From protection OR to risk OR (--condition) chromosome X

[Monserrat I. Morales-Rivera]

Dear all,

I ran a conditional analysis with a top variant but the OR changes from protection to risk for variants that actually are more frequent in controls than cases.

Any suggestions?

Thanks in advance

[Chris Chang]

Always post a full .log file when asking for troubleshooting help.

--
You received this message because you are subscribed to the Google Groups "plink2-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to plink2-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/plink2-users/1f5903fd-1015-4939-b7da-65580afa54dfn%40googlegroups.com.

[Monserrat I. Morales-Rivera]

Thank you

Actually is a trouble presented making an analysis in Plink 1.9, I hope to find some help in this group.

The commands that displayed a protection odds ratio were:

--adjust
  --bfile 
  --chr 23
  --ci 0.95
  --covar
  --hwe 
  --logistic hide-covar
  --out 
  --remove 
  --xchr-model 3

2034 out of 113416 variants loaded from .bim file.
1386 people (425 males, 956 females, 5 ambiguous) loaded from .fam.
Ambiguous sex IDs written to LESpMEZ.XchrModel3.hwe.nosex .
1386 phenotype values loaded from .fam.
--remove: 1277 people remaining.
Warning: Ignoring phenotypes of missing-sex samples.  If you don't want those
phenotypes to be ignored, use the --allow-no-sex flag.
Using 1 thread (no multithreaded calculations invoked).
--covar: 10 covariates loaded.
Before main variant filters, 1277 founders and 0 nonfounders present.
Calculating allele frequencies... done.
Warning: 26728 het. haploid genotypes present (see LESpMEZ.XchrModel3.hwe.hh );
many commands treat these as missing.
Total genotyping rate in remaining samples is 0.998375.
--hwe: 7 variants removed due to Hardy-Weinberg exact test.
2027 variants and 1277 people pass filters and QC.
Among remaining phenotypes, 126 are cases and 1151 are controls.
Writing logistic model association results to
LESpMEZ.XchrModel3.hwe.assoc.logistic ... done.
--adjust: Genomic inflation est. lambda (based on median chisq) = 1.3607.
--adjust values (1987 variants) written to
LESpMEZ.XchrModel3.hwe.assoc.logistic.adjusted .

While the conditional analysis was made with the following:

--adjust
  --bfile 
  --chr 23
  --ci 0.95
  --condition rs
  --covar
  --hwe
  --logistic hide-covar
  --out 
  --remove
  --xchr-model 3

Random number seed: 1753818796
193130 MB RAM detected; reserving 96565 MB for main workspace.
2034 out of 113416 variants loaded from .bim file.
1386 people (425 males, 956 females, 5 ambiguous) loaded from .fam.
Ambiguous sex IDs written to LESpMEZ.XchrModel3.hwe.conditional.nosex .
1386 phenotype values loaded from .fam.
--remove: 1277 people remaining.
Warning: Ignoring phenotypes of missing-sex samples.  If you don't want those
phenotypes to be ignored, use the --allow-no-sex flag.
Using 1 thread (no multithreaded calculations invoked).
--covar: 10 covariates loaded.
Before main variant filters, 1277 founders and 0 nonfounders present.
Calculating allele frequencies... done.
Warning: 26728 het. haploid genotypes present (see
LESpMEZ.XchrModel3.hwe.conditional.hh ); many commands treat these as missing.
Total genotyping rate in remaining samples is 0.998375.
--hwe: 7 variants removed due to Hardy-Weinberg exact test.
2027 variants and 1277 people pass filters and QC.
Among remaining phenotypes, 126 are cases and 1151 are controls.
Writing logistic model association results to
LESpMEZ.XchrModel3.hwe.conditional.assoc.logistic ... done.
--adjust: Genomic inflation est. lambda (based on median chisq) = 1.30051.
--adjust values (1985 variants) written to
LESpMEZ.XchrModel3.hwe.conditional.assoc.logistic.adjusted .

The problem is that variants that showed a protection odds ratio in logistic analysis, changed to risk OR in the conditional. 

Any idea of the trouble origin?

[Chris Chang]

Why is it surprising that, particularly for non-statistically-significant results, adding a --condition covariate could flip the sign of a result?  It would be a lot more surprising if that never happened.

If I don't seem to understanding your question correctly, please post a small dataset (could contain just 1-2 variants, and have many samples removed if you want) and a pair of commands I can run, and point out EXACTLY what seems to be problematic about the result.

To view this discussion visit https://groups.google.com/d/msgid/plink2-users/ca29760f-6ede-4537-b2de-1e978dc2d680n%40googlegroups.com.