Risk factors predicting rectal cancer recurrence by machine learning model

Purpose: Despite advances in treatment modalities, the recurrence event is still a clinical challenge for patients with rectal cancer. Previous studies have shown that fewer studies have compared the clinical benefits of different predictive models to improve predictive outcomes for individual conditions. Method: A total of 3,403 hospital-based cancer registry records by six cancer registries in Taiwan, between 2011 and 2020. Different from traditional statistics analysis, the approach is mainly based on the predicted classification models, which were classification and regression trees, random forests. The modelling performance of predictive models was evaluated by accuracy, sensitivity, specificity, F1 score, Kappa, Matthews correlation coefficient (MCC), Area under the curve (AUC). Using decision curve analysis (DCA) and clinical impact curves, predictive models to evaluate the best clinical benefits. Results: The results of this study showed that risk factors for recurrence are related to lifestyle, genetics, environmental factors, and the use of anti-cancer therapy. Unlike traditional regression model analysis, the risk factor analysis by machine learning methods used in our study could deal with the existence of complex interactions between different predictor variables. Furthermore, these results suggest that effective identification of treatment-associated factors that increase the risk of recurrence is a growing concern for cancer survivors. Conclusion: The results of this study can be used as a reference for clinicians to assist in clinical decision making and provide maximum clinical benefit to patients.