It turns out that most association studies are probably incorrect, but a fraction of reported studies probably represent true associations with genetic variants that have modest effects on disease risk. These variants require large study sizes for associations to be reliably detected, explaining the lack of reproducibility by underpowered studies. Furthermore, the fact that some associations are likely correct means that common genetic variants are likely to contribute to common disease susceptibility.

The paper shows that published reports linking variation in DNA sequence with the risk of disease should initially be viewed with skepticism. But, if the link between a DNA sequence variant and disease can be demonstrated in repeated studies—particularly studies with lots of patients—the link is more likely to be correct. Finally, we show that some DNA variants that are common in the population are likely to affect the risk of disease.

I actually started reviewing the association study literature at the suggestion of my father, who is also in human genetics, and was asked to write a review article on this topic. If I had known how vast the association literature was, I never would have agreed to do the review, but fortunately some talented students helped, and three years later we completed the initial review, published in Genetics in Medicine. We noticed in that review that many association studies were replicated on occasion, but very few were consistently replicated, so one of the students (Kirk Lohmueller) decided to start the meta-analyses described in this paper to figure out the reasons behind that inconsistency.