key: cord-0887351-2q6qmex3
authors: Wu, Xiao; Nethery, Rachel C.; Sabath, Benjamin M.; Braun, Danielle; Dominici, Francesca
title: Exposure to air pollution and COVID-19 mortality in the United States: A nationwide cross-sectional study
date: 2020-04-07
journal: medRxiv : the preprint server for health sciences
DOI: 10.1101/2020.04.05.20054502
sha: 9aa462d97c0fda4cd26f4804a3c5a9825870eb26
doc_id: 887351
cord_uid: 2q6qmex3

Objectives: United States government scientists estimate that COVID-19 may kill tens of thousands of Americans. Many of the pre-existing conditions that increase the risk of death in those with COVID-19 are the same diseases that are affected by long-term exposure to air pollution. We investigated whether long-term average exposure to fine particulate matter (PM2.5) is associated with an increased risk of COVID-19 death in the United States. Design: A nationwide, cross-sectional study using county-level data. Data sources: COVID-19 death counts were collected for more than 3,000 counties in the United States (representing 98% of the population) up to April 22, 2020 from Johns Hopkins University, Center for Systems Science and Engineering Coronavirus Resource Center. Main outcome measures: We fit negative binomial mixed models using county-level COVID-19 deaths as the outcome and county-level long-term average of PM2.5 as the exposure. In the main analysis, we adjusted by 20 potential confounding factors including population size, age distribution, population density, time since the beginning of the outbreak, time since state issuance of the stay-at-home order, hospital beds, number of individuals tested, weather, and socioeconomic and behavioral variables such as obesity and smoking. We included a random intercept by state to account for potential correlation in counties within the same state. We conducted more than 68 additional sensitivity analyses. Results: We found that an increase of only 1 μg/m3 in PM2.5 is associated with an 8% increase in the COVID-19 death rate (95% confidence interval [CI]: 2%, 15%). The results were statistically significant and robust to secondary and sensitivity analyses. Conclusions: A small increase in long-term exposure to PM2.5 leads to a large increase in the COVID-19 death rate. Despite the inherent limitations of the ecological study design, our results underscore the importance of continuing to enforce existing air pollution regulations to protect human health both during and after the COVID-19 crisis. The data and code are publicly available so our analyses can be updated routinely.

The scale of the COVID-19 public health emergency is an unmatched one in our lifetime. It will have grave social and economic consequences. The suddenness and global scope of this pandemic has raised urgent questions that require coordinated and credentialed information to slow its devastation. A critically important public health objective is to identify key modifiable environmental factors, such as ambient air pollution, that could increase the severity of the health outcomes (e.g., ICU hospitalization and death) among individuals with COVID-19.

Our understanding of what causes death in COVID-19 patients is evolving. Early data from China suggests that a majority of COVID-19 deaths occurred in adults aged respiratory distress syndrome (ARDS) which has a mortality rate ranging from 27% to 45%. 4 within a given county. We compute average 2016 PM 2.5 exposure analogously for each county to use in sensitivity analyses.

Potential Confounders: We consider the following sixteen county level variables and one state level variable as potential confounders: population density, percent of the population ≥ 65, percent living in poverty, median household income, percent black, percent Hispanic, percent of the adult population with less than a high school education, median house value, percent of owner-occupied housing, population mean BMI (an indicator of obesity), percent ever-smokers, number of hospital beds, and average daily temperature and relative humidity for summer (June-September) and winter (December-February) for each county, and state level number of COVID-19 tests performed. Additional detail on the creation of all variables used in the analysis is available in the Supplementary Materials.

We fit zero-inflated negative binomial mixed models (ZNB). 22, 23, 24 using COVID-19 deaths as the outcome and PM 2.5 as the exposure of interest. The ZNB is composed of two sub-models. The first is a count sub-model that estimates the association between COVID-19 deaths and PM 2.5 (adjusted by covariates) among counties eligible (e.g., confirmed COVID-19 cases) to experience a COVID-19 death. The second is a zero sub-model that accounts for the excess of zeros that may be generated by counties not yet eligible for COVID-19 deaths (e.g., due to the absence of confirmed COVID-19 cases) and unlikely to have COVID-19 deaths as of April 4, 2020. Additional modeling details are provided in the Supplementary Materials. We include a population size offset and we adjust for all variables listed above. We also include a random intercept by state to account for potential correlation in counties within the same state, due to similar socio-cultural, behavioral, and healthcare system features and similar COVID-19 response and testing policies. We only report the result from the count sub-model. More All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. . https://doi.org/10.1101/2020.04.05.20054502 doi: medRxiv preprint specifically we report exponentiated parameter estimates, the mortality rate ratios (MRR) and 95% CI from the count sub-model. The MRR can be interpreted as the relative increase in the COVID-19 death rate associated with a 1 ߤ g/m 3 increase in long-term average PM 2.5 exposure among the counties eligible to experience a COVID-19 death. We do not report results from the zero sub-model. We carried out all analyses in R statistical software and performed model fitting using the NBZIMM package. 25

