Risk prediction model for waitlist mortality in patients with left ventricular assist devices

Abstract

Background

Left ventricular assist devices (LVAD) are a bridge to heart transplantation (HT). Given limited donor organs, assessment of risk of waitlist mortality is important for waitlist prioritization for HT. We sought to derive and validate a risk prediction model for waitlist mortality in LVAD patients.

Methods

Adult patients with a continuous-flow, centrifugal, durable LVAD listed or likely to be listed for HT in the Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS) were included. The outcome was time to all-cause mortality within 2 years from implant. We considered 41 candidate predictors at 3 months post-implant. Univariate Fine-Gray models and 4 logistic regression techniques (logistic, LASSO, random forest, gradient boosting) were used to select variables for a final survival model using the Fine-Gray method. The model was validated in INTERMACS and in an independent cohort (European Registry for Patients with Mechanical Circulatory Support [EUROMACS]). Model discrimination and calibration were evaluated.

Results

The INTERMACS cohort included 2364 patients with 268 (11%) deaths. A risk prediction model for waitlist mortality at 2 years was derived with area-under-the-curve (AUC) of 0.72 (95% CI 0.67–0.77). The EUROMACS cohort included 577 patients with 70 (12%) deaths. The model AUC was 0.62 (95% CI 0.55–0.70). The model predicted waitlist mortality when divided into low-, medium-, or high-risk groups in the INTERMACS (p<0.001) and EUROMACS (p=0.0099) cohorts.

Conclusions

We derived and validated a risk prediction model for waitlist mortality in LVAD patients using 2 independent cohorts. Our risk assessment model can inform HT prioritization in LVAD patents.