August 7, 2023

expert reaction to study looking at air pollution and antibiotic resistance

A study published in the Lancet Planetary Health looks at particulate matter (PM)_2·5 air pollution and clinical antibiotic resistance.

Prof Julie Morrissey, Professor of Microbial Genetics; and Director of the Centre for Microbial Sciences and Infectious Diseases, University of Leicester, said:

“This observational study shows that globally high levels of air pollution are associated with increased risk of antibiotic resistance.  The dissemination of antibiotic resistance is multi-factorial and although this study shows a strong correlation between air pollution and antibiotic resistance, it is limited by a lack of experimental evidence and consideration of other environmental, social and economic confounding factors.  This is acknowledged by the authors.  However, this is still an important observation that is supported by previous epidemiological studies showing that antibiotic resistance genes and antibiotic resistant bacteria are associated with increased levels of airborne particulate matter.  Additionally, there is an increasing body of experimental evidence that shows that particulate air pollutants directly interact with bacteria, changing global gene expression that can potentiate antibiotic resistance and increase bacterial virulence.  To fully establish the impact of particulate matter on antimicrobial resistance further experimental research into the mechanism of the interaction between bacteria and particulate matter is critical.”

Prof Anna Hansell, Professor of Environmental Epidemiology; Director of the Centre for Environmental Health and Sustainability; and Director of the NIHR Health Protection Research Unit in Environmental Exposures and Health, University of Leicester, said:

“This study looks at associations between air pollution and antibiotic resistance at country-level in 116 countries.  While very detailed modelling has been performed, the underlying study design is ecological (hypothesis-generating) and associations are not necessarily causal.  Known factors that drive antibiotic resistance – such as lax antibiotic prescribing or infection control policies and unrestricted use of antibiotics in animal farming – were not examined in the study.  These may be important determinants of antibiotic resistance in countries that also have higher air pollution levels.  More confirmatory data, including laboratory studies to look at mechanisms and within-country studies, are needed to investigate the plausibility of the associations demonstrated.”

Dr Andrew Singer, Principal Scientist, UK Centre for Ecology & Hydrology, said:

“The whole premise of the paper is based on a statistical correlation between PM2.5 and antibiotic resistance within a set of pre-defined pathogens.  They observe that these two factors correlate globally (116 countries), however, one might expect lots of factors to co-correlate with high AMR.  It’s not an experimental study nor does it generate data on the mechanism for the phenomenon being proposed, however, they do acknowledge this limitation in the paper.  The study convincingly raises PM2.5 as an intriguing possible driver of AMR—one that hasn’t yet been examined in any great detail to date.  Like all good research, it raises more questions than it answers – but it will hopefully drive the research community to examine this in greater detail, especially from a mechanistic perspective, and it is possible that future research could confirm yet more reasons why we should reduce our PM2.5 load in the atmosphere, though this study itself isn’t able to prove air pollution causes AMR.”

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

“This study reports that the researchers collected a very wide range of data, and certainly the data that they used was complicated.  But I think the results require a lot of care in interpretation.  Quite a few of the quantities that the researchers mention in their paper could not actually be used in the data analysis anyway, for various reasons.  The researchers found interesting correlations and associations between measures of air pollution, antibiotic resistance, and other factors, but there remain doubts about cause and effect.  And the modelling of the possible future scenarios makes good sense only if the associations in the researchers’ model are at least approximately measuring cause and effect.  Since we can’t be sure of that, the scenario modelling might reflect what could happen, or it might not.

“Overall, this observational data analysis and modelling research does indicate that it might well be worth looking further at a role for air pollution in relation to antibiotic resistance, but at this stage there remains a very large amount of uncertainty about what’s really going on.  I’d say the new research raises more questions than it answers.

“Two aspects of the doubt about cause and effect are that the study is observational, and that it is based overwhelmingly on aggregate data for whole countries.  The main findings are associations between a country’s level of antibiotic resistance and its level of air pollution by fine particles, so-called PM_2.5.  Basically, the higher the PM_2.5 level was in a country, the higher was the level of antibiotic resistance, and that association appeared to get stronger as time passed.  The researchers looked separately at different bacteria in some of their analyses, and sometimes they also looked at resistance to different antibiotics for each type of bacterium.  But several of the main analyses just used an overall aggregate measure of antibiotic resistance, for whole countries and taking into account all the bacteria and all the antibiotics.  That’s a reasonable thing to do overall, I’d say, though it means some detail might be lost.  Probably more importantly, they used only one measure of air pollution, PM_2.5, despite having acquired data on other pollutants, probably because they did not have data on other pollutants for anywhere near all the countries they wanted to consider.  That’s probably not a huge issue, because there tend to be quite high correlations between the levels of different air pollutants, but again it means some relevant detail might be lost.

