External validation of the Netherlands Heart Registration (NHR) prediction model for early mortality after transcatheter aortic valve implantation (TAVI).

We have read with interest the paper by Al-Farra et al.1 recently published in the Journal. Early mortality following TAVI, although being rare nowadays, is a key safety outcome related to the procedure. Risk stratification of early mortality in patients undergoing TAVI may be useful for better patient selection and adequate procedural planning. The NHR risk score developed by Al-Farra et al.1 may be useful in this regard. External validation of a risk score, however, is crucial to evaluate its performance before clinical implementation.2 We performed therefore an independent external validation of the NHR risk score using data collected in a large Swedish registry. We used the Swedish Transcatheter Cardiac Intervention Registry (SWENTRY), which is part of the Swedish System for enhancement and development of evidence-based care in heart disease evaluated to recommended therapies (SWEDEHEART)3 and holds pre- and periprocedural data on all Swedish patients undergoing TAVI in the country. We identified all Swedish patients undergoing TAVI between 2008 and 2021 (n = 8463). The registry was further linked to the Population Register and the National Prescribed Drugs Register (NPDR). No exclusion criteria were adopted. All variables were retrieved directly from SWENTRY except for prescribed diabetic medications where information was collected from the NDPR. The number of missing data was low in the registry. Short-term mortality in SWENTRY was 2.9%, and the NHR risk score had moderate discrimination with a concordance index (c-index) of 0.67. Model calibration was accurate up to a mortality risk of 4%–5%, which included the majority of patients (Figure 1). For higher risk categories, the model tended to underestimate the observed risk. A re-estimated model on our data set indicated differences in variable contribution. Procedure acuteness was associated with greater risk. However, it only applied to procedures outside office hours or in a critical patient. Age, impaired left ventricular ejection fraction, and a higher New York Heart Association class were associated with greater risk than that estimated in the original model, while a critical perioperational state, diabetes without medication, and body surface area had a lower impact on mortality. Direct aortic access was also associated with a higher risk for death than the transapical approach, which was different than in the NHR risk score. However, altogether, the predictors managed a fair discrimination. Two sensitivity analyses were performed. One with the inclusion or exclusion of only SGLT2 inhibitors as diabetes medication, as the original model, included patients between 2013 and 2018, where treatment with SGLT2 inhibitors was infrequently used, and one analysis on patients between the years of 2019 and 2021 as the variable of left ventricular ejection fraction changed from classes to continuous in SWENTRY. The results were consistent in both analyses, although the underestimation of risk in the higher risk interval was more evident in the latest years. In conclusion, the TAVI-NHR risk model had acceptable discrimination for short-term mortality in a large cohort of Swedish patients undergoing TAVI. These findings support the possibility of transporting the model to a Scandinavian setting. Stefan James reports proctoring fees from Medtronic unrelated to the present study. The remaining authors report no conflict of interest.