Table 3.

Criteria for inclusion in meta-analysis of moderators

Criteria	Necessary for inclusion in meta-analysis confirming moderator effects	Necessary for inclusion in meta-analysis exploring moderator effects	Criteria for the judgment of yes	Exceptions
Was the analysis a- priori	√		Mention of explicit hypothesis planned in protocol stating which sub-groups will be tested for which outcome	Criterion is not fulfilled in cases where the protocol includes a considerably large set of stated hypotheses or vague hypotheses (e.g. psychological factors will interact with treatment allocation')
Was selection of factors for analysis clinically plausible and either or both:	√		A description of theoretical background, or reference to other published evidence leading to the hypothesis	Is not fulfilled in cases where the meta-analyst considers the theory/evidence to be weak, but should not form reason for exclusion.
i) Theory based
ii) Evidence based
Were moderators measured prior to randomisation?	√	√	Specific statement that baseline measurement occurred prior to randomization	Not applicable for baseline factors that do not change over time, such as gender, or for cluster randomization.
Adequate quality of measurement of baseline factors	√	√	If there is published evidence to support good measurement properties of measurements for target population, according to meta-analysts' protocol.	Is not fulfilled where there is inadequate variability in baseline measure.
Contains an explicit test of the interaction between moderator and treatment	√	√	Ideally, Report a pooled effect size with 95% confidence intervals. Other acceptable analysis includes regression etc.	Not fulfilled when sub-groups are tested separately, or in excessive multiple testing.