Suppose I am meta-analyzing proportions using their logit transformations. We did a meta-regression of the covariates age, gender, and disease severity score in studies and get respective beta coefficients and intercept with corresponding 95% CI. Then how would I convert this equation to say pooled proportion adjusted for age, gender, and severity score is ....xyz??
Nipun Verma Assuming that you have you have obtained the results for k studies, labeled i = 1,…,k. Each study consists of a treatment group (T) and a control group (C). One of these counts will be zero which might cause some calculation problems. In order to avoid this, add a small value of 0.5 or 0.25 to all counts so that zero counts do not occur. Attached a paper for clarity on the process.
Good luck.
Nipun Verma Assuming that you have you have obtained the results for k studies, labeled i = 1,…,k. Each study consists of a treatment group (T) and a control group (C). One of these counts will be zero which might cause some calculation problems. In order to avoid this, add a small value of 0.5 or 0.25 to all counts so that zero counts do not occur. Attached a paper for clarity on the process.
Good luck.
Thanks sir. i know this fact of continuity correction.
my question is different.
i have a pooled estimate of a proportion of mortality for a disease without a comparator
now I am doing meta regression to what factors can causes variance in estimates, which I can get using meta regression
But now I want to say that after adjusting for a variable let’s say age gender etc. the pooled proportion is .....??
Hello Nipun,
Not sure if we are on the same page here, but I will take a stab. The estimate (i.e., beta coefficient) for disease will be the estimate for individuals that are the mean age of your age variable (assuming this is a continuous factor) and what ever the reference level of the gender variable is (typically it is male if male=0 and female=1). So if you want the effect of disease for a 60 year-old female (let's call it "60F"), for example, you would apply something similar to the following (assuming reference level for gender = male) 60F = disease estimate + gender estimate + (age estimate x (60-mean age in study)). Don't forget that if you want more intuitive numbers you will have to back transform from the logit after you calculate for the adjusted estimate [odds = exp(logit) and then prob = odds / (1 + odds)]. I hope at least some of this helps.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis