Abstract
Background Prospectively collected frailty markers are associated with an incremental 1-year mortality risk after transcatheter aortic valve replacement (TAVR) compared with comorbidities alone. Whether information on frailty markers captured retrospectively in administrative billing data is similarly predictive of long-term mortality after TAVR is unknown. We sought to characterize the prognostic importance of frailty factors as identified in healthcare billing records in comparison to validated measures of frailty for the prediction of long-term mortality after TAVR. Methods and Results Adult patients undergoing TAVR between August 25, 2011, and September 29, 2015, were identified among Medicare fee-for-service beneficiaries. The Johns Hopkins Claims-based Frailty Indicator was used to identify frail patients. We used nested Cox regression models to identify claims-based predictors of mortality up to 4 years post-procedure. Four groups of variables, including cardiac risk factors, noncardiac risk factors, patient procedural risk factors, and nontraditional markers of frailty, were introduced sequentially, and their integrated discrimination improvement was assessed. A total of 52 338 TAVR patients from 558 clinical sites were identified, with a mean follow-up time period of 16 months. In total, 14 174 (27.1%) patients died within the study period. The mortality rate was 53.9% at 4 years post-TAVR. A total of 34 863 (66.6%) patients were defined as frail. The discrimination of each of the 4 models was 0.60 (95% CI, 0.59-60), 0.65 (95% CI, 0.64-0.65), 0.68 (95% CI, 0.67-0.68), and 0.70 (95% CI, 0.69-0.70), respectively. The addition of nontraditional frailty markers as identified in claims improved mortality prediction above and beyond traditional risk factors (integrated discrimination improvement: 0.019; P<0.001). Conclusions Risk prediction models that include frailty as identified in claims data can be used to predict long-term mortality risk after TAVR. Linkage to claims data may allow enhanced mortality risk prediction for studies that do not collect information on frailty.