Interpretation of "negative" studies in occupational epidemiology

Two criteria for an interpretation of noneffect are that the relative risk estimate be near unity and that the confidence interval be narrow; lack of statistical significance has no bearing on this issue. A further requirement is that the relative risk estimate not be near the null value as an effect of non-differential exposure misclassification, negative confounding, or some other systematic error. Occasionally, when results are unexpected or difficult to accept, studies are "underinterpreted" as negative or inconclusive on the basis of arguments such as "confounding," "crude exposure assessment," or "lack of a known mechanism." The present paper supports the position that these arguments are commonly invalid. Scientific standards should be used to separate causal associations from noncausal ones, but in public health decisions this practice has to be balanced by the principle that the "benefit of the doubt" should be given to the persons subject to potential risk.