Difference between revisions of "Concentration-response to PM2.5"

Revision as of 07:51, 8 May 2015

Question

What is the the quantitative dose-response relationships between outdoor air PM2.5 concentration and mortality due to cardio-pulmonary, lung cancer, and other non-accidental causes (index Cause of death 2)?

Answer

This code gets the ovariable of this page and calculates some basic results.

library(OpasnetUtils) # A package with Opasnet functionalities
library(xtable)       # A package with table formats
library(ggplot2)      # A package with fancy graph formats

objects.latest("Op_en2202", "calculations") # Get the latest ovariables from code calculations on page Op_en2202.

ConcentrationresponsetoPM2.5 <- EvalOutput(ConcentrationresponsetoPM2.5)

print(xtable(ConcentrationresponsetoPM2.5)@output), type = 'html') # Show a result table

ggplot(ConcentrationresponsetoPM2.5@output, # Show a bar chart
	aes(x = Exposuremetric, y = ERFResult)) +
	geom_bar(position="dodge") + 
	theme_grey(base_size = 24)

Rationale

⇤#: This paper should be included in the text.

• Burnett RT, Pope CA 3rd, Ezzati M, Olives C, Lim SS, Mehta S, Shin HH, Singh G, Hubbell B, Brauer M, Anderson HR, Smith KR, Balmes JR, Bruce NG, Kan H, Laden F, Prüss-Ustün A, Turner MC, Gapstur SM, Diver WR, Cohen A. An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure. Environ Health Perspect. 2014 Apr;122(4):397-403. doi:10.1289/ehp.1307049 Epub 2014 Feb 7. Erratum in Environ Health Perspect. 2014 Sep;122(9):A235. [1] [2] --Jouni (talk) 14:19, 26 January 2015 (UTC)

Data

Concentration-response function

PM_2.5 are fine particles less than 2.5 μm in diameter. Exposure-response function can be derived from exposure modelling, animal toxicology, small clinical or panel studies, and epidemiological studies. Exposed population can be divided into subpopulations (e.g. adults, children, infants, the elderly), and exposure is assessed per certain time period (e.g. daily or annual exposure).

• Health effects related to short-term exposure
  • respiratory symptoms
  • adverse cardiovascular effects
  • increased medication usage
  • increased number of hospital admissions
  • increased mortality
• Health effects related to long-term exposure (more relevance to public health)
  • increased incidence of respiratory symptoms
  • reduction in lung function
  • increased incidence of chronic obstructive pulmonary disease (COPD)
  • reduction in life expectancy
    • increased cardiopulmonary mortality
    • increased lung cancer mortality

Sensitive subgroups: children, the elderly, individuals with heart and lung disease, individuals who are active outdoors.

Mortality effects of long-term (chronic) exposure to ambient air

In principle the ERFs for long-term exposure (produced by cohort studies) should also capture the mortality effects of short-term exposure (ERFs produced by time-series studies). In practice it is likely that they do not do so fully. This is due to the so-called "harvesting" phenomenon, i.e. it is possible that acute exposure, at least to some extent, only brings forward deaths that would have happened shortly in any case. However, adding effects of acute exposure to effects of long-term exposure is problematic because the risk of double-counting. ^[1]

Pope et al. (2002) ^[2]

• 6% increase in the risk of deaths from all causes (excluding violent death) (95% CI 2-11%) per 10 µg/m3 PM2.5 in age group 30+
• 12% increase in the risk of death from cardiovascular diseases and diabetes (95% CI 8-15%) per 10 µg/m3 PM2.5 in age group 30+
• 14% increase in the risk of death from lung cancer (95% CI 4-23%) per 10 µg/m3 PM_2.5 in age group 30+

Woodruff et al (1997) ^[3]

• 4% (95% Cl 2%-7%) increase in all-cause infant mortality per 10 µg/m3 PM₁₀ (age 1 month to 1 year)

Tuomisto et al. 2008:^[4]

• A structured expert judgement study of the population mortality effects of PM_2.5 air pollution.
• Opinions of six European air pollution experts were elicited.
• Percent increase per 1 µg/m³ increase in PM_2.5:
  • Equal-weight decision-maker
    • Best estimate 0.97
    • 95% quantile 4.54
    • 5% quantile 0.02
  • Performance-based decision-maker
    • Best estimate 0.60
    • 95% quantile 3.80
    • 5% quantile 0.06

'Mortality effects of short-term (acute) exposure to ambient air PM

Anderson et al. 2004 ^[5]

• 0.6% (95% Cl 0.4%-0.8%) increase in all-cause mortality (excluding accidents) per 10 µg/m3 PM₁₀ in all ages
• 1.3% (95% Cl 0.5%-2%) increase in respiratory mortality per 10 µg/m3 PM₁₀ in all ages
• 0.9% (95% Cl 0.5%-1.3%) increase in cardiovascular mortality per 10 µg/m3 PM₁₀ in all ages

These coefficients are defined as distributions around estimates of central tendency for each cause of death.

Relative increase of mortality per 1 μgm-3 increase of outdoor PM2.5 concentration. Values were drawn with equal probability from the two distributions reported in ^[6], ^[7]

Uncertainties:

• Mortality estimate from Hoek et al. (2002)^[8] was not included due to many confounding factors related to mortality, e.g. road noise.
• Probability for PM2.5 assumed to be the true cause of the effects in 70 %, 90 %, and 10 % for cardiopulmonary, lung cancer and all other mortality, respectively (author judgement).
• Toxicity differences between ambient air particles and the particles generated by different bus types were not taken into account due to lack of comprehensive data. ^{[9] [10]}

• No threshold was assumed in the dose-response relationship. ^{[11] [12]}

Effect of infiltration

The mortality due to indoor exposure to outdoor pollutants can be described with this equation:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): log \Delta M_{in,j} = \frac{\Delta C_{out-in} t_{in}}{\Delta C_{out} t_{out} + \Delta C_{out-in} t_{in}} log \Delta M_{all,j} (1)

where logΔM_{all, j} is the increase in mortality due to the j^th outcome associated with total PM exposure for each 10 μg/m³ increase in PM₁₀ or PM_2.5.

j represents three major health outcomes: all-cause, cardiovascular, and respiratory mortality.

ΔC_out is the increase in outdoor PM₁₀ or PM_2.5 concentrations, which is set as 10 μg/m³.

ΔC_out-in is the increase in outdoor-originated PM₁₀ or PM_2.5 concentrations found in the indoor environment.

t_out is the duration of direct exposure to outdoor PM pollution.

t_in is the duration of indoor exposure to PM of outdoor origin.

logΔM_{in, j} estimates the increase in mortality due to the j^th outcome associated with indoor exposure to outdoor-origin PM for each 10 μg/m³ increase in PM₁₀ or PM_2.5

When one compares changes in mortality in different countries and different infiltration factors, this is what I think actually happens:

If infiltration factor is relatively low such as for PM₁₀ in China, the people are exposed less than predicted based on outdoor concentrations. When an epidemiologic study is performed, the observed mortality rate is lower than in other places with higher infiltration factors, and this can be falsely interpreted that the dose-response slope (ΔM_all) is less steep. When this mortality estimate was combined with the lower exposure estimate (due to low infiltration) in equation 1, you get too low an estimate for the mortality due to indoor exposure to outdoor PM. In contrast, the local infiltration factor could and should be used to adjust the observed mortality rate.

In other words, what you actually observe in epidemiological studies is not ΔM_all but ΔM_obs, which is something close to ΔM_in because so much time is spent indoors. The work should be about estimating ΔM_all based on ΔM_obs and infiltration and not vice versa. (Here I interpret ΔM_all as the true risk estimate that we would observe, if the exposure measure was correct.)

More precisely, let's look at the mathematics of log-linear regression. We assume that the logarithms of the observed probabilities (or rates) of disease have the probability distribution of

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): log(p) \sim a + b C,

where p is the probability of disease,

a is a constant describing the background probability,

b is a risk coefficient for the exposure, and

C is the exposure concentration in the population.

What happens with PM is that we observe the differences in the probability of disease due to differences in exposure C. However, we do NOT observe C itself (the actual exposure concentration) but only the surrogate C_obs, which in this case is the outdoor concentration of PM C_out. With a given difference in the probability of disease between the exposed and non-exposed groups, we therefore make a biased conclusion about b, and this bias can be described as:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): log(p_E) - log(p_0) \sim (a + b C_E) - (a + b C_0) = b (C_E - C_0) = b_{obs} (C_{E,obs} - C_{0,obs})

where E is the exposed group,

0 is the non-exposed group, and

obs is the biased observed variable (in contrast to the actual variable we would observe if all measurements were correct). Let's look at the ratio of the biased and correct risk estimates:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \frac{b_{obs}}{b} = \frac{C_E - C_0}{C_{E,obs} - C_{0,obs}} = \frac{\Sigma_i C_{E,i} t_i - \Sigma_i C_{0,i} t_i}{C_{E,out} - C_{0,out}}

where i means different microenvironments and t_i is the fraction of time spent in each microenvironment. If we use F_i to denote the relative exposure concentrations in different microenvironments i (in the case of only indoor and outdoor microenvironments F is 1 for outdoor and equal to infiltration factor for indoor), we get

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \frac{b_{obs}}{b} = \frac{\Sigma_i F_i C_{E,out} t_i - \Sigma_i F_i C_{0,out} t_i}{C_{E,out} - C_{0,out}} = \frac{(C_{E,out} - C_{0,out}) \Sigma_i F_i t_i}{C_{E,out} - C_{0,out}} = \Sigma_i F_i t_i

As we can see, the observed b is biased downward if the population spends a lot of time in microenvironments with low infiltration factor. Therefore, it is a mistake to assume that the b_obs would reflect logΔM_all. Instead,

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): log \Delta M_{all} = \frac{log \Delta M_{obs}}{\Sigma_i F_i t_i}.

and this should be used in any related calculations.

Calculations

This code defines and stores the ovariable of this page (ConcentrationresponsetoPM2.5).

library(OpasnetUtils)
library(xtable)

ConcentrationresponsetoPM2.5<- Ovariable("ConcentrationresponsetoPM2.5", ddata = "Op_en2202", getddata = FALSE, save = TRUE) 

cat("Ovariable ConcentrationresponsetoPM2.5 saved for later use. Page name: Op_en2202, code name: calculations\n")

print(xtable(EvalOutput(ConcentrationresponsetoPM2.5)@output), type = "html")

Unit

m³/μg D↷

See also

• Relative risks of mortality -page in Heande
• Health aspects of air pollution. Results from the WHO project "Systematic review of health aspects of air pollution in Europe". World Health Organization, 2004.
• Pope et al. 2004. Cardiovascular mortality and long-term exposure to particulate air pollution. Circulation (109), 71-77.
• C. Puett, Joel Schwartz, Jaime E. Hart, Jeff D. Yanosky, Frank E. Speizer, Helen Suh, Christopher J. Paciorek, Lucas M. Neas and Francine Laden: Chronic Particulate Exposure, Mortality, and Coronary Heart Disease in the Nurses’ Health Study. American Journal of Epidemiology, doi:10.1093/aje/kwn232
• NEEDS - New Energy Externalities Developments for Sustainability, Deliverable 3.7 "A set of concentration-response function", Integrated Project, Sixth Framework Programme, Project no. 502687.

References