select search filters
roundups & rapid reactions
factsheets & briefing notes
before the headlines
Fiona fox's blog

air pollution and death

Researchers, publishing in PNAS, looked at global estimates of mortality associated with longterm exposure air pollution.

A roundup accompanied this Before the Headlines.


Title, Date of Publication & Journal

Global estimates of mortality associated with long, 2018, PNAS


Study’s main claims – and are they supported by the data

The paper does not support the main claim that the health benefits of reducing a specific kind of particulate matter are much greater than previously thought.

The study builds on previous work on the health impact of PM2.5s. PM2.5 stands for ‘particulate matter sized 2.5 micrometres or less’ and is a particular kind of pollution thought to have an especially harmful effect on the lungs.

The authors look at PM2.5 concentrations in sixteen regions of the world, and calculate that non-accidental, non-communicable deaths increase by about 60% in the regions with the highest outdoor PM2.5 concentrations (~80 micrograms per metre cubed) compared to the region with the lowest concentrations (2.4 micrograms per metre cubed). They conclude that this therefore means PM2.5 are responsible for increasing death rates.

This is an unsound conclusion from the data which the authors have assembled, as they have not ruled out the presence of a third variable which controls both air pollution levels and death rates. This mistake is sometimes described as mistaking ‘correlation for causation’. There are many reasons why death rates and pollution levels might be linked that are more complex than simply ‘pollution causes death’. For example, the wealth of a country might correlate both with the quality of fuel that it burns for power (poorer countries buy more polluting fuel) and the quality of treatment for pollution-related diseases such as heart attacks (poorer countries have less access to high-tech medicine). In this scenario, countries with more pollution would also have higher mortality, even if pollution had nothing to do with mortality. This is a plausible alternate explanation for the pattern the authors see, since the regions with the highest PM2.5 concentrations are also those regions which have most recently industrialised (India, Middle East and China). What is slightly complex here is that pollution is known to cause deaths (from other studies), so there is – probably – correlation and causation occurring at the same time.

Even if their findings were supported by evidence, the results are of limited importance in the UK. Western Europe already has some of the lowest PM2.5 concentrations in the world (~13.4 micrograms per metre cubed) and therefore the estimates the authors give of the global number of lives which could be saved by reducing PM2.5 concentrations are unlikely to be directly relevant to European decision makers.

It should also be emphasised that your risk of death in any given year is very low unless you are quite old or have a pre-existing health condition. Therefore, even the worst-case scenario of increasing this low risk by 60% will not have a noticeable impact on any individual person’s life.


The main strength of this study is extending existing estimates of the risk of outdoor PM2.5 levels to new countries – the model now covers around 97% of the world, according to the authors.

The main limitation is that, for reasons data availability, the authors have only fitted their model to the levels of outdoor PM2.5. This will almost completely ignore indoor PM2.5 sources including second hand smoke, household air pollution from use of solid fuels, and active smoking. The authors do not indicate how serious this omission is, but ignoring smoking completely in estimates of lung-based disease would not normally be considered acceptable without some discussion.

The authors do, however, make the argument that models of indoor PM2.5 are quite inaccurate and therefore their methods are appropriate; I lack the subject-specific expertise to judge whether their argument is adequate, but it is a striking limitation of the study.

Furthermore, the methodology for accounting for deaths is slightly unusual. The authors count any death that is not an accident or a communicable disease (unless it is a respiratory tract infection) as being caused by pollution. This would seem likely to overestimate the rate of deaths due to pollution, as – for example – a post-partum haemorrhage is non-accidental, non-communicable but clearly nothing to do with pollution.


PM2.5 – ‘Particulate matter sized 2.5 micrometers or less’; a particular kind of pollution thought to have an especially harmful effect on the lungs
Any specific expertise relevant to studied paper (beyond statistical)?
I have no expertise in this area beyond statistical.




Any specific expertise relevant to studied paper (beyond statistical)?

I have no expertise in this area beyond statistical

in this section

filter Headlines by year

search by tag