Development of a Logistic Regression Model for Colorectal Cancer Relapse Prediction in Sabah, Malaysia

Colorectal cancer (CRC) remains one of the leading causes of cancer-related deaths in Malaysia, with relapse contributing substantially to poor survival outcomes. Despite advances in treatment, relapse surveillance continues to rely on non-individualized schedules. This study aimed to develop and internally validate an interpretable logistic regression model for predicting relapse among Malaysian CRC survivors using routinely available clinical and pathological variables. A retrospective case-control study was conducted using data from hospital-based cancer registries and oncology records across selected public hospitals in Sabah. Patients diagnosed between 2015 and 2020 who completed curative-intent treatment with at least five years of follow-up were included. Ten routinely collected clinicopathological variables were evaluated as candidate predictors using multivariable logistic regression. Model performance was assessed with 10-fold cross-validation. Discrimination was measured using the area under the receiver operating characteristic curve (AUC), and calibration was assessed across risk strata. The optimal probability threshold was identified using the Youden index. Six predictors remained significant in the final model, including tumor stage, lymphovascular and perineural invasion, carcinoembryonic antigen level, tumor grade, and completeness of chemotherapy (p < 0.02). The final model demonstrated strong discrimination (AUC = 0.85, 95% CI 0.81-0.89) with good calibration (Hosmer-Lemeshow p = 0.42). Sensitivity and specificity were balanced, and internal validation confirmed model stability across folds. Predicted probabilities were stratified into low, moderate, and high-risk categories to support tailored follow-up planning. This validated model offers a clinically practical approach to relapse risk stratification among Malaysian CRC survivors. Its reliance on routinely available data ensures scalability, particularly in resource-limited settings. Integration into the digital CARE-CRC tool may enable risk-adapted surveillance and improve long-term outcomes within Malaysia's public healthcare system.