We conduct six secondary analyses to assess the robustness of our results to the confounder set used, potential unmeasured confounders, and outliers.

• First, because New York state has experienced the most severe COVID-19 outbreak in the United States to date and has five times higher COVID-19 deaths than the next highest state, we anticipate that it will strongly influence our analysis. As a result, we repeat the analysis excluding all counties in New York state.

• Second, existing COVID-19 testing and case count data are unable to accurately capture the size of an outbreak in a given county, and the inability to fully adjust for this factor could induce confounding in our analyses (e.g., if counties with high PM 2.5 exposure also tend to have large outbreaks relative to the population size, then their death rates per unit population could appear differentially elevated, inducing a spurious correlation with PM 2.5 ). To explore how this may impact our results, we conduct analyses excluding counties with less than 10 confirmed COVID-19 cases.

• Finally, we fit models omitting the following potential confounders from the model 

We conduct several sensitivity analyses to assess the robustness of our results to data and modeling choices.

• First, we repeat all the analyses using alternative methods to estimate exposure to PM 2.5 . 26 • Second, because our study relies on observational data, our results could be sensitive to modeling choices (e.g., distributional assumptions or assumptions of linearity). We evaluate sensitivity to such choices by conducting analyses: ○ Treating PM 2.5 as a categorical variable (categorized at empirical quintiles) ○ Adjusting for population density as a categorical variable (categorized at empirical quintiles) ○ Using a negative binomial model without accounting for zero-inflation.

Results for the sensitivity analyses are shown in Supplementary Materials.

Our study utilized data from 3,080 counties, of which 2,395 (77.8%) have reported zero COVID-19 deaths at the time of this analysis. Table 2 describes the data used in our analyses. All COVID-19 death counts are cumulative counts up to April 4, 2020. These spatial patterns in COVID-19 death rates generally mimic patterns in both high population density and high PM 2.5 exposure areas. In the Supplementary Materials, we provide additional visualizations and data diagnostics that justify the use of the ZNB model for our analyses. After All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. In secondary analyses shown in 

This is the first nationwide study in the United States that estimates the relationship between long-term exposure to PM 2.5 and COVID-19 death rates. We found statistically significant evidence that an increase of 1 ߤ g/m 3 in long-term PM 2.5 exposure is associated with a 15% increase in the COVID-19 mortality rate. Our results are adjusted for a large set of All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. . https://doi.org/10.1101/2020.04.05.20054502 doi: medRxiv preprint socioeconomic, demographic, weather, behavioral, and healthcare-related confounders and demonstrated robustness across a wide range of sensitivity analyses.

In our previous study 27 of 60 million Americans older than 65 years, we found that a of 1 mg/m3 in long-term PM 2.5 exposure is associated with a 0.73% increase in the rate of all-cause mortality. Therefore, a small increase in long-term exposure to PM 2.5 leads to a large increase in COVID-19 death rate of a magnitude that is 20 times the one estimated for all-cause mortality.

This analysis provides a timely characterization of the relationship between historical exposure to air pollution COVID-19 deaths in the United States. Research on how modifiable factors may exacerbate COVID-19 symptoms and increase mortality risk is essential to guide policies and behaviors to minimize fatality related to the outbreak. Our analysis relies on up-to-date population-level COVID-19 data and well-validated air pollution exposure measures. 27 Strengths of this analysis include adjusting for a wide range of potential confounders and a demonstrated robustness of results to different model choices. Moreover, the analyses rely exclusively on data and code that are publicly available. This provides a platform for the scientific community to continue updating and expanding these analyses as the pandemic evolves, and data accumulate.

