GS04 - Construction and Evaluation of a Predictive Model for Remission in Rectal Cancer Patients Undergoing Neoadjuvant Chemoradiotherapy

Image or Table (Do Not add text here):

Purpose/Background: Neoadjuvant chemoradiotherapy (nCRT) is standard for locally advanced rectal cancer (LARC), but response heterogeneity exists (only 15%-27% achieve pCR, 20% treatment-insensitive), and existing models lack long-term (≥5 years) prognosis evaluation. This study aimed to identify independent predictors of pathologic response (pR: CR+PR) after nCRT, construct/validate a predictive model, and explore its value in predicting 8-year recurrence, metastasis, and overall survival (OS) for individualized treatment.

Methods/Interventions: Retrospective analysis of 196 LARC patients (2013–2023, First Affiliated Hospital of Naval Medical University) who received nCRT (long-course radiotherapy: 45–50.4 Gy/25–28 fractions + concurrent capecitabine; partial with consolidation chemotherapy) followed by radical surgery; inclusion: tumor ≤15 cm from anus, pretreatment MRI T3/T4/N (no distant metastasis), complete data; exclusion: incomplete nCRT, emergency surgery, pre-treatment inflammation. Additionally, 58 LARC patients who met the same inclusion/exclusion criteria and were treated at a tertiary Grade A hospital during the same period were selected as the external validation cohort. Statistical analysis: SPSS 26.0 for univariate/multivariate Logistic regression (predictor screening); R software (pROC, rms) for cut-off values, nomogram, ROC (AUC), DCA; 10-fold cross-validation; Kaplan-Meier for 5-year and 7-year PFS/OS.

Results/Outcomes: Patients included 132 males (67%, mean age 61.2 years). Multivariate analysis identified 4 independent pR predictors (P < 0.01): preoperative CEA ≥3 ng/mL (OR=0.390), maximum tumor diameter (MTD) ≥3 cm (OR=0.367), clinical T3–4 (OR=0.354 vs T1–2), clinical N1–2 (OR=0.364 vs N0). The nomogram had AUC=0.833 (95%CI:0.735–0.884); 10-fold cross-validation: accuracy=0.71, Kappa=0.42; calibration curves/DCA confirmed agreement and clinical benefit. For the external validation cohort, the AUC was 0.808 (95% CI: 0.722–0.864) with an accuracy of 0.69.
Among 155 followed patients (median 85.6 months), 21.94% had recurrence/metastasis, 16.77% died; 5-year PFS=83.67%, OS=89.53%; 7-year PFS=78.06%, OS=83.23%; high-risk patients showed worse PFS trend (HR=3.01, P=0.034) but no significant PFS/OS differences (OS HR=1.55, P=0.307).

Conclusion/Discussion: The nomogram constructed based on preoperative CEA, MTD, and clinical T/N stages exhibited good performance in predicting pR in both the modeling cohort (196 cases) and the 58-case external validation cohort (AUC = 0.808, accuracy = 0.69 in external validation), and it has clinical applicability. However, its long-term survival prediction is limited (no significant PFS/OS differences) due to single-center bias and small event rates; future multicenter validation, multi-omics integration, and 10-year follow-up are needed to optimize LARC individualized treatment.