I have 5 studies each with a range of treatments - some of which do not overlap with other studies. So, for example:

Study 1: tx 1 tx 2

Study 2: tx 1 tx 3

Study 3: tx 1 tx 2 tx 3

I was told that five studies is not enough power to conduct a network meta analysis. So, my question is how do I determine how many studies I need to have enough power to conduct a NMA?

  • 1
    $\begingroup$ If you want to calculate the precise power you need to define the effect size of interest. However, my perspective is that you simply need at least 3 studies on a total of at least 3 alternative treatments linked in some way (star or closed-loop) to perform a NMA. Of course, the more the studies, the merrier. But even a NMA of three large and precise studies could be of interest. $\endgroup$ Commented Jun 13, 2018 at 11:50
  • 1
    $\begingroup$ Great, thanks. This is helpful! Do you know of any citations/examples that have used NMA in a few number of studies? $\endgroup$
    – sahanm
    Commented Jun 13, 2018 at 12:38
  • $\begingroup$ Of course: ncbi.nlm.nih.gov/pubmed/20828843 $\endgroup$ Commented Jun 13, 2018 at 17:14
  • $\begingroup$ A problem I see with few studies is that heterogeneity and inconsistency may become undistinguishable or nearly so. In particular, if you have only one study per comparison, you can't estimate heterogeneity, and my understanding is that all the variability in results across studies is interpreted as inconsistency. $\endgroup$ Commented Nov 22, 2022 at 21:04


Your Answer

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

Browse other questions tagged or ask your own question.