In-Hospital Mortality Prediction using Machine Learning and Stacked Ensemble Learning of Asian Women with ST-Elevation Myocardial Infarction (STEMI)

Abstract

Abstract Predictions of mortality in Asian women following STEMI have been the subject of limited studies. This study aims to develop and validate prediction models for in-hospital mortality following STEMI in Asian women using machine learning (ML) and stacked ensemble learning (EL) techniques, and to compare the performance of the algorithms to that of a conventional risk scoring method. From 2006 to 2016, data on multi-ethnic Asian women admitted with STEMI from the Malaysian National Cardiovascular Disease Database (NCVD-ACS) registry were collected. Developed algorithms were compared to the Thrombolysis in Myocardial Infarction Risk score (TIMI) and a ML model constructed using data from the general STEMI population. Predictors for ML models were selected using iterative feature selection comprises of feature importance and sequential backward elimination. The machine learning models developed using ML feature selection (AUC ranging from 0.60–0.93) outperforms the conventional risk score, TIMI (AUC 0.81). Individual ML model, SVM Linear with selected features performed better than the best performed stacked EL model (AUC:0.934, CI: 0.893–0.975 vs AUC: 0.914, CI: 0.871–0.957). The women specific model also performs better than the general non-gender specific model (AUC: 0.919, CI: 0.874–0.965). Systolic blood pressure, Killip class, fasting blood glucose, beta-blocker, ACE inhibitor, and oral hypoglycemic agent are identified as common predictors of mortality for women. In multi-ethnic populations, Asian women with STEMI were more accurately classified by ML and stacked EL than by the TIMI risk score. It has also been determined that women-specific ML models perform better than the standard STEMI model. In the future, ongoing testing and validation can improve the clinical care provided to women with STEMI.