Table 1 EPOC study designs considered for inclusion

Patient randomized controlled trials

(P-RCT):

The individual patients are allocated by random to the intervention or control group. Individual randomisation facilitates equally distributed patient characteristics and comparability. Only the exposure to the intervention should be the factor that distinguishes between both groups.

Cluster randomized controlled trials (C-RCT):

This is a robust study design that prevents contamination of professionals by randomising groups of professionals (i.e. different practices, wards or hospitals). However, this means the fundamental assumption of independence is violated because patients within a cluster are more likely to respond in a similar manner. This lack of independence, statistically called “intracluster correlation,” also means a specific adjustment for clustering effects is required to assure comparability with individually randomized trials.

Non-randomized controlled trials (CCTs):

Patient or cluster trials where allocation to experimental and control groups is quasi-random (i.e. alternated allocation).

Controlled before and after studies (CBAs):

CBAs are experimental studies with two or more control groups compared with one or more experimental groups but allocation is not random. Data is collected on the control and intervention groups before the intervention is introduced and then further data is collected after the intervention has been introduced. The reliability of the intervention effect is questionable because there may be unidentified differences between the experimental intervention and control groups which may have modified the observed effect. Note: EPOC has recently changed the policy about inclusion of CBA studies with only one intervention site. Specific details about design criteria can be found at the website (http://www.epoc.cochrane.org)

Interrupted time series designs (ITS):

This represents a robust method of measuring the effect of an intervention as a trend over time. It is a useful design when recruitment of a control cohort is impractical, e.g. due to changes in hospital policy. Three or more data points are collected before and after the intervention as a minimum standard. The intervention effect is measured against the pre-intervention trend.