RevMan Web: Running a Meta-Analysis in Cochrane’s Tool

RevMan (Review Manager) is Cochrane’s own software for writing, analyzing, and formatting a systematic review, and RevMan Web is the current browser-based version of it. It sits downstream of the screening/extraction tools already covered elsewhere on this site – Covidence, Rayyan, and DistillerSR handle study identification and data extraction, but a review’s actual pooled effect estimates, forest plots, and Cochrane-formatted review text are produced in RevMan. Most generic RevMan explainers describe the interface without touching what changed recently in how it pools data, or how it connects to the certainty-of-evidence step that follows – this guide covers both, with the specific mechanics of setting up an analysis.

Where RevMan Web fits in the review workflow

Cochrane’s own RevMan Web documentation organizes the software around three phases rather than a flat menu of features: Plan (defining the review’s comparisons and pre-specifying risk-of-bias domains before data collection starts), Prepare (data extraction templates and the risk-of-bias assessment tools), and Populate (importing study references and entering the actual outcome data and statistical methods that generate the review’s analyses). That ordering matters in practice: a comparison’s risk-of-bias domains and outcome list are meant to be locked in during Plan, before anyone starts entering study data in Populate – setting them up retroactively after data entry is a common source of rework. RevMan Web also supports diagnostic test accuracy reviews, summary-of-findings tables, and PRISMA flow diagrams as native features rather than exports built elsewhere. For the wider landscape of screening and extraction tools RevMan sits alongside, see Systematic Literature Review Tools: A Comparison Guide and Covidence vs. Rayyan vs. DistillerSR.

Setting up a comparison and adding an analysis

Inside a review, each outcome you want to pool is entered as an analysis within a comparison – for example, “Intervention vs. control” as the comparison, with “Mortality at 12 months” as one of its outcomes. For a dichotomous outcome, you enter the number of events and the total number of participants in each arm, per included study; for a continuous outcome, you enter the mean, standard deviation, and sample size per arm. RevMan Web computes each study’s effect estimate and variance from those raw numbers, derives its inverse-variance weight, and renders the pooled result as a forest plot automatically – you do not calculate the weights or the pooled estimate by hand. What you do choose explicitly is the effect measure (risk ratio, odds ratio, risk difference for dichotomous data; mean difference or standardized mean difference for continuous data) and the pooling model. Getting the effect measure right before entering data matters more than it looks: switching it after data entry does not just reformat the output, it changes what RevMan is actually computing from the same raw numbers. See Forest plot for how to read the output this step produces, and Standardized Mean Difference (SMD) and Hedges’ g in Meta-Analysis if your outcomes were measured on different scales across studies.

Choosing a pooling method: DerSimonian-Laird, REML, and when Hartung-Knapp kicks in

This is the part most RevMan walkthroughs skip, and it changed recently. Per the current Cochrane Handbook (Chapter 10, section 10.10.4.4): “Until 2024, only the DerSimonian and Laird ‘moment-based’ method… was implemented in RevMan. As of 2024, a restricted maximum likelihood (REML) method is also available,” and “the default option for estimating the between-study variance is REML, while the DerSimonian and Laird moment-based method remains an available option.” If you’re comparing a RevMan Web analysis against an older tutorial, older published review, or a colleague’s file created before that change, a different default tau² estimator – not a data-entry error – is a likely reason the confidence interval doesn’t match. The Handbook chapter walks through the full mechanics in Inverse-Variance Weighting in Meta-Analysis: DerSimonian-Laird, REML and Hartung-Knapp.

RevMan Web also prompts you toward the Hartung-Knapp-Sidik-Jonkman (HKSJ) adjustment under a specific, documented rule: per the same Handbook section, “users are prompted to use the HKSJ method when the between-study estimate is greater than zero and the number of study results is greater than two,” because it reduces the risk of an overly narrow interval when tau² is estimated at zero, or an overly wide, unstable one from relying on too few studies. Section 10.10.4.5 separately cautions that HKSJ’s own performance is limited with only two or three studies, and recommends a sensitivity analysis comparing HKSJ against the standard interval in that narrow-k range rather than trusting either alone. Whether random-effects pooling is the right choice at all – before any of this tau²-estimator detail matters – is covered in Fixed-Effect vs. Random-Effects Meta-Analysis, and how to read the heterogeneity statistics RevMan reports alongside the pooled estimate is in Heterogeneity in Meta-Analysis: I², τ², and Prediction Intervals.

