If we are running a meta-analysis of sub groups, we will have effect size estimate and pvalues for each sub-group. Using these p values directly is not reccomened. Adjsuting the p values is a more conservative approach. Let's say we have two groups (duration of intervention and tailoring). Duration of intervention is
Let's say we have two groups (duration of intervention and tailoring). Duration of intervention is classified into three subgroups (3 months, 9 months, and 12 months). Tailoring of the interventions is subgrouped as tailored/not tailored. Two separate sub group meta-analysis was run for the two main subgroups providing effect size estimates, p values and hetrogeneity statistics.
How do we then adjust the resulted p values? Do we adjust the pvalues from the total group(in this case 2) or from each subcategory of the sub group (in our example 3, and 2)? Thanks.
I don't know if I understand your question. How many subgroups are you wanting to analyze at one time? It sounds like you are wanting to use the p values associated with each subgroup. Those are generally not very useful. They don't tell you if the subgroups are different from one another. They only tell you if a subgroup's effect size is statistically different from zero (or one).The preferred way to run the subgroup analysis is with all subgroups included and you examine the between-groups p-value. If the between-groups p-value is less than 0.05, then you run pairwise comparisons (again using the between-groups p-value) for all possible subgroup pairs while adjusting the alpha level using a Bonferroni correction, Benjamini-Hochberg false-discovery rate correction, etc. This will tell you which subgroups are different from one another. This is analogous to running a one-way ANOVA with 3 or more groups.
Dear Professor Warren,
Thanks for this helpful answer. Yes, I was doing sub group analysis one-by-one. As you pointed out, it only tells whether the effect sizes are statistically different from the null value. I would like to include all the subgroups in the analysis. However, how do I run multiple subgroups in one subgroup analysis using STATA? The by(var) function of STATA allows me only to add one subgroup at once. How do I enter multiple subgroups? Thank you very much.
Dear Mihiretu
You can examine multiple meta-analysis sub-groups in STATA - in the data sheet, add a variable name the in top, such as 'dosage' and then add text next to each author name / treatment arm, such as the 3 sub-grpups 'high', 'medium' and 'low' as in the example below.
In the data sheet:
Author nInt meanInt sdInt nCon meanCon sdCon dosage
Mihiretu2017 40 31 12 40 10 41 high
Warren2016 30 45 14 30 19 46 medium
Stausholm2013 20 45 18 20 15 41 low
then run a command like: metan nInt meanInt sdInt nCon meanCon sdCon, by(dosage)
Dear Martin @Martin,
Thanks for your response. STATA allows us only to do one subgroup at a time such as dosage(high, medium, low) as you mentioned. By multiple subgroups analysis, I mean performing the meta-analysis for multiple groups at a time. For example by(dosage, sex, age category,...). There is one article from the New England Journal of Medicine which used that kind of analysis. However, they have done IPD meta-analysis. I only have summary data from each study. With that kind of subgroup analysis, it is possible to adjust the subgroup interaction effect and apply Bonferroni correction to downgrade the cut-off point for significance level. I have extensively read about it, I think with STATA it is impossible or complicated (probably nobody has done it yet). I am not sure whether it is possible to do this with CMA (Comprehensive Meta-analysis).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis