LAUSR

The use of statistical risk models has become an integral part of cardiovascular medicine. Risk assessment for transcatheter aortic valve replacement (TAVR) is particularly challenging because of the need to predict risk for surgical aortic valve replacement as well as the risk of the TAVR procedure itself. In the United States, Society of Thoracic Surgeons (STS) risk models are typically used to estimate surgical aortic valve replacement risk to categorize TAVR candidates into low-, intermediate-, or high-risk groups. These STS models are fully disclosed in the literature and provided to all STS database participants, thereby making them readily available for widespread use.1 As pointed out by Rogers et al,2 the models must be used carefully, with full recognition of their designed features. Temporal changes in the patient population and procedural outcomes are known to occur, so the models are frequently updated to accurately reflect these changes. As illustrated by Rogers et al,2 inaccuracies and misclassification are likely if models developed for a certain year are applied to a patient population from a different year. This clearly constitutes a misuse of the STS surgical aortic valve replacement risk models. Another misuse of STS models is seen in the unfortunate tendency to use STS models to predict the risk of the TAVR procedure itself. There is widespread recognition that this yields inadequate estimates of TAVR risk.3,4 Statistical risk models are designed to predict risk only for the population used to develop the models. To predict the risk of TAVR procedures, then, one should use risk models that have been developed solely for a TAVR population.3,5 Such models based entirely on a TAVR population have recently been published,6-8 and heart teams should be encouraged to use these models to estimate the risk of TAVR procedures. Of these, the model from the STS/American College of Cardiology Transcatheter Valve Therapy Registry is based on the largest patient cohort and uses a “real-world” experience to predict inhospitalTAVRmortality.8 Othermodelsattempttopredictlongerterm mortality, albeit from a select clinical trial population.6 Prediction of adverse outcomes is clearly important, but especially in the TAVR population, there is a compelling need to also predict patient benefit. Even if there is a low chance of adverse events, the procedure should not be performed unless there is a reasonable probability of providing a tangible benefit. It may be possible to infer benefit by comparing preprocedure and postprocedure measurements of parameters that reflect physiologic status. Examples include grip strength and the 5-m walk test to measure frailty9 and the Kansas City Cardiomyopathy Questionnaire10 to quantitate symptoms of heart failure. If these metrics can be collected before and after the procedure, it then becomes possible to develop statistical models to predict preprocedure/postprocedure difference based on specific patient characteristics. If the preprocedure/ postprocedure difference is a reflection of “benefit,” then developing predictive models of patient benefit should be straightforward. The ability to objectively predict patient benefit as well as patient risk will be critically important in the highrisk TAVR population. Statistical models to predict patient benefit are being developed by the Transcatheter Valve Therapy Registry and should provide a significant upgrade in our ability to evaluate candidates for the TAVR procedure. The unique aspects of TAVR have hastened the development of a new generation of predictive models. Reliance on models designed for other purposes should appropriately diminish as these newer, more procedure-specific models come on line as the standard for TAVR risk assessment.