Prediction of Recurrence After Endometrial Polypectomy in Women of Childbearing Age Based on Machine Learning Algorithm

Objective: Research on machine learning (ML) prediction models for recurrence after endometrial polypectomy remains unexplored. In this study, we aim to establish an ML-driven model based on multi-biomarkers to predict recurrence after endometrial polypectomy. Methods: A retrospective cohort of 606 patients who underwent endometrial polypectomy was analyzed. Variables including age, BMI, reproductive history, prior uterine surgery, coexisting uterine disorders, polyp characteristics, and routine blood indices were extracted. Seven ML algorithms (logistic regression, SVM, MLP, random forest, KNN, Naïve Bayes, and decision tree) were trained using 10-fold cross-validation. Performance was evaluated by AUC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score. Results: Of the 606 patients, 179 (29.5%) developed recurrence within one year postoperatively. The cohort was randomly divided into a training set (n = 424) and a validation set (n = 182). In the training set, the random forest (RF) algorithm achieved the best performance (AUC = 0.838, accuracy = 79.5%, specificity = 0.930, F1 score = 0.576). In the validation set, RF remained superior (AUC = 0.760, accuracy = 75.3%, specificity = 0.875, F1 score = 0.526), underscoring its strong generalizability. SHAP analysis identified age, posterior-wall polyp location, prior uterine surgery, histopathological subtype, and hemoglobin level as the most influential predictors of recurrence. Conclusion: The RF-based model, using demographic, clinical, and hematologic features, showed high accuracy in predicting recurrence risk after endometrial polypectomy. This interpretable ML framework can help clinicians identify high-risk patients early and personalize postoperative surveillance.