Introduction/Objective: To examine problems of sample size estimation in randomised controlled trial in group randomised studies and comparisons of care delivered by different types of health professional. In a typical randomised controlled trial the interventions are randomised to patients individually. Sometimes interventions are randomised to groups of patients for example the implementation of clinical guidelines. In this trial design the outcomes for patients within the same group are not statistically independent. In some randomised studies each intervention is delivered by a different type of health professional. An example of this type of study would be the comparison of chiropractic manipulation and physiotherapy in the treatment of lower back pain. Outcomes for patients treated by the same health professional will tend to be correlated. Sample size estimates in these two types of study will need to be increased to adjust for the lack of independence in outcome often measured by the intra-cluster correlation coefficient (ICC).

Methods: The adjustment of sample size calculation for these types of studies will be described. At the design stage a major problem is to obtain estimates of the ICC. Data from randomised and observational studies in primary care will be examined to obtain estimates of the ICC. Practical strategies will be considered for estimating sample size where precise estimates of the ICC are not available.

Results: Empirical evidence suggests that the ICC is generally small (0.05) and rarely exceed 0.1.

Discussion: Provided the numbers of patients in each randomised group or treated by the same health professional are small, sample size inflation will not be great. In the absence of precise estimates of the ICC, a sensitivity analysis on a range of estimate of the ICC may inform sample size estimation.