Synthea (Synthetic Patient Generator)

In one line

Synthea makes lifelike patients who never existed — full medical histories you can export as FHIR or C-CDA — so engineers can develop and test against real-shaped data with zero privacy risk. Licence: Apache 2.0 (fully open).

The problem it solves

You can't build or test a health system without data — but you can't use real patient data freely, because it's PHI: regulated, risky, and slow to get approved. That's a chicken-and-egg trap for every new project. Synthea breaks it by generating patients who are statistically realistic but completely fictional — safe to share, commit to a repo, and seed into a demo at any scale.

How it generates patients

Synthea (from MITRE) simulates each person's whole life, birth to death:

• Disease-progression modules — clinically-informed state machines (the Generic Module Framework) model how conditions develop and are treated over time, so a synthetic diabetic accrues a believable history of encounters, HbA1c results, medications and complications — not random noise.
• Demographics & geography — populations are generated against real demographic and prevalence statistics (US-based by default; configurable).
• Standards-conformant output — exports as FHIR R4 Bundles, C-CDA, CSV, and bulk-FHIR (NDJSON), so the data drops straight into real tools.

Why synthetic data matters

Because no real person is involved, the data carries no HIPAA/DPDP constraints — it can be shared, versioned, and reproduced exactly. This is the cleanest answer to the de-identification problem: rather than risk re-identifying anonymised real data, use data that was never real. It is the principle this whole platform runs on — synthetic data only in the labs.

Where it shows up in digital health

• Populating a FHIR server — seeding HAPI or Medplum for development; a real deployment of the FHIR Sandbox or FHIR Mapper lab would load Synthea Bundles.
• Demos & sales — showing an app with a believable patient panel, no privacy review.
• Load & integration testing — generate thousands of patients to stress a pipeline.
• Research & ML prototyping — develop methods on synthetic data before applying for access to real cohorts (OMOP conversions exist).

Limits to keep in mind

• It's a model, not reality — distributions are US-centric by default and won't capture every local pattern; conclusions drawn only from Synthea don't transfer to real populations.
• Not for statistical inference — it's for building and testing software, not for estimating true disease epidemiology.
• Correlations are simplified — the modules are realistic but not exhaustive; rare multi-morbidity interactions may be missing.

Key takeaways

• Synthea generates realistic, fully synthetic patient histories — FHIR, C-CDA, CSV, bulk.
• It removes the PHI barrier from development, demos, and testing entirely.
• Built on clinically-informed state machines, so histories are coherent, not random.
• Use it to seed FHIR servers and labs — but remember it models, and doesn't measure, reality.