Machine learning-based personalisation of renin-angiotensin system inhibitors to reduce 3-year mortality in patients with acute myocardial infarction: insights from the KAMIR-NIH Registry
K H Kim, J WiAbstract
Background
Selecting the optimal pharmacotherapy after acute myocardial infarction (AMI) is critical for long-term survival. Although renin-angiotensin system inhibitors (RASI) are standard of care, individual responses vary. We aimed to develop a machine learning (ML)-based decision support model to predict 3-year mortality and guide personalized RASI selection using a large-scale nationwide registry.
Methods
We analyzed 12,598 patients with AMI who underwent successful percutaneous coronary intervention (PCI) and were discharged alive from the KAMIR-NIH (Korea AMI Registry) between 2011 and 2015. Three ML algorithms—Logistic Regression (LR), AdaBoost, and Support Vector Machine (SVM)—were implemented to predict 3-year all-cause mortality. The models incorporated 25 clinical variables, including age, LV function, diabetes mellitus (DM), and medication profiles.
Results
The overall 3-year mortality rate was significantly lower in patients receiving RASI (ACE inhibitors or ARBs) compared to those who did not (8.2% vs. 25.0%, p < 0.001). Patients with DM exhibited a nearly two-fold higher risk of 3-year mortality compared to non-diabetic patients (18.0% vs. 9.8%, p < 0.001). The ML models successfully predicted 3-year mortality with high precision (LR-based probability outputs showed distinct risk stratification). By simulating different RASI strategies (No RASI vs. ACEI vs. ARB), the ML model identified that specific subgroups—particularly those with DM or impaired LV function—derived varying degrees of survival benefit from different RASI classes, suggesting that a one-size-fits-all approach may be suboptimal.
Conclusions
Machine learning algorithms can effectively predict long-term mortality and provide personalized drug recommendations for AMI patients. Implementation of such ML-driven decision support systems may optimize post-AMI pharmacotherapy and significantly improve 3-year survival outcomes by identifying the most beneficial treatment for individual patient profiles.