How the Harvard Covid-19 Study Became the Center of a Partisan Uproar
The shock of Covid-19 hit in different ways for Xiao Wu, a 27-year-old Ph.D. student in Harvard’s public health school, and his academic adviser, biostatistician Francesca Dominici.
Wu, a former national high school math champion in China, was shaken in January as he absorbed reports from his native country about Covid-19’s exponential spread. “Considering that infection rate, it was almost guaranteed this would be a global pandemic,” he said. “It was transmitted too fast.”
But reality didn’t sink in for Dominici until she walked across Harvard Yard on March 10, the day the university sent students home. “These kids were crying,” she recalled. “They were sitting on the curb, literally, with their suitcases. I think that’s when I realized—gosh, this is bad. This is huge.”
When Dominici went home, she saw her work—quantifying the health effects of pollution—in the stark new light of a respiratory illness that was likely to be the biggest public health crisis of her lifetime.
She said she could only think about one thing: “What can I do about it?” “How can I help?”
Dominici soon emailed Wu and other members of their research group at Harvard T.H. Chan School of Public Health with an idea. Could they build upon their previous work—documenting links between fine particle air pollution, known as PM 2.5, and premature deaths—and examine whether there was a relationship between air pollution and Covid-19?
The study that emerged less than a month later, with Wu as lead author and Dominici as senior author, has become both a spark for advocacy and a lightning rod for critics. Although the paper, posted online on the health sciences platform MedRxiv on April 7, had not yet been peer reviewed, it made waves immediately.
The researchers found that Covid-19 patients who lived in polluted areas were more likely to die than those in less polluted areas, with a small increase in PM 2.5 associated with a large increase in the risk of premature death. If the results held up, it would be striking evidence of the need for a more expansive public policy to control Covid-19 that included environmental, as well as personal, protection.
But the paper instantly touched off a political scrum, in large part because the Harvard team’s conclusion was directly at odds with the Trump administration’s drive to aggressively roll back environmental regulation. Democrats and environmental advocates cited the research in their battle against the Trump agenda, while Andrew Wheeler, the administrator of Trump’s Environmental Protection Agency (EPA), Wheeler’s science advisers, Republican members of Congress and anti-regulation lobbyists all blasted the study.
When the Harvard researchers revised their findings in an update three weeks after the original paper was released, and came up with a slightly smaller, but still significant, association between PM 2.5 and Covid-19 mortality, their critics pounced on the change, characterizing it, erroneously, as tantamount to a retraction.
Now, even as the results of other studies emerge, from different teams in different countries, that support a relationship between air pollution and Covid-19 deaths, there already are signs that the response to the science is dividing predictably along partisan lines. In one sense, the Harvard paper is a case study in how quickly science becomes politicized in the Trump era. But because the study, like so much Covid-19 research, was released online as a “preprint” before it underwent peer review, it also has become an object lesson in the peril and promise of the rapid-fire dissemination of research in the midst of a public health crisis. Even some advocates of stronger air pollution regulation have criticized the Harvard team for releasing the paper prior to its formal review by other scientists.
Dominici maintains the controversy was inescapable, given that the researchers were tackling not only the charged topic of Covid-19, but PM 2.5—the fine soot pollution, with particles less than 2.5 microns in diameter, caused by combustion. The study brought attention to the risks of burning fossil fuels—a primary contributor to climate change—and to the disproportionate health burden on minority communities, where PM 2.5 exposure is greatest and which are seeing the highest Covid-19 death rates.
The study landed “in the middle of the controversy over Covid, climate change, and environmental justice and racism,” said Dominici. “In a way, I couldn’t have had a better strike, dealing with all three at the same time.”
A Controversial Review
As the spread of the new coronavirus upended life around the world in the first half of 2020, another public health drama was unfolding behind the scenes in Washington, D.C. The Trump administration was racing to complete a series of air pollution policy decisions, having made clear its goal of finalizing them by the end of the president’s first term.
