Development and validation of a predictive model for tumor regression following neoadjuvant chemoradiotherapy in...

The standard treatment for locally advanced rectal cancer (LARC) involves neoadjuvant chemoradiotherapy (nCRT) followed by surgery. While nCRT improves tumor downstaging and R0 resection rates, heterogeneous responses among patients necessitate predictive tools to identify poor responders. This study aimed to develop a predictive model for tumor regression after nCRT using baseline clinical indicators. We retrospectively analyzed LARC patients treated with nCRT and total mesorectal excision (TME) at the Second Affiliated Hospital of Nanchang University (2016-2020). Baseline clinical data, including laboratory tests, tumor markers, and MRI findings, were collected. Tumor regression was evaluated post-nCRT using the RECIST criteria. Multivariate logistic regression identified predictors, and a predictive model was constructed. Model performance was assessed via ROC curves, decision curve analysis (DCA), calibration curves, and tenfold cross-validation. Finally, perform external validation on the model. Among 158 patients, 98 achieved objective response (CR/PR; 62%), while 60 showed poor regression (SD/PD). Independent predictors of tumor regression included pretreatment tumor diameter (P < 0.001), interval to surgery post-nCRT (P = 0.006), D-dimer (P = 0.010), prognostic nutrition index (PNI, P = 0.035), carcinoembryonic antigen (CEA, P = 0.004), and extramural vascular invasion (EMVI, P = 0.026). The predictive model demonstrated an AUC of 0.84 (95% CI 0.780-0.899), sensitivity of 85.0%, and specificity of 72.4%. Calibration curves indicated high accuracy, and DCA confirmed clinical utility. External validation demonstrated the model's reasonably good predictive value (AUC = 0.864, 95% CI 0.787-0.879). Pretreatment tumor diameter, interval to surgery, D-dimer, PNI, CEA, and EMVI are independent predictors of tumor regression after nCRT. The proposed model effectively stratifies LARC patients, guiding personalized treatment decisions to optimize outcomes.