2

I am conducting a meta-analysis in which some of the studies compare two or more interventions to the same control. The Cochrane Handbook recommends:

  1. Combine groups to create a single pair-wise comparison (recommended).
  2. Select one pair of interventions and exclude the others.
  3. Split the ‘shared’ group into two or more groups with smaller sample size, and include two or more (reasonably independent) comparisons.
  4. Include two or more correlated comparisons and account for the correlation.

The third option is the only feasible one. How do I calculate the mean and SD of the split control group?

Giuseppe Biondi-Zoccai
  • 2,244
  • 3
  • 19
  • 48
Eimear
  • 21
  • 3

3 Answers3

1

Definitely you need to go beyond the traditional paradigm of pairwise meta-analysis, and instead conduct a network meta-analysis (in this case an adjusted indirect comparison given the star-shaped evidence network).

You can find ample references on this, such as:

  1. a recent question on CrossValidated;
  2. a comprehensive website;
  3. the Cochrane toolkit;
  4. a recent book;
  5. an upcoming book.

You can easily do network meta-analysis with R, Stata, or WinBUGS. I recommend you to use R, for instance the netmeta package.

Community
  • 1
Giuseppe Biondi-Zoccai
  • 2,244
  • 3
  • 19
  • 48
1

If you split the control group you end up with estimates which are not independent. If there is only one study out of many this is probably not an issue but if it occurs often in your data-set then you need to fit a multi-level model with study as a random effect. This can be done in R using metafor There is an example here.

mdewey
  • 16,541
  • 22
  • 30
  • 57
  • 1
    And also see: http://www.metafor-project.org/doku.php/analyses:gleser2009 This explains how to compute the covariance of the sampling errors when multiple interventions groups are compared against a common control group. – Wolfgang Aug 07 '16 at 19:10
0

While a network meta-analysis is one way to go. The simpler way is to use the recommendations of the Cochrane Handbook. If you're down the route of splitting the comparator group then divide the 'N' by half. That way you give half the weight of the comparator to each comparison.

I made a mock meta-analysis below whereby study 2 is a three arm study.

enter image description here

abousetta
  • 1,030
  • 5
  • 9
  • 1
    Yes this is exactly what I want to do! But is it okay to assume the same mean and SD for both halves of the split control group? – Eimear Aug 08 '16 at 17:12
  • Generally speaking, it's not the best option, but it is acceptable. Also, any assumption can be tested in a sensitivity analysis. – abousetta Aug 09 '16 at 01:02