Abstract 4145242: Beta-Blocker Efficacy in Acute Myocardial Infarction: A Machine Learning Approach using the KAMIR-NIH Dataset

Background: Acute myocardial infarction (AMI) is a critical global health issue requiring effective secondary prevention strategies. Current guidelines advocate for beta-blockers (BB) in STEMI and NSTEMI cases due to their mortality benefits. However, these guidelines often overlook individual patient variability, necessitating personalized treatment approaches. This study aims to develop a machine learning (ML) model to predict individualized mortality in AMI patients and evaluate the relative benefits of BB across diverse patient profiles to enhance clinical outcomes. Methods: We analyzed 12,599 AMI patients in the Korea AMI Registry (KAMIR)-National Institutes of Health (NIH) dataset. Patients were randomly divided into a training set (n = 8,467) and a testing set (n = 4,132) by a 2:1 ratio. Patients were categorized into four quartiles based on Mortality Risk Differences determined by our ML model, indicating increasing benefits from BB therapy. Quartiles Q1 and Q2 (Low-Risk group) showed smaller differences in mortality risks, while Q3 and Q4 (High-Risk group) showed larger benefits. Results: Among various ML models, Binary GLM Logistic Regression performed best, achieving an AUC of 0.8428. Our evaluations focused on patients with anemia, those aged over 65, eGFR below 60 and eGFR below 90. These factors were selected because they demonstrated the greatest information gain according to the 'information gain attribute evaluation' in the high-risk group. when analyzing the total test group within these groups, there was a noticeable survival advantage for patients in the BB group compared to those not on BB. (p = 0.009, p < 0.001, p = 0.003, p < 0.001). However, there was no significant difference noted among the groups within the low-risk group. Conclusion: This study highlights the potential of ML to enhance personalized medicine in AMI management, particularly in optimizing BB treatment. By utilizing detailed patient data, our model facilitates more personalized treatments that are specifically aligned with individual patient needs, thereby improving clinical outcomes. This approach not only enhances the effectiveness of interventions but also embodies the principles of precision medicine, adapting treatment strategies to optimally suit each patient's unique clinical profile.