Research preparation with Medicare-like synthetic data

Explore ideas before restricted Medicare access.

Medicare-Synth helps researchers work with artificial, Medicare-shaped datasets for structure checks and methods testing. For one limited study setting, it can adjust synthetic populations toward selected published statistics. It is not a source of Medicare population estimates.

Not Medicare findings
Structural validity and agreement with selected published rates do not authorize citing synthetic results as empirical Medicare statistics, clinical conclusions, or policy evidence.

What you can do today

Status Meaning
Works today Study file structure and joins; run fixture-based code exercises; export ordinary teaching CSV/Parquet with developer help
Limited prototype Generate and compare 2021 FFS-like diabetes / any-inpatient worlds adjusted toward three selected public targets
Not intended One-click calibrated website downloads; any disease/year as a calibrated product; population, causal, or clinical claims

Full matrix: What you can do.

Three researcher tasks

Explore available data

Learn how person, enrollment, condition, and claims-style records connect—and why patient count is not claim count.

Understanding the data →

Plan an analysis

Walk the supported diabetes and any-inpatient example before you invest in restricted extracts.

Worked example →

Compare assumptions

Same selected calibration targets can still produce different outcomes when free assumptions change.

Interpreting results →

Flagship example (stored Phase 5 illustration)

Artificial 2021 Original Medicare FFS-like adults aged 65+, with diabetes and heart-failure indicators and any inpatient admission (not diagnosis-specific). Selected anchors include approximate diabetes prevalence 27.7% (2018 source), heart failure 12% (2022 source), and ~13% inpatient risk among people with diabetes (weaker industry prior).

In the checked-in Phase 5 family at 6,000 simulated beneficiaries, two worlds that share those calibration targets show 404 versus 572 inpatient claim records under different free background-risk assumptions. Use that contrast to practice sensitivity—not to bound Medicare truth.

→ Diabetes–hospitalization example

How to begin

Building or extending the toolkit?

Developer documentation
Installation, CLI, scenarios, data model, calibrated-world reproduction, and contribution guides live under For developers. Canonical technical state remains in the repository root (SPEC.md, ARCHITECTURE.md, ROADMAP.md).

Scope notice

Phases 0–5 are complete for the pinned MVP; Phase 6 independent evaluation is active. Read limitations before using any release or synthetic summary in a manuscript or proposal.