PSA-based machine learning model improves prostate cancer risk stratification in a screening population.

Area of Science:

• Urology
• Oncology
• Machine Learning in Medicine

Background:

• Prostate cancer diagnosis heavily relies on Prostate Specific Antigen (PSA) testing, but its accuracy is limited.
• Improving prostate cancer risk stratification requires incorporating patient demographics and additional biochemical markers.
• There is a need for accurate, cost-effective models using objective measures for better prostate cancer risk assessment.

Purpose of the Study:

• To develop a contemporary, accurate, and cost-effective model for prostate cancer risk stratification.
• To enhance the diagnostic accuracy of prostate cancer screening beyond traditional PSA testing.
• To leverage machine learning for improved prediction of prostate cancer.

Main Methods:

• A machine learning model was developed using data from a local institution and the Prostate, Lung, Colorectal and Ovarian Cancer (PLCO) screening trial.
• The model was trained on a dataset of 3638 patients, incorporating Prostate Specific Antigen (PSA), free-PSA, age, and the free-PSA to total PSA (FTR) ratio.
• The model's performance was validated on a separate set of 910 patients.

Main Results:

• The developed machine learning model demonstrated improved prediction for prostate cancer with an Area Under the Curve (AUC) of 0.72.
• This performance surpasses that of PSA alone (AUC 0.63), age (AUC 0.52), free-PSA (AUC 0.50), and FTR alone (AUC 0.65).
• At 80% sensitivity, the model achieved 45.3% specificity, indicating a significant improvement in diagnostic accuracy.

Conclusions:

• A dense neural network model significantly improved diagnostic accuracy in screening for prostate cancer.
• Machine learning methods offer additional utility in prostate cancer risk stratification when combined with biochemical parameters.
• The findings support the integration of advanced computational models for more precise prostate cancer detection.