Research on the prediction of slow blood flow in pPCI of STEMI patients based on CatBoost.

backgroundIn recent years, the incidence of ST-segment elevation myocardial infarction (STEMI) has been on the rise, leading to an increase in the number of patients undergoing direct percutaneous coronary intervention (pPCI). However, some patients experience slow blood flow after pPCI, significantly raising the risk of death. To help doctors intervene early in cases of slow blood flow in pPCI patients, we employ appropriate methods for data preprocessing, model training, and prediction. This approach enables doctors to diagnose, develop, and implement treatment plans earlier. The study also demonstrates the practical application of machine learning and deep learning algorithms in analyzing slow blood flow, providing valuable references and guidance for other researchers. METHODS AND

resultsData of patients undergoing direct percutaneous coronary intervention (pPCI) were collected from four public tertiary hospitals in Nanning, City, China. The paper employed 4 methods for imputing missing data, 9 methods for addressing data imbalance, and 8 integrated machine learning methods for prediction. By pairing each of the 4 imputation methods with each of the 9 balancing methods, 36 combined CatBoost methods were created. These 36 combinations were then each paired with one of the 8 integrated machine learning methods, resulting in a total of 288 prediction combination methods. ROC curve, AUC, and F1 score were primarily used as evaluation indicators to compare the results. In the end, the Bayesian-based Optuna method was used to optimize the hyperparameters of the CatBoost. The results showed that the AUC value obtained with GAN data imputation was the highest, and the F1 score was also elevated. After hyperparameter optimization, the AUC value of the CatBoost model improved significantly. The CatBoost prediction model identified creatine kinase isozyme, hypersensitive C-reactive protein, time to first medical contact, and stent length as the most important factors affecting slow blood flow.

conclusionsThe prediction of slow blood flow in STEMI patients undergoing direct percutaneous coronary intervention (pPCI) using CatBoost combined with GAN can achieve promising results. This approach provides a theoretical basis for doctors to intervene in advance.