Story generator for leiomyosarcoma patient CSV: person/condition_occurrence/visit_occurrence/procedure_occurrence

[myyong]

Summary

Implemented patient_story() in examples/leiomyosarcoma/story_generators.py to drive synthetic OMOP data generation from examples/leiomyosarcoma/leiomyosarcoma_patient.csv, a flat event log (person_id, event_datetime, days_after_diagnosis, event_type, event_name, concept_id, visit_occurrence_id) for a single leiomyosarcoma patient.

What was built

• Typed row dataclasses (Person, ConditionOccurrence, VisitOccurrence, ProcedureOccurrence) mirroring each table's columns in orm.yaml, with a shared Row.to_row() that only emits explicitly-set fields — unset fields fall through to the table's normal row generator defaults.
• CSV loading: _load_patient_events() reads the CSV once and groups rows by (source) person_id; patient_story() consumes one patient's events per call via a module-level iterator.
• condition_occurrence: one row per CSV event with event_type == "Condition", concept_id → condition_concept_id.
• visit_occurrence: one row per event_type == "Visit" event, concept_id → visit_concept_id.
• procedure_occurrence: for each visit, procedures on the same days_after_diagnosis are matched and inserted with visit_occurrence_id set from the visit's actual (DB-assigned) FK, returned via the story generator's yield/send protocol.

Open design decisions / assumptions made

The CSV has no absolute dates and is missing several NOT NULL OMOP fields, so we filled gaps with fixed values pending review:

Field	Assumption
condition_start_date / visit_start_date / procedure_date	Derived as days_after_diagnosis offset from an anchor date of today (no real anchor diagnosis date available)
visit_end_date	Set equal to visit_start_date (treats all visits as single-day)
condition_type_concept_id	Hardcoded to 32840
visit_type_concept_id	Hardcoded to 32817
procedure_type_concept_id	Hardcoded to 32817

These should be revisited once real anchor dates / source type-concept mappings are available.

Depends on

• concept is configured as vocabulary_table: true, so real OMOP concept_ids (e.g. 40482829 for Leiomyosarcoma) referenced by the story rely on make-vocab/create-vocab having been run against a source DB that actually contains those concepts. Not yet verified end-to-end against a live database.

Next steps

• Run create-vocab + create-data against a real destination DB to validate the full FK chain
• Confirm/replace the hardcoded type-concept and anchor-date assumptions above
• Extend to remaining tables noted in patient_story's docstring (death, observation_period, measurement, device_exposure, observation, specimen, drug_exposure) if needed for this dataset

---

🤖 Generated with Claude Code