Good for women, good for men, bad for people: Simpson's paradox and the importance of sex-specific analysis in observational studies.

Abstract

Even if a medial intervention has a beneficial effect in both men and women, an observational study that combines data from men and women can lead to the incorrect conclusion that treatment has a harmful effect. This is an example of Simpson's paradox, which although uncommon in practice, does, in fact, occur (Wainer H. Simpson's paradox. Chance 1999;12:43). More importantly, it is likely that in an observational study, a related result will occur; namely, ignoring sex in the analysis will lead to biased results. To better understand why Simpson's paradox and the related result occur, we present a graphic explanation.