Cost determinants among adults hospitalized with respiratory syncytial virus in the United States, 2017–2019

Abstract Background Respiratory syncytial virus (RSV) infections are common in adults, but data describing the cost of RSV‐associated hospitalization are lacking due to inconsistency in diagnostic coding and incomplete case ascertainment. We evaluated costs of RSV‐associated hospitalization in adult patients with laboratory‐confirmed, community‐onset RSV. Methods We included adults ≥ 18 years of age admitted to three hospital systems in New York during two RSV seasons who were RSV‐positive by polymerase chain reaction (PCR) and had more than or equal to two acute respiratory infection symptoms or exacerbation of underlying cardiopulmonary disease. We abstracted costs from hospital finance systems or converted hospital charges to cost using cost‐charge ratios. We converted cost into 2020 US dollars and extrapolated to the United States. We used a generalized linear model to determine predictors of hospitalization cost, stratified by admission to intensive care units (ICU). Results Cost data were available for 79% (601/756) of eligible patients. The mean total cost of hospitalization was $8403 (CI95 $7240–$9741). The highest costs were those attributed to ICU services $7885 (CI95 $5877–$10,240), whereas the lowest were radiology $324 (CI95 $275–$376). Other than longer length of stay, predictors of higher cost included having chronic liver disease (odds ratio [OR] 1.38 [CI95 1.05–1.80]) for patients without ICU admission and antibiotic use (OR 1.49 [CI95 1.10–2.03]) for patients with ICU admission. The annual US cost was estimated to be $1.2 (CI95 0.9–1.4) billion. Conclusion The economic burden of RSV hospitalization of adults ≥ 18 years of age in the United States is substantial. RSV vaccine programs may be useful in reducing this economic burden.

In the United States, hospitalization costs associated with RSV infection in adults are estimated to be between one and five billion dollars annually. 1,3 Available hospitalization cost data have been primarily derived by extrapolating costs from administrative databases, or by using influenza and pneumonia cost data to estimate costs for RSV. 1,3,5 In one study utilizing national administrative databases to compare RSV-associated hospitalization costs with influenza-and unspecified viral pneumonia-associated hospitalization costs, the average length of stay (LOS) was longer for RSV hospitalizations, and the mean adjusted cost for RSV hospitalization ($38,828) was more than double than that for influenza ($14,519) or unspecified viral pneumonia ($18,051). 1 Cost data from hospital administrative database studies are not optimal because RSV testing in adults is often incomplete, thereby resulting in underdiagnosis. Datta et al found that discharge diagnoses in adults underestimated RSV-associated hospitalization by as much as 50%. 6 Patient-level cost data from prospective studies in which all suspect RSV cases are tested and ascertainment of RSV is complete could reduce bias and improve generalizability, but there are no published data using this approach. We designed a nested study within a large, prospective, multi-center, multi-season, populationbased, active surveillance study of RSV-associated hospitalization in adults ≥ 18 years of age 4 and collected epidemiologic and cost data, to estimate the cost of RSV-associated hospitalizations and determine factors associated with higher costs. Informed consent to access cost data was required at the Rochester sites for inclusion in the study.
To identify patients with laboratory-confirmed RSV infection, study staff reviewed infection control databases and clinical virology laboratory logs to ascertain the results of PCR tests ordered as standard of care for patients admitted with acute respiratory illness (ARI).
Additionally, during periods of active surveillance, study staff reviewed admission logs and/or emergency department logs to identify patients meeting the screening case definition that included more than or equal to two ARI symptoms, or patients admitted with congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), or asthma preceded by ARI symptoms within the past 14 days. ARI was defined as two or more of the following criteria: presence of fever (≥37.8 C) or feeling feverish, new or worsening cough, new or worsening sputum production, new or worsening dyspnea, or sore throat, runny nose/nasal congestion, and/or body aches.
For those who met the screening case definition, but were not tested for RSV by treating clinicians, the patient or their legally authorized representative was approached to obtain written informed consent for RSV testing in the research laboratory. As a further check for patients with RSV missed by active surveillance, when the surveillance seasons ended, the electronic medical record (EMR) was queried for all positive RSV tests in hospitalized adults during the study period.
Eligible patients were adults ≥ 18 years of age with laboratoryconfirmed RSV who were hospitalized for ≥24 h and met the screening case definition. Patients with healthcare-associated RSV infections, defined as RSV detected ≥3 days after admission, were not included in the study. 4

| Cost calculations
The primary outcome for this study was the cost of RSV-associated hospitalization from admission to discharge or in-hospital death. We extracted costs and/or charges from hospital billing and financial databases. For two of the three hospital systems, costs were directly provided. For the remaining hospital system, costs were estimated using inpatient charges. Charges were multiplied by the hospital's publicly available cost-to-charge ratio (CCR) from the 2018 Hospital Cost Report Data file maintained by the Centers for Medicare & Medicaid Services. 9 We updated all costs to 2020 US dollars using the medical care component of the consumer price index. 10 We determined total costs by combining direct costs of patient care (e.g., nursing, room and board, medications, and supplies) and overhead costs (e.g., administrative expenses to comply with federal and state regulatory requirements and maintain medical records). 11 We did not include patients' productivity loss (e.g., loss of working hours) nor physician or professional fees as they were not captured in hospital discharge cost datasets. 12 We described the costs attributed to the following categories set by the hospitals: Emergency Department (e.g., ED care prior to inpatient admission), nursing (e.g., bed in non-ICU ward), pharmacy (medications administered), laboratory (laboratory tests and processing), therapy (e.g., respiratory interventions such as mechanical ventilation), ICU (bed in ICU), radiology (imaging studies), and other (e.g., endoscopy, dialysis, and echocardiography).

