Review Manager, universally shortened to RevMan, is the software Cochrane built for preparing and maintaining systematic reviews. It structures a review into comparisons, outcomes, and studies, accepts the numbers you extract from each trial, and runs the pairwise meta-analysis that produces the forest plots readers associate with Cochrane. The current version is RevMan Web, a browser-based platform that has replaced the old RevMan 5 desktop program, and it is used both by Cochrane author teams and, under a subscription, by researchers producing reviews outside Cochrane.
From RevMan 5 to RevMan Web
For two decades RevMan was a desktop application, and RevMan 5 became the default meta-analysis tool taught in countless methods courses. Cochrane ended active development of the desktop line around 2020 and moved everything to the web platform, and the desktop version is no longer supported centrally. RevMan Web is now where new Cochrane reviews are written, and it is the version you should learn: it stores the review in the cloud, lets several authors work on the same file without emailing versions around, tracks changes, and has native support for newer methods content such as the RoB 2 risk of bias tool. If you have an old RevMan 5 file, RevMan Web can import it, but starting a new project in the retired desktop program is a bad idea. The broader landscape of options is mapped in our comparison of meta-analysis software packages, which places RevMan alongside R, Stata, and the commercial tools.
Who can use RevMan Web
Access depends on what you are writing. Authors of a registered Cochrane review use RevMan Web through their Cochrane account as part of the editorial process; the software is the mandated production environment for those reviews. For everyone else, Cochrane sells RevMan Web as a subscription product. At the time of writing there are institutional licences for universities and organisations, individual subscriptions for single researchers, a low-cost student option, and free access for researchers in many low-income countries through the Research4Life programme, with discounts for existing Cochrane authors working on non-Cochrane projects. Prices and tiers change, so check Cochrane’s current subscription page rather than relying on a figure from an older guide. The practical point is that RevMan is no longer a free download the way RevMan 5 once was, which is one reason many teams now weigh it against open-source alternatives before committing.
Setting up the review structure
RevMan forces you to think in the hierarchy a synthesis actually needs. A review contains one or more comparisons (intervention versus comparator), each comparison contains outcomes, and each outcome can be split into subgroups that hold the individual studies. Before any data entry, you add your included studies with their identifying details, then build the comparison tree to match the analysis plan in your protocol. This structure is worth getting right early, because it determines how results are grouped on every plot. A well-specified outcome states the measure, the timepoint, and the direction of benefit; a vague one produces a forest plot nobody can read. The thinking behind that planning stage is covered in our walkthrough of how to do a meta-analysis from protocol to pooled estimate. RevMan Web also carries the non-statistical parts of a review: the text sections, the characteristics of included studies tables, the references, and the risk of bias assessments, including structured support for RoB 2 judgements, all live in the same file as the analyses, which keeps the manuscript and its numbers from drifting apart.
Entering dichotomous and continuous data
Data entry is deliberately simple. For a dichotomous outcome such as mortality, you enter the number of events and the number of participants in each arm, and RevMan computes a risk ratio, odds ratio, or risk difference per study. For a continuous outcome such as a pain score, you enter the mean, standard deviation, and sample size per arm, and it computes a mean difference or a standardised mean difference when studies used different scales. A third route, the generic inverse variance outcome, accepts an effect estimate and its standard error directly, which is how you pool hazard ratios, adjusted estimates, or anything the standard formats cannot hold. The built-in calculator earns its keep here: it converts a confidence interval or a p value into a standard error, derives standard deviations from standard errors, and combines arms, so you can salvage studies that report their results awkwardly instead of discarding them.
Handling awkward study designs
Real reviews rarely contain only tidy two-arm parallel trials, and this is where RevMan requires methodological judgement rather than typing. A crossover trial analysed as if it were parallel wastes its within-person design, so the recommended route is a paired analysis entered through the generic inverse variance outcome. A cluster randomised trial entered with individual-level numbers commits a unit-of-analysis error that makes the study look far more precise than it is; the counts must be deflated by the design effect first. Trials with multiple intervention arms sharing one control group risk double counting the control participants if each pairwise comparison is entered separately, so arms are combined or the control group is split. And time-to-event outcomes enter either as log hazard ratios through generic inverse variance or through the observed minus expected and variance format. RevMan will not warn you about any of these; the software accepts whatever numbers you type, which is why a synthesis is only as sound as the person preparing the data.
