I am doing a meta-analysis to assess the effect of serious games on memory. Some included studies use more than one measure to assess the memory. All measures are relevant to the paper. My question is which measures to use? Is it possible to run a meta-analysis including all measures? if so, how? I am using RevMan to conduct meta-analyses, can I use it to sort out this issue?
I think there are multiple facets to your questions:
- 1 :Which measure to use: this is relevant to the specific field of research you are conducting but basically the best measure is a measure that is meaningful for researcher and end-users of the research. This measure has to be validated (eg robust between different operators (ie intra-inter rater agreement).
- 2 : Can you combine different measure on the same model? My first answer is NO. You can not combine pear and orange in a meta-analysis model with the aim of obtaining meaningful results. If the measures are not equivalent putting them on the same model would definitely give a result that has no practical meaning. running different models (one for each type of outcome reported) would be definitively a safer approach.
- 3: Adding more than one outcome for a specific study. I am not familiar with your research field. However, in my field (veterinary science), we commonly have different experiments in the same paper. What can be done in this case is to account for these different experiment accounting from the fact they come from the same study (see more details on random-effect meta-analysis). However, this approach does not hold where the same patients have been measured repeatedly because in this case adding them as additionnal findings would definitely be more challenging and would require more complex approach (that I'm not familiar with). What is generally done in this case is to take a random set of data among all of the reported effects of the same experiment (that can be challenged later in sensitivity analysis of the study).
-4: revMan. as soon as you go to more complex analyses and models RevMan is not the best freeware to use. RevMan is a good one for building the systematic review protocol, ROB and flowchart but you need to use other software. My preference is for R freeware where different packages for meta-analysis have been reported (eg metafor or meta package but there are others). The Bookdown site of Harrer and al on how to do a meta-analysis with R is a very practical and up to date place to address most of the topics and question you may ask yourself. (https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/).
-5: finding a collaborator who has done meta-analysis before is also very important to help you in this study. This will save you a lot of time and helps in writing the protocol and making complex analyses.
Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal of statistical software, 36(3), 1-48.
Schwarzer, G. (2007). meta: An R package for meta-analysis. R news, 7(3), 40-45.
You cannot combine different measures. If all measures are important, you may do meta-analyses separately for each measures.
Multiple outcome measurements can not be combined in a single forest plot because it will duplicate the participants and create a unit-of-analysis issue. The best approach would be to create a hierarchical list of measures that you have and extract the highest measure on that list from each study. Different measures can be combined using standardized mean difference (if continuous data).
Another approach would be to use a multilevel meta-analysis to account for correlation between the effect sizes. In that case, you can use all measures from each study to combine into one analysis but that is more complex and RevMan can not run that. STATA 17 or R can be used for that instead.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis