Statistical analyses
Full details of the analysis plan have been published.20 We used
intention to treat analyses as our main analyses, with missing
data at baseline dealt with by including an indicator variable
for those with missing data, as suggested by White and
others.44-46 We also did a series of sensitivity analyses to test the
assumptions that we had made regarding the nature of missing
baseline and outcome data (see supplementary table A and
detailed analysis plan for detailed discussion of these
assumptions and the sensitivity analyses20).44 45 We used
multilevel regression models to account for the clustering