Insights / Topic · Software & tools
Choosing reference managers, screening platforms and analysis software for a review, from Zotero and Rayyan to R, Stata, RevMan and GRADEpro.
Software choices made at the start of a review are hard to reverse later. A reference manager that mishandles duplicates, a screening platform that cannot export decisions, or an analysis package that cannot fit the model the data need can each cost weeks once the project is under way.
Each stage has its own tools. Reference managers such as Zotero and EndNote store and deduplicate search results. Screening platforms such as Rayyan, Covidence and EPPI-Reviewer manage independent title, abstract and full-text decisions and record reasons for exclusion. Data extraction can live in a structured spreadsheet or in the screening platform itself. Risk of bias judgements are recorded against the official RoB 2 and ROBINS-I templates, and robvis draws the summary plots.
For analysis, R offers the widest range: metafor and meta for pairwise models, meta-regression and small-study tests, netmeta for frequentist network meta-analysis, and gemtc or multinma for Bayesian networks. Stata's meta suite covers most pairwise needs with a consistent syntax. RevMan suits standard Cochrane-style reviews but is limited for meta-regression and networks. MetaInsight provides network meta-analysis in a browser. GRADEpro GDT produces summary of findings tables. Code-based tools have one further advantage: the analysis can be shared and re-run, which reviewers increasingly ask for.
Reviewers increasingly ask for the materials behind the analysis: the search exports, the screening decisions, the extraction sheet and the code. A workflow built around tools that export cleanly makes this a matter of attaching files. One built around manual copying between programs makes it a reconstruction exercise, and errors introduced in transfer are hard to find once figures and tables no longer match each other.
The notes in this topic compare tools at each stage, explain where each becomes limiting, show how to move data between them without losing information, and set out a workflow that keeps every step reproducible from search export to final figure.
Explainer
RevMan, GRADE, Cochrane, Nested Knowledge, TrialScout