Contrasting a Longitudinal Factor Model with a Linear Mixed-Effects Model to Address Incomplete Data on Repeated Measures in an AIDS Prevention Study

• Activity Number: 171 - Missing Data
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Panteha Hayati Rezvan* and Xiang Lu and Thomas Belin
• Companies: University of California Los Angeles and University of California Los Angeles and UCLA
• Keywords: Missing data; joint modeling; multiple imputation; longitudinal factor model; linear mixed-effects model; multivariate normal imputation

Abstract

Missing data are frequently encountered in clustered/longitudinal studies and have the potential to induce bias and reduce precision if not handled properly. Using data from a clustered randomized controlled trial, we illustrate the application of a newly developed factor-based joint-modeling strategy for handling incomplete high-dimensional data. Substantive interest in the study focused on the effect of maternal HIV infection on early child growth outcomes in HIV-uninfected children. We investigated this association using a commonly used linear mixed-effects model (LMM) for analysis of longitudinal data as well as using joint multivariate normal imputation (MVNI) and the longitudinal factor (LF) imputation to handle incomplete data. We observed convergence failure when applying MVNI, apparently due to high correlation between repeated measures on time-dependent variables. The LMM and LF methods provided similar estimated regression coefficients and 95% CIs for analysis models we investigated; however, in some cases the methods yielded slightly different findings that affected interpretation of the results. Implications for other settings will be discussed in the presentation.