Machine Learning for Predicting Mortality Following Transcatheter Aortic Valve Replacement: A Scoping Review of Current Literature

Abstract

This scoping review aims to consolidate the current evidence on the use of machine learning (ML) algorithms to predict mortality following transcatheter aortic valve replacement (TAVR) procedures. This review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. Electronic databases, including Medline, Embase, Cochrane, and ClinicalTrials.gov, were searched for studies assessing the use of ML models for the prediction of mortality in patients undergoing TAVR. A total of 1076 citations were screened, of which 21 studies underwent full-text review. Ultimately, 11 studies comprising 101,030 patients were included for data extraction. Study sample sizes ranged from 178 to 54,739 patients, with reported mortality intervals ranging from 30 days to 5 years after surgery. The parameters that were included in these ML models included demographic and clinical variables, electrocardiographic findings, laboratory results, computed tomography scans, and the type and size of the selected TAVR device. A total of 13 ML models were assessed across the included studies. Random Forest appeared to be the highest performing ML model in predicting 30-day and 1-year survival following TAVR, with reported area under the receiver operating characteristic (AUROC) scores ranging from 0.66 to 0.97. In addition, the gradient-boosted machine seemed to outperform the CoreValve and TAVI2-SCORE in predicting 1-year survival. ML models may allow for accurate prediction of mortality following TAVR and may offer advantages over traditional clinical risk scores. However, limited datasets and variability in algorithm metric scores remain the main limitations to the implementation of these models in clinical practice.