Subgroup and sensitivity analyses in practice
Within its limits, RevMan handles planned exploration well. Any outcome can be split into subgroups, by population, dose, risk of bias, or any categorical characteristic, and RevMan pools each subgroup separately and reports a test for subgroup differences so you can judge whether the effects genuinely differ rather than eyeballing overlapping intervals. The logic and the many ways such analyses mislead are covered in our guide to subgroup analysis and meta-regression. Sensitivity analyses, rerunning the synthesis without high risk of bias studies, switching between fixed and random effects, or excluding estimates that relied on imputed statistics, are done manually by duplicating an analysis and toggling studies in or out. It is more laborious than a scripted leave-one-out loop, but for a handful of pre-specified checks it works, and the principles of choosing them are set out in our piece on sensitivity analysis in meta-analysis.
The analyses RevMan supports
RevMan covers the standard pairwise toolkit. For dichotomous data it offers the Mantel-Haenszel method, the default and the better performer when events are sparse, plus the Peto odds ratio for rare events and inverse variance weighting; continuous and generic outcomes use inverse variance. Each can be run as a fixed-effect or a random-effects analysis, with the random-effects model based on the long-serving DerSimonian and Laird estimator of tau-squared. That choice is a methodological decision, not a software default to accept blindly, and our guide to fixed-effect versus random-effects models explains how to make it before you see the results. RevMan reports the I-squared and chi-squared statistics with every pooled analysis, and what those numbers do and do not tell you is unpacked in our piece on heterogeneity in meta-analysis. Subgroup analyses with a test for subgroup differences and simple sensitivity reruns are supported; anything more ambitious is not.
Forest plots, funnel plots, and reading the output
The forest plot is where RevMan shines. Every study appears with its raw numbers, effect estimate, confidence interval, and weight, with the pooled diamond beneath and the heterogeneity statistics printed underneath, in the exact house style journals recognise from Cochrane reviews. Plots export in vector and image formats suitable for submission. RevMan also draws a funnel plot for any outcome with enough studies, plotting effect against standard error so you can inspect small-study asymmetry, although it does not run Egger-style asymmetry tests for you. Reading the output takes more care than producing it: the pooled diamond deserves attention to its width as much as its position, the weights column reveals whether one large trial is quietly deciding the answer, and a statistically significant pooled effect above substantial unexplained heterogeneity is a finding to interrogate, not to celebrate. If you need a refresher on what the shapes mean, see our guides to reading a forest plot and interpreting a funnel plot.
Summary of Findings tables through GRADEpro
A modern review pairs its pooled estimates with a Summary of Findings table that rates how much confidence readers can place in each outcome. RevMan does not build these alone; it integrates with GRADEpro GDT, the companion tool for GRADE certainty ratings. You link the review, pull the pooled results across, make the judgements about risk of bias, inconsistency, indirectness, imprecision, and publication bias in GRADEpro, and return the finished table to the review. We cover that workflow end to end in our GRADEpro walkthrough, because the table is where many otherwise sound reviews lose marks.
RevMan’s real limitations
Honesty about the ceiling matters. RevMan offers no meta-regression, so you cannot model how a continuous study characteristic such as dose or baseline risk relates to effect size. Its random-effects machinery is essentially one estimator with no Hartung-Knapp adjustment, no REML, and no prediction interval on the standard output. There is no multivariate or network meta-analysis, no dose-response modelling, and no serious toolkit for publication bias beyond the visual funnel plot. None of this is a flaw for the reviews RevMan was designed for, but when your synthesis needs those methods the answer is to export your data and move to a statistical environment. Our guides to meta-analysis in R and meta-analysis in Stata pick up exactly where RevMan stops.
Where RevMan fits in 2026
The sensible way to see RevMan Web is as a disciplined production environment rather than a statistics package. It enforces the structure a review needs, makes data entry hard to get wrong, draws the plots journals expect, and connects cleanly to GRADEpro. In exchange you accept a closed set of models and, outside Cochrane, a subscription. For a standard intervention review with pairwise comparisons it remains one of the fastest routes from extracted data to a submittable synthesis; for anything methodologically adventurous, treat it as the starting point, not the destination. A reasonable rule of thumb: if your analysis plan mentions meta-regression, prediction intervals, alternative tau-squared estimators, or comparisons across more than two interventions, plan for a statistical package from the outset rather than discovering the ceiling halfway through. Whichever tool runs the numbers, the reporting standards stay the same, so effort invested in clean outcome definitions inside RevMan transfers wholesale. If you would rather not learn any of it, our meta-analysis service delivers the completed analysis with every model choice documented.