RoB 2.0: A revised tool to assess risk of bias in randomized trials

Risk of Bias 2 (RoB 2) is a revised, more structured framework for judging bias in randomized trials—built to assess whether specific numerical results are likely affected by systematic error, not just whether a study was “well conducted.” The tool’s central shift is from broad judgments about trial quality to result-level assessments that map directly onto the mechanisms that can distort treatment effects. That matters because bias can vary across outcomes and across parts of a trial, so a single study-level label can miss the real risk to the estimate reviewers plan to use.

RoB 2 distinguishes bias from imprecision, reporting quality, and general study conduct. Imprecision reflects random error—often visible in wide confidence intervals—while bias reflects systematic error that can skew results. The framework also pushes back on the common conflation of “quality” with “bias.” A trial can be executed competently yet still be vulnerable to bias, such as when blinding participants and personnel is difficult (for example, in surgical trials). Likewise, poor reporting alone doesn’t automatically imply bias, since papers rarely contain exhaustive methodological detail; RoB 2 instead focuses on believability of the result.

The motivation for RoB 2 grew out of experience with the earlier RoB tool (RoB 1): inconsistent use, frequent “unclear” judgments, overly complex domains, weak fit for crossover and cluster designs, and the absence of an overall risk-of-bias judgment. RoB 2 was designed to be more comprehensive, more usable, and aligned with evolving bias science. It also aims to produce an overall risk-of-bias rating for each result, which can feed directly into sensitivity analyses.

A key design feature is that RoB 2 is “result-based.” Reviewers apply the tool to a specific numerical outcome estimate (e.g., a mean difference in depression score at 12 weeks in a cognitive behavior therapy trial), rather than to the study as a whole. The five mandatory domains correspond to major sources of systematic error: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in measurement of the outcome, and (5) bias in selection of the reported result. Domain labels are intentionally framed around causes of bias rather than older terms like selection or performance bias.

RoB 2 uses signaling questions with structured response options (yes/probably yes/no/probably no/no information) to drive domain-level judgments of low risk, some concerns, or high risk. An algorithm then combines domain judgments into an overall rating for the specific result, with rules such as “low overall” requiring low risk across all five domains. The framework also includes preliminary steps that force reviewers to specify the study type, the comparison, the outcome, the exact numerical result, and the effect of interest.

For deviations from intended interventions (Domain 2), the tool differentiates between the effect of assignment to intervention (intention-to-treat) and the effect of adhering to intervention (per-protocol). Deviations matter differently depending on which estimand is targeted, and blinding of participants and trial personnel can change whether deviations are considered biasing. The webinar also highlighted practical implementation within Cochrane workflows, including guidance resources on riskofbias.info, visualization support via Rob Viz, and the use of RevMan Web for proper implementation. Cochrane’s rollout is planned as a managed, slow pilot focused on new reviews, with support for teams adopting RoB 2 through a pilot program and training materials.