Li et al. 2000; Z. Li et al. 2000; Li et al. 2001; Schwartz 2001; Chang et al. 2003; Kelsh et al. 2003; Lamm 2003; Buffler et al. 2004; Gibbs 2004a,b). The ecologic study of Morgan and Cassady (2002) assessed cancer incidence in a community exposed to both perchlorate and trichloroethylene in drinking water. No epidemiologic studies, either published or unpublished, have measured both thyroid outcomes and perchlorate exposure from drinking water in the same people.

Many of the epidemiologic data related to the effects of perchlorate are derived from ecologic studies. The smallest units on which exposure or outcome data are available are geographically defined units, most commonly counties, states, or countries; exposure data, outcome data, or both are available only at that level, not on individual subjects. Because ecologic studies do not include information about exposure and outcome in individuals, they are considered to be the weakest type of observational studies. In ecologic studies, comparisons are made between exposures and outcomes among large units. An example of an ecologic study would be one that compares the mean serum concentration of thyroid-stimulating hormone (TSH) in infants born in a county in which the drinking-water supply contains perchlorate with the mean serum TSH concentration in infants born in a county in which the drinking water does not contain perchlorate.

That design is subject to what is referred to as the ecologic fallacy: associations observed at the ecologic level may not apply at the individual level. For example, an observation that the average serum TSH concentration is higher in newborns in a city in which the water supply contains perchlorate than in newborns in cities in which the drinking water does not contain perchlorate would be compatible with a perchlorate effect on thyroid function. In that example, being compatible with an association assumes that no other important variations between the cities could account for the difference in serum TSH concentrations. However, it is not known from such a study whether infants who have high TSH concentrations were themselves exposed to perchlorate during gestation. How well ecologic studies are able to characterize individual exposure depends, in part, on how much variability of exposure there is in the geographic unit. For example, ecologic studies of the correlation of dietary fat intake and breast-cancer mortality by country would undoubtedly suffer from considerable variation in the average dietary fat intake among the persons in a given country. The use of average community exposures for ecologic studies of perchlorate in municipal drinking water may be less subject to error, depending on the population coverage of the water supply. In many of the exposed communities included in the literature, an entire community’s water supply has

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement