A third limitation of the ecologic design is that disease outcome usually is not confirmed at the individual level. Most studies rely on routine reporting, either of mortality through death certificates or of cancer incidence through cancer registration and surveillance systems. Such sources of information vary in their degree of accuracy and completeness, and they can sometimes vary in relation to the surrogate measures being used to define exposure (e.g., geographic area). This can lead to the identification of spurious associations.
Fourth, ecologic studies seldom estimate or account for population migration or movement. This, too, can result in the appearance of spurious associations if aggregate or population measures of radiation exposure actually reflect underlying changes in population mobility with factors such as time, age, or geographic area.
Finally, descriptive studies are often based on a small number of cases of disease. Such studies have low statistical power to detect an association if it truly exists, and they are very sensitive to random fluctuations in the spatial and/or temporal distribution(s) of the disease(s) under study. This is especially true for diseases such as cancer, particularly childhood cancer, which are relatively uncommon on a population basis.
There have also been attempts to evaluate the effect of environmental radiation exposures using the two most common analytical study designs employed in epidemiology: the case-control and the cohort study. Such studies are almost always based on individual-level data and thus are not subject to many of the limitations summarized above for ecologic studies. Nevertheless, each of these study designs is subject to specific weaknesses and limitations. Of most concern in case-control studies is the potential bias that can result in relation to the selection of cases and controls, such that the two groups are differentially representative of the same underlying population. A second important source of bias can be differential recall of information about exposure for cases relative to controls. In cohort studies, a common limitation is the relatively small number of cases for uncommon disease outcomes and the resultant low statistical power. A second concern is the completeness of follow-up of the cohort under study, and equal follow-up and determination of disease status according to exposure. Such limitations of both types of analytic epidemiologic studies may be particularly problematic in investigations of low doses and relatively small increases in disease risk. Under such circumstances, the magnitude of the impact on risk estimates of small or modest biases may be as great or greater than the magnitude of the true disease risk.
In summary, most existing published studies of environmental radiation exposure are ecologic in design. Such studies are limited in their usefulness in defining the risk of disease in relation to radiation exposure or dose. They can sometimes be informative in generating new hypotheses or suggesting directions of study but seldom, if ever, are of value in testing specific hypotheses or providing quantitative estimates of risk in relation to specific sources of environmental radiation. Epidemiologic studies, in general, have limited ability to define the shape of the radiation dose-response curve and to provide quantitative estimates of risk in relation to radiation dose, especially for relatively low doses. To even attempt to do so, a study should (1) be based on accurate, individual dose estimates, preferably to the organ of interest; (2) contain substantial numbers of people in the dose range of interest; (3) have long enough follow-up to include adequate numbers of cases of the disease under study; and (4) have complete and unbiased follow-up. Unfortunately, the published literature on environmental radiation exposures is not characterized by studies with such features.
The accompanying tables provide a summary of the principal studies of environmental radiation exposure published since the BEIR V report (NRC 1990). Articles included in this summary were identified principally from searching the PubMed database of published articles from 1990 through July 2004. Searches were restricted to human studies and were broadly defined: key words included radiation; neoplasms; radiation-induced; radioactive fallout; and environmental radiation. Searches specific to the Chernobyl accident included Chernobyl, Russia, Ukraine, and Belarus as key words. Articles were also identified from UNSCEAR (2000b) and from the usual scientific interactions with other investigators. The tables are organized according to the type of exposure situation under study as follows: (1) populations living around nuclear facilities; (2) populations exposed from atmospheric testing, fallout, or other environmental releases of radiation; (3) populations exposed from the Chernobyl accident; (4) populations exposed from natural background; and (5) children of adults exposed to radiation. Within each type of exposure situation, the tables are further grouped according to study design: ecologic studies, case-control studies, and cohort studies. Each table contains a brief description of the principal design features and results of each study. The principal criteria used to assess the utility of each study in evaluating the risk of disease in relation to radiation exposure were the following: (1) Was there a quantitative estimate of radiation dose; (2) if so, was the estimate for individuals in the study (i.e., individual-level estimates of radiation dose received); and (3) was there a quantitative estimate of disease risk in relation to radiation dose?
Table 9-1A lists 16 ecologic studies of populations living around nuclear facilities, 13 of the locations being outside the United States. Most define exposure, or potential for exposure, based on a measure of distance from the facility, although the two studies of exposures at Three Mile Island by Hatch (1992) utilized some information on measurements