I am conducting a systematic review about a community-based intervention to prevent hospitalizations in older adults. We identified 3 studies that reported on the outcome of interest and I am not sure whether I can pool their results in a meta-analysis. The first study reported data on the comparison between community A and community B for year 2006. The second study reported data on the comparison between the same community A and a group of 5 surrounding communities (community B was not included in that comparison) for years 2006, 2007 and 2008. The third study reported data on the comparison between communities C and D for the year 2006.

My understanding is that it would be appropriate to pool the results of the 3 studies for all the years because ultimately they report effect estimates concerning different sets of comparisons, even if community A was part of those comparisons for a specific year in the first and second study. Is that understanding correct? Would someone have any reference supporting or refuting that approach?


The gold standard approach would be to take account of the dependence introduced by having A in two of the primary studies. If you have ample studies comparing A B C D but not all of them together in all of the studies then you would use network meta-analysis (also called multiple treatment comparison). However with three studies set out as you describe that is not going to work. As it stands you will have to proceed cautiously with a section in your discussion about the possible limitation induced by having one group repeated.

  • $\begingroup$ Many thanks for your answer! I have performed a sensitivity analysis excluding study 1, and another analysis excluding study 2. The interpretation of the findings did not change, only the magnitude of the effect estimate. I will read about network meta-analysis. $\endgroup$ – user156625 Apr 18 '17 at 20:43
  • $\begingroup$ Learning about NMA will be a good experience but it will not help you with this data-set as you would need more studies (as I hinted in the answer). $\endgroup$ – mdewey Apr 19 '17 at 8:03
  • $\begingroup$ I have just read a book chapter and 2 articles about network meta-analysis and I am not certain that for the context that I intended it would be helpful even if I had many studies at hand. Network meta-analysis is helpful when comparing multiple treatments (e.g., A, B and C). My problem does not refer to a comparison among different treatments in different study populations, but between the results of the comparisons of the same treatment among different communities, when one of the communities was reported in different comparisons by two different studies. I will try to make another question $\endgroup$ – user156625 Apr 23 '17 at 16:36
  • $\begingroup$ I have made a new question trying to express the context of my problem in a clearer way. The link to the new question is here stats.stackexchange.com/questions/275368/… $\endgroup$ – user156625 Apr 23 '17 at 17:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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