April 21, 2020

expert reaction to preprint on air pollution in England and COVID-19 severity

A preprint, posted on medRxiv, reports on air pollution in England and COVID-19 severity

Prof Paul Monks, Professor of Atmospheric Chemistry and Earth Observation Science, and Head of Science and Engineering College, University of Leicester, said:

“The analysis needs more consideration as the results seem heavily biased by the data points in London, that drive the observed correlations between deaths and air-pollution.  Further, the chemical effects described may not be causal in terms of the relationships between NO2 and Ozone which would be expected to anti-correlate for chemical reasons.  Though it might instinctive that air pollution, as it can cause inflammation, might be a co-factor with COVID-19, the analysis presented is in my view inconclusive as to a strong effect in the UK.”

Prof Keith Neal, Emeritus Professor of the Epidemiology of Infectious Diseases, University of Nottingham, said:

“This study demonstrates those cities with most cases are those with the highest levels of air pollution.  The study uses the number of deaths and not the death rate per head of population.  Just using the number of deaths means large cities with larger populations will always look worse. London and the Midlands have populations around 10 million, the other regions in the order of 5 million.

“It is also apparent that international mixing is much greater in places such as London, Lombardy and New York so the epidemic started earlier and  has had more time to cause more cases.

“COVID-19 around the world has affected areas with the greatest population densities to a much greater degree.  The population density and more people to get infected could well explain their results rather than the direct effects of air pollution.”

Prof Anna Hansell, Professor in Environmental Epidemiology, University of Leicester, said:

“This paper shows a correlation between regional air pollution levels and COVID-19 cases and deaths in England. It is a preprint paper and has not been through peer review. It provides a preliminary and broad-brush analysis. Limitations in the study design mean findings cannot be interpreted as necessarily causal.

“The authors use regional air pollution concentrations, but most air pollution researchers now use concentrations at home address to best reflect people’s exposure – levels of pollutants can vary a lot e.g. between someone living near a busy main road vs. someone living in the countryside. Using regional concentrations will result in a lot of uncertainty in the analysis.

“There are differences between regions other than air pollution that may explain the findings. The statistical analysis does not make any corrections for these differences – such as population density, ethnicity, deprivation and age distribution of the population. It’s important also to remember we do not know the true underlying coronavirus infection rate in the population – the testing regime in the UK to date has mainly been to confirm the diagnosis in people with more severe disease. Regions with higher deaths and cases may just reflect the fact that (i) the epidemic started earlier in those regions, (ii) they include densely populated areas where social distancing is more difficult so the disease is like to have spread more quickly (but we don’t have data on this) and (iii) they include very deprived areas where people have higher rates of chronic illness that predispose to more severe COVID-19 -so they would be more likely to be detected and to die from the disease. Finally, the London and Midlands tend to have higher air pollution levels as they include very densely populated areas – individuals are an important source of air pollution!

“Air pollution researchers often notice associations that appear to be inverse between ozone and NO2/NO as in this paper. This doesn’t mean that ozone is protective against COVID-19. It relates to the chemistry between the two – nitrogen oxides are precursors of ozone, while ozone can be in turn ‘scavenged’ by them, so the concentration patterns may appear inverse and this gets reflected in the associations.

“We know that air pollution increases the risk of cardiovascular and respiratory diseases and that these diseases predispose to worse outcomes with COVID-19, which gives us an indirect link with air pollution. We need detailed research to explore whether there is a direct link between air pollution and COVID-19.”

Prof Roy Harrison, Professor of Environmental Health, University of Birmingham, said:

“The headline finding of this study, that air pollution exposure exacerbates the severity of COVID-19 disease is very plausible.  Air pollutant exposure is known to harm lung development, reduce lung function and increase the prevalence of other lung diseases.  It would therefore be expected to increase the impact of an infection which attacks the lungs.

“This, however is a very superficial study which does little to advance knowledge.  It is solely observational, so, as acknowledged by the authors, cannot prove cause and effect.  It is expected that disease transmission and hence case numbers and attributable deaths would be higher in highly populated regions where air pollution levels are also elevated.  One therefore has to look at case fatality rates (i.e. the number of deaths as a ratio to the number of cases), ideally adjusted for the demographics of the exposed population, to understand whether air pollution is having an effect.  The paper examines crude fatality numbers, but not case fatality rates.

