Understanding the data
Medicare-like research files usually separate who the person is, whether they were enrolled, condition indicators, and service/claim events. Medicare-Synth mirrors that shape in its fixtures and MVP worlds so analysis code can practice joins and definitions.
A minimal relationship picture
Beneficiary (one row per person in the extract)
│
├── Enrollment (membership over time / year rules)
├── Chronic-condition indicators (e.g. diabetes, heart failure flags)
└── Claims / encounters (zero, one, or many rows per person)
└── optional claim lines / detail rows in richer fixtures
| Idea | Plain meaning |
|---|---|
| What one row represents | The grain: person, enrollment spell, claim, or line |
| How records connect | Shared identifiers (keys) linking persons to events |
| Administrative indicator | A coded flag or algorithm result—not identical to a clinician’s chart diagnosis |
| Claim presence | An event occurred in the synthetic extract—not proof of clinical necessity |
Why counts differ
- Patient count ≠ claim count. One hospitalized person may contribute one or many inpatient rows depending on the generator and fixture rules.
- In the current MVP generator, a hospitalized simulated beneficiary has at most one inpatient claim in the simplified world—useful for code tests, not for realistic readmission trajectories.
- Condition indicators can be present without a matching diagnosis-coded inpatient stay; the MVP outcome is all-cause admission presence, not diabetes-principal hospitalization.
Where technical field definitions live
For table families, keys, and source-to-release flow, use the developer data model guide. For lineage statuses and what validation does (and does not) prove, see provenance and validation.