Reduction in Measurement Error Confounds Cumulative Pollution Exposure
Environ Health Perspect. doi:10.1289/ehp.11804 available via http://dx.doi.org [Online 26 September 2008]
Referencing: Air Pollution, Airway Inflammation, and Lung Function in a Cohort Study of Mexico City Schoolchildren
Barraza-Villarreal et al. (2008) showed a convincing link between increased air pollution and reduced forced expiratory volume in 1 sec (FEV1). However, the apparent stronger association between reduced FEV1 and cumulative exposure over 1–5 days may be due in part to a reduction in measurement error of particulate matter < 2.5 µm (PM2.5) and not a true cumulative effect (Barraza-Villarreal et al.'s Figure 3).
Air pollution studies are prone to measurement error. In the study of Barraza-Villarreal et al. (2008)—as in most others—the estimates of air pollution came from a network of fixed monitors. Each child's day-to-day exposure was assigned using the closest monitor, and no monitors were > 5 km from the child's home or school. However, even with a monitor near the child's location, the estimate cannot be perfect because of variation in individual exposure (e.g., time spent outdoors).
I evaluatedthe effect of measurement error using a simulation study. I assumed that the 158 asthmatic children had a PM2.5 exposure given by
PMc2.5 = 28.9 + bc,
c = 1, …, 158,
bc ~ N(0, σ2b).
The mean PM2.5 exposure is 28.9 µg/m3, and each child (c) varies around this mean (b). This between-child variation means that some children live in more polluted areas than others.
The children's FEV1 was observed at repeated times, which was simulated using
FEVct1 = 1.89 + fct + αPMc2.5,
c = 1, …, 158,
t = 1, …, nc,
fct ~ N(0, σ2f),
where 1.89 L/sec is the mean FEV1, t is time, nc is the number of observations for child c, and fct is the measurement error in FEV1. The parameter α controls the change in FEV1 due to PM2.5 exposure.
In the study of Barraza-Villarreal et al. (2008), FEV1 was dependent on PM2.5 exposure from the previous 1–5 days. Daily PM2.5 values are subject to measurement error (e), which I simulated using
PMct2.5 = PMc2.5 + ect,
ect ~ N(0, σ2e).
Barraza-Villarreal et al. (2008) used a mixed model to estimate the effect of PM2.5 on FEV1 and controlled for the repeated FEV1 results from the same child. They also controlled for a number of covariates; however, for this simulation study I simply regressed the simulated daily values, FEV1ct, against the simulated daily pollution values, PMct2.5, and included a random intercept for each child.
Figure 1. Increase in the estimated effect of PM2.5 with increasing lag using a simulation study. Vertical lines are the mean estimate and 95% confidence interval.
|
I assumed a between-child variation in PM2.5 of σ2b = 2.82 and an equal measurement error in PM2.5 of σ2e = 2.82 (by naively using the standard deviation in PM2.5). I assumed a measurement error variation in FEV1 of σ2e = 0.662. I simulated data for 158 children and random sampled the number of observations per child (nc) by rounding a randomly generated value from a normal distribution N(11,2.22).
The results of 100 simulations are shown in Figure 1. Longer exposure lags gave estimated reductions that more closely approximated the true effect. On face value, longer exposure appears to be more damaging to health, but the simulated data had no cumulative effect. The stronger effect occurred because of the regression dilution bias and a reduction in the measurement error of PM2.5 exposure from using multiple days (MacMahon et al. 1990). Although different simulation results can be obtained by varying the strength of the pollution effect and measurement errors, the trend will always be to increased effects with increasing exposure periods.
The results of this simulation show that care should be taken when summing repeated measurements. Cumulative measurements are confounded by reductions in measurement error, which makes interpretation difficult.
The results of this simulation in no way invalidate the results found by Barraza-Villarreal et al. (2008). There is strong evidence that increased exposure to air pollution damages lung function. However, it is difficult to estimate how much of this reduction is due to a cumulative effect, thus requiring methodological development.
The author declares he has no competing financial interests.
Adrian Barnett
School of Public Health
Queensland University of Technology
Kelvin Grove, Queensland, Australia
E-mail:
a.barnett@qut.edu.au
References
Barraza-Villarreal A, Sunyer J, Hernandez-Cadena L, Escamilla-Nuñez MC, Sienra-Monge JJ, Ramírez-Aguilar M, et al. 2008. Air pollution, airway inflammation, and lung function in a cohort study of Mexico City schoolchildren. Environ Health Perspect 116:832–838.
MacMahon S, Peto R, Cutler J, Collins R, Sorlie P, Neaton J, et al. 1990. Blood pressure, stroke, and coronary heart disease. Part 1, prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias. Lancet 335:765–774.
