Random forest-based model for the recurrence prediction of borderline ovarian tumor: clinical development and validation
This study aims to develop an effective machine learning (ML)-based predictive model for the recurrence of borderline ovarian tumor (BOT), and provide the guidelines of accurate clinical diagnosis and precise treatment for patients. A total of 660 patients diagnosed with BOT were included in this study. Statistical testing methods were employed to identify the most influential factors. At the same time, five machine learning-based models-random forest (RF), logistic regression (LR), gradient boosting (GB), multilayer perceptron (MLP), and support vector machine (SVM)-were utilized to construct recurrence prediction models. Model validity was assessed using five metrics: area under the curve (AUC), positive predictive value (PPV), accuracy (ACC), recall (REC), specificity (SPE), and the optimal model was selected based on these performance metrics. The calibration curve further illustrates the reliability of the model. Then, the optimal ML-based model determined the importance of features using SHAP values. Additionally, CIC and DCA, along with recurrence-free survival analysis, were employed to further assess the clinical value of the optimal model. The RF model demonstrated superior predictive performance. Additionally, the SHAP analysis of the RF-based model provides the key clinical factors associated with the recurrence of BOT. Furthermore, the DCA and CIC shows the clinical application value of the RF-based model. Moreover, random forest-recurrence free survival (rf-RFS) model validate the effectiveness of the proposed method personalized treatment strategies and informed clinical decision-making of the recurrence of BOT. The RF-based model offers an effective tool for predicting BOT recurrence, with a user-friendly web-based calculator developed to aid clinical decision-making.