Machine Learning With Genetic and Clinical Data to Predict Ischemic Outcomes After PCI

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Pharmacogenomics

Background:

• Ischemic events post-percutaneous coronary intervention (PCI) are infrequent but serious.
• Optimizing dual antiplatelet therapy (DAPT) intensity is crucial to balance ischemic event prevention and bleeding risk.
• Current risk scores lack pharmacogenetic integration, limiting personalized treatment strategies.

Purpose of the Study:

• To develop and validate machine learning (ML) models predicting 1-year ischemic outcomes after PCI.
• To integrate clinical, demographic, and CYP2C19 genetic data for enhanced risk prediction.
• To refine clinical decision-making for DAPT intensity post-PCI.

Main Methods:

• Analysis of 8317 patients from the TAILOR-PCI trial and Precision PCI registry.
• Development of ML models using Boruta feature selection (11 predictors) and algorithms like SVM and XGBoost.
• External validation using cross-validation and SMOTE, assessing performance via AUC, sensitivity, and specificity.

Main Results:

• A support vector machine (SVM) model achieved an AUC of 0.667 (sensitivity 0.871, specificity 0.282).
• XGBoost offered a balanced profile with an AUC of 0.619 (sensitivity 0.442, specificity 0.688).
• All 11 selected predictors demonstrated high importance in the SVM model.

Conclusions:

• ML models integrating clinical and genetic data can identify high-risk patients post-PCI.
• This risk stratification aids in safely de-escalating DAPT for many patients, reducing bleeding risk.
• A multimodal approach to risk stratification is essential for personalized post-PCI care.