Regulation and the use of "negative" results from human reproductive studies: the case of ethylene dibromide.

This paper examines the regulatory usefulness of five studies of the reproductive effects of occupational exposure to ethylene dibromide. None of the studies found a statistically significant adverse effect. Results from an epidemiologic study can find no adverse effect either because there is in fact no association between the outcome and the exposure being studied or because the study was not powerful enough to detect an association between the two. This distinction is especially important when "negative" studies are offered as evidence that a chemical is not harmful to workers. General reasons for low power in reproductive studies are discussed, and each of the studies is examined to see if the results were in fact negative and what factors were operating to affect the power of each study. Four of the studies showed potential reproductive impairments that were not large enough to be judged statistically significant by the authors. The power of all of these studies was then examined and found to be low for a number of reasons. These studies therefore do not represent evidence that ethylene dibromide is harmless; on the contrary, they may indicate the opposite. In general, when regulating potentially harmful chemicals in a low-power situation, power should be raised by raising the alpha level of the study, especially when animal data show a strong positive effect.