Machine Learning Models Integrating CYP2C19 Genotypes for Long-Term Risk Prediction and Personalized Antiplatelet...

Area of Science:

• Cardiovascular Medicine
• Pharmacogenomics
• Artificial Intelligence in Healthcare

Background:

• Personalizing antiplatelet therapy post-myocardial infarction (MI) is crucial for secondary prevention.
• Integrating clinical data with genetic information, like CYP2C19 genotype, offers potential for improved risk prediction.
• Current risk stratification models may not fully capture individual patient variability.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting long-term major adverse cardiac events (MACEs) in MI patients.
• To utilize clinical features and CYP2C19 genotype for personalized P2Y12 inhibitor selection.
• To assess the prognostic significance of genetic variants and clinical factors in MI outcomes.

Main Methods:

• Prospective observational study of 218 MI patients with up to 9-year MACE follow-up.
• Development and evaluation of ML models (including CatBoost) using clinical, genetic, and angiographic variables.
• Application of uplift modeling and SHAP (Shapley Additive Explanations) for feature importance and treatment effect analysis.

Main Results:

• The optimal ML model achieved an area under the ROC curve of 0.721, demonstrating stable discrimination.
• Key predictors of MACE included age, comorbidity index, coronary lesion count, P2Y12 inhibitor, stent type, and CYP2C19 variants.
• CYP2C19 variants were significant MACE predictors; drug-eluting stents and ticagrelor were associated with lower risk.

Conclusions:

• An ML framework integrating CYP2C19 genotype effectively stratifies long-term MACE risk in MI patients.
• Genetic testing holds prognostic utility for guiding antiplatelet therapy selection.
• This data-driven approach supports the potential role of ticagrelor in genetically defined high-risk MI patients for secondary prevention.