Table 1.

Defining characteristics of cluster randomised trials (CRT), non-randomised cluster trials (NCT), controlled before-and-after study (CBA), and before-and-after trials (BA)

	CRT	NCT	CBA	BA
Defining features	Randomisation of adequate number of clusters to allow statistical between-arm comparison	Non-random allocation of adequate number of clusters to allow statistical between arm comparison	Random or non-random allocation of small number of clusters - too few to allow statistical between-arm comparison	Before and after assessment of outcomes in the absence of a control group
Appropriate setting	Adequate resources, randomisation feasible	Adequate resources, randomisation is politically or logistically not possible	Resource limited evaluations or where number of clusters is naturally constrained (e.g. a district-level intervention in a province that only has 4 districts)	Mass media campaign where no unexposed group can be found
Detectable effect size given adequate sample size	Small	Small to moderate (depending on baseline comparability and temporal stability of outcome in control arm)	Moderate to large (depending on baseline comparability and temporal stability of outcome in control arm)	Large
Number of clusters	At least 4–6 clusters per arm, higher if effect size is small	At least 4–6 clusters per arm, higher if effect size is small	At least 2 clusters per arm, unless outcome is assessed repeatedly before and after	Study power is determined by number of participants and number of pre/post measures
Baseline measure of outcome of interest	Not required, but may increase study power and allow adjusting for imbalances	Required	Required	Required
Outcome assessment at multiple time-points	Not required, but may increase study power	Usually not required, but may increase study power	Desirable, required if there is only one cluster per arm	Required
Statistical analysis	Direct comparison between intervention and control	Direct comparison between intervention and control, by adjusting for baseline measure of outcome, by calculating change scores or by calculating the difference-in- difference	Comparison before versus after; control arm only serves to demonstrate absence of trends. Analysis of slope and intercept of trends in the outcome measure (if multiple pre/post measures are available)	Comparison before versus after. Analysis of slope and intercept of trends in the outcome measure (if multiple pre/post measures are available)
Special types	Stepped wedge design Trials using systematic allocation (“quasi-randomised”)	Controlled interrupted time series analysis Per-protocol analysis within a CRT		Interrupted time series analysis Pre/post adopter versus non-adopter analysis