On the list was a legally required review of the adequacy of the national air quality standards for PM 2.5, which three decades of science had linked to premature death due to heart and lung disease. There was alarming new science documenting PM 2.5’s deadly health effects, even at levels that met the current U.S. standards.
Dominici was the principal investigator on one of the new PM 2.5 studies.
In one of the largest epidemiological studies ever conducted, she and her team at Harvard had analyzed the health records of more than 60 million Medicare recipients. The study, published in 2017, indicated that strengthening the nation’s air pollution standard so that the maximum allowed concentration was 11 micrograms per cubic meter instead of 12, would save 12,000 American lives annually. The research also found that the risk of premature PM 2.5-related death for Blacks was three times greater than the national average.
In January—the same week the World Health Organization (WHO) issued its first situation report on a new coronavirus that was causing respiratory illness in Wuhan, China—career staff scientists at the EPA recommended that the Trump administration adopt a stronger PM 2.5 standard.
The process of reviewing the standard was fraught with controversy, with the Trump EPA’s hand-picked outside science advisers casting doubt on the agency’s long-standing conclusions about PM 2.5’s health harms. The Trump administration’s new chairman of EPA’s Clean Air Science Advisory Committee, statistician Louis Anthony Cox, a Denver consultant for the American Petroleum Institute and other industry groups, urged that the agency require a much higher burden of proof when judging the risks of pollutants.
Dominici was prominent among the public health experts criticizing Cox’s proposed approach. “We risk the health of thousands breathing unhealthy air,” she said last year in a commentary in the journal Science she wrote with Gretchen Goldman, research director for the Union of Concerned Scientists’ Center for Science and Democracy.
The Trump EPA did not overhaul its scientific assessment process, as Cox had proposed. But by early March, Wheeler had made clear his plan was to reject the call for a stronger PM 2.5 standard.
Dominici and her colleagues at Harvard were following developments at the EPA closely. They planned to weigh in when the public comment period on this and other proposals weakening environmental regulations formally opened. But before that could happen, the World Health Organization declared Covid-19 a pandemic, Trump announced a national state of emergency, and the scientists—like their fellow citizens—retreated to their homes and watched the crisis unfold.
Would Pollution Make Covid-19 Worse?
Dominici immediately began thinking about air pollution’s interplay with Covid-19. She and her colleagues had published research in 2019 showing that long-term exposure to PM 2.5 and ozone were associated with an increased risk of hospitalization for acute respiratory distress syndrome (ARDS) in older U.S. adults, and ARDS was killing Covid-19 victims.
“We have an enormous amount of knowledge and data regarding the adverse health effects of fine particulate matter,” Dominici said. “We know that it penetrates very deep into the lungs and causes serious inflammation. At the same time, Covid is attacking the lungs.”
She added, “From a biological standpoint, from an epidemiological standpoint, and just from a common sense standpoint, where you have two agents attacking the lungs at the same time, to me it was very natural to pose the question of whether exposure to air pollution would make Covid worse.”
Harvard’s public health school researchers know well the challenges of air pollution studies. Scientists can’t control exposure in such research, but instead have to study real-world conditions, taking into account as many confounding factors as possible. Over two decades, the Harvard team had built a research platform that allowed them to control for other possible causes of illness and premature death, like population density, age distribution, smoking and obesity. “We have, I would say, really good scientific tools to do this kind of study,” said Wu.
To do a study on air pollution’s effects, though, the Harvard researchers would need to know the locations of Covid-19’s victims.
That data first became available on March 23, when the Johns Hopkins University’s coronavirus resource center, which had been tracking the spread of the disease globally, began breaking down information on Covid-19 cases and fatalities in the United States by county.
County-level data is often used in epidemiological studies. But the Harvard researchers knew it was not ideal. In the United States, the size of counties vary widely. Some are quite large and include both lightly polluted rural areas and heavily polluted cities. Instead of being able to pinpoint pollution hot spots in those counties, the researchers only could consider average pollution county-wide.
