Abstract 17725: Improved Discriminatory Power to Predict Early LVAD Mortality: The Mechanical Circulatory Support Network Risk Score

Abstract

Introduction: Various risk models have been used to predict mortality after implantation of continuous-flow ventricular assist devices (CF-VADs). Discriminatory power and predictive accuracy vary, and there is a need to develop and validate a contemporary risk score in a large multicenter patient cohort. Hypothesis: We sought to develop an early mortality risk score utilizing a large voluntary research registry of patients implanted with CF-VADs, with the focus on increased discriminatory power. Methods: From May 2004 to May 2015, 1150 patients in the Mechanical Circulatory Support Research Network (MCSRN) registry were randomly divided into equal-sized derivation and validation samples (n=575, each group). Candidate predictors from the derivation sample were subjected to backward stepwise logistic regression until the model with lowest Akaike information criterion value was identified for predicting 90-day mortality. A risk score was developed based on the variables and their respective regression coefficients in the final model. Discriminatory power of the risk score was evaluated using receiver operating characteristic analysis, and calibration was assessed using Hosmer-Lemeshow goodness of fit test. Results: Overall 90-day mortality was 9.4% (n=99). The final logistic regression model consisted of 9 predictors: Age at implant, preoperative mechanical ventilation, serum total bilirubin, hemoglobin, tricuspid valve regurgitation severity, left ventricular end-diastolic diameter, cardiac index, pulmonary capillary wedge pressure, and destination therapy (DT) as an indication for pump placement. The area under the curve was 0.86 in the derivation sample (95% CI, 0.75-0.97) and 0.75 in the validation sample (95% CI, 0.63-0.86). Hosmer-Lemeshow goodness of fit test yielded insignificant values in both samples (p=0.75, p=0.58) consistent with good calibration. Conclusions: The proposed 90-day mortality MCSRN risk score provides highest-to-date discriminatory power when compared to already published models. The risk score is applicable to a contemporary large cohort of patients implanted with CF-VADs for either indication (bridge or DT).