Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Questions about output when running meta-analysis in STATA

    Hi all,

    I have been working on a meta-analysis for the studies in my area of research. I have now got to the point of testing for differences between different subgroups that may have an effect on the HR.

    Now, when I run the subgroup analysis, the output includes a heterogeneity summary as below (I have excluded the output of the subgroup analysis here, although for reference the two HRs derived were 1.01 (0.92-1.11) and 0.70 (0.45-1.11):


    Heterogeneity summary
    -----------------------------------------------------------------------------
    Group | df Q P > Q tau2 % I2 H2
    ---------------+-------------------------------------------------------------
    No | 13 44.66 0.000 0.01 53.75 2.16
    Yes | 1 1.79 0.181 0.05 44.23 1.79
    ---------------+-------------------------------------------------------------
    Overall | 15 49.37 0.000 0.02 69.62 3.29
    -----------------------------------------------------------------------------
    Test of group differences: Q_b = chi2(1) = 2.33 Prob > Q_b = 0.127


    My question is the following-does the p-value here (0.127) simply relate to the heterogeneity summary, or is it formally testing for differences between the two groups I am interested in examining? I was under the impression that to formally test for subgroup differences in a meta-analysis, we should be running a meta-regression (which I've done as well), although the p-value derived from a meta-regression differs slightly from the p-value derived here in the heterogeneity summary (0.075). What method should I be using to formally test for differences between groups in a meta-analysis?

    Warm regards, Oliver

  • #2
    Can anyone please help me with my question? Thanks!

    Comment


    • #3
      The test of group differences shown in the output is equivalent to a fixed-effect meta-regression model with a binary moderator.

      If you fitted a random-effects meta-regression, then the results will not be similar, unless the estimated tau^2 is close to zero. Both approaches have different assumptions. If you think that within subgroups there is no robust evidence for heterogeneity, then the simple test of group difference is OK. Otherwise, use the random-effects meta-regression model.

      Comment

      Working...
      X