Also, without the medical records of the individual research subjects in their study, the Harvard researchers would only be able to draw limited conclusions. The study would be much stronger if they knew the ages and races of the each of the Covid-19 patients, which of them had diabetes or other underlying conditions, which ones smoked, and which were obese—all potentially confounding factors.
“If you don’t have the individual-level data, clearly the research is less accurate,” Dominici said. “Unfortunately, at a time of a pandemic, we live in a less-than-ideal world.”
At some point, researchers and health institutions may be able to work out health privacy agreements that allow access to enough medical data about individual patients with Covid-19 to more precisely study the impact of air pollution on the pandemic. (The Trump administration meanwhile is working on rules that would instead limit the EPA’s use of such individual-level studies—a proposal vigorously opposed by the scientific community, including the public health science team at Harvard.)
“I think eventually we might be able to get there, but it’s not something that’s going to be available in a month or two, or even a year,” Dominici said, referring to the data on individual patients. “So I think in the meantime, I think we need to do the best that we can with the data that we have and act responsibly.
“That’s how science progresses,” Dominici said. “You have to put out there the best you can do at the time, and be honest about the limitations of your study. Then you can find out what further data you need, and you can get it and do better. But we’ve got to start somewhere.”
A Hard Decision
The previous work on air pollution by Dominici and her colleagues gave her team an enormous head start on the Covid-19 analysis.
Thanks to the large Medicare study, they already had well-validated estimates for pollution exposure across the entire continental United States. Based on a combination of satellite and ground-level monitors, they included data about PM 2.5 pollution at a level of detail less than a mile square. This enabled them to produce pollution estimates for more than 3,000 U.S. counties, encompassing 98 percent of the population, and meld it with Johns Hopkins’ Covid data, adjusting the results to account for more than 20 possible confounding factors.
By early April, when Covid-19 cases had risen to more than 200,000 in the United States, with the number in New York alone surpassing cases in China, they had their initial result, and it was significant: An increase of only 1 microgram per cubic meter in PM 2.5, they calculated, was associated with a 15 percent increase in the death rate.
Now the team faced the question of when and how to publish their results—a decision made far more complex both by the pandemic and by rapidly changing practice in the biomedical world. Typically, it takes months to get research published, as manuscripts are peer reviewed for scientific accuracy, novelty and importance. Authors may be asked to make revisions, and may have to submit their work to more than one journal. The New England Journal of Medicine (NEJM), perhaps the most prestigious medical journal, only publishes 5 percent of the 16,000 papers submitted to it each year.
But there was another option. Scientists increasingly were disseminating their work more quickly by posting their papers online on so-called “preprint” servers, soliciting feedback from colleagues around the world before formal peer review and publication. Biomedical science had come late to this practice (social scientists and engineers had been doing it for years), but now scores of new Covid-19 research papers were appearing daily on a health sciences preprint server called medRxiv (pronounced like “med-archive”), established in June, 2019 by Yale University and the nonprofit Cold Spring Harbor Laboratory.
In an editorial in February, NEJM said it was taking steps to expedite peer review, but encouraged authors of Covid-19 manuscripts to post on preprint servers, noting, “Practitioners and public health authorities need actionable information as soon as possible.”
Wu said the Harvard study team knew that preprints remained controversial for many of their colleagues. Furthermore, the subject of their study would put the work instantly in the spotlight.
“Research like this, as soon as it’s out on preprint, we will have a lot of attention,” Wu said. “So we had discussions. It was a hard decision. But as long as the research is helpful in fighting against the Covid-19 pandemic, we decided we should let the public know as soon as possible.”
As Covid-19 cases and fatalities were escalating in the United States, the Trump administration took several actions on pollution that—in the researchers’ view—lent urgency to their work. On March 26, EPA Administrator Wheeler took the extraordinary step of announcing a suspension of environmental law enforcement because of Covid-19—relief that had been sought by the oil and gas industry. Wheeler insisted that companies were only being relieved of reporting requirements, not their pollution limits. But in fact, without self-reporting by the companies EPA would not know whether they were violating pollution limits or not.