The inability to accurately quantify the number of COVID-19 cases due to limited testing capacity presents a potential limitation in that we are unable to fully adjust for the size of the outbreak in a given county, which could be a source of unmeasured confounding. We instead adjust by total population size. However, we anticipate that by using a zero-inflated model we provide partial adjustment for this, as we estimate the MRR accounting for the fact that some counties were not eligible to experience a COVID-19 death because they still had zero All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. . https://doi.org/10.1101/2020.04.05.20054502 doi: medRxiv preprint confirmed cases. Moreover, the constantly evolving COVID-19 response measures creates a fluid scenario with regards to the availability of medical resources in a given area. The availability of these resources is likely to impact fatality rates. Because real-time, nationwide data on such measures and resources are not available, we are unable to adjust for this feature.

As a result, this could represent another source of confounding.

The results of this paper suggest that long-term exposure to air pollution increases vulnerability to experiencing the most severe COVID-19 outcomes. These findings align with the known relationship between PM 2.5 exposure and many of the cardiovascular and respiratory comorbidities that dramatically increase the risk of death in COVID-19 patients. They are also consistent with findings that air pollution exposure dramatically increased the risk of death during the Severe Acute Respiratory Syndrome (SARS) outbreak in 2003, which is caused by another type of coronavirus. 28 This study provides a motive for expanded follow-up investigations as more and higher quality COVID-19 data become available. These would include validating these results in other data sources and study types and studies of biological mechanisms, impacts of PM 2.5 exposure timing, and relationships between PM 2.5 and other COVID-19 outcomes such as hospitalization.

The results of this study also underscore the importance of continuing to enforce existing air pollution regulations during the COVID-19 crisis. Based on our result, we anticipate a failure to do so can potentially increase the COVID-19 death toll and hospitalizations, further burdening our healthcare system and drawing resources away from COVID-19 patients.

All rights reserved. No reuse allowed without permission.

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint (which this version posted April 7, 2020. Figure 2 : Mortality Risk Ratios (MRR) and 95% confidence intervals. The MRR can be interpreted as percentage increase in the COVID-19 death rate associated with a 1 ߤ g/m 3 increase in long term average PM 2.5. The MRR from the main analysis is adjusted by seventeen potential confounders (population density, percent of the population >65, percent living in poverty, median household income, percent black, percent Hispanic, percent of the adult population with less than a high school education, median house value, percent of owneroccupied housing, population mean BMI, percent ever-smokers, summer/winter temperature/relative humidity, and number of hospital beds for each county, and number of COVID-19 tests performed for each state). In secondary analyses, we (a) omit number of hospital beds; (b) omit number of tested cases in each state; (c) omit smoking rate and BMI from BRFSS; (d) omit summer/winter temperature/relative humidity; (e) exclude data from New York state; (f) exclude data from counties with less than 10 confirmed cases. All COVID-19 death counts are cumulative counts up to and including April 04, 2020.

for the China Medical Treatment Expert Group for Covid-19

Estimating clinical severity of COVID-19 from the transmission dynamics in Wuhan, China

No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

COVID-19) -United States

Report of the WHO-China Joint Mission on Coronavirus Disease

Acute Respiratory Distress Syndrome (ARDS)

Preliminary Estimates of the Prevalence of Selected Underlying Health Conditions Among Patients with Coronavirus Disease 2019 -United States

Air pollution and cardiovascular disease: a statement for healthcare professionals from the Expert Panel on Population and Prevention Science of the

Air pollution and mortality in the Medicare population

Chronic fine and coarse particulate exposure, mortality, and coronary heart disease in the Nurses' Health Study

Fine particulate air pollution and hospital admission for cardiovascular and respiratory diseases

Ambient air pollution and the risk of acute ischemic stroke

Ambient Air Pollution and Pregnancy Outcomes: A

Impact of Long-Term Exposures to Ambient PM 2.5 and Ozone on ARDS Risk for Older Adults in the United States CHEST

Integrated Science Assessment (ISA) for Particulate Matter

A national case-crossover analysis of the short-term effect of PM2.5 on hospitalizations and mortality in subjects with diabetes and neurological disorders

Long-term PM2.5 exposure and neurological hospital admissions in the Northeastern United States

Air Pollution and Mortality in the Medicare Population

Regional estimates of chemical com-position of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors. Environmental science & technology

Regional estimates of chemical com-position of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors. Environmental science & technology

Modern Applied Statistics with S

Negative Binomial Mixed Models for Analyzing Microbiome Count Data

Negative binomial mixed models for analyzing microbiome count data

An ensemble-based model of PM2.5 concentration across the contiguous United States with high spatiotemporal resolution

No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint