Statistical Analysis
We used descriptive statistics and bivariate tests to characterize distributions of covariates across groups for the overall cohort and among those hospitalized. Variables associated with the outcome at a p-value <0.20 in bivariate tests were included in multivariable analyses. We used multilevel logistic regression models for all outcomes to account for clustering of patient outcomes among individual hospitals in the health system. Models estimated odds ratios and 95% confidence intervals (CI).
In sensitivity analyses, we analyzed all-cause mortality and hospitalization using a multilevel Cox proportional hazards model instead. Further, we analyzed AKI and ICU using a multilevel Cox proportional hazards model with mortality as a competing risk. The Cox proportional hazards assumption was verified graphically by generating a plot of the Schoenfeld residuals. In all instances, our primary results were robust and sensitivity analyses yielded similar inferences. Analyses were conducted with R, Version 4.0.0, and used 2-sided statistical tests with a p value < 0.05 as statistically significant in final analyses.