Is RoB 1 still acceptable?

It is not wrong, but it is superseded. Cochrane recommends RoB 2 for randomized trials in new reviews, and many journals and funders expect it. RoB 1 remains acceptable when updating a review that used it, and the judgements in existing reviews remain valid for what they measured.

Can I convert RoB 1 judgements to RoB 2?

No. The domains, the questions and the rules for combining them differ. Re-assess with RoB 2 or report the RoB 1 judgements on their own scale.

Does Some concerns mean the same as Unclear?

No. Unclear meant the reviewer could not judge. Some concerns is a substantive judgement that the result may be affected by bias without the evidence supporting High. RoB 2 handles missing information through the No information response to individual signalling questions instead.

Why did RoB 2 drop the other bias domain?

To keep the tool to specific, defined sources of bias, each with its own signalling questions. Open-ended domains were applied inconsistently. Design-specific issues moved to the cluster-randomized and crossover variants; applicability, imprecision and similar concerns belong in GRADE.

Is RoB 2 harder to apply?

Yes. It needs more documents, per-result assessment and more time. Empirical studies found low agreement without calibration and much better agreement once teams wrote down review-specific instructions. Plan for a calibration round and two independent reviewers.

Which tool should a non-Cochrane review use?

RoB 2, unless the journal or funder specifies otherwise. The Cochrane Handbook describes RoB 2 as the recommended tool, and there is no methodological reason to prefer RoB 1 for a new review.

How do I make a traffic-light plot for RoB 2?

The robvis package and web app (McGuinness and Higgins, 2021) has templates for both RoB 1 and RoB 2. CoRATES exports RoB 2 assessments as traffic-light and weighted bar plots directly.

Does CoRATES support RoB 1?

No. CoRATES implements RoB 2, with the official algorithms proposing domain and overall judgements, and supports independent assessment by several reviewers followed by reconciliation.

RoB 1 vs RoB 2: what changed in the Cochrane risk-of-bias tool

Choosing an appraisal tool

The original Cochrane risk-of-bias tool was introduced in 2008 and revised in 2011. Most people now call it RoB 1, although that name was never official. RoB 2 was published in 2019 and is the recommended tool for randomized trials in Cochrane Reviews. The two cover the same ground, but RoB 2 is a replacement rather than a new edition: it restructured the domains, replaced free judgements with signalling questions and algorithms, dropped the Unclear category, and moved the unit of assessment from the trial to the result.

This page maps the old domains onto the new ones, explains where the two tools genuinely disagree about the same trial, and covers the practical question that generates most searches: what to do with a review, or an update, that used RoB 1.

The short answer

Starting a new review

RoB 2

Cochrane recommends it for randomized trials, and most journals and funders now expect it. The variants for cluster-randomized and crossover trials are also RoB 2.

Updating a review that used RoB 1

Either, but not both

Cochrane says it will never be mandatory to switch during an update. Switch if you expect many new studies and can re-assess the old ones; continue with RoB 1 if few are expected. Decide before you start and apply one tool to every result.

Reading a review that used RoB 1

Its judgements stand

They are valid for the tool that produced them. Do not translate them: Unclear is not Some concerns, and a High for blinding under RoB 1 does not predict the RoB 2 verdict.

How the domains map

The mapping below is approximate by necessity. RoB 2 did not rename the old domains; it asked different questions about the same sources of bias and moved some concerns between them.

Random sequence generation becomes Domain 1, Randomization process
Allocation concealment becomes Domain 1, Randomization process
Blinding (participants and personnel) becomes Domain 2, Deviations from intended interventions
Blinding (outcome assessment) becomes Domain 4, Measurement of the outcome
Incomplete outcome data becomes Domain 3, Missing outcome data
Incomplete outcome data in part becomes Domain 2, Deviations from intended interventions
Selective reporting becomes Domain 5, Selection of the reported result
Other bias has no equivalent

Dashed lines: exclusions after randomization, previously part of incomplete outcome data, are now considered under Domain 2; other bias has no equivalent in RoB 2.

