Development and validation of a machine learning model for predicting unplanned removal of totally implantable venous access ports in patients with breast cancer

Objective: Analyze risk factors for unplanned removal of totally implantable venous access ports (TIVAP) in breast cancer patients and develop a machine learning prediction model. Methods: A single-center, retrospective study was conducted, including breast cancer patients(n=1,258) who underwent TIVAP removal at the Department of Breast and Thyroid Surgery of a tertiary hospital in Chongqing between October 2013 and March 2023.Variable screening was performed using univariate logistic regression and the Least Absolute Shrinkage and Selection Operator (LASSO) method to identify potential risk factors. The variables selected through this process were then incorporated into a multivariate logistic regression for further analysis. The dataset was randomly split into a training set (n=881) and a validation set (n=377) in a 7:3 ratio. Subsequently, four machine learning prediction models-random forest, decision tree, logistic regression, and XGBoost-were developed and evaluated. The model demonstrating the best performance was selected as the final predictive model. Results: The XGBoost model developed in this study demonstrated optimal performance in predicting the risk of unplanned removal of totally implantable venous access ports (TIVAP) in breast cancer patients. In the training set, the area under the receiver operating characteristic curve (AUC) was 0.826 (95% CI: 0.783-0.869), with a specificity of 0.829 and a sensitivity of 0.683. In the validation set, the AUC was 0.751 (95% CI: 0.704-0.839), with a specificity of 0.820 and a sensitivity of 0.636, indicating that the model possesses strong discriminative ability and robust predictive performance. Furthermore, the study identified independent risk factors for unplanned TIVAP removal, primarily including body mass index (BMI), TNM stage, implantation route, number of successful puncture attempts, coagulation function indices, neutropenia, and catheter indwelling time. The calibration curve shows that the XGBoost model is well calibrated and fitted, and DCA indicates that the model has good clinical net benefits. Conclusion: This study developed a machine learning prediction model for the risk of unplanned removal of totally implantable venous access ports (TIVAP) in breast cancer patients. The model demonstrated favorable discrimination and calibration, with robust predictive performance. It contributes to the early identification of high-risk patients in clinical practice, enabling targeted interventions and preventive measures to reduce the incidence of unplanned TIVAP removal.