ROBINS-I V2

Assessing risk of bias in the results of follow-up (cohort) studies that compare the health effects of interventions and were not assigned by randomization. ROBINS-I V2 is the recommended risk-of-bias tool for non-randomized studies of interventions in Cochrane systematic reviews.

• Cohort studies (the primary design ROBINS-I V2 is currently scoped for)
• Other follow-up designs that compare intervention groups without random assignment
• Non-randomized studies of interventions (NRSI) where confounding is the dominant source of bias

ROBINS-I V2 organises the assessment around six bias domains. For each domain, reviewers answer signalling questions that lead to a domain-level judgement, and the domain judgements are then combined into an overall risk-of-bias judgement for the result being assessed. The domain names below match the official ROBINS-I V2 instrument; for the actual signalling questions and detailed guidance, consult the official tool linked under Reference Documents.

1. 1.Bias due to confounding

   Considers whether differences in baseline characteristics between intervention groups, or in characteristics that change after the start of the intervention but are influenced by it, could distort the estimated effect. Confounding is typically the dominant source of bias in non-randomized studies and is the domain in which ROBINS-I V2 differs most from tools designed for randomized trials.

2. 2.Bias arising from classification of intervention

   Considers whether participants were correctly classified into the intervention groups being compared. Misclassification can occur when intervention status is unclear, changes over time, or is recorded retrospectively from imperfect sources.

3. 3.Bias in selection of participants into the study

   Considers whether the way participants were selected into the study, or into the analysis, could be related to both the intervention and the outcome. This includes prevalent-user bias and immortal-time bias, which V2 explicitly addresses with additional signalling questions across multiple domains.

4. 4.Bias due to missing data

   Considers the extent of missing data on outcomes, interventions, and confounders, and whether the strategy used to handle missing data could distort the result.

5. 5.Bias arising from measurement of the outcome

   Considers whether outcome measurement methods were appropriate, comparable across groups, and unlikely to be influenced by knowledge of the intervention received.

6. 6.Bias in selection of the reported result

   Considers whether the reported effect estimate was selected from among multiple analyses of the same data on the basis of the result, including selective reporting of outcomes, populations, or analyses.

Use ROBINS-I V2 when you are assessing risk of bias in an individual non-randomized cohort study of an intervention as part of a systematic review or evidence synthesis. The tool is most appropriate when the studies you are appraising have a follow-up design and when the comparison of interest is between intervention groups that were not randomly assigned. Each ROBINS-I V2 assessment should be performed for a specific result of interest (a particular outcome and effect estimate), not for the study as a whole. Reviewers should pre-specify the target trial that the study is attempting to emulate, and should make their judgements relative to that target trial rather than to an idealised observational study.

When not to use it

ROBINS-I V2 is not appropriate for randomized trials, where RoB 2 should be used instead. It is also not the right tool for assessing reviews of exposures (rather than interventions); for those, use ROBINS-E (Higgins et al., 2024). Do not use ROBINS-I V2 to assess case-control studies or case series without careful consideration; the tool is currently scoped to follow-up designs. If you are assessing the methodological quality of a systematic review (rather than the risk of bias of a primary study), use AMSTAR 2 instead.

ROBINS-I V2 sits alongside two other Cochrane risk-of-bias tools: RoB 2 (for randomized trials) and ROBINS-E (for non-randomized studies of exposures, published in 2024). All three tools share a common structure of signalling questions feeding domain-level and overall judgements, but they differ in the specific bias domains they assess and the algorithms used to derive judgements. For systematic reviews that include both randomized and non-randomized studies of the same intervention, it is appropriate to apply RoB 2 to the trials and ROBINS-I V2 to the non-randomized studies, then synthesise the results with appropriate consideration of the differing risk-of-bias assessments.

• RoB 2 vs ROBINS-I: which risk-of-bias tool for your study?
• What's new in ROBINS-I V2

CoRATES presents ROBINS-I V2 as a structured digital workflow rather than a spreadsheet or PDF form. Each signalling question is rendered in context with its response options, reviewers can attach rationale and quotes from the source study to each judgement, and the domain-level and overall judgements are computed automatically from responses using the official algorithms. Multiple reviewers can complete independent assessments of the same study in parallel; CoRATES then surfaces the disagreements in a side-by-side reconciliation view so reviewers can resolve them before locking the assessment. Completed assessments can be exported as publication-ready risk-of-bias visual summaries.

