The Cluster Randomised Trial and beyond: achievements and future research- Part 3

Tim Peters (University of Bristol)
Funding issues associated with statistical designs in primary care research

I will present some of the process and results from undertaking a meta-analysis in collaboration with behavioural psychology colleagues at the Cambridge Primary Care Unit. The example, which contained several interesting statistical issues, was later taken up by NICE as part of public health guidance on smoking in pregnancy, and was developed into a short course for students and researchers. It was a systematic review and meta-analysis of self-help smoking cessation interventions in pregnancy. This is an area where there is sparse data, arising from a low proportion meeting the binary endpoint of quitting smoking, and where there is considerable effect heterogeneity. The use of sensitivity analysis was shown to be beneficial in partly understanding this. We opted for a DerSimonian and Laird random effects model approach. In more recent general work with statistical colleagues we have begun developing methods for handling missing statistics in multiple Cochrane reviews, where the relationship between these two approaches will be important to exploit.