CT Biomarkers Improve Frailty-Based Prediction of One-Year Mortality After Surgery

Researchers evaluated whether automated CT biomarkers improve postoperative mortality prediction beyond the ICD-based Risk Analysis Index (RAI-ICD). The retrospective cohort linked abdominopelvic CT scans obtained during 2013–2018 at a tertiary academic center to non-emergent surgery within six months. An XGBoost model, termed the Unified Multimodal Model (UMM), combined RAI-ICD with muscle, adiposity, bone, and aortic calcification biomarkers. The primary outcome was one-year all-cause mortality.

Among 7,672 patients, one-year mortality was 12.3%; primary mortality models included 7,638 patients. RAI-ICD alone outperformed every individual imaging biomarker, with an AUROC of 0.75. Adding all imaging biomarkers increased AUROC to 0.79 (difference, 0.04; P<0.001). Calibration also improved, particularly at higher predicted risk: calibration slope was 1.00 for UMM versus 0.72 for RAI-ICD, and integrated calibration index was 0.01 versus 0.03. At the Youden-optimal threshold, UMM had a positive predictive value of 0.24 and negative predictive value of 0.96.

Decision-curve analysis showed greater net benefit overall and across Robust, Frail, and Very Frail strata. These results suggest routinely acquired CT imaging may complement frailty assessment, but they demonstrate predictive performance rather than improved patient outcomes. The retrospective, single-center design limits generalizability, and external validation is required before implementation.