Machine learning approaches improve risk stratification for secondary cardiovascular disease prevention in multiethnic patients

Area of Science:

• Medical Informatics
• Cardiology
• Machine Learning

Background:

• Effective cardiovascular disease (CVD) prevention relies on accurate identification of high-risk patients.
• The utility of electronic health record (EHR)-based machine learning (ML) models for CVD risk stratification compared to traditional scores is not well-established.

Purpose of the Study:

• To compare the performance of EHR-based ML models against the Thrombolysis in Myocardial Infarction Risk Score for Secondary Prevention (TRS 2°P) for CVD risk stratification.
• To evaluate the ability of ML models to predict 5-year CVD event risk in a real-world, multiethnic population.

Main Methods:

• A cohort of 32,192 patients from a large health system was identified, with 80% used for training and 20% for testing ML models.
• Various ML models (Random Forests, Gradient-Boosted Machines, XGBoost, Logistic Regression) were trained using EHR data to predict 5-year CVD event risk.
• Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and compared with TRS 2°P.

Main Results:

• ML models demonstrated good performance, with XGBoost achieving an AUC of 0.70 in the full CVD cohort and 0.71 in the atherosclerotic CVD (ASCVD) cohort.
• The traditional TRS 2°P score performed poorly, with AUCs of 0.51 and 0.50 for CVD and ASCVD patients, respectively.
• ML models identified novel predictive variables, including education level and primary care visit frequency.

Conclusions:

• EHR-based ML models significantly enhance CVD risk stratification for secondary prevention in a real-world, multiethnic setting.
• ML approaches offer a superior alternative to traditional risk scores for identifying high-risk patients for CVD prevention strategies.