I am conducting a meta-analysis on the effect of compression garments on exercise performance. The included studies have used a variety of methods to measure endurance performance. This means that for some outcome measures, a larger value would indicate a positive effect of the intervention (i.e. longer time to exhaustion = improved performance), whilst for others, a lower value would demonstrate a positive effect (i.e. shorter 5k time trial time = better performance).
I have struggled to find guidance on how to account for this when using the Cochrane RevMan software to conduct the analysis. I wondered if anyone else had encountered something similar, and knows how overcome the issue?
The included data for meta should be consistent, otherwise, the outcomes were skeptical.
https://training.cochrane.org/handbook/current/chapter-10
Lauren Rhodes
That is, if it's possible to convert study results into the same measures, do so. For example, if some studies had percent mortality and others percent survival (as a proportion of a group), you could easily convert them all into one or the other and then do the analysis. It doesn't sound like your studies will convert easily to a common measure, so you might have to split them into two analyses.
First of all, you have to convert the outcome measures in a way that they all go in the same direction (higher value stands for better outcome or viceversa): to do so, you have to multiply by -1, so simply including a minus (or removing it, if it is already there) before the mean value (assuming you have means and standard deviations) of those outcome measures going in the "other direction".
Then, you can combine them using the standardized mean difference (SMD), but be careful to assess that all the different outcome measures are for the same outcome and the comparisons (intervention vs control) are the same.
Hope this could help.
Thank you all for your responses! Based off the answers, I think it would be possible for me to convert my the values for the time trial (time to complete a set distance measured in minutes), into a speed (using the speed = distance/time equation). That way the higher value will indicate better performance. This will then show improvement in the same direction as the results for time to exhaustion trials.
Very appreciative of your comments! Thank you again!
Hi Lauren,.
I've only just come across this question, but yes you response above is the correct approach - convert all performance times to speeds (or alternatively, power outputs) and then you have all your "improvements" heading in the same direction.
Cheers.
Ted.
Ted Polglaze Thank you for your response - good to have clarification and reassurance on that!
@ Stefano Giuseppe Lazzarini,
I have exactly the same issue in meta-analysis using Revman 5.4 showing the diamond box in the opposite direction. I am analyzing students' scores. Pretest and posttest score analysis is not recommended. As this conflict exists, I intentionally use pretest and posttest scores because I know posttest is definitely better. But the output result showed pretest is more effective. I already imagined the outcome of the analysis that posttest scores are more effective. When I multiply the data by -1 as suggested by you, now the diamond box moved to the posttest score. It solves in this particular case. I am also doing analysis for other data. In this new case, the Experimental group scores are better than the Control group in most cases. My question here is, should I multiply by -1 or not. How do I decide to multiply by -1 in some cases and not to multiply in some cases?
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis