A sample size of 93 patients in each group was required to detect a five point change in the functional status score and 20% change in vaccination rate (power of 80%, a =0.05). This estimate was inflated to allow for potential clustering as follows; inflation factor=1+(n-1),[w1] where (rho) is the intraclass correlation and n is the mean size of the cluster. On the basis of previous surveys, =0.05 was selected. The estimate of 93 was inflated by 1.3. Allowing for 15% dropout, we aimed to recruit 42 doctors and 280 patients.

1. Donner A, Birkett N, Buck C. Randomization by cluster; size requirements and analysis. Am J Epidemiol 1981;114:906-14.