Healthcare and life sciences

Data mapping

Entity

patient (pseudonymised identifier)

Group

hospital or clinic

Profile

generic; configure time windows with verticals.register

Reference implementation

from datetime import datetime, timedelta
from vertiq import plugins, rules
from vertiq.mapping import Mapping, apply, repair
from vertiq.tasks import predict_event, predict_value

# rows: the source export as a list of dicts, e.g. list(csv.DictReader(open("export.csv")))
mapping = Mapping.parse({"entity": "Patienten Nr", "time": "Admit Date", "type": "Encounter Type", "value": "LOS Days", "group": "Klinik", "attrs": ["Diagnosis Code", "Ward", "Follow-up Due", "Follow-up Done"], "date_format": "%d/%m/%Y"})
m, notes = repair(mapping, rows)      # validates column names, value types and date format against the rows
ds, problems = apply(m, rows)         # rows that fail to load are reported, not imputed

@plugins.feature("clinic")
def chronic_codes(entity, before, cut):   # receives only events recorded before the cutoff
    chronic = {"E11", "I50", "J44"}
    return {"chronic": float(any(e.attrs.get("Diagnosis Code") in chronic for e in before))}

task = predict_event(ds, target="admission", window_days=30, pack="clinic")
print(task.status, task.report())     # "trained" only if the model outperformed the baseline on held-out entities
if task.status == "trained":
    print(task.score("P29999"))  # probability with contributing factors

size = predict_value(ds, target="admission", window_days=120, pack="clinic")
print(size.status, size.report())

flags = rules.run(ds, [
    rules.repeat(30, type="admission"),   # readmission within 30 days
    rules.overdue("Follow-up Due", "Follow-up Done", tolerance_days=3),   # follow-up completed more than 3 days late
])
print(flags["summary"])

This code is executed by the test suite (tests/test_guides.py) against generated records with these column names.