2672 - Integration of PhenoAge and AI-Derived FaceAge to Improve Prognostic Stratification in Metastatic Cancer Patients Receiving Palliative Radiotherapy
Presenter(s)
J. Zeng1, A. Warrington1, F. Haugg2, G. Lee3, D. Bontempi4, D. S. Bitterman1, S. Pai1, C. V. Guthier1, Y. H. Chen5, L. M. Hertan6, T. A. Balboni7, B. H. Kann1, H. Aerts1, M. S. Krishnan8, and R. H. Mak1; 1Department of Radiation Oncology, Mass General Brigham/Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, 2Brigham & Women's Hospital, Boston, MA, 3Inova Schar Cancer Institute, Fairfax, VA, 4Department of Radiation Oncology (MAASTRO), Maastricht University, Maastricht, Netherlands, 5Dana Farber Cancer Institute, Boston, MA, 6Department of Radiation Oncology, Beth Israel Deaconess Medical Center, Boston, MA, 7Brigham and Women's Hospital, Boston, MA, 8Department of Radiation Oncology, Dana-Farber Cancer Institute/Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
Purpose/Objective(s): Accurate prognostication in metastatic cancer is critical for guiding palliative care decisions. Traditional models such as TEACHH rely on chronological age, which may not fully capture physiologic aging. We evaluated the prognostic performance of PhenoAge, a biomarker of biological age derived from laboratory data, in addition to FaceAge, an AI-based facial biomarker of biological age, for predicting overall survival and enhancing TEACHH risk stratification.
Materials/Methods: We retrospectively analyzed 182 patients with metastatic cancer receiving palliative radiotherapy. FaceAge was estimated from frontal facial photographs using a validated deep learning model. PhenoAge was calculated from 9 clinical laboratory values. TEACHH scores were computed using chronological age, FaceAge, and PhenoAge, and reclassification events were identified. Overall survival (OS) was compared across TEACHH groups. Model performance was evaluated using Akaike Information Criterion (AIC), Cox proportional hazards regression, and net reclassification improvement (NRI).
Results: Substituting PhenoAge for chronological age led to TEACHH reclassification in 25.7% of patients (95/182), with most movement occurring from Group B to Group C (62 patients) and from Group A to Group B (21 patients). Median overall survival (OS) for TEACHH Groups B and C was 228 and 80 days using chronological age; when PhenoAge was incorporated, median OS for reclassified patients in Groups B and C was 485 and 35 days, respectively, indicating improved separation between intermediate- and poor-risk groups.
In Cox proportional hazards models, PhenoAge (HR 1.02, p = 0.028) and FaceAge (HR 1.01, p < 0.001) were independently associated with overall survival, whereas chronological age alone was not (HR 1.01, p = 0.074). Model fit improved with incorporation of PhenoAge: the AIC for Chronologic Age + PhenoAge was 782.97 compared with 792.13 for Chronologic Age + FaceAge and 792.57 for Chronologic Age alone. Net reclassification improvement (NRI) analysis demonstrated that PhenoAge improved risk classification relative to chronological age (NRI = 0.445) and FaceAge (NRI = 0.423).Conclusion: PhenoAge and FaceAge capture complementary aspects of biologic aging that are not fully represented by chronological age. Incorporating PhenoAge into TEACHH further improves prognostic discrimination and risk reclassification, identifying patients whose survival is misestimated by traditional models. These findings support the potential role of AI and lab biomarker-derived biologic age measures to enhance prognostic stratification and inform palliative care planning in metastatic oncology.
| TEACHH | PhenoAge A | PhenoAge B | PhenoAge C | Total |
| A | 9 | 21 | 0 | 30 |
| B | 2 | 77 | 62 | 141 |
| C | 0 | 0 | 11 | 11 |
| Total | 11 | 98 | 73 | 182 |