A Risk Model for 1-Year Mortality After Transcatheter Aortic Valve Replacement From the J-TVT Registry

Background Although transcatheter aortic valve replacement (TAVR) has demonstrated favorable outcomes in randomized studies, there remains a sizable group of patients in whom TAVR may be futile. Characterizing the survival rate in a wide array of patients undergoing TAVR can help develop effective strategies for improving the allocation of medial resources. Objectives The aim of this study was to develop a risk model to estimate 1-year mortality after TAVR from a representative nationwide registry in Japan. Methods The J-TVT (Japan Transcatheter Valve Therapies) registry contains complete data, including 1-year outcomes, on patients undergoing TAVR in Japan. A total of 17,655 patients underwent TAVR between 2013 and 2018. They were randomly divided into 2 groups in a 7:3 ratio to form a derivation cohort of 12,316 patients and a validation cohort of 5,339 patients. A risk model was constructed for 1-year mortality in the derivation cohort, and its discrimination and calibration were assessed in the validation cohort. Results The mean age of all registered patients was 84.4 years, and 68.8% were women. The mean body size area was 1.43 m2, and the mean Society of Thoracic Surgeons Predicted Risk of Mortality score was 7.3%. The estimated 1-year survival was 91.8%; 202 and 1,316 deaths were observed at 30 days and 1 year, respectively; The estimated C index for the developed model was 0.733 (95% CI: 0.709-0.757) in the validation cohort, with good calibration. Conclusions A prediction model for 1-year survival following TAVR derived from a national clinical database performed well and should aid physicians managing TAVR patients.

T ranscatheter aortic valve replacement (TAVR)  To date, several risk models have been developed to risk-stratify TAVR patients. However, most of these scoring systems provide information on short-term outcomes, including 30-day or in-hospital outcomes. [5][6][7][8][9] However, previously reported quantitative risk assessment of long-term mortality following TAVR has been contemplated mainly using models developed from a limited number of cases at a single center or a few centers and therefore has limited generalizability. [10][11][12][13] An accurate large longitudinal dataset is a prerequisite for developing a long-term risk model.  Facilities must be approved as TAVR implementation facilities through a rigorous selection process by  Values are mean AE SD or n (%).
AV ¼ aortic valve; BMI ¼ body mass index; BSA ¼ body surface area; CABG ¼ coronary artery bypass grafting; CAD ¼ coronary artery disease; COPD ¼ chronic obstructive pulmonary disease; DM ¼ diabetes mellitus; LVEF ¼ left ventricular ejection fraction; NYHA ¼ New York Heart Association; PCI ¼ percutaneous coronary intervention; STS ¼ Society of Thoracic Surgeons. The number of events in the full cohort was 1,316, and the total observation period was 5,567,375 person-days. To assess the performance of the developed model, the model from the complete case analysis was then applied to the patients in the validation cohort to predict their 1-year mortality using the Breslow estimator. 16 We evaluated the model's discriminatory accuracy using Harrell's C index (concordance statistic) and depicted a time-dependent receiveroperating characteristic curve at last event occurrence. 17 Using the validation cohort, we also depicted the calibration plot for predicted vs observed 1-year mortality outcomes in 10 equally sized groups ranked by their predicted values.   performance of the STS PROM score and logistic EuroSCORE using receiver-operating characteristic curves among patients with these data available in the database.
We also conducted an additional analysis with multiple imputation in the whole cohort, using the fully conditional specification approach in the PROC MI package in SAS (SAS Institute), creating 50 datasets from all variables presented in Table 1. We constructed Cox proportional hazards models for 1-year mortality in the 50 development datasets, the estimates were pooled using Rubin's rule, and the differences were assessed against those from the complete case analysis. The performance of the models from multiple imputation analysis were also assessed in the corresponding 50 validation datasets using the pooled C index.
All analyses were performed using SAS version 9.4.    Figure 1). Figure 3 shows the cumulative incidence function for 1-year mortality among 5 patient groups according to their predicted 1-year mortality probabilities in the validation cohort.

RESULTS
Observed 1-year mortality rates were 2.7%, 6.4%, 16.2%, and 27.2% among patients with predicted mortality rates of <5%, 5% to <10%, 10% to <20%, and $20%, respectively. When dividing the patients in the validation cohort into 10 equally sized groups, the model showed good calibration between the predicted and observed mortality estimates from the Kaplan-Meier method (Figure 4), with a slight underestimation of risk in the lower risk groups.
ADDITIONAL ANALYSIS. The variable estimates from the multiple imputation analysis were quite similar to those from the primary analysis (Supplemental Table 2). The pooled C index from the 50 imputed validation cohorts was 0.730 (95% CI: 0.705-0.755). following TAVR and its continued-access registry. 22 The model developed in this study included parameters that were associated with poor prognosis after TAVR. Sex (male), procedural acuity, NYHA functional class III or IV, presence of chronic obstructive pulmonary disease, peripheral vessel disease, porcelain aorta, cancer, and high serum creatinine were previously reported to be predictors following TAVR. 7,12,13,23,24 Furthermore, frailty is a well-known prognostic value after TAVR 12,20,25-27 and is indispensable for developing risk models following TAVR. Frailty assessment includes various indexes. 28 From the PARTNER trial, Green et al 26  Surprisingly, similar mortality rates have been achieved since 2013, when TAVR began in Japan. 14

FUNDING SUPPORT AND AUTHOR DISCLOSURES
The authors have reported that they have no relationships relevant to the contents of this paper to disclose.