Answers are grounded in this paper and cite their sources.
Background Conventional staging does not fully capture host-related vulnerability in colorectal cancer (CRC). We evaluated artificial intelligence-assisted multilevel computed tomography (CT) body-composition assessment for recurrence-risk stratification. Methods This retrospective multicenter study included 1,328 patients with pathological stage I-III, M0 colorectal cancer who underwent curative-intent R0 resection at three centers, together with an exploratory cohort of 50 patients with unresectable CRC receiving PD-1/PD-L1 antibody-based combination immunotherapy. Center A was used for sex-stratified model development with nested cross-validation, and Centers B and C were used as geographically separate evaluation cohorts. Preoperative portal-venous-phase CT was used to quantify subcutaneous fat, torso fat, skeletal muscle, and intermuscular fat across T8-L5. Model performance was evaluated at 36 months using the Harrell C-index, calibration, censoring-adjusted Brier score, and exploratory decision-curve analysis. Results After adjustment for age, sex, height, BMI category, and center, none of 17 prespecified body-composition measurements showed an FDR-supported linear trend across pTNM stages. Sex-stratified RFS analyses identified several adjusted body-composition associations. The locked clinicopathological-plus-multilevel model achieved C-indices of 0.630, 0.758, and 0.779 in men and 0.604, 0.762, and 0.783 in women in Center A nested out-of-fold evaluation and the Center B and C geographic evaluations, respectively. In the exploratory immunotherapy cohort, four muscle measurements were associated with disease control after FDR correction. Conclusion Multilevel CT body-composition profiling identified exploratory host-related prognostic signals in CRC.
No takes yet. Share an insight, caveat, or question.
Fang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: