Evaluating exposure cutpoint bias in epidemiologic studies of electric and magnetic fields.

Epidemiologists who study the association between exposure to electric or magnetic fields and adverse health outcomes often classify their subjects as "exposed" and "unexposed," and they report results based on an odds ratio. The exposure classification rule--or dichotomy rule--is typically based on a priori assumptions or arbitrary considerations. We show that results may vary substantially with selection of different cutpoints by which to dichotomize exposure. Further, interpretation and comparison of studies is dependent on the choice of cutpoint. We suggest the use of probability plots as a more informative method of data representation. To demonstrate the utility of probability plots, we re-analyze data reported by Savitz et al. [1988, Am J Epidemiol 128:21-38]. Using a higher exposure cutpoint than that of the original analysis, we obtained larger odds ratios, two of which achieved statistical significance. More important, probability plots of these data showed 1) consistency of results with measures of magnetic fields in both low- and high-power-use situations, and 2) discordance with results based on measures of electric fields. Given these observations, we recommend further study, especially that focused on the most highly exposed individuals.