|Home | About | Journals | Submit | Contact Us | Français|
Experience from observational studies and randomised trials teaches us that sex was often misinterpreted as a modifying factor in the risk of disease or response to treatment. For example, on the basis of one underpowered subgroup analysis, we believed for more than a decade that aspirin didn't work for secondary prevention of stroke in women, while we knew it worked for men.
A group of researchers now set out to test the reliability of claims about sex differences in genetic studies. Their systematic review found 77 relevant reports whose authors made 432 claims of gene-sex interactions. In two thirds of these claims, the analyses according to sex were prespecified, rather than post hoc, but just under 13% of the claims were appropriately documented and reported. The authors reanalysed the 188 comparisons for which they had raw data and found that 105 were not statistically significant. Of the 83 that were significant more than half had modest P values—between 0.01 and 0.05. Furthermore, of the 60 claims researchers deemed most internally valid, only one was repeated in at least two other studies.
Ideally, gene-sex interactions should be explored in prespecified and adequately powered subgroup analyses, but post hoc analyses can be useful too. The key to reaching valid evidence is in clear and complete reporting, and in repeated testing of results by several studies.