External Validation Study of a Chest Radiograph-Derived Aging Biomarker for ICU Mortality Prediction

Background: Accurate ICU mortality predictions are essential for guiding clinical decisions and resource allocation. While traditional scoring systems rely on chronological age, this metric often fails to capture the diverse physiological reserves of critically ill patients. We aimed to determine whether integrating a deep learning-based chest radiograph biomarker, Xp-age, into SAPS II, APACHE II, and SOFA would enhance mortality prediction. Methods: In this retrospective external validation study at single-center ICU, we analyzed ICU data from patients aged ≥18. Patients lacking chest radiographs or complete SAPS II, APACHE II, or SOFA scores were excluded. Chest radiographs were analyzed using a deep learning model trained on over 36,000 healthy adults to derive Xp-age. Chronological age was replaced by Xp-age in SAPS II and APACHE II, and added to SOFA. Predictive performance for ICU mortality was assessed via 10-fold cross-validation with XGBoost. Results: Data were available for SAPS II (n =3,933), APACHE II (n =1,707), and SOFA (n =5,590). In SAPS II, the AUROC increased from 0.79 (95% CI, 0.76–0.82) to 0.81 (95% CI, 0.79–0.84), (p =0.015) and in APACHE II from 0.77 (95% CI, 0.73–0.81) to 0.81 (95% CI, 0.77–0.84), (p <0.001). For SOFA, adding Xp-age improved the AUROC from 0.83 (95% CI, 0.81–0.85) to 0.85 (95% CI, 0.83–0.87), (p =0.012). Survival analyses revealed distinct high- and low-risk groups of Xp-age models (p <0.001). Conclusions: Xp-age improves ICU mortality prediction, supporting more personalized risk stratification and targeted interventions in critical care.