From a pooled estimate to a GRADE Summary of Findings table

A pooled effect on its own is not a finished Cochrane review section – it needs a certainty-of-evidence rating. GRADEpro GDT, the standard tool for that step, takes the comparator’s baseline risk plus RevMan’s pooled relative effect, calculates the corresponding absolute effects, walks the five-domain GRADE assessment, and exports a formatted Summary of Findings table that links directly back into RevMan for Cochrane reviews specifically – so the analysis you build in this guide is usually the direct input to that next step, not a separate exercise. The mechanics of that table (what the standard columns mean, and how a downgrade gets justified) are covered in GRADE Summary of Findings Table: Building and Interpreting One. See also GRADE (Evidence Certainty Rating).

Who can actually use RevMan Web

RevMan Web is not gated to a single “free or paid” answer – Cochrane’s own documentation structures access into separate paths for Cochrane authors and staff, individual subscribers, and organizational subscribers, which suggests real differences in what each tier gets rather than one uniform account type. If you’re planning a non-Cochrane systematic review and deciding whether RevMan Web is worth adopting over a general-purpose statistics package, check Cochrane’s current subscription terms directly for your situation rather than assuming either “free for everyone” or “Cochrane-authors-only” – this guide won’t guess at pricing or eligibility rules that change independently of the software itself. What is straightforward: RevMan Web is the browser-based successor to the older desktop application, RevMan 5, and Cochrane has been directing new review production toward the web version rather than the legacy desktop build.

Setup mistakes worth catching before they reach the forest plot

• Skipping the Plan phase. Adding outcomes and risk-of-bias domains only after data entry begins in Populate, instead of defining them in Plan first, is a common source of having to re-enter or restructure a comparison midway through.
• Mixing arm-level and study-level data entry conventions. Dichotomous data needs events and totals per arm, not a pre-calculated risk or rate – entering a summary statistic instead of the raw counts silently changes what RevMan computes, and it is easy to miss in a quick visual check of the forest plot.
• Not re-checking the pooling model after importing an older file. Because the REML default only applies going forward from the 2024 change described above, a comparison built in an older RevMan file may still be set to DerSimonian-Laird – worth confirming explicitly rather than assuming the current default applies retroactively.
• Treating unit-of-analysis errors as a RevMan bug. If a cluster-randomized or crossover trial is entered as if it were an independent-groups trial, RevMan will still compute and plot a result – it has no way to detect that the effective sample size is wrong, so this has to be caught during data extraction, not at the analysis step.

Frequently asked questions

Is RevMan Web free to use?

Access is tiered rather than uniformly free or paid – Cochrane’s documentation separates Cochrane authors/staff, individual subscribers, and organizational subscribers into distinct access paths. Check Cochrane’s current terms for your specific situation rather than assuming one answer applies to everyone.

Do I have to use RevMan Web if my systematic review isn’t a Cochrane review?

No – RevMan is Cochrane’s required format for reviews published in the Cochrane Database of Systematic Reviews, but a non-Cochrane review can run its meta-analysis in any capable tool, including R packages such as metafor (see Meta-Analysis in R with metafor). Some authors still choose RevMan for its built-in forest-plot formatting and Cochrane-familiar output even outside a Cochrane submission.

What replaced desktop RevMan 5?

RevMan Web is Cochrane’s browser-based successor to the older desktop application, RevMan 5. New review production has moved toward the web version rather than the legacy desktop build.

Does RevMan Web calculate risk of bias for me?

It provides the structured tools and templates for entering a risk-of-bias assessment (including RoB 2 and ROBINS-I domains) as part of the Prepare phase, but the judgment itself – low/some concerns/high risk per domain – is made by the reviewers, not generated automatically. See Risk of Bias Assessment: RoB 2, ROBINS-I and the Traffic-Light Plot.