47 Inadequacy of Existing Clinical Prediction Models for Predicting Mortality Post Transcatheter Aortic Valve Implantation. (3rd June 2016)

• Record Type: Journal Article
• Authors: Martin, Glen; Sperrin, Matthew; Ludman, Peter; Belder, Mark de; Gale, Chris; Toff, William; Buchan, Iain; Mamas, Mamas
• Is Part Of: Heart. Volume 102(2016)Supplement 6
• Page Start: A34
• Page End: A34
• Publication Date: 2016-06-03
• Subjects: TAVI -- Risk Prediction -- Aortic Stenosis; Heart -- Diseases -- Treatment -- Periodicals; Cardiology -- Periodicals; 616.12
• DOI: 10.1136/heartjnl-2016-309890.47
• Languages: English
• ISSNs: 1355-6037

Abstract

Abstract : Introduction: Clinical prediction models (CPMs) form the cornerstone of risk stratification for patients undergoing invasive procedures, helping to guide both treatment allocation and the consent process by enabling discussion of the magnitude of risks associated with the proposed treatment strategy. Several CPMs are used to predict mortality after Transcatheter Aortic Valve Implantation (TAVI). However, their performance in national cohorts distinct from those where they have been derived is unknown. We examined the performance of the Logistic EuroSCORE (LES), EuroSCORE II, Society of Thoracic Surgeons (STS), German Aortic Valve, FRANCE-2 TAVI and OBSERVANT-TAVI risk models in the British Cardiovascular Intervention Society UK TAVI registry, to examine if any can be used for practical use in external national cohorts of TAVI patients. Methods: The risk of 30-day mortality given by each CPM was retrospectively calculated, based on the published regression coefficients, for all 3980 patients in the UK TAVI registry from 2007 to 2012. Life status information was provided by the Office for National Statistics for each patient. Performance of each CPM was investigated through calibration and discrimination. Calibration is the agreement between the expected and observed event rates and was analysed by a calibration plot. Discrimination is how well the CPM can distinguish between those who will experience an event and those who will not, which was analysed by the area under the ROC curve (AUC). Furthermore, the level of agreement in risk classification between each of the considered CPMs was analysed at an individual patient level. Results: The observed 30-day mortality rate was 6.23%. All of the CPMs were miscalibrated and grossly over-predicted risk of 30-day mortality; the LES over predicted by a factor of three, while the STS underestimated risk. Discrimination was also poor, with AUC values of 0.58, 0.57, 0.62, 0.60, 0.62 and 0.57 for the LES, EuroSCORE II, STS, the German Aortic Valve, FRANCE-2 TAVI and OBSERVANT models, respectively. Additionally, risk classification agreement was low in both the surgical models and the TAVI specific models, an example is given in Figure 1 . Conclusions: The performance of CPMs currently used in TAVI risk prediction was low when applied in a cohort of TAVI patient independent to those they were developed on. The TAVI specific FRANCE-2 CPM performed better than cardiac surgery based CPMs. Thus, development of TAVI CPMs are recommended either by updating existing CPMs or developing new risk scores in populations of interest.