RoB 1 domain	RoB 2 domain	What changed
Random sequence generation	Domain 1: bias arising from the randomization process	Merged with allocation concealment, and a signalling question on baseline imbalance was added as evidence about whether the randomization process was sound.
Allocation concealment	Domain 1: bias arising from the randomization process	As above.
Blinding of participants and personnel	Domain 2: bias due to deviations from intended interventions	Blinding is no longer a domain in its own right. Domain 2 asks whether awareness of assignment led to deviations from the intended intervention that were unbalanced and likely to affect the outcome, and whether an appropriate analysis was used. An open-label trial is not automatically at high risk.
Blinding of outcome assessment	Domain 4: bias in measurement of the outcome	Asks whether outcome assessors were aware of assignment and, if so, whether the assessment could have been influenced by that knowledge, which depends on the type of outcome.
Incomplete outcome data	Domain 3: bias due to missing outcome data	Moves from the proportion missing to whether the result is likely to be biased by the missingness: were data available for nearly all participants, is there evidence the result was not biased, and could missingness depend on the true value of the outcome. Exclusions after randomization are now considered under Domain 2.
Selective reporting	Domain 5: bias in selection of the reported result	Narrowed to selection among multiple measurements or analyses of the same outcome, which requires checking a pre-specified plan. Non-reporting of whole outcomes is handled at the synthesis level rather than within the tool.
Other bias	No equivalent	Dropped. Design-specific concerns for cluster-randomized and crossover trials are handled by the RoB 2 variants for those designs. Baseline imbalance moved into Domain 1. Concerns that are not biases in the RoB 2 sense, such as applicability or imprecision, belong in GRADE instead.

Side by side

	RoB 1	RoB 2
Published	2008 in the Cochrane Handbook; revised 2011 (Higgins et al., BMJ).	2019 (Sterne et al., BMJ).
Unit of assessment	The trial, with some domains assessed per outcome.	Each result: one outcome, time point and analysis.
Domains	Seven, including an open-ended other bias domain.	Five fixed domains. No other bias domain.
How judgements are reached	Reviewer judgement per domain, supported by quoted text and the Handbook criteria.	Signalling questions answered Yes, Probably yes, Probably no, No or No information, and a published algorithm that proposes the judgement. Reviewers can override with a recorded reason.
Judgement categories	Low risk, High risk, Unclear risk.	Low risk, Some concerns, High risk.
Missing information	Unclear risk.	No information is a response to a signalling question, not a verdict. The algorithm turns it into Some concerns or High depending on where the gap is.
Overall judgement	No formal rule. Reviewers summarised across domains in different ways.	Defined rules. High if any domain is High, or if Some concerns in several domains substantially lowers confidence; Some concerns if any domain is Some concerns and none is High; Low only if every domain is Low.
Effect of interest	Not specified.	Chosen in advance: the effect of assignment to intervention or the effect of adhering to it. Domain 2 differs between them.
Trial designs	Handbook guidance for cluster-randomized and crossover trials.	Dedicated variants for cluster-randomized and crossover trials that add design-specific questions.
Documents needed	Usually the trial report.	The trial report plus the protocol, registry entry or statistical analysis plan for Domain 5.
Cochrane status	Superseded. Still permitted when updating a review that used it.	Recommended for randomized trials in Cochrane Reviews.
In CoRATES	Not supported.	Supported (parallel-group version).

Unclear is gone, and Some concerns is not its replacement

Unclear in RoB 1 covered two different situations: information not reported, and information reported but ambiguous. RoB 2 separates them. No information is an answer to a single signalling question, and the algorithm decides what it implies. For a question that would reveal a problem, no information means no evidence of that problem; for something a trial should have reported, it raises concern. A trial that was Unclear across the board under RoB 1 may come out as Some concerns or as High under RoB 2, and there is no way to tell which without re-assessing it.

Blinding is an input, not a verdict

Under RoB 1, an open-label trial usually received High risk for both blinding domains. Under RoB 2 the question is whether the lack of blinding mattered for this result. An open-label trial with an objective outcome, no differential deviations and an intention-to-treat analysis can be Low in Domains 2 and 4. This is the single biggest source of changed judgements when trials are re-assessed, and it usually moves them towards lower risk.

Domain 5 often moves the other way. If no protocol, registration or analysis plan can be found, RoB 2 does not let a result reach Low for selection of the reported result, where RoB 1 reviewers frequently recorded Unclear and moved on.

