Machine Learning Models for Predicting Permanent Pacemaker Implantation After Transcatheter Aortic Valve Replacement:...

Permanent pacemaker (PPM) implantation has been reported in up to 26% of patients undergoing transcatheter aortic valve replacement (TAVR). Machine learning (ML) models have increasingly been developed to predict the need for PPM after TAVR, offering the potential to improve preprocedural risk stratification and optimize perioperative management. We conducted a systematic literature search of MEDLINE, Embase, Cochrane Central, ClinicalTrials.gov, and Google Scholar to identify primary studies evaluating ML models for predicting PPM implantation following TAVR. Data extracted included study characteristics, ML methodologies, predictor variables, and model performance metrics. Seven studies comprising 4528 patients who underwent TAVR met the inclusion criteria. The evaluated TAVR devices included balloon-expandable, self-expandable, and mechanically expandable Lotus valves. The incidence of PPM implantation ranged from 14% to 41.7%. ML approaches included random forest, gradient boosting, neural networks, support vector machines, and logistic regression. Predictive performance varied across studies, with area under the receiver operating characteristic curve values ranging from 0.61 to 0.92. Frequently identified predictors of PPM implantation included prosthetic valve size, preprocedural right bundle branch block, atrioventricular block, leaflet calcification, larger left ventricular outflow tract diameter, prior aortic valve interventions, and use of self-expanding valves. Overall, ML models demonstrate promising predictive capability for identifying patients at risk of requiring PPM after TAVR, although model performance remains inconsistent across studies. Further external validation, standardization of model development, and prospective evaluation are needed before these tools can be confidently incorporated into routine clinical decision-making and risk stratification following TAVR.