2009年最新进展：Updated and new commands for meta-analysis in Stata.pdf

Updated and mands for meta-analysis in Stata™

Michael J. Bradburn
Cancer Research UK Medical Statistics Group,
Centre for Statistics in Medicine, Oxford, UK

This document and the Stata programs referred to build on previous work done in collaboration
with Jon Deeks and Doug Altman from the Centre for Statistics in Medicine, Oxford. The dialog box
interface was written by Tom Steichen.

1 Introduction

Meta-analysis is the statistical methodology used bine data from two or more studies that
have addressed the same question. Several methods for this type of analysis have been
implemented in the Stata software, including basic analyses of aggregate data (meta and metan
commands), cumulative meta-analysis (metacum), diagnostics (metainf, metabias, metatrim,
galbr, funnel) and meta-regression (metareg). All of these programs were originally written for
the Stata Technical Bulletin, and are described in detail in the respective submissions. Chapter 18
of Egger et al. (2001; available online at ks/sysrev/) forms
a plete introduction to meta-analysis in Stata.

This text describes only a subset of these methods. The article falls in two parts: to illustrate
enhancements to the metan, funnel and labbe programs previously submitted to the Stata
Technical Bulletin in 1998 (Bradburn et al.), and to introduce a mand, metannt. Section
2 starts by recapping on what the existing programs do before summarising the new additions,
together with examples that demonstrate their use, and will also introduce the new dialog box
interface. In section 3, the new routine will be introduced and demonstrated, and some final
remarks made in section 4. A full list of options available for all four the programs is given in the
appendix.


2. The metan, funnel and mands

Aggregate data meta-analysis (metan)

Present capabilities

The original 1998 version of bines 2x2 tables or mean differences, and pre