I am conducting a meta-analysis of continuous data using RevMan 5.4.
Included studies express their results as mean, SD at baseline and end of study for intervention and control arms. With these data I can impute change from baseline which I will use to perform a meta-analysis of change scores.
However, in a few studies, due to patients lost to follow-up, the number of patients at the end of the trial is lower than at baseline. RevMan 5.4 requires mean change from baseline, SD of change and sample size to perform the meta-analysis. Which number of patients should I use to perform the meta-analysis (the sample size at baseline or the follow-up)? Or would it be better to exclude these studies?
Thanks in advance for your help.
To get a true comparison, it would be best to exclude the cases missing in the followups from the baseline and recalculate the baseline statistics for comparison.
Another possibility would be to try to get an estimate using cox regression. Repeat the search on the attachment for references. Best wishes David Booth
PS Both cases would seem to require access to the original data. Best wishes David Booth
Studies with bigger sample sizes should have higher weights, that's the only logical thought I can imagine here.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size