Can machine learning bring cardiovascular risk assessment to the next level? A methodological study using FOURIER...

Area of Science:

• Cardiovascular disease research
• Biostatistics
• Machine learning applications in healthcare

Background:

• Cardiovascular disease remains a leading cause of mortality.
• Accurate risk stratification is crucial for effective secondary prevention strategies.
• Traditional statistical models have limitations in handling complex, high-dimensional data.

Purpose of the Study:

• To evaluate the added value of machine learning (ML) methods for building cardiovascular risk scores.
• To determine conditions under which ML models outperform traditional linear statistical models.
• To explore the utility of ML in patient risk stratification for secondary cardiovascular event prevention.

Main Methods:

• Comparison of linear models, neural networks, random forest, and gradient boosting machines.
• Utilized extensive cardiovascular clinical data from the FOURIER randomized clinical trial.
• Evaluated model performance using restricted subsets of patient data and covariates.

Main Results:

• ML methods, particularly gradient boosting, significantly outperformed linear models (c-index 0.67 vs. 0.62).
• Gradient boosting requires fewer patients and excels with numerous variables.
• Linear models showed limitations when trained on excessive variables, necessitating careful covariate selection.

Conclusions:

• ML methods offer consistent improvements in cardiovascular risk assessment.
• These complex ML models are interpretable and can identify essential covariates for optimal performance.
• ML holds promise for enhanced patient risk stratification and treatment allocation using electronic health records.