Relationship between neighborhood census-tract level socioeconomic status and respiratory syncytial virus-associated hospitalizations in U.S. adults, 2015–2017

The EIP is a network of state health departments and their collaborators in local health departments, public health laboratories, and clinical laboratories; infection control professionals; healthcare providers; and academic institutions. This network provides a national resource for infectious disease surveillance and response, conducts applied epidemiologic and laboratory research, identifies measures to prevent and control emerging infectious diseases, and strengthens national public health infrastructure [26]. RSV surveillance began during the 2016–2017 respiratory season (October 1–April 30) with retrospective surveillance for the 2014–2015 and 2015–2016 respiratory seasons in 6 states (California, Georgia, Minnesota, New York, Oregon, and Tennessee). Maryland was added in 2017–18. A case is defined as an adult ≥18 years that is a resident of the catchment area admitted ≤14 days after a laboratory-confirmed RSV test during the respiratory season to a hospital where catchment area residents receive care. Hospitalization is defined as admission to an inpatient ward of the hospital or a stay in an observation unit that is ≥24 h or a combined stay in an emergency department and observation unit for ≥24 h. Patients who are seen only in the Emergency Department or outpatient clinic (even if for more than 24 h) are not cases. Patients who are admitted to and discharged from the hospital on the same day are considered hospitalized (i.e. an overnight stay is not required). A patient was considered to have an RSV infection if RSV was detected by: rapid antigen, molecular assay, viral culture, or fluorescent antibody. Cases were identified using hospital and state public health laboratory databases, hospital infection control practitioner (ICP) databases/logs, reportable condition databases, electronic medical records (EMR), and/or review of hospital discharge records. Trained surveillance officers completed medical record abstraction using a standardized case report form.

For this analysis, cases from 33 counties in California, Georgia, Maryland, Minnesota, New York, and Tennessee were included. Oregon was excluded due to lack of geocoding data. Demographic and clinical variables collected included patient age, sex, race, ethnicity, residential street address; the presence of comorbid conditions; any substance use; and clinical course including ICU admission and in-hospital death. Participants’ addresses were geocoded using ArcGIS software.

The percentage of the population living in poverty and crowding was calculated at the census-tract level using data from the most recent five-year (2013–2017) American Community Survey (ACS). Census tracts were then divided into four levels of poverty by percentage of people living below the poverty level: 0–4.9%, 5–9.9%, 10–19.9%, and ≥ 20%. These categories were used per the recommendations of the Public Health Disparities Geocoding Project [14, 15]. Data on crowding were extracted for the relevant census tracts by the percentage of households with > 1 occupant/room. This was divided into four levels of crowding by percentage of households within a given census-tract with > 1 occupant per room as follows: low = 0–0.9%, medium low = 1–2.9%, medium high = 3–4.9%, high = ≥5%. Levels were selected as per previous studies [18, 19]. The 2010 US census was used to determine the denominator data of population of each census tract.

For calculation of incidence rates, numerators were obtained from the surveillance data by matching each patient’s geocoded address to their corresponding census-tract. Denominator data were taken from the US census 2010. Patients were divided into four age categories: 18 to < 50 years old, 50 to < 65 years old, 65 to < 80 years old, and ≥ 80 years old and age-adjusted incidence rates were calculated using these age groups and the US census 2010 data as the standard population. A Cuzick test for trend across ordered groups was conducted to compare incidence rates. Incidence rate ratios were calculated comparing each level of poverty/crowding to the lowest level of the respective category.

Log binomial regression models were used to examine the association between census-tract level poverty and crowding with severe RSV-associated illness, defined as either in-hospital death or ICU admission. First, separate unadjusted models were fit with independent variables considered to be potentially associated with severe RSV disease: age, sex, race, ethnicity, obesity as defined as body mass index (BMI) ≥30 kg/m², smoking (current and/or former), presence of an immunocompromised condition (i.e. HIV, immunosuppressive therapy use, primary immune deficiency, graft versus host disease, solid organ transplant, bone marrow transplant, and cancer diagnosis), chronic lung disease (i.e. chronic bronchitis, chronic obstructive pulmonary disease/emphysema, asthma/reactive airway disease and cystic fibrosis), cardiovascular disease (i.e. coronary artery disease, congestive heart failure, cardiomyopathy, history of coronary artery bypass grafting, cerebral vascular accident/stroke/transient ischemic attack, atherosclerosis), and surveillance site. The medical conditions included have previously been identified as possibly being associated with an increased risk of RSV infection or severity of RSV, or other respiratory viral infection [4‐9]. Variables that had a statistical association with death or ICU admission at a p-value < 0.2 (i.e., age, obesity, and study site) were included along with either census-tract level poverty or crowding in the multivariate model. In the final multivariate model, variables that had an association with death or ICU admission at a p-value of < 0.05 were considered statistically significant. Analyses were conducted using STATA, version 15.1.