Effects of Air Pollution and Other Environmental Exposures on Estimates of Severe Influenza Illness, Washington, USA

Ranjani Somayaji; Moni B. Neradilek; Adam A. Szpiro; Kathryn H. Lofy; Michael L. Jackson; Christopher H. Goss; Jeffrey S. Duchin; Kathleen M. Neuzil; Justin R. Ortiz


Emerging Infectious Diseases. 2020;26(5):920-929. 

In This Article

Abstract and Introduction


Ecologic models of influenza burden may be confounded by other exposures that share winter seasonality. We evaluated the effects of air pollution and other environmental exposures in ecologic models estimating influenza-associated hospitalizations. We linked hospitalization data, viral surveillance, and environmental data, including temperature, relative humidity, dew point, and fine particulate matter for 3 counties in Washington, USA, for 2001–2012. We used negative binomial regression models to estimate the incidence of influenza-associated respiratory and circulatory (RC) hospitalizations and to assess the effect of adjusting for environmental exposures on RC hospitalization estimates. The modeled overall incidence rate of influenza-associated RC hospitalizations was 31/100,000 person-years. The environmental parameters were statistically associated with RC hospitalizations but did not appreciably affect the event rate estimates. Modeled influenza-associated RC hospitalization rates were similar to published estimates, and inclusion of environmental covariates in the model did not have a clinically important effect on severe influenza estimates.


Seasonal influenza is associated with an estimated 3,300–48,000 annual deaths in the United States[1] and has a major global impact on economies and health.[2–4] Prospective surveillance with specific laboratory testing for influenza is expensive and may underestimate the true burden of influenza if such tests are underused or insensitive or if influenza results in complications or hospitalizations beyond the period in which virus may be detected in patient samples.[5] Therefore, the Centers for Disease Control and Prevention (CDC) and other public health organizations use modeling studies to estimate the incidence of severe influenza illness to inform public health actions.[1,3,6–10] Typically, modeling of the influenza disease burden links aggregate data for outcomes identified in vital statistics or hospitalization administrative databases to influenza virologic surveillance data over time. The difference between estimates with and without influenza covariates is attributed to influenza activity. Such models have been used extensively in the United States,[10–14] in other countries,[15–17] and to produce global estimates of influenza disease burden.[2,3,18–20] The resulting estimates of excess influenza-associated events inform public health actions, such as vaccine or treatment recommendations and patient and healthcare provider communications.

In the United States, influenza and most other respiratory infections are seasonal and follow an approximately sinusoidal curve with winter peaks. Climatic and air pollutant parameters, such as temperature, humidity, and ambient fine particulate matter, vary during the putative influenza season and are associated with acute respiratory infections.[21] Because these other factors share a seasonality similar to influenza, neglecting them may overestimate the effects of influenza on health outcomes. Influenza models that include meteorological data have improved predictive accuracy for viral circulation and peak seasonality.[21–23] National and global models of influenza disease burden do not account for environmental and meteorological parameters, which may be important confounding variables.[1–3,6–8]

Given the importance of influenza disease burden estimates on public health decision making and the reliance on ecologic models for estimation that exclude environmental exposure covariates, we undertook this study to evaluate the effect of including environmental exposures in traditional models on estimates of influenza disease. We hypothesized that environmental exposures would be associated with severe respiratory and circulatory (RC) hospitalizations and that adjustment for these covariates would have a substantial effect on estimates of severe influenza disease incidence.