounter'],
                                                        gmTools.coalesce (
                                                                enc['reason_for_encounter'],
                                                                '',
                                                                ' \u00BB%s\u00AB' + (' (%s)' % _('RFE'))
                                                        ),
                                                        ' \u00BB%s\u00AB' + (' (%s)' % _('AOE'))
                                                )
                                        ))
                                del encs
                                # spell out last encounter
                                if last_encounter is not None:
                                        lines.append('')
                                        lines.append(_('Progress notes in most recent encounter:'))
                                        lines.extend(last_encounter.format_soap (
                                                episodes = [ self._payload['pk_episode'] ],
                                                left_margin = left_margin,
                                                soap_cats = 'soapu',
                                                emr = emr
                                        ))

                # documents
                if with_documents:
                        doc_folder = patient.get_document_folder()
                        docs = doc_folder.get_documents (
                                pk_episodes = [ self._payload['pk_episode'] ]
                        )
                        if len(docs) > 0:
                                lines.append('')
                                lines.append(_('Documents: %s') % len(docs))
                        for d in docs:
                                lines.append(' ' + d.format(single_line = True))
                        del docs

                # hospitalizations
                if with_hospital_stays:
                        stays = emr.get_hospital_stays(episodes = [ self._payload['pk_episode'] ])
                        if len(stays) > 0:
                                lines.append('')
                                lines.append(_('Hospitalizations: %s') % len(stays))
                        for s in stays:
                                lines.append(s.format(left_margin = (left_margin + 1)))
                        del stays

                # procedures
                if with_procedures:
                        procs = emr.get_performed_procedures(episodes = [ self._payload['pk_episode'] ])
                        if len(procs) > 0:
                                lines.append('')
                                lines.append(_('Procedures performed: %s') % len(procs))
                        for p in procs:
                                lines.append(p.format (
                                        left_margin = (left_margin + 1),
                                        include_episode = False,
                                        include_codes = True
                                ))
                        del procs

                # family history
                if with_family_history:
                        fhx = emr.get_family_history(episodes = [ self._payload['pk_episode'] ])
                        if len(fhx) > 0:
                                lines.append('')
                                lines.append(_('Family History: %s') % len(fhx))
                        for f in fhx:
                                lines.append(f.format (
                                        left_margin = (left_margin + 1),
                                        include_episode = False,
                                        include_comment = True,
                                        include_codes = True
                                ))
                        del fhx

                # test results
                if with_tests:
                        tests = emr.get_test_results_by_date(episodes = [ self._payload['pk_episode'] ])
                        if len(tests) > 0:
                                lines.append('')
                                lines.append(_('Measurements and Results:'))
                        for t in tests:
                                lines.append(' ' + t.format_concisely(date_format = '%Y %b %d', with_notes = True))
                        del tests

                # vaccinations
                if with_vaccinations:
                        vaccs = emr.get_vaccinations (
                                episodes = [ self._payload['pk_episode'] ],
                                order_by = 'date_given DESC, vaccine'
                        )
                        if len(vaccs) > 0:
                                lines.append('')
                                lines.append(_('Vaccinations:'))
                        for vacc in vaccs:
                                lines.extend(vacc.format (
                                        with_indications = True,
                                        with_comment = True,
                                        with_reaction = True,
                                        date_format = '%Y-%m-%d'
                                ))
                        del vaccs

                lines = gmTools.strip_trailing_empty_lines(lines = lines, eol = '\n')
                if return_list:
                        return lines

                left_margin = ' ' * left_margin
                eol_w_margin = '\n%s' % left_margin
                return left_margin + eol_w_margin.join(lines) + '\n'

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_best_guess_clinical_start_date(self):
                return get_best_guess_clinical_start_date_for_episode(pk_episode = self.pk_obj)

        best_guess_clinical_start_date = property(_get_best_guess_clinical_start_date)

        #--------------------------------------------------------
        def _get_best_guess_clinical_end_date(self):
                return get_best_guess_clinical_end_date_for_episode(self.pk_obj)

        best_guess_clinical_end_date = property(_get_best_guess_clinical_end_date)

        #--------------------------------------------------------
        def _get_formatted_clinical_duration(self):
                return format_clinical_duration_of_episode (
                        start = get_best_guess_clinical_start_date_for_episode(self.pk_obj),
                        end = get_best_guess_clinical_end_date_for_episode(self.pk_obj)
                )

        formatted_clinical_duration = property(_get_formatted_clinical_duration)

        #--------------------------------------------------------
        def _get_latest_access_date(self):
                cmd = """SELECT MAX(latest) FROM (
                        -- last modification, latest = when last changed to the current state
                        (SELECT c_epi.modified_when AS latest, 'clin.episode.modified_when'::text AS candidate FROM clin.episode c_epi WHERE c_epi.pk = %(pk)s)

                                UNION ALL

                        -- last modification of encounter in which created, latest = initial creation of that encounter
                        -- DO NOT USE: just because one corrects a typo does not mean the episode took longer
                        --(SELECT c_enc.modified_when AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
                        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                        --))

                        -- end of encounter in which created, latest = explicitly set
                        -- DO NOT USE: we can retrospectively create episodes which
                        -- DO NOT USE: are long since finished
                        --(SELECT c_enc.last_affirmed AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
                        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                        --))

                        -- latest end of encounters of clinical items linked to this episode
                        (SELECT
                                MAX(last_affirmed) AS latest,
                                'clin.episode.pk = clin.clin_root_item,fk_episode -> .fk_encounter.last_affirmed'::text AS candidate
                         FROM clin.encounter
                         WHERE pk IN (
                                SELECT fk_encounter FROM clin.clin_root_item WHERE fk_episode = %(pk)s
                        ))
                                UNION ALL

                        -- latest explicit .clin_when of clinical items linked to this episode
                        (SELECT
                                MAX(clin_when) AS latest,
                                'clin.episode.pk = clin.clin_root_item,fk_episode -> .clin_when'::text AS candidate
                         FROM clin.clin_root_item
                         WHERE fk_episode = %(pk)s
                        )

                        -- latest modification time of clinical items linked to this episode
                        -- this CAN be used since if an item is linked to an episode it can be
                        -- assumed the episode (should have) existed at the time of creation
                        -- DO NOT USE, because typo fixes should not extend the episode
                        --(SELECT MIN(modified_when) AS latest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

                        -- not sure about this one:
                        -- .pk -> clin.clin_root_item.fk_encounter.modified_when

                ) AS candidates"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return rows[0][0]

        latest_access_date = property(_get_latest_access_date)

        #--------------------------------------------------------
        def _get_diagnostic_certainty_description(self):
                return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])

        diagnostic_certainty_description = property(_get_diagnostic_certainty_description)

        #--------------------------------------------------------
        def _get_formatted_revision_history(self):
                cmd = """SELECT
                                '<N/A>'::TEXT as audit__action_applied,
                                NULL AS audit__action_when,
                                '<N/A>'::TEXT AS audit__action_by,
                                pk_audit,
                                row_version,
                                modified_when,
                                modified_by,
                                pk, fk_health_issue, description, is_open, fk_encounter,
                                diagnostic_certainty_classification,
                                summary
                        FROM clin.episode
                        WHERE pk = %(pk_episode)s
                UNION ALL (
                        SELECT
                                audit_action as audit__action_applied,
                                audit_when as audit__action_when,
                                audit_by as audit__action_by,
                                pk_audit,
                                orig_version as row_version,
                                orig_when as modified_when,
                                orig_by as modified_by,
                                pk, fk_health_issue, description, is_open, fk_encounter,
                                diagnostic_certainty_classification,
                                summary
                        FROM audit.log_episode
                        WHERE pk = %(pk_episode)s
                )
                ORDER BY row_version DESC
                """
                args = {'pk_episode': self.pk_obj}
                title = _('Episode: %s%s%s') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['description'],
                        gmTools.u_right_double_angle_quote
                )
                return '\n'.join(self._get_revision_history(cmd, args, title))

        formatted_revision_history = property(_get_formatted_revision_history)

        #--------------------------------------------------------
        def _get_generic_codes(self):
                if len(self._payload['pk_generic_codes']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2episode WHERE fk_item = %(epi)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'epi': self._payload['pk_episode'],
                                        'codes': self._payload['pk_generic_codes']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2episode (fk_item, fk_generic_code) VALUES (%(epi)s, %(pk_code)s)',
                                'args': {
                                        'epi': self._payload['pk_episode'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                return

        generic_codes = property(_get_generic_codes, _set_generic_codes)

        #--------------------------------------------------------
        def _get_has_narrative(self):
                cmd = """SELECT EXISTS (
                        SELECT 1 FROM clin.clin_narrative
                        WHERE
                                fk_episode = %(epi)s
                                        AND
                                fk_encounter IN (
                                        SELECT pk FROM clin.encounter WHERE fk_patient = %(pat)s
                                )
                )"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'epi': self._payload['pk_episode']
                }
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return rows[0][0]

        has_narrative = property(_get_has_narrative)

        #--------------------------------------------------------
        def _get_health_issue(self):
                if self._payload['pk_health_issue'] is None:
                        return None
                return cHealthIssue(self._payload['pk_health_issue'])

        health_issue = property(_get_health_issue)

#============================================================
def create_episode(pk_health_issue=None, episode_name=None, is_open=False, allow_dupes=False, encounter=None, link_obj=None):
        """Creates a new episode for a given patient's health issue.

        pk_health_issue - given health issue PK
        episode_name - name of episode
        is_open - whether a *new* episode is to be open (don't tinker with existing ones)
        """
        if not allow_dupes:
                try:
                        return cEpisode(name = episode_name, health_issue = pk_health_issue, encounter = encounter, link_obj = link_obj)

                except gmExceptions.ConstructorError:
                        pass
        queries = []
        cmd = "INSERT INTO clin.episode (fk_health_issue, description, is_open, fk_encounter) VALUES (%s, %s, %s::boolean, %s)"
        queries.append({'cmd': cmd, 'args': [pk_health_issue, episode_name, is_open, encounter]})
        queries.append({'cmd': cEpisode._cmd_fetch_payload % "currval('clin.episode_pk_seq')"})
        rows = gmPG2.run_rw_queries(link_obj = link_obj, queries = queries, return_data=True)
        return cEpisode(row = {'data': rows[0], 'pk_field': 'pk_episode'})

#-----------------------------------------------------------
def delete_episode(episode=None):
        if isinstance(episode, cEpisode):
                pk = episode['pk_episode']
        else:
                pk = int(episode)

        cmd = 'DELETE FROM clin.episode WHERE pk = %(pk)s'

        try:
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': {'pk': pk}}])
        except gmPG2.dbapi.IntegrityError:
                # should be parsing pgcode/and or error message
                _log.exception('cannot delete episode, it is in use')
                return False

        return True

#-----------------------------------------------------------
def episode2problem(episode=None, allow_closed=False):
        return cProblem (
                aPK_obj = {
                        'pk_patient': episode['pk_patient'],
                        'pk_episode': episode['pk_episode'],
                        'pk_health_issue': episode['pk_health_issue']
                },
                try_potential_problems = allow_closed
        )

#-----------------------------------------------------------
_SQL_best_guess_clinical_start_date_for_episode = """
        SELECT MIN(earliest) FROM (
                -- modified_when of episode,
                -- earliest possible thereof = when created,
                -- should actually go all the way back into audit.log_episode
                (SELECT c_epi.modified_when AS earliest FROM clin.episode c_epi WHERE c_epi.pk = %(pk)s)

                        UNION ALL

                -- last modification of encounter in which created,
                -- earliest-possible thereof = initial creation of that encounter
                (SELECT c_enc.modified_when AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = (
                        SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                ))
                        UNION ALL

                -- start of encounter in which created,
                -- earliest-possible thereof = explicitly set by user
                (SELECT c_enc.started AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = (
                        SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                ))
                        UNION ALL

                -- start of encounters of clinical items linked to this episode,
                -- earliest-possible thereof = explicitly set by user
                (SELECT MIN(started) AS earliest FROM clin.encounter WHERE pk IN (
                        SELECT fk_encounter FROM clin.clin_root_item WHERE fk_episode = %(pk)s
                ))
                        UNION ALL

                -- .clin_when of clinical items linked to this episode,
                -- earliest-possible thereof = explicitly set by user
                (SELECT MIN(clin_when) AS earliest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

                        UNION ALL

                -- earliest modification time of clinical items linked to this episode
                -- this CAN be used since if an item is linked to an episode it can be
                -- assumed the episode (should have) existed at the time of creation
                (SELECT MIN(modified_when) AS earliest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

                        UNION ALL

                -- there may not be items, but there may still be documents ...
                (SELECT MIN(clin_when) AS earliest FROM blobs.doc_med WHERE fk_episode = %(pk)s)
        ) AS candidates
"""

def get_best_guess_clinical_start_date_for_episode(pk_episode=None):
        assert (pk_episode is not None), '<pk_episode> must not be None'
        query = {
                'cmd': _SQL_best_guess_clinical_start_date_for_episode,
                'args': {'pk': pk_episode}
        }
        rows = gmPG2.run_ro_queries(queries = [query])
        return rows[0][0]

#-----------------------------------------------------------
_SQL_best_guess_clinical_end_date_for_episode = """
        SELECT
                CASE WHEN
                        -- if open episode ...
                        (SELECT is_open FROM clin.episode WHERE pk = %(pk)s)
                THEN
                        -- ... no end date
                        NULL::timestamp with time zone
                ELSE (
                        SELECT COALESCE (
                                (SELECT
                                        latest --, source_type
                                 FROM (
                                        -- latest explicit .clin_when of clinical items linked to this episode
                                        (SELECT
                                                MAX(clin_when) AS latest,
                                                'clin.episode.pk = clin.clin_root_item.fk_episode -> .clin_when'::text AS source_type
                                         FROM clin.clin_root_item
                                         WHERE fk_episode = %(pk)s
                                        )
                                                UNION ALL
                                        -- latest explicit .clin_when of documents linked to this episode
                                        (SELECT
                                                MAX(clin_when) AS latest,
                                                'clin.episode.pk = blobs.doc_med.fk_episode -> .clin_when'::text AS source_type
                                         FROM blobs.doc_med
                                         WHERE fk_episode = %(pk)s
                                        )
                                 ) AS candidates
                                 ORDER BY latest DESC NULLS LAST
                                 LIMIT 1
                                ),
                                -- last ditch, always exists, only use when no clinical items or documents linked:
                                -- last modification, latest = when last changed to the current state
                                (SELECT c_epi.modified_when AS latest --, 'clin.episode.modified_when'::text AS source_type
                                 FROM clin.episode c_epi WHERE c_epi.pk = %(pk)s
                                )
                        )
                )
                END
"""

def get_best_guess_clinical_end_date_for_episode(pk_episode=None):
        assert (pk_episode is not None), '<pk_episode> must not be None'
        query = {
                'cmd': _SQL_best_guess_clinical_end_date_for_episode,
                'args': {'pk': pk_episode}
        }
        rows = gmPG2.run_ro_queries(queries = [query])
        return rows[0][0]

#-----------------------------------------------------------
def format_clinical_duration_of_episode(start=None, end=None):
        assert (start is not None), '<start> must not be None'

        if end is None:
                start_end_str = '%s-%s' % (
                        gmDateTime.pydt_strftime(start, "%b'%y"),
                        gmTools.u_ellipsis
                )
                start_end_str_long = '%s - %s' % (
                        gmDateTime.pydt_strftime(start, '%b %d %Y'),
                        gmTools.u_ellipsis
                )
                duration_str = _('%s so far') % gmDateTime.format_interval_medically(gmDateTime.pydt_now_here() - start)
                return (start_end_str, start_end_str_long, duration_str)

        duration_str = gmDateTime.format_interval_medically(end - start)
        # year different:
        if end.year != start.year:
                start_end_str = '%s-%s' % (
                        gmDateTime.pydt_strftime(start, "%b'%y"),
                        gmDateTime.pydt_strftime(end, "%b'%y")
                )
                start_end_str_long = '%s - %s' % (
                        gmDateTime.pydt_strftime(start, '%b %d %Y'),
                        gmDateTime.pydt_strftime(end, '%b %d %Y')
                )
                return (start_end_str, start_end_str_long, duration_str)
        # same year:
        if end.month != start.month:
                start_end_str = '%s-%s' % (
                        gmDateTime.pydt_strftime(start, '%b'),
                        gmDateTime.pydt_strftime(end, "%b'%y")
                )
                start_end_str_long = '%s - %s' % (
                        gmDateTime.pydt_strftime(start, '%b %d'),
                        gmDateTime.pydt_strftime(end, '%b %d %Y')
                )
                return (start_end_str, start_end_str_long, duration_str)

        # same year and same month
        start_end_str = gmDateTime.pydt_strftime(start, "%b'%y")
        start_end_str_long = gmDateTime.pydt_strftime(start, '%b %d %Y')
        return (start_end_str, start_end_str_long, duration_str)

#============================================================
class cEpisodeMatchProvider(gmMatchProvider.cMatchProvider_SQL2):
        """Find episodes for patient."""

        _SQL_episode_start = _SQL_best_guess_clinical_start_date_for_episode % {'pk': 'c_vpe.pk_episode'}
        _SQL_episode_end = _SQL_best_guess_clinical_end_date_for_episode % {'pk': 'c_vpe.pk_episode'}

        _SQL_open_episodes = """
                SELECT
                        c_vpe.pk_episode,
                        c_vpe.description
                                AS episode,
                        c_vpe.health_issue,
                        1 AS rank,
                        (%s) AS episode_start,
                        NULL::timestamp with time zone AS episode_end
                FROM
                        clin.v_pat_episodes c_vpe
                WHERE
                        c_vpe.episode_open IS TRUE
                                AND
                        c_vpe.description %%(fragment_condition)s
                        %%(ctxt_pat)s
        """ % _SQL_episode_start

        _SQL_closed_episodes = """
                SELECT
                        c_vpe.pk_episode,
                        c_vpe.description
                                AS episode,
                        c_vpe.health_issue,
                        2 AS rank,
                        (%s) AS episode_start,
                        (%s) AS episode_end
                FROM
                        clin.v_pat_episodes c_vpe
                WHERE
                        c_vpe.episode_open IS FALSE
                                AND
                        c_vpe.description %%(fragment_condition)s
                        %%(ctxt_pat)s
        """ % (
                _SQL_episode_start,
                _SQL_episode_end
        )

        #--------------------------------------------------------
        def __init__(self):
                query = """
                        (
                                %s
                        ) UNION ALL (
                                %s
                        )
                        ORDER BY rank, episode
                        LIMIT 30""" % (
                                cEpisodeMatchProvider._SQL_open_episodes,
                                cEpisodeMatchProvider._SQL_closed_episodes
                        )
                ctxt = {'ctxt_pat': {'where_part': 'AND pk_patient = %(pat)s', 'placeholder': 'pat'}}
                super().__init__(queries = [query], context = ctxt)

        #--------------------------------------------------------
        def _find_matches(self, fragment_condition):
                try:
                        pat = self._context_vals['pat']
                except KeyError:
                        pat = None
                if not pat:
                        # not patient, no search
                        return (False, [])

                return super()._find_matches(fragment_condition)

        #--------------------------------------------------------
        def _rows2matches(self, rows):
                matches = []
                for row in rows:
                        match = {
                                'weight': 0,
                                'data': row['pk_episode']
                        }
                        label = '%s (%s)%s' % (
                                row['episode'],
                                format_clinical_duration_of_episode (
                                        start = row['episode_start'],
                                        end = row['episode_end']
                                )[0],
                                gmTools.coalesce (
                                        row['health_issue'],
                                        '',
                                        ' - %s'
                                )
                        )
                        match['list_label'] = label
                        match['field_label'] = label
                        matches.append(match)
                return matches

#============================================================
# encounter API
#============================================================
SQL_get_encounters = "SELECT * FROM clin.v_pat_encounters WHERE %s"

class cEncounter(gmBusinessDBObject.cBusinessDBObject):
        """Represents one encounter."""

        _cmd_fetch_payload = SQL_get_encounters % 'pk_encounter = %s'
        _cmds_store_payload = [
                """UPDATE clin.encounter SET
                                started = %(started)s,
                                last_affirmed = %(last_affirmed)s,
                                fk_location = %(pk_org_unit)s,
                                fk_type = %(pk_type)s,
                                reason_for_encounter = gm.nullify_empty_string(%(reason_for_encounter)s),
                                assessment_of_encounter = gm.nullify_empty_string(%(assessment_of_encounter)s)
                        WHERE
                                pk = %(pk_encounter)s AND
                                xmin = %(xmin_encounter)s
                        """,
                # need to return all fields so we can survive in-place upgrades
                "SELECT * FROM clin.v_pat_encounters WHERE pk_encounter = %(pk_encounter)s"
        ]
        _updatable_fields = [
                'started',
                'last_affirmed',
                'pk_org_unit',
                'pk_type',
                'reason_for_encounter',
                'assessment_of_encounter'
        ]
        #--------------------------------------------------------
        def set_active(self):
                """Set the encounter as the active one.

                "Setting active" means making sure the encounter
                row has the youngest "last_affirmed" timestamp of
                all encounter rows for this patient.
                """
                self['last_affirmed'] = gmDateTime.pydt_now_here()
                self.save()
        #--------------------------------------------------------
        def lock(self, exclusive=False, link_obj=None):
                return lock_encounter(self.pk_obj, exclusive = exclusive, link_obj = link_obj)
        #--------------------------------------------------------
        def unlock(self, exclusive=False, link_obj=None):
                return unlock_encounter(self.pk_obj, exclusive = exclusive, link_obj = link_obj)
        #--------------------------------------------------------
        def transfer_clinical_data(self, source_episode=None, target_episode=None):
                """
                Moves every element currently linked to the current encounter
                and the source_episode onto target_episode.

                @param source_episode The episode the elements are currently linked to.
                @type target_episode A cEpisode intance.
                @param target_episode The episode the elements will be relinked to.
                @type target_episode A cEpisode intance.
                """
                if source_episode['pk_episode'] == target_episode['pk_episode']:
                        return True

                cmd = """
                        UPDATE clin.clin_root_item
                        SET fk_episode = %(trg)s
                        WHERE
                                fk_encounter = %(enc)s AND
                                fk_episode = %(src)s
                        """
                gmPG2.run_rw_queries(queries = [{
                        'cmd': cmd,
                        'args': {
                                'trg': target_episode['pk_episode'],
                                'enc': self.pk_obj,
                                'src': source_episode['pk_episode']
                        }
                }])
                self.refetch_payload()
                return True

        #--------------------------------------------------------
        def transfer_all_data_to_another_encounter(self, pk_target_encounter=None):
                if pk_target_encounter == self.pk_obj:
                        return True
                cmd = "SELECT clin.transfer_all_encounter_data(%(src)s, %(trg)s)"
                args = {
                        'src': self.pk_obj,
                        'trg': pk_target_encounter
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

#       conn = gmPG2.get_connection()
#       curs = conn.cursor()
#       curs.callproc('clin.get_hints_for_patient', [pk_identity])
#       rows = curs.fetchall()
#       idx = gmPG2.get_col_indices(curs)
#       curs.close()
#       conn.rollback()

        #--------------------------------------------------------
        def same_payload(self, another_object=None):

                relevant_fields = [
                        'pk_org_unit',
                        'pk_type',
                        'pk_patient',
                        'reason_for_encounter',
                        'assessment_of_encounter'
                ]
                for field in relevant_fields:
                        if self._payload[field] != another_object[field]:
                                _log.debug('mismatch on [%s]: "%s" vs. "%s"', field, self._payload[field], another_object[field])
                                return False

                relevant_fields = [
                        'started',
                        'last_affirmed',
                ]
                for field in relevant_fields:
                        if self._payload[field] is None:
                                if another_object[field] is None:
                                        continue
                                _log.debug('mismatch on [%s]: here="%s", other="%s"', field, self._payload[field], another_object[field])
                                return False

                        if another_object[field] is None:
                                return False

                        # compares at seconds granularity
                        if self._payload[field].strftime('%Y-%m-%d %H:%M:%S') != another_object[field].strftime('%Y-%m-%d %H:%M:%S'):
                                _log.debug('mismatch on [%s]: here="%s", other="%s"', field, self._payload[field], another_object[field])
                                return False

                # compare codes
                # 1) RFE
                if another_object['pk_generic_codes_rfe'] is None:
                        if self._payload['pk_generic_codes_rfe'] is not None:
                                return False
                if another_object['pk_generic_codes_rfe'] is not None:
                        if self._payload['pk_generic_codes_rfe'] is None:
                                return False
                if (
                        (another_object['pk_generic_codes_rfe'] is None)
                                and
                        (self._payload['pk_generic_codes_rfe'] is None)
                ) is False:
                        if set(another_object['pk_generic_codes_rfe']) != set(self._payload['pk_generic_codes_rfe']):
                                return False
                # 2) AOE
                if another_object['pk_generic_codes_aoe'] is None:
                        if self._payload['pk_generic_codes_aoe'] is not None:
                                return False
                if another_object['pk_generic_codes_aoe'] is not None:
                        if self._payload['pk_generic_codes_aoe'] is None:
                                return False
                if (
                        (another_object['pk_generic_codes_aoe'] is None)
                                and
                        (self._payload['pk_generic_codes_aoe'] is None)
                ) is False:
                        if set(another_object['pk_generic_codes_aoe']) != set(self._payload['pk_generic_codes_aoe']):
                                return False

                return True
        #--------------------------------------------------------
        def has_clinical_data(self):
                cmd = """
select exists (
        select 1 from clin.v_pat_items where pk_patient = %(pat)s and pk_encounter = %(enc)s
                union all
        select 1 from blobs.v_doc_med where pk_patient = %(pat)s and pk_encounter = %(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]

        #--------------------------------------------------------
        def has_narrative(self):
                cmd = """
select exists (
        select 1 from clin.v_pat_items where pk_patient=%(pat)s and pk_encounter=%(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
        #--------------------------------------------------------
        def has_soap_narrative(self, soap_cats=None):
                """soap_cats: <space> = admin category"""

                if soap_cats is None:
                        soap_cats = 'soap '
                else:
                        soap_cats = soap_cats.casefold()

                cats = []
                for cat in soap_cats:
                        if cat in 'soapu':
                                cats.append(cat)
                                continue
                        if cat == ' ':
                                cats.append(None)

                cmd = """
                        SELECT EXISTS (
                                SELECT 1 FROM clin.clin_narrative
                                WHERE
                                        fk_encounter = %(enc)s
                                                AND
                                        soap_cat = ANY(%(cats)s)
                                LIMIT 1
                        )
                """
                args = {'enc': self._payload['pk_encounter'], 'cats': cats}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd,'args': args}])
                return rows[0][0]
        #--------------------------------------------------------
        def has_documents(self):
                cmd = """
select exists (
        select 1 from blobs.v_doc_med where pk_patient = %(pat)s and pk_encounter = %(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
        #--------------------------------------------------------
        def get_latest_soap(self, soap_cat=None, episode=None):

                if soap_cat is not None:
                        soap_cat = soap_cat.casefold()

                if episode is None:
                        epi_part = 'fk_episode is null'
                else:
                        epi_part = 'fk_episode = %(epi)s'

                cmd = """
select narrative
from clin.clin_narrative
where
        fk_encounter = %%(enc)s
                and
        soap_cat = %%(cat)s
                and
        %s
order by clin_when desc
limit 1
                """ % epi_part

                args = {'enc': self.pk_obj, 'cat': soap_cat, 'epi': episode}

                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                if len(rows) == 0:
                        return None

                return rows[0][0]
        #--------------------------------------------------------
        def get_episodes(self, exclude=None):
                cmd = """
                        SELECT * FROM clin.v_pat_episodes
                        WHERE pk_episode IN (
                                        SELECT DISTINCT fk_episode
                                        FROM clin.clin_root_item
                                        WHERE fk_encounter = %%(enc)s

                                                UNION

                                        SELECT DISTINCT fk_episode
                                        FROM blobs.doc_med
                                        WHERE fk_encounter = %%(enc)s
                        ) %s"""
                args = {'enc': self.pk_obj}
                if exclude is not None:
                        cmd = cmd % 'AND pk_episode <> ALL(%(excluded)s)'
                        args['excluded'] = exclude
                else:
                        cmd = cmd % ''
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]

        episodes = property(get_episodes)

        #--------------------------------------------------------
        def add_code(self, pk_code=None, field=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                if field == 'rfe':
                        cmd = "INSERT INTO clin.lnk_code2rfe (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
                elif field == 'aoe':
                        cmd = "INSERT INTO clin.lnk_code2aoe (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
                else:
                        raise ValueError('<field> must be one of "rfe" or "aoe", not "%s"', field)
                args = {
                        'item': self._payload['pk_encounter'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def remove_code(self, pk_code=None, field=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                if field == 'rfe':
                        cmd = "DELETE FROM clin.lnk_code2rfe WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                elif field == 'aoe':
                        cmd = "DELETE FROM clin.lnk_code2aoe WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                else:
                        raise ValueError('<field> must be one of "rfe" or "aoe", not "%s"', field)
                args = {
                        'item': self._payload['pk_encounter'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        # data formatting
        #--------------------------------------------------------
        def format_soap(self, episodes=None, left_margin=0, soap_cats='soapu', emr=None, issues=None):

                lines = []
                for soap_cat in gmSoapDefs.soap_cats_str2list(soap_cats):
                        soap_cat_narratives = emr.get_clin_narrative (
                                episodes = episodes,
                                issues = issues,
                                encounters = [self._payload['pk_encounter']],
                                soap_cats = [soap_cat]
                        )
                        if soap_cat_narratives is None:
                                continue
                        if len(soap_cat_narratives) == 0:
                                continue

                        lines.append('%s%s %s %s' % (
                                gmTools.u_box_top_left_arc,
                                gmTools.u_box_horiz_single,
                                gmSoapDefs.soap_cat2l10n_str[soap_cat],
                                gmTools.u_box_horiz_single * 5
                        ))
                        for soap_entry in soap_cat_narratives:
                                txt = gmTools.wrap (
                                        text = soap_entry['narrative'],
                                        width = 75,
                                        initial_indent = '',
                                        subsequent_indent = (' ' * left_margin)
                                )
                                lines.append(txt)
                                when = gmDateTime.pydt_strftime (
                                        soap_entry['date'],
                                        format = '%Y-%m-%d %H:%M',
                                        accuracy = gmDateTime.acc_minutes
                                )
                                txt = '%s%s %.8s, %s %s' % (
                                        ' ' * 40,
                                        gmTools.u_box_horiz_light_heavy,
                                        soap_entry['modified_by'],
                                        when,
                                        gmTools.u_box_horiz_heavy_light
                                )
                                lines.append(txt)
                                lines.append('')

                return lines

        #--------------------------------------------------------
        def format_latex(self, date_format=None, soap_cats=None, soap_order=None):

                nothing2format = (
                        (self._payload['reason_for_encounter'] is None)
                                and
                        (self._payload['assessment_of_encounter'] is None)
                                and
                        (self.has_soap_narrative(soap_cats = 'soapu') is False)
                )
                if nothing2format:
                        return ''

                if date_format is None:
                        date_format = '%A, %b %d %Y'

                tex =  '% -------------------------------------------------------------\n'
                tex += '% much recommended: \\usepackage(tabu)\n'
                tex += '% much recommended: \\usepackage(longtable)\n'
                tex += '% best wrapped in: "\\begin{longtabu} to \\textwidth {lX[,L]}"\n'
                tex += '% -------------------------------------------------------------\n'
                tex += '\\hline \n'
                tex += '\\multicolumn{2}{l}{%s: %s ({\\footnotesize %s - %s})} \\tabularnewline \n' % (
                        gmTools.tex_escape_string(self._payload['l10n_type']),
                        gmTools.tex_escape_string (
                                gmDateTime.pydt_strftime (
                                        self._payload['started'],
                                        date_format,
                                        accuracy = gmDateTime.acc_days
                                )
                        ),
                        gmTools.tex_escape_string (
                                gmDateTime.pydt_strftime (
                                        self._payload['started'],
                                        '%H:%M',
                                        accuracy = gmDateTime.acc_minutes
                                )
                        ),
                        gmTools.tex_escape_string (
                                gmDateTime.pydt_strftime (
                                        self._payload['last_affirmed'],
                                        '%H:%M',
                                        accuracy = gmDateTime.acc_minutes
                                )
                        )
                )
                tex += '\\hline \n'

                if self._payload['reason_for_encounter'] is not None:
                        tex += '%s & %s \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('RFE')),
                                gmTools.tex_escape_string(self._payload['reason_for_encounter'])
                        )
                if self._payload['assessment_of_encounter'] is not None:
                        tex += '%s & %s \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('AOE')),
                                gmTools.tex_escape_string(self._payload['assessment_of_encounter'])
                        )

                for epi in self.get_episodes():
                        soaps = epi.get_narrative(soap_cats = soap_cats, encounters = [self.pk_obj], order_by = soap_order)
                        if len(soaps) == 0:
                                continue
                        tex += '\\multicolumn{2}{l}{\\emph{%s: %s%s}} \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('Problem')),
                                gmTools.tex_escape_string(epi['description']),
                                gmTools.tex_escape_string (
                                        gmTools.coalesce (
                                                value2test = diagnostic_certainty_classification2str(epi['diagnostic_certainty_classification']),
                                                return_instead = '',
                                                template4value = ' {\\footnotesize [%s]}',
                                                none_equivalents = [None, '']
                                        )
                                )
                        )
                        if epi['pk_health_issue'] is not None:
                                tex += '\\multicolumn{2}{l}{\\emph{%s: %s%s}} \\tabularnewline \n' % (
                                        gmTools.tex_escape_string(_('Health issue')),
                                        gmTools.tex_escape_string(epi['health_issue']),
                                        gmTools.tex_escape_string (
                                                gmTools.coalesce (
                                                        value2test = diagnostic_certainty_classification2str(epi['diagnostic_certainty_classification_issue']),
                                                        return_instead = '',
                                                        template4value = ' {\\footnotesize [%s]}',
                                                        none_equivalents = [None, '']
                                                )
                                        )
                                )
                        for soap in soaps:
                                tex += '{\\small %s} & {\\small %s} \\tabularnewline \n' % (
                                        gmTools.tex_escape_string(gmSoapDefs.soap_cat2l10n[soap['soap_cat']]),
                                        gmTools.tex_escape_string(soap['narrative'], replace_eol = True)
                                )
                        tex += ' & \\tabularnewline \n'

                return tex

        #--------------------------------------------------------
        def __format_header_fancy(self, left_margin=0):
                lines = []

                lines.append('%s%s: %s - %s (@%s)%s [#%s]' % (
                        ' ' * left_margin,
                        self._payload['l10n_type'],
                        self._payload['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                        self._payload['last_affirmed_original_tz'].strftime('%H:%M'),
                        self._payload['source_time_zone'],
                        gmTools.coalesce (
                                self._payload['assessment_of_encounter'],
                                '',
                                ' %s%%s%s' % (gmTools.u_left_double_angle_quote, gmTools.u_right_double_angle_quote)
                        ),
                        self._payload['pk_encounter']
                ))

                lines.append(_('  your time: %s - %s  (@%s = %s%s)\n') % (
                        self._payload['started'].strftime('%Y-%m-%d %H:%M'),
                        self._payload['last_affirmed'].strftime('%H:%M'),
                        gmDateTime.current_local_timezone_name,
                        gmTools.bool2subst (
                                gmDateTime.dst_currently_in_effect,
                                gmDateTime.py_dst_timezone_name,
                                gmDateTime.py_timezone_name
                        ),
                        gmTools.bool2subst(gmDateTime.dst_currently_in_effect, ' - ' + _('daylight savings time in effect'), '')
                ))

                if self._payload['praxis_branch'] is not None:
                        lines.append(_('Location: %s (%s)') % (self._payload['praxis_branch'], self._payload['praxis']))

                if self._payload['reason_for_encounter'] is not None:
                        lines.append('%s: %s' % (_('RFE'), self._payload['reason_for_encounter']))
                        codes = self.generic_codes_rfe
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')

                if self._payload['assessment_of_encounter'] is not None:
                        lines.append('%s: %s' % (_('AOE'), self._payload['assessment_of_encounter']))
                        codes = self.generic_codes_aoe
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')
                        del codes
                return lines

