Methodological bias in cluster randomised trials