Hi all

I am currently analysing some lab data from a RCT. The primary outcome is a binary variable and a composite of two secondary outcomes that are also binary. These two secondary outcomes must both be affiliative for the primary outcome to be marked as positive, if one or both of these secondary outcomes are negative then the primary outcome is classed as negative, and if one or both of the secondary outcomes are missing then the primary outcome is set to missing. I have analysed the primary outcome and the two secondary outcomes using a log-Poisson generalised linear mixed model. First I adjusted for randomised group as a fixed effect, and site as a random effect to obtain unadjusted relative risks. I then adjusted for randomised group, and stratification variables such as gender as a fixed effect, and site as a random effect to obtain adjusted relative risk. Site was adjusted for a random effect in both.

My predicament is this, the adjusted confidential interval is pretty much the same, just slightly narrower than the corresponding unadjusted confidential interval for the primary outcome and one of the secondary outcomes, which was expected, however the adjusted confidential interval for the other secondary outcome is much wider than the corresponding unadjusted confidential interval. The significance does not change.

I have been asked to discuss the reason the adjusted confidential interval is much wider than the unadjusted confidential interval for the secondary outcome and not for the primary outcome or the other secondary outcome. I was wondering the reason for this and what this means in a statistical sense.

Thanks