Predicting Right Ventricular Failure Following Left Ventricular Assist Device Support: A Derivation-Validation Multicenter Risk Score

Purpose: Despite several models predicting right ventricular failure (RVF) after durable left ventricular assist device (LVAD) support, poor performance when externally validated has limited their widespread use. We sought to derive a predictive model for RVF after LVAD implantation, and ascertain its performance in an independent cohort. Methods: End-stage heart failure (HF) patients requiring continuous-flow LVAD were prospectively enrolled at one US program (n=477, derivation cohort), with two other US medical centers forming the validation cohort (n=321). The primary outcome was RVF incidence, defined as the need for right ventricular assist device or inotropes for \u3e14 days. Multivariable logistic regression in the derivation set yielded a RVF predictive model, which was subsequently applied to the validation cohort, and a risk score was ultimately developed. Results: Derivation cohort included patients less likely to be African-Americans (7% vs 37%; p\u3c0.001), Hispanics (7% vs 30%; p\u3c0.001), have a remote history of hypertension (49% vs 60%; p=0.002) or be bridged with short-term MCS (8% vs 16%; p=0.001), compared to the validation set. RVF incidence was 16% in the derivation and 36% in the validation cohort (p\u3c0.001). Multivariable analysis identified 7 variables (Figure) as predictive of RVF, with the model achieving a C statistic of 0.734 (95% CI=0.674-0.794) in the derivation and 0.709 (95% CI=0.651-0.767) in the heterogeneous validation cohort. Patients were stratified into 3 RVF risk groups (all comparisons; p\u3c0.001) (Figure). Conclusion: We propose a novel scoring system to predict post-LVAD RVF, achieving high discriminative performance in distinct, heterogeneous LVAD cohorts