Statistical Methods for Health System-Based Cluster Trials: Experimenting in a Finite Sample
MICHAEL HARHAY – UNIVERSITY OF PENNSYLVANIA, SCHOOL OF MEDICINE
Pragmatic and cluster trials in health systems must navigate logistic pressures that don’t generally allow the ideal trial to be designed. In this talk, I will review recent work on estimands, leveraging baseline covariates to maximize statistical power, missing data methods for health system cluster trials, and, if time allows, some work related to compliance that is ongoing with our research team. By the end of the presentation, the audience will understand: 1) Estimands relevant to cluster trials, i.e., cluster versus participant average treatment effects, and issues with informative cluster size. 2) Benefits and optimal strategies for incorporating baseline covariate adjustment in cluster trials. 3) Advantages and disadvantages of different approaches to imputing multilevel missing data in cluster trials
Michael Harhay, Ph.D., is an Assistant Professor of Epidemiology and Medicine (Pulmonary and Critical Care), and the Director of the NIH- and PCORI-funded Clinical Trials Methods and Outcomes Lab at the PAIR (Palliative and Advanced Illness Research) Center at the University of Pennsylvania Perelman School of Medicine. The primary focus of his current research is the development and application of causal inference and Bayesian statistical methods to improve pragmatic and cluster-randomized trials, particularly those embedded in health systems and for patients with critical and severe illnesses.