Pre-existing and machine learning-based models for cardiovascular risk prediction.

Area of Science:

• Cardiology
• Biostatistics
• Artificial Intelligence in Healthcare

Background:

• Cardiovascular disease (CVD) risk prediction is crucial for primary prevention strategies.
• Large, complex healthcare datasets necessitate advanced analytical methods like machine learning.
• Evaluating existing CVD risk prediction models is essential for identifying areas for improvement.

Purpose of the Study:

• To assess the discrimination and calibration of established cardiovascular risk prediction models.
• To develop and evaluate novel machine learning-based algorithms for cardiovascular risk prediction.
• To compare the performance of machine learning models against traditional models in a Korean population.

Main Methods:

• Analysis of data from 222,998 Korean adults aged 40-79 without prior CVD or lipid-lowering therapy.
• Comparative assessment of pre-existing models (e.g., Pooled Cohort Equation, Framingham Risk Score) and machine learning algorithms (logistic regression, treebag, random forest, adaboost, neural network).
• Evaluation metrics included C-statistics for discrimination and Hosmer-Lemeshow test for calibration.

Main Results:

• Pre-existing models demonstrated moderate to good discrimination (C-statistics 0.70-0.80).
• The neural network model achieved a higher C-statistic (0.751) than the Pooled Cohort Equation (0.738).
• The neural network model showed improved calibration and agreement between predicted and observed risks compared to traditional models.

Conclusions:

• Machine learning-based algorithms, particularly neural networks, offer superior performance for cardiovascular risk prediction in statin-naïve Korean adults compared to existing models.
• These advanced models can enhance accuracy in risk assessment for primary prevention.
• The developed machine learning model is suitable for adoption in clinical decision-making processes.