Study population and variables
We analyzed data of hospital discharge records of patients with the nationally representative Kids’ Inpatient Database (KID) for 2003, 2006, 2009, 2012 and 2016. The data was compiled by the Agency for Healthcare Research and Quality (AHRQ) and was generated for Healthcare Cost and Utilization Project (HCUP) in collaboration with public and private statewide data organizations. The KID is a stratified, cross-sectional database that includes discharge data for approximately 10% of newborn discharges and 80% of other discharges in the United States. KID is published every 3–4 years and the latest data available is from 2016.
Using International Classification of Disease, Ninth and Tenth Revision, Clinical Modification (ICD-9-CM and ICD-10-CM), we identified hospitalized children with CHD, influenza infection, and other variables studied. To include only CHD, we excluded patent ductus arteriosus, single umbilical artery and other anomalies of peripheral vascular system from analysis. Please see Supplementary Table 1 for ICD-9-CM and ICD-10-CM codes used in the manuscript. We included children 1 year and older, excluding the infants(< 1 year) to minimize the confounding effect on the outcome variables, as most of the surgeries for CHD occur during infancy and we also wanted to minimize the effect of patent ductus arteriosus especially on preterm infants, which is minimal after infancy.
Outcome variables
The primary outcome of interest was comparison of in-hospital mortality between those children with influenza infection with and without concomitant CHD. Secondary outcomes were acute respiratory failure, acute kidney injury, need for invasive mechanical ventilation (IMV), non-invasive mechanical ventilation (NIMV), myocarditis, tachyarrhythmias, heart block, sudden cardiac arrest, need for ECMO and length of hospital stay.
Statistical analysis
We performed descriptive and inferential statistics using the KID complex survey design, taking into account for clusters, strata, and weighting. For continuous variables such as age and length of stay, we reported median with interquartile range (IQR). Weight-adjusted Chi-square tests were used for categorical variables and weight-adjusted Wilcoxon signed rank tests for continuous variables as the continuous variables were not evenly distributed.
The variables used in the multivariable analysis were carefully selected after rigorous review of the literature. During selection, we were careful to identify variables that had ICD codes that were reliable and consistent. For regression modeling, univariable analyses of each variable of interest were performed first, followed by a multivariable analysis incorporating additional variables (age, sex, race/ethnicity, discharge quarter, year of admission, history of asthma, presence of respiratory and musculoskeletal congenital anomalies, presence of chromosomal anomalies) to determine the effects of covariates and confounding variables on the outcome of interest. We performed logistic regression analysis for the odds ratios (ORs) of the risk of mortality, acute respiratory failure, acute kidney injury and need for mechanical ventilation with and without concomitant CHD. We assessed differences in length of hospital stay by using multiple linear regression analysis.
We then performed analysis to compare the risk between severe and non-severe CHD. Severe CHD includes truncus arteriosus, d-transposition of great arteries(d-TGA), double outlet right ventricle(DORV), l-transposition of great arteries(L-TGA), tetralogy of Fallot, hypoplastic left heart syndrome(HLHS), other single ventricles, atrio-ventricular septal defect(AVSD), pulmonary atresia, tricuspid atresia, interrupted aortic arch, total anomalous pulmonary venous return(TAPVR). Every congenital heart disease without ICD codes for severe CHD were included as non-severe CHD. In addition, if a child has a diagnosis of severe CHD and non-severe CHD, they are counted as severe CHD, e.g. if a child has d-transposition of great arteries(d-TGA) and VSD, they are counted as d-TGA and not as VSD.
Weights provided by HCUP were used in all analyses to account for the complex sampling design and clustering for the analysis. All statistical analyses were performed using Stata statistical software (version 15.1), R version 3.6.0 [5] and R Studio 1.2 [6]. (http://www.R-project.org) Complex survey design of the KID was accounted for using the survey package [7]. Figures were produced using the ggplot2 and patchwork packages. Tables were generated using the tableone package [8].