Selection bias in cluster trial

The large cluster randomised controlled trial by Davies et al suffers from recruitment bias as a result of poor allocation concealment,¹ which is crucial in both individually randomised trials and cluster trials. Allocation was not concealed from the people doing the recruiting, so there is a danger of recruitment bias. Indeed, the nature of the intervention—an educational package—would be likely to increase recruitment bias—a form of selection bias. This possibly occurred in this trial as more of the intervention practices recruited participants and they each recruited more participants than the control practices. Even if the numbers had been similar, we could not be sure that participants were similar in unknown characteristics.

This design flaw has been pointed out in the past,² and it can be dealt with by using someone who is blind to the allocation and study hypothesis to recruit participants.³ We can only, at best, treat data from this study as good observational data. A systematic review of cluster trials published in the BMJ and other leading medical journals some years ago found that 40% of them had some form of bias because of poor design.³ It seems that this poor design practice is still ongoing in BMJ reported cluster trials.