FaceAge as a biomarker for prognosis and treatment stratification in a large-scale oncology cohort

BACKGROUND: Humans age at different rates, and facial characteristics may yield insight into biological age and physiologic health. FaceAge, a deep learning system designed to estimate biological age based on facial photographs, has shown potential as a biomarker for cancer prognosis. This study investigated the prognostic value of extreme discordance between FaceAge and chronological age (FaceAge - Age) in predicting survival and early mortality across a large clinical dataset of 28 cancer types. METHODS: Data from 24 556 patients with cancer aged 60 years or older treated with radiation therapy between 2008 and 2023 were analyzed. FaceAge estimates were compared with chronological age across different diagnoses/clinical contexts, and survival analyses were performed. All tests were 2-sided. RESULTS: The FaceAge estimate was older than the chronological age in 65% of participants (median FaceAge = 74 years vs chronological age = 70 years). Younger patients, female patients, patients with diagnoses associated with worse prognosis, and patients receiving treatment with palliative intent had a higher likelihood of FaceAge - Age ≥10 years. Patients with FaceAge - Age ≥10 years had statistically significantly worse survival, while patients with FaceAge - Age ≤‒5 years had better survival. On multivariate analysis, FaceAge - Age ≥10 years predicted higher mortality risk (hazard ratio = 1.26; P < .001) and early mortality at 30 days (odds ratio = 1.38, P = .004) and 60 days (odds ratio = 1.33, P < .001), whereas FaceAge - Age ≤‒5 years predicted lower mortality risk (hazard ratio = 0.90; P = .002). CONCLUSIONS: Patients with more advanced cancers tend to have statistically significantly older FaceAge than chronological age, and extreme discordance between FaceAge and chronological age is a novel, independent predictor of survival and early mortality. These findings support further development of facial health assessments for clinical prognostication models and personalized treatment decision making.