We obtained patient-level hospital discharge data for 2005-2018 in California and New York, and 2010-2018 in Georgia, from hospital associations and state health departments at the daily level. ED visits included both outpatient visits with discharges directly from the ED and inpatient visits with admissions via the ED, excluding scheduled admissions. Records included information on admission date, age in years, patient’s residential ZIP code, and ICD (International Classification of Diseases) diagnosis codes. ICD-9 (International Classification of Diseases, Ninth Revision) codes were used before October 1, 2015, after which ICD-10 (International Statistical Classification of Diseases, Tenth Revision) codes were adopted. Both primary and secondary ICD diagnosis codes were used to identify cause-specific ED visits for overall respiratory diseases (RESP, ICD-9 460-519; ICD-10 J00-J99) and influenza (FLU, ICD-9 487-488; ICD-10: J09-J11). This approach captures encounters in which respiratory conditions contributed to the visit but were not coded as the principal diagnosis. For instance, influenza can exacerbate chronic or systemic symptoms. Restricting analyses to primary diagnoses alone could underestimate the true respiratory disease burden, particularly among patients with comorbidities, as shown in previous hospital-based surveillance studies [20-22].

For RESP, FLU, and non-FLU RESP, we created a time series of daily ED visit counts for each county by aggregating ED visits based on the patient’s residential ZIP code that overlapped with these counties. For ZIP codes that overlapped with multiple counties, we assigned ED visits to the county with which the ZIP code had the largest population overlap using the 2010 ZIP Code Tabulation Area (ZCTA) County Relationship File from the US Census Bureau. In cases where population overlap could not be determined (less than 1% of ZIP codes), we assigned the county based on the shortest centroid-to-county distance. There was a total of 279 counties (California n=58, Georgia n=159, and New York n=62).

The daily influenza activity proxy was constructed from the same hospital discharge dataset of ICD-coded influenza ED visits (ICD-9 487–488; ICD-10 J09–J11). We calculated the daily influenza-related ED visit rates per 100,000 population for each county during the influenza season (the Centers for Disease Control and Prevention defines an influenza season typically runs from Morbidity and Mortality Weekly Report (MMWR) week 40 to 18 of the following year [2]), using county-level population data from the US Census Bureau. This variable reflects community-level influenza activity used as the exposure term (x_t) in the quasi-Poisson model described in the Statistical Analysis section. To prevent circularity, influenza-coded visits were removed from the respiratory-disease outcome series when estimating associations. The 2009-2010 influenza A (H₁N₁) pandemic season was excluded because it exhibited 2 distinct epidemic peaks and atypical temporal patterns that differed markedly from regular influenza seasons. Including this year could bias the estimation of county-level associations that assume relatively consistent seasonal dynamics. Our analysis thus had 11 influenza seasons analyzed from 2005 to 2018.

Daily ambient air pollution data for 24-hour average PM_2.5, 1-hour maximum NO₂, and 8-hour maximum ozone concentrations from 2005 to 2018 were based on a data product with complete spatial-temporal coverage across the United States at a 12 km × 12 km spatial resolution [23]. These estimates used monitoring data to bias-correct Community Multiscale Air Quality (CMAQ) model simulations that integrate emissions, meteorological conditions, and atmospheric chemistry to improve air quality assessments.

Daily maximum and minimum temperature in degrees Celsius, and water vapor pressure (as a proxy for humidity) in pascals at a 1 km × 1 km spatial resolution were obtained from Daymet [24]. The daily average temperature was calculated as the mean of the daily maximum and minimum.

Air pollution and meteorology data were first aggregated to the ZIP code level, and then up-scaled to the county-level to match the health outcome with population-weighted averaging ZIP code-level exposures. Population size data were obtained from the US Census at the ZCTA level for 2010 to 2020, and we estimated annual populations via linear interpolation of decadal Census data. A 3-day moving average was applied to air pollution and meteorological variables to account for short-term cumulative exposure effects, as the health impacts of environmental exposures are typically not immediate and may manifest over several subsequent days. This approach is consistent with previous studies [18,25,26] on air pollution and meteorological influences on respiratory ED visits that have adopted similar averaging strategies.