“There are weaknesses in the analysis of air quality data which combines roadside , urban background and rural sites without any population weighting.  It is therefore unlikely to well reflect population exposure.  The spatial resolution is very coarse, and the treatment of atmospheric chemistry is also weak.

“To my mind, this kind of study which has been rushed out without adequate consideration of the properties of the data is not helpful.”

Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:

“There are some big problems in making sense of these findings. Given the findings, and those of other research in other places, I think the researchers’ suggestion of doing more detailed research on the subject is good. But, though it certainly sounds plausible that air pollution might have an effect on a respiratory disease like COVID-19, I don’t believe that this particular piece of research, on its own, yet provides clear evidence for such a link.

“At best, all this research has found is an association, a correlation, between the observed levels of COVID-19 confirmed cases and COVID-19 deaths in the eight NHS regions in England, and measures of average levels of two air pollutants (nitrogen oxides and ozone) in those regions from two years before the pandemic. (They looked at a third pollutant, nitrogen dioxide, but did not find statistically significant correlations.) They found that the higher the average level of nitrogen oxides measured in a region, the higher its number of cases and deaths tend to be – and the higher the average level of ozone, the lower a region’s cases and deaths tend to be. The trouble is that there’s no way these findings show that it’s the nitrogen oxide levels that actually cause the differences in cases and deaths, or indeed that higher ozone levels protect against cases and deaths. On ozone, the researchers themselves give reasons why high ozone very probably won’t be causing low numbers of cases and deaths – the picture is more complicated, as they explain. But what about nitrogen oxides? The problem is that the NHS regions differ from each other in a great range of ways other than air pollution levels, and there’s no way to tell whether the differences in numbers of cases and of deaths are caused by these other differences, by nitrogen oxide air pollution, or (as seems likely to me) by a large range of different factors. For instance, in these data the numbers of cases and deaths were greatest in the London region and the Midlands region, the regions that contain England’s two biggest cities. Air pollution does tend to be worse in big cities, but there are many other differences between big cities and smaller places, and between regions containing really big cities and regions that don’t, apart from air pollution levels. This research just can’t investigate what the causes actually are, because it only looked at one aspect, air pollution. A fuller investigation will take much more detailed and complex research, that can’t be done in a rush.

“Then there are some issues about the data that were used. We’ll all have heard of issues with the accuracy of data on confirmed cases and on deaths, but I think these provide at least a broad picture of what’s happening. But I find it surprising that these researchers analysed the counts of cases and deaths in the 8 regions, without taking any account of the fact that they differ in total population to a considerable extent. The Midlands region, which includes both East and West Midlands, has a population of over ten and a half million. The North-East region’s population is only about two and a half million, about a quarter the population size of the Midlands. So, if everything else was equal (which it isn’t), you’d expect about four times as many COVID-19 cases and deaths in the Midlands region than in the North-East. The other regions vary in population between about 5.5 million and about 9 million. As far as I can see, the researchers on this study took no account of these different population sizes. Then on the average measurements of the three pollutants involved, the researchers probably got the best data they could find, but to make a single measure of average air pollution over a region that (for the Midlands) goes all the way from Hereford to Lincoln via several major conurbations, and where air pollution levels vary hugely from place to place, isn’t to my mind likely to give an adequately representative measure of air pollution that applies to the whole region. There are well-known statistical problems in trying to draw conclusions about individual places or cities from data that are available only for wide regions.

“It’s worth mentioning again that the research has not yet been peer reviewed by other scientists. I think that there are several issues about it that would at least need to be revised during peer review. And I should point out that the sort of review that I and other commenters will have had been able to do, given the time constraints, is not so thorough as would happen in a proper peer review.”

Dr Rosie Cornish, Senior Research Associate in Population Health Sciences at the Bristol Medical School, University of Bristol, said:

“This is a study which shows a correlation/association.  Confounders have not been taken account of.  The places with the highest levels of pollution are also the places with the highest population densities, which is obviously a major risk factor for spread of COVID-19.  They do not and cannot show that increased air pollution causes more severe COVID in those who get it OR even that increased air pollution causes greater incidence of COVID-19 or deaths due to COVID-19.  They have hugely overstated their findings.”

