Overview
- The study was published in The Lancet Digital Health and presents two tools from Mass General Brigham called FAHR‑FaceAge and FAHR‑FaceSurvival that were built from a foundation facial model trained on more than 40 million images and fine‑tuned with over 30,000 cancer patient photos.
- Researchers released a public demo site after publication and reported about 50,000 photo submissions, reflecting immediate public interest but no change to the tool’s experimental status.
- In retrospective tests the survival model strongly stratified short‑term risk with roughly 52% of the highest‑scoring group dying within 90 days compared with about 2% in the lowest‑scoring group, and the FaceAge tool associated appearing ≥5 years older with about a 21% higher mortality risk and appearing ≥5 years younger with about a 17% lower risk.
- The two models attend to different facial regions when making predictions and identify correlations with smoking, alcohol and drug use, but authors and independent experts stress these are statistical patterns, not causal biomarkers, and the models must be tested prospectively and compared head‑to‑head with clinicians before any clinical use.
- Commentators and the study team highlight unresolved ethical, bias and privacy issues including the risk of encoding racial disparities, the need for transparent auditing and consent rules, and the requirement for regulatory and ethical guardrails to prevent misuse such as insurance discrimination.