Machine learning for predicting outcomes of transcatheter aortic valve implantation: A systematic review

Background

Transcatheter aortic valve implantation (TAVI) therapy has demonstrated its clear benefits such as low invasiveness, to treat aortic stenosis. Despite associated benefits, still post-procedural complications might occur. The severity of these complications depends on pre-existing clinical conditions and patient specific complex anatomical features. Accurate prediction of TAVI outcomes will assist in the precise risk assessment for patients undergoing TAVI. Throughout the past decade, different machine learning (ML) approaches have been utilized to predict outcomes of TAVI. This systematic review aims to assess the application of ML in TAVI for the purpose of outcome prediction.

Methods

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline was adapted for searching the PubMed and Scopus databases on ML use in TAVI outcomes prediction. Once the studies that meet the inclusion criteria were identified, data from these studies were retrieved and were further examined. 17 parameters relevant to TAVI outcomes were carefully identified for assessing the quality of the included studies.

Results

Following the search of the mentioned databases, 78 studies were initially retrieved, and 17 of these studies were included for further assessment. Most of the included studies focused on mortality prediction, utilizing datasets of varying sizes and diverse ML algorithms. The most employed ML algorithms were random forest, logistics regression, and gradient boosting. Among the studied parameters, serum creatinine, age, BMI, hemoglobin, and aortic valve mean gradient were identified as key predictors for TAVI outcomes. These predictors were found to be well aligned with established associations in current literature.

Conclusion

ML presents a promising opportunity for improving the success and safety of TAVI and enhancing patient-centered care. While currently retrospective studies with low generalizability and heterogeneity form the basis of ML TAVI research, future prospective investigations with highly heterogeneous patient TAVI cohorts will be critically important for firmly establishing the applicability of ML in predicting TAVI outcomes.