Clustered clinical trial designs, in which all participants belong to a cluster of individuals whose outcomes are expected to be correlated, are well established. However, many clinical trials involve partially clustered data, in which only some individuals belong to a cluster. Common examples of such trials include group-based treatment designs in which an intervention is delivered in groups compared to an ungrouped control arm.

Although group-based treatment designs are relatively well known, partial clustering also occurs in fields such as neonatology, where participants include infants from both singleton and multiple births, and ophthalmology, where one or two eyes per participant may need treatment. Such trials may involve interventions at the cluster or individual level, and designs vary in how the clustered observations are randomized. This talk focuses on the implications of partial clustering for sample size requirements and analysis of the range of trial designs used in practice. Recent research on the efficiency of different partially clustered designs and the performance of clustered analysis methods, including mixed-effects models and generalized estimating equations, will be presented.