A novel, highly discriminatory risk model predicting acute severe right ventricular failure in patients undergoing continuous‐flow left ventricular assist device implant

Abstract

Various risk models with differing discriminatory power and predictive accuracy have been used to predict right ventricular failure (RVF) after left ventricular assist device (LVAD) placement. There remains an unmet need for a contemporary risk score for continuous flow (CF)‐LVADs. We sought to independently validate and compare existing risk models in a large cohort of patients and develop a simple, yet highly predictive risk score for acute, severe RVF. Data from the Mechanical Circulatory Support Research Network (MCSRN) registry, consisting of patients who underwent CF‐LVAD implantation, were randomly divided into equal‐sized derivation and validation samples. RVF scores were calculated for the entire sample, and the need for a right ventricular assist device (RVAD) was the primary endpoint. Candidate predictors from the derivation sample were subjected to backward stepwise logistic regression until the model with lowest Akaike information criterion value was identified. A risk score was developed based on the identified variables and their respective regression coefficients. Between May 2004 and September 2014, 734 patients underwent implantation of CF‐LVADs [HeartMate II LVAD, 76% (n = 560), HeartWare HVAD, 24% (n = 174)]. A RVAD was required in 4.5% (n = 33) of the patients [Derivation cohort, n = 15 (4.3%); Validation cohort, n = 18 (5.2%); P = 0.68)]. 19.5% of the patients (n = 143) were female, median age at implant was 59 years (IQR, 49.4–65.3), and median INTERMACS profile was 3 (IQR, 2–3). RVAD was required in 4.5% (n = 33) of the patients. Correlates of acute, severe RVF in the final model included heart rate, albumin, BUN, WBC, cardiac index, and TR severity. Areas under the curves (AUC) for most commonly used risk predictors ranged from 0.61 to 0.78. The AUC for the new model was 0.89 in the derivation and 0.92 in the validation cohort. Proposed risk model provides very high discriminatory power predicting acute severe right ventricular failure and can be reliably applied to patients undergoing placement of contemporary continuous flow left ventricular assist devices.