Should We Include Gender, Age And Other Known Covariates In Genome-Wide Association Analysis Of Rare Disorders?

Question

[kapoormaan]

I just came cross to paper published in nature genetics , "Including known covariates can reduce power to detect genetic effect in case control studies". http://www.ncbi.nlm.nih.gov/pubmed/22820511

I am not expert in statistics so I am not able to fully grab the results from this paper. I use to think that if there is some relationship among primary phenotype and the covariate and suspected relation between SNP and covariate then it should be included in the model. It would be great if some one can clarify the use of covariate in simple biological terms.

Thanks Maan

[zx8754]

The referenced paper also includes R-functions to assess the effect of a single binary or continuous covariate on power for detecting genetic variants.

Answer 1

[dstram52]

The answer is complicated because the intent of the analysis has effects; Consider controlling for obesity (BMI) in an analysis of type 2 diabetes risk. if you are interested in the independent contribution to diabetes risk not mediated through obesity then put BMI or other obesity variables in the model. If you are just interested in prediction of diabetes without knowledge of obesity then keep them out.

Another purpose, even if the covariate variables are independent of the SNP, of controlling for strong risk factors (strong covariates) is just to reduce the residual variance of the outcome and therefore hopefully improve power.

I haven't looked at the paper but perhaps the authors are saying that the latter is only important if the risk factors are quite strong.