Development and validation of an interpretable machine learning model for predicting Gleason score upgrade in prostate cancer.

Background: The high incidence of Gleason score upgrade (GSU) can lead urologists to underestimate tumor aggressiveness, resulting in suboptimal treatment decisions. This study aimed to develop an interpretable machine learning model to predict the risk of GSU in individuals with prostate cancer (PCa) based on readily available clinical parameters.

Methods: A retrospective analysis was conducted on patients who underwent radical prostatectomy (RP) at Shanghai General Hospital and West China Hospital. Data from Shanghai General Hospital were categorized into a training set (80%) and a test set (20%), while data from West China Hospital were used for external validation. Preoperative clinical and pathological data were collected. Nine machine learning models [including random forest (RF) and light gradient boosting machine (LightGBM)], were developed, and the model demonstrating the best predictive performance was selected as the final model. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, decision curves, and SHapley Additive exPlanations (SHAP) interpretation.

Results: The LightGBM model demonstrated strong predictive performance, achieving an area under the ROC curve of 84.53% in the test set and 76.61% in external validation. Significant factors associated with GSU included the International Society of Urological Pathology (ISUP) grade, age, clinical tumor stage (T stage), body mass index, prostate-specific antigen (PSA), free-to-total PSA ratio (f/t PSA), platelet-to-lymphocyte ratio (PLR), and bilateral tumor involvement. An online prediction tool was developed based on this model.

Conclusions: A machine learning model and an online prediction tool were developed to accurately predict GSU and identify factors associated with this process. This approach may assist clinicians in identifying individuals at high-risk for GSU and facilitating evidence-based treatment decisions.