| Cost predictors
Potential predictors of higher costs were abstracted from the medical record and included patients' demographic characteristics (age, sex, and pre-admission living situation); type and number of comorbid conditions (respiratory, cardiac, immunosuppressive, or neurologic conditions, kidney or liver disease, diabetes mellitus, or obesity); Systemic Inflammatory Response Syndrome (SIRS) at presentation (defined as two or more of the following four criteria: temperature > 38.0 C or <36.0 C, tachycardia > 90 beats/minute, tachypnea > 20 breaths/minute, leukocytosis > 12 Â 10 9 /L or leukopenia < 4 Â 10 9 /L 13,14 ); healthcare resource utilization (LOS, ICU admission, ICU LOS, ventilator use, ventilator-days, and antibiotic use); study site; and RSV season.

| Data analyses
Continuous variables (e.g., LOS) were collapsed into multi-level categorical variables. We evaluated the distribution of continuous variables in predicting costs and categorized each increment into the same bin when the association was similar.
We conducted univariate analyses to examine crude associations between predictors and costs. We also evaluated the distribution of LOS and ICU admission by predictors to associate costs with these healthcare resource utilization indicators. We compared mean costs and LOS against a pre-defined reference level (e.g., no comorbidities) using the Wilcoxon signed-rank test. Categorical variables were compared using chi-squared or Fisher's exact test, as appropriate. In addition, we estimated empirical 95% confidence intervals (CI 95 ) for LOS and costs by generating 10,000 bootstrap samples, recalculating means, and using the 2.5 and 97.5 percentiles from this sample. 15 Finally, we described the distribution of cost in median value and interquartile range.
Cost predictors such as ICU admission and ventilator use can identify the same patients. To avoid multi-collinearity in regression analysis, we conducted a cluster analysis that was set to split a cluster 16 until the minimum proportion of variance explained by the cluster component was 75% of the total variance. Among each cluster, we chose the factor with the strongest predictive properties for high costs; for example, ICU admission was chosen from the cluster of ICU admission, ICU LOS, ventilator use, and ventilator-days.
The conceptual framework for our study was to determine a set of predictors (patient attributes and clinical characteristics and outcomes) of higher hospitalization cost. The reason why we included clinical outcomes that occur during hospitalization such as LOS was to allow us to consider the effects of other predictors while controlling for the effects of these clinical outcomes in the cost model. ICU admission was a very strong predictor of cost, yet a small minority of patients was admitted to the ICU. Building two separate prediction models, one for patients admitted to the ICU and one for patients not admitted to the ICU, allowed us to estimate the effect of predictors on the cost that can vary by ICU admission. For each model, we manually removed correlated variables and those with weak statistical significance (p > 0.10) for which clinical relevance is also limited. We used multivariate generalized linear regression model (GLM), with a gamma cost distribution and a LOG link function to determine predictors of the cost of hospitalization. We used exponentiated beta coefficients (eβ) calculated from multivariate GLM to generate the cost ratios by predictors.
To estimate US annual cost, we extrapolated the age groupspecific mean costs by multiplying it by population-based incidence from the prospective surveillance study 4 and the size of the US adult population in 2019. 17 We calculated CI for the extrapolated cost using the delta method. 18 Finally, as a sensitivity analysis, we repeated univariate and multivariate data analyses after limiting the study cohort to only those who were alive at discharge.
All statistical analyses were conducted using SAS version 9.4. This study was approved by the Columbia University Medical Center and University of Rochester institutional review and privacy boards.

| Distribution of healthcare resource utilization and costs
The mean hospital LOS was 8 days (CI 95 [7][8]

| Nationwide RSV hospitalization costs
We used the mean total cost of hospitalization to extrapolate the annual US RSV-associated hospitalization in adults and estimated this cost to be $1.2 (CI 95 $0.9-$1.4) billion (Table 3).

| Sensitivity analysis of costs among those surviving to discharge
We conducted a sensitivity analysis by excluding 27 patients (4%) who died during hospitalization as those who died had significantly higher costs than surviving patients ($13,741, vs. $7579, Table 1). The sensitivity analysis identified the same cost predictors associated with high costs for surviving patients in both the non-ICU and ICU multivariable models (data not shown). Standard errors were estimated by using the delta method for which the correlation measurement was assumed to be 1. to identify RSV cases. In a recent analysis of adults with laboratoryconfirmed RSV infection seen in the emergency department or hospital, only 6% of medical records listed RSV as the primary discharge diagnosis, and only 51% included RSV in any discharge diagnosis. 6 It is possible that if RSV testing was selectively performed on sicker adult patients who are likely to progress to more severe outcomes, as has been noted for infants, 22

| CONCLUSION
This is the first study that evaluated the patient-level cost of laboratory-confirmed RSV-associated hospitalization in adults identi- also thank all staff of the data team including Gregory Hruby PhD,