Machine learning–based model to predict risks of recurrence after catheter ablation in patients with type 2 diabetes mellitus combined with atrial fibrillation

Abstract BACKGROUND The coexistence of atrial fibrillation (AF) and type 2 diabetes mellitus (T2DM) is associated with an elevated risk of AF recurrence following catheter ablation (CA). However, the determinants influencing AF recurrence in these patients remain incompletely elucidated. OBJECTIVE The study employed machine learning (ML) algorithms to create predictive models for these patients. METHODS In this retrospective study, clinical data from patients at the General Hospital of the Northern Theater Command were preprocessed through zero-variance filtering, multiple imputation, undersampling, and recursive feature elimination. Five machine learning models were subsequently developed and evaluated using multiple performance metrics. Global feature importance was assessed with the SHAP method, followed by hierarchical analysis of key variables. RESULTS A total of 390 patients were included in the study. Following data preprocessing, 26 variables were retained for model development. Among the five models constructed, XGBoost achieved the highest performance (AUC = 0.936 [95% CI, 0.908-0.964]). SHAP analysis revealed stroke, hypertension, left atrial diameter, age, and metformin use as the principal predictors of AF recurrence in this study cohort, with metformin use exerting a protective effect. Subsequently, stratified analysis substantiated the necessity of conducting in-depth research to delineate influencing factors across distinct patient subgroups. CONCLUSION The findings of this study underscore the potential of ML in predicting AF recurrence following CA in patients with T2DM complicated by AF. Moreover, it serves as a valuable adjunct for identifying high-risk patients and facilitating timely interventions.