Machine learning risk stratification for major adverse cardiovascular events after acute myocardial infarction in patients receiving early empagliflozin therapy
Litsenziya: cc-by
Annotatsiya
Sodium-glucose cotransporter-2 (SGLT2) inhibitors are increasingly prescribed after acute myocardial infarction (AMI), yet residual risk on this therapy is rarely examined alongside the drug’s association with outcomes in one cohort. We investigated 311 patients with AMI at one center (January 2024–May 2025): 251 received early empagliflozin 10 mg/day with standard therapy and were followed prospectively for six months; 60 contemporaneous patients without an SGLT2 inhibitor formed a non-randomized control group. The endpoint was six-month major adverse cardiovascular events (MACEs): all-cause death, non-fatal reinfarction, urgent percutaneous coronary intervention, or unplanned cardiovascular hospitalization. Classical and machine learning classifiers were compared under repeated nested cross-validation, with imputation, scaling, and tuning inside each fold; the empagliflozin–MACE association was estimated by propensity-score matching on 20 covariates. MACE occurred in 83 patients (26.7%; 24.7% with empagliflozin vs. 35.0% in controls), including 17 deaths (5.5%). Cross-validated backward elimination reduced 24 candidate predictors to eight: body mass index, prior infarction, current smoking, mineralocorticoid-receptor antagonist use, peak A velocity, left-ventricular end-diastolic volume, empagliflozin exposure, and discretized total cholesterol. On these features, L2-regularized logistic regression reached a pooled out-of-fold area under the curve of 0.673 (95% confidence interval 0.601–0.741) with good calibration (slope 0.88), matching random forest, gradient boosting, XGBoost, support-vector machine, k-nearest neighbors, and a multilayer perceptron. After matching (44 pairs), the odds ratio for MACE with empagliflozin was 0.79 (95% CI 0.32–1.98). Discrimination was moderate and needs external validation; the association between empagliflozin and fewer events was small and hypothesis-generating.
Hali tarjima qilinmagan