Strata Academy

Cochrane ROB 2 explained – risk of bias tool for randomized trials

The complete student guide: five domains, signalling questions, domain-level and overall judgements, and when to use ROB 2 instead of ROBINS-I

1. What is ROB 2?

ROB 2 (Risk of Bias 2) is the Cochrane-recommended tool for assessing risk of bias in randomised controlled trials. It replaced the original Cochrane RoB tool and is designed for parallel-group, cluster, crossover, and split-body trial designs when the effect of assignment to intervention is the target.

Unlike generic checklists, ROB 2 uses signalling questions within predefined domains. Your answers lead to algorithm-guided judgements of low risk of bias, some concerns, or high risk of bias at domain level and overall.

ROB 2 is a risk-of-bias tool – not a reporting checklist. A trial can be well written (good CONSORT reporting) but still high risk of bias if randomisation was flawed or outcome data were missing differentially.

2. When to use ROB 2 (and when not to)

Use ROB 2 when the study randomly assigns participants (or clusters) to intervention and comparator groups and you want to judge whether the estimated effect could be distorted by bias in design, conduct, or analysis.

Do not use ROB 2 for non-randomised studies of interventions, diagnostic accuracy studies, systematic reviews, or single-arm case series. Each of those has a different official framework.

• RCT, cluster-RCT, crossover trial (with appropriate ROB 2 variant) → ROB 2
• Cohort or before–after intervention study without randomisation → ROBINS-I
• Diagnostic test accuracy → QUADAS-2
• Systematic review of RCTs → AMSTAR 2 + ROBIS for the review; ROB 2 for each included trial
• Reporting completeness only → CONSORT (complements but does not replace ROB 2)

3. The five ROB 2 domains

Work through each domain with official signalling questions before assigning an overall judgement.

• Domain 1 – Bias arising from the randomisation process: Was the allocation sequence random? Was allocation concealed? Were there baseline imbalances suggesting a problem?
• Domain 2 – Bias due to deviations from intended interventions: Were participants, carers, or clinicians aware of assignment? Were there deviations from intended interventions that arose because of the experimental context? Was analysis appropriate for deviations?
• Domain 3 – Bias due to missing outcome data: Were outcome data available for nearly all participants? Could missingness depend on the true outcome? Was appropriate analysis used?
• Domain 4 – Bias in measurement of the outcome: Could the outcome measure have differed between groups? Were outcome assessors aware of intervention received? (Awareness matters most for subjective outcomes.)
• Domain 5 – Bias in selection of the reported result: Is the reported result likely selected from multiple analyses? Were pre-specified analysis plans followed?

4. Signalling questions – how judgements are made

For each domain, ROB 2 provides signalling questions (yes / probably yes / probably no / no / no information). The official algorithm maps combinations of answers to a domain-level judgement.

You must use the signalling questions from the current Cochrane ROB 2 materials – not a shortened mnemonic from a blog post. Wording matters because algorithms are tied to specific question sets.

At domain level, the options are: Low risk of bias, Some concerns, or High risk of bias. 'Some concerns' often reflects uncertainty from incomplete reporting rather than proven serious bias – but sustained uncertainty can still lower confidence in the result.

• Read the supplementary appendix and protocol if the main paper omits randomisation or analysis detail.
• Distinguish 'not reported' from 'not done' – but flag unclear reporting as limiting your confidence.
• For subjective outcomes (pain, quality of life), Domain 4 (blinding of outcome assessment) carries more weight.
• For objective outcomes (all-cause mortality), Domain 4 is often lower concern even if unblinded.

5. Overall risk of bias judgement

After rating each domain, ROB 2 guides an overall risk of bias for the result being assessed: Low, Some concerns, or High.

A single high-risk domain can drive the overall judgement to high, depending on which domain and how plausible bias is to affect the direction or magnitude of the effect.

In systematic reviews, reviewers often present traffic-light plots per domain per study. Consistency across studies matters for meta-analysis – one high-risk trial among many low-risk trials may warrant sensitivity analysis.

• Document which result you judged (primary outcome at primary time point).
• If the paper reports multiple outcomes, ROB 2 is applied per result – the overall bias may differ for secondary outcomes.
• In meta-analyses, consider downgrading certainty in GRADE if a large contribution to the pooled estimate comes from high-risk studies.

6. ROB 2 vs ROBINS-I – decision guide

ROB 2 and ROBINS-I are complementary, not interchangeable. ROB 2 is for randomised designs; ROBINS-I is for non-randomised studies of interventions where confounding is the central threat.

If authors compare treated vs untreated groups but participants were not randomly assigned, ROBINS-I is the correct tool even when the paper uses causal language.

• True RCT with concealed allocation → ROB 2
• Propensity-score matched observational cohort → ROBINS-I, not ROB 2
• Before–after study with no concurrent control → ROBINS-I or design-specific tools
• Registry-based comparative effectiveness without randomisation → ROBINS-I

7. Worked example – published RCT

Apply ROB 2 domain by domain to a real trial you can access in full text. The ORACLE trial below is a teaching staple for randomisation, blinding, and outcome reporting.

8. Hypothetical walkthrough (HbA1c trial)

Imagine a parallel-group RCT of Drug A vs placebo for 12-week reduction in HbA1c. The paper reports randomisation by computer-generated sequence with central allocation concealment. Ten per cent are lost to follow-up, balanced between arms. Outcome assessors were aware of treatment for clinic visits but HbA1c was analysed from blinded lab samples.

Domain 1: Low concern if sequence generation and concealment are credible and baselines are balanced.

Domain 2: Some concerns if patients and clinicians were unblinded and could change co-interventions (diet, exercise) differentially.

Domain 3: Some concerns if loss to follow-up is moderate and reasons are unclear – check whether imputation was appropriate.

Domain 4: Possibly low for HbA1c if lab analysis was blinded, despite unblinded visits.

Domain 5: Some concerns if trial was registered but primary outcome in the paper differs from registry without explanation.

Overall: Often 'Some concerns' for this pattern – not automatically high, but not a clean low-risk trial either.

9. Common ROB 2 mistakes

Students and even experienced reviewers repeat the same errors. Avoiding them makes your appraisal more defensible in coursework, theses, and journal club.

• Using ROB 2 on observational intervention studies.
• Treating p > 0.05 for baseline differences as proof randomisation failed (small trials can have imbalance by chance).
• Ignoring selective reporting when outcomes in the registry differ from the paper.
• Scoring blinding as low risk for mortality outcomes when blinding is irrelevant to ascertainment.
• Conflating CONSORT reporting quality with low risk of bias.
• Giving 'low risk' because the journal is prestigious – journal tier is not a ROB 2 domain.

9. How StrataResearch applies ROB 2

StrataResearch detects RCT designs from manuscript structure and routes them to ROB 2-aligned domain scoring. Outputs include per-domain judgements, explicit limitations, and structured feedback tied to signalling-question concepts – not a generic essay about 'strengths and weaknesses'.

You can upload a PDF or paste text via quick analysis to see framework-mapped output on a real paper, then compare your manual ROB 2 worksheet against the structured report.

• Study-type routing reduces framework mismatch (RCT vs SR vs diagnostic).
• Domain-level output supports journal club slides and systematic review evidence tables.
• Saved projects let you revisit appraisals when the paper is revised or when supervisors request changes.

10. What to read next

ROB 2 is one layer in a full appraisal. Pair it with CONSORT for reporting, statistics guides for interpreting effect sizes, and GRADE if you are synthesising multiple RCTs in a review.

Interactive version (quizzes, walkthroughs) loads when JavaScript is enabled.