March 3, 2020

expert reaction to study looking at air pollution and life expectancy

Research, published in Cardiovascular Research, reports that lives are shortened by an average of three years worldwide due to air pollution.

Dr Heather Walton, Senior Lecturer in Environmental Health, Environmental Research Group, King’s College London, said:

Does the press release accurately reflect the science?

“What is in the press release is accurate. There are some points in the original paper that it would have been useful to highlight in the press release.  For example, the paper has a careful discussion of the fact that air pollution as a risk factor has a statistical association with mortality rather than a distinctive one where individual deaths can be identified (such as car accidents). 

Is this good quality research?  Are the conclusions backed up by solid data?

“The paper is reasonably well done.  Along with other papers in the field, it uses Global Burden of Disease methods that have to make some compromises in order to do calculations all over the world.  The models of air pollution concentrations are usually validated with direct measurements of air pollutant concentrations but that is not possible in all parts of the world.  The models are also done at a large spatial scale e.g. one estimated air pollution concentration for an area of 50 square kilometres or more.  This misses increased concentrations besides roads.  PM_2.5 and ozone are regional pollutants so this may not matter as much for these pollutants, but it matters a lot for nitrogen dioxide.  This is not covered in these calculations.  The results might be a little bit higher if nitrogen dioxide was included but there is some overlap between the effects (and some controversy as to whether nitrogen dioxide has health effects independent of the other pollutants).

The ‘counter-factual’ was not clearly identified.  This is the scenario with which the results are being compared.  It was probably the lowest concentration at which effects had been shown in epidemiological studies, as used in Burnett et al 2018 (reference 6 in the paper).

How does this work fit with the existing evidence? 

“This study builds on earlier studies of the global burden of disease.  The overall answer in terms of numbers of deaths is similar to that in a paper published in 2018 (reference 6 in the paper).  The method of modelling the air pollution concentrations is different from that in the 2018 paper and, as all modelling approaches have some level of uncertainty, it is useful to have results using a different method. 

“Lilieveld et al adds to the calculations in Burnett et al (2018) by (i) using a different model for air pollution concentrations; (ii) giving results by country and region (iii) giving results by source of emissions and (iv) giving results by age.  All 4 aspects have been calculated previously but only using an earlier version of the concentration-response function predating that in Burnett et al (2018).  (The new concentration-response function relating pollutant concentration change to health effects was a significant development because it included outdoor air pollution studies at high concentrations for the first time.  Previously, the health response at higher concentrations had been assumed by analogy with indoor air pollution and smoking).

Have the authors accounted for confounders?  Are there important limitations to be aware of?

“This study is not itself an epidemiological study, it is a calculation illustrating the implications of the results of previous epidemiological studies.  The original studies do usually account for confounders where possible.  The paper does discuss several limitations, for example, whether particular sources of particulate matter are more harmful than others.  Some other limitations arise from the compromises needed to do calculations on a global scale (as discussed for modelling of concentrations above).  The method for calculating the health burden is also simpler than some more sophisticated approaches e.g. the numbers of deaths are multiplied by the average life expectancy for a specific age group to give the loss of life-expectancy rather than calculating the loss of life expectancy in a birth cohort over time using a full lifetable approach.  Because of the above compromises, the results may differ from those done individually in specific countries.  These are usually able to use more detailed information e.g. modelling at a finer scale.

What are the implications in the real world?  Is there any overspeculation?  

“By their nature, health impact assessments are intended to illustrate implications in the real world, so these implications are clearer than in some other academic papers.  It would have been useful to have more discussion of the fact that air pollution is only one of several risk factors that may contribute to the same death.  Thus, considering each risk factor separately may be somewhat unrealistic as they probably interact with each other.  However, it is currently difficult to take this into account.

‘Using a new method of modelling the effects of various sources of air pollution on death rates’ – what is this new method? What data is this based on?

