Machine-Learning Models for Multicenter Prostate Cancer Treatment Plans.

Area of Science:

• Oncology
• Health Services Research
• Medical Informatics

Background:

• Clinical factors like T-stage, Gleason score, and PSA stratify prostate cancer (PCa) risk.
• These factors guide treatment selection for heterogeneous PCa.
• Nonclinical factors' impact on PCa treatment adherence is under-explored.

Purpose of the Study:

• To investigate the influence of nonclinical factors on treatment selection and guideline adherence in PCa.
• To identify predictors of adherence to National Comprehensive Cancer Network (NCCN) guidelines for PCa.
• To assess the role of treatment centers in PCa treatment decisions and adherence.

Main Methods:

• Retrospective analysis of 552 intermediate- and high-risk PCa patients treated with radiation +/- androgen deprivation therapy (ADT).
• Data extracted from 34 Veterans Health Administration medical centers (2010-2017).
• Support Vector Machine and Random Forest models used to identify clinical and nonclinical predictors of NCCN guideline adherence.

Main Results:

• Treatment center (nonclinical factor) significantly influenced adherence to NCCN guidelines.
• The treatment center predicted decisions regarding ADT prescription but not its duration or treatment alterations.
• Clinical factors alone were insufficient to fully predict adherence.

Conclusions:

• Nonclinical factors, particularly the treatment center, play a crucial role in PCa treatment adherence and ADT decisions.
• Center-specific variations warrant further investigation into barriers affecting guideline adherence and oncological outcomes.
• Publicly available datasets like SEER may be inadequate for building robust predictive models for PCa treatment plans.