{smcl}
{* *! version 3.1 13jul2015}{...}
{vieweralsosee "mvmeta" "mvmeta"}{...}
{vieweralsosee "Main mvmeta demonstration" "mvmetademo_run"}{...}
{hline}
{cmd:Demonstration for the mvmeta package in Stata: getting the data into mvmeta format}
{cmd:Ian White}
{hline}
{title:How to run this tutorial}
{p}You will need to install the mvmeta package as described in {help mvmetademo_run:the main demonstration}.
{p}{cmd:mvmeta} expects the data to be organised with one line per study,
with each line containing the estimates and their variances for that study.
The estimates must be named as a common stub followed by a unique ending,
and the variances must be named as a (different) common stub followed by a repeated ending.
Ideally, covariances are also included, named as the variance stub followed by two endings.
{p}Let's make that more concrete by inputting the p53 data.
These data are described in
{help mvmetademo_run##p53:the main demonstration}.
The data are the estimated log hazard ratios (lnHR)
for mutant vs. normal p53 gene for two outcomes.
{p}The data for overall survival look like this:
study estimate std.error
1 -.18 .56
2 .79 .24
3 .21 .66
4 -.63 .29
5 1.01 .48
6 -.64 .4
{p}and the corresponding data for disease-free survival
(which was only reported in 3 studies) look like this:
study estimate std.error
1 -.58 .56
4 -1.02 .39
6 -.69 .4
{p}We enter these using lnHR as the stub for the log hazard ratios
and selnHR as a stub for the standard errors,
and using abbreviations os for overall survival and dfs
for disease-free survival:
{stata clear}
{stata input study lnHRos selnHRos lnHRdfs selnHRdfs}
{stata 1 -.18 .56 -.58 .56}
{stata 2 .79 .24 . .}
{stata 3 .21 .66 . .}
{stata 4 -.63 .29 -1.02 .39}
{stata 5 1.01 .48 . .}
{stata 6 -.64 .4 -.69 .4}
{stata end}
{p}Note that we have entered missing values for disease-free survival in studies 2, 3 and 5.
{p}{cmd:mvmeta} requires variables representing the variances.
These must be named as . We choose VlnHR as the stub for the variances, so:
{pstd}{stata gen VlnHRosos = selnHRos^2}
{pstd}{stata gen VlnHRdfsdfs = selnHRdfs^2}
{p}These data are now in the format required by {cmd:mvmeta} and can be analysed
as shown in {help mvmetademo_run##p53:the main demonstration}.
{p}If we also knew the within-study covariances, we could have input them with the main data.
If we knew the within-study correlations, we could have input them with the main data (say as variable corrosdfs)
and then computed the covariances using {stata gen VlnHRosdfs = corrosdfs*selnHRos*selnHRdfs}.