• ROBINS-I V2 official tool and guidance (riskofbias.info)
• Cochrane Bias Methods Group: ROBINS-I overview
• Sterne et al. (2016): Original ROBINS-I publication, BMJ
• Cochrane Handbook Chapter 25: Assessing risk of bias in non-randomized studies

The original ROBINS-I tool was published in 2016 (Sterne et al., BMJ). ROBINS-I V2 was released in November 2024 as a substantive update. Notable structural changes from V1 include: (1) the number of bias domains was reduced from seven to six, with the V1 "deviations from intended intervention" domain removed and its concerns redistributed; (2) signalling questions now use "strong" and "weak" response options to better reflect graded judgements; (3) algorithms were added that map signalling question responses to suggested risk-of-bias judgements at the domain level; (4) a new triage section helps reviewers identify studies at critical risk of bias before completing the full assessment; (5) the specification of the effect of interest was refined; and (6) explicit signalling questions were added to address immortal-time bias across multiple domains. V2 is currently scoped specifically for follow-up (cohort) studies, with extensions to other non-randomized designs anticipated.

• Assessing the study as a whole rather than a specific result. ROBINS-I V2 is intended to be applied per result (per outcome and per effect estimate); applying it to the entire study can mask domain-specific risks that only affect particular outcomes.
• Failing to specify the target trial. Without an explicit hypothetical randomized trial that the study is trying to emulate, judgements about confounding and selection bias become subjective and inconsistent across reviewers.
• Confusing ROBINS-I V2 (interventions) with ROBINS-E (exposures). The two tools have different scopes and assumptions; using the wrong one will produce assessments that do not align with the published guidance.
• Treating "Critical" risk of bias as a default for any non-randomized study. Critical is reserved for studies whose results should not be used in synthesis at all; most non-randomized studies will fall into Moderate or Serious.
• Overriding the algorithm-suggested judgement without recording rationale. The V2 algorithms are an important consistency aid; deviations should be documented so reviewers (and readers of the eventual review) can audit the reasoning.
• Carrying over V1 habits, particularly the now-removed "deviations from intended intervention" domain. Reviewers familiar with the original ROBINS-I should re-orient to the V2 six-domain structure and the explicit immortal-time bias signalling questions.

What does ROBINS-I stand for?

ROBINS-I is an acronym for "Risk Of Bias In Non-randomized Studies - of Interventions". The "V2" in ROBINS-I V2 indicates Version 2, the major revision released in November 2024.

How is ROBINS-I V2 different from the original ROBINS-I?

ROBINS-I V2 reduces the number of bias domains from seven to six (removing the "deviations from intended intervention" domain), introduces algorithms that map signalling-question responses to suggested risk-of-bias judgements, adds "strong" and "weak" response options for graded judgements, includes a new triage section for identifying studies at critical risk of bias, refines how the effect of interest is specified, and adds explicit signalling questions for immortal-time bias. V2 is also currently scoped specifically to follow-up (cohort) studies.

Should I use ROBINS-I V2 or RoB 2?

Use RoB 2 for randomized trials and ROBINS-I V2 for non-randomized cohort studies of interventions. If your systematic review includes both designs, apply each tool to the appropriate subset of studies and synthesise the results with awareness of the differing risk-of-bias structures.

Is ROBINS-I V2 the same as ROBINS-E?

No. ROBINS-I V2 assesses risk of bias in non-randomized studies of interventions. ROBINS-E (for Exposures) is a separate tool, published in 2024, for non-randomized studies of exposures rather than interventions. The two tools have different scopes and should not be substituted for each other.

Can ROBINS-I V2 be applied to case-control studies?

ROBINS-I V2 is currently scoped to follow-up (cohort) study designs. Applying it to case-control studies or other non-cohort designs requires careful judgement and may not align fully with the published guidance. Future extensions of ROBINS-I to additional study designs are anticipated.

Where are the official ROBINS-I V2 signalling questions and guidance?

The official ROBINS-I V2 instrument, signalling questions, and guidance documents are hosted at riskofbias.info. CoRATES does not reproduce the official content; reviewers should consult the official sources alongside their assessments. See the Reference Documents section above for direct links.

Does CoRATES automatically score ROBINS-I V2 assessments?

CoRATES applies the V2 algorithms to map signalling-question responses to domain-level and overall risk-of-bias judgements, and updates them in real time as reviewers answer questions. CoRATES does not modify the official scoring algorithms.

How does CoRATES support multi-reviewer ROBINS-I V2 assessment?

Multiple reviewers can complete independent ROBINS-I V2 assessments of the same study in parallel. CoRATES then presents a side-by-side reconciliation view that highlights disagreements at the signalling-question, domain, and overall-judgement level, so reviewers can discuss and resolve them before locking the assessment.

Members of the Cochrane Bias Methods Group and the Cochrane Non-Randomised Studies Methods Group, led by Jonathan Sterne and Julian Higgins. Development of ROBINS-I V2 was funded in part by Medical Research Council (MRC) grant MR/M025209/1.

• Sterne JA, Hernan MA, Reeves BC, et al. (2016). ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ 2016;355:i4919. View
• Higgins JPT, Morgan RL, Rooney AA, et al. (2024). A tool to assess risk of bias in non-randomized follow-up studies of exposure effects (ROBINS-E). Environment International, 186, 108602. View