Dear RG Members,
When using meta-analysis based on a fixed effects model in which the summary statistic is an odds ratio for a binary outcome, how does one calculate statistical power for obtaining evidence in favor of a significant treatment effect? What could be regarded as a reasonable number of studies and subjects for having sufficient power when combining data from different studies?
-David
I recommend that you read Chapter 29 in the book, Introduction to Meta-Analysis, by Borenstein, Hedges, Higgins, and Rothstein.
Dear sir,
I can't understand your question properly but i'll respond as i understand,
Every study in the Meta-analysis has "Effect size" according to the effect of Drug A over drug B for example. But every study also have a specific weight in the meta-analysis which is calculated according to the statistical method like "mantel haenszel or inverse variance" this weight increases with the study sample because as you increase the sample size you approaching the population values.
The overall effect size measured according to these two main value, the effect size for every study in the MA and its weight
And if you are asking about how to calculate it this is very easy to use RevMan or comprehensive meta-analysis software just by importing the data of every study you will get your overall effect size
Hi
You can do it according to this methods.
First you must calculate the weighted proportion of outcome or exposure (in regard to the study design) in group A and B (it is pA and pB)
2- calculate the ratio of nA / nB=k
3- calculate the Z according to the below formula:
x=log(pooled OR)*sqrt(nB)/sqrt(1/(k*pA*(1-pA))+1/(pB*(1-pB)))
power= pnorm( x - z(1-alpha/2)+ pnorm ( -x - z(1-alpha/2)
pnorm= standard Normal distribution function
My view is that the purpose of meta-analysis is to move away from p value hypothesis testing and embrace magnitude and direction of effect instead. If this is the case then power and P values etc should not be relevant for meta-analysis
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis