I am developing network meta-analysis BUGS model that I will use to compare Hazard Ratios in studies that assume Box Proportional Hazard model.

As input I am given pairs: (log HR, its standard error) that measure magnitude of the effect of difference between two treatments. Naturally I assume normal distribution model for the log HR ("LHR" for short) with its given SE.

Problem is, when I have more than one LHR result from a single study: sometimes these results are based on independent groups - and then I treat them just as if they came from separate studies.

But more often than not these results are based on multiple comparisons in more than two-arms experiment on the same group. It means, that they have e.g. 4 groups of patients: A, B, C and d and report Hazard Ratios A vs B, A vs C, A vd D, B vs C, B vs D and C vs D. These 6 results are not-so-independent, so I can't really treat them as independent in my BUGS model.

So I believe that treating them as separate studies is wrong, but what else can I do?

If that is relevant for you, I can post actual BUGS model.


1 Answer 1


I hope that you find this tutorial helpful (http://www.biomedcentral.com/content/pdf/1471-2288-10-54.pdf). It describes the use of a network meta-analysis for accounting for the lack of independence in multi-arm trials. The field is growing quickly and I expect to see a lot of new work on this area in the upcoming years.

  • $\begingroup$ Yes, this article is very much to the point and I've already read this article. I should have posted it as my own answer, but apparently I forgot. So all the credit goes to you! And thank you. $\endgroup$ Commented May 1, 2013 at 10:52
  • $\begingroup$ Thanks for the feedback, and glad I was able to help (even if confirming what you already knew :) ). Good luck. $\endgroup$
    – abousetta
    Commented May 1, 2013 at 14:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.