2077 - Deep Learning Based Estimation of Facial Age for Prognostication in Patients with Early-Stage Non-Small Cell Lung Cancer Undergoing Definitive Stereotactic Body Radiotherapy

G. Lee¹, F. Haugg², D. Bontempi^2,3, J. He¹, D. S. Bitterman^1,2, S. Pai^2,4, C. V. Guthier^1,2, K. J. Fitzgerald⁵, D. E. Kozono¹, B. H. Kann^1,2, H. Aerts^1,2, and R. H. Mak^1,2; ¹Department of Radiation Oncology, Dana-Farber Cancer Institute/Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, ²Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA, ³Department of Radiation Oncology (MAASTRO), Maastricht University, Maastricht, Netherlands, ⁴Radiology and Nuclear Medicine, CARIM & GROW, Maastricht University, Maastricht, Netherlands, ⁵Department of Radiation Oncology, Brigham and Women’s Hospital/Dana-Farber Cancer Institute, Boston, MA

Purpose/Objective(s): Patients who undergo definitive stereotactic body radiotherapy (SBRT) for early-stage non-small cell lung cancer (NSCLC), in lieu of surgery, are often elderly and have substantial medical comorbidities. Assessment of their frailty and life expectancy is important to make well-informed, personalized treatment decisions. We previously developed a deep-learning model capable of estimating biological age from facial images (FaceAge) which demonstrated potential as a prognostic marker among cancer patients. Here, we investigate the prognostic value of FaceAge in the early-stage NSCLC population. Materials/

Methods: Retrospective review was conducted of 670 patients age =60 who underwent definitive SBRT for early stage (I-II) NSCLC between 2009-2023 at six clinic locations. FaceAge predictions were made on facial identification photographs taken before SBRT. Cox and logistic regression analyses were performed for overall survival (OS) and risk of early mortality within 2 years, respectively.
Results: Median follow-up was 25 months for all and 31 months for surviving patients. Baseline patient characteristics were: 61% female and 39% male; ECOG PS of 0 in 21%, 1 in 49%, and 2+ in 30%; 94% with smoking history (median 45 pack-years); stage IA1-3 in 90%, IB in 7%, and IIA/B in 3%; 42% with prior NSCLC or multifocal disease; 50% with biopsy-confirmed histology (61% adeno, 30% squamous, 9% others). Median age was 77 (range 60-98) and median FaceAge was 79 (range 47-98). 64% had FaceAge older than their chronological age. 17% vs 22% were =85 years old by chronological age vs FaceAge, respectively. Median OS was 47 months and 24% died within 2 years. On multivariable analysis, older FaceAge as a continuous variable was significantly associated with worse OS (HR per decade 1.39 [95% CI 1.14-1.71], p<0.01) while chronological age was not, adjusting for other significant clinical factors including sex, ECOG PS, smoking pack years, and histology. Older FaceAge as a continuous variable (OR per decade 1.52 [95% CI 1.07-2.18], p=0.02) and notably, FaceAge =85 years old (OR 1.81 [95% CI 1.05-3.14], p=0.03) were associated with increased risk for early mortality within 2 years, while chronological age and age =85 years old were not, adjusting for the same clinical factors. 2-year OS for patients with FaceAge =85 vs <85 years old was 67% vs 75% (p<0.01), respectively.
Conclusion: Deep learning-based quantification of biological age from facial images may be an effective biomarker for overall physiological health and early mortality among early-stage NSCLC patients undergoing SBRT. In particular, patients with predicted FaceAge =85 years old had increased risk for early mortality, and such prognostic information may assist in guiding treatment decisions for this high-risk patient cohort with multiple medical comorbidities.