Dr Graham Wheeler, Medical Statistician, UCL, said:

“This study is only assessing the correlation between past UK air pollution levels and the number of COVID-19 cases or deaths in the UK.  It does not establish a causal link between these two measures.

“These analyses are based on 7 data points only, with measurements of air pollution averaged across entire regions.

“For some of these correlations, there is weak evidence to suggest there is a genuine association between air pollution levels and COVID-19 cases or deaths.  The data from the London region appear to have a big effect on these correlations too; with this data point removed, much weaker correlations would likely be reported.

“This work does not account for population density.  Therefore, the results are not surprising: areas with higher pollution have more people in them, and with more people in an area, more COVID-19 cases are likely to be reported.”

Dr Tony Fletcher, Associate Professor in Environmental Epidemiology, London School of Hygiene & Tropical Medicine, said:

“It is certainly plausible that air pollution could exacerbate respiratory disease and mortality due to COVID, either because of pollution having caused relevant pre-existing disease or by pollution affecting exposure due to viral particles attaching to particulate matter – however this paper is of low quality and does not provide evidence of such an association.  It is not yet peer reviewed and one hopes the peer reviewers will pick up these problems.  It comes from a toxicology group who should get in touch with their epidemiological colleagues, to address these basic epidemiological principles which it fails to respect.  They confuse numerator data and rates, claiming that it will compare infectivity and death “rates”, but in fact only comparing numerator data.

“Secondly, they carry out a crude ecological comparison with no adjustment for any other relevant factors, not only population is ignored, but age, socioeconomic status, mobility, and any other factor affecting risk are also excluded.  Of course, London has the most cases, as well as the largest population and highest NO2 levels – and it drives the association they find.”

Dr Liam Brierley, Research Fellow, Department of Biostatistics, University of Liverpool, said:

“This preprint (work that is not yet peer reviewed) examines air pollution levels and total numbers of COVID-19 cases and deaths in each of the 8 English regions.  A slightly earlier work, also not yet peer reviewed, also reported a trend of more severe disease and higher death rate in cities with greater air pollution in China (https://www.medrxiv.org/content/10.1101/2020.04.04.20053595v2).

“However, this study by Travaglio and colleagues does not directly model death rate or case fatality, but numbers of deaths.  This distinction is important to make as it changes the message of the research from what is implied in several places.  The former would suggest that people are more likely to die if developing COVID-19 disease in a region with higher air pollution, whereas what the study actually shows is that people are more likely to develop COVID-19 disease in these regions, but not more likely to die from it once they have developed the disease.

“The patterns presented show that numbers of cases and number of deaths are higher in regions with higher air pollution (e.g. London, Midlands, North West).  However, would be much more convincing if other factors, such as population size were corrected for.  For example, we would expect London to have both high air pollution and high counts of cases/deaths even if air quality and the disease were completely unrelated, simply because London is a large urban centre with many people.

“Other research available does demonstrate that air pollution may be important for COVID-19 as regions with higher air pollution are more likely to have greater prevalence of chronic respiratory diseases including asthma and chronic obstructive pulmonary disease (COPD), which can affect COVID-19 disease severity.  Air pollution is therefore still an important consideration in public health responses to pandemic respiratory infections, though much more detailed research will be necessary to fully understand how this affects COVID-19 risks for the UK.”

Preprint (not a paper): ‘Links between air pollution and COVID-19 in England’ by Marco Travaglio et al. is on medRxiv.  This work is not peer-reviewed.

https://www.medrxiv.org/content/10.1101/2020.04.16.20067405v2

All our previous output on this subject can be seen at this weblink:

www.sciencemediacentre.org/tag/covid-19

Declared interests

Dr Graham Wheeler: “I am employed by UCL.  I am a Fellow, Chartered Statistician and Statistical Ambassador of the Royal Statistical Society, and a voluntary research committee member for Chiltern Music Therapy, a not-for-profit organisation providing music therapy services. I have previously received honoraria from AstraZeneca and Novametrics Consulting Ltd.”

None others received.