Adding additional covariates to plink logistic regression model on HLA returns p-vals etc as NA

Question

[4galaxy77]

I am running a logistic regression association test on the imputed HLA genotypes, using TB as a phenotype, using the HLA-tapas wrapper around plink v1.90.

When I use sex,age,age2,PC1 as covariates, the model returns results:

 CHR                                  SNP         BP   A1       TEST    NMISS         OR       SE      L95      U95         STAT            P
   6                            rs1529749   28000361    C        ADD    55686     0.9457   0.2215   0.6126     1.46      -0.2518       0.8012
   6                           rs17765491   28000610    A        ADD    55686     0.9614   0.2217   0.6226    1.485      -0.1775       0.8591
   6                           rs17708949   28001003    C        ADD    55686     0.9523   0.2264   0.6111    1.484      -0.2159       0.8291

However, when I include additional PCs, sex,age,age2,PC1,PC2,PC3,PC4,PC5,PC6 the output is:

 CHR                                  SNP         BP   A1       TEST    NMISS         OR       SE      L95      U95         STAT            P
   6                            rs1529749   28000361    C        ADD    55686         NA       NA       NA       NA           NA           NA
   6                           rs17765491   28000610    A        ADD    55686         NA       NA       NA       NA           NA           NA
   6                           rs17708949   28001003    C        ADD    55686         NA       NA       NA       NA           NA           NA

That's the only thing I changed between the runs. What might be causing this? Please let me know if you need any additional info.

plink output for 'working' run

PLINK v1.90b3w 64-bit (3 Sep 2015)
Options in effect:
  --allow-no-sex
  --ci 0.95
  --covar ep_aug_2020.txt
  --covar-name sex,age,age2,PC1
  --double-id
  --logistic hide-covar
  --out output/a15
  --pheno ep_aug_2020.txt
  --pheno-name a15
  --vcf chr6.dose.G_group.filtered.vcf.gz

Random number seed: 1651677404
1031781 MB RAM detected; reserving 515890 MB for main workspace.
--vcf: output/a15-temporary.bed + output/a15-temporary.bim +
output/a15-temporary.fam written.
48849 variants loaded from .bim file.
76311 people (0 males, 0 females, 76311 ambiguous) loaded from .fam.
Ambiguous sex IDs written to output/a15.nosex .
55686 phenotype values present after --pheno.
Using 1 thread (no multithreaded calculations invoked.
--covar: 4 out of 363 covariates loaded.
9 people had missing value(s).
Before main variant filters, 76311 founders and 0 nonfounders present.
Calculating allele frequencies... done.
48849 variants and 76311 people pass filters and QC.
Among remaining phenotypes, 135 are cases and 55551 are controls.  (20625
phenotypes are missing.)
Writing logistic model association results to output/a15.assoc.logistic ...
done.

plink output of run that produces NAs:

PLINK v1.90b3w 64-bit (3 Sep 2015)
Options in effect:
  --allow-no-sex
  --ci 0.95
  --covar ep_aug_2020.txt
  --covar-name sex,age,age2,PC1,PC2,PC3,PC4,PC5,PC6
  --double-id
  --logistic hide-covar
  --out output/a15_666666PC
  --pheno ep_aug_2020.txt
  --pheno-name a15
  --vcf chr6.dose.G_group.filtered.vcf.gz

Random number seed: 1651677868
1031781 MB RAM detected; reserving 515890 MB for main workspace.
--vcf: output/a15_666666PC-temporary.bed + output/a15_666666PC-temporary.bim +
output/a15_666666PC-temporary.fam written.
48849 variants loaded from .bim file.
76311 people (0 males, 0 females, 76311 ambiguous) loaded from .fam.
Ambiguous sex IDs written to output/a15_666666PC.nosex .
55686 phenotype values present after --pheno.
Using 1 thread (no multithreaded calculations invoked.
--covar: 9 out of 363 covariates loaded.
9 people had missing value(s).
Before main variant filters, 76311 founders and 0 nonfounders present.
Calculating allele frequencies... done.
48849 variants and 76311 people pass filters and QC.
Among remaining phenotypes, 135 are cases and 55551 are controls.  (20625
phenotypes are missing.)
Writing logistic model association results to
output/a15_666666PC.assoc.logistic ... done.

End time: Wed May  4 16:30:06 2022

[shijxovo]

I also encountered the same problem. How did you solve it

Answer 1

[chrchang523]

Rerun this with plink 2.0: its logistic regression adds an ERRCODE column which indicates a reason for each NA result.

[4galaxy77]

Thanks, it was due to the variance inflation factor of age being too high.