If we have precalculated effect estimates such as md,smd,or,rr with their confidence intervals only for each study. Then, how could i perform meta-analysis?
I have checked STATA, ir says metan command can be used logOR (or logRR) with corresponding SE or 95% CI. However, when I pass these infor to SATA, it showed error ("insufficient variables").
In R using metafor package, one can use forest (defalt) for ploting but it autometically changes the values of confidence intervals.
So, how could i make forest plot either using STATA or R. I saw many articles are publishing this type of plots. Kindly suggest the right way of doing.
Thank you in advance.
The best software for calculating overall effects are the Review Manager (RevMan) from the Cochrane collaboration (it's for free) or Comprehensive Meta-analysis software (CMA, paid). The first is optimal for forest plot, the second is better as it can calculate overall effects from a multitude of data type (means, SD, SE, p value, Confidence intervals and so on). Sometimes you need to work on both (as I do). CMA for calculation and RevMan for graphical presentation.
Hi Ram Bajpai
In metafor package you can use the escalc function, inside the function, you can automatically calculate the effect estimates (included log transformations), and depend of your response variable you can performed the meta-analysis and forest plot.
if you can more specific code for do that, please send a message.
i hope this info help you.
Hi..both,
Thank you for your reply.
My original point is if you have only effect estimate (such as MD or SMD or OR or RR) with corresponding 95% CIs. Then how to create forest plot with which software??
You can check the link below for restructuring the data from the 95% IC, and OR or RR, this will works only on parametric test.
http://www.metafor-project.org/doku.php/tips:assembling_data_or
Hi,
Stata has a comprehensive set of user-written commands to perform meta-analysis of studies with (i) binary outcomes i.e. relative risk, odds ratio; (ii) continuous outcomes i.e. mean differences.
Assuming you have data for six cities in the UK as shown below:
City Log RR standarderror variance
1
2
3
The following command should give you the effect estimate and automatically generate a random forest plot in Stata\;
metan LogRRl standarderror, label (namevar=City)
For further reading see the link belowhttp://www.biostat.jhsph.edu/~fdominic/teaching/bio656/software/meta.analysis.pdf
Thank you Ram, Emma and other colleagues. Now, after doing the meta analysis, how can I generate the Forest plot? I have already data of odds ratios and confidence intervals (and even the % weights of each OR).
Hi Wael, there is an option in both 'metan' and 'ipdmetan' Stata commands using pre-calculated estimates (i.e., OR, SE, and 95%CIs) to create forest plot. On the other hand, you can simply use 'forest' command in Stata to create desired forest plot using these values.
Hi Ram,
First, you have to generate the ln estimates:
gen lnest=ln(Est)
gen lnLCI=ln(LCI)
gen lnUCI=ln(UCI)
The you use metan to create the forest plot:
metan lnest lnLCI lnUCI, eform effect(Estimate) xlab(0.25, 0.5, 1, 2, 4) random label(namevar=Studyname) texts (90).
Let me know if this works!
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis