OBJECTIVES: AI/ML is being rapidly developed in cardiovascular diseases (CVDs) research to predict disease risk, incidence, and outcomes. This TLR investigated the use of AI/ML techniques to understand the outcomes and risk prediction among MI patients in comparison with conventional statistical methods (CSMs).

METHODS: The search was conducted using OVID platform to identify studies reporting AI/ML techniques and/or CSMs in patients with MI published from January 2017 to 9 June 2022.

RESULTS: A total of 1755 studies were identified, of which 38 full texts were included for analysis. Included studies comprised patients with MI and suspected-MI, aged approx. >50 years, 79% were retrospective studies and 21% prospective. Majority of the studies reported supervised learning (87%), followed by unsupervised (5%) and unspecified (8%) learning methods. The use of AI/ML techniques was reported by nine studies (24%) and CSMs by three (8%), whereas 26 studies (68%) reported both AI/ML and CSM methods. For AI/ML models, 20 studies reported the use of random forests, followed by gradient boosting (n=18), neural networks (n=14), support vector machines (n=12), tree-based (n=10), Bayesian techniques (n=9) and k-nearest (n=2). The commonly used CSMs were logistic regression (n=19), least absolute shrinkage and selection operator (LASSO; n=5), GRACE (Global Registry of Acute Coronary Events) score, Cox regression, ridge regression, and elastic net (n=2 each). Most studies presented prediction models for risk of CVDs (n=16), all-cause mortality (n=10), CV-related mortality (n=4), rehospitalization (n=4), major adverse cardiovascular event (n=3), non-CVDs (n=3), and hospitalization (n=2). In majority of the studies, AI/ML-based models were reported as superior to CSMs.

CONCLUSIONS: AI/ML is a transformative technology and has immense potential in healthcare domain. Based on this review, we observed that AI/ML-based models demonstrated better performance over CSMs in MI patients. Given the disparity observed across studies, there is need for reporting standards for AI/ML studies.