Meta-analysis P-value decreases, despite same direction and marginally significant in both populations

[dylkot]

Hi Han,

Thank you so much again for all of your help in getting GMMAT to work so well for my project. I had a quick question about meta-analysis. I am combining P-values from 2 cohorts and getting the following results:

I was a bit surprised to see the meta-analysis P-value decrease despite the association being significant in both cohorts (genome-wide in cohort1, marginally in cohort 2), and the effects pointing in the same direction. It does seem to be computing the aggregated score and var appropriately (by adding them). But I was just wondering if this is surprising to you or if there is anything obvious I am doing wrong. This is the command I ran for meta-analysis:

glmm.score.meta(files = c(in1, in2), outfile = outfn, SNP = c('SNP', 'SNP'), A2 = c('REF', 'REF'), A1 = c('ALT', 'ALT'))

Thank you so much again! Dylan

[hanchenphd]

Hi Dylan,

You mean the meta-analysis p-value increased (compared to the discovery cohort results), correct? I don't think this is surprising. If you do a Wald test on both cohorts and run an inverse-variance weighted fixed-effects meta-analysis, you will probably see the same.

Best, Han

[dylkot]

Hi Han,

Yes, exactly, the p-value increased (became less significant). OK thanks for confirming. I found it confusing that even though the association was significant in Cohort2 and pointed in the same direction as the initial association, the meta-analysis P-value still became less significant. I would have thought it would become more significant because it is only 5% likely to observe the independent association by chance. I have already run the Wald test to obtain odds ratios so can try the inverse-variance weighted fixed-effects meta-analysis.

Thanks again for your help!