Journal of Cardiology and Heart Failure

ABSTRACT

Background

Thoracic endovascular aortic repair (TEVAR) and complex endovascular aortic interventions treat thoracic aortic aneurysm and Stanford type B aortic dissection, but outcomes remain heterogeneous and conventional risk tools provide limited individualized prognostication. We reviewed ML/AI prediction models after TEVAR/complex endovascular repair.

Methods

PubMed, Scopus, and Cochrane were searched from inception to December 2025 (PROSPERO CRD420251267013) per PRISMA 2020. We included adult studies reporting quantitative performance for ML/AI prognostic models and assessed risk of bias using PROBAST.

Results

Five retrospective studies (n=79–10,738) evaluated tree-based algorithms (including XGBoost), decision trees, radiomics-based ML, and deep learning using clinical and/or CT angiography features. Discrimination ranged from approximately AUC 0.70–0.99, and where compared, ML models generally outperformed conventional regression. Calibration and clinical utility were inconsistently reported; only one study used independent external validation. Heterogeneity in populations, predictors, and outcome definitions precluded meta-analysis, and overall risk of bias was moderate-to-high, mainly in the analysis domain.

Conclusion

ML/AI models show promising performance for predicting mortality, adverse events, reintervention, and aortic remodeling after TEVAR, but standardized reporting and multicenter external validation are required before clinical implementation.

Keywords: TEVAR; Thoracic Endovascular Aortic Repair; Complex Endovascular Repair; Machine Learning; Artificial Intelligence; Prediction Model; Radiomics; Deep Learning