Development and External Validation of a Machine Learning-Based Risk Classification Score for Prostate...

Area of Science:

• Oncology
• Cardiovascular Medicine
• Biostatistics

Background:

• Patients with prostate cancer (PC) have increased cardiovascular (CV) risk.
• Existing CV risk scores are often inaccurate for PC patients and require unavailable lab tests.
• This limits practical CV risk assessment in oncology settings.

Purpose of the Study:

• To develop and validate novel, PC-specific cardiovascular risk scores.
• To utilize variables routinely available in oncology clinics for risk stratification.
• To improve the calibration and applicability of CV risk assessment in PC management.

Main Methods:

• Developed risk scores using XGBoost and LASSO on a development cohort (n=1,815).
• Validated scores in two independent cohorts (V1: n=4,022; V2: n=1,729).
• Assessed performance using time-dependent AUC (TDAUC) for cardiovascular disease (CVD), atherosclerotic CVD (ASCVD), and heart failure (HF).

Main Results:

• Composite CVD score achieved 10-year TDAUC of 0.71 in the development cohort and 0.59 in V2.
• ASCVD and HF scores demonstrated good discrimination, with 10-year TDAUCs of 0.66 and 0.70 respectively in the development cohort.
• Scores retained predictive ability in validation cohorts, indicating robustness.

Conclusions:

• These are the first PC-specific CV risk scores developed using oncology-available variables.
• The scores support practical CV risk stratification within PC clinics.
• Further validation and implementation are encouraged to enhance patient care.