“Importantly, though, because it’s an observational study, there were differences between countries with different PM_2.5 levels, and between different years, apart from the PM_2.5 levels themselves.  One or more of these other differences, so-called confounding factors, could be the actual cause of the differences in levels of antibiotic resistance, and not the PM_2.5 levels at all.  The researchers were perfectly aware of that, and made statistical adjustments to allow for several of them, by including them in their statistical models as well as PM_2.5 levels.  For instance, they adjusted for the overall level of antibiotic use in each country, for some aspects of the weather (temperature and rainfall), for a measure of the availability of clean water, some educational measures, and more.  They could not make adjustment for all the factors that they mention in their paper as having been collected, again because for many of them they did not have the data for enough countries.  And there are likely to be other potential confounding factors that they couldn’t collect any data on at all.  Therefore it’s still possible that there are confounding factors at work, and that these are involved in the causation of a country’s level of antibiotic resistance.  So that’s one important reason why we can’t be sure that changes in air pollution levels in a country actually do cause changes in antibiotic resistance.

“The issue with aggregate data is that the data used in this study are for whole countries.  Some countries are extremely large and include regions that differ a great deal from one another.  Air pollution differs from one place to another, usually a lot, even in countries that are geographically small.  Assuming for a minute that there is a cause-and-effect relationship between air pollution and antibiotic resistance, it could be true that, in part of a country, it’s the average level of air pollution that matters in determining the level of antibiotic resistance, rather than the way air pollution varies within a town or city.  But in a country as large and varied as China or Russia or the USA, it could well happen that what goes on in different parts of the country is different enough so that the overall country averages just don’t represent what is going on overall.  The overall averages might show the appropriate pattern, or they might not, and since these researchers did not use unaggregated data for areas smaller than whole countries, you just can’t tell.  After all, the bacteria involved in antibiotic resistance don’t know which country they are in – they react to the local conditions, not the country aggregate.  So that’s another reason why we can’t be sure exactly what’s causing what.

“I’m certainly not saying here that it’s impossible that air pollution is a major cause of antibiotic resistance.  It might be.  What I’m saying is that this study couldn’t establish that air pollution is a major cause of antibiotic resistance, because it’s observational and because it uses aggregate average data for whole countries.  And because the scenario models essentially assume that the associations described in the models are describing cause and effect, there’s no way we can be certain that the numbers produced in the scenario models are close to what would actually happen if the assumptions about future policies do occur.  They might be reasonably close, or they might not.  It’s certainly important to deal with the rise of antibiotic resistance, and there are other good reasons why air pollution should be reduced and why some of the other changes in the scenario models, like better education and more access to clean water, should be made anyway.  But the new research has not shown to any degree of certainty that making these changes would reduce antibiotic resistance in the way that the scenario modelling says it would.  Really the scenario results are telling us more about how the statistical models work than about how the future will work.

“Something that would help in understanding cause and effect here, as the researchers themselves point out, is better knowledge about how small pollutant particles in the air could actually lead to more antibiotic resistance.  Figure 1 in the research paper shows some possible ways this might happen, and the press release draws attention to that Figure, but these are only potential pathways of cause and effect (as the release rightly says), and this new research does not directly look at evidence for or against them.

“The press release, in its final paragraph, does point out some of the limitations of the study that are mentioned in the research paper.  But there’s another limitation that I think could be very important.  The researchers do allude to it in the paper, but I think they don’t mention one very important part of it.  They write that, because their statistical models don’t account for all the differences between countries in antibiotic resistance, “some factors could be almost as important as PM_2.5 in contributing to antibiotic resistance.”  They give a long list of possible factors, that they didn’t have data on, that might be relevant in this way, including food ingestion, veterinary use of antibiotics, and more.  (This is in the penultimate paragraph of the paper.)  What they don’t mention is that including factors like that in their statistical modelling, if they could find the data to do so, could reduce the strength of the association between PM_2.5 and antibiotic resistance, possibly making it disappear altogether.  That’s because the models tell you (among other things) how strong the association is between PM_2.5and antibiotic resistance when all the other factors are kept constant – that is, they tell you the extra association between PM_2.5 and antibiotic resistance, on top of associations between antibiotic resistance and the other factors in the model.  That’s how they adjust for potential confounders, so that you can see what’s left over for air pollution to explain once the confounders have been accounted for.  So if you put more factors in, they might explain statistically the entire association between PM_2.5 and antibiotic resistance, or they might still leave a lot of it there.  It’s even statistically possible (though probably not very likely) that they could make the association larger.  We just can’t tell.”

‘Association between particulate matter (PM)_2·5 air pollution and clinical antibiotic resistance: a global analysis’ by Zhenchao Zhou et al. was published in the Lancet Planetary Health at 23:30 UK time on Monday 7 August 2023.

Declared interests

Prof Julie Morrissey: “No conflicts to declare.”

Prof Anna Hansell: “I do not have conflicts of interest to report.

I am Chair of the Committee on the Medical Effects of Air Pollution (COMEAP), but comments here are in a personal capacity as Professor of Environmental Epidemiology at the University of Leicester.”

Dr Andrew Singer: “No conflicts.”

Prof Kevin McConway: “I am a Trustee of the SMC and a member of its Advisory Committee.  My quote above is in my capacity as an independent professional statistician.”