Clinical-radiological predictive model for preoperative risk stratification in rectal adenocarcinoma

BACKGROUND: Accurate identification of locally advanced rectal cancer (LARC) is crucial for treatment planning, yet conventional Magnetic resonance imaging (MRI) assessment remains subjective and experience-dependent, leading to inconsistent staging and suboptimal treatment decisions. An objective approach for preoperative risk stratification in rectal cancer patients, as an alternative to conventional MRI-based LARC identification, is therefore critically needed. METHOD: We retrospectively analyzed 294 rectal adenocarcinoma patients from three cohorts who underwent preoperative MRI and surgery. Dynamic contrast-enhanced (DCE) MRI based and clinical features were analyzed for correlation with pathology and by Cox regression for feature selection, then used to build survival prediction models. Model performance was compared against MRI- and pathology-based LARC status for predicting postoperative 3-year disease-free survival (DFS). Mediation analysis assessed whether pathological characteristics mediated imaging-clinical feature effects on DFS. RESULTS: The kinetic DCE-MRI feature Washout inversely correlated with pathological T-stage. Preoperative carcinoembryonic antigen (CEA) (HR 1.02; 95 %CI: 1.001-1.039) and Washout (HR 0.014; 95 %CI: 0.001-0.332) were independent predictors of 3-year DFS. High-risk patients identified by the models had significantly worse survival than low-risk patients (p < 0.01). The models outperformed conventional MRI-based assessment (AUC 0.757-0.819 vs 0.600-0.672; C-index 0.755-0.774 vs 0.586-0.673). T/N stage partially mediated effects of CEA (17.7 %) and Washout (51.1 %) on DFS. CONCLUSION: The developed models provide an objective, valuable tool for preoperative risk stratification as alternative to subjective LARC identification, enhancing preoperative risk stratification.