Table 2 Comparing the SWD to the parallel group CRT: aspects of study design and preparation

Equipoise

~

(2a)*

An SWD may be used in a situation where there is a slight preference for the experimental treatment [6, 12, 13]. At the same time, equipoise remains a necessary requirement for all studies including SWDs. Unlike literature suggests, SWDs where equipoise is disturbed from the start should not be undertaken

† Social value

~

(2b)*

A study with an SWD may benefit fewer individuals after completion since it typically takes longer to complete. However, this disadvantage may be offset by faster implementation following the SWD

† Implemen-tation decisions

-

(2c)*

If evidence on the cost-effectiveness of a new intervention is lacking, collecting this evidence may be valuable to support implementation decisions. However, deimplementation following a negative result has worse consequences for SWDs than for parallel group CRTs

Disease

-

(2d)

An SWD is not the design of choice for a study in a rapidly spreading disease. A pandemic requires an efficient, short-term design and analysis [53]

Study design

~

(2e)

An SWD might be logistically easier because of the phased implementation of the intervention rather than implementation of the intervention at (often) half of the clusters simultaneously in a parallel group CRT [5, 6, 8, 9, 11–13, 16, 34, 54]. However, variations of the parallel group CRT have also been mentioned which give the opportunity of phased implementation [18, 19, 21]

†

+

(2f)

The SWD offers the possibility to assess cost-effectiveness over time when the uptake of the intervention is difficult or slow. Even though statistical power to assess time trends may be relatively low, compared to parallel group CRTs the SWD allows a more accurate assessment of the actual long-term costs and effects after implementation barriers have been overcome

~

(2g)

An SWD may take longer to complete [5, 6, 9, 12, 13, 16, 18, 53–55]

†

-

(2h)*

In an SWD it will be difficult to compare more than 2 treatments whereas in a parallel group CRT more treatment arms can be added rather easily. Implementing more than 2 treatments may also be of questionable use in an SWD

Sample size

~

(2i)*

An SWD may require fewer clusters than a parallel group CRT [5, 9, 12–14, 16, 17, 21, 34–36]

~

(2j)*

An SWD may require a larger total number of subjects and/or measurements than a parallel group CRT, depending on cluster size, intracluster correlation (ICC) and number of measurement periods [5, 34, 36]

†

~

(2k)

The effect of incorporating interim analyses on the total sample size for an SWD is not clear yet

Power

~

(2l)

An SWD may have more power than a parallel group CRT due to an increase in the amount of data collected and the possibility of within-cluster comparisons [5, 6, 9, 12, 13, 16–21, 35, 37, 54, 56]

+

(2m)

The ICC has only a minimal effect on power within an SWD (at least in the cross-sectional design) [9, 11, 13, 35, 48]

Participation

+

(2n)

Clusters may be more willing to participate in an SWD as each cluster will switch to the new (promising) intervention during the study [6, 13, 23]

Timing of outcome

-

(2o)

The time between steps in an SWD should be long enough to detect a treatment effect [5, 9, 12, 13, 18]. Hence, if it takes a relatively long time before a treatment effect can be detected, the SWD may require a much longer time period to be completed than the parallel group CRT