Predicting the Risk of Right Ventricular Failure in Patients Undergoing Left Ventricular Assist Device Implantation

Abstract

Background: Right ventricular failure (RVF) is a cause of major morbidity and mortality after left ventricular assist device (LVAD) implantation. It is, therefore, integral to identify patients who may benefit from biventricular support early post-LVAD implantation. Our objective was to explore the performance of risk prediction models for RVF in adult patients undergoing LVAD implantation. Methods: A systematic search was performed on Medline, Embase, Cochrane Central Register of Controlled Trials, and Cochrane Database of Systematic Reviews from inception until August 2019 for all relevant studies. Performance was assessed by discrimination (via C statistic) and calibration if reported. Study quality was assessed using the Prediction Model Risk of Bias Assessment Tool criteria. Results: After reviewing 3878 citations, 25 studies were included, featuring 20 distinctly derived models. Five models were derived from large multicenter cohorts: the European Registry for Patients With Mechanical Circulatory Support, Interagency Registry for Mechanically Assisted Circulatory Support, Kormos, Pittsburgh Bayesian, and Mechanical Circulatory Support Research Network RVF models. Seventeen studies (68%) were conducted in cohorts implanted with continuous-flow LVADs exclusively. The definition of RVF as an outcome was heterogenous among models. Seven derived models (28%) were validated in at least 2 cohorts, reporting limited discrimination (C-statistic range, 0.53–0.65). Calibration was reported in only 3 studies and was variable. Conclusions: Existing RVF prediction models exhibit heterogeneous derivation and validation methodologies, varying definitions of RVF, and are mostly derived from single centers. Validation studies of these prediction models demonstrate poor-to-modest discrimination. Newer models are derived in cohorts implanted with continuous-flow LVADs exclusively and exhibit modest discrimination. Derivation of enhanced discriminatory models and their validations in multicenter cohorts is needed.