Air pollution and poverty: does the sword cut both ways?

T his issue of the journal includes three papers that touch on relations among socioeconomic status (SES), health, and air quality. Jerrett et al considered whether SES differentials in Hamilton, Ontario, modify the temporal relations between daily mortality and either coefficient of haze (COH) or SO2. 1 Martins et al did a similar analysis with respect to PM10 in Sao Paulo, Brazil. 2 The third paper, by Gouveia et al, also involved Sao Paulo but examined cross sectional relations between several pollutants and infant birth weight. As such, it involves SES factors only implicitly, by virtue of the trends seen by Martins et al that link levels of PM10 in Sao Paulo with residence in slums and other SES indicators. Inequitable distribution of environmental impacts within a city or region may raise issues of ‘‘environmental justice’’,* but it may also be possible to get additional insights into the implied health relations by probing a little deeper into the nature and origins of such differential impacts. Most time series studies are based on entire cities and spatially averaged air quality, in order to maximise statistical power and to preclude the necessity of assigning individual deaths to specific air quality monitors. Time series studies avoid SES confounding by design, as those factors do not vary on a daily basis. However, many air pollutants tend to vary in concert, especially those that are co-emitted by common sources, thus making it difficult to identify the most probable causal agent. Cross sectional studies may have less co-pollutant collinearity, but can suffer from SES confounding to the extent that SES may tend to decrease with residential proximity to major pollution sources. All of these issues are in play among these three studies. Jerrett et al matched daily deaths in five zones of Hamilton, Ontario with COH and SO2 data from each of the five available monitoring stations. Total deaths were about the same in all five of the unequally sized zones. They then examined the relations between these mortality responses and each of 12 potential effect modifiers that were based on zone-wide averages (making this an ‘‘ecological’’ study). Although a random effects model showed that the heterogeneity among zonal risk estimates was not significant, the authors concluded that manufacturing employment and educational attainment were significant modifiers of the effect of COH on daily mortality. They proposed three possible rationales for this finding: additional pollutant exposures from the workplace, reduced measurement error because of less mobility, or surrogate effects of material deprivation in general. Interestingly, household income showed no tendency to modify the time series relations, and the mean distance to hospital appeared to rank third (after education), suggesting the importance of access to medical care in these acute situations. The risk estimates were not associated with the mean ambient COH level, even though COH was moderately correlated with both poverty and unemployment. The basic problem here is that it is not possible to identify which of the several different rationales might be worthy of public health scrutiny, based on only five observations. An obvious remedy is to use individual rather than aggregate data, including estimates of exposure. Martins et al defined six subregions of Sao Paulo, each with a radius of 2 km but differing greatly in population. One region had four air quality monitors; the others, only one. PM10 was the only pollutant considered; mean values ranged from about 40 to 70 mg/m and appeared to be correlated with both slum housing and respiratory mortality response rate. Based on the data taken from the paper, an effect modification regression on both slum residency and the PM10 level appeared to be slightly superior to one based on slum residency alone. This outcome may be relevant to the widespread occurrence in Sao Paulo of levels above the former US PM10 annual standard. From that regression, it appears that the PM10 exposure level is just as important as socioeconomic conditions, but we have no way of knowing what other pollutant exposures might also be playing a part. For example, NO2 levels in Sao Paulo are about four times those in the US and another study of daily mortality in Sao Paulo showed large effects for CO and PM10. 4 Comparing this paper with a previous analysis of SES effects in Sao Paulo that was based on city-wide air quality suggests that local exposure (and thus individual) gradients could be very important. Gouveia et al studied the relations between birth weights of Sao Paulo infants and first, second, and third trimester exposures to PM10, CO, SO2, NO2, and O3. 3 A suite of potentially confounding variables was considered, but smoking, alcohol use, and poverty status were not among them. In terms of changes in mean birth weight, only first trimester exposures showed consistent negative effects, for all five pollutants. The decrements associated with mean levels of PM10, CO, NO2, were in the range 83–86 g; effects of SO2 and O3 were smaller and not significant. However, similar findings were not obtained with logistic regressions for the odds of low birth weight (LBW). For example, a birth weight decrement of 85 g applied to the entire population should have created a relative risk for LBW of about 1.5 (assuming a normal distribution), but the values reported were much smaller and SO2 even showed significant beneficial effects in the first trimester. Furthermore, the actual fraction of LBW in Sao Paulo (5%) is less than then typically seen in the United States (7.7%), even though the mean US birth weight is 200 g higher. Furthermore, a study of air quality and birth weight in the north eastern US that controlled for smoking and alcohol found that CO showed the most consistent effects on LBW, with an OR of 1.3 for a 1 ppm increment in the third semester, and a similar study in northern Nevada also found a negative association between third trimester PM10 and mean birth weight but not with the fraction of LBW. These discrepancies make it difficult to accept the Sao Paulo birth weight associations as causal. Differential environmental impacts may result from at least two important . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .