Arsenic and Drinking Water
To the Editor:
A recent paper by Hopenhayn and colleagues1 provides interesting evidence that arsenic in drinking water can lower birthweight based on a prospective comparison of pregnancies in 2 cities in Chile. Antofagasta has levels close to 40 μg/L in the public water supply, whereas Valparaiso water contains almost no arsenic. The crude birthweight means were not different, but following statistical adjustment for factors known to be predictive of birthweight, the estimated difference was 57 g.
Because the assessment of exposure was ecologic, the exposure variation that exists in the populations studied is confounded with city. This issue is not resolved by the individual-level exposure analysis the authors describe: the individual-level measures effectively serve as surrogates for the city-level measures, because almost all of the true exposure variation depends on city. (In this regard, a within-city analysis would have been of great interest if it confirmed the direction of the association.) To interpret differences based on an ecologically measured exposure, one should adjust not only for individual-level covariates, but for possible site-to-site differences in the impact of those covariates on the outcome.
The statistical model applied by Hopenhayn et al.1 imposes the assumption that the effect of each predictor on birthweight is identical for the 2 cities. In particular, the effect of early prenatal care is modeled as adding the same fixed increment to birthweight in both cities. Epidemiologists have recently appreciated that community context can play an important role in health; for example, the effect of income on health can be highly dependent on socioeconomic context.2 Consider the role of early prenatal care in improving birthweight. That well-known relationship may not only reflect a direct effect of prenatal care, but it may also serve as a marker for women who are taking good care of themselves in many ways. Such an “effect” may well vary with the community context, especially between 2 cities with pronounced social differences (as shown in Tables 1 and 2 of the paper by Hopenhayn1). The findings would be more convincing had the investigators allowed the effect of covariates (such as prenatal care) on birthweight to be different for the 2 cities. This could readily be accomplished by including interaction terms for city by covariate.
1.Hopenhayn C, Ferreccio C, Browning SR, et al. Arsenic exposure from drinking water and birth weight. Epidemiology
© 2004 Lippincott Williams & Wilkins, Inc.
2.Diez-Roux A. Invited commentary: places, people and health. Am J Epidemiol