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

CYP2C19 genotype (loss-of-function status) is integrated into machine-learning models with clinical variables to predict ischemic risk.

Ischemic events after percutaneous coronary intervention (PCI) are uncommon but severe, and existing machine-learning risk scores do not incorporate pharmacogenetic data. The authors developed and externally validated models combining clinical, demographic and CYP2C19 genetic information to predict a 1-year composite outcome (cardiovascular death, myocardial infarction, stroke, stent thrombosis), using 8317 patients (TAILOR-PCI trial, n = 4572; Precision PCI registry, n = 3745). Boruta selection retained 11 predictors; the best external performance came from a polynomial-kernel SVM (AUC 0.667; sensitivity 0.871; specificity 0.282), with XGBoost offering a more balanced profile (AUC 0.619). These models can help identify the small subset of high ischemic-risk patients, most of whom may safely de-escalate dual antiplatelet therapy to reduce bleeding risk.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Integrating CYP2C19 genotype, an established determinant of clopidogrel response (CPIC level A recommendation), into multimodal risk stratification is a sensible direction for individualizing antiplatelet intensity. External performance remains modest (AUC < 0.67), and the rare outcome limits positive predictive value. The tool remains proof-of-concept rather than routine-ready.

Analysis by Dr Thibaut Benquey