Correcting for batch effect in survival data

Question

[english.server]

Hi!

I want to calculate cox coefficient for miRNAs in TCGA data. To check the presence of batch effects, I used Mbatch database. I am now wondering should I apply (a) batch correction in the cox formula (b) remove batch effect from expression data and proceed without covariates in cox formula (c) ignore batch effects!

Answer 1

[Kevin Blighe]

Your assumption is that there are batch effects in your miRNA expression data, and you have confirmed that these batch effects exist and that their effects are statistically significantly probable to bias any statistical inferences made on the data.

If this is the case, then please include batch as a covariate in the design formula, as per any regression model.

[the thorough analyst will of course try both strategies, the other being to directly remove the batch effect and then not include batch as covariate]

Kevin