Statistical Analysis:
Continuous variables are presented as means with standard deviation or median with interquartile ranges (IQRs) based on distribution and categorical variables are presented as frequencies (percentages). To address our first aim, the number of ICD hospital admissions per 100,000 pediatric hospitalizations by year was determined and graphed by year. The rates of hospitalizations appeared linear and thus linear regression was used to evaluate the changes in hospitalization rates over time. Linear regression was also used to evaluate changes in hospitalization rates by underlying diagnostic category over time. For the second aim, mortality overall and by grouped year was calculated as deaths over all ICD admissions.
For the third aim, we first performed univariate logistic regression analysis to evaluate associations between independent variables and the outcome of hospital death. Variables associated with mortality with a p <0.2 were then included in a multivariable analysis. To account for hospital clustering, multivariable mixed logistic regression modeling was used, using hospital as a random effect. Given the relatively lower number of outcomes and multiple potential degrees of freedom for the multivariable analysis, we assessed the race/ethnicity variable and noted similar odds in non-Hispanic white, Hispanic, and other compared to Black race. Therefore, race was condensed into a binary variable of Non-Hispanic Black and Non-Black categories to limit the degrees of freedom for the multivariable analysis.
Given that a diagnosis of heart failure was possibly in the mechanistic pathway of death, it was initially omitted from the multivariable model. In a secondary analysis, to evaluate for the possible intermediary effects of heart failure, an exploratory analysis was then performed including heat failure in the multivariable model. To accommodate for multiple testing when including heart failure as a covariate, a Bonferroni correction was applied to the final exploratory model and a p value of <0.008 was considered statistically significant. For all other models and testing, a two-sided p-value of < 0.05 was considered significant. Statistical analysis was performed using SAS version 9.4 (SAS Institute Inc., Cary, NC).