Many of the concepts and much of the algebra of misclassification are applicable to assessing and interpreting errors in exposure and disease misclassification. Differences arise based on the structure of epidemiologic studies, which are designed to assess the impact of exposure on the development of disease and not the reverse. Also, the sources of error and the ways in which disease and exposure are assessed tend to be quite different, and thus the mechanisms by which errors arise are different as well. Health care access, a determinant of diagnosis of disease, does not correspond directly to exposure assessment, for example. Health and disease, not exposure, are the focal points of epidemiology, so that measurement of exposure is driven by its relevance to health. The degree of interest in an exposure rises or falls as the possibility of having an influence on health evolves, whereas the disease is an event with which we are inherently concerned, whether or not a particular exposure is or is not found to affect it. Once an exposure has been clearly linked to disease, e.g., tobacco or asbestos, then it becomes a legitimate target of epidemiologic inquiry even in isolation from studies of its health impact.
The range of exposures of interest is as broad, perhaps even broader, than the spectrum of health outcomes. Exposure, as defined here, includes exogenous agents such as drugs, diet, and pollutants. It also includes genetic attributes that affect ability to metabolize specific compounds; stable attributes such as height or hair color; physiologic characteristics such as blood pressure; behaviors such as physical exercise; mental states such as stress or depression; the social environment, including poverty and discrimination; and participation in health care, such as disease screening and receipt of immunizations. As a consequence of this remarkable range of interests that fall within the scope of epidemiology, there is a corresponding diversity of methods for measuring exposure (Armstrong et al., 1992). Tools include biological assessment based on specimens of blood or urine, physical observation, assessment of the social and physical environment, review of paper or computerized records and a broad array of tools based on self-report and recall, including instruments to evaluate stress, diet, and tobacco use. The distinctive features of the exposure of interest pose specific challenges to accurate measurement, and thus there are many different strategies for evaluating exposure accuracy and misclassification. Nevertheless, some generic principles or questions can be described.
Was this article helpful?