        #--------------------------------------------------------
        def format_header(self, fancy_header=True, left_margin=0, with_rfe_aoe=False):
                lines = []

                if fancy_header:
                        return self.__format_header_fancy(left_margin = left_margin)

                now = gmDateTime.pydt_now_here()
                if now.strftime('%Y-%m-%d') == self._payload['started_original_tz'].strftime('%Y-%m-%d'):
                        start = '%s %s' % (
                                _('today'),
                                self._payload['started_original_tz'].strftime('%H:%M')
                        )
                else:
                        start = self._payload['started_original_tz'].strftime('%Y-%m-%d %H:%M')
                lines.append('%s%s: %s - %s%s%s' % (
                        ' ' * left_margin,
                        self._payload['l10n_type'],
                        start,
                        self._payload['last_affirmed_original_tz'].strftime('%H:%M'),
                        gmTools.coalesce(self._payload['assessment_of_encounter'], '', ' \u00BB%s\u00AB'),
                        gmTools.coalesce(self._payload['praxis_branch'], '', ' @%s')
                ))
                if with_rfe_aoe:
                        if self._payload['reason_for_encounter'] is not None:
                                lines.append('%s: %s' % (_('RFE'), self._payload['reason_for_encounter']))
                        codes = self.generic_codes_rfe
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')
                        if self._payload['assessment_of_encounter'] is not None:
                                lines.append('%s: %s' % (_('AOE'), self._payload['assessment_of_encounter']))
                        codes = self.generic_codes_aoe
                        if len(codes) > 0:
                                lines.append('')
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')
                        del codes

                return lines

        #--------------------------------------------------------
        def format_by_episode(self, episodes=None, issues=None, left_margin=0, patient=None, with_soap=False, with_tests=True, with_docs=True, with_vaccinations=True, with_family_history=True):

                if patient is not None:
                        emr = patient.emr

                lines = []
                if episodes is None:
                        episodes = [ e['pk_episode'] for e in self.episodes ]

                for pk in episodes:
                        epi = cEpisode(aPK_obj = pk)
                        lines.append(_('\nEpisode %s%s%s%s:') % (
                                gmTools.u_left_double_angle_quote,
                                epi['description'],
                                gmTools.u_right_double_angle_quote,
                                gmTools.coalesce(epi['health_issue'], '', ' (%s)')
                        ))

                        # soap
                        if with_soap:
                                if patient.ID != self._payload['pk_patient']:
                                        msg = '<patient>.ID = %s but encounter %s belongs to patient %s' % (
                                                patient.ID,
                                                self._payload['pk_encounter'],
                                                self._payload['pk_patient']
                                        )
                                        raise ValueError(msg)
                                lines.extend(self.format_soap (
                                        episodes = [pk],
                                        left_margin = left_margin,
                                        soap_cats = None,               # meaning: all
                                        emr = emr,
                                        issues = issues
                                ))

                        # test results
                        if with_tests:
                                tests = emr.get_test_results_by_date (
                                        episodes = [pk],
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(tests) > 0:
                                        lines.append('')
                                        lines.append(_('Measurements and Results:'))

                                for t in tests:
                                        lines.append(t.format())

                                del tests

                        # vaccinations
                        if with_vaccinations:
                                vaccs = emr.get_vaccinations (
                                        episodes = [pk],
                                        encounters = [ self._payload['pk_encounter'] ],
                                        order_by = 'date_given DESC, vaccine'
                                )
                                if len(vaccs) > 0:
                                        lines.append('')
                                        lines.append(_('Vaccinations:'))
                                for vacc in vaccs:
                                        lines.extend(vacc.format (
                                                with_indications = True,
                                                with_comment = True,
                                                with_reaction = True,
                                                date_format = '%Y-%m-%d'
                                        ))
                                del vaccs

                        # family history
                        if with_family_history:
                                fhx = emr.get_family_history(episodes = [pk])
                                if len(fhx) > 0:
                                        lines.append('')
                                        lines.append(_('Family History: %s') % len(fhx))
                                for f in fhx:
                                        lines.append(f.format (
                                                left_margin = (left_margin + 1),
                                                include_episode = False,
                                                include_comment = True
                                        ))
                                del fhx

                        # documents
                        if with_docs:
                                doc_folder = patient.get_document_folder()
                                docs = doc_folder.get_documents (
                                        pk_episodes = [pk],
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(docs) > 0:
                                        lines.append('')
                                        lines.append(_('Documents:'))
                                for d in docs:
                                        lines.append(' ' + d.format(single_line = True))
                                del docs

                return lines

        #--------------------------------------------------------
        def format_maximum_information(self, patient=None):
                if patient is None:
                        from Gnumed.business.gmPerson import gmCurrentPatient, cPerson
                        if self._payload['pk_patient'] == gmCurrentPatient().ID:
                                patient = gmCurrentPatient()
                        else:
                                patient = cPerson(self._payload['pk_patient'])

                return self.format (
                        patient = patient,
                        fancy_header = True,
                        with_rfe_aoe = True,
                        with_soap = True,
                        with_docs = True,
                        with_tests = False,
                        with_vaccinations = True,
                        with_co_encountlet_hints = True,
                        with_family_history = True,
                        by_episode = False,
                        return_list = True
                )

        #--------------------------------------------------------
        def format(self, episodes=None, with_soap=False, left_margin=0, patient=None, issues=None, with_docs=True, with_tests=True, fancy_header=True, with_vaccinations=True, with_co_encountlet_hints=False, with_rfe_aoe=False, with_family_history=True, by_episode=False, return_list=False):
                """Format an encounter.

                Args:
                        episode: which episodes, touched upon by this encounter, to include information for
                        with_co_encountlet_hints:
                                - whether to include which *other* episodes were discussed during this encounter
                                - (only makes sense if episodes is not None, since that would preclude information)
                """
                lines = self.format_header (
                        fancy_header = fancy_header,
                        left_margin = left_margin,
                        with_rfe_aoe = with_rfe_aoe
                )

                if patient is None:
                        _log.debug('no patient, cannot load patient related data')
                        with_soap = False
                        with_tests = False
                        with_vaccinations = False
                        with_docs = False

                if by_episode:
                        lines.extend(self.format_by_episode (
                                episodes = episodes,
                                issues = issues,
                                left_margin = left_margin,
                                patient = patient,
                                with_soap = with_soap,
                                with_tests = with_tests,
                                with_docs = with_docs,
                                with_vaccinations = with_vaccinations,
                                with_family_history = with_family_history
                        ))
                else:
                        if with_soap:
                                lines.append('')
                                if patient.ID != self._payload['pk_patient']:
                                        msg = '<patient>.ID = %s but encounter %s belongs to patient %s' % (
                                                patient.ID,
                                                self._payload['pk_encounter'],
                                                self._payload['pk_patient']
                                        )
                                        raise ValueError(msg)
                                emr = patient.emr
                                lines.extend(self.format_soap (
                                        episodes = episodes,
                                        left_margin = left_margin,
                                        soap_cats = None,               # meaning: all
                                        emr = emr,
                                        issues = issues
                                ))

        #               # family history
        #               if with_family_history:
        #                       if episodes is not None:
        #                               fhx = emr.get_family_history(episodes = episodes)
        #                               if len(fhx) > 0:
        #                                       lines.append(u'')
        #                                       lines.append(_('Family History: %s') % len(fhx))
        #                               for f in fhx:
        #                                       lines.append(f.format (
        #                                               left_margin = (left_margin + 1),
        #                                               include_episode = False,
        #                                               include_comment = True
        #                                       ))
        #                               del fhx

                        # test results
                        if with_tests:
                                emr = patient.emr
                                tests = emr.get_test_results_by_date (
                                        episodes = episodes,
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(tests) > 0:
                                        lines.append('')
                                        lines.append(_('Measurements and Results:'))
                                for t in tests:
                                        lines.append(t.format())
                                del tests

                        # vaccinations
                        if with_vaccinations:
                                emr = patient.emr
                                vaccs = emr.get_vaccinations (
                                        episodes = episodes,
                                        encounters = [ self._payload['pk_encounter'] ],
                                        order_by = 'date_given DESC, vaccine'
                                )
                                if len(vaccs) > 0:
                                        lines.append('')
                                        lines.append(_('Vaccinations:'))
                                for vacc in vaccs:
                                        lines.extend(vacc.format (
                                                with_indications = True,
                                                with_comment = True,
                                                with_reaction = True,
                                                date_format = '%Y-%m-%d'
                                        ))
                                del vaccs

                        # documents
                        if with_docs:
                                doc_folder = patient.get_document_folder()
                                docs = doc_folder.get_documents (
                                        pk_episodes = episodes,
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(docs) > 0:
                                        lines.append('')
                                        lines.append(_('Documents:'))
                                for d in docs:
                                        lines.append(' ' + d.format(single_line = True))
                                del docs

                        # co-encountlets
                        if with_co_encountlet_hints:
                                if episodes is not None:
                                        other_epis = self.get_episodes(exclude = episodes)
                                        if len(other_epis) > 0:
                                                lines.append('')
                                                lines.append(_('%s other episodes touched upon during this encounter:') % len(other_epis))
                                        for epi in other_epis:
                                                lines.append(' %s%s%s%s' % (
                                                        gmTools.u_left_double_angle_quote,
                                                        epi['description'],
                                                        gmTools.u_right_double_angle_quote,
                                                        gmTools.coalesce(epi['health_issue'], '', ' (%s)')
                                                ))

                if return_list:
                        return lines

                eol_w_margin = '\n%s' % (' ' * left_margin)
                return '%s\n' % eol_w_margin.join(lines)

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_generic_codes_rfe(self):
                if len(self._payload['pk_generic_codes_rfe']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes_rfe']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes_rfe(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes_rfe']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2rfe WHERE fk_item = %(enc)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'codes': self._payload['pk_generic_codes_rfe']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2rfe (fk_item, fk_generic_code) VALUES (%(enc)s, %(pk_code)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                self.refetch_payload()
                return

        generic_codes_rfe = property(_get_generic_codes_rfe, _set_generic_codes_rfe)
        #--------------------------------------------------------
        def _get_generic_codes_aoe(self):
                if len(self._payload['pk_generic_codes_aoe']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes_aoe']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes_aoe(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes_aoe']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2aoe WHERE fk_item = %(enc)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'codes': self._payload['pk_generic_codes_aoe']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2aoe (fk_item, fk_generic_code) VALUES (%(enc)s, %(pk_code)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                self.refetch_payload()
                return

        generic_codes_aoe = property(_get_generic_codes_aoe, _set_generic_codes_aoe)
        #--------------------------------------------------------
        def _get_praxis_branch(self):
                if self._payload['pk_org_unit'] is None:
                        return None
                return gmPraxis.get_praxis_branch_by_org_unit(pk_org_unit = self._payload['pk_org_unit'])

        praxis_branch = property(_get_praxis_branch)
        #--------------------------------------------------------
        def _get_org_unit(self):
                if self._payload['pk_org_unit'] is None:
                        return None
                return gmOrganization.cOrgUnit(aPK_obj = self._payload['pk_org_unit'])

        org_unit = property(_get_org_unit)
        #--------------------------------------------------------
        def _get_formatted_revision_history(self):
                cmd = """SELECT
                                '<N/A>'::TEXT as audit__action_applied,
                                NULL AS audit__action_when,
                                '<N/A>'::TEXT AS audit__action_by,
                                pk_audit,
                                row_version,
                                modified_when,
                                modified_by,
                                pk, fk_patient, fk_type, fk_location, source_time_zone, reason_for_encounter, assessment_of_encounter, started, last_affirmed
                        FROM clin.encounter
                        WHERE pk = %(pk_encounter)s
                UNION ALL (
                        SELECT
                                audit_action as audit__action_applied,
                                audit_when as audit__action_when,
                                audit_by as audit__action_by,
                                pk_audit,
                                orig_version as row_version,
                                orig_when as modified_when,
                                orig_by as modified_by,
                                pk, fk_patient, fk_type, fk_location, source_time_zone, reason_for_encounter, assessment_of_encounter, started, last_affirmed
                        FROM audit.log_encounter
                        WHERE pk = %(pk_encounter)s
                )
                ORDER BY row_version DESC
                """
                args = {'pk_encounter': self._payload['pk_encounter']}
                title = _('Encounter: %s%s%s') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['l10n_type'],
                        gmTools.u_right_double_angle_quote
                )
                return '\n'.join(self._get_revision_history(cmd, args, title))

        formatted_revision_history = property(_get_formatted_revision_history)

#-----------------------------------------------------------
def create_encounter(fk_patient=None, enc_type=None):
        """Creates a new encounter for a patient.

        fk_patient - patient PK
        enc_type - type of encounter
        """
        if enc_type is None:
                enc_type = 'in surgery'
        # insert new encounter
        queries = []
        try:
                enc_type = int(enc_type)
                cmd = """
                        INSERT INTO clin.encounter (fk_patient, fk_type, fk_location)
                        VALUES (%(pat)s, %(typ)s, %(prax)s) RETURNING pk"""
        except ValueError:
                enc_type = enc_type
                cmd = """
                        INSERT INTO clin.encounter (fk_patient, fk_location, fk_type)
                        VALUES (
                                %(pat)s,
                                %(prax)s,
                                coalesce (
                                        (select pk from clin.encounter_type where description = %(typ)s),
                                        -- pick the first available
                                        (select pk from clin.encounter_type limit 1)
                                )
                        ) RETURNING pk"""
        praxis = gmPraxis.gmCurrentPraxisBranch()
        args = {'pat': fk_patient, 'typ': enc_type, 'prax': praxis['pk_org_unit']}
        queries.append({'cmd': cmd, 'args': args})
        rows = gmPG2.run_rw_queries(queries = queries, return_data = True)
        encounter = cEncounter(aPK_obj = rows[0]['pk'])

        return encounter

#------------------------------------------------------------
def lock_encounter(pk_encounter, exclusive=False, link_obj=None):
        """Used to protect against deletion of active encounter from another client."""
        return gmPG2.lock_row(link_obj = link_obj, table = 'clin.encounter', pk = pk_encounter, exclusive = exclusive)

#------------------------------------------------------------
def unlock_encounter(pk_encounter, exclusive=False, link_obj=None):
        unlocked = gmPG2.unlock_row(link_obj = link_obj, table = 'clin.encounter', pk = pk_encounter, exclusive = exclusive)
        if not unlocked:
                _log.warning('cannot unlock encounter [#%s]', pk_encounter)
                call_stack = inspect.stack()
                call_stack.reverse()
                for idx in range(1, len(call_stack)):
                        caller = call_stack[idx]
                        _log.error('%s[%s] @ [%s] in [%s]', ' ' * idx, caller[3], caller[2], caller[1])
                del call_stack
        return unlocked

#-----------------------------------------------------------
def delete_encounter(pk_encounter):
        """Deletes an encounter by PK.

        - attempts to obtain an exclusive lock which should
          fail if the encounter is the active encounter in
          this or any other client
        - catches DB exceptions which should mostly be related
          to clinical data already having been attached to
          the encounter thus making deletion fail
        """
        conn = gmPG2.get_connection(readonly = False)
        if not lock_encounter(pk_encounter, exclusive = True, link_obj = conn):
                _log.debug('cannot lock encounter [%s] for deletion, it seems in use', pk_encounter)
                return False
        cmd = """DELETE FROM clin.encounter WHERE pk = %(enc)s"""
        args = {'enc': pk_encounter}
        try:
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        except gmPG2.PG_ERROR_EXCEPTION as exc:
                _log.exception('cannot delete encounter [%s]', pk_encounter)
                gmPG2.log_pg_exception_details(exc)
                unlock_encounter(pk_encounter, exclusive = True, link_obj = conn)
                return False
        unlock_encounter(pk_encounter, exclusive = True, link_obj = conn)
        return True

#-----------------------------------------------------------
# encounter types handling
#-----------------------------------------------------------
def update_encounter_type(description=None, l10n_description=None):

        rows = gmPG2.run_rw_queries(
                queries = [{
                'cmd': "select i18n.upd_tx(%(desc)s, %(l10n_desc)s)",
                'args': {'desc': description, 'l10n_desc': l10n_description}
                }],
                return_data = True
        )

        success = rows[0][0]
        if not success:
                _log.warning('updating encounter type [%s] to [%s] failed', description, l10n_description)

        return {'description': description, 'l10n_description': l10n_description}
#-----------------------------------------------------------
def create_encounter_type(description=None, l10n_description=None):
        """This will attempt to create a NEW encounter type."""

        # need a system name, so derive one if necessary
        if description is None:
                description = l10n_description

        args = {
                'desc': description,
                'l10n_desc': l10n_description
        }

        _log.debug('creating encounter type: %s, %s', description, l10n_description)

        # does it exist already ?
        cmd = "select description, _(description) from clin.encounter_type where description = %(desc)s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])

        # yes
        if len(rows) > 0:
                # both system and l10n name are the same so all is well
                if (rows[0][0] == description) and (rows[0][1] == l10n_description):
                        _log.info('encounter type [%s] already exists with the proper translation')
                        return {'description': description, 'l10n_description': l10n_description}

                # or maybe there just wasn't a translation to
                # the current language for this type yet ?
                cmd = "select exists (select 1 from i18n.translations where orig = %(desc)s and lang = i18n.get_curr_lang())"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])

                # there was, so fail
                if rows[0][0]:
                        _log.error('encounter type [%s] already exists but with another translation')
                        return None

                # else set it
                cmd = "select i18n.upd_tx(%(desc)s, %(l10n_desc)s)"
                rows = gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return {'description': description, 'l10n_description': l10n_description}

        # no
        queries = [
                {'cmd': "insert into clin.encounter_type (description) values (%(desc)s)", 'args': args},
                {'cmd': "select i18n.upd_tx(%(desc)s, %(l10n_desc)s)", 'args': args}
        ]
        rows = gmPG2.run_rw_queries(queries = queries)

        return {'description': description, 'l10n_description': l10n_description}

#-----------------------------------------------------------
def get_most_commonly_used_encounter_type():
        cmd = """
                SELECT
                        COUNT(*) AS type_count,
                        fk_type
                FROM clin.encounter
                GROUP BY fk_type
                ORDER BY type_count DESC
                LIMIT 1
        """
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd}])
        if len(rows) == 0:
                return None
        return rows[0]['fk_type']

#-----------------------------------------------------------
def get_encounter_types():
        cmd = """
                SELECT
                        _(description) AS l10n_description,
                        description
                FROM
                        clin.encounter_type
                ORDER BY
                        l10n_description
        """
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd}])
        return rows

#-----------------------------------------------------------
def get_encounter_type(description=None):
        cmd = "SELECT * from clin.encounter_type where description = %s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': [description]}])
        return rows

#-----------------------------------------------------------
def delete_encounter_type(description=None):
        deleted = False
        cmd = "delete from clin.encounter_type where description = %(desc)s"
        args = {'desc': description}
        try:
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                deleted = True
        except gmPG2.dbapi.IntegrityError as e:
                if e.pgcode != gmPG2.PG_error_codes.FOREIGN_KEY_VIOLATION:
                        raise

        return deleted

#============================================================
class cProblem(gmBusinessDBObject.cBusinessDBObject):
        """Represents one problem.

        problems are the aggregation of
                .clinically_relevant=True issues and
                .is_open=True episodes
        """
        _cmd_fetch_payload = ''                                 # will get programmatically defined in __init__
        _cmds_store_payload:list = ["select 1"]
        _updatable_fields:list = []

        #--------------------------------------------------------
        def __init__(self, aPK_obj=None, try_potential_problems=False):
                """Initialize.

                aPK_obj must contain the keys
                        pk_patient
                        pk_episode
                        pk_health_issue
                """
                if aPK_obj is None:
                        raise gmExceptions.ConstructorError('cannot instatiate cProblem for PK: [%s]' % (aPK_obj))

                # As problems are rows from a view of different emr struct items,
                # the PK can't be a single field and, as some of the values of the
                # composed PK may be None, they must be queried using 'is null',
                # so we must programmatically construct the SQL query
                where_parts = []
                pk = {}
                for col_name in aPK_obj:
                        val = aPK_obj[col_name]
                        if val is None:
                                where_parts.append('%s IS NULL' % col_name)
                        else:
                                where_parts.append('%s = %%(%s)s' % (col_name, col_name))
                                pk[col_name] = val

                # try to instantiate from true problem view
                cProblem._cmd_fetch_payload = """
                        SELECT *, False as is_potential_problem
                        FROM clin.v_problem_list
                        WHERE %s""" % ' AND '.join(where_parts)

                try:
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj = pk)
                        return

                except gmExceptions.ConstructorError:
                        _log.exception('actual problem not found, trying "potential" problems')
                        if try_potential_problems is False:
                                raise

                # try to instantiate from potential-problems view
                cProblem._cmd_fetch_payload = """
                        SELECT *, True as is_potential_problem
                        FROM clin.v_potential_problem_list
                        WHERE %s""" % ' AND '.join(where_parts)
                gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj=pk)
        #--------------------------------------------------------
        def get_as_episode(self):
                """
                Retrieve the cEpisode instance equivalent to this problem.
                The problem's type attribute must be 'episode'
                """
                if self._payload['type'] != 'episode':
                        _log.error('cannot convert problem [%s] of type [%s] to episode' % (self._payload['problem'], self._payload['type']))
                        return None
                return cEpisode(aPK_obj = self._payload['pk_episode'])
        #--------------------------------------------------------
        def get_as_health_issue(self):
                """
                Retrieve the cHealthIssue instance equivalent to this problem.
                The problem's type attribute must be 'issue'
                """
                if self._payload['type'] != 'issue':
                        _log.error('cannot convert problem [%s] of type [%s] to health issue' % (self._payload['problem'], self._payload['type']))
                        return None
                return cHealthIssue(aPK_obj = self._payload['pk_health_issue'])
        #--------------------------------------------------------
        def get_visual_progress_notes(self, encounter_id=None):
                if self._payload['type'] == 'issue':
                        latest = cHealthIssue(aPK_obj = self._payload['pk_health_issue']).latest_episode
                        if latest is None:
                                return []

                        pk_episode = latest['pk_episode']
                else:
                        pk_episode = self._payload['pk_episode']
                doc_folder = gmDocuments.cDocumentFolder(aPKey = self._payload['pk_patient'])
                return doc_folder.get_visual_progress_notes(episodes = [pk_episode])

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        # doubles as 'diagnostic_certainty_description' getter:
        def get_diagnostic_certainty_description(self):
                return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])

        diagnostic_certainty_description = property(get_diagnostic_certainty_description)
        #--------------------------------------------------------
        def _get_generic_codes(self):
                if self._payload['type'] == 'issue':
                        cmd = """
                                SELECT * FROM clin.v_linked_codes WHERE
                                        item_table = 'clin.lnk_code2h_issue'::regclass
                                                AND
                                        pk_item = %(item)s
                        """
                        args = {'item': self._payload['pk_health_issue']}
                else:
                        cmd = """
                                SELECT * FROM clin.v_linked_codes WHERE
                                        item_table = 'clin.lnk_code2episode'::regclass
                                                AND
                                        pk_item = %(item)s
                        """
                        args = {'item': self._payload['pk_episode']}

                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        generic_codes = property(_get_generic_codes)

#-----------------------------------------------------------
#-----------------------------------------------------------
#-----------------------------------------------------------
def reclass_problem(problem=None):
        """Transform given problem into either episode or health issue instance.
        """
        if isinstance(problem, (cEpisode, cHealthIssue)):
                return problem

        exc = TypeError('cannot reclass [%s] instance to either episode or health issue' % type(problem))

        if not isinstance(problem, cProblem):
                _log.debug('%s' % problem)
                raise exc

        if problem['type'] == 'episode':
                return cEpisode(aPK_obj = problem['pk_episode'])

        if problem['type'] == 'issue':
                return cHealthIssue(aPK_obj = problem['pk_health_issue'])

        raise exc

#============================================================
_SQL_get_hospital_stays = "select * from clin.v_hospital_stays where %s"

class cHospitalStay(gmBusinessDBObject.cBusinessDBObject):

        _cmd_fetch_payload = _SQL_get_hospital_stays % "pk_hospital_stay = %s"
        _cmds_store_payload = [
                """UPDATE clin.hospital_stay SET
                                clin_when = %(admission)s,
                                discharge = %(discharge)s,
                                fk_org_unit = %(pk_org_unit)s,
                                narrative = gm.nullify_empty_string(%(comment)s),
                                fk_episode = %(pk_episode)s,
                                fk_encounter = %(pk_encounter)s
                        WHERE
                                pk = %(pk_hospital_stay)s
                                        AND
                                xmin = %(xmin_hospital_stay)s
                        RETURNING
                                xmin AS xmin_hospital_stay
                        """
        ]
        _updatable_fields = [
                'admission',
                'discharge',
                'pk_org_unit',
                'pk_episode',
                'pk_encounter',
                'comment'
        ]

        #--------------------------------------------------------
        def format_maximum_information(self, patient=None):
                return self.format (
                        include_procedures = True,
                        include_docs = True
                ).split('\n')

        #-------------------------------------------------------
        def format(self, left_margin=0, include_procedures=False, include_docs=False, include_episode=True):

                if self._payload['discharge'] is not None:
                        discharge = ' - %s' % gmDateTime.pydt_strftime(self._payload['discharge'], '%Y %b %d')
                else:
                        discharge = ''

                episode = ''
                if include_episode:
                        episode = ': %s%s%s' % (
                                gmTools.u_left_double_angle_quote,
                                self._payload['episode'],
                                gmTools.u_right_double_angle_quote
                        )

                lines = ['%s%s%s (%s@%s)%s' % (
                        ' ' * left_margin,
                        gmDateTime.pydt_strftime(self._payload['admission'], '%Y %b %d'),
                        discharge,
                        self._payload['ward'],
                        self._payload['hospital'],
                        episode
                )]

                if include_docs:
                        for doc in self.documents:
                                lines.append('%s%s: %s\n' % (
                                        ' ' * left_margin,
                                        _('Document'),
                                        doc.format(single_line = True)
                                ))

                return '\n'.join(lines)

        #--------------------------------------------------------
        def _get_documents(self):
                return [ gmDocuments.cDocument(aPK_obj = pk_doc) for pk_doc in  self._payload['pk_documents'] ]

        documents = property(_get_documents)

#-----------------------------------------------------------
def get_latest_patient_hospital_stay(patient=None):
        cmd = _SQL_get_hospital_stays % "pk_patient = %(pat)s ORDER BY admission DESC LIMIT 1"
        queries = [{
                # this assumes non-overarching stays
                #'cmd': u'SELECT * FROM clin.v_hospital_stays WHERE pk_patient = %(pat)s ORDER BY admission DESC LIMIT 1',
                'cmd': cmd,
                'args': {'pat': patient}
        }]
        rows = gmPG2.run_ro_queries(queries = queries)
        if len(rows) == 0:
                return None
        return cHospitalStay(row = {'data': rows[0], 'pk_field': 'pk_hospital_stay'})

#-----------------------------------------------------------
def get_patient_hospital_stays(patient=None, ongoing_only=False, return_pks=False):
        args = {'pat': patient}
        if ongoing_only:
                cmd = _SQL_get_hospital_stays % "pk_patient = %(pat)s AND discharge is NULL ORDER BY admission"
        else:
                cmd = _SQL_get_hospital_stays % "pk_patient = %(pat)s ORDER BY admission"

        queries = [{'cmd': cmd, 'args': args}]
        rows = gmPG2.run_ro_queries(queries = queries)
        if return_pks:
                return [ r['pk_hospital_stay'] for r in rows ]
        return [ cHospitalStay(row = {'data': r, 'pk_field': 'pk_hospital_stay'})  for r in rows ]

#-----------------------------------------------------------
def create_hospital_stay(encounter=None, episode=None, fk_org_unit=None):

        queries = [{
                 'cmd': 'INSERT INTO clin.hospital_stay (fk_encounter, fk_episode, fk_org_unit) VALUES (%(enc)s, %(epi)s, %(fk_org_unit)s) RETURNING pk',
                 'args': {'enc': encounter, 'epi': episode, 'fk_org_unit': fk_org_unit}
        }]
        rows = gmPG2.run_rw_queries(queries = queries, return_data = True)

        return cHospitalStay(aPK_obj = rows[0][0])

#-----------------------------------------------------------
def delete_hospital_stay(stay=None):
        cmd = 'DELETE FROM clin.hospital_stay WHERE pk = %(pk)s'
        args = {'pk': stay}
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True

#============================================================
_SQL_get_procedures = "select * from clin.v_procedures where %s"

class cPerformedProcedure(gmBusinessDBObject.cBusinessDBObject):

        _cmd_fetch_payload = _SQL_get_procedures % "pk_procedure = %s"
        _cmds_store_payload = [
                """UPDATE clin.procedure SET
                                soap_cat = 'p',
                                clin_when = %(clin_when)s,
                                clin_end = %(clin_end)s,
                                is_ongoing = %(is_ongoing)s,
                                narrative = gm.nullify_empty_string(%(performed_procedure)s),
                                fk_hospital_stay = %(pk_hospital_stay)s,
                                fk_org_unit = (CASE
                                        WHEN %(pk_hospital_stay)s IS NULL THEN %(pk_org_unit)s
                                        ELSE NULL
                                END)::integer,
                                fk_episode = %(pk_episode)s,
                                fk_encounter = %(pk_encounter)s,
                                fk_doc = %(pk_doc)s,
                                comment = gm.nullify_empty_string(%(comment)s)
                        WHERE
                                pk = %(pk_procedure)s AND
                                xmin = %(xmin_procedure)s
                        RETURNING xmin as xmin_procedure"""
        ]
        _updatable_fields = [
                'clin_when',
                'clin_end',
                'is_ongoing',
                'performed_procedure',
                'pk_hospital_stay',
                'pk_org_unit',
                'pk_episode',
                'pk_encounter',
                'pk_doc',
                'comment'
        ]
        #-------------------------------------------------------
        def __setitem__(self, attribute, value):

                if (attribute == 'pk_hospital_stay') and (value is not None):
                        gmBusinessDBObject.cBusinessDBObject.__setitem__(self, 'pk_org_unit', None)

                if (attribute == 'pk_org_unit') and (value is not None):
                        gmBusinessDBObject.cBusinessDBObject.__setitem__(self, 'pk_hospital_stay', None)

                gmBusinessDBObject.cBusinessDBObject.__setitem__(self, attribute, value)

        #--------------------------------------------------------
        def format_maximum_information(self, left_margin=0, patient=None):
                return self.format (
                        left_margin = left_margin,
                        include_episode = True,
                        include_codes = True,
                        include_address = True,
                        include_comm = True,
                        include_doc = True
                ).split('\n')

        #-------------------------------------------------------
        def format(self, left_margin=0, include_episode=True, include_codes=False, include_address=False, include_comm=False, include_doc=False):

                if self._payload['is_ongoing']:
                        end = _(' (ongoing)')
                else:
                        end = self._payload['clin_end']
                        if end is None:
                                end = ''
                        else:
                                end = ' - %s' % gmDateTime.pydt_strftime(end, '%Y %b %d')

                line = '%s%s%s: %s%s [%s @ %s]' % (
                        (' ' * left_margin),
                        gmDateTime.pydt_strftime(self._payload['clin_when'], '%Y %b %d'),
                        end,
                        self._payload['performed_procedure'],
                        gmTools.bool2str(include_episode, ' (%s)' % self._payload['episode'], ''),
                        self._payload['unit'],
                        self._payload['organization']
                )

                line += gmTools.coalesce(self._payload['comment'], '', '\n' + (' ' * left_margin) + _('Comment: ') + '%s')

                if include_comm:
                        for channel in self.org_unit.comm_channels:
                                line += ('\n%(comm_type)s: %(url)s' % channel)

                if include_address:
                        adr = self.org_unit.address
                        if adr is not None:
                                line += '\n'
                                line += '\n'.join(adr.format(single_line = False, show_type = False))
                                line += '\n'

                if include_doc:
                        doc = self.doc
                        if doc is not None:
                                line += '\n'
                                line += _('Document') + ': ' + doc.format(single_line = True)
                                line += '\n'

                if include_codes:
                        codes = self.generic_codes
                        if len(codes) > 0:
                                line += '\n'
                        for c in codes:
                                line += '%s  %s: %s (%s - %s)\n' % (
                                        (' ' * left_margin),
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                )
                        del codes

                return line

        #--------------------------------------------------------
        def add_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "INSERT INTO clin.lnk_code2procedure (fk_item, fk_generic_code) values (%(issue)s, %(code)s)"
                args = {
                        'issue': self._payload['pk_procedure'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def remove_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "DELETE FROM clin.lnk_code2procedure WHERE fk_item = %(issue)s AND fk_generic_code = %(code)s"
                args = {
                        'issue': self._payload['pk_procedure'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_stay(self):
                if self._payload['pk_hospital_stay'] is None:
                        return None
                return cHospitalStay(aPK_obj = self._payload['pk_hospital_stay'])

        hospital_stay = property(_get_stay)

        #--------------------------------------------------------
        def _get_org_unit(self):
                return gmOrganization.cOrgUnit(self._payload['pk_org_unit'])

        org_unit = property(_get_org_unit)

        #--------------------------------------------------------
        def _get_doc(self):
                if self._payload['pk_doc'] is None:
                        return None
                return gmDocuments.cDocument(aPK_obj = self._payload['pk_doc'])

        doc = property(_get_doc)

        #--------------------------------------------------------
        def _get_generic_codes(self):
                if len(self._payload['pk_generic_codes']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2procedure WHERE fk_item = %(proc)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'proc': self._payload['pk_procedure'],
                                        'codes': self._payload['pk_generic_codes']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2procedure (fk_item, fk_generic_code) VALUES (%(proc)s, %(pk_code)s)',
                                'args': {
                                        'proc': self._payload['pk_procedure'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                return

        generic_codes = property(_get_generic_codes, _set_generic_codes)

#-----------------------------------------------------------
def get_performed_procedures(patient=None, return_pks=False):
        queries = [{
                'cmd': 'SELECT * FROM clin.v_procedures WHERE pk_patient = %(pat)s ORDER BY clin_when',
                'args': {'pat': patient}
        }]
        rows = gmPG2.run_ro_queries(queries = queries)
        if return_pks:
                return [ r['pk_procedure'] for r in rows ]
        return [ cPerformedProcedure(row = {'data': r, 'pk_field': 'pk_procedure'})  for r in rows ]

#-----------------------------------------------------------
def get_procedures4document(pk_document=None, return_pks=False):
        args = {'pk_doc': pk_document}
        cmd = _SQL_get_procedures % 'pk_doc = %(pk_doc)s'
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        if return_pks:
                return [ r['pk_procedure'] for r in rows ]
        return [ cPerformedProcedure(row = {'data': r, 'pk_field': 'pk_procedure'})  for r in rows ]

#-----------------------------------------------------------
def get_latest_performed_procedure(patient=None):
        queries = [{
                'cmd': 'select * FROM clin.v_procedures WHERE pk_patient = %(pat)s ORDER BY clin_when DESC LIMIT 1',
                'args': {'pat': patient}
        }]
        rows = gmPG2.run_ro_queries(queries = queries)
        if len(rows) == 0:
                return None
        return cPerformedProcedure(row = {'data': rows[0], 'pk_field': 'pk_procedure'})

#-----------------------------------------------------------
def create_performed_procedure(encounter=None, episode=None, location=None, hospital_stay=None, procedure=None):

        queries = [{
                'cmd': """
                        INSERT INTO clin.procedure (
                                fk_encounter,
                                fk_episode,
                                soap_cat,
                                fk_org_unit,
                                fk_hospital_stay,
                                narrative
                        ) VALUES (
                                %(enc)s,
                                %(epi)s,
                                'p',
                                %(loc)s,
                                %(stay)s,
                                gm.nullify_empty_string(%(proc)s)
                        )
                        RETURNING pk""",
                'args': {'enc': encounter, 'epi': episode, 'loc': location, 'stay': hospital_stay, 'proc': procedure}
        }]

        rows = gmPG2.run_rw_queries(queries = queries, return_data = True)

        return cPerformedProcedure(aPK_obj = rows[0][0])

#-----------------------------------------------------------
def delete_performed_procedure(procedure=None):
        cmd = 'delete from clin.procedure where pk = %(pk)s'
        args = {'pk': procedure}
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True

#============================================================
def export_emr_structure(patient=None, filename=None):

        if filename is None:
                filename = gmTools.get_unique_filename(prefix = 'gm-emr_struct-%s-' % patient.subdir_name, suffix = '.txt')

        f = open(filename, 'w+', encoding = 'utf8')

        f.write('patient [%s]\n' % patient.description_gender)
        emr = patient.emr
        for issue in emr.health_issues:
                f.write('\n')
                f.write('\n')
                f.write('issue [%s] #%s\n' % (issue['description'], issue['pk_health_issue']))
                f.write(' is active     : %s\n' % issue['is_active'])
                f.write(' has open epi  : %s\n' % issue['has_open_episode'])
                f.write(' possible start: %s\n' % issue.possible_start_date)
                f.write(' safe start    : %s\n' % issue.safe_start_date)
                end = issue.clinical_end_date
                if end is None:
                        f.write(' end           : active and/or open episode\n')
                else:
                        f.write(' end           : %s\n' % end)
                f.write(' latest access : %s\n' % issue.latest_access_date)
                first = issue.first_episode
                if first is not None:
                        first = first['description']
                f.write(' 1st episode   : %s\n' % first)
                last = issue.latest_episode
                if last is not None:
                        last = last['description']
                f.write(' latest episode: %s\n' % last)
                epis = sorted(issue.get_episodes(), key = lambda e: e.best_guess_clinical_start_date)
                for epi in epis:
                        f.write('\n')
                        f.write(' episode [%s] #%s\n' % (epi['description'], epi['pk_episode']))
                        f.write('  is open         : %s\n' % epi['episode_open'])
                        f.write('  best guess start: %s\n' % epi.best_guess_clinical_start_date)
                        f.write('  best guess end  : %s\n' % epi.best_guess_clinical_end_date)
                        f.write('  latest access   : %s\n' % epi.latest_access_date)
                        f.write('  start 1st enc   : %s\n' % epi['started_first'])
                        f.write('  start last enc  : %s\n' % epi['started_last'])
                        f.write('  end last enc    : %s\n' % epi['last_affirmed'])

        f.close()
        return filename

#============================================================
# tools
#------------------------------------------------------------
def check_fk_encounter_fk_episode_x_ref():

        aggregate_result = 0

        fks_linking2enc = gmPG2.get_foreign_keys2column(schema = 'clin', table = 'encounter', column = 'pk')
        tables_linking2enc = set([ r['referencing_table'] for r in fks_linking2enc ])

        fks_linking2epi = gmPG2.get_foreign_keys2column(schema = 'clin', table = 'episode', column = 'pk')
        tables_linking2epi = [ r['referencing_table'] for r in fks_linking2epi ]

        tables_linking2both = tables_linking2enc.intersection(tables_linking2epi)

        tables_linking2enc = {}
        for fk in fks_linking2enc:
                table = fk['referencing_table']
                tables_linking2enc[table] = fk

        tables_linking2epi = {}
        for fk in fks_linking2epi:
                table = fk['referencing_table']
                tables_linking2epi[table] = fk

        for t in tables_linking2both:

                table_file_name = 'x-check_enc_epi_xref-%s.log' % t
                table_file = open(table_file_name, 'w+', encoding = 'utf8')

                # get PK column
                args = {'table': t}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': gmPG2.SQL_get_pk_col_def, 'args': args}])
                pk_col = rows[0][0]
                print("checking table:", t, '- pk col:', pk_col)
                print(' =>', table_file_name)
                table_file.write('table: %s\n' % t)
                table_file.write('PK col: %s\n' % pk_col)

                # get PKs
                cmd = 'select %s from %s' % (pk_col, t)
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd}])
                pks = [ r[0] for r in rows ]
                for pk in pks:
                        args = {'pk': pk, 'tbl': t}
                        enc_cmd = "select fk_patient from clin.encounter where pk = (select fk_encounter from %s where %s = %%(pk)s)" % (t, pk_col)
                        epi_cmd = "select fk_patient from clin.encounter where pk = (select fk_encounter from clin.episode where pk = (select fk_episode from %s where %s = %%(pk)s))" % (t, pk_col)
                        enc_rows = gmPG2.run_ro_queries(queries = [{'cmd': enc_cmd, 'args': args}])
                        epi_rows = gmPG2.run_ro_queries(queries = [{'cmd': epi_cmd, 'args': args}])
                        enc_pat = enc_rows[0][0]
                        epi_pat = epi_rows[0][0]
                        args['pat_enc'] = enc_pat
                        args['pat_epi'] = epi_pat
                        if epi_pat != enc_pat:
                                print(' mismatch: row pk=%s, enc pat=%s, epi pat=%s' % (pk, enc_pat, epi_pat))
                                aggregate_result = -2

                                table_file.write('--------------------------------------------------------------------------------\n')
                                table_file.write('mismatch on row with pk: %s\n' % pk)
                                table_file.write('\n')

                                table_file.write('journal entry:\n')
                                cmd = 'SELECT * from clin.v_emr_journal where src_table = %(tbl)s AND src_pk = %(pk)s'
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                if len(rows) > 0:
                                        table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('row data:\n')
                                cmd = 'SELECT * from %s where %s = %%(pk)s' % (t, pk_col)
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('episode:\n')
                                cmd = 'SELECT * from clin.v_pat_episodes WHERE pk_episode = (select fk_episode from %s where %s = %%(pk)s)' % (t, pk_col)
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('patient of episode:\n')
                                cmd = 'SELECT * FROM dem.v_persons WHERE pk_identity = %(pat_epi)s'
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('encounter:\n')
                                cmd = 'SELECT * from clin.v_pat_encounters WHERE pk_encounter = (select fk_encounter from %s where %s = %%(pk)s)' % (t, pk_col)
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('patient of encounter:\n')
                                cmd = 'SELECT * FROM dem.v_persons WHERE pk_identity = %(pat_enc)s'
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n')

                table_file.write('done\n')
                table_file.close()

        return aggregate_result

#------------------------------------------------------------
def export_patient_emr_structure():
        from Gnumed.business import gmPersonSearch
        gmPraxis.gmCurrentPraxisBranch(branch = gmPraxis.get_praxis_branches()[0])
        pat = gmPersonSearch.ask_for_patient()
        while pat is not None:
                print('patient:', pat.description_gender)
                fname = os.path.expanduser('~/gnumed/gm-emr_structure-%s.txt' % pat.subdir_name)
                print('exported into:', export_emr_structure(patient = pat, filename = fname))
                pat = gmPersonSearch.ask_for_patient()

        return 0

#============================================================
# main - unit testing
#------------------------------------------------------------
if __name__ == '__main__':

        if len(sys.argv) < 2:
                sys.exit()

        if sys.argv[1] != 'test':
                sys.exit()

        gmI18N.activate_locale()
        gmI18N.install_domain('gnumed')
        #--------------------------------------------------------
        # define tests
        #--------------------------------------------------------
        def test_problem():
                print("\nProblem test")
                print("------------")
                prob = cProblem(aPK_obj={'pk_patient': 12, 'pk_health_issue': 1, 'pk_episode': None})
                print(prob)
                fields = prob.get_fields()
                for field in fields:
                        print(field, ':', prob[field])
                print('\nupdatable:', prob.get_updatable_fields())
                epi = prob.get_as_episode()
                print('\nas episode:')
                if epi is not None:
                        for field in epi.get_fields():
                                print('   .%s : %s' % (field, epi[field]))

        #--------------------------------------------------------
        def test_health_issue():
                print("\nhealth issue test")
                print("-----------------")
                #h_issue = cHealthIssue(aPK_obj = 894)
                h_issue = cHealthIssue(aPK_obj = 1)
                print(h_issue)
#               print('possible start:', h_issue.possible_start_date)
#               print('safe start    :', h_issue.safe_start_date)
#               print('end date      :', h_issue.clinical_end_date)

                #print(h_issue.latest_access_date)
#               fields = h_issue.get_fields()
#               for field in fields:
#                       print field, ':', h_issue[field]
#               print "has open episode:", h_issue.has_open_episode()
#               print "open episode:", h_issue.get_open_episode()
#               print "updateable:", h_issue.get_updatable_fields()
#               h_issue.close_expired_episode(ttl=7300)
#               h_issue = cHealthIssue(encounter = 1, name = u'post appendectomy/peritonitis')
#               print h_issue
#               print h_issue.format_as_journal()
                print(h_issue.formatted_revision_history)

        #--------------------------------------------------------       
        def test_episode():
                print("episode test")
                for i in range(1,15):
                        print("------------")
                        episode = cEpisode(aPK_obj = i)#322) #1674) #1354) #1461) #1299)

                        print(episode['description'])
                        print(' start:', episode.best_guess_clinical_start_date)
                        print(' end  :', episode.best_guess_clinical_end_date)
                        #print(' dura :', get_formatted_clinical_duration(pk_episode = i))
                return

                print(episode)
                fields = episode.get_fields()
                for field in fields:
                        print(field, ':', episode[field])
                print("updatable:", episode.get_updatable_fields())
                input('ENTER to continue')

                desc = '1-%s' % episode['description']
                print("==> renaming to", desc)
                successful = episode.rename (
                        description = desc
                )
                if not successful:
                        print("error")
                else:
                        print("success")
                        for field in fields:
                                print(field, ':', episode[field])

                print(episode.formatted_revision_history)

                input('ENTER to continue')

        #--------------------------------------------------------
        def test_encounter():
                print("\nencounter test")
                print("--------------")
                encounter = cEncounter(aPK_obj=1)
                print(encounter)
                fields = encounter.get_fields()
                for field in fields:
                        print(field, ':', encounter[field])
                print("updatable:", encounter.get_updatable_fields())
                #print encounter.formatted_revision_history
                print(encounter.transfer_all_data_to_another_encounter(pk_target_encounter = 2))

        #--------------------------------------------------------
        def test_encounter2latex():
                encounter = cEncounter(aPK_obj=1)
                print(encounter)
                print("")
                print(encounter.format_latex())
        #--------------------------------------------------------
        def test_performed_procedure():
                procs = get_performed_procedures(patient = 12)
                for proc in procs:
                        print(proc.format(left_margin=2))
        #--------------------------------------------------------
        def test_hospital_stay():
                stay = create_hospital_stay(encounter = 1, episode = 2, fk_org_unit = 1)
#               stay['hospital'] = u'Starfleet Galaxy General Hospital'
#               stay.save_payload()
                print(stay)
                for s in get_patient_hospital_stays(12):
                        print(s)
                delete_hospital_stay(stay['pk_hospital_stay'])
                stay = create_hospital_stay(encounter = 1, episode = 4, fk_org_unit = 1)
        #--------------------------------------------------------
        def test_diagnostic_certainty_classification_map():
                tests = [None, 'A', 'B', 'C', 'D', 'E']

                for t in tests:
                        print(type(t), t)
                        print(type(diagnostic_certainty_classification2str(t)), diagnostic_certainty_classification2str(t))
        #--------------------------------------------------------
        def test_episode_codes():
                epi = cEpisode(aPK_obj = 2)
                print(epi)
                print(epi.generic_codes)
        #--------------------------------------------------------
        def test_episode_encounters():
                epi = cEpisode(aPK_obj = 1638)
                print(epi.format())

        #--------------------------------------------------------
        def test_export_emr_structure():
                export_patient_emr_structure()
                #praxis = gmPraxis.gmCurrentPraxisBranch(branch = gmPraxis.get_praxis_branches()[0])
                #from Gnumed.business import gmPerson
                ## 12 / 20 / 138 / 58 / 20 / 5 / 14
                #pat = gmPerson.gmCurrentPatient(gmPerson.cPatient(aPK_obj = 138))
                #fname = os.path.expanduser(u'~/gnumed/emr_structure-%s.txt' % pat.subdir_name)
                #print export_emr_structure(patient = pat, filename = fname)

        #--------------------------------------------------------
        def test_cEpisodeMatchProvider():
                mp = cEpisodeMatchProvider()
                print(mp._find_matches('no patient'))

        #--------------------------------------------------------
        # run them
        #test_episode()
        #test_episode_encounters()
        #test_problem()
        #test_encounter()
        #test_health_issue()
        #test_hospital_stay()
        #test_performed_procedure()
        #test_diagnostic_certainty_classification_map()
        #test_encounter2latex()
        #test_episode_codes()

        #test_export_emr_structure()

        test_cEpisodeMatchProvider()

#============================================================

Functions

def check_fk_encounter_fk_episode_x_ref()
Expand source code
def check_fk_encounter_fk_episode_x_ref():

        aggregate_result = 0

        fks_linking2enc = gmPG2.get_foreign_keys2column(schema = 'clin', table = 'encounter', column = 'pk')
        tables_linking2enc = set([ r['referencing_table'] for r in fks_linking2enc ])

        fks_linking2epi = gmPG2.get_foreign_keys2column(schema = 'clin', table = 'episode', column = 'pk')
        tables_linking2epi = [ r['referencing_table'] for r in fks_linking2epi ]

        tables_linking2both = tables_linking2enc.intersection(tables_linking2epi)

        tables_linking2enc = {}
        for fk in fks_linking2enc:
                table = fk['referencing_table']
                tables_linking2enc[table] = fk

        tables_linking2epi = {}
        for fk in fks_linking2epi:
                table = fk['referencing_table']
                tables_linking2epi[table] = fk

        for t in tables_linking2both:

                table_file_name = 'x-check_enc_epi_xref-%s.log' % t
                table_file = open(table_file_name, 'w+', encoding = 'utf8')

                # get PK column
                args = {'table': t}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': gmPG2.SQL_get_pk_col_def, 'args': args}])
                pk_col = rows[0][0]
                print("checking table:", t, '- pk col:', pk_col)
                print(' =>', table_file_name)
                table_file.write('table: %s\n' % t)
                table_file.write('PK col: %s\n' % pk_col)

                # get PKs
                cmd = 'select %s from %s' % (pk_col, t)
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd}])
                pks = [ r[0] for r in rows ]
                for pk in pks:
                        args = {'pk': pk, 'tbl': t}
                        enc_cmd = "select fk_patient from clin.encounter where pk = (select fk_encounter from %s where %s = %%(pk)s)" % (t, pk_col)
                        epi_cmd = "select fk_patient from clin.encounter where pk = (select fk_encounter from clin.episode where pk = (select fk_episode from %s where %s = %%(pk)s))" % (t, pk_col)
                        enc_rows = gmPG2.run_ro_queries(queries = [{'cmd': enc_cmd, 'args': args}])
                        epi_rows = gmPG2.run_ro_queries(queries = [{'cmd': epi_cmd, 'args': args}])
                        enc_pat = enc_rows[0][0]
                        epi_pat = epi_rows[0][0]
                        args['pat_enc'] = enc_pat
                        args['pat_epi'] = epi_pat
                        if epi_pat != enc_pat:
                                print(' mismatch: row pk=%s, enc pat=%s, epi pat=%s' % (pk, enc_pat, epi_pat))
                                aggregate_result = -2

                                table_file.write('--------------------------------------------------------------------------------\n')
                                table_file.write('mismatch on row with pk: %s\n' % pk)
                                table_file.write('\n')

                                table_file.write('journal entry:\n')
                                cmd = 'SELECT * from clin.v_emr_journal where src_table = %(tbl)s AND src_pk = %(pk)s'
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                if len(rows) > 0:
                                        table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('row data:\n')
                                cmd = 'SELECT * from %s where %s = %%(pk)s' % (t, pk_col)
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('episode:\n')
                                cmd = 'SELECT * from clin.v_pat_episodes WHERE pk_episode = (select fk_episode from %s where %s = %%(pk)s)' % (t, pk_col)
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('patient of episode:\n')
                                cmd = 'SELECT * FROM dem.v_persons WHERE pk_identity = %(pat_epi)s'
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('encounter:\n')
                                cmd = 'SELECT * from clin.v_pat_encounters WHERE pk_encounter = (select fk_encounter from %s where %s = %%(pk)s)' % (t, pk_col)
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n\n')

                                table_file.write('patient of encounter:\n')
                                cmd = 'SELECT * FROM dem.v_persons WHERE pk_identity = %(pat_enc)s'
                                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                                table_file.write(gmTools.format_dict_like(rows[0], left_margin = 1, tabular = False, value_delimiters = None))
                                table_file.write('\n')

                table_file.write('done\n')
                table_file.close()

        return aggregate_result
def create_encounter(fk_patient=None, enc_type=None)

Creates a new encounter for a patient.

fk_patient - patient PK enc_type - type of encounter

Expand source code
def create_encounter(fk_patient=None, enc_type=None):
        """Creates a new encounter for a patient.

        fk_patient - patient PK
        enc_type - type of encounter
        """
        if enc_type is None:
                enc_type = 'in surgery'
        # insert new encounter
        queries = []
        try:
                enc_type = int(enc_type)
                cmd = """
                        INSERT INTO clin.encounter (fk_patient, fk_type, fk_location)
                        VALUES (%(pat)s, %(typ)s, %(prax)s) RETURNING pk"""
        except ValueError:
                enc_type = enc_type
                cmd = """
                        INSERT INTO clin.encounter (fk_patient, fk_location, fk_type)
                        VALUES (
                                %(pat)s,
                                %(prax)s,
                                coalesce (
                                        (select pk from clin.encounter_type where description = %(typ)s),
                                        -- pick the first available
                                        (select pk from clin.encounter_type limit 1)
                                )
                        ) RETURNING pk"""
        praxis = gmPraxis.gmCurrentPraxisBranch()
        args = {'pat': fk_patient, 'typ': enc_type, 'prax': praxis['pk_org_unit']}
        queries.append({'cmd': cmd, 'args': args})
        rows = gmPG2.run_rw_queries(queries = queries, return_data = True)
        encounter = cEncounter(aPK_obj = rows[0]['pk'])

        return encounter
def create_encounter_type(description=None, l10n_description=None)

This will attempt to create a NEW encounter type.

Expand source code
def create_encounter_type(description=None, l10n_description=None):
        """This will attempt to create a NEW encounter type."""

        # need a system name, so derive one if necessary
        if description is None:
                description = l10n_description

        args = {
                'desc': description,
                'l10n_desc': l10n_description
        }

        _log.debug('creating encounter type: %s, %s', description, l10n_description)

        # does it exist already ?
        cmd = "select description, _(description) from clin.encounter_type where description = %(desc)s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])

        # yes
        if len(rows) > 0:
                # both system and l10n name are the same so all is well
                if (rows[0][0] == description) and (rows[0][1] == l10n_description):
                        _log.info('encounter type [%s] already exists with the proper translation')
                        return {'description': description, 'l10n_description': l10n_description}

                # or maybe there just wasn't a translation to
                # the current language for this type yet ?
                cmd = "select exists (select 1 from i18n.translations where orig = %(desc)s and lang = i18n.get_curr_lang())"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])

                # there was, so fail
                if rows[0][0]:
                        _log.error('encounter type [%s] already exists but with another translation')
                        return None

                # else set it
                cmd = "select i18n.upd_tx(%(desc)s, %(l10n_desc)s)"
                rows = gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return {'description': description, 'l10n_description': l10n_description}

        # no
        queries = [
                {'cmd': "insert into clin.encounter_type (description) values (%(desc)s)", 'args': args},
                {'cmd': "select i18n.upd_tx(%(desc)s, %(l10n_desc)s)", 'args': args}
        ]
        rows = gmPG2.run_rw_queries(queries = queries)

        return {'description': description, 'l10n_description': l10n_description}
def create_episode(pk_health_issue=None, episode_name=None, is_open=False, allow_dupes=False, encounter=None, link_obj=None)

Creates a new episode for a given patient's health issue.

pk_health_issue - given health issue PK episode_name - name of episode is_open - whether a new episode is to be open (don't tinker with existing ones)

Expand source code
def create_episode(pk_health_issue=None, episode_name=None, is_open=False, allow_dupes=False, encounter=None, link_obj=None):
        """Creates a new episode for a given patient's health issue.

        pk_health_issue - given health issue PK
        episode_name - name of episode
        is_open - whether a *new* episode is to be open (don't tinker with existing ones)
        """
        if not allow_dupes:
                try:
                        return cEpisode(name = episode_name, health_issue = pk_health_issue, encounter = encounter, link_obj = link_obj)

                except gmExceptions.ConstructorError:
                        pass
        queries = []
        cmd = "INSERT INTO clin.episode (fk_health_issue, description, is_open, fk_encounter) VALUES (%s, %s, %s::boolean, %s)"
        queries.append({'cmd': cmd, 'args': [pk_health_issue, episode_name, is_open, encounter]})
        queries.append({'cmd': cEpisode._cmd_fetch_payload % "currval('clin.episode_pk_seq')"})
        rows = gmPG2.run_rw_queries(link_obj = link_obj, queries = queries, return_data=True)
        return cEpisode(row = {'data': rows[0], 'pk_field': 'pk_episode'})
def create_health_issue(description=None, encounter=None, patient=None, link_obj=None)

Creates a new health issue for a given patient.

description - health issue name

Expand source code
def create_health_issue(description=None, encounter=None, patient=None, link_obj=None):
        """Creates a new health issue for a given patient.

        description - health issue name
        """
        try:
                h_issue = cHealthIssue(name = description, encounter = encounter, patient = patient)
                return h_issue

        except gmExceptions.NoSuchBusinessObjectError:
                pass

        queries = []
        cmd = "insert into clin.health_issue (description, fk_encounter) values (%(desc)s, %(enc)s)"
        queries.append({'cmd': cmd, 'args': {'desc': description, 'enc': encounter}})
        cmd = "select currval('clin.health_issue_pk_seq')"
        queries.append({'cmd': cmd})
        rows = gmPG2.run_rw_queries(queries = queries, return_data = True, link_obj = link_obj)
        h_issue = cHealthIssue(aPK_obj = rows[0][0])
        return h_issue
def create_hospital_stay(encounter=None, episode=None, fk_org_unit=None)
Expand source code
def create_hospital_stay(encounter=None, episode=None, fk_org_unit=None):

        queries = [{
                 'cmd': 'INSERT INTO clin.hospital_stay (fk_encounter, fk_episode, fk_org_unit) VALUES (%(enc)s, %(epi)s, %(fk_org_unit)s) RETURNING pk',
                 'args': {'enc': encounter, 'epi': episode, 'fk_org_unit': fk_org_unit}
        }]
        rows = gmPG2.run_rw_queries(queries = queries, return_data = True)

        return cHospitalStay(aPK_obj = rows[0][0])
def create_performed_procedure(encounter=None, episode=None, location=None, hospital_stay=None, procedure=None)
Expand source code
def create_performed_procedure(encounter=None, episode=None, location=None, hospital_stay=None, procedure=None):

        queries = [{
                'cmd': """
                        INSERT INTO clin.procedure (
                                fk_encounter,
                                fk_episode,
                                soap_cat,
                                fk_org_unit,
                                fk_hospital_stay,
                                narrative
                        ) VALUES (
                                %(enc)s,
                                %(epi)s,
                                'p',
                                %(loc)s,
                                %(stay)s,
                                gm.nullify_empty_string(%(proc)s)
                        )
                        RETURNING pk""",
                'args': {'enc': encounter, 'epi': episode, 'loc': location, 'stay': hospital_stay, 'proc': procedure}
        }]

        rows = gmPG2.run_rw_queries(queries = queries, return_data = True)

        return cPerformedProcedure(aPK_obj = rows[0][0])
def delete_encounter(pk_encounter)

Deletes an encounter by PK.

  • attempts to obtain an exclusive lock which should fail if the encounter is the active encounter in this or any other client
  • catches DB exceptions which should mostly be related to clinical data already having been attached to the encounter thus making deletion fail
Expand source code
def delete_encounter(pk_encounter):
        """Deletes an encounter by PK.

        - attempts to obtain an exclusive lock which should
          fail if the encounter is the active encounter in
          this or any other client
        - catches DB exceptions which should mostly be related
          to clinical data already having been attached to
          the encounter thus making deletion fail
        """
        conn = gmPG2.get_connection(readonly = False)
        if not lock_encounter(pk_encounter, exclusive = True, link_obj = conn):
                _log.debug('cannot lock encounter [%s] for deletion, it seems in use', pk_encounter)
                return False
        cmd = """DELETE FROM clin.encounter WHERE pk = %(enc)s"""
        args = {'enc': pk_encounter}
        try:
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        except gmPG2.PG_ERROR_EXCEPTION as exc:
                _log.exception('cannot delete encounter [%s]', pk_encounter)
                gmPG2.log_pg_exception_details(exc)
                unlock_encounter(pk_encounter, exclusive = True, link_obj = conn)
                return False
        unlock_encounter(pk_encounter, exclusive = True, link_obj = conn)
        return True
def delete_encounter_type(description=None)
Expand source code
def delete_encounter_type(description=None):
        deleted = False
        cmd = "delete from clin.encounter_type where description = %(desc)s"
        args = {'desc': description}
        try:
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                deleted = True
        except gmPG2.dbapi.IntegrityError as e:
                if e.pgcode != gmPG2.PG_error_codes.FOREIGN_KEY_VIOLATION:
                        raise

        return deleted
def delete_episode(episode=None)
Expand source code
def delete_episode(episode=None):
        if isinstance(episode, cEpisode):
                pk = episode['pk_episode']
        else:
                pk = int(episode)

        cmd = 'DELETE FROM clin.episode WHERE pk = %(pk)s'

        try:
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': {'pk': pk}}])
        except gmPG2.dbapi.IntegrityError:
                # should be parsing pgcode/and or error message
                _log.exception('cannot delete episode, it is in use')
                return False

        return True
def delete_health_issue(health_issue=None)
Expand source code
def delete_health_issue(health_issue=None):
        if isinstance(health_issue, cHealthIssue):
                args = {'pk': health_issue['pk_health_issue']}
        else:
                args = {'pk': int(health_issue)}
        try:
                gmPG2.run_rw_queries(queries = [{'cmd': 'DELETE FROM clin.health_issue WHERE pk = %(pk)s', 'args': args}])
        except gmPG2.dbapi.IntegrityError:
                # should be parsing pgcode/and or error message
                _log.exception('cannot delete health issue')
                return False

        return True
def delete_hospital_stay(stay=None)
Expand source code
def delete_hospital_stay(stay=None):
        cmd = 'DELETE FROM clin.hospital_stay WHERE pk = %(pk)s'
        args = {'pk': stay}
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def delete_performed_procedure(procedure=None)
Expand source code
def delete_performed_procedure(procedure=None):
        cmd = 'delete from clin.procedure where pk = %(pk)s'
        args = {'pk': procedure}
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def diagnostic_certainty_classification2str(classification)
Expand source code
def diagnostic_certainty_classification2str(classification):

        global __diagnostic_certainty_classification_map

        if __diagnostic_certainty_classification_map is None:
                __diagnostic_certainty_classification_map = {
                        None: '',
                        'A': _('A: Sign'),
                        'B': _('B: Cluster of signs'),
                        'C': _('C: Syndromic diagnosis'),
                        'D': _('D: Scientific diagnosis')
                }

        try:
                return __diagnostic_certainty_classification_map[classification]
        except KeyError:
                return _('<%s>: unknown diagnostic certainty classification') % classification
def episode2problem(episode=None, allow_closed=False)
Expand source code
def episode2problem(episode=None, allow_closed=False):
        return cProblem (
                aPK_obj = {
                        'pk_patient': episode['pk_patient'],
                        'pk_episode': episode['pk_episode'],
                        'pk_health_issue': episode['pk_health_issue']
                },
                try_potential_problems = allow_closed
        )
def export_emr_structure(patient=None, filename=None)
Expand source code
def export_emr_structure(patient=None, filename=None):

        if filename is None:
                filename = gmTools.get_unique_filename(prefix = 'gm-emr_struct-%s-' % patient.subdir_name, suffix = '.txt')

        f = open(filename, 'w+', encoding = 'utf8')

        f.write('patient [%s]\n' % patient.description_gender)
        emr = patient.emr
        for issue in emr.health_issues:
                f.write('\n')
                f.write('\n')
                f.write('issue [%s] #%s\n' % (issue['description'], issue['pk_health_issue']))
                f.write(' is active     : %s\n' % issue['is_active'])
                f.write(' has open epi  : %s\n' % issue['has_open_episode'])
                f.write(' possible start: %s\n' % issue.possible_start_date)
                f.write(' safe start    : %s\n' % issue.safe_start_date)
                end = issue.clinical_end_date
                if end is None:
                        f.write(' end           : active and/or open episode\n')
                else:
                        f.write(' end           : %s\n' % end)
                f.write(' latest access : %s\n' % issue.latest_access_date)
                first = issue.first_episode
                if first is not None:
                        first = first['description']
                f.write(' 1st episode   : %s\n' % first)
                last = issue.latest_episode
                if last is not None:
                        last = last['description']
                f.write(' latest episode: %s\n' % last)
                epis = sorted(issue.get_episodes(), key = lambda e: e.best_guess_clinical_start_date)
                for epi in epis:
                        f.write('\n')
                        f.write(' episode [%s] #%s\n' % (epi['description'], epi['pk_episode']))
                        f.write('  is open         : %s\n' % epi['episode_open'])
                        f.write('  best guess start: %s\n' % epi.best_guess_clinical_start_date)
                        f.write('  best guess end  : %s\n' % epi.best_guess_clinical_end_date)
                        f.write('  latest access   : %s\n' % epi.latest_access_date)
                        f.write('  start 1st enc   : %s\n' % epi['started_first'])
                        f.write('  start last enc  : %s\n' % epi['started_last'])
                        f.write('  end last enc    : %s\n' % epi['last_affirmed'])

        f.close()
        return filename
def export_patient_emr_structure()
Expand source code
def export_patient_emr_structure():
        from Gnumed.business import gmPersonSearch
        gmPraxis.gmCurrentPraxisBranch(branch = gmPraxis.get_praxis_branches()[0])
        pat = gmPersonSearch.ask_for_patient()
        while pat is not None:
                print('patient:', pat.description_gender)
                fname = os.path.expanduser('~/gnumed/gm-emr_structure-%s.txt' % pat.subdir_name)
                print('exported into:', export_emr_structure(patient = pat, filename = fname))
                pat = gmPersonSearch.ask_for_patient()

        return 0
def format_clinical_duration_of_episode(start=None, end=None)
Expand source code
def format_clinical_duration_of_episode(start=None, end=None):
        assert (start is not None), '<start> must not be None'

        if end is None:
                start_end_str = '%s-%s' % (
                        gmDateTime.pydt_strftime(start, "%b'%y"),
                        gmTools.u_ellipsis
                )
                start_end_str_long = '%s - %s' % (
                        gmDateTime.pydt_strftime(start, '%b %d %Y'),
                        gmTools.u_ellipsis
                )
                duration_str = _('%s so far') % gmDateTime.format_interval_medically(gmDateTime.pydt_now_here() - start)
                return (start_end_str, start_end_str_long, duration_str)

        duration_str = gmDateTime.format_interval_medically(end - start)
        # year different:
        if end.year != start.year:
                start_end_str = '%s-%s' % (
                        gmDateTime.pydt_strftime(start, "%b'%y"),
                        gmDateTime.pydt_strftime(end, "%b'%y")
                )
                start_end_str_long = '%s - %s' % (
                        gmDateTime.pydt_strftime(start, '%b %d %Y'),
                        gmDateTime.pydt_strftime(end, '%b %d %Y')
                )
                return (start_end_str, start_end_str_long, duration_str)
        # same year:
        if end.month != start.month:
                start_end_str = '%s-%s' % (
                        gmDateTime.pydt_strftime(start, '%b'),
                        gmDateTime.pydt_strftime(end, "%b'%y")
                )
                start_end_str_long = '%s - %s' % (
                        gmDateTime.pydt_strftime(start, '%b %d'),
                        gmDateTime.pydt_strftime(end, '%b %d %Y')
                )
                return (start_end_str, start_end_str_long, duration_str)

        # same year and same month
        start_end_str = gmDateTime.pydt_strftime(start, "%b'%y")
        start_end_str_long = gmDateTime.pydt_strftime(start, '%b %d %Y')
        return (start_end_str, start_end_str_long, duration_str)
def get_best_guess_clinical_end_date_for_episode(pk_episode=None)
Expand source code
def get_best_guess_clinical_end_date_for_episode(pk_episode=None):
        assert (pk_episode is not None), '<pk_episode> must not be None'
        query = {
                'cmd': _SQL_best_guess_clinical_end_date_for_episode,
                'args': {'pk': pk_episode}
        }
        rows = gmPG2.run_ro_queries(queries = [query])
        return rows[0][0]
def get_best_guess_clinical_start_date_for_episode(pk_episode=None)
Expand source code
def get_best_guess_clinical_start_date_for_episode(pk_episode=None):
        assert (pk_episode is not None), '<pk_episode> must not be None'
        query = {
                'cmd': _SQL_best_guess_clinical_start_date_for_episode,
                'args': {'pk': pk_episode}
        }
        rows = gmPG2.run_ro_queries(queries = [query])
        return rows[0][0]
def get_dummy_health_issue()
Expand source code
def get_dummy_health_issue():
        issue = {
                'pk_health_issue': None,
                'description': _('Unattributed episodes'),
                'age_noted': None,
                'laterality': 'na',
                'is_active': True,
                'clinically_relevant': True,
                'is_confidential': None,
                'is_cause_of_death': False,
                'is_dummy': True,
                'grouping': None
        }
        return issue
def get_encounter_type(description=None)
Expand source code
def get_encounter_type(description=None):
        cmd = "SELECT * from clin.encounter_type where description = %s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': [description]}])
        return rows
def get_encounter_types()
Expand source code
def get_encounter_types():
        cmd = """
                SELECT
                        _(description) AS l10n_description,
                        description
                FROM
                        clin.encounter_type
                ORDER BY
                        l10n_description
        """
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd}])
        return rows
def get_latest_patient_hospital_stay(patient=None)
Expand source code
def get_latest_patient_hospital_stay(patient=None):
        cmd = _SQL_get_hospital_stays % "pk_patient = %(pat)s ORDER BY admission DESC LIMIT 1"
        queries = [{
                # this assumes non-overarching stays
                #'cmd': u'SELECT * FROM clin.v_hospital_stays WHERE pk_patient = %(pat)s ORDER BY admission DESC LIMIT 1',
                'cmd': cmd,
                'args': {'pat': patient}
        }]
        rows = gmPG2.run_ro_queries(queries = queries)
        if len(rows) == 0:
                return None
        return cHospitalStay(row = {'data': rows[0], 'pk_field': 'pk_hospital_stay'})
def get_latest_performed_procedure(patient=None)
Expand source code
def get_latest_performed_procedure(patient=None):
        queries = [{
                'cmd': 'select * FROM clin.v_procedures WHERE pk_patient = %(pat)s ORDER BY clin_when DESC LIMIT 1',
                'args': {'pat': patient}
        }]
        rows = gmPG2.run_ro_queries(queries = queries)
        if len(rows) == 0:
                return None
        return cPerformedProcedure(row = {'data': rows[0], 'pk_field': 'pk_procedure'})
def get_most_commonly_used_encounter_type()
Expand source code
def get_most_commonly_used_encounter_type():
        cmd = """
                SELECT
                        COUNT(*) AS type_count,
                        fk_type
                FROM clin.encounter
                GROUP BY fk_type
                ORDER BY type_count DESC
                LIMIT 1
        """
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd}])
        if len(rows) == 0:
                return None
        return rows[0]['fk_type']
def get_patient_hospital_stays(patient=None, ongoing_only=False, return_pks=False)
Expand source code
def get_patient_hospital_stays(patient=None, ongoing_only=False, return_pks=False):
        args = {'pat': patient}
        if ongoing_only:
                cmd = _SQL_get_hospital_stays % "pk_patient = %(pat)s AND discharge is NULL ORDER BY admission"
        else:
                cmd = _SQL_get_hospital_stays % "pk_patient = %(pat)s ORDER BY admission"

        queries = [{'cmd': cmd, 'args': args}]
        rows = gmPG2.run_ro_queries(queries = queries)
        if return_pks:
                return [ r['pk_hospital_stay'] for r in rows ]
        return [ cHospitalStay(row = {'data': r, 'pk_field': 'pk_hospital_stay'})  for r in rows ]
def get_performed_procedures(patient=None, return_pks=False)
Expand source code
def get_performed_procedures(patient=None, return_pks=False):
        queries = [{
                'cmd': 'SELECT * FROM clin.v_procedures WHERE pk_patient = %(pat)s ORDER BY clin_when',
                'args': {'pat': patient}
        }]
        rows = gmPG2.run_ro_queries(queries = queries)
        if return_pks:
                return [ r['pk_procedure'] for r in rows ]
        return [ cPerformedProcedure(row = {'data': r, 'pk_field': 'pk_procedure'})  for r in rows ]
def get_procedures4document(pk_document=None, return_pks=False)
Expand source code
def get_procedures4document(pk_document=None, return_pks=False):
        args = {'pk_doc': pk_document}
        cmd = _SQL_get_procedures % 'pk_doc = %(pk_doc)s'
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        if return_pks:
                return [ r['pk_procedure'] for r in rows ]
        return [ cPerformedProcedure(row = {'data': r, 'pk_field': 'pk_procedure'})  for r in rows ]
def health_issue2problem(health_issue=None, allow_irrelevant=False)
Expand source code
def health_issue2problem(health_issue=None, allow_irrelevant=False):
        return cProblem (
                aPK_obj = {
                        'pk_patient': health_issue['pk_patient'],
                        'pk_health_issue': health_issue['pk_health_issue'],
                        'pk_episode': None
                },
                try_potential_problems = allow_irrelevant
        )
def lock_encounter(pk_encounter, exclusive=False, link_obj=None)

Used to protect against deletion of active encounter from another client.

Expand source code
def lock_encounter(pk_encounter, exclusive=False, link_obj=None):
        """Used to protect against deletion of active encounter from another client."""
        return gmPG2.lock_row(link_obj = link_obj, table = 'clin.encounter', pk = pk_encounter, exclusive = exclusive)
def reclass_problem(problem=None)

Transform given problem into either episode or health issue instance.

Expand source code
def reclass_problem(problem=None):
        """Transform given problem into either episode or health issue instance.
        """
        if isinstance(problem, (cEpisode, cHealthIssue)):
                return problem

        exc = TypeError('cannot reclass [%s] instance to either episode or health issue' % type(problem))

        if not isinstance(problem, cProblem):
                _log.debug('%s' % problem)
                raise exc

        if problem['type'] == 'episode':
                return cEpisode(aPK_obj = problem['pk_episode'])

        if problem['type'] == 'issue':
                return cHealthIssue(aPK_obj = problem['pk_health_issue'])

        raise exc
def unlock_encounter(pk_encounter, exclusive=False, link_obj=None)
Expand source code
def unlock_encounter(pk_encounter, exclusive=False, link_obj=None):
        unlocked = gmPG2.unlock_row(link_obj = link_obj, table = 'clin.encounter', pk = pk_encounter, exclusive = exclusive)
        if not unlocked:
                _log.warning('cannot unlock encounter [#%s]', pk_encounter)
                call_stack = inspect.stack()
                call_stack.reverse()
                for idx in range(1, len(call_stack)):
                        caller = call_stack[idx]
                        _log.error('%s[%s] @ [%s] in [%s]', ' ' * idx, caller[3], caller[2], caller[1])
                del call_stack
        return unlocked
def update_encounter_type(description=None, l10n_description=None)
Expand source code
def update_encounter_type(description=None, l10n_description=None):

        rows = gmPG2.run_rw_queries(
                queries = [{
                'cmd': "select i18n.upd_tx(%(desc)s, %(l10n_desc)s)",
                'args': {'desc': description, 'l10n_desc': l10n_description}
                }],
                return_data = True
        )

        success = rows[0][0]
        if not success:
                _log.warning('updating encounter type [%s] to [%s] failed', description, l10n_description)

        return {'description': description, 'l10n_description': l10n_description}

Classes

class cEncounter (aPK_obj: int | dict = None, row: dict = None, link_obj=None)

Represents one encounter.

Call init from child classes like so:

    super().__init__(aPK_obj = aPK_obj, row = row, link_obj = link_obj)

Args

aPK_obj
retrieve data from backend
  • an scalar value the ._cmd_fetch_payload WHERE condition must be a simple column: "… WHERE pk_col = %s"
  • a dictionary of values the ._cmd_fetch_payload WHERE condition must consume the dictionary and produce a unique row
row
must hold the fields
  • data: list of column values for the row selected by ._cmd_fetch_payload (as returned by cursor.fetchone() in the DB-API)
  • pk_field: the name of the primary key column OR
  • pk_obj: a dictionary suitable for being passed to cursor.execute and holding the primary key values, used for composite PKs
  • for example:
    row = {
            'data': rows[0],
            'pk_field': 'pk_XXX (the PK column name)',
            'pk_obj': {'pk_col1': pk_col1_val, 'pk_col2': pk_col2_val}
    }
    
    rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
    objects = [ cChildClass(row = {'data': r, 'pk_field': 'the PK column name'}) for r in rows ]
    
Expand source code
class cEncounter(gmBusinessDBObject.cBusinessDBObject):
        """Represents one encounter."""

        _cmd_fetch_payload = SQL_get_encounters % 'pk_encounter = %s'
        _cmds_store_payload = [
                """UPDATE clin.encounter SET
                                started = %(started)s,
                                last_affirmed = %(last_affirmed)s,
                                fk_location = %(pk_org_unit)s,
                                fk_type = %(pk_type)s,
                                reason_for_encounter = gm.nullify_empty_string(%(reason_for_encounter)s),
                                assessment_of_encounter = gm.nullify_empty_string(%(assessment_of_encounter)s)
                        WHERE
                                pk = %(pk_encounter)s AND
                                xmin = %(xmin_encounter)s
                        """,
                # need to return all fields so we can survive in-place upgrades
                "SELECT * FROM clin.v_pat_encounters WHERE pk_encounter = %(pk_encounter)s"
        ]
        _updatable_fields = [
                'started',
                'last_affirmed',
                'pk_org_unit',
                'pk_type',
                'reason_for_encounter',
                'assessment_of_encounter'
        ]
        #--------------------------------------------------------
        def set_active(self):
                """Set the encounter as the active one.

                "Setting active" means making sure the encounter
                row has the youngest "last_affirmed" timestamp of
                all encounter rows for this patient.
                """
                self['last_affirmed'] = gmDateTime.pydt_now_here()
                self.save()
        #--------------------------------------------------------
        def lock(self, exclusive=False, link_obj=None):
                return lock_encounter(self.pk_obj, exclusive = exclusive, link_obj = link_obj)
        #--------------------------------------------------------
        def unlock(self, exclusive=False, link_obj=None):
                return unlock_encounter(self.pk_obj, exclusive = exclusive, link_obj = link_obj)
        #--------------------------------------------------------
        def transfer_clinical_data(self, source_episode=None, target_episode=None):
                """
                Moves every element currently linked to the current encounter
                and the source_episode onto target_episode.

                @param source_episode The episode the elements are currently linked to.
                @type target_episode A cEpisode intance.
                @param target_episode The episode the elements will be relinked to.
                @type target_episode A cEpisode intance.
                """
                if source_episode['pk_episode'] == target_episode['pk_episode']:
                        return True

                cmd = """
                        UPDATE clin.clin_root_item
                        SET fk_episode = %(trg)s
                        WHERE
                                fk_encounter = %(enc)s AND
                                fk_episode = %(src)s
                        """
                gmPG2.run_rw_queries(queries = [{
                        'cmd': cmd,
                        'args': {
                                'trg': target_episode['pk_episode'],
                                'enc': self.pk_obj,
                                'src': source_episode['pk_episode']
                        }
                }])
                self.refetch_payload()
                return True

        #--------------------------------------------------------
        def transfer_all_data_to_another_encounter(self, pk_target_encounter=None):
                if pk_target_encounter == self.pk_obj:
                        return True
                cmd = "SELECT clin.transfer_all_encounter_data(%(src)s, %(trg)s)"
                args = {
                        'src': self.pk_obj,
                        'trg': pk_target_encounter
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

#       conn = gmPG2.get_connection()
#       curs = conn.cursor()
#       curs.callproc('clin.get_hints_for_patient', [pk_identity])
#       rows = curs.fetchall()
#       idx = gmPG2.get_col_indices(curs)
#       curs.close()
#       conn.rollback()

        #--------------------------------------------------------
        def same_payload(self, another_object=None):

                relevant_fields = [
                        'pk_org_unit',
                        'pk_type',
                        'pk_patient',
                        'reason_for_encounter',
                        'assessment_of_encounter'
                ]
                for field in relevant_fields:
                        if self._payload[field] != another_object[field]:
                                _log.debug('mismatch on [%s]: "%s" vs. "%s"', field, self._payload[field], another_object[field])
                                return False

                relevant_fields = [
                        'started',
                        'last_affirmed',
                ]
                for field in relevant_fields:
                        if self._payload[field] is None:
                                if another_object[field] is None:
                                        continue
                                _log.debug('mismatch on [%s]: here="%s", other="%s"', field, self._payload[field], another_object[field])
                                return False

                        if another_object[field] is None:
                                return False

                        # compares at seconds granularity
                        if self._payload[field].strftime('%Y-%m-%d %H:%M:%S') != another_object[field].strftime('%Y-%m-%d %H:%M:%S'):
                                _log.debug('mismatch on [%s]: here="%s", other="%s"', field, self._payload[field], another_object[field])
                                return False

                # compare codes
                # 1) RFE
                if another_object['pk_generic_codes_rfe'] is None:
                        if self._payload['pk_generic_codes_rfe'] is not None:
                                return False
                if another_object['pk_generic_codes_rfe'] is not None:
                        if self._payload['pk_generic_codes_rfe'] is None:
                                return False
                if (
                        (another_object['pk_generic_codes_rfe'] is None)
                                and
                        (self._payload['pk_generic_codes_rfe'] is None)
                ) is False:
                        if set(another_object['pk_generic_codes_rfe']) != set(self._payload['pk_generic_codes_rfe']):
                                return False
                # 2) AOE
                if another_object['pk_generic_codes_aoe'] is None:
                        if self._payload['pk_generic_codes_aoe'] is not None:
                                return False
                if another_object['pk_generic_codes_aoe'] is not None:
                        if self._payload['pk_generic_codes_aoe'] is None:
                                return False
                if (
                        (another_object['pk_generic_codes_aoe'] is None)
                                and
                        (self._payload['pk_generic_codes_aoe'] is None)
                ) is False:
                        if set(another_object['pk_generic_codes_aoe']) != set(self._payload['pk_generic_codes_aoe']):
                                return False

                return True
        #--------------------------------------------------------
        def has_clinical_data(self):
                cmd = """
select exists (
        select 1 from clin.v_pat_items where pk_patient = %(pat)s and pk_encounter = %(enc)s
                union all
        select 1 from blobs.v_doc_med where pk_patient = %(pat)s and pk_encounter = %(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]

        #--------------------------------------------------------
        def has_narrative(self):
                cmd = """
select exists (
        select 1 from clin.v_pat_items where pk_patient=%(pat)s and pk_encounter=%(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
        #--------------------------------------------------------
        def has_soap_narrative(self, soap_cats=None):
                """soap_cats: <space> = admin category"""

                if soap_cats is None:
                        soap_cats = 'soap '
                else:
                        soap_cats = soap_cats.casefold()

                cats = []
                for cat in soap_cats:
                        if cat in 'soapu':
                                cats.append(cat)
                                continue
                        if cat == ' ':
                                cats.append(None)

                cmd = """
                        SELECT EXISTS (
                                SELECT 1 FROM clin.clin_narrative
                                WHERE
                                        fk_encounter = %(enc)s
                                                AND
                                        soap_cat = ANY(%(cats)s)
                                LIMIT 1
                        )
                """
                args = {'enc': self._payload['pk_encounter'], 'cats': cats}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd,'args': args}])
                return rows[0][0]
        #--------------------------------------------------------
        def has_documents(self):
                cmd = """
select exists (
        select 1 from blobs.v_doc_med where pk_patient = %(pat)s and pk_encounter = %(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
        #--------------------------------------------------------
        def get_latest_soap(self, soap_cat=None, episode=None):

                if soap_cat is not None:
                        soap_cat = soap_cat.casefold()

                if episode is None:
                        epi_part = 'fk_episode is null'
                else:
                        epi_part = 'fk_episode = %(epi)s'

                cmd = """
select narrative
from clin.clin_narrative
where
        fk_encounter = %%(enc)s
                and
        soap_cat = %%(cat)s
                and
        %s
order by clin_when desc
limit 1
                """ % epi_part

                args = {'enc': self.pk_obj, 'cat': soap_cat, 'epi': episode}

                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                if len(rows) == 0:
                        return None

                return rows[0][0]
        #--------------------------------------------------------
        def get_episodes(self, exclude=None):
                cmd = """
                        SELECT * FROM clin.v_pat_episodes
                        WHERE pk_episode IN (
                                        SELECT DISTINCT fk_episode
                                        FROM clin.clin_root_item
                                        WHERE fk_encounter = %%(enc)s

                                                UNION

                                        SELECT DISTINCT fk_episode
                                        FROM blobs.doc_med
                                        WHERE fk_encounter = %%(enc)s
                        ) %s"""
                args = {'enc': self.pk_obj}
                if exclude is not None:
                        cmd = cmd % 'AND pk_episode <> ALL(%(excluded)s)'
                        args['excluded'] = exclude
                else:
                        cmd = cmd % ''
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]

        episodes = property(get_episodes)

        #--------------------------------------------------------
        def add_code(self, pk_code=None, field=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                if field == 'rfe':
                        cmd = "INSERT INTO clin.lnk_code2rfe (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
                elif field == 'aoe':
                        cmd = "INSERT INTO clin.lnk_code2aoe (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
                else:
                        raise ValueError('<field> must be one of "rfe" or "aoe", not "%s"', field)
                args = {
                        'item': self._payload['pk_encounter'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def remove_code(self, pk_code=None, field=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                if field == 'rfe':
                        cmd = "DELETE FROM clin.lnk_code2rfe WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                elif field == 'aoe':
                        cmd = "DELETE FROM clin.lnk_code2aoe WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                else:
                        raise ValueError('<field> must be one of "rfe" or "aoe", not "%s"', field)
                args = {
                        'item': self._payload['pk_encounter'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        # data formatting
        #--------------------------------------------------------
        def format_soap(self, episodes=None, left_margin=0, soap_cats='soapu', emr=None, issues=None):

                lines = []
                for soap_cat in gmSoapDefs.soap_cats_str2list(soap_cats):
                        soap_cat_narratives = emr.get_clin_narrative (
                                episodes = episodes,
                                issues = issues,
                                encounters = [self._payload['pk_encounter']],
                                soap_cats = [soap_cat]
                        )
                        if soap_cat_narratives is None:
                                continue
                        if len(soap_cat_narratives) == 0:
                                continue

                        lines.append('%s%s %s %s' % (
                                gmTools.u_box_top_left_arc,
                                gmTools.u_box_horiz_single,
                                gmSoapDefs.soap_cat2l10n_str[soap_cat],
                                gmTools.u_box_horiz_single * 5
                        ))
                        for soap_entry in soap_cat_narratives:
                                txt = gmTools.wrap (
                                        text = soap_entry['narrative'],
                                        width = 75,
                                        initial_indent = '',
                                        subsequent_indent = (' ' * left_margin)
                                )
                                lines.append(txt)
                                when = gmDateTime.pydt_strftime (
                                        soap_entry['date'],
                                        format = '%Y-%m-%d %H:%M',
                                        accuracy = gmDateTime.acc_minutes
                                )
                                txt = '%s%s %.8s, %s %s' % (
                                        ' ' * 40,
                                        gmTools.u_box_horiz_light_heavy,
                                        soap_entry['modified_by'],
                                        when,
                                        gmTools.u_box_horiz_heavy_light
                                )
                                lines.append(txt)
                                lines.append('')

                return lines

        #--------------------------------------------------------
        def format_latex(self, date_format=None, soap_cats=None, soap_order=None):

                nothing2format = (
                        (self._payload['reason_for_encounter'] is None)
                                and
                        (self._payload['assessment_of_encounter'] is None)
                                and
                        (self.has_soap_narrative(soap_cats = 'soapu') is False)
                )
                if nothing2format:
                        return ''

                if date_format is None:
                        date_format = '%A, %b %d %Y'

                tex =  '% -------------------------------------------------------------\n'
                tex += '% much recommended: \\usepackage(tabu)\n'
                tex += '% much recommended: \\usepackage(longtable)\n'
                tex += '% best wrapped in: "\\begin{longtabu} to \\textwidth {lX[,L]}"\n'
                tex += '% -------------------------------------------------------------\n'
                tex += '\\hline \n'
                tex += '\\multicolumn{2}{l}{%s: %s ({\\footnotesize %s - %s})} \\tabularnewline \n' % (
                        gmTools.tex_escape_string(self._payload['l10n_type']),
                        gmTools.tex_escape_string (
                                gmDateTime.pydt_strftime (
                                        self._payload['started'],
                                        date_format,
                                        accuracy = gmDateTime.acc_days
                                )
                        ),
                        gmTools.tex_escape_string (
                                gmDateTime.pydt_strftime (
                                        self._payload['started'],
                                        '%H:%M',
                                        accuracy = gmDateTime.acc_minutes
                                )
                        ),
                        gmTools.tex_escape_string (
                                gmDateTime.pydt_strftime (
                                        self._payload['last_affirmed'],
                                        '%H:%M',
                                        accuracy = gmDateTime.acc_minutes
                                )
                        )
                )
                tex += '\\hline \n'

                if self._payload['reason_for_encounter'] is not None:
                        tex += '%s & %s \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('RFE')),
                                gmTools.tex_escape_string(self._payload['reason_for_encounter'])
                        )
                if self._payload['assessment_of_encounter'] is not None:
                        tex += '%s & %s \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('AOE')),
                                gmTools.tex_escape_string(self._payload['assessment_of_encounter'])
                        )

                for epi in self.get_episodes():
                        soaps = epi.get_narrative(soap_cats = soap_cats, encounters = [self.pk_obj], order_by = soap_order)
                        if len(soaps) == 0:
                                continue
                        tex += '\\multicolumn{2}{l}{\\emph{%s: %s%s}} \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('Problem')),
                                gmTools.tex_escape_string(epi['description']),
                                gmTools.tex_escape_string (
                                        gmTools.coalesce (
                                                value2test = diagnostic_certainty_classification2str(epi['diagnostic_certainty_classification']),
                                                return_instead = '',
                                                template4value = ' {\\footnotesize [%s]}',
                                                none_equivalents = [None, '']
                                        )
                                )
                        )
                        if epi['pk_health_issue'] is not None:
                                tex += '\\multicolumn{2}{l}{\\emph{%s: %s%s}} \\tabularnewline \n' % (
                                        gmTools.tex_escape_string(_('Health issue')),
                                        gmTools.tex_escape_string(epi['health_issue']),
                                        gmTools.tex_escape_string (
                                                gmTools.coalesce (
                                                        value2test = diagnostic_certainty_classification2str(epi['diagnostic_certainty_classification_issue']),
                                                        return_instead = '',
                                                        template4value = ' {\\footnotesize [%s]}',
                                                        none_equivalents = [None, '']
                                                )
                                        )
                                )
                        for soap in soaps:
                                tex += '{\\small %s} & {\\small %s} \\tabularnewline \n' % (
                                        gmTools.tex_escape_string(gmSoapDefs.soap_cat2l10n[soap['soap_cat']]),
                                        gmTools.tex_escape_string(soap['narrative'], replace_eol = True)
                                )
                        tex += ' & \\tabularnewline \n'

                return tex

        #--------------------------------------------------------
        def __format_header_fancy(self, left_margin=0):
                lines = []

                lines.append('%s%s: %s - %s (@%s)%s [#%s]' % (
                        ' ' * left_margin,
                        self._payload['l10n_type'],
                        self._payload['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                        self._payload['last_affirmed_original_tz'].strftime('%H:%M'),
                        self._payload['source_time_zone'],
                        gmTools.coalesce (
                                self._payload['assessment_of_encounter'],
                                '',
                                ' %s%%s%s' % (gmTools.u_left_double_angle_quote, gmTools.u_right_double_angle_quote)
                        ),
                        self._payload['pk_encounter']
                ))

                lines.append(_('  your time: %s - %s  (@%s = %s%s)\n') % (
                        self._payload['started'].strftime('%Y-%m-%d %H:%M'),
                        self._payload['last_affirmed'].strftime('%H:%M'),
                        gmDateTime.current_local_timezone_name,
                        gmTools.bool2subst (
                                gmDateTime.dst_currently_in_effect,
                                gmDateTime.py_dst_timezone_name,
                                gmDateTime.py_timezone_name
                        ),
                        gmTools.bool2subst(gmDateTime.dst_currently_in_effect, ' - ' + _('daylight savings time in effect'), '')
                ))

                if self._payload['praxis_branch'] is not None:
                        lines.append(_('Location: %s (%s)') % (self._payload['praxis_branch'], self._payload['praxis']))

                if self._payload['reason_for_encounter'] is not None:
                        lines.append('%s: %s' % (_('RFE'), self._payload['reason_for_encounter']))
                        codes = self.generic_codes_rfe
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')

                if self._payload['assessment_of_encounter'] is not None:
                        lines.append('%s: %s' % (_('AOE'), self._payload['assessment_of_encounter']))
                        codes = self.generic_codes_aoe
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')
                        del codes
                return lines

        #--------------------------------------------------------
        def format_header(self, fancy_header=True, left_margin=0, with_rfe_aoe=False):
                lines = []

                if fancy_header:
                        return self.__format_header_fancy(left_margin = left_margin)

                now = gmDateTime.pydt_now_here()
                if now.strftime('%Y-%m-%d') == self._payload['started_original_tz'].strftime('%Y-%m-%d'):
                        start = '%s %s' % (
                                _('today'),
                                self._payload['started_original_tz'].strftime('%H:%M')
                        )
                else:
                        start = self._payload['started_original_tz'].strftime('%Y-%m-%d %H:%M')
                lines.append('%s%s: %s - %s%s%s' % (
                        ' ' * left_margin,
                        self._payload['l10n_type'],
                        start,
                        self._payload['last_affirmed_original_tz'].strftime('%H:%M'),
                        gmTools.coalesce(self._payload['assessment_of_encounter'], '', ' \u00BB%s\u00AB'),
                        gmTools.coalesce(self._payload['praxis_branch'], '', ' @%s')
                ))
                if with_rfe_aoe:
                        if self._payload['reason_for_encounter'] is not None:
                                lines.append('%s: %s' % (_('RFE'), self._payload['reason_for_encounter']))
                        codes = self.generic_codes_rfe
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')
                        if self._payload['assessment_of_encounter'] is not None:
                                lines.append('%s: %s' % (_('AOE'), self._payload['assessment_of_encounter']))
                        codes = self.generic_codes_aoe
                        if len(codes) > 0:
                                lines.append('')
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        if len(codes) > 0:
                                lines.append('')
                        del codes

                return lines

        #--------------------------------------------------------
        def format_by_episode(self, episodes=None, issues=None, left_margin=0, patient=None, with_soap=False, with_tests=True, with_docs=True, with_vaccinations=True, with_family_history=True):

                if patient is not None:
                        emr = patient.emr

                lines = []
                if episodes is None:
                        episodes = [ e['pk_episode'] for e in self.episodes ]

                for pk in episodes:
                        epi = cEpisode(aPK_obj = pk)
                        lines.append(_('\nEpisode %s%s%s%s:') % (
                                gmTools.u_left_double_angle_quote,
                                epi['description'],
                                gmTools.u_right_double_angle_quote,
                                gmTools.coalesce(epi['health_issue'], '', ' (%s)')
                        ))

                        # soap
                        if with_soap:
                                if patient.ID != self._payload['pk_patient']:
                                        msg = '<patient>.ID = %s but encounter %s belongs to patient %s' % (
                                                patient.ID,
                                                self._payload['pk_encounter'],
                                                self._payload['pk_patient']
                                        )
                                        raise ValueError(msg)
                                lines.extend(self.format_soap (
                                        episodes = [pk],
                                        left_margin = left_margin,
                                        soap_cats = None,               # meaning: all
                                        emr = emr,
                                        issues = issues
                                ))

                        # test results
                        if with_tests:
                                tests = emr.get_test_results_by_date (
                                        episodes = [pk],
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(tests) > 0:
                                        lines.append('')
                                        lines.append(_('Measurements and Results:'))

                                for t in tests:
                                        lines.append(t.format())

                                del tests

                        # vaccinations
                        if with_vaccinations:
                                vaccs = emr.get_vaccinations (
                                        episodes = [pk],
                                        encounters = [ self._payload['pk_encounter'] ],
                                        order_by = 'date_given DESC, vaccine'
                                )
                                if len(vaccs) > 0:
                                        lines.append('')
                                        lines.append(_('Vaccinations:'))
                                for vacc in vaccs:
                                        lines.extend(vacc.format (
                                                with_indications = True,
                                                with_comment = True,
                                                with_reaction = True,
                                                date_format = '%Y-%m-%d'
                                        ))
                                del vaccs

                        # family history
                        if with_family_history:
                                fhx = emr.get_family_history(episodes = [pk])
                                if len(fhx) > 0:
                                        lines.append('')
                                        lines.append(_('Family History: %s') % len(fhx))
                                for f in fhx:
                                        lines.append(f.format (
                                                left_margin = (left_margin + 1),
                                                include_episode = False,
                                                include_comment = True
                                        ))
                                del fhx

                        # documents
                        if with_docs:
                                doc_folder = patient.get_document_folder()
                                docs = doc_folder.get_documents (
                                        pk_episodes = [pk],
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(docs) > 0:
                                        lines.append('')
                                        lines.append(_('Documents:'))
                                for d in docs:
                                        lines.append(' ' + d.format(single_line = True))
                                del docs

                return lines

        #--------------------------------------------------------
        def format_maximum_information(self, patient=None):
                if patient is None:
                        from Gnumed.business.gmPerson import gmCurrentPatient, cPerson
                        if self._payload['pk_patient'] == gmCurrentPatient().ID:
                                patient = gmCurrentPatient()
                        else:
                                patient = cPerson(self._payload['pk_patient'])

                return self.format (
                        patient = patient,
                        fancy_header = True,
                        with_rfe_aoe = True,
                        with_soap = True,
                        with_docs = True,
                        with_tests = False,
                        with_vaccinations = True,
                        with_co_encountlet_hints = True,
                        with_family_history = True,
                        by_episode = False,
                        return_list = True
                )

        #--------------------------------------------------------
        def format(self, episodes=None, with_soap=False, left_margin=0, patient=None, issues=None, with_docs=True, with_tests=True, fancy_header=True, with_vaccinations=True, with_co_encountlet_hints=False, with_rfe_aoe=False, with_family_history=True, by_episode=False, return_list=False):
                """Format an encounter.

