The model, FAHR-Face, was trained on more than 40 million facial images and adapted into two tools: FAHR-FaceAge estimates biological age, while FAHR-FaceSurvival predicts mortality risk directly from a facial image.
The tools use different facial information. In testing, the tools produced distinct but complementary prognostic information, and the highest-risk group identified by FAHR-FaceSurvival had more than three times the mortality of the lowest-risk group after adjustment for clinical factors.
Among cancer patients who received radiotherapy, those judged by FAHR-FaceAge to appear at least five years older than their chronological age had a 21% higher risk of death during an average four-year follow-up. The model’s age estimates remained broadly stable across changes in lighting, pose, facial expression and makeup, with only a small shift observed after rhinoplasty in the tested procedures.
The findings build on research showing that cancer patients often appear older than their chronological age and that older-appearing patients have worse survival outcomes. The researchers also found that smoking, heavy alcohol use and drug use were associated with older biological-age estimates in facial photographs.
A separate analysis examined whether changes in facial age over time were associated with survival. It used routine clinical photographs from 2,276 cancer patients who underwent at least two radiation-therapy courses, with photographs taken between 10 days and four years apart. The resulting Face Aging Rate was associated with worse survival after adjustment for the interval between photographs, sex, race and diagnosis.
These findings describe associations with outcomes and do not establish that a photograph can determine an individual patient’s survival.