Potential source of bias	Assessment criteria
Was recruitment bias adequately prevented?	Low risk of bias: Individuals were not recruited to the trial after the clusters had been randomised.
	High risk of bias: Individuals were recruited to the trial after the clusters had been randomised (the knowledge of whether each cluster is an ‘intervention’ or ‘control’ cluster could affect the types of participants recruited).
	Unclear risk of bias: Insufficient information to permit judgement.
Were baseline imbalances (in terms of either the clusters or the individuals) adequately addressed?	Low risk of bias: The randomised groups were similar at baseline; or the randomised groups were imbalanced at baseline but finally controlled for at the design (such as using stratified or pair matched randomisation of clusters) or analysis stage of the study.
	High risk of bias: There were baseline imbalances between the randomised groups, but finally they were not controlled for at the design or analysis stage of the study.
	Unclear risk of bias: Insufficient information to permit judgement.
Were loss of clusters and participants adequately addressed?	See Appendix 2: "Incomplete outcome data" for criteria of how we will assess this domain.
Was the study analysed by correct statistical methods (i.e. taking the clustering into account)?	Low risk of bias: The cluster‐randomised trial was analysed by correct statistical methods, taking the clustering into account. Ways to avoid unit‐of‐analysis errors in cluster‐randomised trials are (see Cochrane Handbook 16.3.3, Higgins 2011b): to conduct the analysis at the same level as the allocation; to conduct the analysis at the level of the individual while accounting for the clustering in the data. Such an analysis might be based on a ‘multilevel model’, a ‘variance components analysis’ or a ‘generalized estimating equations (GEEs)’, among other techniques.
	High risk of bias: The cluster‐randomised trial was analysed by incorrect statistical methods, not taking the clustering into account. Such analyses tend to create a ‘unit of analysis error’ and produce over‐precise results (the standard error of the estimated intervention effect is too small) and P values that are too small. Although they do not lead to biased estimates of effect, if they remain uncorrected, they will receive too much weight in a meta‐analysis.
	Unclear risk of bias: insufficient information to permit judgement.