External Validation of Mortality Prediction Models in Japanese Transcatheter Aortic Valve Replacement Registry

Abstract

Objectives: Mortality prediction models (MPMs) play a crucial role in risk assessment for transcatheter aortic valve replacement (TAVR), but their external validity in Japanese patients remains unclear. This study evaluated the performance of existing short-term (30-day) and long-term (1-year) MPMs with Japanese TAVR patients.

Methods: We analysed patients who underwent TAVR between 2016 and 2023 in the KPUM transcatheter aortic valve implantation (TAVI) registry, a prospective multicentre registry in Japan. Five short-term (30-day) MPMs (Society of Thoracic Surgeons Predicted Risk of Mortality [STS-PRoM], French Aortic National CoreValve and Edwards registry [FRANCE-2], Observational Study Of Appropriateness, Efficacy And Effectiveness of AVR-TAVR Procedures For the Treatment Of Severe Symptomatic Aortic Stenosis registry [OBSERVANT], American College of Cardiology-TAVI [ACC-TAVI], and Netherlands Heart Registration [NHR] models) and 3 long-term (1-year) MPMs (Optimized CathEter vAlvular iNtervention-TAVI [OCEAN-TAVI], Osaka University, and TAVR-Risk [TARI] models) were validated. Model performance was assessed using the area under the receiver operating characteristic curve (AU-ROC) for discrimination and calibration plots/slope/intercept for calibration. Intercept recalibration was performed for short-term MPMs.

Results: Among 1756 patients analysed for short-term mortality, the mortality risk was 1.4% (n = 25), while among 1235 patients analysed for long-term mortality, the mortality risk was 14.5% (n = 179), respectively. Among short-term MPMs, NHR (AU-ROC: 0.82) and STS-PRoM (0.78) showed the highest discrimination, though all models overestimated mortality risk, which improved after intercept recalibration. Among long-term MPMs, the OCEAN-TAVI and Osaka University models (AU-ROC: 0.75) showed the highest discrimination, with the OCEAN-TAVI demonstrating the best calibration.

Conclusions: In Japanese TAVR patients, STS-PRoM and NHR models showed superior short-term performance after recalibration, while OCEAN-TAVI demonstrated the best overall long-term performance. Future research should explore machine learning-based models to improve risk prediction accuracy and clinical applicability.

Keywords: aortic stenosis; calibration; discrimination; external validation; mortality; prediction model; transcatheter aortic valve replacement (TAVR).