Selecting indices of occupational exposure for epidemiologic studies
Year: 1999
Type: other
Abstract: Decisions about how best to describe the magnitude or timing of exposure are challenging in most every epidemiological study because the number of possible representations of any epidemiologic relationship is large, but knowledge is limited, This paper seeks to promote a discussion of problems surrounding the selection of indices of exposure for epidemiologic studies in occupational health. Five themes are considered: 1) general approaches to selecting exposure indicators; 2) measurement error; 3) exposure-time or latency, analysis, 4) dosimetrie models, and 5) philosophical issues in exposure-response modeling. The predominant methods for selecting exposure indicators are based on operational criteria: one strategy (the maximum estimate criterion) is to choose indicators that maximize the observed association of exposure and disease: a second is to select ones that maximize the goodness-of-fit of statistical models (the best fit criterion). Results based on these approaches can be contradictory. Recent research shows the maximum estimate criterion is not justified as a general model-building strategy. It is reliable only with simple regressions with random, nondifferential error in the single predictor variable or single 2 × 2 tables with nondifferential exposure misclassification. When measurement error is systematic or confounders are measured with error, the maximum-estimate criterion is inappropriate. It is also questionable when exposure-time analysis is used to define latent periods. The best fit criterion may give a valid statistical description of the data, but it is supported only by statistical arguments which may be invalid in the absence of biological foundations. Optimizing the grouping of subjects to improve the accuracy of exposure assignment has greater theoretical justification, but further research is needed on the properties of different grouping procedures. The expanding use of dosimetric models and other techniques for increasing biological resolution may help penetrate the epidemiologic black box. but introduces new methodological and ethical challenges. Model building is a feedback process involving theory, knowledge and beliefs, on the one hand, and sets of data on the other. Consequently, some consideration of the philosophical basis of that process is essential. Future research might focus on the properties of measurement error, dosimetric models, and other advanced exposure estimation techniques. A need also exists for more deliberate development of epidemiological theory and an ongoing examination of research methods in relation to goals for science and public health.
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2154622273