Validation of clinical scores for right ventricular failure prediction after implantation of continuous-flow left ventricular assist devices.

Background: Several clinical prediction schemes for right ventricular failure (RVF) risk after left ventricular assist device (LVAD) implantation have been developed in both the pulsatile- and continuous-flow LVAD eras. The performance of these models has not been evaluated systematically in a continuous-flow LVAD cohort.

Methods: We evaluated 6 clinical RVF prediction models (Michigan, Penn, Utah, Kormos et al, CRITT, Pittsburgh Decision Tree) in 116 patients (age 51 ± 13 years; 41.4% white and 56.0% black; 66.4% men; 56.0% bridge to transplant, 37.1% destination therapy, 17.4% bridge to decision) who received a continuous-flow LVAD (HeartMate II: 79 patients, HeartWare: 37 patients) between 2008 and 2013.

Results: Overall, 37 patients (31.9%) developed RVF, defined: as pulmonary vasodilator use for ≥48 hours or inotrope use for ≥14 days post-operatively; re-institution of inotropes; multi-organ failure due to RVF; or need for mechanical RV support. Median (Quartile 1 to Quartile 3) time to initial discontinuation of inotropes was 6 (range 4 to 8) days. Among scores, the Michigan score reached significance for RVF prediction but discrimination was modest (C = 0.62 [95% CI 0.52 to 0.72], p = 0.021; positive predictive value [PPV] 60.0%; negative predictive value [NPV] 75.8%), followed by CRITT (C = 0.60 [95% CI 0.50 to 0.71], p = 0.059; PPV 40.5%; NPV 72.2%). Other models did not significantly discriminate RVF. The newer, INTERMACS 3.0 definition for RVF, which includes inotropic support beyond 7 days, was reached by 57 patients (49.1%). The Kormos model performed best with this definition (C = 0.62 [95% CI 0.54 to 0.71], p = 0.005; PPV 64.3%; NPV 59.5%), followed by Penn (C = 0.61), Michigan (C = 0.60) and CRITT (C = 0.60), but overall score performance was modest.

Conclusion: Current schemes for post-LVAD RVF risk prediction perform only modestly when applied to external populations.