101 - FaceAge Deviation and Area Deprivation Index: Associated and Synergistic Predictors of Mortality in Patients Undergoing Radiation Therapy
Presenter(s)
E. B. Wang1, F. Haugg2, A. Warrington3, J. Zeng3, G. Lee3, L. L. Thompson3, D. S. Bitterman3, P. J. Catalano4, C. V. Guthier3, B. H. Kann3, H. Aerts3, and R. H. Mak3; 1Department of Internal Medicine, Brigham and Women's Hospital, Boston, MA, 2Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA, 3Department of Radiation Oncology, Mass General Brigham/Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, 4Department of Data Science, Dana-Farber Cancer Institute, Boston, MA
Purpose/Objective(s): Accurate prognostication is central to decision making in radiation oncology. Yet, existing prediction models are often complex, costly, and reliant on invasive testing. Foundational AI for Health Recognition (FAHR)-FaceAge, a fine-tuned deep learning model that estimates biological age (FaceAge) from facial photographs has exhibited promise in this domain. In this study, we investigate social factors associated with FaceAge and evaluate whether integrating FaceAge with clinical and socioeconomic data enhances its prognostic utility.
Materials/Methods: We analyzed 33,462 patients treated with radiation therapy at an academic medical center and network sites from 2008 to 2023. Advanced biological aging was quantified using Face Age Deviation (FAD), calculated as FaceAge minus chronological age. Socioeconomic disadvantage was represented using Area Deprivation Index (ADI). Associations between FAD and ADI and between patient characteristics and ADI were evaluated with linear regression. FAD, ADI, and clinical variables were assessed using univariate Cox regression, with significant variables included multivariable analysis. Sub-group analyses among patients with DCIS, non-metastatic, and metastatic cancer were performed.
Results: Across all patients, median age, FaceAge, and FAD were 64.5, 65.8, and 1.23 years respectively. Median ADI was 0.20 (Range 0-1). Greater FAD was associated with higher ADI (social disadvantage) on multivariate linear regression (p < 0.01). In multivariate Cox regression, both FAD (HR 1.154 per 1 SD, p < 0.01) and ADI (HR 1.017 per 1 SD, p < 0.01) were associated with increased mortality while controlling for each other, age, sex, race, smoking, alcohol, drug use, disease status (i.e. benign, non-metastatic, metastatic), and course intent. Evaluating FAD and ADI independent of other covariates, model discrimination was greater when ADI and FAD were combined (c-index: 0.573) than with each variable alone (FAD c-index: 0.573, ADI c-index: 0.512). Multivariate model discrimination was greater in the DCIS cohort (c-index: 0.92) than in the non-metastatic (c-index: 0.73) and metastatic (c-index: 0.67) cohorts.
Conclusion: Among patients receiving radiation therapy, FAD remained an independent predictor of overall survival after adjusting for ADI and other covariates. FAD demonstrated greater prognostic influence than ADI, exhibiting larger effect sizes in multivariate Cox regression. Nonetheless, FAD and ADI functioned synergistically, with improved model discrimination when both variables were included. Of note, model performance was exceptional among patients with DCIS, suggesting that, in addition to complementing the social determinants of health captured by ADI, FAD may capture latent biological vulnerabilities unrelated to cancer severity. Prospective validation is needed to clarify the role of FAD in clinical practice and to further investigate the biological mechanisms underlying its prognostic value.