Application of machine learning algorithms to predict coronary artery calcification with a sibship-based design

Area of Science:

• Genetics
• Epidemiology
• Cardiovascular Disease

Background:

• Coronary artery calcification (CAC) is a marker of subclinical atherosclerosis.
• Identifying genetic and clinical predictors of high CAC burden is crucial for understanding cardiovascular disease risk.
• Hypertension is a significant risk factor for cardiovascular events.

Purpose of the Study:

• To identify genetic and clinical predictors of high CAC burden in hypertensive individuals using machine learning.
• To assess the replicability of identified predictors using two distinct datasets.
• To evaluate the efficacy of machine learning algorithms in predicting subclinical coronary atherosclerosis.

Main Methods:

• Utilized data from the Genetic Epidemiology Network of Arteriopathy study, including 471 single nucleotide polymorphisms (SNPs) and 17 risk factors.
• Applied Random Forests and RuleFit machine learning algorithms to two independent datasets of hypertensive sibships.
• Employed five-fold cross-validation and permutation tests to assess prediction accuracy and identify significant predictors.

Main Results:

• Machine learning models achieved approximately 70% sensitivity and 60% specificity in predicting high CAC burden.
• Predictability was comparable using 287 tagSNPs versus all 471 SNPs.
• Identified two replicable tagSNPs (in GPR35 and NOS3 genes) and 12 replicable risk factors associated with high CAC burden.

Conclusions:

• Machine learning methods are effective in identifying important and replicable predictors of subclinical coronary atherosclerosis in sibship studies.
• The identified genetic variants and clinical factors contribute to understanding CAC burden in hypertensive populations.
• This study highlights the potential of integrating genetic and clinical data for personalized cardiovascular risk assessment.