On March 30, the EPA finalized a rollback in vehicle fuel economy standards, a step the administration’s own regulatory analysis concluded would increase adverse health impacts due to PM 2.5. Meanwhile, any day Wheeler was expected to open for public comment his decision not to strengthen the national air quality standard for PM 2.5.
“At the time of a pandemic that attacks the lungs, it’s really inconceivable not to clean the air,” Dominici said. “I wanted to put this study out there as soon as possible—considering all of the limitations, and being very open about discussing the limitations—to raise the urgency and the importance of this issue.
“We are in the middle of a pandemic, which honestly, we don’t know how to fight, even without what I think are really unwise decisions by the Trump administration,” she said.
Dominici checked with some of the major scientific journals, and was assured that they would still be eligible to submit their manuscript for peer review and publication if it was posted as a preprint first.
In New York, mobile morgues were set up outside strained hospitals. U.S. Surgeon General Jerome Adams said that it would be the “hardest and the saddest” week for “most Americans’ lives.”
Shortly after 8 a.m. on April 7, New York City’s deadliest day in the pandemic, with 597 coronavirus fatalities, the Harvard team’s study was posted on MedRxiv. The paper didn’t mention the Trump administration by name, but took its Covid enforcement policy to task.
“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 authors wrote.
By noon, The New York Times posted a story about the new study, noting it could have “far-reaching implications” in regard to the Trump administration’s environmental rollbacks.
Controversy, and a Revision of Findings
Reaction to the Harvard study was swift, and divided.
Joe Biden, then close to securing the Democratic presidential nomination (Bernie Sanders would drop out of the race the next day), shared the Times story on Twitter. “We’re starting to see evidence that long-term exposure to air pollution—which disproportionately affects communities of color & low-income communities—is linked to COVID-19 death rates,” he posted. “It’s past time we treat pollution like the threat to our health it is.”
Fifteen environmental groups cited the Harvard research in the lawsuit they filed the following week against the Trump administration over its pandemic-related rollback of environmental enforcement. “EPA’s policy is especially dangerous in light of recent research showing that increased air pollution causes a statistically significant increase in the COVID-19 death rate,” said the complaint led by the Natural Resources Defense Council.
Meanwhile, former coal industry executive Steve Milloy, who served on Trump’s EPA transition team and is a vocal supporter of the president’s environmental rollbacks, blasted the study as “totally absurd” on his web site, junkscience.com. He repeated his oft-stated position, which conflicted with the consensus science and even the stance of Trump’s EPA, that PM 2.5’s link to premature death was not “biologically plausible.”
For his part, EPA Administrator Wheeler dismissed the Harvard results as preliminary, then lit into the researchers.
“We look forward to reviewing the Harvard study once it’s completed and peer reviewed, but at this point, it’s not,” he said on April 14, the day he announced his long-anticipated decision not to strengthen the air quality standard for PM 2.5.
“I have to say, at least in the press, that the scientists seem to have a bias,” Wheeler added. “They are talking about our enforcement discretion memo, and either they didn’t read it or they didn’t understand it.” Wheeler repeated that there would be no pollution increase due to the enforcement memo, an assertion impossible to document at that point.
Dominici, in an interview, defended the rigor of the team’s scientific results, given the limitations of the available data. But she bristled at the notion that concern over the health impacts of air pollution was an improper bias.
“I’ve worked with EPA and I’ve been funded by EPA,” said Dominici. “My own bias is that I’m biased in favor of protecting the public. And I’m biased in favor of protecting the most vulnerable. So yes, you know, this is a bias I’m really proud to have.”
But the Harvard researchers also faced criticism from within the scientific community.
Two scientists who have published extensively on the associations between air pollution and adverse health effects said the Harvard research had so many shortcomings they were not convinced of its results. In an essay on April 21 for The Conversation, Paul Villeneuve, an epidemiologist at Carleton University, and Mark Goldberg, a professor of medicine at McGill University, noted why county-level data was inadequate for drawing conclusions. They brought up a number of potentially problematic factors that the Harvard study had not taken into account, including varying physical distancing policies and the possible underreporting of deaths. Interactions among the factors were so complex, they argued, it was impossible to account for them without information about individuals.