Per result, not per trial

A trial with a blinded clinician-assessed primary outcome and an unblinded patient-reported secondary outcome gets a separate RoB 2 assessment for each result, and may get different verdicts. This multiplies the work but produces judgements that map onto the specific rows of the meta-analysis. RoB 1 let reviewers assess per outcome for some domains but was usually applied once per trial.

The overall judgement has rules

RoB 1 never defined how domain judgements combine, so reviews summarised them in incompatible ways. RoB 2 does: any High domain makes the result High overall, and Some concerns in several domains can too. The Handbook stresses that risk of bias means risk of material bias, so a Some concerns judgement should reflect a plausible effect on the result rather than a formality.

It takes longer, and agreement does not improve by itself

Minozzi et al. (2020) measured about half an hour per result and only slight agreement on the overall judgement among four experienced raters. Their follow-up (2022) found that a calibration exercise and a written, review-specific implementation document raised agreement to moderate and cut the time per result substantially. The tool rewards teams that plan for it and punishes teams that treat it as a longer RoB 1.

What to do with an existing review that used RoB 1

Do not convert. There is no valid mapping from Low, High and Unclear to Low, Some concerns and High, because the tools ask different questions and combine them differently. Re-assess with RoB 2 or keep the RoB 1 judgements as they are.

For a Cochrane update, the editorial position is that switching to RoB 2 will never be mandatory if the previous version used the original tool. The suggested rule of thumb is to switch when many new studies are expected and to continue with RoB 1 when few are. If you switch, update the methods before starting and assess every result, old and new, with RoB 2. Mixing the two tools within one review is not acceptable.

For GRADE, the risk-of-bias domain is a judgement about a body of evidence, so it can draw on either tool. State which tool applies to which trials, and remember that a RoB 2 overall of Some concerns is not automatically a downgrade; the question is whether the concerns are likely to have changed the pooled result.

Appraise with RoB 2 in CoRATES

Open an appraisal in your browser and work through it item by item. Nothing to install and no account needed. RoB 1 is not available in CoRATES.

Start a RoB 2 appraisal

Frequently asked questions

Is RoB 1 still acceptable?: It is not wrong, but it is superseded. Cochrane recommends RoB 2 for randomized trials in new reviews, and many journals and funders expect it. RoB 1 remains acceptable when updating a review that used it, and the judgements in existing reviews remain valid for what they measured.
Can I convert RoB 1 judgements to RoB 2?: No. The domains, the questions and the rules for combining them differ. Re-assess with RoB 2 or report the RoB 1 judgements on their own scale.
Does Some concerns mean the same as Unclear?: No. Unclear meant the reviewer could not judge. Some concerns is a substantive judgement that the result may be affected by bias without the evidence supporting High. RoB 2 handles missing information through the No information response to individual signalling questions instead.
Why did RoB 2 drop the other bias domain?: To keep the tool to specific, defined sources of bias, each with its own signalling questions. Open-ended domains were applied inconsistently. Design-specific issues moved to the cluster-randomized and crossover variants; applicability, imprecision and similar concerns belong in GRADE.
Is RoB 2 harder to apply?: Yes. It needs more documents, per-result assessment and more time. Empirical studies found low agreement without calibration and much better agreement once teams wrote down review-specific instructions. Plan for a calibration round and two independent reviewers.
Which tool should a non-Cochrane review use?: RoB 2, unless the journal or funder specifies otherwise. The Cochrane Handbook describes RoB 2 as the recommended tool, and there is no methodological reason to prefer RoB 1 for a new review.
How do I make a traffic-light plot for RoB 2?: The robvis package and web app (McGuinness and Higgins, 2021) has templates for both RoB 1 and RoB 2. CoRATES exports RoB 2 assessments as traffic-light and weighted bar plots directly.
Does CoRATES support RoB 1?: No. CoRATES implements RoB 2, with the official algorithms proposing domain and overall judgements, and supports independent assessment by several reviewers followed by reconciliation.

Reference Documents

Related guides

This page describes the tools and cites their official sources. It does not reproduce signalling questions, items or scoring tables, which are the intellectual property of their original authors. Consult the official publications and guidance linked above when applying any tool.