A novel machine learning model to predict high on-treatment platelet reactivity on clopidogrel in Asianpatients after percutaneous coronary intervention.

Abstract

Various genetic and nongenetic variables influence the high on-treatment platelet reactivity (HTPR) in patients taking clopidogrel. This study aimed to develop a novel machine learning (ML) model to predict HTPR in Chinese patients after percutaneous coronary intervention (PCI). This cohort study collected information on 507 patients taking clopidogrel. Data were randomly divided into a training set (90%) and a testing set (10%). Nine candidate Machine learning (ML) models and multiple logistic regression (LR) analysis were developed on the training set. Their performance was assessed according to the area under the receiver operating characteristic curve, precision, recall, F1 score, and accuracy on the test set. Model interpretations were generated using importance scores by transforming model variables into scaled features and representing in radar plots. Finally, we established a prediction platform for the prediction of HTPR. A total of 461 patients (HTPR rate: 19.52%) were enrolled in building the prediction model for HTPR. The XGBoost model had an optimized performance, with an AUC of 0.82, a precision of 0.80, a recall of 0.44, an F1 score of 0.57, and an accuracy of 0.87, which was superior to those of LR. Furthermore, the XGBoost method identified 7 main predictive variables. To facilitate the application of the model, we established an XGBoost prediction platform consisting of 7 variables and all variables for the HTPR prediction. A ML-based approach, such as XGBoost, showed optimum performance and might help predict HTPR on clopidogrel after PCI and guide clinical decision-making. Further validated studies will strengthen this finding.