metan: a comprehensive update to the popular user-written meta-analysis package for Stata
2Department of Population Health Sciences, University of Bristol, UK
3Bristol Medical School, Department of Population Health Sciences, University of Bristol, UK
Background: Whilst the ‘metan’ package for Stata has provided high-quality user-contributed support for aggregate-data meta-analysis since 1998, until recently, the package had not been maintained for nearly a decade and lacked various standard features available in other applications.
Objectives: In collaboration with previous co-authors and members of the meta-analysis community, we aimed to comprehensively update the ‘metan’ package and thereafter to track and quantify changes in usage, opinion and impact.
Methods: We introduced new options and syntax variations, whilst ensuring backward compatibility. We tested robustly for numerical errors and bugs. All changes were documented and agreed by the authorship group. User documentation was updated and clarified. Using the ‘ssccount’ package, we tracked number of downloads per month and assessed time trends. User engagement was quantified via the ‘Statalist’ online forums and direct email contact.
Results: ‘metan’ version 4 was released in December 2020. New random-effects options were added, such as restricted maximum likelihood (REML) and the Hartung-Knapp-Sidik-Jonkman (HKSJ) adjustment. Support for cumulative and influence meta-analysis and for proportions was added. Forest plots are drawn with a separate program, enabling options (e.g., colours) to be specified for individual elements and text to be flexibly manipulated. ‘metan’ is available via the Statistical Software Components (SSC) archive and via GitHub, where testing scripts and error logs are also stored. The textbook ‘Systematic Reviews in Healthcare: Meta-analysis in Context, 3rd edition’ (2021) includes a detailed chapter on its use. There was a clear upward trend in usage, from ~250 monthly downloads in 2010 to a stable ~3,500 since 2019. User engagement, via forums and email, occurs around once per month. There was no evidence of a drop in engagement in response to new official meta-analysis capabilities within Stata v16+.
Conclusions: The ‘metan’ package is familiar to users and remains well-used. We will continue to maintain, test and support it going forward and provide additional documentation and examples. This will ensure it remains a key tool for those conducting and presenting meta-analyses of aggregate data. Relevance and importance to patients: Robust, flexible and easy-to-use meta-analysis packages, supported and maintained by their authors, are of paramount importance to evidence synthesis and production.