Transcatheter Aortic Valve Implantation Futility Risk Model - Development and Validation among Treated Aortic Stenosis Patients - The 64th Annual Conference of the Israel Heart Society & The Israel Society of Cardiothoracic Surgery, 25-26 April 2017

Uri Landes ¹ Oren Zusman ¹ Katia Orvin ¹ Amos Levi ¹ Guy Witberg ¹ Adi Lador ¹ Abid Assali ¹ Hana Vaknin-Assa ¹ Ram Sharony ² Ashraf Hamdan ¹ Yaron Shapira ¹ Mordehay Vaturi ¹ Shmuel Schwartzenberg ¹ Alexander Sagie ¹ Ran Kornowski ¹

¹Department of Cardiology, Rabin Medical Center, Petach Tikva and Sackler Faculty of Medicine, Tel Aviv University
²Department of Cardiothoracic Surgery, Rabin Medical Center, Petach Tikva and Sackler Faculty of Medicine, Tel Aviv University

Background:
Optimal patients` selection for transcatheter-aortic-valve-implantation (TAVI) is still unclear. A sizeable group of patients do not benefit from the intervention despite successful procedure. We sought to develop a risk model in order to predict those ‘futile’ TAVI procedures.

Methods:
We included all patients that underwent TAVI in our medical center with device success and with no major and/or debilitating complications as defined by the VARC-2, leaving those with an uncomplicated optimal TAVI for the present analysis. We examined various demographic data, clinical details and echo-cardiographic findings, including patients’ frailty assessment. The outcome was defined as 1-year composite score of mortality, stroke, NYHA functional-class lack of improvement (vs. baseline), and repeat (≥1 month) cardiovascular admissions. We used logistic regression and support vector machine (SVM, a machine learning method) to fit the prediction model. We used 10-fold cross-validation to verify our results.

Results:
Out of 567 patients, 482 met the inclusion criteria. Mean age 82 (±6.5) years, 56% females. At 1-year, 102/482 (21%) patients experienced the adverse futility related outcome. The final model included age, sex, weight, height, frailty, albumin, creatinine, hemoglobin, aortic and mitral insufficiency grade, history of diabetes, COPD, and baseline ECG conduction-delay. Using a regression model, the area under the curve (AUC) was 0.75 and was reduced to 0.69 after validation, with 96% specificity and 2.9% false (+). With SVM, the AUC after validation was also 0.69, but specificity rose to 99.7% with only one patient misclassified as false-positive, and with positive/negative predictive values of 0.97/0.85, respectively. Using the model could identify futile TAVI upfront in a third of the cases (31% sensitivity).

Conclusions:
Without depriving treatment from those who may need it, the TAVI futility model could provide important insight and aid identifying patients who, despite successful/uncomplicated procedure, might not derive benefit and/or longevity following TAVI.