April 1, 2020
View Full PaperSystematic review and meta-analysis of current risk models in predicting short-term mortality after transcatheter aortic valve replacement
Key Result
Current risk stratification models for TAVR showed poor discrimination (C-statistic <0.7), but the STS/ACC TVT and STS models demonstrated accurate calibration for short-term mortality.
Study Design
Type
Meta-Analysis (n=68,215)
Structured PICO
How do current risk stratification models perform in predicting short-term mortality after transcatheter aortic valve replacement?
P
Population
68,215 patients across 24 studies undergoing transcatheter aortic valve replacement (TAVR)
I
Intervention
Risk stratification models (11 different models including STS/ACC TVT, STT, and STS)
O
Outcome
30-day or in-hospital mortality (assessed via discrimination/C-statistics and calibration/OERs)hard clinical
While current TAVR risk models have accurate calibration useful for center-level comparisons, their ability to discriminate individual patient mortality risk remains limited.
Abstract
AIMS: The aim of this study was to evaluate the performance of risk stratification models (RSMs) in predicting short-term mortality after transcatheter aortic valve replacement (TAVR). METHODS AND RESULTS: MEDLINE and Scopus were queried to identify studies which validated RSMs designed to assess 30-day or in-hospital mortality after TAVR. Discrimination and calibration were assessed using C-statistics and observed/expected ratios (OERs), respectively. C-statistics were pooled using a random-effects inverse-variance method, while OERs were pooled using the Peto odds ratio. A good RSM is defined as one with a C-statistic >0.7 and an OER close to 1.0. Twenty-four studies (n=68,215 patients) testing 11 different RSMs were identified. Discrimination of all RSMs was poor (C-statistic <0.7); however, certain TAVR-specific RSMs such as the in-hospital STS/ACC TVT (C-statistic=0.65) and STT (C-statistic=0.66) predicted individual mortality more reliably than surgical models (C-statistic range=0.59-0.61). A good calibration was demonstrated by the in-hospital STS/ACC TVT (OER=0.99), 30-day STS/ACC TVT (OER=1.08) and STS (OER=1.01) models. Baseline dialysis (OER: 2.64 1.88, 3.70; p<0.001) was the strongest predictor of mortality. CONCLUSIONS: This study demonstrates that the STS/ACC TVT model (in-hospital and 30-day) and the STS model have accurate calibration, making them useful for comparison of centre-level risk-adjusted mortality. In contrast, the discriminative ability of currently available models is limited.