Inference and Optimal Design for Longitudinal Cluster-Randomized Clinical Trials Given a Small Number of Clusters with Application to a Serious Mental Illness Intervention Study

• Activity Number: 71 - Statistical Methods for Personalized Medicine
• Type: Contributed
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 5:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): CHAE RYON KANG* and DI ZHANG
• Companies: University of Pittsburgh and University of Pittsburgh
• Keywords: Longitudinal cluster randomization controlled trials; Three-level hierarchical structure ; Small number of clusters; Doubly robust estimator

Abstract

Longitudinal cluster-randomization design has been frequently used in pragmatic clinical trials. It features in three-level hierarchical structure with multiple observations per participant nested within a cluster. The existing literature on statistical methods have focused on the two-level or three-level hierarchical structure without longitudinally measured data. The methodologies for the three-level hierarchical design are not well understood, especially when the study interest is in the interaction effect between treatment and time. Additional challenges include a small number of available clusters, the missing data in longitudinal outcomes, and the imbalanced marker distributions among clusters. Our study focuses on two aspects of the three-level hierarchical design. First, we evaluate the performance of several sandwich type variance estimators on the inference of treatment-by-time interaction given a small number of clusters. Second, we attempt to determine the optimal balance between the number of clusters and cluster sizes under budget constraints using a simulation-based method. The proposed approach is illustrated using the study of adults with serious mental Illness.