                Args:
                        episode: which episodes, touched upon by this encounter, to include information for
                        with_co_encountlet_hints:
                                - whether to include which *other* episodes were discussed during this encounter
                                - (only makes sense if episodes is not None, since that would preclude information)
                """
                lines = self.format_header (
                        fancy_header = fancy_header,
                        left_margin = left_margin,
                        with_rfe_aoe = with_rfe_aoe
                )

                if patient is None:
                        _log.debug('no patient, cannot load patient related data')
                        with_soap = False
                        with_tests = False
                        with_vaccinations = False
                        with_docs = False

                if by_episode:
                        lines.extend(self.format_by_episode (
                                episodes = episodes,
                                issues = issues,
                                left_margin = left_margin,
                                patient = patient,
                                with_soap = with_soap,
                                with_tests = with_tests,
                                with_docs = with_docs,
                                with_vaccinations = with_vaccinations,
                                with_family_history = with_family_history
                        ))
                else:
                        if with_soap:
                                lines.append('')
                                if patient.ID != self._payload['pk_patient']:
                                        msg = '<patient>.ID = %s but encounter %s belongs to patient %s' % (
                                                patient.ID,
                                                self._payload['pk_encounter'],
                                                self._payload['pk_patient']
                                        )
                                        raise ValueError(msg)
                                emr = patient.emr
                                lines.extend(self.format_soap (
                                        episodes = episodes,
                                        left_margin = left_margin,
                                        soap_cats = None,               # meaning: all
                                        emr = emr,
                                        issues = issues
                                ))

        #               # family history
        #               if with_family_history:
        #                       if episodes is not None:
        #                               fhx = emr.get_family_history(episodes = episodes)
        #                               if len(fhx) > 0:
        #                                       lines.append(u'')
        #                                       lines.append(_('Family History: %s') % len(fhx))
        #                               for f in fhx:
        #                                       lines.append(f.format (
        #                                               left_margin = (left_margin + 1),
        #                                               include_episode = False,
        #                                               include_comment = True
        #                                       ))
        #                               del fhx

                        # test results
                        if with_tests:
                                emr = patient.emr
                                tests = emr.get_test_results_by_date (
                                        episodes = episodes,
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(tests) > 0:
                                        lines.append('')
                                        lines.append(_('Measurements and Results:'))
                                for t in tests:
                                        lines.append(t.format())
                                del tests

                        # vaccinations
                        if with_vaccinations:
                                emr = patient.emr
                                vaccs = emr.get_vaccinations (
                                        episodes = episodes,
                                        encounters = [ self._payload['pk_encounter'] ],
                                        order_by = 'date_given DESC, vaccine'
                                )
                                if len(vaccs) > 0:
                                        lines.append('')
                                        lines.append(_('Vaccinations:'))
                                for vacc in vaccs:
                                        lines.extend(vacc.format (
                                                with_indications = True,
                                                with_comment = True,
                                                with_reaction = True,
                                                date_format = '%Y-%m-%d'
                                        ))
                                del vaccs

                        # documents
                        if with_docs:
                                doc_folder = patient.get_document_folder()
                                docs = doc_folder.get_documents (
                                        pk_episodes = episodes,
                                        encounter = self._payload['pk_encounter']
                                )
                                if len(docs) > 0:
                                        lines.append('')
                                        lines.append(_('Documents:'))
                                for d in docs:
                                        lines.append(' ' + d.format(single_line = True))
                                del docs

                        # co-encountlets
                        if with_co_encountlet_hints:
                                if episodes is not None:
                                        other_epis = self.get_episodes(exclude = episodes)
                                        if len(other_epis) > 0:
                                                lines.append('')
                                                lines.append(_('%s other episodes touched upon during this encounter:') % len(other_epis))
                                        for epi in other_epis:
                                                lines.append(' %s%s%s%s' % (
                                                        gmTools.u_left_double_angle_quote,
                                                        epi['description'],
                                                        gmTools.u_right_double_angle_quote,
                                                        gmTools.coalesce(epi['health_issue'], '', ' (%s)')
                                                ))

                if return_list:
                        return lines

                eol_w_margin = '\n%s' % (' ' * left_margin)
                return '%s\n' % eol_w_margin.join(lines)

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_generic_codes_rfe(self):
                if len(self._payload['pk_generic_codes_rfe']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes_rfe']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes_rfe(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes_rfe']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2rfe WHERE fk_item = %(enc)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'codes': self._payload['pk_generic_codes_rfe']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2rfe (fk_item, fk_generic_code) VALUES (%(enc)s, %(pk_code)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                self.refetch_payload()
                return

        generic_codes_rfe = property(_get_generic_codes_rfe, _set_generic_codes_rfe)
        #--------------------------------------------------------
        def _get_generic_codes_aoe(self):
                if len(self._payload['pk_generic_codes_aoe']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes_aoe']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes_aoe(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes_aoe']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2aoe WHERE fk_item = %(enc)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'codes': self._payload['pk_generic_codes_aoe']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2aoe (fk_item, fk_generic_code) VALUES (%(enc)s, %(pk_code)s)',
                                'args': {
                                        'enc': self._payload['pk_encounter'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                self.refetch_payload()
                return

        generic_codes_aoe = property(_get_generic_codes_aoe, _set_generic_codes_aoe)
        #--------------------------------------------------------
        def _get_praxis_branch(self):
                if self._payload['pk_org_unit'] is None:
                        return None
                return gmPraxis.get_praxis_branch_by_org_unit(pk_org_unit = self._payload['pk_org_unit'])

        praxis_branch = property(_get_praxis_branch)
        #--------------------------------------------------------
        def _get_org_unit(self):
                if self._payload['pk_org_unit'] is None:
                        return None
                return gmOrganization.cOrgUnit(aPK_obj = self._payload['pk_org_unit'])

        org_unit = property(_get_org_unit)
        #--------------------------------------------------------
        def _get_formatted_revision_history(self):
                cmd = """SELECT
                                '<N/A>'::TEXT as audit__action_applied,
                                NULL AS audit__action_when,
                                '<N/A>'::TEXT AS audit__action_by,
                                pk_audit,
                                row_version,
                                modified_when,
                                modified_by,
                                pk, fk_patient, fk_type, fk_location, source_time_zone, reason_for_encounter, assessment_of_encounter, started, last_affirmed
                        FROM clin.encounter
                        WHERE pk = %(pk_encounter)s
                UNION ALL (
                        SELECT
                                audit_action as audit__action_applied,
                                audit_when as audit__action_when,
                                audit_by as audit__action_by,
                                pk_audit,
                                orig_version as row_version,
                                orig_when as modified_when,
                                orig_by as modified_by,
                                pk, fk_patient, fk_type, fk_location, source_time_zone, reason_for_encounter, assessment_of_encounter, started, last_affirmed
                        FROM audit.log_encounter
                        WHERE pk = %(pk_encounter)s
                )
                ORDER BY row_version DESC
                """
                args = {'pk_encounter': self._payload['pk_encounter']}
                title = _('Encounter: %s%s%s') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['l10n_type'],
                        gmTools.u_right_double_angle_quote
                )
                return '\n'.join(self._get_revision_history(cmd, args, title))

        formatted_revision_history = property(_get_formatted_revision_history)

Ancestors

Instance variables

var episodes
Expand source code
def get_episodes(self, exclude=None):
        cmd = """
                SELECT * FROM clin.v_pat_episodes
                WHERE pk_episode IN (
                                SELECT DISTINCT fk_episode
                                FROM clin.clin_root_item
                                WHERE fk_encounter = %%(enc)s

                                        UNION

                                SELECT DISTINCT fk_episode
                                FROM blobs.doc_med
                                WHERE fk_encounter = %%(enc)s
                ) %s"""
        args = {'enc': self.pk_obj}
        if exclude is not None:
                cmd = cmd % 'AND pk_episode <> ALL(%(excluded)s)'
                args['excluded'] = exclude
        else:
                cmd = cmd % ''
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]
var formatted_revision_history
Expand source code
def _get_formatted_revision_history(self):
        cmd = """SELECT
                        '<N/A>'::TEXT as audit__action_applied,
                        NULL AS audit__action_when,
                        '<N/A>'::TEXT AS audit__action_by,
                        pk_audit,
                        row_version,
                        modified_when,
                        modified_by,
                        pk, fk_patient, fk_type, fk_location, source_time_zone, reason_for_encounter, assessment_of_encounter, started, last_affirmed
                FROM clin.encounter
                WHERE pk = %(pk_encounter)s
        UNION ALL (
                SELECT
                        audit_action as audit__action_applied,
                        audit_when as audit__action_when,
                        audit_by as audit__action_by,
                        pk_audit,
                        orig_version as row_version,
                        orig_when as modified_when,
                        orig_by as modified_by,
                        pk, fk_patient, fk_type, fk_location, source_time_zone, reason_for_encounter, assessment_of_encounter, started, last_affirmed
                FROM audit.log_encounter
                WHERE pk = %(pk_encounter)s
        )
        ORDER BY row_version DESC
        """
        args = {'pk_encounter': self._payload['pk_encounter']}
        title = _('Encounter: %s%s%s') % (
                gmTools.u_left_double_angle_quote,
                self._payload['l10n_type'],
                gmTools.u_right_double_angle_quote
        )
        return '\n'.join(self._get_revision_history(cmd, args, title))
var generic_codes_aoe
Expand source code
def _get_generic_codes_aoe(self):
        if len(self._payload['pk_generic_codes_aoe']) == 0:
                return []

        cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
        args = {'pks': self._payload['pk_generic_codes_aoe']}
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]
var generic_codes_rfe
Expand source code
def _get_generic_codes_rfe(self):
        if len(self._payload['pk_generic_codes_rfe']) == 0:
                return []

        cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
        args = {'pks': self._payload['pk_generic_codes_rfe']}
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]
var org_unit
Expand source code
def _get_org_unit(self):
        if self._payload['pk_org_unit'] is None:
                return None
        return gmOrganization.cOrgUnit(aPK_obj = self._payload['pk_org_unit'])
var praxis_branch
Expand source code
def _get_praxis_branch(self):
        if self._payload['pk_org_unit'] is None:
                return None
        return gmPraxis.get_praxis_branch_by_org_unit(pk_org_unit = self._payload['pk_org_unit'])

Methods

def add_code(self, pk_code=None, field=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def add_code(self, pk_code=None, field=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        if field == 'rfe':
                cmd = "INSERT INTO clin.lnk_code2rfe (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
        elif field == 'aoe':
                cmd = "INSERT INTO clin.lnk_code2aoe (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
        else:
                raise ValueError('<field> must be one of "rfe" or "aoe", not "%s"', field)
        args = {
                'item': self._payload['pk_encounter'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def format(self, episodes=None, with_soap=False, left_margin=0, patient=None, issues=None, with_docs=True, with_tests=True, fancy_header=True, with_vaccinations=True, with_co_encountlet_hints=False, with_rfe_aoe=False, with_family_history=True, by_episode=False, return_list=False)

Format an encounter.

Args

episode
which episodes, touched upon by this encounter, to include information for

with_co_encountlet_hints: - whether to include which other episodes were discussed during this encounter - (only makes sense if episodes is not None, since that would preclude information)

Expand source code
def format(self, episodes=None, with_soap=False, left_margin=0, patient=None, issues=None, with_docs=True, with_tests=True, fancy_header=True, with_vaccinations=True, with_co_encountlet_hints=False, with_rfe_aoe=False, with_family_history=True, by_episode=False, return_list=False):
        """Format an encounter.

        Args:
                episode: which episodes, touched upon by this encounter, to include information for
                with_co_encountlet_hints:
                        - whether to include which *other* episodes were discussed during this encounter
                        - (only makes sense if episodes is not None, since that would preclude information)
        """
        lines = self.format_header (
                fancy_header = fancy_header,
                left_margin = left_margin,
                with_rfe_aoe = with_rfe_aoe
        )

        if patient is None:
                _log.debug('no patient, cannot load patient related data')
                with_soap = False
                with_tests = False
                with_vaccinations = False
                with_docs = False

        if by_episode:
                lines.extend(self.format_by_episode (
                        episodes = episodes,
                        issues = issues,
                        left_margin = left_margin,
                        patient = patient,
                        with_soap = with_soap,
                        with_tests = with_tests,
                        with_docs = with_docs,
                        with_vaccinations = with_vaccinations,
                        with_family_history = with_family_history
                ))
        else:
                if with_soap:
                        lines.append('')
                        if patient.ID != self._payload['pk_patient']:
                                msg = '<patient>.ID = %s but encounter %s belongs to patient %s' % (
                                        patient.ID,
                                        self._payload['pk_encounter'],
                                        self._payload['pk_patient']
                                )
                                raise ValueError(msg)
                        emr = patient.emr
                        lines.extend(self.format_soap (
                                episodes = episodes,
                                left_margin = left_margin,
                                soap_cats = None,               # meaning: all
                                emr = emr,
                                issues = issues
                        ))

#               # family history
#               if with_family_history:
#                       if episodes is not None:
#                               fhx = emr.get_family_history(episodes = episodes)
#                               if len(fhx) > 0:
#                                       lines.append(u'')
#                                       lines.append(_('Family History: %s') % len(fhx))
#                               for f in fhx:
#                                       lines.append(f.format (
#                                               left_margin = (left_margin + 1),
#                                               include_episode = False,
#                                               include_comment = True
#                                       ))
#                               del fhx

                # test results
                if with_tests:
                        emr = patient.emr
                        tests = emr.get_test_results_by_date (
                                episodes = episodes,
                                encounter = self._payload['pk_encounter']
                        )
                        if len(tests) > 0:
                                lines.append('')
                                lines.append(_('Measurements and Results:'))
                        for t in tests:
                                lines.append(t.format())
                        del tests

                # vaccinations
                if with_vaccinations:
                        emr = patient.emr
                        vaccs = emr.get_vaccinations (
                                episodes = episodes,
                                encounters = [ self._payload['pk_encounter'] ],
                                order_by = 'date_given DESC, vaccine'
                        )
                        if len(vaccs) > 0:
                                lines.append('')
                                lines.append(_('Vaccinations:'))
                        for vacc in vaccs:
                                lines.extend(vacc.format (
                                        with_indications = True,
                                        with_comment = True,
                                        with_reaction = True,
                                        date_format = '%Y-%m-%d'
                                ))
                        del vaccs

                # documents
                if with_docs:
                        doc_folder = patient.get_document_folder()
                        docs = doc_folder.get_documents (
                                pk_episodes = episodes,
                                encounter = self._payload['pk_encounter']
                        )
                        if len(docs) > 0:
                                lines.append('')
                                lines.append(_('Documents:'))
                        for d in docs:
                                lines.append(' ' + d.format(single_line = True))
                        del docs

                # co-encountlets
                if with_co_encountlet_hints:
                        if episodes is not None:
                                other_epis = self.get_episodes(exclude = episodes)
                                if len(other_epis) > 0:
                                        lines.append('')
                                        lines.append(_('%s other episodes touched upon during this encounter:') % len(other_epis))
                                for epi in other_epis:
                                        lines.append(' %s%s%s%s' % (
                                                gmTools.u_left_double_angle_quote,
                                                epi['description'],
                                                gmTools.u_right_double_angle_quote,
                                                gmTools.coalesce(epi['health_issue'], '', ' (%s)')
                                        ))

        if return_list:
                return lines

        eol_w_margin = '\n%s' % (' ' * left_margin)
        return '%s\n' % eol_w_margin.join(lines)
def format_by_episode(self, episodes=None, issues=None, left_margin=0, patient=None, with_soap=False, with_tests=True, with_docs=True, with_vaccinations=True, with_family_history=True)
Expand source code
def format_by_episode(self, episodes=None, issues=None, left_margin=0, patient=None, with_soap=False, with_tests=True, with_docs=True, with_vaccinations=True, with_family_history=True):

        if patient is not None:
                emr = patient.emr

        lines = []
        if episodes is None:
                episodes = [ e['pk_episode'] for e in self.episodes ]

        for pk in episodes:
                epi = cEpisode(aPK_obj = pk)
                lines.append(_('\nEpisode %s%s%s%s:') % (
                        gmTools.u_left_double_angle_quote,
                        epi['description'],
                        gmTools.u_right_double_angle_quote,
                        gmTools.coalesce(epi['health_issue'], '', ' (%s)')
                ))

                # soap
                if with_soap:
                        if patient.ID != self._payload['pk_patient']:
                                msg = '<patient>.ID = %s but encounter %s belongs to patient %s' % (
                                        patient.ID,
                                        self._payload['pk_encounter'],
                                        self._payload['pk_patient']
                                )
                                raise ValueError(msg)
                        lines.extend(self.format_soap (
                                episodes = [pk],
                                left_margin = left_margin,
                                soap_cats = None,               # meaning: all
                                emr = emr,
                                issues = issues
                        ))

                # test results
                if with_tests:
                        tests = emr.get_test_results_by_date (
                                episodes = [pk],
                                encounter = self._payload['pk_encounter']
                        )
                        if len(tests) > 0:
                                lines.append('')
                                lines.append(_('Measurements and Results:'))

                        for t in tests:
                                lines.append(t.format())

                        del tests

                # vaccinations
                if with_vaccinations:
                        vaccs = emr.get_vaccinations (
                                episodes = [pk],
                                encounters = [ self._payload['pk_encounter'] ],
                                order_by = 'date_given DESC, vaccine'
                        )
                        if len(vaccs) > 0:
                                lines.append('')
                                lines.append(_('Vaccinations:'))
                        for vacc in vaccs:
                                lines.extend(vacc.format (
                                        with_indications = True,
                                        with_comment = True,
                                        with_reaction = True,
                                        date_format = '%Y-%m-%d'
                                ))
                        del vaccs

                # family history
                if with_family_history:
                        fhx = emr.get_family_history(episodes = [pk])
                        if len(fhx) > 0:
                                lines.append('')
                                lines.append(_('Family History: %s') % len(fhx))
                        for f in fhx:
                                lines.append(f.format (
                                        left_margin = (left_margin + 1),
                                        include_episode = False,
                                        include_comment = True
                                ))
                        del fhx

                # documents
                if with_docs:
                        doc_folder = patient.get_document_folder()
                        docs = doc_folder.get_documents (
                                pk_episodes = [pk],
                                encounter = self._payload['pk_encounter']
                        )
                        if len(docs) > 0:
                                lines.append('')
                                lines.append(_('Documents:'))
                        for d in docs:
                                lines.append(' ' + d.format(single_line = True))
                        del docs

        return lines
def format_header(self, fancy_header=True, left_margin=0, with_rfe_aoe=False)
Expand source code
def format_header(self, fancy_header=True, left_margin=0, with_rfe_aoe=False):
        lines = []

        if fancy_header:
                return self.__format_header_fancy(left_margin = left_margin)

        now = gmDateTime.pydt_now_here()
        if now.strftime('%Y-%m-%d') == self._payload['started_original_tz'].strftime('%Y-%m-%d'):
                start = '%s %s' % (
                        _('today'),
                        self._payload['started_original_tz'].strftime('%H:%M')
                )
        else:
                start = self._payload['started_original_tz'].strftime('%Y-%m-%d %H:%M')
        lines.append('%s%s: %s - %s%s%s' % (
                ' ' * left_margin,
                self._payload['l10n_type'],
                start,
                self._payload['last_affirmed_original_tz'].strftime('%H:%M'),
                gmTools.coalesce(self._payload['assessment_of_encounter'], '', ' \u00BB%s\u00AB'),
                gmTools.coalesce(self._payload['praxis_branch'], '', ' @%s')
        ))
        if with_rfe_aoe:
                if self._payload['reason_for_encounter'] is not None:
                        lines.append('%s: %s' % (_('RFE'), self._payload['reason_for_encounter']))
                codes = self.generic_codes_rfe
                for c in codes:
                        lines.append(' %s: %s (%s - %s)' % (
                                c['code'],
                                c['term'],
                                c['name_short'],
                                c['version']
                        ))
                if len(codes) > 0:
                        lines.append('')
                if self._payload['assessment_of_encounter'] is not None:
                        lines.append('%s: %s' % (_('AOE'), self._payload['assessment_of_encounter']))
                codes = self.generic_codes_aoe
                if len(codes) > 0:
                        lines.append('')
                for c in codes:
                        lines.append(' %s: %s (%s - %s)' % (
                                c['code'],
                                c['term'],
                                c['name_short'],
                                c['version']
                        ))
                if len(codes) > 0:
                        lines.append('')
                del codes

        return lines
def format_latex(self, date_format=None, soap_cats=None, soap_order=None)
Expand source code
def format_latex(self, date_format=None, soap_cats=None, soap_order=None):

        nothing2format = (
                (self._payload['reason_for_encounter'] is None)
                        and
                (self._payload['assessment_of_encounter'] is None)
                        and
                (self.has_soap_narrative(soap_cats = 'soapu') is False)
        )
        if nothing2format:
                return ''

        if date_format is None:
                date_format = '%A, %b %d %Y'

        tex =  '% -------------------------------------------------------------\n'
        tex += '% much recommended: \\usepackage(tabu)\n'
        tex += '% much recommended: \\usepackage(longtable)\n'
        tex += '% best wrapped in: "\\begin{longtabu} to \\textwidth {lX[,L]}"\n'
        tex += '% -------------------------------------------------------------\n'
        tex += '\\hline \n'
        tex += '\\multicolumn{2}{l}{%s: %s ({\\footnotesize %s - %s})} \\tabularnewline \n' % (
                gmTools.tex_escape_string(self._payload['l10n_type']),
                gmTools.tex_escape_string (
                        gmDateTime.pydt_strftime (
                                self._payload['started'],
                                date_format,
                                accuracy = gmDateTime.acc_days
                        )
                ),
                gmTools.tex_escape_string (
                        gmDateTime.pydt_strftime (
                                self._payload['started'],
                                '%H:%M',
                                accuracy = gmDateTime.acc_minutes
                        )
                ),
                gmTools.tex_escape_string (
                        gmDateTime.pydt_strftime (
                                self._payload['last_affirmed'],
                                '%H:%M',
                                accuracy = gmDateTime.acc_minutes
                        )
                )
        )
        tex += '\\hline \n'

        if self._payload['reason_for_encounter'] is not None:
                tex += '%s & %s \\tabularnewline \n' % (
                        gmTools.tex_escape_string(_('RFE')),
                        gmTools.tex_escape_string(self._payload['reason_for_encounter'])
                )
        if self._payload['assessment_of_encounter'] is not None:
                tex += '%s & %s \\tabularnewline \n' % (
                        gmTools.tex_escape_string(_('AOE')),
                        gmTools.tex_escape_string(self._payload['assessment_of_encounter'])
                )

        for epi in self.get_episodes():
                soaps = epi.get_narrative(soap_cats = soap_cats, encounters = [self.pk_obj], order_by = soap_order)
                if len(soaps) == 0:
                        continue
                tex += '\\multicolumn{2}{l}{\\emph{%s: %s%s}} \\tabularnewline \n' % (
                        gmTools.tex_escape_string(_('Problem')),
                        gmTools.tex_escape_string(epi['description']),
                        gmTools.tex_escape_string (
                                gmTools.coalesce (
                                        value2test = diagnostic_certainty_classification2str(epi['diagnostic_certainty_classification']),
                                        return_instead = '',
                                        template4value = ' {\\footnotesize [%s]}',
                                        none_equivalents = [None, '']
                                )
                        )
                )
                if epi['pk_health_issue'] is not None:
                        tex += '\\multicolumn{2}{l}{\\emph{%s: %s%s}} \\tabularnewline \n' % (
                                gmTools.tex_escape_string(_('Health issue')),
                                gmTools.tex_escape_string(epi['health_issue']),
                                gmTools.tex_escape_string (
                                        gmTools.coalesce (
                                                value2test = diagnostic_certainty_classification2str(epi['diagnostic_certainty_classification_issue']),
                                                return_instead = '',
                                                template4value = ' {\\footnotesize [%s]}',
                                                none_equivalents = [None, '']
                                        )
                                )
                        )
                for soap in soaps:
                        tex += '{\\small %s} & {\\small %s} \\tabularnewline \n' % (
                                gmTools.tex_escape_string(gmSoapDefs.soap_cat2l10n[soap['soap_cat']]),
                                gmTools.tex_escape_string(soap['narrative'], replace_eol = True)
                        )
                tex += ' & \\tabularnewline \n'

        return tex
def format_maximum_information(self, patient=None)
Expand source code
def format_maximum_information(self, patient=None):
        if patient is None:
                from Gnumed.business.gmPerson import gmCurrentPatient, cPerson
                if self._payload['pk_patient'] == gmCurrentPatient().ID:
                        patient = gmCurrentPatient()
                else:
                        patient = cPerson(self._payload['pk_patient'])

        return self.format (
                patient = patient,
                fancy_header = True,
                with_rfe_aoe = True,
                with_soap = True,
                with_docs = True,
                with_tests = False,
                with_vaccinations = True,
                with_co_encountlet_hints = True,
                with_family_history = True,
                by_episode = False,
                return_list = True
        )
def format_soap(self, episodes=None, left_margin=0, soap_cats='soapu', emr=None, issues=None)
Expand source code
def format_soap(self, episodes=None, left_margin=0, soap_cats='soapu', emr=None, issues=None):

        lines = []
        for soap_cat in gmSoapDefs.soap_cats_str2list(soap_cats):
                soap_cat_narratives = emr.get_clin_narrative (
                        episodes = episodes,
                        issues = issues,
                        encounters = [self._payload['pk_encounter']],
                        soap_cats = [soap_cat]
                )
                if soap_cat_narratives is None:
                        continue
                if len(soap_cat_narratives) == 0:
                        continue

                lines.append('%s%s %s %s' % (
                        gmTools.u_box_top_left_arc,
                        gmTools.u_box_horiz_single,
                        gmSoapDefs.soap_cat2l10n_str[soap_cat],
                        gmTools.u_box_horiz_single * 5
                ))
                for soap_entry in soap_cat_narratives:
                        txt = gmTools.wrap (
                                text = soap_entry['narrative'],
                                width = 75,
                                initial_indent = '',
                                subsequent_indent = (' ' * left_margin)
                        )
                        lines.append(txt)
                        when = gmDateTime.pydt_strftime (
                                soap_entry['date'],
                                format = '%Y-%m-%d %H:%M',
                                accuracy = gmDateTime.acc_minutes
                        )
                        txt = '%s%s %.8s, %s %s' % (
                                ' ' * 40,
                                gmTools.u_box_horiz_light_heavy,
                                soap_entry['modified_by'],
                                when,
                                gmTools.u_box_horiz_heavy_light
                        )
                        lines.append(txt)
                        lines.append('')

        return lines
def get_episodes(self, exclude=None)
Expand source code
def get_episodes(self, exclude=None):
        cmd = """
                SELECT * FROM clin.v_pat_episodes
                WHERE pk_episode IN (
                                SELECT DISTINCT fk_episode
                                FROM clin.clin_root_item
                                WHERE fk_encounter = %%(enc)s

                                        UNION

                                SELECT DISTINCT fk_episode
                                FROM blobs.doc_med
                                WHERE fk_encounter = %%(enc)s
                ) %s"""
        args = {'enc': self.pk_obj}
        if exclude is not None:
                cmd = cmd % 'AND pk_episode <> ALL(%(excluded)s)'
                args['excluded'] = exclude
        else:
                cmd = cmd % ''
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]
def get_latest_soap(self, soap_cat=None, episode=None)
Expand source code
        def get_latest_soap(self, soap_cat=None, episode=None):

                if soap_cat is not None:
                        soap_cat = soap_cat.casefold()

                if episode is None:
                        epi_part = 'fk_episode is null'
                else:
                        epi_part = 'fk_episode = %(epi)s'

                cmd = """
select narrative
from clin.clin_narrative
where
        fk_encounter = %%(enc)s
                and
        soap_cat = %%(cat)s
                and
        %s
order by clin_when desc
limit 1
                """ % epi_part

                args = {'enc': self.pk_obj, 'cat': soap_cat, 'epi': episode}

                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                if len(rows) == 0:
                        return None

                return rows[0][0]
def has_clinical_data(self)
Expand source code
        def has_clinical_data(self):
                cmd = """
select exists (
        select 1 from clin.v_pat_items where pk_patient = %(pat)s and pk_encounter = %(enc)s
                union all
        select 1 from blobs.v_doc_med where pk_patient = %(pat)s and pk_encounter = %(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
def has_documents(self)
Expand source code
        def has_documents(self):
                cmd = """
select exists (
        select 1 from blobs.v_doc_med where pk_patient = %(pat)s and pk_encounter = %(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
def has_narrative(self)
Expand source code
        def has_narrative(self):
                cmd = """
select exists (
        select 1 from clin.v_pat_items where pk_patient=%(pat)s and pk_encounter=%(enc)s
)"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'enc': self.pk_obj
                }
                rows = gmPG2.run_ro_queries (
                        queries = [{
                                'cmd': cmd,
                                'args': args
                        }]
                )
                return rows[0][0]
def has_soap_narrative(self, soap_cats=None)

soap_cats: = admin category

Expand source code
def has_soap_narrative(self, soap_cats=None):
        """soap_cats: <space> = admin category"""

        if soap_cats is None:
                soap_cats = 'soap '
        else:
                soap_cats = soap_cats.casefold()

        cats = []
        for cat in soap_cats:
                if cat in 'soapu':
                        cats.append(cat)
                        continue
                if cat == ' ':
                        cats.append(None)

        cmd = """
                SELECT EXISTS (
                        SELECT 1 FROM clin.clin_narrative
                        WHERE
                                fk_encounter = %(enc)s
                                        AND
                                soap_cat = ANY(%(cats)s)
                        LIMIT 1
                )
        """
        args = {'enc': self._payload['pk_encounter'], 'cats': cats}
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd,'args': args}])
        return rows[0][0]
def lock(self, exclusive=False, link_obj=None)
Expand source code
def lock(self, exclusive=False, link_obj=None):
        return lock_encounter(self.pk_obj, exclusive = exclusive, link_obj = link_obj)
def remove_code(self, pk_code=None, field=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def remove_code(self, pk_code=None, field=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        if field == 'rfe':
                cmd = "DELETE FROM clin.lnk_code2rfe WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
        elif field == 'aoe':
                cmd = "DELETE FROM clin.lnk_code2aoe WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
        else:
                raise ValueError('<field> must be one of "rfe" or "aoe", not "%s"', field)
        args = {
                'item': self._payload['pk_encounter'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def set_active(self)

Set the encounter as the active one.

"Setting active" means making sure the encounter row has the youngest "last_affirmed" timestamp of all encounter rows for this patient.

Expand source code
def set_active(self):
        """Set the encounter as the active one.

        "Setting active" means making sure the encounter
        row has the youngest "last_affirmed" timestamp of
        all encounter rows for this patient.
        """
        self['last_affirmed'] = gmDateTime.pydt_now_here()
        self.save()
def transfer_all_data_to_another_encounter(self, pk_target_encounter=None)
Expand source code
def transfer_all_data_to_another_encounter(self, pk_target_encounter=None):
        if pk_target_encounter == self.pk_obj:
                return True
        cmd = "SELECT clin.transfer_all_encounter_data(%(src)s, %(trg)s)"
        args = {
                'src': self.pk_obj,
                'trg': pk_target_encounter
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def transfer_clinical_data(self, source_episode=None, target_episode=None)

Moves every element currently linked to the current encounter and the source_episode onto target_episode.

@param source_episode The episode the elements are currently linked to. @type target_episode A cEpisode intance. @param target_episode The episode the elements will be relinked to. @type target_episode A cEpisode intance.

Expand source code
def transfer_clinical_data(self, source_episode=None, target_episode=None):
        """
        Moves every element currently linked to the current encounter
        and the source_episode onto target_episode.

        @param source_episode The episode the elements are currently linked to.
        @type target_episode A cEpisode intance.
        @param target_episode The episode the elements will be relinked to.
        @type target_episode A cEpisode intance.
        """
        if source_episode['pk_episode'] == target_episode['pk_episode']:
                return True

        cmd = """
                UPDATE clin.clin_root_item
                SET fk_episode = %(trg)s
                WHERE
                        fk_encounter = %(enc)s AND
                        fk_episode = %(src)s
                """
        gmPG2.run_rw_queries(queries = [{
                'cmd': cmd,
                'args': {
                        'trg': target_episode['pk_episode'],
                        'enc': self.pk_obj,
                        'src': source_episode['pk_episode']
                }
        }])
        self.refetch_payload()
        return True
def unlock(self, exclusive=False, link_obj=None)
Expand source code
def unlock(self, exclusive=False, link_obj=None):
        return unlock_encounter(self.pk_obj, exclusive = exclusive, link_obj = link_obj)

Inherited members

class cEpisode (aPK_obj=None, id_patient=None, name='xxxDEFAULTxxx', health_issue=None, row=None, encounter=None, link_obj=None)

Represents one clinical episode.

Call init from child classes like so:

    super().__init__(aPK_obj = aPK_obj, row = row, link_obj = link_obj)

Args

aPK_obj
retrieve data from backend
  • an scalar value the ._cmd_fetch_payload WHERE condition must be a simple column: "… WHERE pk_col = %s"
  • a dictionary of values the ._cmd_fetch_payload WHERE condition must consume the dictionary and produce a unique row
row
must hold the fields
  • data: list of column values for the row selected by ._cmd_fetch_payload (as returned by cursor.fetchone() in the DB-API)
  • pk_field: the name of the primary key column OR
  • pk_obj: a dictionary suitable for being passed to cursor.execute and holding the primary key values, used for composite PKs
  • for example:
    row = {
            'data': rows[0],
            'pk_field': 'pk_XXX (the PK column name)',
            'pk_obj': {'pk_col1': pk_col1_val, 'pk_col2': pk_col2_val}
    }
    
    rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
    objects = [ cChildClass(row = {'data': r, 'pk_field': 'the PK column name'}) for r in rows ]
    
Expand source code
class cEpisode(gmBusinessDBObject.cBusinessDBObject):
        """Represents one clinical episode.
        """
        _cmd_fetch_payload = "select * from clin.v_pat_episodes where pk_episode=%s"
        _cmds_store_payload = [
                """update clin.episode set
                                fk_health_issue = %(pk_health_issue)s,
                                is_open = %(episode_open)s::boolean,
                                description = %(description)s,
                                summary = gm.nullify_empty_string(%(summary)s),
                                diagnostic_certainty_classification = gm.nullify_empty_string(%(diagnostic_certainty_classification)s)
                        where
                                pk = %(pk_episode)s and
                                xmin = %(xmin_episode)s""",
                """select xmin_episode from clin.v_pat_episodes where pk_episode = %(pk_episode)s"""
        ]
        _updatable_fields = [
                'pk_health_issue',
                'episode_open',
                'description',
                'summary',
                'diagnostic_certainty_classification'
        ]
        #--------------------------------------------------------
        def __init__(self, aPK_obj=None, id_patient=None, name='xxxDEFAULTxxx', health_issue=None, row=None, encounter=None, link_obj=None):
                pk = aPK_obj
                if pk is None and row is None:

                        where_parts = ['description = %(desc)s']

                        if id_patient is not None:
                                where_parts.append('pk_patient = %(pat)s')

                        if health_issue is not None:
                                where_parts.append('pk_health_issue = %(issue)s')

                        if encounter is not None:
                                where_parts.append('pk_patient = (SELECT fk_patient FROM clin.encounter WHERE pk = %(enc)s)')

                        args = {
                                'pat': id_patient,
                                'issue': health_issue,
                                'enc': encounter,
                                'desc': name
                        }

                        cmd = 'SELECT * FROM clin.v_pat_episodes WHERE %s' % ' AND '.join(where_parts)

                        rows = gmPG2.run_ro_queries (
                                link_obj = link_obj,
                                queries = [{'cmd': cmd, 'args': args}],
                        )

                        if len(rows) == 0:
                                raise gmExceptions.NoSuchBusinessObjectError('no episode for [%s:%s:%s:%s]' % (id_patient, name, health_issue, encounter))

                        r = {'data': rows[0], 'pk_field': 'pk_episode'}
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, row=r)

                else:
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj=pk, row=row, link_obj = link_obj)

        #--------------------------------------------------------
        # external API
        #--------------------------------------------------------
        @classmethod
        def from_problem(cls, problem:'cProblem') -> 'cEpisode':
                """Initialize episode from episody-type problem."""
                if isinstance(problem, cEpisode):
                        return problem

                assert isinstance(problem, cProblem), 'cannot convert [%s] to episode' % problem
                assert problem['type'] == 'episode', 'cannot convert [%s] to episode' % problem
                return cls(aPK_obj = problem['pk_episode'])

        #--------------------------------------------------------
        def get_patient(self):
                return self._payload['pk_patient']

        #--------------------------------------------------------
        def get_narrative(self, soap_cats=None, encounters=None, order_by = None):
                return gmClinNarrative.get_narrative (
                        soap_cats = soap_cats,
                        encounters = encounters,
                        episodes = [self.pk_obj],
                        order_by = order_by
                )

        #--------------------------------------------------------
        def rename(self, description=None):
                """Method for episode editing, that is, episode renaming.

