Statistical Analysis
As recommended by the Agency for Healthcare Research and Quality,
weighted data were used for all statistical analyses. Temporal trends in
LVAD utilization as well as post-LVAD mortality were assessed using the
average annual growth rate formula. Baseline characteristics and
post-LVAD outcomes were compared using the Pearson Chi-Squared (χ2)
tests for categorical variables, independent samples T-test for
parametric continuous variables, and Mann-Whitney U test for
non-parametric continuous variables. We considered statistical
significance when p value was below 0.001. Categorical variables of
interest included hypertension (HTN), diabetes mellitus (DM),
malnutrition, coronary artery disease (CAD), atrial tachyarrhythmias
(Atach) which include atrial fibrillation and atrial flutter, peripheral
arterial disease (PAD), chronic kidney disease (CKD), history tobacco
use, history of stroke, dyslipidemia, and chronic liver disease (CLD).
Continuous variables of interest included mean age, total cost of
hospitalization, and hospital length of stay. NIS provides median
household income for patient’s ZIP code divided by percentile.
High-income patients were considered between the 51stto 100th percentile and low-income individuals were
considered between 0 to 50th percentile. The
association between income and post-LVAD mortality was analyzed using
multivariable logistic regression. All multivariable regression models
were created using generalized estimating equations. Missing data for
race were handled using multiple imputation as recommended by Healthcare
Cost and Utilization Project. Missing primary payer status in patients
65 years old or older was imputed to Medicare, whereas missing data for
all other variables were imputed to the dominant category (See
supplemental material). All statistical analyses were performed using
SPSS (IBM SPSS Statistics for Mac, Version 23.0. Armonk, NY: IBM Corp.).