Separation of blinding.

In the earlier version, biases related to blinding of participants, personnel and outcome assessors were all assessed within a single domain (although they may have been assessed separately for different outcomes). In the revised tool, bias related to blinding of participants and personnel is assessed in a domain separately from bias related to blinding of outcome assessment.

Nature of the judgement.

The judgements are now expressed simply as ‘Low risk’, ‘High risk’ or ‘Unclear risk’ of bias. The questions have been removed, along with the responses ‘Yes’ indicating low risk of bias and ‘No’ indicating high risk of bias.

Minor rewording. 

The items have been renamed in RevMan with the removal of question-based judgements:

Adequate sequence generation? becomes Random sequence generation.

Allocation concealment? becomes Allocation concealment

Blinding? becomes Blinding of participants and personnel and Blinding of outcome assessment.

Incomplete outcome data addressed? becomes Incomplete outcome data.

Free of selective reporting? becomes Selective reporting.

Free of other bias? becomes Other bias.

Insertion of categories of bias.

The revised tool clarifies the category of bias within which each domain falls: selection bias (random sequence generation and allocation concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective reporting) and other bias.

Reconsideration of eligible issues for ‘other bias’, including early stopping of a trial.

The guidance for the ‘other bias’ domain has been edited to strengthen the guidance that additional items should be used only exceptionally, and that these items should relate to issues that may lead directly to bias. In particular, the mention of early stopping of a trial has been removed, because (i) simulation evidence suggests that inclusion of stopped early trials in meta-analyses will not lead to substantial bias, and (ii) exclusion of stopped early trials has the potential to bias meta-analyses towards the null (as well as leading to loss of precision).