                @param description
                        - the new descriptive name for the encounter
                @type description
                        - a string instance
                """
                # sanity check
                if description.strip() == '':
                        _log.error('<description> must be a non-empty string instance')
                        return False
                # update the episode description
                old_description = self._payload['description']
                self._payload['description'] = description.strip()
                self._is_modified = True
                successful, data = self.save_payload()
                if not successful:
                        _log.error('cannot rename episode [%s] to [%s]' % (self, description))
                        self._payload['description'] = old_description
                        return False
                return True

        #--------------------------------------------------------
        def add_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""

                if pk_code in self._payload['pk_generic_codes']:
                        return

                cmd = """
                        INSERT INTO clin.lnk_code2episode
                                (fk_item, fk_generic_code)
                        SELECT
                                %(item)s,
                                %(code)s
                        WHERE NOT EXISTS (
                                SELECT 1 FROM clin.lnk_code2episode
                                WHERE
                                        fk_item = %(item)s
                                                AND
                                        fk_generic_code = %(code)s
                        )"""
                args = {
                        'item': self._payload['pk_episode'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return

        #--------------------------------------------------------
        def remove_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "DELETE FROM clin.lnk_code2episode WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                args = {
                        'item': self._payload['pk_episode'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a'):
                rows = gmClinNarrative.get_as_journal (
                        episodes = [self.pk_obj],
                        order_by = 'pk_encounter, clin_when, scr, src_table'
                        #order_by = u'pk_encounter, scr, clin_when, src_table'
                )

                if len(rows) == 0:
                        return ''

                lines = []

                lines.append(_('Clinical data generated during encounters within this episode:'))

                left_margin = ' ' * left_margin

                prev_enc = None
                for row in rows:
                        if row['pk_encounter'] != prev_enc:
                                lines.append('')
                                prev_enc = row['pk_encounter']

                        when = row['clin_when'].strftime(date_format)
                        top_row = '%s%s %s (%s) %s' % (
                                gmTools.u_box_top_left_arc,
                                gmTools.u_box_horiz_single,
                                gmSoapDefs.soap_cat2l10n_str[row['real_soap_cat']],
                                when,
                                gmTools.u_box_horiz_single * 5
                        )
                        soap = gmTools.wrap (
                                text = row['narrative'],
                                width = 60,
                                initial_indent = '  ',
                                subsequent_indent = '  ' + left_margin
                        )
                        row_ver = ''
                        if row['row_version'] > 0:
                                row_ver = 'v%s: ' % row['row_version']
                        bottom_row = '%s%s %s, %s%s %s' % (
                                ' ' * 40,
                                gmTools.u_box_horiz_light_heavy,
                                row['modified_by'],
                                row_ver,
                                gmDateTime.pydt_strftime(row['modified_when'], date_format),
                                gmTools.u_box_horiz_heavy_light
                        )

                        lines.append(top_row)
                        lines.append(soap)
                        lines.append(bottom_row)

                eol_w_margin = '\n%s' % left_margin
                return left_margin + eol_w_margin.join(lines) + '\n'

        #--------------------------------------------------------
        def format_maximum_information(self, patient=None):
                if patient is None:
                        from Gnumed.business.gmPerson import gmCurrentPatient, cPerson
                        if self._payload['pk_patient'] == gmCurrentPatient().ID:
                                patient = gmCurrentPatient()
                        else:
                                patient = cPerson(self._payload['pk_patient'])

                return self.format (
                        patient = patient,
                        with_summary = True,
                        with_codes = True,
                        with_encounters = True,
                        with_documents = True,
                        with_hospital_stays = True,
                        with_procedures = True,
                        with_family_history = True,
                        with_tests = False,             # does not inform on the episode itself
                        with_vaccinations = True,
                        with_health_issue = True,
                        return_list = True
                )

        #--------------------------------------------------------
        def format(self, left_margin=0, patient=None,
                with_summary=True,
                with_codes=True,
                with_encounters=True,
                with_documents=True,
                with_hospital_stays=True,
                with_procedures=True,
                with_family_history=True,
                with_tests=True,
                with_vaccinations=True,
                with_health_issue=False,
                return_list=False
        ):

                if patient is not None:
                        if patient.ID != self._payload['pk_patient']:
                                msg = '<patient>.ID = %s but episode %s belongs to patient %s' % (
                                        patient.ID,
                                        self._payload['pk_episode'],
                                        self._payload['pk_patient']
                                )
                                raise ValueError(msg)
                        emr = patient.emr
                else:
                        with_encounters = False
                        with_documents = False
                        with_hospital_stays = False
                        with_procedures = False
                        with_family_history = False
                        with_tests = False
                        with_vaccinations = False

                lines = []

                # episode details
                lines.append (_('Episode %s%s%s   [#%s]') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['description'],
                        gmTools.u_right_double_angle_quote,
                        self._payload['pk_episode']
                ))

                enc = cEncounter(aPK_obj = self._payload['pk_encounter'])
                lines.append (' ' + _('Created during encounter: %s (%s - %s)   [#%s]') % (
                        enc['l10n_type'],
                        enc['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                        enc['last_affirmed_original_tz'].strftime('%H:%M'),
                        self._payload['pk_encounter']
                ))

                if patient is not None:
                        range_str, range_str_verb, duration_str = self.formatted_clinical_duration
                        lines.append(_(' Duration: %s (%s)') % (duration_str, range_str_verb))

                lines.append(' ' + _('Status') + ': %s%s' % (
                        gmTools.bool2subst(self._payload['episode_open'], _('active'), _('finished')),
                        gmTools.coalesce (
                                value2test = diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification']),
                                return_instead = '',
                                template4value = ', %s',
                                none_equivalents = [None, '']
                        )
                ))

                if with_health_issue:
                        lines.append(' ' + _('Health issue') + ': %s' % gmTools.coalesce (
                                self._payload['health_issue'],
                                _('none associated')
                        ))

                if with_summary:
                        if self._payload['summary'] is not None:
                                lines.append(' %s:' % _('Synopsis'))
                                lines.append(gmTools.wrap (
                                                text = self._payload['summary'],
                                                width = 60,
                                                initial_indent = '  ',
                                                subsequent_indent = '  '
                                        )
                                )

                # codes
                if with_codes:
                        codes = self.generic_codes
                        if len(codes) > 0:
                                lines.append('')
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        del codes

                lines.append('')

                # encounters
                if with_encounters:
                        encs = emr.get_encounters(episodes = [self._payload['pk_episode']])
                        if encs is None:
                                lines.append(_('Error retrieving encounters for this episode.'))
                        elif len(encs) == 0:
                                #lines.append(_('There are no encounters for this issue.'))
                                pass
                        else:
                                first_encounter = emr.get_first_encounter(episode_id = self._payload['pk_episode'])
                                last_encounter = emr.get_last_encounter(episode_id = self._payload['pk_episode'])
                                lines.append(_('Last worked on: %s\n') % last_encounter['last_affirmed_original_tz'].strftime('%Y-%m-%d %H:%M'))
                                if len(encs) < 4:
                                        line = _('%s encounter(s) (%s - %s):')
                                else:
                                        line = _('1st and (up to 3) most recent (of %s) encounters (%s - %s):')
                                lines.append(line % (
                                        len(encs),
                                        first_encounter['started'].strftime('%m/%Y'),
                                        last_encounter['last_affirmed'].strftime('%m/%Y')
                                ))
                                lines.append(' %s - %s (%s):%s' % (
                                        first_encounter['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                                        first_encounter['last_affirmed_original_tz'].strftime('%H:%M'),
                                        first_encounter['l10n_type'],
                                        gmTools.coalesce (
                                                first_encounter['assessment_of_encounter'],
                                                gmTools.coalesce (
                                                        first_encounter['reason_for_encounter'],
                                                        '',
                                                        ' \u00BB%s\u00AB' + (' (%s)' % _('RFE'))
                                                ),
                                                ' \u00BB%s\u00AB' + (' (%s)' % _('AOE'))
                                        )
                                ))
                                if len(encs) > 4:
                                        lines.append(_(' %s %s skipped %s') % (
                                                gmTools.u_ellipsis,
                                                (len(encs) - 4),
                                                gmTools.u_ellipsis
                                        ))
                                for enc in encs[1:][-3:]:
                                        lines.append(' %s - %s (%s):%s' % (
                                                enc['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                                                enc['last_affirmed_original_tz'].strftime('%H:%M'),
                                                enc['l10n_type'],
                                                gmTools.coalesce (
                                                        enc['assessment_of_encounter'],
                                                        gmTools.coalesce (
                                                                enc['reason_for_encounter'],
                                                                '',
                                                                ' \u00BB%s\u00AB' + (' (%s)' % _('RFE'))
                                                        ),
                                                        ' \u00BB%s\u00AB' + (' (%s)' % _('AOE'))
                                                )
                                        ))
                                del encs
                                # spell out last encounter
                                if last_encounter is not None:
                                        lines.append('')
                                        lines.append(_('Progress notes in most recent encounter:'))
                                        lines.extend(last_encounter.format_soap (
                                                episodes = [ self._payload['pk_episode'] ],
                                                left_margin = left_margin,
                                                soap_cats = 'soapu',
                                                emr = emr
                                        ))

                # documents
                if with_documents:
                        doc_folder = patient.get_document_folder()
                        docs = doc_folder.get_documents (
                                pk_episodes = [ self._payload['pk_episode'] ]
                        )
                        if len(docs) > 0:
                                lines.append('')
                                lines.append(_('Documents: %s') % len(docs))
                        for d in docs:
                                lines.append(' ' + d.format(single_line = True))
                        del docs

                # hospitalizations
                if with_hospital_stays:
                        stays = emr.get_hospital_stays(episodes = [ self._payload['pk_episode'] ])
                        if len(stays) > 0:
                                lines.append('')
                                lines.append(_('Hospitalizations: %s') % len(stays))
                        for s in stays:
                                lines.append(s.format(left_margin = (left_margin + 1)))
                        del stays

                # procedures
                if with_procedures:
                        procs = emr.get_performed_procedures(episodes = [ self._payload['pk_episode'] ])
                        if len(procs) > 0:
                                lines.append('')
                                lines.append(_('Procedures performed: %s') % len(procs))
                        for p in procs:
                                lines.append(p.format (
                                        left_margin = (left_margin + 1),
                                        include_episode = False,
                                        include_codes = True
                                ))
                        del procs

                # family history
                if with_family_history:
                        fhx = emr.get_family_history(episodes = [ self._payload['pk_episode'] ])
                        if len(fhx) > 0:
                                lines.append('')
                                lines.append(_('Family History: %s') % len(fhx))
                        for f in fhx:
                                lines.append(f.format (
                                        left_margin = (left_margin + 1),
                                        include_episode = False,
                                        include_comment = True,
                                        include_codes = True
                                ))
                        del fhx

                # test results
                if with_tests:
                        tests = emr.get_test_results_by_date(episodes = [ self._payload['pk_episode'] ])
                        if len(tests) > 0:
                                lines.append('')
                                lines.append(_('Measurements and Results:'))
                        for t in tests:
                                lines.append(' ' + t.format_concisely(date_format = '%Y %b %d', with_notes = True))
                        del tests

                # vaccinations
                if with_vaccinations:
                        vaccs = emr.get_vaccinations (
                                episodes = [ self._payload['pk_episode'] ],
                                order_by = 'date_given DESC, vaccine'
                        )
                        if len(vaccs) > 0:
                                lines.append('')
                                lines.append(_('Vaccinations:'))
                        for vacc in vaccs:
                                lines.extend(vacc.format (
                                        with_indications = True,
                                        with_comment = True,
                                        with_reaction = True,
                                        date_format = '%Y-%m-%d'
                                ))
                        del vaccs

                lines = gmTools.strip_trailing_empty_lines(lines = lines, eol = '\n')
                if return_list:
                        return lines

                left_margin = ' ' * left_margin
                eol_w_margin = '\n%s' % left_margin
                return left_margin + eol_w_margin.join(lines) + '\n'

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_best_guess_clinical_start_date(self):
                return get_best_guess_clinical_start_date_for_episode(pk_episode = self.pk_obj)

        best_guess_clinical_start_date = property(_get_best_guess_clinical_start_date)

        #--------------------------------------------------------
        def _get_best_guess_clinical_end_date(self):
                return get_best_guess_clinical_end_date_for_episode(self.pk_obj)

        best_guess_clinical_end_date = property(_get_best_guess_clinical_end_date)

        #--------------------------------------------------------
        def _get_formatted_clinical_duration(self):
                return format_clinical_duration_of_episode (
                        start = get_best_guess_clinical_start_date_for_episode(self.pk_obj),
                        end = get_best_guess_clinical_end_date_for_episode(self.pk_obj)
                )

        formatted_clinical_duration = property(_get_formatted_clinical_duration)

        #--------------------------------------------------------
        def _get_latest_access_date(self):
                cmd = """SELECT MAX(latest) FROM (
                        -- last modification, latest = when last changed to the current state
                        (SELECT c_epi.modified_when AS latest, 'clin.episode.modified_when'::text AS candidate FROM clin.episode c_epi WHERE c_epi.pk = %(pk)s)

                                UNION ALL

                        -- last modification of encounter in which created, latest = initial creation of that encounter
                        -- DO NOT USE: just because one corrects a typo does not mean the episode took longer
                        --(SELECT c_enc.modified_when AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
                        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                        --))

                        -- end of encounter in which created, latest = explicitly set
                        -- DO NOT USE: we can retrospectively create episodes which
                        -- DO NOT USE: are long since finished
                        --(SELECT c_enc.last_affirmed AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
                        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                        --))

                        -- latest end of encounters of clinical items linked to this episode
                        (SELECT
                                MAX(last_affirmed) AS latest,
                                'clin.episode.pk = clin.clin_root_item,fk_episode -> .fk_encounter.last_affirmed'::text AS candidate
                         FROM clin.encounter
                         WHERE pk IN (
                                SELECT fk_encounter FROM clin.clin_root_item WHERE fk_episode = %(pk)s
                        ))
                                UNION ALL

                        -- latest explicit .clin_when of clinical items linked to this episode
                        (SELECT
                                MAX(clin_when) AS latest,
                                'clin.episode.pk = clin.clin_root_item,fk_episode -> .clin_when'::text AS candidate
                         FROM clin.clin_root_item
                         WHERE fk_episode = %(pk)s
                        )

                        -- latest modification time of clinical items linked to this episode
                        -- this CAN be used since if an item is linked to an episode it can be
                        -- assumed the episode (should have) existed at the time of creation
                        -- DO NOT USE, because typo fixes should not extend the episode
                        --(SELECT MIN(modified_when) AS latest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

                        -- not sure about this one:
                        -- .pk -> clin.clin_root_item.fk_encounter.modified_when

                ) AS candidates"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return rows[0][0]

        latest_access_date = property(_get_latest_access_date)

        #--------------------------------------------------------
        def _get_diagnostic_certainty_description(self):
                return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])

        diagnostic_certainty_description = property(_get_diagnostic_certainty_description)

        #--------------------------------------------------------
        def _get_formatted_revision_history(self):
                cmd = """SELECT
                                '<N/A>'::TEXT as audit__action_applied,
                                NULL AS audit__action_when,
                                '<N/A>'::TEXT AS audit__action_by,
                                pk_audit,
                                row_version,
                                modified_when,
                                modified_by,
                                pk, fk_health_issue, description, is_open, fk_encounter,
                                diagnostic_certainty_classification,
                                summary
                        FROM clin.episode
                        WHERE pk = %(pk_episode)s
                UNION ALL (
                        SELECT
                                audit_action as audit__action_applied,
                                audit_when as audit__action_when,
                                audit_by as audit__action_by,
                                pk_audit,
                                orig_version as row_version,
                                orig_when as modified_when,
                                orig_by as modified_by,
                                pk, fk_health_issue, description, is_open, fk_encounter,
                                diagnostic_certainty_classification,
                                summary
                        FROM audit.log_episode
                        WHERE pk = %(pk_episode)s
                )
                ORDER BY row_version DESC
                """
                args = {'pk_episode': self.pk_obj}
                title = _('Episode: %s%s%s') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['description'],
                        gmTools.u_right_double_angle_quote
                )
                return '\n'.join(self._get_revision_history(cmd, args, title))

        formatted_revision_history = property(_get_formatted_revision_history)

        #--------------------------------------------------------
        def _get_generic_codes(self):
                if len(self._payload['pk_generic_codes']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2episode WHERE fk_item = %(epi)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'epi': self._payload['pk_episode'],
                                        'codes': self._payload['pk_generic_codes']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2episode (fk_item, fk_generic_code) VALUES (%(epi)s, %(pk_code)s)',
                                'args': {
                                        'epi': self._payload['pk_episode'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                return

        generic_codes = property(_get_generic_codes, _set_generic_codes)

        #--------------------------------------------------------
        def _get_has_narrative(self):
                cmd = """SELECT EXISTS (
                        SELECT 1 FROM clin.clin_narrative
                        WHERE
                                fk_episode = %(epi)s
                                        AND
                                fk_encounter IN (
                                        SELECT pk FROM clin.encounter WHERE fk_patient = %(pat)s
                                )
                )"""
                args = {
                        'pat': self._payload['pk_patient'],
                        'epi': self._payload['pk_episode']
                }
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return rows[0][0]

        has_narrative = property(_get_has_narrative)

        #--------------------------------------------------------
        def _get_health_issue(self):
                if self._payload['pk_health_issue'] is None:
                        return None
                return cHealthIssue(self._payload['pk_health_issue'])

        health_issue = property(_get_health_issue)

Ancestors

Static methods

def from_problem(problem: cProblem) ‑> cEpisode

Initialize episode from episody-type problem.

Expand source code
@classmethod
def from_problem(cls, problem:'cProblem') -> 'cEpisode':
        """Initialize episode from episody-type problem."""
        if isinstance(problem, cEpisode):
                return problem

        assert isinstance(problem, cProblem), 'cannot convert [%s] to episode' % problem
        assert problem['type'] == 'episode', 'cannot convert [%s] to episode' % problem
        return cls(aPK_obj = problem['pk_episode'])

Instance variables

var best_guess_clinical_end_date
Expand source code
def _get_best_guess_clinical_end_date(self):
        return get_best_guess_clinical_end_date_for_episode(self.pk_obj)
var best_guess_clinical_start_date
Expand source code
def _get_best_guess_clinical_start_date(self):
        return get_best_guess_clinical_start_date_for_episode(pk_episode = self.pk_obj)
var diagnostic_certainty_description
Expand source code
def _get_diagnostic_certainty_description(self):
        return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])
var formatted_clinical_duration
Expand source code
def _get_formatted_clinical_duration(self):
        return format_clinical_duration_of_episode (
                start = get_best_guess_clinical_start_date_for_episode(self.pk_obj),
                end = get_best_guess_clinical_end_date_for_episode(self.pk_obj)
        )
var formatted_revision_history
Expand source code
def _get_formatted_revision_history(self):
        cmd = """SELECT
                        '<N/A>'::TEXT as audit__action_applied,
                        NULL AS audit__action_when,
                        '<N/A>'::TEXT AS audit__action_by,
                        pk_audit,
                        row_version,
                        modified_when,
                        modified_by,
                        pk, fk_health_issue, description, is_open, fk_encounter,
                        diagnostic_certainty_classification,
                        summary
                FROM clin.episode
                WHERE pk = %(pk_episode)s
        UNION ALL (
                SELECT
                        audit_action as audit__action_applied,
                        audit_when as audit__action_when,
                        audit_by as audit__action_by,
                        pk_audit,
                        orig_version as row_version,
                        orig_when as modified_when,
                        orig_by as modified_by,
                        pk, fk_health_issue, description, is_open, fk_encounter,
                        diagnostic_certainty_classification,
                        summary
                FROM audit.log_episode
                WHERE pk = %(pk_episode)s
        )
        ORDER BY row_version DESC
        """
        args = {'pk_episode': self.pk_obj}
        title = _('Episode: %s%s%s') % (
                gmTools.u_left_double_angle_quote,
                self._payload['description'],
                gmTools.u_right_double_angle_quote
        )
        return '\n'.join(self._get_revision_history(cmd, args, title))
var generic_codes
Expand source code
def _get_generic_codes(self):
        if len(self._payload['pk_generic_codes']) == 0:
                return []

        cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
        args = {'pks': self._payload['pk_generic_codes']}
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]
var has_narrative
Expand source code
def _get_has_narrative(self):
        cmd = """SELECT EXISTS (
                SELECT 1 FROM clin.clin_narrative
                WHERE
                        fk_episode = %(epi)s
                                AND
                        fk_encounter IN (
                                SELECT pk FROM clin.encounter WHERE fk_patient = %(pat)s
                        )
        )"""
        args = {
                'pat': self._payload['pk_patient'],
                'epi': self._payload['pk_episode']
        }
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return rows[0][0]
var health_issue
Expand source code
def _get_health_issue(self):
        if self._payload['pk_health_issue'] is None:
                return None
        return cHealthIssue(self._payload['pk_health_issue'])
var latest_access_date
Expand source code
def _get_latest_access_date(self):
        cmd = """SELECT MAX(latest) FROM (
                -- last modification, latest = when last changed to the current state
                (SELECT c_epi.modified_when AS latest, 'clin.episode.modified_when'::text AS candidate FROM clin.episode c_epi WHERE c_epi.pk = %(pk)s)

                        UNION ALL

                -- last modification of encounter in which created, latest = initial creation of that encounter
                -- DO NOT USE: just because one corrects a typo does not mean the episode took longer
                --(SELECT c_enc.modified_when AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
                --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                --))

                -- end of encounter in which created, latest = explicitly set
                -- DO NOT USE: we can retrospectively create episodes which
                -- DO NOT USE: are long since finished
                --(SELECT c_enc.last_affirmed AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
                --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
                --))

                -- latest end of encounters of clinical items linked to this episode
                (SELECT
                        MAX(last_affirmed) AS latest,
                        'clin.episode.pk = clin.clin_root_item,fk_episode -> .fk_encounter.last_affirmed'::text AS candidate
                 FROM clin.encounter
                 WHERE pk IN (
                        SELECT fk_encounter FROM clin.clin_root_item WHERE fk_episode = %(pk)s
                ))
                        UNION ALL

                -- latest explicit .clin_when of clinical items linked to this episode
                (SELECT
                        MAX(clin_when) AS latest,
                        'clin.episode.pk = clin.clin_root_item,fk_episode -> .clin_when'::text AS candidate
                 FROM clin.clin_root_item
                 WHERE fk_episode = %(pk)s
                )

                -- latest modification time of clinical items linked to this episode
                -- this CAN be used since if an item is linked to an episode it can be
                -- assumed the episode (should have) existed at the time of creation
                -- DO NOT USE, because typo fixes should not extend the episode
                --(SELECT MIN(modified_when) AS latest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

                -- not sure about this one:
                -- .pk -> clin.clin_root_item.fk_encounter.modified_when

        ) AS candidates"""
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
        return rows[0][0]

Methods

def add_code(self, pk_code=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def add_code(self, pk_code=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""

        if pk_code in self._payload['pk_generic_codes']:
                return

        cmd = """
                INSERT INTO clin.lnk_code2episode
                        (fk_item, fk_generic_code)
                SELECT
                        %(item)s,
                        %(code)s
                WHERE NOT EXISTS (
                        SELECT 1 FROM clin.lnk_code2episode
                        WHERE
                                fk_item = %(item)s
                                        AND
                                fk_generic_code = %(code)s
                )"""
        args = {
                'item': self._payload['pk_episode'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return
def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a')
Expand source code
def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a'):
        rows = gmClinNarrative.get_as_journal (
                episodes = [self.pk_obj],
                order_by = 'pk_encounter, clin_when, scr, src_table'
                #order_by = u'pk_encounter, scr, clin_when, src_table'
        )

        if len(rows) == 0:
                return ''

        lines = []

        lines.append(_('Clinical data generated during encounters within this episode:'))

        left_margin = ' ' * left_margin

        prev_enc = None
        for row in rows:
                if row['pk_encounter'] != prev_enc:
                        lines.append('')
                        prev_enc = row['pk_encounter']

                when = row['clin_when'].strftime(date_format)
                top_row = '%s%s %s (%s) %s' % (
                        gmTools.u_box_top_left_arc,
                        gmTools.u_box_horiz_single,
                        gmSoapDefs.soap_cat2l10n_str[row['real_soap_cat']],
                        when,
                        gmTools.u_box_horiz_single * 5
                )
                soap = gmTools.wrap (
                        text = row['narrative'],
                        width = 60,
                        initial_indent = '  ',
                        subsequent_indent = '  ' + left_margin
                )
                row_ver = ''
                if row['row_version'] > 0:
                        row_ver = 'v%s: ' % row['row_version']
                bottom_row = '%s%s %s, %s%s %s' % (
                        ' ' * 40,
                        gmTools.u_box_horiz_light_heavy,
                        row['modified_by'],
                        row_ver,
                        gmDateTime.pydt_strftime(row['modified_when'], date_format),
                        gmTools.u_box_horiz_heavy_light
                )

                lines.append(top_row)
                lines.append(soap)
                lines.append(bottom_row)

        eol_w_margin = '\n%s' % left_margin
        return left_margin + eol_w_margin.join(lines) + '\n'
def format_maximum_information(self, patient=None)
Expand source code
def format_maximum_information(self, patient=None):
        if patient is None:
                from Gnumed.business.gmPerson import gmCurrentPatient, cPerson
                if self._payload['pk_patient'] == gmCurrentPatient().ID:
                        patient = gmCurrentPatient()
                else:
                        patient = cPerson(self._payload['pk_patient'])

        return self.format (
                patient = patient,
                with_summary = True,
                with_codes = True,
                with_encounters = True,
                with_documents = True,
                with_hospital_stays = True,
                with_procedures = True,
                with_family_history = True,
                with_tests = False,             # does not inform on the episode itself
                with_vaccinations = True,
                with_health_issue = True,
                return_list = True
        )
def get_narrative(self, soap_cats=None, encounters=None, order_by=None)
Expand source code
def get_narrative(self, soap_cats=None, encounters=None, order_by = None):
        return gmClinNarrative.get_narrative (
                soap_cats = soap_cats,
                encounters = encounters,
                episodes = [self.pk_obj],
                order_by = order_by
        )
def remove_code(self, pk_code=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def remove_code(self, pk_code=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        cmd = "DELETE FROM clin.lnk_code2episode WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
        args = {
                'item': self._payload['pk_episode'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def rename(self, description=None)

Method for episode editing, that is, episode renaming.

@param description - the new descriptive name for the encounter @type description - a string instance

Expand source code
def rename(self, description=None):
        """Method for episode editing, that is, episode renaming.

        @param description
                - the new descriptive name for the encounter
        @type description
                - a string instance
        """
        # sanity check
        if description.strip() == '':
                _log.error('<description> must be a non-empty string instance')
                return False
        # update the episode description
        old_description = self._payload['description']
        self._payload['description'] = description.strip()
        self._is_modified = True
        successful, data = self.save_payload()
        if not successful:
                _log.error('cannot rename episode [%s] to [%s]' % (self, description))
                self._payload['description'] = old_description
                return False
        return True

Inherited members

class cEpisodeMatchProvider

Find episodes for patient.

Expand source code
class cEpisodeMatchProvider(gmMatchProvider.cMatchProvider_SQL2):
        """Find episodes for patient."""

        _SQL_episode_start = _SQL_best_guess_clinical_start_date_for_episode % {'pk': 'c_vpe.pk_episode'}
        _SQL_episode_end = _SQL_best_guess_clinical_end_date_for_episode % {'pk': 'c_vpe.pk_episode'}

        _SQL_open_episodes = """
                SELECT
                        c_vpe.pk_episode,
                        c_vpe.description
                                AS episode,
                        c_vpe.health_issue,
                        1 AS rank,
                        (%s) AS episode_start,
                        NULL::timestamp with time zone AS episode_end
                FROM
                        clin.v_pat_episodes c_vpe
                WHERE
                        c_vpe.episode_open IS TRUE
                                AND
                        c_vpe.description %%(fragment_condition)s
                        %%(ctxt_pat)s
        """ % _SQL_episode_start

        _SQL_closed_episodes = """
                SELECT
                        c_vpe.pk_episode,
                        c_vpe.description
                                AS episode,
                        c_vpe.health_issue,
                        2 AS rank,
                        (%s) AS episode_start,
                        (%s) AS episode_end
                FROM
                        clin.v_pat_episodes c_vpe
                WHERE
                        c_vpe.episode_open IS FALSE
                                AND
                        c_vpe.description %%(fragment_condition)s
                        %%(ctxt_pat)s
        """ % (
                _SQL_episode_start,
                _SQL_episode_end
        )

        #--------------------------------------------------------
        def __init__(self):
                query = """
                        (
                                %s
                        ) UNION ALL (
                                %s
                        )
                        ORDER BY rank, episode
                        LIMIT 30""" % (
                                cEpisodeMatchProvider._SQL_open_episodes,
                                cEpisodeMatchProvider._SQL_closed_episodes
                        )
                ctxt = {'ctxt_pat': {'where_part': 'AND pk_patient = %(pat)s', 'placeholder': 'pat'}}
                super().__init__(queries = [query], context = ctxt)

        #--------------------------------------------------------
        def _find_matches(self, fragment_condition):
                try:
                        pat = self._context_vals['pat']
                except KeyError:
                        pat = None
                if not pat:
                        # not patient, no search
                        return (False, [])

                return super()._find_matches(fragment_condition)

        #--------------------------------------------------------
        def _rows2matches(self, rows):
                matches = []
                for row in rows:
                        match = {
                                'weight': 0,
                                'data': row['pk_episode']
                        }
                        label = '%s (%s)%s' % (
                                row['episode'],
                                format_clinical_duration_of_episode (
                                        start = row['episode_start'],
                                        end = row['episode_end']
                                )[0],
                                gmTools.coalesce (
                                        row['health_issue'],
                                        '',
                                        ' - %s'
                                )
                        )
                        match['list_label'] = label
                        match['field_label'] = label
                        matches.append(match)
                return matches

Ancestors

Inherited members

class cHealthIssue (aPK_obj=None, encounter=None, name='xxxDEFAULTxxx', patient=None, row=None)

Represents one health issue.

Call init from child classes like so:

    super().__init__(aPK_obj = aPK_obj, row = row, link_obj = link_obj)

Args

aPK_obj
retrieve data from backend
  • an scalar value the ._cmd_fetch_payload WHERE condition must be a simple column: "… WHERE pk_col = %s"
  • a dictionary of values the ._cmd_fetch_payload WHERE condition must consume the dictionary and produce a unique row
row
must hold the fields
  • data: list of column values for the row selected by ._cmd_fetch_payload (as returned by cursor.fetchone() in the DB-API)
  • pk_field: the name of the primary key column OR
  • pk_obj: a dictionary suitable for being passed to cursor.execute and holding the primary key values, used for composite PKs
  • for example:
    row = {
            'data': rows[0],
            'pk_field': 'pk_XXX (the PK column name)',
            'pk_obj': {'pk_col1': pk_col1_val, 'pk_col2': pk_col2_val}
    }
    
    rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
    objects = [ cChildClass(row = {'data': r, 'pk_field': 'the PK column name'}) for r in rows ]
    
Expand source code
class cHealthIssue(gmBusinessDBObject.cBusinessDBObject):
        """Represents one health issue."""

        #_cmd_fetch_payload = u"select *, xmin_health_issue from clin.v_health_issues where pk_health_issue=%s"
        _cmd_fetch_payload = "select * from clin.v_health_issues where pk_health_issue = %s"
        _cmds_store_payload = [
                """update clin.health_issue set
                                description = %(description)s,
                                summary = gm.nullify_empty_string(%(summary)s),
                                age_noted = %(age_noted)s,
                                laterality = gm.nullify_empty_string(%(laterality)s),
                                grouping = gm.nullify_empty_string(%(grouping)s),
                                diagnostic_certainty_classification = gm.nullify_empty_string(%(diagnostic_certainty_classification)s),
                                is_active = %(is_active)s,
                                clinically_relevant = %(clinically_relevant)s,
                                is_confidential = %(is_confidential)s,
                                is_cause_of_death = %(is_cause_of_death)s
                        WHERE
                                pk = %(pk_health_issue)s
                                        AND
                                xmin = %(xmin_health_issue)s""",
                "select xmin as xmin_health_issue from clin.health_issue where pk = %(pk_health_issue)s"
        ]
        _updatable_fields = [
                'description',
                'summary',
                'grouping',
                'age_noted',
                'laterality',
                'is_active',
                'clinically_relevant',
                'is_confidential',
                'is_cause_of_death',
                'diagnostic_certainty_classification'
        ]

        #--------------------------------------------------------
        def __init__(self, aPK_obj=None, encounter=None, name='xxxDEFAULTxxx', patient=None, row=None):
                pk = aPK_obj

                if (pk is not None) or (row is not None):
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj=pk, row=row)
                        return

                if patient is None:
                        cmd = """select *, xmin_health_issue from clin.v_health_issues
                                        where
                                                description = %(desc)s
                                                        and
                                                pk_patient = (select fk_patient from clin.encounter where pk = %(enc)s)"""
                else:
                        cmd = """select *, xmin_health_issue from clin.v_health_issues
                                        where
                                                description = %(desc)s
                                                        and
                                                pk_patient = %(pat)s"""

                queries = [{'cmd': cmd, 'args': {'enc': encounter, 'desc': name, 'pat': patient}}]
                rows = gmPG2.run_ro_queries(queries = queries)

                if len(rows) == 0:
                        raise gmExceptions.NoSuchBusinessObjectError('no health issue for [enc:%s::desc:%s::pat:%s]' % (encounter, name, patient))

                pk = rows[0][0]
                r = {'data': rows[0], 'pk_field': 'pk_health_issue'}

                gmBusinessDBObject.cBusinessDBObject.__init__(self, row=r)

        #--------------------------------------------------------
        # external API
        #--------------------------------------------------------
        @classmethod
        def from_problem(cls, problem:'cProblem') -> 'cHealthIssue':
                """Initialize health issue from issue-type problem."""
                if isinstance(problem, cHealthIssue):
                        return problem

                assert isinstance(problem, cProblem), 'cannot convert [%s] to health issue' % problem
                assert problem['type'] == 'issue', 'cannot convert [%s] to health issue' % problem
                return cls(aPK_obj = problem['pk_health_issue'])

        #--------------------------------------------------------
        def rename(self, description:str=None) -> bool:
                """Rename health issue.

