Joint Modeling of Multivariate Longitudinal Hearing Loss Data Ascertained at Multiple Frequencies

• Activity Number: 304
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Mulugeta Gebregziabher*+ and Mark Eckert and Lois Mattews and Judy Dubno
• Companies: Medical Univ. of South Carolina and MUSC and MUSC and MUSC
• Keywords: correlated data ; conventional frequency ; extended frequency ; GLMM ; hearing loss ; missing data

Abstract

Typical hearing loss studies involve determination of hearing loss levels (dB) at multiple frequencies. These thresholds are commonly measured longitudinally for each subject's ear creating a truly multivariate longitudinal data. The most common analysis of such data involves separate modeling of the outcomes at each frequency for each ear. However, this ignores the correlation among the outcomes and could lead to inefficient use of data. Another approach is joint modeling of the outcomes. In many situations the parameters of the joint model are subject to constraints due to missing data as a result of dysfunction of cochlear cells. There is also computational challenge due to the increase in the dimension of the outcome vector when thresholds from each ear are considered separate outcomes. We propose a unified joint modeling approach that deals with these constraints and provides valid inference in the presence of missing data. We use data from a cohort of 1375 healthy subjects who completed a test-battery of thresholds at conventional and extended frequencies and were followed at the Medical University of South Carolina to demonstrate the proposed method.