Overall our results suggest that there was little benefit to stratifying depression by either sex or recurrence for currently available data sizes. Extreme differences between sexes, such as opposite directions of effect in the two sexes, would have to exist to necessitate their analysis separately. The power implications of stratifying on these traits is likely to out-weigh the identification of such loci. In situations where the effect of a SNP is only found in one sex and zero effect in the other, such as rs4778037 in this study, it is still better to analyze sexes together to reduce the multiple testing burden of separate analyses. The increased trait variance explained demonstrated by using the largest available training and discovery datasets (PGC MDD29 and UKB MDD, respectively) in polygenic profiling supports increasing the sample size over phenotypic refinement. Similarly, the lack of discernible difference between h2SNP and rG estimates between depression subtypes suggests that the best approach, currently, is to maximize the sample size in order to reduce sampling error and obtain more accurate point estimates. However, addressing recurrence, sex and
