Skip to content

Risk of bias 2 tool: How to assess trial results

The risk of bias 2 tool (RoB 2) assesses whether a specific result from a randomized trial may be biased; it does not assign one rating to the trial as a whole. The result is defined by its outcome, time point, comparison, and analysis, so different results from the same trial can receive different judgements. [1][2]

For systematic reviews, using RoB 2 well means specifying the result and effect of interest, choosing a version suited to the trial design, and documenting how evidence supports each judgement. The tool structures reviewer decisions; it does not remove the need for methodological interpretation. [1][3]

How do I use the risk of bias 2 tool in a systematic review?

Use RoB 2 by defining in advance which trial results you will assess, selecting the appropriate tool version, and answering its signalling questions using evidence from the trial report and other relevant sources. [1][3]

First, identify each result by its outcome, time point, comparison, and analysis. Then specify the effect of interest: usually the effect of assignment to intervention, but sometimes the effect of adhering to the intervention. This choice matters because it affects assessment of deviations from intended interventions. [1]

For each result, reviewers answer signalling questions and consider the evidence bearing on each domain. Record the source and rationale for important decisions rather than treating the tool as a mechanical checklist. The Cochrane Handbook and official RoB 2 materials provide the detailed questions and decision guidance; a short overview cannot substitute for them. [1][3]

What are the domains of the risk of bias 2 tool?

The standard RoB 2 version for individually randomized, parallel-group trials has five required domains: the randomization process; deviations from intended interventions; missing outcome data; outcome measurement; and selection of the reported result. The cluster-randomized version adds a domain covering the timing of participant identification and recruitment. [1][6]

These domains organize assessment around possible sources of bias in the result being reviewed. Use the required domains for the chosen version rather than adding custom domains. The Handbook’s five-domain description applies to the standard version, while design-specific guidance covers the variants. [1][6]

The domains do not function as independent checklist scores. Reviewers use the signalling questions and the evidence to judge whether a domain raises concerns about the particular result. To apply the questions correctly—including design-specific guidance—consult the full Handbook and the relevant RoB 2 tool. [1][3]

Which RoB 2 version and effect of interest should I specify?

Choose the RoB 2 version that matches the trial design, and specify the effect of interest before assessing results. The standard version covers individually randomized, parallel-group trials; separate versions are available for cluster-randomized and crossover trials. [1][3]

A review should state which results it will assess and which version it will use for each applicable design. The effect of interest is usually assignment to intervention, but a review may instead be concerned with adherence to intervention. Because this choice changes how reviewers assess deviations from intended interventions, it should be explicit rather than inferred from the analysis label. [1]

An “intention-to-treat” label alone does not establish that all randomized participants were included or that missing outcomes were handled appropriately. Reviewers should examine what the analysis actually included and how the result was produced. [1][2]

How do I interpret results from the risk of bias 2 tool?

RoB 2’s algorithm proposes a domain judgement of Low risk, Some concerns, or High risk, but reviewers should check the proposed judgement against the evidence, explain any override, and report the supporting rationale. [1]

The overall judgement follows defined rules: Low risk when all domains are Low; Some concerns when at least one domain has Some concerns and none is High; and High risk when any domain is High or several concerns together substantially reduce confidence in the result. [1]

These categories summarize the assessment; they are not a numerical measure of bias or certainty. A defensible report makes the path from evidence to signalling-question answers, domain judgements, and the overall judgement visible to readers. [1][2][4]

How should I assess missing outcome data and other common challenges?

Assess missing outcome data in context; there is no universally valid percentage cutoff that determines whether missingness creates risk of bias. [1]

Consider the outcome, event frequency, reasons data are missing, differences between groups, and plausible outcomes among participants whose data are missing. A single threshold cannot account for these factors, so a percentage alone is not a sufficient basis for judgement. [1]

Likewise, lack of blinding or a baseline imbalance does not automatically mean a result is at high risk of bias. Ask whether the feature could materially bias that specific result. An isolated baseline difference or a p-value near 0.1, by itself, does not show that randomization failed. [1][2]

Implementation feedback also signals why defining the unit of assessment matters: nearly 43% of the published reviews examined appeared to assess whole studies rather than specific results. That figure describes the reviews examined in that source and should not be generalized to all reviews. [2]

How can review teams apply RoB 2 consistently, and where can automation help?

Review teams can improve consistency by piloting assessments on varied results, assessing independently where feasible, discussing disagreements, and writing review-specific decision rules. Transparent reporting should include signalling-question responses and detailed rationales. [2][4]

When evaluating a systematic-review platform, check whether it can retain source evidence, signalling-question responses, reviewer rationales, and records of disagreements. These are practical workflow requirements derived from the guidance on transparent assessment, not evidence that any particular platform can reliably answer signalling questions or replace reviewer interpretation. [1][2]

Key takeaways

  • RoB 2 assesses a defined result from a randomized trial, not the trial as a whole. [1][2]
  • Match the tool version to the trial design and specify the effect of interest before assessment. [1][3]
  • Use the required domains and treat algorithmic judgements as proposals to verify against evidence. [1]
  • Judge missing outcome data in context; there is no universal percentage cutoff. [1]
  • Document rationales and disagreements so readers can follow how each judgement was reached. [2][4]
  • Automation can organize and flag assessment information, but reviewer judgement remains necessary. [1][2]

FAQ

Are there training resources for the risk of bias 2 tool?

Yes. The Cochrane Handbook chapter and official RoB 2 materials provide the tool and detailed guidance for its use. Practical implementation guidance also discusses piloting assessments, resolving disagreements, and making decision rules transparent. [1][2][3][4]

Should I use RoB 2 to assess non-randomized studies?

This guidance concerns RoB 2 versions for randomized trials, including individually randomized, parallel-group, cluster-randomized, and crossover designs. It does not establish which tool is appropriate for non-randomized studies; choose a method based on the study design and relevant methodological guidance. [1][3]

Can an automated system make the final RoB 2 judgement?

The cited guidance supports structured assessment and emphasizes that reviewers should verify algorithmic proposals, justify overrides, and report evidence. It does not establish that automation can replace reviewers’ interpretation or make reliable final judgements on its own. [1][2]

Why does RoB 2 assess a specific result rather than an entire trial?

A result is defined by its outcome, time point, comparison, and analysis, and different results from the same trial can receive different assessments. Assessing the specific result keeps the judgement focused on the evidence relevant to that finding. [1][2]

Sources

  1. Chapter 8: Assessing risk of bias in a randomized trial | Cochrane
  2. Ten tips for successful assessment of risk of bias in randomized trials using the RoB 2 tool: Early lessons from Cochrane - PMC
  3. Risk of bias tools - RoB 2 tool
  4. Using Risk of Bias 2 to assess results from randomised controlled trials: guidance from Cochrane
  5. Risk of bias | Cochrane
  6. Chapter 23: Including variants on randomized trials | Cochrane

Related articles

Robins-I Tool Challenges: Applying It Consistently

Learn why ROBINS-I judgments vary and how version control, target-trial definitions, confounders, and clear rationales support reproducible assessments.

5 min read ·

prisma 2020 checklist pdf: How to Find and Complete It

Find the official PRISMA 2020 checklist PDF and learn how to record manuscript locations and verify them against review records.

10 min read ·

AI Abstract Screening Challenges: Risks and Validation

Learn how AI screening can affect recall, precision, workload, and missed studies, and how to validate and document review-specific workflows.

6 min read ·