Acute myocardial infarction (AMI) is a global health burden with an incidence of 70-100 cases per 100,000 people annually. Despite standard treatments, AMI patients face high risks, with recurrent AMI rates of 20-30% within five years and major adverse cardiovascular events (MACE) rates of 10-15% within the first year. High-risk groups, such as those with multiple cardiovascular risk factors or chronic kidney disease, have even greater mortality rates. Traditional prediction models focus on clinical and demographic factors, often lacking comprehensive medication data, such as SGLT2 inhibitors. Inspired by recent findings suggesting that SGLT2 inhibitorsmay impact AMI prognosis, this study emphasizes the importance of integrating these cardioprotective drugs into risk assessment models, particularly for Asian populations.Aim:This study aims to develop a machine learning-based risk prediction model for Asian populations, using long-term,multi-center data. It integrates clinical, laboratory, and detailed medication data, including SGLT2 inhibitors, antiplatelet agents, lipid-lowering drugs,and asthma treatments. Through extended follow-up and comprehensive medication assessment, the model seeks to improve high-risk AMI patient identification and evaluate the protective role of SGLT2 inhibitors.

This retrospective study analyzed data from 7,081 AMI patients treated over six years across three Taipei Medical University-affiliated hospitals. In Patients with a confirmed AMI diagnosis and successful percutaneous coronary intervention (PCI) were included, while those with prior PCI or coronary bypass surgery were excluded. The dataset encompassed clinical characteristics, lab values, and medication records. Primary outcomes were MACE, recurrent AMI, and all-cause mortality.Data processing involved multiple imputation for missing values and variable standardization. The data was split into a training set (70%) and a test set (30%), and models applied included random forest, support vector machine, and artificial neural networks, with hyperparameters optimized via7-fold cross-validation. Model performance was evaluated using AUC, accuracy, recall, and F1 score, and feature importance was analyzed using SHAP values to identify critical predictors of MACE and mortality.

The results and conclusions of the BEST CLINICAL ABSTRACTS will be released after the presentation on April 25.

The results and conclusions of the BEST CLINICAL ABSTRACTS will be released after the presentation on April 25.