Analysis of paired data from a cross-over trial, incorporating one or more covariates


Thank you for this wonderful software. I am a clinician-researcher interested in understanding whether I can use Metaboanalyst to find differential features under two conditions in a cross-over clinical trial, but incorporating a covariate.

I have been able to do paired analysis, and I have been able to see how I could use a different input method to incorporate a covariate (but without paired analysis).

Is this a way to incorporate a covariate (or more than one) in a paired-sample study design in Metaboanalyst?


Not sure if it will work for your data. Maybe you can try the linear models with three metadata factors

  • Treatment
  • Pair
  • Covariate

Set Pair as “blocking factor”.

1 Like

Thank you very much. I have tried this approach, and I am finding that 1, -1, 2, -2, etc. pairing approach is not recognized as a valid blocking factor. The system assumes it’s a continuous variable, and changing it to categorical or removing the minus signs to form two-item groups doesn’t seem to work.

Perhaps it requires that I recode the pairing variable somehow in the metadata sheet.


Yes, this is different from time-series. I suppose 4 columns in your metadata:

Sample IDs, Treatment, Subject IDs, Covariate

Pair information is essentially “Subject IDs”.

1 Like