Negative and non-positive epidemiological studies.

Abstract

The aim of this study was to identify and discuss validity aspects on so called negative and non-positive studies. Arguments and examples are drawn from experiences in occupational health epidemiology regarding the interpretation of more or less equivocal study results. A negative study may be defined as showing a result that goes against the investigated hypothesis of an increased (or prevented) risk. Traditionally, studies with a risk estimate (relative risk or odds ratio) above, but close to unity are also referred to as negative, given a narrow confidence interval (CI) that includes unity. A risk estimate above unity with the CI including unity is non-positive, however, but an estimate below unity with upper CI bond exceeding unity might be seen as possibly negative or non-negative. A weaker "significance" than usually required should perhaps be accepted when evaluating serious hazards. In contrast to positive studies, the negative and non-positive studies tend to escape criticism in spite of questionable validity that may have obscured existing risks (or preventive effects). Even stronger arguments can be made in criticising negative and non-positive studies than positive studies, for example, regarding selection phenomena, and observational problems regarding exposure and outcome. Negative confounding should be considered although usually weak. In case-control studies, so called over-matching may obscure an existing risk as could the "healthy worker effect" in cohort studies. Small scale non-positive studies should be made available for meta-analyses and when considering studies that do not convincingly show a risk; those who are exposed should be given the "benefit of the doubt".