Currently I am working on a systematic review with meta-analysis. I will be writing review about the neuroprotective efficacy on a drug. From the extracted papers I collected infarct volume and neurobehavioral score. But now I am facing a problem, some of the results are in mean +/- SD format (7.24 ± 1.04 versus 12.59 ± 1.44mm3) and some are in percentage improvement (47.07 +/- 7.63% versus 26.27 +/- 5.63%). Is there any way I can combine these data to find out effect size and plot a graph? If yes, could you suggest a method or software?
I'm not sure about your target results but when you would like to combine the data you should subtract data to mean and S.D. only. I'm not sure about what is percentage used for(such as 95% CI coverage or just minimum and maximum=>you should clarify it by reading the paper carefully. I recommend you to use STATA or any software you are familiar with.
Hi. I would try to analyze the data in both ways (mean + SD and percentage changes) as I believe that both are relevant decision makers. In order to get those data I would contact the author(s) and ask for the missing information. They have the data and should make it available to you.
There is three options to deal with this matter:
1. You can use Standardized mean difference (SMD) to put together measurement in one forest plot.
2. analyse separately the measures of mean and SD in one forest plot (weighted mean difference WMD) and the percentage in another forest plot (WMD)
3. convert the change of the mean and SD into percentage or the reverse if you have the baseline measurement convert the percentage change into mean and SD
I hope that this reply will help
kind regards
Dear Charles:
I agree with Ammar about using SMD. REVMAN can do this for free.
Cheers,
J
You can pool the effect as SMD (Standardized Mean Difference) in RevMan (http://tech.cochrane.org/revman/download ).
You will pool (Mean,SD,N) for both groups. They will be represented as SMD (Mean/SD), So the % will be omitted from the effect estimate.
Using SMD is better especially if the effect is measured on different scales.
For the Meta analysis part you may use RevMan.. The software is easy to use
Thank you very much for all your answers. I did contact the authors for raw data. If it wont work I'll go for the methods you people suggested. Thank you guys. cheers.
is there provision for entering non interventional data like prevalence data in Rev Man, I cant find it
I agree with Dr. Knittle and recommend the text of Sharon-Lise T Norman "meta-analysis: Formulating, Evaluating, combining, and reporting" published in Stat Med. 1999 and the software RevMan.
I agree but would also recommend having a look at the fantastic meta-analysis package metafor for R, which includes a function to calculate effect sizes from different indicators (escalc()
Please, see http://www.jstatsoft.org/v30/i07/ (where transformation process are automated and then you can add to RevMan as continuous and analyse there).
The latest version is avalable at
http://www.statanalysis.co.uk/meta-analysis.html
MetaEasy software in which prior transformation to a common metric is not necessary.
Please, see more details in Kontopantelis & Reeves, 2009.
Thank you.
I too have used comprehensive meta-analysis, and I think its the best and easiest software to work with.
- Badges
- Science topic