Predicting Positive Surgical Margins in Robot-Assisted Prostatectomy Using Machine Learning Models.

OBJECTIVE

To develop a machine learning (ML) model predicting positive surgical margins (PSM) after robot-assisted radical prostatectomy (RARP).

METHODS

We conducted a single-centre retrospective analysis of 301 patients with RARP (211 for training and 90 for validation). Twenty-four features were reduced to five using the Boruta algorithm. Seven ML models were developed and evaluated, and the optimal model was interpreted using SHAP.

RESULTS

PSM incidence was 42.0%. The artificial neural network (ANN) performed best (AUC 0.808, accuracy 0.811, sensitivity 0.789, F1 0.779). SHAP identified clinical T stage, creatinine, high-risk status, and positive biopsy percentage as risk factors, while neoadjuvant therapy was protective.

CONCLUSION

The ANN model demonstrates potential in predicting PSM following RARP. Currently, it serves as an exploratory predictive tool for risk stratification, requiring further external validation prior to direct clinical application.