Reduction in Measurement Error: Barraza-Villarreal et al. Respond
Environ Health Perspect. doi:10.1289/ehp.11804R available via http://dx.doi.org [Online 26 September 2008]
We thank Barnett for his comments on our article (Barraza-Villarreal et al. 2008), in which we reported associations between ambient air pollution and adverse lung function outcomes in a cohort of schoolchildren in Mexico City, Mexico. In the last several years, the adverse effects of air pollution on lung function, such as decrement in forced expiratory volume in 1 sec (FEV1) has been clearly demonstrated (Gauderman et al. 2007; Romieu et al. 1997). Before our study, there were reports of associations between cumulative particulate matter [PM < 10 µm (PM10) and < 2.5 µm (PM2.5 ) in aerodynamic diameter] and gaseous (ozone, sulfur dioxide, and nitrogen dioxide) air pollutant exposure and decrease in lung function in other studies (Downs et al. 2007; Romieu et al. 2006). Replication of these findings in different populations under different conditions of exposure is an important aspect of epidemiologic research, with consistency of results strengthening the weight of evidence for a true association between exposure and outcome.
Figure 1. Same day (A,B) and 2-day cumulative (C,D) PM2.5 distributions. (A,C) original data. (B,D) Simulated data.
|
However, air pollution exposure assessment is always a critical factor in environmental epidemiology. Like other studies of air pollution and lung health, our study (Barraza-Villarreal et al. 2008) relied on ecologic rather than personal indicators of exposure. Exposure misclassification due to the use of fixed-site ambient monitors rather than personal dosimeters is likely to underestimate rather than overestimate the effect of air pollution on lung function.
In his letter Barnett mentions that "the apparent stronger association between reduced FEV1 and cumulative exposure over 1–5 days may be due in part to a reduction in measurement error of particulate matter < 2.5 µm (PM2.5) and not a true cumulative effect." He attempted to verify this assertion by carrying out a simulation study; however, we see several problems with it. First, in his simulations, Barnett assumed a normal distribution (Figure 1). Several distributions have been reported as adequate for PM2.5, among them log-logistic, log-normal, and gamma. Using the data from our study (Barraza-Villarreal et al. 2008), we carried out an exercise similar to Barnett's, but we fitted different distributions (data not shown). The one that best fit our data was the gamma distribution. Second, when considering cumulative exposure, it is important to take into account the correlation between the observations on consecutive days; it is not enough to simulate from a distribution and then add the exposure. The models presented by Barnett did not take into account this correlation. Third, we reproduced the simulation of FEV1 as presented by Barnett (data not shown) and observed that it could produce negative value for FEV1 because it does not take into account the correlation of observations within children, although a sample size for each child was simulated and an artificial mixed model was fitted.
In conclusion, because Barnett's simulation of PM2.5 was based on a normal distribution, it does not reproduce the original structure of our data (Figure 1) (Barraza-Villarreal et al. 2008); therefore, the conclusions obtained are not applicable.
The authors declare they have no competing financial interests.
Albino Barraza-Villarreal
Consuelo Escamilla-Nuñez
Leticia Hernández-Cadena
Isabelle Romieu
Instituto Nacional de Salud Publica de México
Cuernavaca, Morelos, México
E-mail:
abarraza@insp.mx
Silvia Ruiz-Velasco
Instituto de Investigaciones en Matemáticas Aplicadas y en Sistemas
Universidad Nacional Autónoma de México
México, D.F., México
Jordi Sunyer
Environmental Epidemiological Research Centre
Institut Municipal d'Investigació Médica
Barcelona, Spain
References
Barraza-Villarreal A, Sunyer J, Hernandez-Cadena L, Escamilla-Nuñez MC, Sienra-Monge JJ, Ramírez-Aguilar M, et al. 2008. Air pollution, airway inflammation, and lung function in a cohort study of Mexico City schoolchildren. Environ Health Perspect 116:832–838.
Downs SH, Schindler C, Liu LJ, Keidel D, Bayer-Oglesby L, Brutsche MH, et al. 2007. Reduced exposure to PM10 and attenuated age-related decline in lung function. N Engl J Med 357(23):2395–2397.
Gauderman WJ, Vora H, McConnell R, Berhane K, Gilliland F, Thomas D, et al. 2007. Effect of exposure to traffic on lung development from 10 to 18 years of age: a cohort study. Lancet 369(9561):571–577.
Romieu I, Meneses F, Ruiz S, Huerta J, Sienra JJ, White M, et al. 1997. Effects of intermittent ozone exposure on peak expiratory flow and respiratory symptoms among asthmatic children in Mexico City. Arch Environ Health 52(2): 368–376.
Romieu I, Ramirez-Aguilar M, Sienra-Monge JJ, Moreno-Macías H, del Rio-Navarro BE, David G, et al. 2006. GSTM1 and GSTP1 and respiratory health in asthmatic children exposed to ozone. Eur Respir J 28(5):953–959.