Table 4 Comparing the SWD to the parallel group CRT: aspects of data analysis and interpretation

Effect estimate

-

(4a)

In an SWD, the unidirectional crossover strategy complicates the statistical analysis [6, 9, 11–13, 36]. Any temporal trends or fluctuations may (partially) invalidate the statistical analyses used by default [36]. If temporal trends or fluctuations are expected or found, a simple within-cluster analysis can provide a biased estimate of the treatment effect [9]. Calendar time is a potential confounder and should be adjusted for in the analysis [37]. Since incorporation of the effect of time requires a modelling approach, assumption-free analysis methods (nonparametric methods) cannot be used to analyse SWD data

+

(4b)*

In an SWD the effect measure of interest (e.g. difference in means or relative risk) can be calculated for each cluster, and the (in)consistency in effect estimates across clusters can be examined [37]

†

+

(4c)

In an SWD learning and decay effects over can be assessed over time, i.e. due to more experience with the intervention outcomes may become better over time. However it could also be that the intervention is well adopted just after implementation but ‘forgotten’ about after some time (e.g. if the intervention consists of new guidelines)

† Interim analyses

-

(4d)*

Interim analyses within an SWD are less efficient due to the unequal numbers of measurements under the different treatment arms during the study. For parallel group CRTs these numbers are more comparable during the entire trial period

† Number of measurements

+

(4e)

Collecting evidence on outcomes at several time steps may allow assessment of the (changes in) these outcomes during a longer follow-up period in those clusters that crossed over early in the study. This might benefit subsequent statistical and health economic analyses, for example, when extrapolating beyond the trial horizon

† Unrelated studies

+

(4f)

Collecting health economic evidence in an SWD might also provide insight into general barriers and facilitators to implementation and into changes in cost-effectiveness when moving from a clinical to a routine care setting. In an SWD more evidence on implementation is collected than in parallel group CRTs, as the process of implementing the new intervention can be observed during the study, for all clusters, as opposed to parallel group CRTs where half of the clusters do not get the intervention during the study, and studying changes in implementation over time is more limited. This additional evidence might be valuable in the design and execution of other studies, for example, studies on other interventions in the same disease area