How do I assess risk of bias with RoB 2?
Assess one result at a time, not one study at a time. For each result you work through five domains, answer the signalling questions in each from the trial report, and let those answers drive a domain-level judgement of low risk, some concerns, or high risk. The overall judgement is governed by the worst domain — not by a total, because there is no total.
The unit of assessment is a result, not a study
This is the change people most often miss, and it drives everything else. A single trial can be at low risk of bias for its primary outcome and high risk for a secondary one — different amounts of missing data, different blinding of the assessor, different degrees of selective reporting.
So "we assessed risk of bias for the 14 included studies" is the wrong sentence. You assessed it for each result you are pooling. If you carry one judgement per study into your synthesis, you are attaching an appraisal of the primary outcome to outcomes it was never about.
The five domains
- Randomisation process
- Was the sequence random, was allocation concealed, and do the baseline characteristics suggest a problem with either?
- Deviations from intended interventions
- Depends on your question: are you asking about assignment to the intervention, or about adhering to it? The answer changes which deviations matter.
- Missing outcome data
- Is enough of the data present, and where it is missing, could the reason for missingness depend on the outcome itself?
- Measurement of the outcome
- Was the method appropriate, was the assessor aware of the assignment, and could that awareness have influenced a subjective outcome?
- Selection of the reported result
- Was the analysis pre-specified, or does the reported result look chosen from several that were run? A registered protocol is the evidence here.
It is not a score — do not add it up
The three judgements are ordered categories, not numbers. Counting how many domains were "low" and averaging, or converting to points and summing to a total out of five, destroys exactly the information the tool exists to preserve: which specific threat to validity applies, and how it would bias this result.
A result is at high overall risk if any single domain is high, or if enough domains raise concerns that the result is doubtful taken together. One fatal flaw is fatal regardless of how clean the other four domains look.
Record the rationale and the quotation from the report that drove each signalling answer. Two reviewers will disagree, and the disagreement is only resolvable if both wrote down what they saw.
What to do with the judgements
Risk of bias feeds the certainty rating for the outcome, and it can justify a sensitivity analysis restricted to low-risk results. If the effect disappears when you exclude high-risk results, that is a finding, and it belongs in the results rather than in a limitations paragraph.
Where this answer stops
The tool assesses what the trial report lets you see. A trial can be conducted impeccably and reported so thinly that several domains land on "some concerns", and that is the correct answer even though it is a judgement about the paper as much as about the study.