General Information

Please paste a link to the paper or a citation here:

Link: Xiao Wu, Rachel C. Nethery, Benjamin M. Sabath, Danielle Braun, Francesca Dominici. (2020). Exposure to air pollution and COVID-19 mortality in the United States. medRxiv 2020.04.05.20054502

What is the paper's Manubot-style citation?

Citation: doi:10.1101/2020.04.05.20054502

Is this paper primarily relevant to Background or Pathogenesis?

[x] Background
[ ] Pathogenesis
[ ] Methods

Please list some keywords (3-10) that help identify the relevance of this paper to COVID-19

air pollution
policy
preexisting conditions
acute respiratory distress syndrome

Which areas of expertise are particularly relevant to the paper?

[ ] virology
[x] epidemiology
[ ] biostatistics
[ ] immunology
[ ] pharmacology
[ ] other:

dziakj1 commented 4 years ago

Summary

Suggested questions to answer about each paper:

What did they analyze? They studied county-level deaths for 3,000 counties in the United States through early April 2020, adjusting for covariates regarding the population size, health care system, and demographic and health variables.
What methods did they use? They fit a zero-inflated negative binomial model, adjusting for county-level covariates, to publicly available county-level data. This isn't a formal causal analysis (e.g., no propensity scores) but is intended to provide some indirect evidence for causality. They also made their data and code publicly available here:
https://github.com/wxwx1993/PM_COVID
Does this paper study COVID-19, SARS-CoV-2, or a related disease and/or virus?

Yes, it studies COVID-19.
What is the main finding (or a few main takeaways)?

They found that counties with higher levels of particulate matter, a form of air pollution (PM2.5), had significantly higher numbers of COVID-19 deaths in a zero-inflated negative binomial mixed model after correcting for covariates and a random effect of state. A zero-inflated negative binomial distribution is commonly used in statistical modeling of count outcomes, and is thought to be more realistic in many cases than the classic Poisson distribution for count outcomes. Specifically, the following covariates were considered 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." This was in addition to population size, which was built into the model as an offset. They did notice that the results were no longer significant if the number of hospital beds available was not accounted for. They cite their past research linking general mortality to particulate matter pollution, as a background against which these results were considered reasonable to expect.
What does this paper tell us about the background and/or diagnostics/therapeutics for COVID-19 / SARS-CoV-2?

Although it says little about COVID-19 directly, it is very relevant to where COVID-19 fits into the big picture of public health as a whole, and of social determinants of health and the importance of environmental protection. They report, "we found that an increase of only 1 [microgram per cubic meter] in long-term average PM2.5 is associated with a statistically significant increase of 15% in the COVID-19 death rate.
Do you have any concerns about methodology or the interpretation of these results beyond this analysis?

Obviously, it only pertains to the United States. Also, the analysis is at the county level and does not provide detailed information on how the mechanism works at the individual level. (although I think the authors are suggest it is because chronic preexisting conditions were worsened by the poor air quality, and/or the risk of acute respiratory distress syndrome was higher in such settings). In particular, they don't address whether chronic exposure to pollution makes people more likely to become infected, or more likely to be seriously harmed by the infection, although they seem to believe the latter. Also, they note that their model could be viewed as having two components: the probability of a county having any susceptibility to COVID-19 deaths at all at the current time (i.e., structural zeroes, the zero-inflation part of the model), and modeling the count of deaths in counties where COVID-19 deaths were possible. They say they only report the latter results (presumably being of more lasting interest?). Fortunately, they report that they did several alternative analyses (e.g., they tried leaving out New York) and sensitivity analyses to demonstrate the robustness of their findings. This article is a MedRxiv preprint and not yet peer-reviewed.
Any comments or notes?

Francesca Dominici, the senior author, is a biostatistics professor and Senior Associate Dean for Research at the Chan School of Public Health in Harvard and Co-Director of the Data Science Initiative at Harvard University. She has been active in studying the contribution of air pollution to mortality in the United States. I heard her give a talk at the Joint Statistical Meetings and was very impressed. She sometimes gets involved in advocating policies to reduce air pollution, but I would imagine that her science is quite solid. They also made their data and code publicly available here:
https://github.com/wxwx1993/PM_COVID

agitter commented 4 years ago

Thanks @dziakj1. I edited the original post to change the citation to doi:10.1101/2020.04.05.20054502 so that it can be copy/pasted into the manuscript.