2

I'm a bit confused about a meta-analysis performed with PLINK on two studies.

The statistics look like this for a specific SNP in each study:

Study_1: OR = 3.657, SE = 0.336, p= 0.0001137
Study_2: OR = 9.08E-009, SE= 8363.831118, p= 0.998

Using PLINK you get following results (randomized fixed effects):

Combined p = 0.0001139
Combined OR = 3.657

My question is, how is it possible that this specific SNP still got a low meta-analysis p value < 0.05 considering that in study 2 (ID2) the SNP got very bad statistics, high p, low OR and very high SE?

eXpander
  • 524
  • 1
  • 6
  • 15
  • How many individuals / non-missing alleles were in studies 1 and 2 respectively? It seems that you may have a lot of missing data in study 2, which is why it could potentially only nudge the $P$-value and OR. Also, whats the MAF of the SNP? – Chris C Nov 04 '15 at 12:31
  • @ChrisC: Because I do not have access to study 2 details I cannot give information about the SNP missingness. 1000 samples were analyzed. The MAF is higher than 10 %. The SNP callrate in study 1 is approximately 98%. – eXpander Nov 04 '15 at 12:47
  • 1
    Interesting. Is it an additive model, or did you do dom/rec? It seems (to me) that study 2 may have very few samples, leading to a ridiculous OR/SE and $P$. If it's a dominant model, you could just have very few AA with an MAF of 10%, but that really shouldn't be an issue. Regardless of speculation, study 2 seems to be almost non-informative as it barely moved the $P$ or the $OR$, so something is definitely happening to basically invalidate it. I would email the authors of study 2 and ask if they could provide raw data or at least genotypic counts for study 2. – Chris C Nov 04 '15 at 12:55
  • Can you clarify how PLINK does this? Does it combine the OR and then get a p-value or does it combine the p-values separately? – mdewey Apr 24 '17 at 13:02

0 Answers0