Leveraging Machine Learning Models to Explore Disparities in Prostate Cancer Diagnosis, Treatment, and Survival

Area of Science:

• Oncology
• Data Science
• Public Health

Background:

• Prostate cancer (CaP) is a leading cause of cancer mortality globally, with significant disparities in the US.
• Medically underserved regions like Appalachia face higher CaP mortality due to access barriers.
• Machine learning (ML) can identify complex survival predictors in CaP.

Purpose of the Study:

• To apply ML models to identify key predictors of CaP survival.
• To assess the impact of sociodemographic and clinical factors on CaP outcomes.
• To analyze CaP survival using Kentucky Cancer Registry data.

Main Methods:

• Retrospective analysis of 37,893 CaP cases (2010-2022) from the Kentucky Cancer Registry.
• Utilized Kaplan-Meier, Random Survival Forest (RSF), and Elastic Net regression for survival estimation and variable importance.
• Employed multiple imputation for missing data and leave-one-out cross-validation.

Main Results:

• Age at diagnosis, treatment modality, and smoking status were top survival predictors.
• Age was ~2.5 times more influential than treatment type in RSF models.
• Younger patients (<60) had better outcomes; insurance status, tumor grade, and lymph node involvement also showed disparities.

Conclusions:

• ML models effectively identified age, smoking, and treatment as critical CaP survival predictors.
• Findings underscore the need for early screening and equitable treatment access.
• Interventions should address disparities in high-risk regions like Kentucky and Appalachia.