Meta-Analysis effect size is not in the same direction as GWAS effect sizes.

Question

[lamia_203]

Hi, I am conducting a meta-analysis and the output doesn't make sense. I ran gwas on several different treatments and some treatments work well, other do not. The effect size for the 3 gwas results are 0.016,0.012 and 0.011 with respective SE of 0.005,0.005,0.004.

However the effect size for the meta-analysis is -0.013 and SE of 0.0025. The effect size of the meta-analysis is of a complete different direction than the gwas's. I ran the gwas using plink and the meta-analysis using METAL. I used the weights effect size estimates using the inverse of the corresponding standard errors. So if the SE are small and all three studies are of similar range, I do not understand why the meta-analysis effect size completely opposite direction. Any help would be much appreciated many thanks!

[Kevin Blighe]

Are these even statistically significant?

[lamia_203]

Yes, I've pulled the most significant SNP from the meta-analysis then used that SNP to select from the other studies. I've managed to sort it out - the allele direction was flipped hence the correct absolute effect size but incorrect direction. I'm not sure how that has happened to some files and not all files...

Answer 1

[Lemire]

I think it's just your interpretation of the output that is incorrect, with respect to which allele the effect size is based on. Read and re-read the documentation, it can be confusing sometimes.

With regards to your meta-analysis estimates, they seem correct (albeit the sign), so it's not a software issue, it's an interpretation issue.

Using R:

library(metafor)
y<- c( 0.016,0.012, 0.011) ; s <- c( 0.005,0.005,0.004) 
rma( yi=y, sei=s )

Random-Effects Model (k = 3; tau^2 estimator: REML)

tau^2 (estimated amount of total heterogeneity): 0 (SE = 0.0000)
tau (square root of estimated tau^2 value):      0
I^2 (total heterogeneity / total variability):   0.00%
H^2 (total variability / sampling variability):  1.00

Test for Heterogeneity: 
Q(df = 2) = 0.6358, p-val = 0.7277

Model Results:

estimate      se    zval    pval   ci.lb   ci.ub     
  0.0127  0.0026  4.7882  <.0001  0.0075  0.0179  ***

---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Same estimates, up to rounding errors.

[lamia_203]

Thank your for the response and confirming the estimate values in R. I've looked at the meta analysis alleles and they are flipped. Not sure how that has occurred but it makes sense why your estimate value is the same as mine but the sign is the opposite. Many thanks for your help!