Background: In some general practice intervention trials, patients must be randomized in practices rather than individually, and this must be taken into account in the analysis.
Objectives: In this article we aim to show how failure to do this may lead to spurious statistical significance and CIs which are narrower than they should be, and to describe the use of summary measures for each practice as a simple method of analysis.
Method: The statistical issues are demonstrated by an example of a trial in general practice.
Discussion: The choice of unit of analysis will be most important where there are large numbers of patients recruited from each practice or a high degree of variability between practices.