“Proper peer review must not be bypassed—and the onus for respecting its role falls not just on journalists but also on scientists to communicate the correct information accurately,” Villeneuve and Goldberg wrote.
The following week, on April 27, the Harvard team updated and revised their results. They added three more weeks worth of coronavirus fatalities and nearly doubled the number of potentially confounding variables in the analysis—including, for example, the stage of the pandemic and the state stay-at-home orders. The result was a slightly weaker but still significant association between PM 2.5 and Covid-19 deaths: an 8 percent—instead of 15 percent—greater risk of death for every 1 microgram per cubic meter increase in PM 2.5.
“As soon as you get the preprint out, you can receive feedback from scientists all around the world,” said Wu. “As a result we have much stronger models in the second version of the paper.”
The American Petroleum Institute, however, argued that the revision was evidence of the study’s underlying weakness. The update was “an action that suggests some fragility in the overall study results, and sets a precedent that the findings will likely continue to change due to the preliminary nature of the work,” API said, in an analysis of the Harvard study that the industry trade group submitted to EPA’s science advisory committee.
Cox, chair of the EPA’s Clean Air Scientific Advisory Committee, had harsher words, in an interview with The Washington Examiner.
“This whole thing is a bogus piece of analysis,” he said. “It’s technically unsound. It has sensational policy implications, none of which are trustworthy.”
Views on the Harvard study—and on the connection between air pollution and Covid-19—were hardening along political fault lines.
“In his inaugural address, Trump spoke of ‘American carnage.’ He is delivering just that,” said former New York Mayor Michael Bloomberg and NRDC president Gina McCarthy, who served as EPA administrator under Obama, in an editorial on May 4. Citing the Harvard study, they blasted Trump for “his push to increase air pollution in the midst of an unprecedented respiratory pandemic—jeopardizing people’s ability to survive it.”
The reliably conservative editorial page of The Wall Street Journal shot back the same day, lamenting that “opponents of fossil fuels are flogging a sloppy study that ties pollutants to coronavirus deaths.
“It may be that long-term exposure to PM 2.5 modestly raises the coronavirus death risk, but more rigorous studies are needed to prove this, and shutting down fossil-fuel production now won’t save lives,” the Wall Street Journal said. “It will deny livelihoods to hundreds of thousands of people.”
‘A Bit of a Wild West Out There’
Somewhat lost in the debate over the Harvard study was the fact that more research was emerging on the links between air pollution and Covid-19.
Some of these studies, like the Harvard study, were preprints that had not yet been peer reviewed—including research out of Italy, England, and China. But scientific journals were catching up, and publishing the first peer-reviewed studies showing an association between air pollution and Covid-19.
A particularly striking peer-reviewed study came out of the Netherlands, where researchers were able to use detailed demographic and air pollution data available at the level of small municipalities, much smaller than the U.S. county-level data available to the Harvard group. The study found a 13 percent to 16.6 percent increase in Covid mortality related to a 1 microgram per cubic meter increase in PM 2.5, a risk that their findings suggested was not just due to the dense population of cities. Some of the highest PM 2.5 levels, and associated Covid-19 risk, were in rural areas where the pollution was related to intensive agricultural activity.
“We’re kind of wiping out that big city complication, and we still find a very strong relationship, making it even more intriguing,” said Matthew Cole, an economist at the University of Birmingham in England.
Cole believes that individual-level medical data will still be necessary for scientists to draw more definitive conclusions on the link between air pollution and Covid-19. But he believes it is worthwhile for researchers to begin to explore the question with the imperfect data now available. “Anything that increases our understanding of the patterns of the impact of Covid across regions is obviously very useful,” he said.
