Delphi-2M points to a next phase for synthetic data in healthcare: models that generate longitudinal patient histories and use them to forecast population risk at scale. The upside is faster analytics with less direct exposure of real patient data—if teams can prove utility and privacy in practice.
Delphi-2M simulates decades of health trajectories using synthetic patient records
Delphi-2M is a generative AI system trained on health datasets from the UK and Denmark that uses synthetic patient records to predict disease risk across more than 1,000 conditions. The model is described as simulating decades of health trajectories, with an emphasis on capturing interactions between diseases to improve the realism and usefulness of the generated records for research and analytics.
The story frames Delphi-2M as an enabler for predictive population health management—supporting more tailored public health strategies and creating opportunities for vendors to monetize “tailored insights” derived from synthetic populations, rather than distributing or repeatedly querying sensitive raw patient data.
- Population analytics without constant access to PHI: If synthetic cohorts can stand in for real longitudinal data in common workflows, teams can reduce how often analysts and downstream systems touch identifiable health records.
- Validation becomes the product: For founders and data leads, the differentiator shifts to proving synthetic fidelity for specific tasks (risk prediction, stratification, policy simulation), not just generating “realistic-looking” rows.
- Privacy and compliance don’t disappear: “Synthetic” is not automatically safe—teams still need clear privacy guarantees, documentation, and governance around training data provenance and release/usage controls.
- New buyers, new expectations: Public health authorities and insurers adopting precision prevention will expect reproducible evaluation, drift monitoring, and clear limits on what the synthetic model can and cannot support.