“Previous work used a statistical model combining information from ground level measurements, satellite data and chemical transport model predictions.  This study used a chemical transport model alone (which allows clearer attribution to sources).  A chemical transport model takes account of chemical reactions that occur in the atmosphere as emissions travel through space.

Do you agree that ‘the findings suggest the world is facing an air pollution “pandemic”.’

“The press release (but not the paper) comments on air pollution as a ‘pandemic’.  Pandemic means ‘affecting everyone’ and this is certainly true of air pollution, which is a difficult risk factor to avoid.  However, the term ‘pandemic’ is more typically used for new infections that spread suddenly.  The effects of air pollution on health have been around for a long time.  That does not change the fact that it is an important public health problem that needs to be addressed.”

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

“This new study is, in my view, statistically sound. What’s mainly new about it is its use of a relatively new statistical model, GEMM (standing for Global Exposure Mortality Model), to describe the relationship between exposure to air pollutants and death rates. This new model was published in 2018 and is based on a wider range of data from more countries than previously used. In particular it reduced the uncertainty about death rates in areas with high levels of pollution from fine particulate matter. The data behind models like this generally comes from observational studies, and so there must always be some doubt about the extent to which the excess deaths are actually caused by air pollution, as opposed to being caused by some other characteristics of people who live in highly polluted places. And the excess deaths described in studies like this come from statistical calculations – faced with individuals who died from strokes, you can’t usually tell that this one’s stroke was caused by air pollution and that one’s stroke wasn’t. But, in aggregate, they are real deaths, and we don’t have a better method of estimating the numbers.

“The comparisons with other factors that shorten life, like tobacco smoking, violence, and HIV/AIDS, are useful to put things in context. I don’t think we can be confident that ambient air pollution kills more people than tobacco smoking, even though in this study the best estimates of the numbers of deaths and of the shortening of life do indicate that. There is too much uncertainty about the exact numbers, for statistical reasons and because there is still lack of knowledge about cause and about the details of how air pollutants affect health – and, though the evidence of the harm from tobacco smoking is better established, there’s some uncertainty about the exact numbers for that too. But I do think that what Professor Lelieveld says, that the numbers of deaths and loss of life expectancy from air pollution ‘rival’ those from tobacco smoking, is probably a fair summary. The pattern of cause and effect is clearer for tobacco, but the overall numbers are pretty similar to those for air pollution.

“However, the comparison between air pollution and tobacco certainly doesn’t mean that it’s just as bad for you to go out into a street with high air pollution as it is to smoke some cigarettes. These figures are all averages across whole populations. On average, across the world, according to figures quoted in this study, the loss of life expectancy from tobacco smoking is 2.2 years. But this is the average over people who aren’t exposed to tobacco smoke at all, those who might be affected only by passive smoking, and those who smoke themselves. The loss of life expectancy of smokers is at least 10 years, compared to non-smokers. Those not exposed to tobacco smoke won’t lose any life because of smoking, and the average loss of life from passive smoking isn’t huge. Not everyone is exposed to tobacco smoke, and the 2.2 years figure for loss of life expectancy is an average across people who lose no years of life and people who lose a large number because of their smoking habit. The effect on an individual’s lifespan of breathing polluted air will, on average, be considerably less than the effect of smoking a pack of cigarettes a day. But, in many places, the proportion of people exposed to polluted air will be higher than the proportion of smokers, because we all have to breathe the air where we live. That ubiquity of exposure to air pollution is a key reason why the average effect on the lifespan is high.

“The comparison with deaths from violence is even starker. People living in a war zone, or a country with very high levels of criminal violence, can have a horribly high chance of having their lives shortened by violence. But, thankfully, most of the people in the world don’t live in such places, and the average loss of life worldwide from violence, 0.3 years according to this study, is therefore rather low. The effects of air pollution are nowhere near as dramatic on an individual basis, but far more people in the world breathe polluted air every day than live in war zones.”

Dr Samuel Cai, Senior Epidemiologist, The George Institute for Global Health, University of Oxford, said:

“The press release is generally accurate. I found this study is of very good quality and the conclusion sounds correct. The global mortality rate of 8.8 million/year generated from this modelling study is higher than what previous studies showed, this is mainly because that this new study has now 1) included more comprehensive epidemiological data in different parts of the world, for example, mortality from hypertension/diabetes were also considered in this study; 2) used more advanced models to extract air pollution data and its emission sources more accurately. Convincingly, this 8.8m deaths is in agreement with what Burnett et al reported in 2018¹, who reported 8.9m deaths. So, I believe these are the most up-to-date estimation based on data the world has to offer. 

“Just like any modelling studies, there will always be some degrees of uncertainties. In particular, when it comes to modelling health effects from particulate air pollution, it is very important to take into account the toxicity of these particles into account, as toxicity of these particles varies across different regions within and across countries, depending on the emission sources/intensity. Also, how defense systems in human body react and process these inhaled particles still needs further investigation. These are the key challenges that many scientists in the field are working hard to find answers.

“The authors combined recent advanced models to answer their scientific questions. For example, they used the Global Exposure Mortality Model (GEMM), which is based on an expanded list of studies from around the world, in particular from regions where the concentration of air pollution is in the higher end (e.g China). Put simply, now the GEMM model is able to cover a wide range of air pollution concentrations (from very low to very high). Previously this was not possible due to a lack of high-quality studies from regions where air pollution is severe. Then the researchers combined this GEMM model with a highly sophisticated air pollution model, which has the ability to distinguish different emission sources based on data-driven methods.

“This study once again shows that air pollution is a leading risk factor for health worldwide. It is not a secret that air pollution is the ‘new tobacco’, so the public health implication is very clear – authorities need to act swiftly/comprehensively to protect their citizens from air pollution through science-based policy and healthcare communities should be fully aware of this and act accordingly to advise and protect the most vulnerable groups. I sort of agree that this is a ‘pandemic’ if this is the trendy term, as WHO put it, 9 out 10 people in the world affected by harmful levels of air pollution. So by definition, and the fact that millions of people indirectly die from it every year, it sounds like the word ‘pandemic’ qualifies.

“I do not feel there is any overspeculation in the findings of this study as they are supported by the good science we see in this study and some recent studies like Burnett et al.”

1. ‘Global estimates of mortality associated with long-term exposure to outdoor fine particulate matter’ by Burnett et al. https://www.pnas.org/content/115/38/9592

Prof Stefan Reis, Science Area Head for Atmospheric Chemistry and Effects, UK Centre for Ecology & Hydrology, said:

“This new study adds further evidence of the harmful effects of air pollution on human health at the global scale. Exposure to air pollution may affect the incidence and severity of other non-communicable diseases, this is highlighted by the findings. In addition, adverse health effects at lower concentrations than previously considered contribute health impacts. This makes the case to reduce population exposure even when current guideline and limit values for air pollutant concentrations are attained much stronger. It is worth noting that emissions from natural wildfires are estimated to amount to just 10% of total global biomass burning. The recent severe bushfires in Australia and wide-spread burning in South America and South Asia raise the question of how much of man-made (or human activity influenced) biomass burning emissions can be reduced. And following on from that, to what extent climate change may affect the rate of natural wildfires and contribute to further health impacts in some world regions will need careful consideration.”

‘Loss of life expectancy from air pollution compared to other risk factors: a worldwide perspective’ by Lelieveld et al. was published in Cardiovascular Research at 00:05 UK time on Tuesday 3 March.

DOI: 10.1093/cvr/cvaa025

Declared interests

Prof Kevin McConway: “Prof McConway is a member of the SMC Advisory Committee, but his quote above is in his capacity as a professional statistician.”

Dr Samuel Cai: “I declared no COI.”

Prof Stefan Reis: “No conflicts of interest”

None others received