Strata Academy

RoB 2 Checklist: How to Score Cochrane Risk of Bias

Score all five Cochrane domains with signalling questions — low, some concerns, or high risk — plus when to use RoB 2 instead of ROBINS-I

• Five domains – each judged separately before overall risk.
• Reporting quality (CONSORT) ≠ low risk of bias.
• Subjective outcomes: Domain 4 (blinding of assessment) matters most.

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.

Tip: If the paper is not an RCT – for example a cohort study of a treatment policy – use ROBINS-I instead. Applying ROB 2 to the wrong design is one of the most common student errors.

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)

Note: Authors sometimes label a study 'randomised' when allocation was quasi-random (e.g. alternate days). Read the methods carefully. If true randomisation with concealed allocation is not credible, consider whether ROBINS-I is more appropriate.

Practice domain judgements in the free RoB 2 assistant, or upload an RCT PDF for RoB 2–aligned domain scoring. Open RoB 2 checklist tool · Appraise an RCT PDF

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?

Tip: Work domain by domain. Students who jump straight to an overall 'gut feeling' almost always miss missing-data and selective-reporting issues.

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

Tip: Use the framework picker guide or ROB 2 assistant tool to practise before appraising a full PDF.

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.

Note: This example is illustrative. Your judgement must follow the official signalling questions for the specific trial you are appraising.

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.

10. Pairing ROB 2 with CONSORT

Complete a CONSORT pass before ROB 2 domain scoring. CONSORT gaps force 'unclear' or 'some concerns' in ROB 2 when randomisation, blinding, or outcome pre-specification are not reported.

Item 8–10 (randomisation and blinding) feed ROB 2 Domains 1, 2, and 4. Item 12 (sample size and interim analyses) links to Domain 3 and 5. The flow diagram supports attrition assessment in Domain 3.

Writing 'CONSORT complete therefore low ROB' is a common marker comment — reporting transparency and internal validity are related but not identical.

11. ROB 2 in systematic reviews and GRADE

Reviewers apply ROB 2 to each included RCT, then summarise risk of bias in forest plots or tables. Consistency matters — one high-risk trial driving a pooled effect warrants sensitivity analysis.

GRADE downgrades certainty for risk of bias when a large contribution to the effect estimate comes from studies rated high or unclear ROB 2 risk.

Publication bias (funnel plot asymmetry, missing registered trials) is a GRADE concern separate from ROB 2 — do not conflate the two downgrades.

When writing SoF tables, state explicitly which ROB 2 domains drove any 'serious' or 'very serious' risk-of-bias downgrade.

12. 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.

13. 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.

Frequently asked questions

Is there a RoB 2 checklist?

Yes — RoB 2 is Cochrane's domain checklist for randomised trials. Work through five domains with official signalling questions, assign low / some concerns / high per domain, then derive an overall risk-of-bias judgement for the result you are scoring.

How do you score RoB 2?

Answer the signalling questions for each domain (yes / probably yes / probably no / no / no information), map answers to a domain judgement with the Cochrane algorithm, then combine domain ratings into an overall low, some concerns, or high risk judgement for that outcome.

What are the five ROB 2 domains?

Randomisation process; deviations from intended interventions; missing outcome data; measurement of the outcome; and selection of the reported result. Each domain is judged separately before an overall rating.

What are ROB 2 signalling questions?

Official yes/no/partial questions within each domain that guide your judgement — for example, whether allocation was concealed or outcome assessors were blinded. Answers map to low risk, some concerns, or high risk for that domain.

When should I use ROB 2 instead of ROBINS-I?

Use ROB 2 for randomised trials (and some cluster-RCT designs per Cochrane guidance). Use ROBINS-I for non-randomised studies of interventions, and QUADAS-2 for diagnostic accuracy studies.

Interactive walkthroughs and quizzes load when JavaScript is enabled — the checklist and tables above are fully readable without it.