Conclusions
Systematic reviews of complex behaviour change interventions in healthcare may include a heterogeneous set of studies in terms of content, context, trial design and setting. The measures of behaviour change may also vary which leads to difficulty in attempts to synthesise the data, as well as increased heterogeneity.
In this paper we have presented 4 different methods for combining behavioural outcome measures from trials, described the strengths and weaknesses of each method, and the problems inherent with combining heterogeneous outcome measures with mixed levels of clustering. Each of the methods presented has advantages and disadvantages, summarised in table 4, and we recommend that reviewers chose their methods carefully based on the needs of their review, and plan methods and data conversion policies in advance to avoid selective reporting. We observed that for our data, conclusions would remain robust regardless of the methods of analysis chosen; however the estimated magnitude of the treatment effect varied quite markedly according to the method chosen. We view the methods presented as useful when trying to convert all outcome measures to the same scale and to provide an overall summary, but results should be interpreted extremely cautiously given the limitations. We would recommend that results are used as an aid in summarising the evidence and generating future hypotheses rather than to infer future effects.