I am looking to do a meta analysis on intervention RCTs but all of the papers have provided baseline and post intervention Mean (SD) for the groups. I have looked at the Cochrane page as I am using Revman ( https://handbook-5-1.cochrane.org/index.htm#chapter_16/16_1_3_2_imputing_standard_deviations_for_changes_from_baseline.htm )
However, I am not from a statistical background so this is getting quite complex. Just wanted to see if there are any resources to guide as this must be very common, and I would hope there are fairly simple ways to deal with it. I have come across using the SD from baseline or post-intervention as the SD for the change but obviously am hesitant of just going along with this.
As only baseline and post-intervention Mean and SD are reported for the majority of the studies, I am thinking I may need to leave it at the narrative synthesis and leave out the meta analysis. But given that the trials are all randomised with similar baseline characteristics and biomarker parameters, could I just enter the final measurements in both groups, rather than the mean change and SD from baseline?
Appreciate of any input.
This is a important problem when we calculate the changes from baseline to endpoint. If you want to calculate the SD of changes, a R score is needed. In the Cochrane handbook, the R score of 0.5 is recommended. There are several data transformation templets for meta-analysis. You can search it in internet. The formula of SD for changes is already embedded, and you just need to input the mean and SD of baseline and endpoint.
I hope this answer could help you.
Dear Qasim,
I think you should go on with meta-analyses. I see 3 options to carry them out:
1) You can calculate the correlation coefficient to impute the change from baseline SDs. The correlation coefficient can be calculated from a study that is reported in considerable detail or by reasoned argument. If you have detailed information (baseline, final and change from baseline SDs) from some of the studies, I would suggest you to go on with the correlation coefficient calculation and then using this coefficient to calculate the SDs of the mean change. I´m attaching a sheet we used in a similar meta-analysis (see it here: Article Effects of oral supplementation with probiotics or synbiotic...
)2) If you do not have information on baseline, final and change from baseline SDs from any study, you can argue about a reasonable value of correlation coefficient between –1 and 1 (to next calculate the SDs of the mean changes). A correlation coefficient of 1 indicates that there is no variation on the baseline and final measurements across participants, which is generally not observed. So I suggest you to choose a value lower than 1 (a more conservative value).
3) If the included RCTs are at low risk of bias on selection and attrition domains and you are confident that the groups are well balanced, you can go on with meta-analyses comparing only final means (and final SDs) between groups.
Good luck!
-Erica
Erica Aranha Suzumura Thank you for your time taken to provide a comprehensive answer, this is very useful. Are final means and final SDs commonly used between groups if they do not violate bias in the domains you mentioned?
Han Qi Thank-you, I did see mention of using an R score, and am not aware that templates for formulae can be used in RevMan
The cochrane handbook does say that imputation should be used sparingly and I am hesitant to do this, and given the nature of randomised controlled trials, the cochrane handbook also states "An alternative to these methods is simply to use a comparison of final measurements, which in a randomized trial in theory estimates the same quantity as the comparison of changes from baseline." I think I am drawn towards using the final means (and final SDs) across all studies as this is provided in all the studies, especially as the number of trials in the review are in the 20s.
I also found this page useful: https://handbook-5-1.cochrane.org/chapter_9/9_4_5_2_meta_analysis_of_change_scores.htm
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis