PM2.5 Dose-Response: Function or Fiction?

The objectives of the GBD are

“systematically incorporate information on non-fatal outcomes in health status assessment” and “ensure that all estimates and projections were derived on the basis of objective epidemiological and demographic methods, which were not influenced by defenders.” [emphasis added].

This dose-response function (DRF) extends from 1 μg/m3 up to 600μg/m3. It is the basis for GBD’s prediction of around 7 million excess global deaths annually, about 25% more than the latest COVID estimate.

P.M2.5 it is a collection of airborne particles defined solely by their size (aerodynamic diameter of less than 2.5 μm). It is measured by collecting and weighing an air sample in a filter. This size limit was based on the probability of particle deposition in specific regions of the respiratory tract. P.M2.5 it is relatively easy to measure and can be linked to a wide variety of emission sources; it is basically a regulatory construct.

The Clean Air Act charges the EPA with regulating emissions from various sources to keep ambient concentrations within a national standard, that is, below a threshold for adverse health effects. The chemical composition of regulated pollutants is specified for gases but not for particulates, which are not limited to known toxicants.

The scientific issues here include:

  • The mathematical function used as the basis for risk prediction (logarithmic vs. linear)
  • Previous risk estimates used to support the DRF.
  • Differences in PM composition2.5 on those estimates.
  • Differences in the types of exposures involved.
  • Existence of thresholds.

Math function. I started my evaluation by replicating the GBD DRF (the red line) and the data used to derive it. The use of the logarithmic exposure scale is perhaps the most fundamental issue here; implies that additional exposures multiply instead of add to mortality risk. This may be the case for infectious diseases where one victim can infect many others, but has no role in toxicology, where thresholds of no effects followed by additive effects are expected. The Clean Air Act relies on such thresholds, but the EPA has not identified one for PM2.5, nor does it have the GBD DRF.

The GBD logarithmic function was intended to link three disparate sources of airborne particles.

  • ambient air pollution [AAP]) from urban traffic, combustion sources and windblown dust
  • second-hand tobacco smoke (HSS)
  • sources of indoor air pollution such as cooking and heating (HAP).

A logarithmic DRF underestimates the risks at high concentrations and overestimates them at lower concentrations. The GBD DRF would predict that an order of magnitude increase of 10 to 100 or 60 to 600 μg/m3 would double the risk of mortality, while a linear DRF would not predict any risk at 10 μg/m3 and a hazard ratio of about 20 at the highest concentration level.

The plot indicates that a linear DRF (the dashed line) could also fit these data, with correlation coefficients > 0.9.

Additional data. Below I have included additional risk estimates based on London data from the 1960s, primarily to broaden the range of environmental exposures. [1] The data set now indicates two divergent sets of risk estimates, ambient air quality data with much lower risks and household air pollution with much higher risks. Indoor or home hazards may be greater because they involve exposures directly to people’s breathing zones. In contrast, the lowest environmental data is based on centrally located outdoor measurements for which individual exposures are problematic.

Secondhand smoke (SHS) is a crucial data point in the original GBD DRF, linked to a PM2.5 50 μg/m level3 and well beyond the range of measured interior levels. I have never seen a long term indoor PM2.5 level as high as 50 μg/m3, and changed the estimated SHS exposure to 20 μg/m3 to match available measurements. However, the main effect of SHS may be to alter the chemical composition of indoor air, a factor not considered by the GBD.

Characteristics and exposures of PM. The three types of particles for which the GBD DRF was built vary substantially in size, composition, and exposure. The GBD and almost all regulatory assessments take the big step of equating emission rates or ambient concentration levels with actual human exposures, as is the case with the GBD particulate categories.

  • Housewives are likely to be directly exposed to indoor pollutants, but for a limited time.
  • Long-term exposures to secondhand smoke may be limited to living with a regular smoker.
  • Ambient air pollution poses the most significant exposure uncertainties because it is only measured outdoors. In contrast, we tend to spend about 85% of our time indoors, where air quality can be quite different from what is measured at a centrally located outdoor monitoring station [2].
  • Ambient air pollution depends on emission sources in the area, which vary widely between rural, suburban, industrial, or high-density cities and can be seasonal. The ambient air pollution data I added is mostly from London in the 1960s, categorized as “British smoke” and measured as the blackness of soot-stained filters rather than the mass collected. Such particles are primarily elemental carbon and are more harmful than the indiscriminate types of particles collected as PM.2.5.
  • Indoor pollutants and secondhand smoke are source-specific and presumably independent of geography and time period.
  • Indoor air pollution comes predominantly from the use of solid fuels indoors, often under incomplete combustion (smoldering) conditions. Somewhat larger dust particles may also be involved.
  • Secondhand smoke includes thousands of pollutants of unknown toxicity, such as carbon monoxide, benzene, formaldehyde, and various hydrocarbons. SHS particles tend to be much smaller than 2.5 μm, often called nanoparticles, and are therefore capable of entering the bloodstream and traveling to multiple organs.

Therefore, there is no physical or chemical similarity between the three categories of particles.

Alternative dose-response functions. I replotted these data with a linear exposure relationship and limited to 300 μg/m3 (an order of magnitude higher than current urban environmental levels); only indoor contamination demonstrates a dose-response relationship, and that relationship is weak.

A valid overall risk estimate must be based on all available evidence after accounting for the uncertainties in each of those determinations. In particular GBD PM2.5 risk estimates do not include confidence limits.

The GBD DRF fails on all counts:

  • The logarithmic function does not agree with toxicological principles.
  • The composition of the particles is not controlled.
  • The exposure mode is not controlled.
  • The length of exposure is not considered.
  • The risks of high exposures are underestimated.
  • The validity of individual risk estimates has not been established.

Furthermore, it is hard to imagine a practical use for such global risk estimates. Excessive exposure to air pollution is already known to portend health risks, as does smoking, but it is unclear which emission sources need to be controlled to reduce ambient PM.2.5. Maybe the Sub Pink The objective of the GBD was to generate fear accounts in the media and thus continue the process.

The original warning still holds:

Adequate direct evidence is lacking to identify the shape of mortality RR (relative risk) functions at the high ambient concentrations observed in many locations around the world.

[1] Frederick W. Lipfert. Air pollution and community health. Willie, 1994

[2] Frederick W. Lipfert, Indoor Air Quality. DOI: 10.1080/10962247.2017.1349010

Leave a Comment

Your email address will not be published. Required fields are marked *