Predicting intermediate-risk prostate cancer using machine learning

Area of Science:

• Urology
• Oncology
• Machine Learning in Medicine

Background:

• Intermediate-risk prostate cancer (IR PCa) represents the most frequent risk group for localized prostate cancer.
• Accurate risk stratification is crucial for appropriate patient management and avoiding overtreatment.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting the probability of IR PCa.
• To compare the performance of the ML model against traditional clinical models in identifying IR PCa.

Main Methods:

• A generalized linear model (GLM) was developed using pre-biopsy predictors including age, PSA, digital rectal exam findings, prostate volume, and PSA density (PSAD).
• The model's discriminatory ability was assessed using the area under the receiver operating characteristic curve (AUC).
• Patient data from January 2017 to December 2022, including biopsy outcomes, were analyzed.

Main Results:

• The study included 729 patients, with 120 (16.5%) diagnosed with IR PCa.
• The novel ML model achieved an AUC of 0.806, significantly outperforming the clinical model's AUC of 0.669 (p=0.018).
• The GLM showed potential for a 44.3% reduction in unnecessary biopsies, with PSAD identified as the most significant predictor.

Conclusions:

• A GLM utilizing pre-biopsy features was successfully developed to predict IR PCa.
• The developed model demonstrates strong discriminatory performance and clinical applicability.
• This ML tool can aid urologists in better determining the necessity of prostate biopsies.