                Args:
                        description: the new descriptive name for the issue
                """
                if description.strip() == '':
                        return False

                # update the issue description
                old_description = self._payload['description']
                self._payload['description'] = description.strip()
                self._is_modified = True
                successful, data = self.save_payload()
                if not successful:
                        _log.error('cannot rename health issue [%s] with [%s]' % (self, description))
                        self._payload['description'] = old_description
                        return False

                return True

        #--------------------------------------------------------
        def get_episodes(self) -> list['cEpisode']:
                """The episodes linked to this health issue."""
                cmd = "SELECT * FROM clin.v_pat_episodes WHERE pk_health_issue = %(pk)s"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]

        episodes = property(get_episodes)

        #--------------------------------------------------------
        def close_expired_episode(self, ttl:int=180) -> bool:
                """Close open episode if "older" than the Time To Live.

                Args:
                        ttl: maximum "age" of episode in days
                """
                open_episode = self.open_episode
                if open_episode is None:
                        return True

                #clinical_end = open_episode.best_guess_clinical_end_date
                clinical_end = open_episode.latest_access_date          # :-/
                ttl = datetime.timedelta(ttl)
                now = datetime.datetime.now(tz = clinical_end.tzinfo)
                if (clinical_end + ttl) > now:
                        return False

                open_episode['episode_open'] = False
                success, data = open_episode.save_payload()
                if success:
                        return True

                return False

        #--------------------------------------------------------
        def close_episode(self) -> bool:
                """Unconditionally close the open episode, if any."""
                open_episode = self.get_open_episode()
                open_episode['episode_open'] = False
                success_state, data = open_episode.save_payload()
                return success_state

        #--------------------------------------------------------
        def get_open_episode(self) -> 'cEpisode':
                SQL = "select pk FROM clin.episode WHERE fk_health_issue = %(pk_issue)s AND is_open IS True LIMIT 1"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': SQL, 'args': {'pk_issue': self.pk_obj}}])
                if rows:
                        return cEpisode(aPK_obj = rows[0][0])

                return None

        open_episode = property(get_open_episode)

        #--------------------------------------------------------
        def age_noted_human_readable(self):
                if self._payload['age_noted'] is None:
                        return '<???>'

                # since we've already got an interval we are bound to use it,
                # further transformation will only introduce more errors,
                # later we can improve this deeper inside
                return gmDateTime.format_interval_medically(self._payload['age_noted'])

        #--------------------------------------------------------
        def add_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "INSERT INTO clin.lnk_code2h_issue (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
                args = {
                        'item': self._payload['pk_health_issue'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def remove_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "DELETE FROM clin.lnk_code2h_issue WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
                args = {
                        'item': self._payload['pk_health_issue'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a'):
                rows = gmClinNarrative.get_as_journal (
                        issues = [self.pk_obj],
                        order_by = 'pk_episode, pk_encounter, clin_when, scr, src_table'
                )

                if len(rows) == 0:
                        return ''

                left_margin = ' ' * left_margin

                lines = []
                lines.append(_('Clinical data generated during encounters under this health issue:'))

                prev_epi = None
                for row in rows:
                        if row['pk_episode'] != prev_epi:
                                lines.append('')
                                prev_epi = row['pk_episode']

                        when = gmDateTime.pydt_strftime(row['clin_when'], date_format)
                        top_row = '%s%s %s (%s) %s' % (
                                gmTools.u_box_top_left_arc,
                                gmTools.u_box_horiz_single,
                                gmSoapDefs.soap_cat2l10n_str[row['real_soap_cat']],
                                when,
                                gmTools.u_box_horiz_single * 5
                        )
                        soap = gmTools.wrap (
                                text = row['narrative'],
                                width = 60,
                                initial_indent = '  ',
                                subsequent_indent = '  ' + left_margin
                        )
                        row_ver = ''
                        if row['row_version'] > 0:
                                row_ver = 'v%s: ' % row['row_version']
                        bottom_row = '%s%s %s, %s%s %s' % (
                                ' ' * 40,
                                gmTools.u_box_horiz_light_heavy,
                                row['modified_by'],
                                row_ver,
                                gmDateTime.pydt_strftime(row['modified_when'], date_format),
                                gmTools.u_box_horiz_heavy_light
                        )

                        lines.append(top_row)
                        lines.append(soap)
                        lines.append(bottom_row)

                eol_w_margin = '\n%s' % left_margin
                return left_margin + eol_w_margin.join(lines) + '\n'

        #--------------------------------------------------------
        def format (self, left_margin=0, patient=None,
                with_summary=True,
                with_codes=True,
                with_episodes=True,
                with_encounters=True,
                with_medications=True,
                with_hospital_stays=True,
                with_procedures=True,
                with_family_history=True,
                with_documents=True,
                with_tests=True,
                with_vaccinations=True,
                with_external_care=True
        ):

                lines = []

                lines.append(_('Health Issue %s%s%s%s   [#%s]') % (
                        '\u00BB',
                        self._payload['description'],
                        '\u00AB',
                        gmTools.coalesce (
                                value2test = self.laterality_description,
                                return_instead = '',
                                template4value = ' (%s)',
                                none_equivalents = [None, '', '?']
                        ),
                        self._payload['pk_health_issue']
                ))

                if self._payload['is_confidential']:
                        lines.append('')
                        lines.append(_(' ***** CONFIDENTIAL *****'))
                        lines.append('')

                if self._payload['is_cause_of_death']:
                        lines.append('')
                        lines.append(_(' contributed to death of patient'))
                        lines.append('')

                enc = cEncounter(aPK_obj = self._payload['pk_encounter'])
                lines.append (_(' Created during encounter: %s (%s - %s)   [#%s]') % (
                        enc['l10n_type'],
                        enc['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                        enc['last_affirmed_original_tz'].strftime('%H:%M'),
                        self._payload['pk_encounter']
                ))

                if self._payload['age_noted'] is not None:
                        lines.append(_(' Noted at age: %s') % self.age_noted_human_readable())

                lines.append(' ' + _('Status') + ': %s, %s%s' % (
                        gmTools.bool2subst(self._payload['is_active'], _('active'), _('inactive')),
                        gmTools.bool2subst(self._payload['clinically_relevant'], _('clinically relevant'), _('not clinically relevant')),
                        gmTools.coalesce (
                                value2test = diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification']),
                                return_instead = '',
                                template4value = ', %s',
                                none_equivalents = [None, '']
                        )
                ))

                if with_summary:
                        if self._payload['summary'] is not None:
                                lines.append(' %s:' % _('Synopsis'))
                                lines.append(gmTools.wrap (
                                        text = self._payload['summary'],
                                        width = 60,
                                        initial_indent = '  ',
                                        subsequent_indent = '  '
                                ))

                # codes ?
                if with_codes:
                        codes = self.generic_codes
                        if len(codes) > 0:
                                lines.append('')
                        for c in codes:
                                lines.append(' %s: %s (%s - %s)' % (
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                ))
                        del codes

                lines.append('')

                # patient/emr dependant
                if patient is not None:
                        if patient.ID != self._payload['pk_patient']:
                                msg = '<patient>.ID = %s but health issue %s belongs to patient %s' % (
                                        patient.ID,
                                        self._payload['pk_health_issue'],
                                        self._payload['pk_patient']
                                )
                                raise ValueError(msg)
                        emr = patient.emr

                        # episodes
                        if with_episodes:
                                epis = self.get_episodes()
                                if epis is None:
                                        lines.append(_('Error retrieving episodes for this health issue.'))
                                elif len(epis) == 0:
                                        lines.append(_('There are no episodes for this health issue.'))
                                else:
                                        lines.append (
                                                _('Episodes: %s (most recent: %s%s%s)') % (
                                                        len(epis),
                                                        gmTools.u_left_double_angle_quote,
                                                        emr.get_most_recent_episode(issue = self._payload['pk_health_issue'])['description'],
                                                        gmTools.u_right_double_angle_quote
                                                )
                                        )
                                        for epi in epis:
                                                lines.append(' \u00BB%s\u00AB (%s)' % (
                                                        epi['description'],
                                                        gmTools.bool2subst(epi['episode_open'], _('ongoing'), _('closed'))
                                                ))
                                lines.append('')

                        # encounters
                        if with_encounters:
                                first_encounter = emr.get_first_encounter(issue_id = self._payload['pk_health_issue'])
                                last_encounter = emr.get_last_encounter(issue_id = self._payload['pk_health_issue'])

                                if first_encounter is None or last_encounter is None:
                                        lines.append(_('No encounters found for this health issue.'))
                                else:
                                        encs = emr.get_encounters(issues = [self._payload['pk_health_issue']])
                                        lines.append(_('Encounters: %s (%s - %s):') % (
                                                len(encs),
                                                first_encounter['started_original_tz'].strftime('%m/%Y'),
                                                last_encounter['last_affirmed_original_tz'].strftime('%m/%Y')
                                        ))
                                        lines.append(_(' Most recent: %s - %s') % (
                                                last_encounter['started_original_tz'].strftime('%Y-%m-%d %H:%M'),
                                                last_encounter['last_affirmed_original_tz'].strftime('%H:%M')
                                        ))

                        # medications
                        if with_medications:
                                meds = emr.get_current_medications (
                                        issues = [ self._payload['pk_health_issue'] ],
                                        order_by = 'is_currently_active DESC, started, substance'
                                )
                                if len(meds) > 0:
                                        lines.append('')
                                        lines.append(_('Medications and Substances'))
                                for m in meds:
                                        lines.append(m.format(left_margin = (left_margin + 1)))
                                del meds

                        # hospitalizations
                        if with_hospital_stays:
                                stays = emr.get_hospital_stays (
                                        issues = [ self._payload['pk_health_issue'] ]
                                )
                                if len(stays) > 0:
                                        lines.append('')
                                        lines.append(_('Hospitalizations: %s') % len(stays))
                                for s in stays:
                                        lines.append(s.format(left_margin = (left_margin + 1)))
                                del stays

                        # procedures
                        if with_procedures:
                                procs = emr.get_performed_procedures (
                                        issues = [ self._payload['pk_health_issue'] ]
                                )
                                if len(procs) > 0:
                                        lines.append('')
                                        lines.append(_('Procedures performed: %s') % len(procs))
                                for p in procs:
                                        lines.append(p.format(left_margin = (left_margin + 1)))
                                del procs

                        # family history
                        if with_family_history:
                                fhx = emr.get_family_history(issues = [ self._payload['pk_health_issue'] ])
                                if len(fhx) > 0:
                                        lines.append('')
                                        lines.append(_('Family History: %s') % len(fhx))
                                for f in fhx:
                                        lines.append(f.format (
                                                left_margin = (left_margin + 1),
                                                include_episode = True,
                                                include_comment = True,
                                                include_codes = False
                                        ))
                                del fhx

                        epis = self.get_episodes()
                        if len(epis) > 0:
                                epi_pks = [ e['pk_episode'] for e in epis ]

                                # documents
                                if with_documents:
                                        doc_folder = patient.get_document_folder()
                                        docs = doc_folder.get_documents(pk_episodes = epi_pks)
                                        if len(docs) > 0:
                                                lines.append('')
                                                lines.append(_('Documents: %s') % len(docs))
                                        del docs

                                # test results
                                if with_tests:
                                        tests = emr.get_test_results_by_date(episodes = epi_pks)
                                        if len(tests) > 0:
                                                lines.append('')
                                                lines.append(_('Measurements and Results: %s') % len(tests))
                                        del tests

                                # vaccinations
                                if with_vaccinations:
                                        vaccs = emr.get_vaccinations(episodes = epi_pks, order_by = 'date_given, vaccine')
                                        if len(vaccs) > 0:
                                                lines.append('')
                                                lines.append(_('Vaccinations:'))
                                        for vacc in vaccs:
                                                lines.extend(vacc.format(with_reaction = True))
                                        del vaccs

                        del epis

                if with_external_care:
                        care = self._get_external_care(order_by = 'organization, unit, provider')
                        if len(care) > 0:
                                lines.append('')
                                lines.append(_('External care:'))
                        for item in care:
                                lines.append(' %s%s@%s%s' % (
                                        gmTools.coalesce(item['provider'], '', '%s: '),
                                        item['unit'],
                                        item['organization'],
                                        gmTools.coalesce(item['comment'], '', ' (%s)')
                                ))

                left_margin = ' ' * left_margin
                eol_w_margin = '\n%s' % left_margin
                lines = gmTools.strip_trailing_empty_lines(lines = lines, eol = '\n')
                return left_margin + eol_w_margin.join(lines) + '\n'
        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_external_care(self, order_by=None):
                return gmExternalCare.get_external_care_items(pk_health_issue = self.pk_obj, order_by = order_by)

        external_care = property(_get_external_care)

        #--------------------------------------------------------
        def _get_first_episode(self):

                args = {'pk_issue': self.pk_obj}

                cmd = """SELECT
                        earliest, pk_episode
                FROM (
                                -- .modified_when of all episodes of this issue,
                                -- earliest-possible thereof = when created,
                                -- should actually go all the way back into audit.log_episode
                                (SELECT
                                        c_epi.modified_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM clin.episode c_epi
                                 WHERE c_epi.fk_health_issue = %(pk_issue)s
                                )
                        UNION ALL

                                -- last modification of encounter in which episodes of this issue were created,
                                -- earliest-possible thereof = initial creation of that encounter
                                (SELECT
                                        c_enc.modified_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.episode c_epi
                                                INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                                INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- start of encounter in which episodes of this issue were created,
                                -- earliest-possible thereof = set by user
                                (SELECT
                                        c_enc.started AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.episode c_epi
                                                INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                                INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- start of encounters of clinical items linked to episodes of this issue,
                                -- earliest-possible thereof = explicitly set by user
                                (SELECT
                                        c_enc.started AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.encounter c_enc ON (c_cri.fk_encounter = c_enc.pk)
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- .clin_when of clinical items linked to episodes of this issue,
                                -- earliest-possible thereof = explicitly set by user
                                (SELECT
                                        c_cri.clin_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- earliest modification time of clinical items linked to episodes of this issue
                                -- this CAN be used since if an item is linked to an episode it can be
                                -- assumed the episode (should have) existed at the time of creation
                                (SELECT
                                        c_cri.modified_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- there may not be items, but there may still be documents ...
                                (SELECT
                                        b_dm.clin_when AS earliest,
                                        c_epi.pk AS pk_episode
                                 FROM
                                        blobs.doc_med b_dm
                                                INNER JOIN clin.episode c_epi ON (b_dm.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                  WHERE c_hi.pk = %(pk_issue)s
                                )
                ) AS candidates
                ORDER BY earliest NULLS LAST
                LIMIT 1"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                if len(rows) == 0:
                        return None
                return cEpisode(aPK_obj = rows[0]['pk_episode'])

        first_episode = property(_get_first_episode)

        #--------------------------------------------------------
        def _get_latest_episode(self):

                # explicit always wins:
                if self._payload['has_open_episode']:
                        return self.open_episode

                args = {'pk_issue': self.pk_obj}

                # cheap query first: any episodes at all ?
                cmd = "SELECT 1 FROM clin.episode WHERE fk_health_issue = %(pk_issue)s"
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                if len(rows) == 0:
                        return None

                cmd = """SELECT
                        latest, pk_episode
                FROM (
                                -- .clin_when of clinical items linked to episodes of this issue,
                                -- latest-possible thereof = explicitly set by user
                                (SELECT
                                        c_cri.clin_when AS latest,
                                        c_epi.pk AS pk_episode,
                                        1 AS rank
                                 FROM
                                        clin.clin_root_item c_cri
                                                INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- .clin_when of documents linked to episodes of this issue
                                (SELECT
                                        b_dm.clin_when AS latest,
                                        c_epi.pk AS pk_episode,
                                        1 AS rank
                                 FROM
                                        blobs.doc_med b_dm
                                                INNER JOIN clin.episode c_epi ON (b_dm.fk_episode = c_epi.pk)
                                                        INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )
                        UNION ALL

                                -- last_affirmed of encounter in which episodes of this issue were created,
                                -- earliest-possible thereof = set by user
                                (SELECT
                                        c_enc.last_affirmed AS latest,
                                        c_epi.pk AS pk_episode,
                                        2 AS rank
                                 FROM
                                        clin.episode c_epi
                                                INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                                INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                                 WHERE c_hi.pk = %(pk_issue)s
                                )

                ) AS candidates
                WHERE
                        -- weed out NULL rows due to episodes w/o clinical items and w/o documents
                        latest IS NOT NULL
                ORDER BY
                        rank,
                        latest DESC
                LIMIT 1
                """
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                if len(rows) == 0:
                        # there were no episodes for this issue
                        return None
                return cEpisode(aPK_obj = rows[0]['pk_episode'])

        latest_episode = property(_get_latest_episode)

        #--------------------------------------------------------
        # Steffi suggested we divide into safe and assumed (= possible) start dates
        def _get_safe_start_date(self):
                """This returns the date when we can assume to safely KNOW
                   the health issue existed (because the provider said so)."""

                args = {
                        'enc': self._payload['pk_encounter'],
                        'pk': self._payload['pk_health_issue']
                }
                cmd = """SELECT COALESCE (
                        -- this one must override all:
                        -- .age_noted if not null and DOB is known
                        (CASE
                                WHEN c_hi.age_noted IS NULL
                                THEN NULL::timestamp with time zone
                                WHEN
                                        (SELECT d_i.dob FROM dem.identity d_i WHERE d_i.pk = (
                                                SELECT c_enc.fk_patient FROM clin.encounter c_enc WHERE c_enc.pk = %(enc)s
                                        )) IS NULL
                                THEN NULL::timestamp with time zone
                                ELSE
                                        c_hi.age_noted + (
                                                SELECT d_i.dob FROM dem.identity d_i WHERE d_i.pk = (
                                                        SELECT c_enc.fk_patient FROM clin.encounter c_enc WHERE c_enc.pk = %(enc)s
                                                )
                                        )
                        END),

                        -- look at best_guess_clinical_start_date of all linked episodes

                        -- start of encounter in which created, earliest = explicitly set
                        (SELECT c_enc.started AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = c_hi.fk_encounter)
                )
                FROM clin.health_issue c_hi
                WHERE c_hi.pk = %(pk)s"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                start = rows[0][0]
                # leads to a loop:
                #end = self.clinical_end_date
                #if start > end:
                #       return end
                return start

        safe_start_date = property(_get_safe_start_date)

        #--------------------------------------------------------
        def _get_possible_start_date(self):
                args = {'pk': self._payload['pk_health_issue']}
                cmd = """
SELECT MIN(earliest) FROM (
        -- last modification, earliest = when created in/changed to the current state
        (SELECT modified_when AS earliest FROM clin.health_issue WHERE pk = %(pk)s)

UNION ALL
        -- last modification of encounter in which created, earliest = initial creation of that encounter
        (SELECT c_enc.modified_when AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = (
                SELECT c_hi.fk_encounter FROM clin.health_issue c_hi WHERE c_hi.pk = %(pk)s
        ))

UNION ALL
        -- earliest explicit .clin_when of clinical items linked to this health_issue
        (SELECT MIN(c_vpi.clin_when) AS earliest FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s)

UNION ALL
        -- earliest modification time of clinical items linked to this health issue
        -- this CAN be used since if an item is linked to a health issue it can be
        -- assumed the health issue (should have) existed at the time of creation
        (SELECT MIN(c_vpi.modified_when) AS earliest FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s)

UNION ALL
        -- earliest start of encounters of clinical items linked to this episode
        (SELECT MIN(c_enc.started) AS earliest FROM clin.encounter c_enc WHERE c_enc.pk IN (
                SELECT c_vpi.pk_encounter FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s
        ))

-- here we should be looking at
-- .best_guess_clinical_start_date of all episodes linked to this encounter

) AS candidates"""

                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return rows[0][0]

        possible_start_date = property(_get_possible_start_date)

        #--------------------------------------------------------
        def _get_clinical_end_date(self):
                if self._payload['is_active']:
                        return None
                if self._payload['has_open_episode']:
                        return None
                latest_episode = self.latest_episode
                if latest_episode is not None:
                        return latest_episode.best_guess_clinical_end_date
                # apparently, there are no episodes for this issue
                # and the issue is not active either
                # so, we simply do not know, the safest assumption is:
                return self.safe_start_date

        clinical_end_date = property(_get_clinical_end_date)

        #--------------------------------------------------------
        def _get_latest_access_date(self):
                cmd = """
SELECT
        MAX(latest)
FROM (
        -- last modification, latest = when last changed to the current state
        -- DO NOT USE: database upgrades may change this field
        (SELECT modified_when AS latest FROM clin.health_issue WHERE pk = %(pk)s)

        --UNION ALL
        -- last modification of encounter in which created, latest = initial creation of that encounter
        -- DO NOT USE: just because one corrects a typo does not mean the issue took any longer
        --(SELECT c_enc.modified_when AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
        -- )
        --)

        --UNION ALL
        -- end of encounter in which created, latest = explicitly set
        -- DO NOT USE: we can retrospectively create issues which
        -- DO NOT USE: are long since finished
        --(SELECT c_enc.last_affirmed AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
        -- )
        --)

        UNION ALL
        -- latest end of encounters of clinical items linked to this issue
        (SELECT
                MAX(last_affirmed) AS latest
         FROM clin.encounter
         WHERE pk IN (
                SELECT pk_encounter FROM clin.v_pat_items WHERE pk_health_issue = %(pk)s
         )
        )

        UNION ALL
        -- latest explicit .clin_when of clinical items linked to this issue
        (SELECT
                MAX(clin_when) AS latest
         FROM clin.v_pat_items
         WHERE pk_health_issue = %(pk)s
        )

        -- latest modification time of clinical items linked to this issue
        -- this CAN be used since if an item is linked to an issue it can be
        -- assumed the issue (should have) existed at the time of modification
        -- DO NOT USE, because typo fixes should not extend the issue
        --(SELECT MIN(modified_when) AS latest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

) AS candidates"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return rows[0][0]

        latest_access_date = property(_get_latest_access_date)

        #--------------------------------------------------------
        def _get_laterality_description(self):
                try:
                        return laterality2str[self._payload['laterality']]
                except KeyError:
                        return '<?>'

        laterality_description = property(_get_laterality_description)

        #--------------------------------------------------------
        def _get_diagnostic_certainty_description(self):
                return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])

        diagnostic_certainty_description = property(_get_diagnostic_certainty_description)

        #--------------------------------------------------------
        def _get_formatted_revision_history(self):
                cmd = """SELECT
                                '<N/A>'::TEXT as audit__action_applied,
                                NULL AS audit__action_when,
                                '<N/A>'::TEXT AS audit__action_by,
                                pk_audit,
                                row_version,
                                modified_when,
                                modified_by,
                                pk,
                                description,
                                laterality,
                                age_noted,
                                is_active,
                                clinically_relevant,
                                is_confidential,
                                is_cause_of_death,
                                fk_encounter,
                                grouping,
                                diagnostic_certainty_classification,
                                summary
                        FROM clin.health_issue
                        WHERE pk = %(pk_health_issue)s
                UNION ALL (
                        SELECT
                                audit_action as audit__action_applied,
                                audit_when as audit__action_when,
                                audit_by as audit__action_by,
                                pk_audit,
                                orig_version as row_version,
                                orig_when as modified_when,
                                orig_by as modified_by,
                                pk,
                                description,
                                laterality,
                                age_noted,
                                is_active,
                                clinically_relevant,
                                is_confidential,
                                is_cause_of_death,
                                fk_encounter,
                                grouping,
                                diagnostic_certainty_classification,
                                summary
                        FROM audit.log_health_issue
                        WHERE pk = %(pk_health_issue)s
                )
                ORDER BY row_version DESC
                """
                args = {'pk_health_issue': self.pk_obj}
                title = _('Health issue: %s%s%s') % (
                        gmTools.u_left_double_angle_quote,
                        self._payload['description'],
                        gmTools.u_right_double_angle_quote
                )
                return '\n'.join(self._get_revision_history(cmd, args, title))

        formatted_revision_history = property(_get_formatted_revision_history)
        #--------------------------------------------------------
        def _get_generic_codes(self):
                if len(self._payload['pk_generic_codes']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2h_issue WHERE fk_item = %(issue)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'issue': self._payload['pk_health_issue'],
                                        'codes': self._payload['pk_generic_codes']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2h_issue (fk_item, fk_generic_code) VALUES (%(issue)s, %(pk_code)s)',
                                'args': {
                                        'issue': self._payload['pk_health_issue'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                return

        generic_codes = property(_get_generic_codes, _set_generic_codes)

Ancestors

Static methods

def from_problem(problem: cProblem) ‑> cHealthIssue

Initialize health issue from issue-type problem.

Expand source code
@classmethod
def from_problem(cls, problem:'cProblem') -> 'cHealthIssue':
        """Initialize health issue from issue-type problem."""
        if isinstance(problem, cHealthIssue):
                return problem

        assert isinstance(problem, cProblem), 'cannot convert [%s] to health issue' % problem
        assert problem['type'] == 'issue', 'cannot convert [%s] to health issue' % problem
        return cls(aPK_obj = problem['pk_health_issue'])

Instance variables

var clinical_end_date
Expand source code
def _get_clinical_end_date(self):
        if self._payload['is_active']:
                return None
        if self._payload['has_open_episode']:
                return None
        latest_episode = self.latest_episode
        if latest_episode is not None:
                return latest_episode.best_guess_clinical_end_date
        # apparently, there are no episodes for this issue
        # and the issue is not active either
        # so, we simply do not know, the safest assumption is:
        return self.safe_start_date
var diagnostic_certainty_description
Expand source code
def _get_diagnostic_certainty_description(self):
        return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])
var episodes : list[cEpisode]

The episodes linked to this health issue.

Expand source code
def get_episodes(self) -> list['cEpisode']:
        """The episodes linked to this health issue."""
        cmd = "SELECT * FROM clin.v_pat_episodes WHERE pk_health_issue = %(pk)s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
        return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]
var external_care
Expand source code
def _get_external_care(self, order_by=None):
        return gmExternalCare.get_external_care_items(pk_health_issue = self.pk_obj, order_by = order_by)
var first_episode
Expand source code
def _get_first_episode(self):

        args = {'pk_issue': self.pk_obj}

        cmd = """SELECT
                earliest, pk_episode
        FROM (
                        -- .modified_when of all episodes of this issue,
                        -- earliest-possible thereof = when created,
                        -- should actually go all the way back into audit.log_episode
                        (SELECT
                                c_epi.modified_when AS earliest,
                                c_epi.pk AS pk_episode
                         FROM clin.episode c_epi
                         WHERE c_epi.fk_health_issue = %(pk_issue)s
                        )
                UNION ALL

                        -- last modification of encounter in which episodes of this issue were created,
                        -- earliest-possible thereof = initial creation of that encounter
                        (SELECT
                                c_enc.modified_when AS earliest,
                                c_epi.pk AS pk_episode
                         FROM
                                clin.episode c_epi
                                        INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                        INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- start of encounter in which episodes of this issue were created,
                        -- earliest-possible thereof = set by user
                        (SELECT
                                c_enc.started AS earliest,
                                c_epi.pk AS pk_episode
                         FROM
                                clin.episode c_epi
                                        INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                        INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- start of encounters of clinical items linked to episodes of this issue,
                        -- earliest-possible thereof = explicitly set by user
                        (SELECT
                                c_enc.started AS earliest,
                                c_epi.pk AS pk_episode
                         FROM
                                clin.clin_root_item c_cri
                                        INNER JOIN clin.encounter c_enc ON (c_cri.fk_encounter = c_enc.pk)
                                        INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- .clin_when of clinical items linked to episodes of this issue,
                        -- earliest-possible thereof = explicitly set by user
                        (SELECT
                                c_cri.clin_when AS earliest,
                                c_epi.pk AS pk_episode
                         FROM
                                clin.clin_root_item c_cri
                                        INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- earliest modification time of clinical items linked to episodes of this issue
                        -- this CAN be used since if an item is linked to an episode it can be
                        -- assumed the episode (should have) existed at the time of creation
                        (SELECT
                                c_cri.modified_when AS earliest,
                                c_epi.pk AS pk_episode
                         FROM
                                clin.clin_root_item c_cri
                                        INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- there may not be items, but there may still be documents ...
                        (SELECT
                                b_dm.clin_when AS earliest,
                                c_epi.pk AS pk_episode
                         FROM
                                blobs.doc_med b_dm
                                        INNER JOIN clin.episode c_epi ON (b_dm.fk_episode = c_epi.pk)
                                                INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                          WHERE c_hi.pk = %(pk_issue)s
                        )
        ) AS candidates
        ORDER BY earliest NULLS LAST
        LIMIT 1"""
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        if len(rows) == 0:
                return None
        return cEpisode(aPK_obj = rows[0]['pk_episode'])
var formatted_revision_history
Expand source code
def _get_formatted_revision_history(self):
        cmd = """SELECT
                        '<N/A>'::TEXT as audit__action_applied,
                        NULL AS audit__action_when,
                        '<N/A>'::TEXT AS audit__action_by,
                        pk_audit,
                        row_version,
                        modified_when,
                        modified_by,
                        pk,
                        description,
                        laterality,
                        age_noted,
                        is_active,
                        clinically_relevant,
                        is_confidential,
                        is_cause_of_death,
                        fk_encounter,
                        grouping,
                        diagnostic_certainty_classification,
                        summary
                FROM clin.health_issue
                WHERE pk = %(pk_health_issue)s
        UNION ALL (
                SELECT
                        audit_action as audit__action_applied,
                        audit_when as audit__action_when,
                        audit_by as audit__action_by,
                        pk_audit,
                        orig_version as row_version,
                        orig_when as modified_when,
                        orig_by as modified_by,
                        pk,
                        description,
                        laterality,
                        age_noted,
                        is_active,
                        clinically_relevant,
                        is_confidential,
                        is_cause_of_death,
                        fk_encounter,
                        grouping,
                        diagnostic_certainty_classification,
                        summary
                FROM audit.log_health_issue
                WHERE pk = %(pk_health_issue)s
        )
        ORDER BY row_version DESC
        """
        args = {'pk_health_issue': self.pk_obj}
        title = _('Health issue: %s%s%s') % (
                gmTools.u_left_double_angle_quote,
                self._payload['description'],
                gmTools.u_right_double_angle_quote
        )
        return '\n'.join(self._get_revision_history(cmd, args, title))
var generic_codes
Expand source code
def _get_generic_codes(self):
        if len(self._payload['pk_generic_codes']) == 0:
                return []

        cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
        args = {'pks': self._payload['pk_generic_codes']}
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]
var laterality_description
Expand source code
def _get_laterality_description(self):
        try:
                return laterality2str[self._payload['laterality']]
        except KeyError:
                return '<?>'
var latest_access_date
Expand source code
        def _get_latest_access_date(self):
                cmd = """
SELECT
        MAX(latest)
FROM (
        -- last modification, latest = when last changed to the current state
        -- DO NOT USE: database upgrades may change this field
        (SELECT modified_when AS latest FROM clin.health_issue WHERE pk = %(pk)s)

        --UNION ALL
        -- last modification of encounter in which created, latest = initial creation of that encounter
        -- DO NOT USE: just because one corrects a typo does not mean the issue took any longer
        --(SELECT c_enc.modified_when AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
        -- )
        --)

        --UNION ALL
        -- end of encounter in which created, latest = explicitly set
        -- DO NOT USE: we can retrospectively create issues which
        -- DO NOT USE: are long since finished
        --(SELECT c_enc.last_affirmed AS latest FROM clin.encounter c_enc WHERE c_enc.pk = (
        --      SELECT fk_encounter FROM clin.episode WHERE pk = %(pk)s
        -- )
        --)

        UNION ALL
        -- latest end of encounters of clinical items linked to this issue
        (SELECT
                MAX(last_affirmed) AS latest
         FROM clin.encounter
         WHERE pk IN (
                SELECT pk_encounter FROM clin.v_pat_items WHERE pk_health_issue = %(pk)s
         )
        )

        UNION ALL
        -- latest explicit .clin_when of clinical items linked to this issue
        (SELECT
                MAX(clin_when) AS latest
         FROM clin.v_pat_items
         WHERE pk_health_issue = %(pk)s
        )

        -- latest modification time of clinical items linked to this issue
        -- this CAN be used since if an item is linked to an issue it can be
        -- assumed the issue (should have) existed at the time of modification
        -- DO NOT USE, because typo fixes should not extend the issue
        --(SELECT MIN(modified_when) AS latest FROM clin.clin_root_item WHERE fk_episode = %(pk)s)

) AS candidates"""
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
                return rows[0][0]
var latest_episode
Expand source code
def _get_latest_episode(self):

        # explicit always wins:
        if self._payload['has_open_episode']:
                return self.open_episode

        args = {'pk_issue': self.pk_obj}

        # cheap query first: any episodes at all ?
        cmd = "SELECT 1 FROM clin.episode WHERE fk_health_issue = %(pk_issue)s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        if len(rows) == 0:
                return None

        cmd = """SELECT
                latest, pk_episode
        FROM (
                        -- .clin_when of clinical items linked to episodes of this issue,
                        -- latest-possible thereof = explicitly set by user
                        (SELECT
                                c_cri.clin_when AS latest,
                                c_epi.pk AS pk_episode,
                                1 AS rank
                         FROM
                                clin.clin_root_item c_cri
                                        INNER JOIN clin.episode c_epi ON (c_cri.fk_episode = c_epi.pk)
                                                INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- .clin_when of documents linked to episodes of this issue
                        (SELECT
                                b_dm.clin_when AS latest,
                                c_epi.pk AS pk_episode,
                                1 AS rank
                         FROM
                                blobs.doc_med b_dm
                                        INNER JOIN clin.episode c_epi ON (b_dm.fk_episode = c_epi.pk)
                                                INNER JOIN clin.health_issue c_hi ON (c_epi.fk_health_issue = c_hi.pk)
                         WHERE c_hi.pk = %(pk_issue)s
                        )
                UNION ALL

                        -- last_affirmed of encounter in which episodes of this issue were created,
                        -- earliest-possible thereof = set by user
                        (SELECT
                                c_enc.last_affirmed AS latest,
                                c_epi.pk AS pk_episode,
                                2 AS rank
                         FROM
                                clin.episode c_epi
                                        INNER JOIN clin.encounter c_enc ON (c_enc.pk = c_epi.fk_encounter)
                                        INNER JOIN clin.health_issue c_hi ON (c_hi.pk = c_epi.fk_health_issue)
                         WHERE c_hi.pk = %(pk_issue)s
                        )

        ) AS candidates
        WHERE
                -- weed out NULL rows due to episodes w/o clinical items and w/o documents
                latest IS NOT NULL
        ORDER BY
                rank,
                latest DESC
        LIMIT 1
        """
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        if len(rows) == 0:
                # there were no episodes for this issue
                return None
        return cEpisode(aPK_obj = rows[0]['pk_episode'])
var open_episodecEpisode
Expand source code
def get_open_episode(self) -> 'cEpisode':
        SQL = "select pk FROM clin.episode WHERE fk_health_issue = %(pk_issue)s AND is_open IS True LIMIT 1"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': SQL, 'args': {'pk_issue': self.pk_obj}}])
        if rows:
                return cEpisode(aPK_obj = rows[0][0])

        return None
var possible_start_date
Expand source code
        def _get_possible_start_date(self):
                args = {'pk': self._payload['pk_health_issue']}
                cmd = """
SELECT MIN(earliest) FROM (
        -- last modification, earliest = when created in/changed to the current state
        (SELECT modified_when AS earliest FROM clin.health_issue WHERE pk = %(pk)s)

UNION ALL
        -- last modification of encounter in which created, earliest = initial creation of that encounter
        (SELECT c_enc.modified_when AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = (
                SELECT c_hi.fk_encounter FROM clin.health_issue c_hi WHERE c_hi.pk = %(pk)s
        ))

UNION ALL
        -- earliest explicit .clin_when of clinical items linked to this health_issue
        (SELECT MIN(c_vpi.clin_when) AS earliest FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s)

UNION ALL
        -- earliest modification time of clinical items linked to this health issue
        -- this CAN be used since if an item is linked to a health issue it can be
        -- assumed the health issue (should have) existed at the time of creation
        (SELECT MIN(c_vpi.modified_when) AS earliest FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s)

UNION ALL
        -- earliest start of encounters of clinical items linked to this episode
        (SELECT MIN(c_enc.started) AS earliest FROM clin.encounter c_enc WHERE c_enc.pk IN (
                SELECT c_vpi.pk_encounter FROM clin.v_pat_items c_vpi WHERE c_vpi.pk_health_issue = %(pk)s
        ))

-- here we should be looking at
-- .best_guess_clinical_start_date of all episodes linked to this encounter

) AS candidates"""

                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return rows[0][0]
var safe_start_date

This returns the date when we can assume to safely KNOW the health issue existed (because the provider said so).

