Cardiovascular Risk Prediction Using Machine Learning in a Large Japanese Cohort

Area of Science:

• Cardiology
• Machine Learning
• Public Health

Background:

• Cardiovascular disease (CVD) screening requires precise event prediction for risk stratification and resource allocation.
• Employer-mandated health checkups offer a valuable data source for CVD risk assessment.

Purpose of the Study:

• To identify incident CVD markers in Japanese adults using random survival forests (RSF).
• To develop and evaluate simplified RSF models for individual-level CVD risk prediction.

Main Methods:

• Utilized RSF on biomarker, health history, medication, and lifestyle data from 155,108 adults (≥40 years).
• Examined coronary artery disease (CAD) and atherosclerotic CVD (ASCVD) events over 6 years.
• Split data into training (70%) and test (30%) sets for model validation.

Main Results:

• RSF identified key predictors: heart disease history, age, blood pressure medication, HbA1c, fasting blood sugar, and HDL.
• Simplified RSF models with top 20 predictors showed strong performance (AUC >84% for CAD, >82% for ASCVD).
• Observed 1,642 CAD and 2,164 ASCVD events during follow-up.

Conclusions:

• Machine learning, specifically RSF, offers an accurate method for cardiovascular risk assessment.
• Employer health checkup data can be leveraged for efficient CVD prevention.
• The developed algorithm generates individual risk curves, aiding clinical decision-making.