Impact of coinfection status and comorbidity on disease severity in adult emergency department patients with influenza B

Abstract Background Influenza B accounts for approximately one fourth of the seasonal influenza burden. However, research on the importance of influenza B has received less attention compared to influenza A. We sought to describe the association of both coinfections and comorbidities with disease severity among adults presenting to emergency departments (ED) with influenza B. Methods Nasopharyngeal samples from patients found to be influenza B positive in four US and three Taiwanese ED over four consecutive influenza seasons (2014–2018) were tested for coinfections with the ePlex RP RUO panel. Multivariable logistic regressions were fitted to model adjusted odds ratios (aOR) for two severity outcomes separately: hospitalization and pneumonia diagnosis. Adjusting for demographic factors, underlying health conditions, and the National Early Warning Score (NEWS), we estimated the association of upper respiratory coinfections and comorbidity with disease severity (including hospitalization or pneumonia). Results Amongst all influenza B positive individuals (n = 446), presence of another upper respiratory pathogen was associated with an increased likelihood of hospitalization (aOR = 2.99 [95% confidence interval (95% CI): 1.14–7.85, p = 0.026]) and pneumonia (aOR = 2.27 [95% CI: 1.25–4.09, p = 0.007]). Chronic lung diseases (CLD) were the strongest predictor for hospitalization (aOR = 3.43 [95% CI: 2.98–3.95, p < 0.001]), but not for pneumonia (aOR = 1.73 [95% CI: 0.80–3.78, p = 0.166]). Conclusion Amongst ED patients infected with influenza B, the presence of other upper respiratory pathogens was independently associated with both hospitalization and pneumonia; presence of CLD was also associated with hospitalization. These findings may be informative for ED clinician's in managing patients infected with influenza B.

to 650 000 deaths globally 1 and account for between 9 and 45 million cases of illnesses annually in the United States alone. 2 Seasonal epidemics are caused by two types of influenza viruses: influenza A and B. 3 Historically, research has largely focused on influenza A due to its genetic variability, seasonal dominance, and pandemic potential. [4][5][6][7] Influenza B viruses primarily infect humans without continuous circulation in animal reservoirs and undergo slower genetic variability over time. 8 Thus, their pandemic potential is perceived to be less than influenza A. 7 However, influenza B viruses have the potential to predominate seasonal circulation as demonstrated during the 2019-2020 season 2 and have been shown to be associated with severe disease, in both pediatric and young adult populations. 5,9,10 The presence of comorbidities, including cardiovascular diseases and chronic lung diseases, represent known risk factors for severe influenza complications, 2 but these risk factors have primarily been derived from studies of patients with influenza A. 7,[11][12][13] In addition, the presence of coinfections with bacterial and/or viral pathogens has been found to be associated with increased disease severity, from studies of individuals with influenza A. [14][15][16] Systematic epidemiologic analyses regarding the association of comorbidity and coinfections in those infected influenza B have been relatively limited. 5,7,17 In this investigation, we retrospectively evaluated a large biorepository of nasopharyngeal specimens (NPs) taken from ED   For all subjects who tested positive for influenza, a structured data collection form was completed by trained research coordinators using information from the electronic health records (EHR).
Data gathered included demographic characteristics, clinical information (including comorbidities), and clinical outcomes (including oxygen supplementation, disposition, length of hospital stay, and pneumonia diagnosis, as determined by the attending physician through radiological interpretation). Each patient was assigned a deidentified number. All de-identified data were entered into a secure Research Electronic Data Capture database, and the dataset underwent rigorous quality control measures to ensure accuracy.

| Data analysis
Assessment of coinfections with 17 common viral and four atypical bacterial respiratory pathogens represented the primary explanatory variable for severe influenza B disease in our analyses. All models were adjusted for demographic factors and underlying health conditions according to the Centers for Disease Control and Prevention (CDC) classification of individuals at increased risk of influenza complications. 2,19 To control for clinical risk upon presentation to the ED, we adjusted for the National Early Warning Score for acutely ill patients (NEWS) 20 as a validated risk predictor for respiratory infections. 21 Classifications based on the NEWS range between 0 and greater 7. 20,21 Hospitalization and pneumonia diagnosis were used here as our outcome measure for disease severity; each has previously been reported as measure of disease severity, 12,22,23 and both were systematically collected across all years and all study sites.
Disposition and pneumonia diagnosis were dichotomized (admitted v. discharged patients and pneumonia v. no pneumonia diagnosis as determined by radiological findings).

