Abstract
Objective: To evaluate whether preoperative body composition metricsautomatically extracted from CT scans can predict postoperative outcomes aftercolectomy, either alone or combined with clinical variables or existing riskpredictors. Main outcomes and measures: The primary outcome was the predictiveperformance for 1-year all-cause mortality following colectomy. A Coxproportional hazards model with 1-year follow-up was used, and performance wasevaluated using the concordance index (C-index) and Integrated Brier Score(IBS). Secondary outcomes included postoperative complications, unplannedreadmission, blood transfusion, and severe infection, assessed using AUC andBrier Score from logistic regression. Odds ratios (OR) described associationsbetween individual CT-derived body composition metrics and outcomes. Over 300features were extracted from preoperative CTs across multiple vertebral levels,including skeletal muscle area, density, fat areas, and inter-tissue metrics.NSQIP scores were available for all surgeries after 2012.