How to Read a RoB 2 Risk-of-Bias Assessment

RoB 2 is the current Cochrane tool for rating how much a randomized trial's design and conduct might have biased a particular result. It works through five domains, from the randomization process to selective reporting, and assigns each a rating of low risk, some concerns, or high risk. Because the overall judgment follows the weakest domain, a single serious flaw can lower confidence in a result no matter how clean the rest of the trial was.

What RoB 2 is for

RoB 2 is the current Cochrane tool for judging how much the design and conduct of a randomized trial could have biased its result. It does not score how well a study is written or how relevant it is to your question. It asks something narrower: given how this trial was actually run, how much should you worry that its estimated effect is distorted by the way patients were assigned, treated, measured, or selected for reporting.

One habit is worth building early. RoB 2 rates a specific result, not the whole paper. The same trial can be low risk for its primary outcome and high risk for a secondary one, because missing data or measurement problems can differ from outcome to outcome.

The five domains

RoB 2 walks through five domains in order. First, bias arising from the randomization process, which covers whether the sequence was truly random and whether the upcoming assignment was concealed. Second, bias due to deviations from the intended interventions, which is where blinding and what people did after assignment live. Third, bias due to missing outcome data. Fourth, bias in measurement of the outcome, which turns on who assessed it and whether they knew the assignment. Fifth, bias in selection of the reported result, meaning cherry-picking among analyses or outcomes.

Each domain is reached through a set of signalling questions with structured answers such as yes, probably yes, probably no, no, or no information. Those answers feed an algorithm that produces the domain-level judgment, which keeps the process transparent rather than a matter of taste.

The two questions a trial can answer

The second domain forces a choice many readers skip. RoB 2 asks whether you care about the effect of being assigned to a treatment or the effect of actually adhering to it. These are different questions, and they change which deviations count as bias.

The effect of assignment is the intention-to-treat spirit and is usually the more robust target. The effect of adhering is closer to a per-protocol question and is harder to estimate without reintroducing the very confounding that randomization removed. A good assessment states which one it is judging.

How the judgments roll up

Each domain earns one of three labels: low risk, some concerns, or high risk. The overall judgment is not an average of the five. It is driven by the worst domain, roughly so that any single domain at high risk, or several domains raising some concerns, pushes the overall rating toward high risk.

This is deliberate. A trial can be careful in four domains and undone by a fifth, and averaging would hide that. The rule protects readers from being reassured by an otherwise tidy study that has one disqualifying weakness.

Why the traffic light is not the whole story

Reviewers present RoB 2 as a colored table, and it is tempting to read only the colors. The reasoning behind each judgment matters more. A rating of some concerns with a clear explanation tells you exactly where to look; a low risk rating with no supporting detail tells you almost nothing.

Empirical work long predating the tool showed that trials with inadequate allocation concealment produced exaggerated effect estimates on average. That is precisely the failure the first domain is built to catch, which is why the explanation attached to a domain is worth reading rather than skimming past the color.

How to use it as a reader

Read the domain judgments before you look at the effect size, then decide whether the flaws could plausibly move the result in a direction that matters to you. Ask whether a high risk rating on measurement, for example, would inflate a subjective outcome more than an objective one.

RoB 2 does not hand you a number to subtract from the effect. It gives you a structured reason to be more or less confident. Treat it as a map of where a result might be soft, not as a final verdict on the whole paper.