A Doubly Robust Method to Handle Missing Multilevel Outcome Data with Application to a Cluster-Sampled Population-Based Study

• Activity Number: 179 - Statistical Methods for Measurement Error and Missing Data in Covariates/Exposures
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Nicole Butera* and Donglin Zeng and Annie Green Howard and Penny Gordon-Larsen and Jianwen Cai
• Companies: The University of North Carolina at Chapel Hill and UNC Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: Clustering; Doubly robust; Hierarchical modeling; Longitudinal; Missing data; Multilevel

Abstract

Missing data are common in multilevel longitudinal cohort studies and can lead to bias, particularly in studies with nonrandom missingness. Many common methods for handling missing data require correctly specifying a missingness model. Although doubly robust methods exist, the working models in these methods do not explicitly incorporate multilevel data structures, and so are likely to be misspecified. We developed a doubly robust method to consistently estimate regression coefficients in the presence of missing multilevel outcome data, assuming correct specification of either (1) the probability of missingness or (2) the outcome model. This method involves specification of separate multilevel models for missingness and for the outcome, conditional on observed variables and cluster-specific random effects, to account for correlation among observations. We showed this proposed estimator is doubly robust and derived its asymptotic distribution, conducted simulation studies to compare the method to an existing doubly robust method that ignored clustering, and applied the method to data from the China Health and Nutrition Survey, a multilevel population-based study.