We considered 3 categories of county-level covariates: SES, environmental exposures, and chronic health condition prevalence. SES indicators such as the percent of population living under the poverty line and the percent of uninsured individuals were obtained as 5-year averages (2011-2016) from the American Community Survey. Long-term county-level environmental exposures were obtained by averaging levels of PM_2.5, NO₂, O₃, humidity, and temperature across all influenza seasons (2005-2018). Chronic health conditions were obtained from the 2022 Behavioral Risk Factor Surveillance System, including the prevalence rates of individuals with disability, and other major chronic diseases such as stroke, diabetes, and high blood pressure. Table S1 in Multimedia Appendix 1 summarizes the data sources, variable descriptions, and time units for all covariates.

This study used secondary, deidentified hospital discharge data obtained under active data use agreements with state health departments and hospital associations. The dataset spans multiple US states and years (2005-2018) and includes both outpatient and inpatient discharge records. All data were provided under institutional agreements for environmental health research and stored on secure university servers. The study protocol was reviewed and approved by the Emory University Institutional Review Board (STUDY00006348, 2024P007407). Informed consent was not required because the data were fully deidentified before access and analysis. Only ZIP code–level admission and diagnostic information were available, and all analyses were performed using aggregated county-level data. All datasets were anonymized before analysis, and no attempt was made to reidentify individuals. No participant compensation was applicable for this study.

We analyzed influenza-attributable respiratory ED visits from 2005-2006 to 2017-2018 across 3 US states and identified substantial geographic heterogeneity associated with socioeconomic, environmental, and chronic disease factors. By leveraging statewide hospital discharge data spanning more than a decade, our study provided a comprehensive assessment of influenza-related respiratory disease burden. The meta-regression framework within the Bayesian spatial modeling approach enabled the quantification of geographic variability and offered a deeper understanding of county-level influenza-attributable respiratory visit burdens.

Substantial regional heterogeneity in ED burden was observed, with Georgia exhibiting rates over twice those of California and New York. This may reflect differences in SES, health care access, population health, and health-seeking behaviors. Georgia also had the highest proportion of all respiratory ED visits coded as influenza (306,234/7,481,121, 4.1%) and higher baseline rates of poverty and chronic health conditions.

Meta-regression analysis revealed that county-level higher poverty rates and lower insurance rates were associated with higher rates of respiratory ED visits attributable to influenza. This aligns with previous research indicating that social determinants of health play a crucial role in chronic respiratory disease disparities [30], with lower SES being associated with severe respiratory conditions and more ED visits.

For environmental exposures, higher levels of PM_2.5, humidity, and temperature were associated with increased influenza-attributable respiratory ED visit rates in univariate models. Among these factors, humidity exhibited the strongest association, followed by temperature. However, these findings contrast with previous research [31,32], suggesting peak influenza transmission occurs during colder, drier conditions. One explanation is that ED visits capture both new infections and exacerbations of existing conditions, whereas previous studies have primarily focused on influenza virus transmission alone. Moreover, elevated humidity, particularly during the influenza season, which largely overlaps with the winter period in our study, may worsen respiratory illness through mechanisms such as indoor dampness and mold [33]. Even modest increases in wintertime temperature may lead to more outdoor activity or allergen exposure, triggering respiratory events [34]. Additionally, previous studies have generally examined temperature and humidity across the full annual range, whereas our analysis focused on modest rises within the influenza season. This distinction may explain why our findings differ: even though influenza transmission is typically higher in winter compared with warmer periods of the year, higher temperature and humidity within the influenza season appear to have positive associations with influenza-attributable respiratory outcomes. The secondary sensitivity analysis’s results (see Table S7 in Multimedia Appendix 1) support these observed positive associations by showing that the positive associations between temperature, humidity, and respiratory ED visits remained consistent across stratified levels. Both higher temperature and humidity within the influenza season and extremely cold conditions were associated with increased ED visit rates.

A small but positive association between PM_2.5 levels and influenza-attributable respiratory ED visit rates was observed in the univariate models. This finding is consistent with previous research [26] showing a positive link between ambient PM_2.5 and respiratory diseases and respiratory-related ED visits. After adjusting for nonlinear population effects, we observed a statistically significant positive association between ozone and influenza-attributable respiratory ED visit rates, suggesting that ozone may contribute to respiratory burden when factors associated with population size are accounted for. This finding is supported by previous epidemiological studies that link ambient ozone exposures to airway inflammation and increased hospital admissions for asthma and other respiratory conditions [35]. However, the association between PM_2.5 and respiratory ED visit rates became significantly negative after adjusting for both the nonlinear population size and 2 SES factors. We did not find other studies reporting a protective effect between PM_2.5 and health outcomes. This unexpected result may be due to the positive correlations between population size and PM_2.5, and the negative association of PM_2.5 may be due to model misspecification of the population association. Further investigation is needed to elucidate these associations.

