Synthetic Patients generates privacy‑preserving synthetic patient records by training models on real clinical data under differential privacy, ensuring no individual information can be recovered. The resulting datasets are validated for statistical similarity, clinical consistency, and downstream utility, and can be exported in FHIR R4/R5 or custom formats. This enables healthcare AI, pharma trial planning, hospital research, and digital‑health product testing without handling protected health information.
Funding
Funding not disclosed
Founders
Product
Problem
Healthcare organizations and researchers need realistic patient datasets for AI development, trial simulation, and product testing, but sharing real electronic health records is restricted by privacy regulations such as HIPAA, GDPR, and the EU AI Act. Existing de‑identification methods still risk re‑identification, limiting data accessibility and slowing innovation.
Solution
Synthetic Patients offers a differentially private generative model that learns population‑level patterns from real clinical records while ensuring that no individual’s information can be recovered. The trained model can be exported to a customer’s environment, where it produces synthetic electronic health records in standard FHIR R4/R5 formats or custom structures. Generated cohorts are validated on three pillars—statistical similarity to the source data, adherence to clinical rules, and downstream utility for machine‑learning tasks—so users receive data that behaves like real patients without containing protected health information. The service integrates with regulated workflows via API access, supports on‑premise deployment with SSO, RBAC, and audit logging, and complies with GDPR, HIPAA, and the EU AI Act.
Target Audience
Primary customers are healthcare AI developers, pharmaceutical companies planning clinical trials, hospital research teams, and digital‑health product engineers who need realistic, privacy‑safe patient data for development, testing, and validation.
Features
- Differential‑privacy training guarantees that the model cannot leak any individual’s data, allowing the model itself to be transferred safely
- Validation framework covering statistical similarity, clinical rule consistency, and downstream task performance
- Output in native FHIR R4/R5 or custom formats for seamless integration with existing health‑IT systems
- On‑premise deployment option with SSO, role‑based access control, and audit logs to meet strict compliance requirements
- API with tiered rate limits and subscription plans for flexible usage across research, pharma, and digital‑health projects
- Support for reinforcement‑learning environments (K01 Trajectory) and bespoke model training within a client’s secure infrastructure