Risk of Bias Assessment: RoB 2 vs. ROBINS-I
July 9, 2026 · Dr. Amara Chen
One of the most common — and most avoidable — errors in a submitted systematic review is applying the wrong risk-of-bias tool to the included studies. It happens more than it should, usually because a review includes a mix of randomized and non-randomized studies and the same tool gets applied to both. The two are not interchangeable, and reviewers who work with evidence synthesis daily will catch the mismatch immediately.
Why the tool has to match the study design
Risk-of-bias tools are built around the specific ways bias can enter a particular study design. Randomized controlled trials and non-randomized (observational) studies fail in structurally different ways — randomization itself controls for a category of confounding that observational studies simply don't have. A tool built for one doesn't transfer cleanly to the other.
RoB 2: for randomized trials
RoB 2 (the Cochrane Risk of Bias tool, second version) is built specifically for randomized controlled trials. It assesses bias across five domains: the randomization process itself, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each domain gets a judgment of low risk, some concerns, or high risk, and these roll up into an overall study-level judgment. RoB 2 assumes randomization occurred — applying it to a non-randomized study doesn't make sense, because the very first domain it assesses isn't applicable.
ROBINS-I: for non-randomized studies
ROBINS-I (Risk Of Bias In Non-randomized Studies of Interventions) exists because observational studies need a tool that accounts for confounding and selection bias in ways RCTs generally don't. It assesses seven domains, including confounding and selection of participants into the study — domains that don't appear in RoB 2 at all, because randomization is designed to handle exactly those problems in a trial. ROBINS-I judgments run on a five-point scale from low to critical risk of bias, reflecting that observational evidence typically carries more, and more varied, sources of bias than a well-run trial.
The practical decision
The choice is determined by study design, not by preference or convenience:
Randomized controlled trials, including cluster-randomized and crossover trials — use RoB 2.
Cohort studies, case-control studies, and other non-randomized designs — use ROBINS-I.
A systematic review that includes both types of studies needs to apply both tools, matched correctly to each study, and should present the results separately rather than pooling risk-of-bias judgments across fundamentally different designs. This also affects how findings get synthesized: mixing high-confidence RCT evidence with lower-confidence observational evidence in a single pooled estimate, without separating or clearly weighting them, is itself a methodological weakness worth flagging in the review's limitations.
What this affects downstream
Risk-of-bias judgments feed directly into GRADE certainty ratings, which in turn shape how confidently a review's conclusions can be stated. Getting this step wrong doesn't just risk a reviewer comment — it can understate or overstate the actual reliability of the evidence base the review is built on. It's worth the extra care at this stage, because everything downstream depends on getting it right the first time.