| Univariate statistics and bivariate analyses
Characteristics of the study population were assessed using summary statistics, while distributions were inspected via Q-Q-plots and Shapiro-Wilk test for normality. 24

| Multivariable logistic regression models
All baseline models were adjusted for age, gender, race, and ethnicity.
The covariate for age was centered at the average of 43.2 years for better interpretability of adjusted estimates. The indicators for race and ethnicity were coded with White and non-Hispanic/non-Latino participants constituting the reference group, respectively. Separate multivariable logistic regressions for each of the two severity outcomes were built by iteratively including predictor variables and assessing model fit improvement via likelihood ratio test (LRT) statistics. These manually fitted models were validated by using computational algorithms for forward and backward selection based on Aikaike Information Criterion (AIC) values as well as best subset variable selection algorithms. 25 Models for hospitalization were fitted using linear indicators for NEWS and underlying health conditions.
For the pneumonia models, binary indicators with cut-offs at NEWS = 5 and one underlying health condition were selected based on superior model performance and previous literature. 26 To adjust for heterogeneity across study sites, robust variance estimates 27 based on Huber 28 and White 29 were used.
Interactions between coinfection status and other severity predictors were inspected by testing interaction terms and assessing changes in model fit via LRT statistics. Adjusted probabilities for the two severity outcomes by coinfection status were predicted from the fitted main models, holding other predictors constant at their mean.

| Comorbidity analysis
A secondary analysis on the impact of individual comorbidities on disease severity was conducted. Respiratory conditions, including chronic obstructive pulmonary disease, asthma, cystic fibrosis, tuberculosis, and emphysema, were combined into one predictor as chronic lung diseases (CLD). Adjusted probabilities were estimated analogously to the primary models from the fitted regression models.

| Model validation
Goodness of model fit was assessed using Hosmer-Lemeshow test statistics, 30 and collinearity of predictors was inspected through variance inflation factors (VIF). As test statistics for both models were non-significant (p > 0.05) and VIF consistently ranked below VIF = 5, adequate fit for the estimated models can be assumed. Validation of model predictions was performed through k-fold cross-validation techniques by iteratively leaving out one observation.
All statistical analyses were performed in Stata 15.1 (StataCorp LLC), and figures were refined using Prism9 (GraphPad software).

| Summary of the study population
A summary of the study population characteristics, comorbidities, and major disease severity outcomes is displayed in Table 1

| Models for hospitalization
Adjusted models for estimating the odds of clinical outcomes by severity predictors are shown in Table 2 Figure 2A).

| Interactions with coinfection status
All assessed interactions between coinfection status and other predictors yielded statistically nonsignificant results across the two severity outcome models.

| Secondary analysis for chronic lung diseases
In-depth analyses assessing the effect of specific comorbidities found that CLD was the strongest individual predictor variable for hospitali-  Figure 2B).

| DISCUSSION
In this large sample of individuals infected with influenza B (across 5 large hospitals and multiple influenza seasons), we found that having an upper viral and/or bacterial coinfection was strongly associated with more severe disease (as measured by either hospitalization or pneumonia). Having chronic lung diseases (CLD) was independently associated with hospitalization but did not increase the likelihood of being diagnosed with pneumonia.
Global influenza surveillance 1,2,31 and multiple scientific  In conclusion, biomedical research has been focused on influenza A, resulting in substantial knowledge gaps concerning the epidemiology and pathogenesis associated with influenza B, especially regarding coinfections. 5,7,41 Our study helps to improve understanding of these gaps. We identified coinfections and chronic lung diseases as the most important risk factors for severe disease complications associated with influenza B. Despite representing one of the largest epidemiologic analyses on disease severity of influenza B in association with coinfections and comorbidities to date, our results require further investigation and confirmation.
F I G U R E 2 Observed and predicted adjusted probabilities (including 95% confidence intervals) for hospitalization and pneumonia diagnosis comparing patients with influenza B (white bars) and patients with influenza B coinfections with other respiratory pathogens (light gray bars; panel (A)) and comparing patients with influenza B and chronic lung diseases (dark gray bars; panel (B)). Adjusted probabilities by coinfection and chronic lung disease status were predicted from the fitted logistic regression models for the two severity outcomes, holding other predictors constant at their mean. Thin bars represent 95% confidence intervals F I G U R E 3 Forest plots for adjusted odds ratios (including 95% confidence intervals and p values) by severity predictor for (A) hospitalization and (B) pneumonia diagnosis. Adjusted odds ratios were estimated using the multivariable logistic regression models for the two severity outcomes. OR, odds ratio; 95% CI, 95% confidence interval; NEWS, National Early Warning Score for acutely ill patients

ETHICS STATEMENT
The parent studies underlying this analysis were reviewed and approved by all participating institutions' Institutional Review Boards (IRB). Johns Hopkins University School of Medicine IRB approved protocols: IRB00135664, IRB00041233, IRB00141101, IRB00052743, and IRB00091667.

PATIENT CONSENT STATEMENT
All participants in the parent studies consented to have their samples made available and used for future research.

SOURCES
Does not apply to this study. The authors declare that this manuscript does not contain any previously published material (including figures/ diagrams, or short extracts, or content taken from websites), and all figures and tables are original.