Our findings demonstrated strong associations between the prevalence of county-level chronic health conditions and influenza-attributable respiratory ED visit rates. Among these conditions, stroke exhibited the strongest estimated association across all models. This may be attributed to the long-term effects of cerebrovascular disease, including compromised immune function [36], increased susceptibility to infection, and difficulties in clearing respiratory pathogens due to weakened muscular function [37]. Previous studies have shown that stroke survivors are at an elevated risk of pneumonia and other respiratory infections [38].

Diabetes and COPD rates also showed the strongest associations. These findings align with existing evidence that diabetes and COPD significantly contribute to respiratory health complications. Diabetes impairs immune response, increasing susceptibility to infections, including influenza and pneumonia [39], while COPD is characterized by chronic lung inflammation and reduced pulmonary function [40], making individuals more prone to severe respiratory infections and exacerbations requiring emergency care.

Adjusting for poverty and uninsured rates attenuated most associations, suggesting that part of the association between county-level chronic disease and influenza-attributable respiratory ED visits may reflect underlying socioeconomic disadvantage. The effects of poverty and uninsured prevalence remained similar and increased in some 2-SES adjusted models. If chronic conditions are a mediating factor between SES and influenza-attributable respiratory ED visits, we would expect SES effects to decrease when controlling for health status. The observed stronger association points to an independent and possibly more direct role of SES in shaping emergency care use. Although individuals with lower SES often face barriers to seeking health care, they also experience higher disease burden and tend to rely on EDs as their primary point of care due to limited access to preventive services and insurance coverage [41,42]. This reliance on EDs as safety-net facilities may explain why poverty and uninsured prevalence remain positively associated with respiratory ED visit rates, even after accounting for chronic disease burden [43]. The persistent SES effect likely reflects both greater underlying need and structural inequities in healthcare access and continuity.

Our findings have important implications for public health practice and intervention. County-level identification of areas with elevated influenza-attributable respiratory ED visit rates can help local health departments prioritize limited resources, such as staffing, testing capacity, and antiviral distribution, during peak influenza seasons. In addition, integrating spatial patterns of socioeconomic disadvantage and chronic-disease burden can inform more equitable vaccination outreach and community-based prevention programs. For example, counties with persistently high respiratory ED burden and high poverty or uninsured rates may benefit from tailored vaccination drives, mobile clinics, or public health messaging to enhance early care-seeking behaviors. These results demonstrate how spatially resolved influenza burden estimates can support targeted mitigation and preparedness strategies at the community level.

This study has several limitations. First, our reliance on ICD-coded influenza ED visit data as a proxy for influenza burden may still be subject to misclassification bias, given that coding practices vary across hospitals and over time. However, we conducted a validation analysis by comparing ICD-coded influenza hospitalizations from our datasets to laboratory-confirmed influenza cases reported by the US Influenza Hospitalization Surveillance Network (FluSurv-NET) surveillance system [44]. In counties with overlapping data, we aligned weekly counts from both sources using the MMWR-week framework. Pearson correlation analysis (see Figure S3 in Multimedia Appendix 1) demonstrated good agreement between ICD-coded hospitalizations and FluSurv-NET data.

Second, our analysis areas may not fully generalize to other regions with different health care access, demographics, or environmental exposures. Third, although we considered key environmental and socioeconomic variables, important factors such as vaccination status, healthcare-seeking behavior, and medical history were unavailable at the county level.

This study provides a county-level assessment of influenza-attributable respiratory ED visits across California, Georgia, and New York. Our findings highlight the importance of SES, environmental exposures, and chronic health conditions as key spatial predictors of local influenza-related respiratory ED visit rates. Counties with higher poverty and uninsured rates, greater humidity and temperature, and elevated prevalence of chronic conditions such as stroke, COPD, and others, were strongly linked to higher influenza-attributable respiratory ED visits. These findings highlight the need for targeted interventions, such as improving health care access, mitigating environmental risks, and managing chronic diseases, to reduce influenza severity and demonstrate the value of incorporating spatial data into public health planning.