Expand source code
def _get_safe_start_date(self):
        """This returns the date when we can assume to safely KNOW
           the health issue existed (because the provider said so)."""

        args = {
                'enc': self._payload['pk_encounter'],
                'pk': self._payload['pk_health_issue']
        }
        cmd = """SELECT COALESCE (
                -- this one must override all:
                -- .age_noted if not null and DOB is known
                (CASE
                        WHEN c_hi.age_noted IS NULL
                        THEN NULL::timestamp with time zone
                        WHEN
                                (SELECT d_i.dob FROM dem.identity d_i WHERE d_i.pk = (
                                        SELECT c_enc.fk_patient FROM clin.encounter c_enc WHERE c_enc.pk = %(enc)s
                                )) IS NULL
                        THEN NULL::timestamp with time zone
                        ELSE
                                c_hi.age_noted + (
                                        SELECT d_i.dob FROM dem.identity d_i WHERE d_i.pk = (
                                                SELECT c_enc.fk_patient FROM clin.encounter c_enc WHERE c_enc.pk = %(enc)s
                                        )
                                )
                END),

                -- look at best_guess_clinical_start_date of all linked episodes

                -- start of encounter in which created, earliest = explicitly set
                (SELECT c_enc.started AS earliest FROM clin.encounter c_enc WHERE c_enc.pk = c_hi.fk_encounter)
        )
        FROM clin.health_issue c_hi
        WHERE c_hi.pk = %(pk)s"""
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        start = rows[0][0]
        # leads to a loop:
        #end = self.clinical_end_date
        #if start > end:
        #       return end
        return start

Methods

def add_code(self, pk_code=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def add_code(self, pk_code=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        cmd = "INSERT INTO clin.lnk_code2h_issue (fk_item, fk_generic_code) values (%(item)s, %(code)s)"
        args = {
                'item': self._payload['pk_health_issue'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def age_noted_human_readable(self)
Expand source code
def age_noted_human_readable(self):
        if self._payload['age_noted'] is None:
                return '<???>'

        # since we've already got an interval we are bound to use it,
        # further transformation will only introduce more errors,
        # later we can improve this deeper inside
        return gmDateTime.format_interval_medically(self._payload['age_noted'])
def close_episode(self) ‑> bool

Unconditionally close the open episode, if any.

Expand source code
def close_episode(self) -> bool:
        """Unconditionally close the open episode, if any."""
        open_episode = self.get_open_episode()
        open_episode['episode_open'] = False
        success_state, data = open_episode.save_payload()
        return success_state
def close_expired_episode(self, ttl: int = 180) ‑> bool

Close open episode if "older" than the Time To Live.

Args

ttl
maximum "age" of episode in days
Expand source code
def close_expired_episode(self, ttl:int=180) -> bool:
        """Close open episode if "older" than the Time To Live.

        Args:
                ttl: maximum "age" of episode in days
        """
        open_episode = self.open_episode
        if open_episode is None:
                return True

        #clinical_end = open_episode.best_guess_clinical_end_date
        clinical_end = open_episode.latest_access_date          # :-/
        ttl = datetime.timedelta(ttl)
        now = datetime.datetime.now(tz = clinical_end.tzinfo)
        if (clinical_end + ttl) > now:
                return False

        open_episode['episode_open'] = False
        success, data = open_episode.save_payload()
        if success:
                return True

        return False
def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a')
Expand source code
def format_as_journal(self, left_margin=0, date_format='%Y %b %d, %a'):
        rows = gmClinNarrative.get_as_journal (
                issues = [self.pk_obj],
                order_by = 'pk_episode, pk_encounter, clin_when, scr, src_table'
        )

        if len(rows) == 0:
                return ''

        left_margin = ' ' * left_margin

        lines = []
        lines.append(_('Clinical data generated during encounters under this health issue:'))

        prev_epi = None
        for row in rows:
                if row['pk_episode'] != prev_epi:
                        lines.append('')
                        prev_epi = row['pk_episode']

                when = gmDateTime.pydt_strftime(row['clin_when'], date_format)
                top_row = '%s%s %s (%s) %s' % (
                        gmTools.u_box_top_left_arc,
                        gmTools.u_box_horiz_single,
                        gmSoapDefs.soap_cat2l10n_str[row['real_soap_cat']],
                        when,
                        gmTools.u_box_horiz_single * 5
                )
                soap = gmTools.wrap (
                        text = row['narrative'],
                        width = 60,
                        initial_indent = '  ',
                        subsequent_indent = '  ' + left_margin
                )
                row_ver = ''
                if row['row_version'] > 0:
                        row_ver = 'v%s: ' % row['row_version']
                bottom_row = '%s%s %s, %s%s %s' % (
                        ' ' * 40,
                        gmTools.u_box_horiz_light_heavy,
                        row['modified_by'],
                        row_ver,
                        gmDateTime.pydt_strftime(row['modified_when'], date_format),
                        gmTools.u_box_horiz_heavy_light
                )

                lines.append(top_row)
                lines.append(soap)
                lines.append(bottom_row)

        eol_w_margin = '\n%s' % left_margin
        return left_margin + eol_w_margin.join(lines) + '\n'
def get_episodes(self) ‑> list[cEpisode]

The episodes linked to this health issue.

Expand source code
def get_episodes(self) -> list['cEpisode']:
        """The episodes linked to this health issue."""
        cmd = "SELECT * FROM clin.v_pat_episodes WHERE pk_health_issue = %(pk)s"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': {'pk': self.pk_obj}}])
        return [ cEpisode(row = {'data': r, 'pk_field': 'pk_episode'})  for r in rows ]
def get_open_episode(self) ‑> cEpisode
Expand source code
def get_open_episode(self) -> 'cEpisode':
        SQL = "select pk FROM clin.episode WHERE fk_health_issue = %(pk_issue)s AND is_open IS True LIMIT 1"
        rows = gmPG2.run_ro_queries(queries = [{'cmd': SQL, 'args': {'pk_issue': self.pk_obj}}])
        if rows:
                return cEpisode(aPK_obj = rows[0][0])

        return None
def remove_code(self, pk_code=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def remove_code(self, pk_code=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        cmd = "DELETE FROM clin.lnk_code2h_issue WHERE fk_item = %(item)s AND fk_generic_code = %(code)s"
        args = {
                'item': self._payload['pk_health_issue'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def rename(self, description: str = None) ‑> bool

Rename health issue.

Args

description
the new descriptive name for the issue
Expand source code
def rename(self, description:str=None) -> bool:
        """Rename health issue.

        Args:
                description: the new descriptive name for the issue
        """
        if description.strip() == '':
                return False

        # update the issue description
        old_description = self._payload['description']
        self._payload['description'] = description.strip()
        self._is_modified = True
        successful, data = self.save_payload()
        if not successful:
                _log.error('cannot rename health issue [%s] with [%s]' % (self, description))
                self._payload['description'] = old_description
                return False

        return True

Inherited members

class cHospitalStay (aPK_obj: int | dict = None, row: dict = None, link_obj=None)

Represents business objects in the database.

Rules

  • instances ARE ASSUMED TO EXIST in the database
  • PK construction (aPK_obj): DOES verify its existence on instantiation (fetching data fails)
  • Row construction (row): allowed by using a dict of pairs of field name: field value (PERFORMANCE improvement)
  • does NOT verify FK target existence
  • does NOT create new entries in the database
  • does NOT lazy-fetch fields on access

Class scope SQL commands and variables:

_cmd_fetch_payload:

  • must return exactly one row
  • WHERE clause argument values are expected in self.pk_obj (taken from init(aPK_obj))
  • must return xmin of all rows that _cmds_store_payload will be updating, so views must support the xmin columns of their underlying tables

_cmds_store_payload:

  • one or multiple "update … set … where xmin_ = … and pk = …" statements which actually update the database from the data in self._payload,
  • the last query must refetch at least the XMIN values needed to detect concurrent updates, their field names had better be the same as in _cmd_fetch_payload,
  • the last query CAN return other fields which is particularly useful when those other fields are computed in the backend and may thus change upon save but will not have been set by the client code explicitly - this is only really of concern if the saved subclass is to be reused after saving rather than re-instantiated
  • when subclasses tend to live a while after save_payload() was called and they support computed fields (say, _(some_column) you need to return all columns (see cEncounter)

_updatable_fields:

  • a list of fields available for update via object['field']

Call init from child classes like so:

    super().__init__(aPK_obj = aPK_obj, row = row, link_obj = link_obj)

Args

aPK_obj
retrieve data from backend
  • an scalar value the ._cmd_fetch_payload WHERE condition must be a simple column: "… WHERE pk_col = %s"
  • a dictionary of values the ._cmd_fetch_payload WHERE condition must consume the dictionary and produce a unique row
row
must hold the fields
  • data: list of column values for the row selected by ._cmd_fetch_payload (as returned by cursor.fetchone() in the DB-API)
  • pk_field: the name of the primary key column OR
  • pk_obj: a dictionary suitable for being passed to cursor.execute and holding the primary key values, used for composite PKs
  • for example:
    row = {
            'data': rows[0],
            'pk_field': 'pk_XXX (the PK column name)',
            'pk_obj': {'pk_col1': pk_col1_val, 'pk_col2': pk_col2_val}
    }
    
    rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
    objects = [ cChildClass(row = {'data': r, 'pk_field': 'the PK column name'}) for r in rows ]
    
Expand source code
class cHospitalStay(gmBusinessDBObject.cBusinessDBObject):

        _cmd_fetch_payload = _SQL_get_hospital_stays % "pk_hospital_stay = %s"
        _cmds_store_payload = [
                """UPDATE clin.hospital_stay SET
                                clin_when = %(admission)s,
                                discharge = %(discharge)s,
                                fk_org_unit = %(pk_org_unit)s,
                                narrative = gm.nullify_empty_string(%(comment)s),
                                fk_episode = %(pk_episode)s,
                                fk_encounter = %(pk_encounter)s
                        WHERE
                                pk = %(pk_hospital_stay)s
                                        AND
                                xmin = %(xmin_hospital_stay)s
                        RETURNING
                                xmin AS xmin_hospital_stay
                        """
        ]
        _updatable_fields = [
                'admission',
                'discharge',
                'pk_org_unit',
                'pk_episode',
                'pk_encounter',
                'comment'
        ]

        #--------------------------------------------------------
        def format_maximum_information(self, patient=None):
                return self.format (
                        include_procedures = True,
                        include_docs = True
                ).split('\n')

        #-------------------------------------------------------
        def format(self, left_margin=0, include_procedures=False, include_docs=False, include_episode=True):

                if self._payload['discharge'] is not None:
                        discharge = ' - %s' % gmDateTime.pydt_strftime(self._payload['discharge'], '%Y %b %d')
                else:
                        discharge = ''

                episode = ''
                if include_episode:
                        episode = ': %s%s%s' % (
                                gmTools.u_left_double_angle_quote,
                                self._payload['episode'],
                                gmTools.u_right_double_angle_quote
                        )

                lines = ['%s%s%s (%s@%s)%s' % (
                        ' ' * left_margin,
                        gmDateTime.pydt_strftime(self._payload['admission'], '%Y %b %d'),
                        discharge,
                        self._payload['ward'],
                        self._payload['hospital'],
                        episode
                )]

                if include_docs:
                        for doc in self.documents:
                                lines.append('%s%s: %s\n' % (
                                        ' ' * left_margin,
                                        _('Document'),
                                        doc.format(single_line = True)
                                ))

                return '\n'.join(lines)

        #--------------------------------------------------------
        def _get_documents(self):
                return [ gmDocuments.cDocument(aPK_obj = pk_doc) for pk_doc in  self._payload['pk_documents'] ]

        documents = property(_get_documents)

Ancestors

Instance variables

var documents
Expand source code
def _get_documents(self):
        return [ gmDocuments.cDocument(aPK_obj = pk_doc) for pk_doc in  self._payload['pk_documents'] ]

Methods

def format_maximum_information(self, patient=None)
Expand source code
def format_maximum_information(self, patient=None):
        return self.format (
                include_procedures = True,
                include_docs = True
        ).split('\n')

Inherited members

class cPerformedProcedure (aPK_obj: int | dict = None, row: dict = None, link_obj=None)

Represents business objects in the database.

Rules

  • instances ARE ASSUMED TO EXIST in the database
  • PK construction (aPK_obj): DOES verify its existence on instantiation (fetching data fails)
  • Row construction (row): allowed by using a dict of pairs of field name: field value (PERFORMANCE improvement)
  • does NOT verify FK target existence
  • does NOT create new entries in the database
  • does NOT lazy-fetch fields on access

Class scope SQL commands and variables:

_cmd_fetch_payload:

  • must return exactly one row
  • WHERE clause argument values are expected in self.pk_obj (taken from init(aPK_obj))
  • must return xmin of all rows that _cmds_store_payload will be updating, so views must support the xmin columns of their underlying tables

_cmds_store_payload:

  • one or multiple "update … set … where xmin_ = … and pk = …" statements which actually update the database from the data in self._payload,
  • the last query must refetch at least the XMIN values needed to detect concurrent updates, their field names had better be the same as in _cmd_fetch_payload,
  • the last query CAN return other fields which is particularly useful when those other fields are computed in the backend and may thus change upon save but will not have been set by the client code explicitly - this is only really of concern if the saved subclass is to be reused after saving rather than re-instantiated
  • when subclasses tend to live a while after save_payload() was called and they support computed fields (say, _(some_column) you need to return all columns (see cEncounter)

_updatable_fields:

  • a list of fields available for update via object['field']

Call init from child classes like so:

    super().__init__(aPK_obj = aPK_obj, row = row, link_obj = link_obj)

Args

aPK_obj
retrieve data from backend
  • an scalar value the ._cmd_fetch_payload WHERE condition must be a simple column: "… WHERE pk_col = %s"
  • a dictionary of values the ._cmd_fetch_payload WHERE condition must consume the dictionary and produce a unique row
row
must hold the fields
  • data: list of column values for the row selected by ._cmd_fetch_payload (as returned by cursor.fetchone() in the DB-API)
  • pk_field: the name of the primary key column OR
  • pk_obj: a dictionary suitable for being passed to cursor.execute and holding the primary key values, used for composite PKs
  • for example:
    row = {
            'data': rows[0],
            'pk_field': 'pk_XXX (the PK column name)',
            'pk_obj': {'pk_col1': pk_col1_val, 'pk_col2': pk_col2_val}
    }
    
    rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
    objects = [ cChildClass(row = {'data': r, 'pk_field': 'the PK column name'}) for r in rows ]
    
Expand source code
class cPerformedProcedure(gmBusinessDBObject.cBusinessDBObject):

        _cmd_fetch_payload = _SQL_get_procedures % "pk_procedure = %s"
        _cmds_store_payload = [
                """UPDATE clin.procedure SET
                                soap_cat = 'p',
                                clin_when = %(clin_when)s,
                                clin_end = %(clin_end)s,
                                is_ongoing = %(is_ongoing)s,
                                narrative = gm.nullify_empty_string(%(performed_procedure)s),
                                fk_hospital_stay = %(pk_hospital_stay)s,
                                fk_org_unit = (CASE
                                        WHEN %(pk_hospital_stay)s IS NULL THEN %(pk_org_unit)s
                                        ELSE NULL
                                END)::integer,
                                fk_episode = %(pk_episode)s,
                                fk_encounter = %(pk_encounter)s,
                                fk_doc = %(pk_doc)s,
                                comment = gm.nullify_empty_string(%(comment)s)
                        WHERE
                                pk = %(pk_procedure)s AND
                                xmin = %(xmin_procedure)s
                        RETURNING xmin as xmin_procedure"""
        ]
        _updatable_fields = [
                'clin_when',
                'clin_end',
                'is_ongoing',
                'performed_procedure',
                'pk_hospital_stay',
                'pk_org_unit',
                'pk_episode',
                'pk_encounter',
                'pk_doc',
                'comment'
        ]
        #-------------------------------------------------------
        def __setitem__(self, attribute, value):

                if (attribute == 'pk_hospital_stay') and (value is not None):
                        gmBusinessDBObject.cBusinessDBObject.__setitem__(self, 'pk_org_unit', None)

                if (attribute == 'pk_org_unit') and (value is not None):
                        gmBusinessDBObject.cBusinessDBObject.__setitem__(self, 'pk_hospital_stay', None)

                gmBusinessDBObject.cBusinessDBObject.__setitem__(self, attribute, value)

        #--------------------------------------------------------
        def format_maximum_information(self, left_margin=0, patient=None):
                return self.format (
                        left_margin = left_margin,
                        include_episode = True,
                        include_codes = True,
                        include_address = True,
                        include_comm = True,
                        include_doc = True
                ).split('\n')

        #-------------------------------------------------------
        def format(self, left_margin=0, include_episode=True, include_codes=False, include_address=False, include_comm=False, include_doc=False):

                if self._payload['is_ongoing']:
                        end = _(' (ongoing)')
                else:
                        end = self._payload['clin_end']
                        if end is None:
                                end = ''
                        else:
                                end = ' - %s' % gmDateTime.pydt_strftime(end, '%Y %b %d')

                line = '%s%s%s: %s%s [%s @ %s]' % (
                        (' ' * left_margin),
                        gmDateTime.pydt_strftime(self._payload['clin_when'], '%Y %b %d'),
                        end,
                        self._payload['performed_procedure'],
                        gmTools.bool2str(include_episode, ' (%s)' % self._payload['episode'], ''),
                        self._payload['unit'],
                        self._payload['organization']
                )

                line += gmTools.coalesce(self._payload['comment'], '', '\n' + (' ' * left_margin) + _('Comment: ') + '%s')

                if include_comm:
                        for channel in self.org_unit.comm_channels:
                                line += ('\n%(comm_type)s: %(url)s' % channel)

                if include_address:
                        adr = self.org_unit.address
                        if adr is not None:
                                line += '\n'
                                line += '\n'.join(adr.format(single_line = False, show_type = False))
                                line += '\n'

                if include_doc:
                        doc = self.doc
                        if doc is not None:
                                line += '\n'
                                line += _('Document') + ': ' + doc.format(single_line = True)
                                line += '\n'

                if include_codes:
                        codes = self.generic_codes
                        if len(codes) > 0:
                                line += '\n'
                        for c in codes:
                                line += '%s  %s: %s (%s - %s)\n' % (
                                        (' ' * left_margin),
                                        c['code'],
                                        c['term'],
                                        c['name_short'],
                                        c['version']
                                )
                        del codes

                return line

        #--------------------------------------------------------
        def add_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "INSERT INTO clin.lnk_code2procedure (fk_item, fk_generic_code) values (%(issue)s, %(code)s)"
                args = {
                        'issue': self._payload['pk_procedure'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        def remove_code(self, pk_code=None):
                """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
                cmd = "DELETE FROM clin.lnk_code2procedure WHERE fk_item = %(issue)s AND fk_generic_code = %(code)s"
                args = {
                        'issue': self._payload['pk_procedure'],
                        'code': pk_code
                }
                gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
                return True

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        def _get_stay(self):
                if self._payload['pk_hospital_stay'] is None:
                        return None
                return cHospitalStay(aPK_obj = self._payload['pk_hospital_stay'])

        hospital_stay = property(_get_stay)

        #--------------------------------------------------------
        def _get_org_unit(self):
                return gmOrganization.cOrgUnit(self._payload['pk_org_unit'])

        org_unit = property(_get_org_unit)

        #--------------------------------------------------------
        def _get_doc(self):
                if self._payload['pk_doc'] is None:
                        return None
                return gmDocuments.cDocument(aPK_obj = self._payload['pk_doc'])

        doc = property(_get_doc)

        #--------------------------------------------------------
        def _get_generic_codes(self):
                if len(self._payload['pk_generic_codes']) == 0:
                        return []

                cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
                args = {'pks': self._payload['pk_generic_codes']}
                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        def _set_generic_codes(self, pk_codes):
                queries = []
                # remove all codes
                if len(self._payload['pk_generic_codes']) > 0:
                        queries.append ({
                                'cmd': 'DELETE FROM clin.lnk_code2procedure WHERE fk_item = %(proc)s AND fk_generic_code = ANY(%(codes)s)',
                                'args': {
                                        'proc': self._payload['pk_procedure'],
                                        'codes': self._payload['pk_generic_codes']
                                }
                        })
                # add new codes
                for pk_code in pk_codes:
                        queries.append ({
                                'cmd': 'INSERT INTO clin.lnk_code2procedure (fk_item, fk_generic_code) VALUES (%(proc)s, %(pk_code)s)',
                                'args': {
                                        'proc': self._payload['pk_procedure'],
                                        'pk_code': pk_code
                                }
                        })
                if len(queries) == 0:
                        return
                # run it all in one transaction
                gmPG2.run_rw_queries(queries = queries)
                return

        generic_codes = property(_get_generic_codes, _set_generic_codes)

Ancestors

Instance variables

var doc
Expand source code
def _get_doc(self):
        if self._payload['pk_doc'] is None:
                return None
        return gmDocuments.cDocument(aPK_obj = self._payload['pk_doc'])
var generic_codes
Expand source code
def _get_generic_codes(self):
        if len(self._payload['pk_generic_codes']) == 0:
                return []

        cmd = gmCoding._SQL_get_generic_linked_codes % 'pk_generic_code = ANY(%(pks)s)'
        args = {'pks': self._payload['pk_generic_codes']}
        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]
var hospital_stay
Expand source code
def _get_stay(self):
        if self._payload['pk_hospital_stay'] is None:
                return None
        return cHospitalStay(aPK_obj = self._payload['pk_hospital_stay'])
var org_unit
Expand source code
def _get_org_unit(self):
        return gmOrganization.cOrgUnit(self._payload['pk_org_unit'])

Methods

def add_code(self, pk_code=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def add_code(self, pk_code=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        cmd = "INSERT INTO clin.lnk_code2procedure (fk_item, fk_generic_code) values (%(issue)s, %(code)s)"
        args = {
                'issue': self._payload['pk_procedure'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True
def format_maximum_information(self, left_margin=0, patient=None)
Expand source code
def format_maximum_information(self, left_margin=0, patient=None):
        return self.format (
                left_margin = left_margin,
                include_episode = True,
                include_codes = True,
                include_address = True,
                include_comm = True,
                include_doc = True
        ).split('\n')
def remove_code(self, pk_code=None)

must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)

Expand source code
def remove_code(self, pk_code=None):
        """<pk_code> must be a value from ref.coding_system_root.pk_coding_system (clin.lnk_code2item_root.fk_generic_code)"""
        cmd = "DELETE FROM clin.lnk_code2procedure WHERE fk_item = %(issue)s AND fk_generic_code = %(code)s"
        args = {
                'issue': self._payload['pk_procedure'],
                'code': pk_code
        }
        gmPG2.run_rw_queries(queries = [{'cmd': cmd, 'args': args}])
        return True

Inherited members

class cProblem (aPK_obj=None, try_potential_problems=False)

Represents one problem.

problems are the aggregation of .clinically_relevant=True issues and .is_open=True episodes

Initialize.

aPK_obj must contain the keys pk_patient pk_episode pk_health_issue

Expand source code
class cProblem(gmBusinessDBObject.cBusinessDBObject):
        """Represents one problem.

        problems are the aggregation of
                .clinically_relevant=True issues and
                .is_open=True episodes
        """
        _cmd_fetch_payload = ''                                 # will get programmatically defined in __init__
        _cmds_store_payload:list = ["select 1"]
        _updatable_fields:list = []

        #--------------------------------------------------------
        def __init__(self, aPK_obj=None, try_potential_problems=False):
                """Initialize.

                aPK_obj must contain the keys
                        pk_patient
                        pk_episode
                        pk_health_issue
                """
                if aPK_obj is None:
                        raise gmExceptions.ConstructorError('cannot instatiate cProblem for PK: [%s]' % (aPK_obj))

                # As problems are rows from a view of different emr struct items,
                # the PK can't be a single field and, as some of the values of the
                # composed PK may be None, they must be queried using 'is null',
                # so we must programmatically construct the SQL query
                where_parts = []
                pk = {}
                for col_name in aPK_obj:
                        val = aPK_obj[col_name]
                        if val is None:
                                where_parts.append('%s IS NULL' % col_name)
                        else:
                                where_parts.append('%s = %%(%s)s' % (col_name, col_name))
                                pk[col_name] = val

                # try to instantiate from true problem view
                cProblem._cmd_fetch_payload = """
                        SELECT *, False as is_potential_problem
                        FROM clin.v_problem_list
                        WHERE %s""" % ' AND '.join(where_parts)

                try:
                        gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj = pk)
                        return

                except gmExceptions.ConstructorError:
                        _log.exception('actual problem not found, trying "potential" problems')
                        if try_potential_problems is False:
                                raise

                # try to instantiate from potential-problems view
                cProblem._cmd_fetch_payload = """
                        SELECT *, True as is_potential_problem
                        FROM clin.v_potential_problem_list
                        WHERE %s""" % ' AND '.join(where_parts)
                gmBusinessDBObject.cBusinessDBObject.__init__(self, aPK_obj=pk)
        #--------------------------------------------------------
        def get_as_episode(self):
                """
                Retrieve the cEpisode instance equivalent to this problem.
                The problem's type attribute must be 'episode'
                """
                if self._payload['type'] != 'episode':
                        _log.error('cannot convert problem [%s] of type [%s] to episode' % (self._payload['problem'], self._payload['type']))
                        return None
                return cEpisode(aPK_obj = self._payload['pk_episode'])
        #--------------------------------------------------------
        def get_as_health_issue(self):
                """
                Retrieve the cHealthIssue instance equivalent to this problem.
                The problem's type attribute must be 'issue'
                """
                if self._payload['type'] != 'issue':
                        _log.error('cannot convert problem [%s] of type [%s] to health issue' % (self._payload['problem'], self._payload['type']))
                        return None
                return cHealthIssue(aPK_obj = self._payload['pk_health_issue'])
        #--------------------------------------------------------
        def get_visual_progress_notes(self, encounter_id=None):
                if self._payload['type'] == 'issue':
                        latest = cHealthIssue(aPK_obj = self._payload['pk_health_issue']).latest_episode
                        if latest is None:
                                return []

                        pk_episode = latest['pk_episode']
                else:
                        pk_episode = self._payload['pk_episode']
                doc_folder = gmDocuments.cDocumentFolder(aPKey = self._payload['pk_patient'])
                return doc_folder.get_visual_progress_notes(episodes = [pk_episode])

        #--------------------------------------------------------
        # properties
        #--------------------------------------------------------
        # doubles as 'diagnostic_certainty_description' getter:
        def get_diagnostic_certainty_description(self):
                return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])

        diagnostic_certainty_description = property(get_diagnostic_certainty_description)
        #--------------------------------------------------------
        def _get_generic_codes(self):
                if self._payload['type'] == 'issue':
                        cmd = """
                                SELECT * FROM clin.v_linked_codes WHERE
                                        item_table = 'clin.lnk_code2h_issue'::regclass
                                                AND
                                        pk_item = %(item)s
                        """
                        args = {'item': self._payload['pk_health_issue']}
                else:
                        cmd = """
                                SELECT * FROM clin.v_linked_codes WHERE
                                        item_table = 'clin.lnk_code2episode'::regclass
                                                AND
                                        pk_item = %(item)s
                        """
                        args = {'item': self._payload['pk_episode']}

                rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
                return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

        generic_codes = property(_get_generic_codes)

Ancestors

Instance variables

var diagnostic_certainty_description
Expand source code
def get_diagnostic_certainty_description(self):
        return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])
var generic_codes
Expand source code
def _get_generic_codes(self):
        if self._payload['type'] == 'issue':
                cmd = """
                        SELECT * FROM clin.v_linked_codes WHERE
                                item_table = 'clin.lnk_code2h_issue'::regclass
                                        AND
                                pk_item = %(item)s
                """
                args = {'item': self._payload['pk_health_issue']}
        else:
                cmd = """
                        SELECT * FROM clin.v_linked_codes WHERE
                                item_table = 'clin.lnk_code2episode'::regclass
                                        AND
                                pk_item = %(item)s
                """
                args = {'item': self._payload['pk_episode']}

        rows = gmPG2.run_ro_queries(queries = [{'cmd': cmd, 'args': args}])
        return [ gmCoding.cGenericLinkedCode(row = {'data': r, 'pk_field': 'pk_lnk_code2item'}) for r in rows ]

Methods

def get_as_episode(self)

Retrieve the cEpisode instance equivalent to this problem. The problem's type attribute must be 'episode'

Expand source code
def get_as_episode(self):
        """
        Retrieve the cEpisode instance equivalent to this problem.
        The problem's type attribute must be 'episode'
        """
        if self._payload['type'] != 'episode':
                _log.error('cannot convert problem [%s] of type [%s] to episode' % (self._payload['problem'], self._payload['type']))
                return None
        return cEpisode(aPK_obj = self._payload['pk_episode'])
def get_as_health_issue(self)

Retrieve the cHealthIssue instance equivalent to this problem. The problem's type attribute must be 'issue'

Expand source code
def get_as_health_issue(self):
        """
        Retrieve the cHealthIssue instance equivalent to this problem.
        The problem's type attribute must be 'issue'
        """
        if self._payload['type'] != 'issue':
                _log.error('cannot convert problem [%s] of type [%s] to health issue' % (self._payload['problem'], self._payload['type']))
                return None
        return cHealthIssue(aPK_obj = self._payload['pk_health_issue'])
def get_diagnostic_certainty_description(self)
Expand source code
def get_diagnostic_certainty_description(self):
        return diagnostic_certainty_classification2str(self._payload['diagnostic_certainty_classification'])
def get_visual_progress_notes(self, encounter_id=None)
Expand source code
def get_visual_progress_notes(self, encounter_id=None):
        if self._payload['type'] == 'issue':
                latest = cHealthIssue(aPK_obj = self._payload['pk_health_issue']).latest_episode
                if latest is None:
                        return []

                pk_episode = latest['pk_episode']
        else:
                pk_episode = self._payload['pk_episode']
        doc_folder = gmDocuments.cDocumentFolder(aPKey = self._payload['pk_patient'])
        return doc_folder.get_visual_progress_notes(episodes = [pk_episode])

Inherited members