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

Area of Science:

• Urology
• Oncology
• Machine Learning
• Medical Informatics

Background:

• Gleason score upgrade (GSU) can lead to underestimation of prostate cancer (PCa) aggressiveness.
• This can result in suboptimal treatment decisions for patients.

Purpose of the Study:

• Develop an interpretable machine learning model to predict GSU risk.
• Utilize readily available clinical parameters for prediction.

Main Methods:

• Retrospective analysis of radical prostatectomy (RP) patients.
• Development and evaluation of nine machine learning models, including LightGBM.
• Performance assessment using ROC curves, calibration curves, decision curves, and SHAP interpretation.

Main Results:

• LightGBM model achieved 84.53% AUC in the test set and 76.61% in external validation.
• Key predictors for GSU include ISUP grade, age, T stage, BMI, PSA, f/t PSA, PLR, and bilateral tumor involvement.
• An online prediction tool was created based on the model.

Conclusions:

• An accurate machine learning model and online tool for GSU prediction were developed.
• Identified key factors associated with GSU.
• This approach aids clinicians in identifying high-risk patients for informed treatment planning.