MACHINE LEARNING COMPARED TO CONVENTIONAL STATISTICAL MODELS FOR PREDICTING MYOCARDIAL INFARCTION READMISSION AND MORTALITY: A SYSTEMATIC REVIEW.

Abstract

BACKGROUND\nTo review the performance of machine learning (ML) methods compared to conventional statistical models (CSM) for predicting readmission and mortality in patients with myocardial infarction (MI).\n\n\nMETHODS\nFollowing PRISMA guidelines, we systematically reviewed the literature search using MEDLINE, EPUB, Cochrane CENTRAL, EMBASE, INSPEC, ACM Library, and Web of Science. Eligible studies included primary research articles published between January 2000 and March 2020, comparing ML and CSM for prognostication after MI.\n\n\nRESULTS\nOf 7,348 articles, 112 underwent full-text review, with the final set comprised of 24 articles and 374,365 patients. ML methods included artificial neural networks (n=12 studies), random forests (n=11), decision trees (n=8), support vector machines (n=8) and Bayesian techniques (n=7). CSM included logistic regression (n=19 studies), existing CSM-derived risk scores (n=12) and Cox regression (n=2). Thirteen of 19 studies examining mortality reported higher c-indices using ML compared to CSM. One study examined readmissions at two different time points, with c-indices that were higher for ML than CSM. Across all studies, a total of 29 comparisons were performed, but the majority (n=26, 90%) found small (< 0.05) absolute differences in the c-index between ML and CSM. Using a modified CHARMS checklist, sources of bias were identifiable in the majority of studies, and only 2 were externally validated.\n\n\nCONCLUSION\nAlthough ML algorithms tended to have higher c-indices than CSM for predicting death or readmission after MI, these studies exhibited threats to internal validity and were often unvalidated. Further comparisons are needed, with adherence to clinical quality standards for prognosis research.