“I must admit, I’ve seen some awful Covid papers,” said Cole. “Truly awful data and analysis—junk, essentially, and you should infer nothing from it. And it’s a little difficult for a non-academic to judge whether a piece of work is meaningful or not. But actually, when you find a useful preprint done by a credible group, then it gives you some evidence that there’s something that maybe we need to look into further.”
He added, “I think we need to be a little careful that we don’t dismiss all of these studies as being flawed, because I think increasingly, some of this evidence is looking quite convincing.”
Some efforts are underway to speed up the peer review process to address the need for reliable scientific information in the response to Covid-19. Scientific journals do have a rapid review process that has been used for many coronavirus-related studies, primarily studies that could inform life-saving changes in medical practice. The MIT Press and University of California, Berkeley late last month launched an online journal that is designed to accelerate peer review of Covid-19-related studies released as preprints.
“All of a sudden we have a global health emergency, yet with the complete lack of scientific information, and so a desperate hunger for scientific information about every aspect of this virus—from the basic science of it to the socio-economic correlations,” said Stefano Bertozzi, dean emeritus of the UC Berkeley School of Public Health, who will co-edit the journal. “And, as a result, the preprint servers really came into their own, with all their warts.”
Bertozzi said the journal is “an experiment,” an effort to provide some scientific verification that will be useful to policymakers, the public health community and journalists as they navigate the proliferation of pre-publication Covid-19 studies now circulating.
“I mean, it’s a bit of a Wild West out there,” he said.
A Long-Term Challenge
Dominici said the attention to her team’s Covid-19 study has been “totally unexpected,” and unlike anything the group had faced previously in its high-profile research, but she defended the decision to release the results as soon as they could.
“It makes you uncomfortable to go out there and have so much attention for something that hasn’t gone through peer review,” she said. “In retrospect, I’m comfortable with this decision. It’s not like I’m recommending a drug that could harm you. The worst I’m saying is that air pollution is bad for you, which we know is true.”
The Harvard team submitted the study to NEJM which ultimately passed on publishing it. The manuscript is currently under review at another journal. Dominici said the team is re-running its analyses to take into account the deaths that have occurred since they first posted the study. So far, she said, the results are consistent with the earlier findings and she anticipates their conclusions will remain the same.
One unanticipated advantage of the early release is that the Harvard group inspired researchers around the world to do similar studies, making use of the data sets and models that her team made public when they posted their preprint. Teams are working on studies on air pollution and Covid-19 in Brazil, India, China, the United Kingdom and the Netherlands, drawing on their work. Dominici said she also had a group of graduate and undergraduate students who meet by video conference weekly to work together on future and deeper research on whether environmental factors are significant in Covid’s spread or severity.
“In April, I thought this was going to be really bad, but it was going to pass,” Dominici recalls. “I thought, ‘Let’s write this paper and feel like we’re being helpful.’ Now, I’ve been shifting my mindset completely, to where I realize we’re going to be with this for a very long time. This is not a single project, but this is going to be my research for the next few years.”
Wu and another of the five co-authors of the paper recently gave a video presentation on their work to a meeting of an All-Party Parliamentary Group on Air Pollution in the United Kingdom that is considering environmental policy related to Covid-19. In an interview, Wu said he believes the research also could help inform policymakers on where to prioritize social distancing measures, personal protective equipment distribution, and other measures to stem the spread of the virus.
“The takeaway message is that we wanted this message to be delivered to the public as soon as possible,” Wu said.
The Trump administration, for its part, has not shown any inclination to move quickly.
“I have no thoughts on what EPA’s role should be, other than to note that, to date, the papers I have seen relating Covid-19 risks to air pollution do not provide reason to conclude (or suspect) that Covid-19 risks depend on air pollution,” said Cox, the chair of the agency’s clean air science advisory committee, in an email.
At EPA, officials said the agency is tracking and reviewing studies, but those published to date cannot show causation. They can only identify hypotheses for further research.
Asked whether the EPA plans any such studies, an EPA spokeswoman said by email, “We currently do not have research underway in this area.”
Top Photo Credit: Stephanie Keith/Getty Images
