Methods for Joint Clustering of Longitudinal HIV Biomarker Trajectories in the Presence of Missing and Censored Observations

• Activity Number: 649
• Type: Contributed
• Date/Time: Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Miranda Lynch* and Marianna Baum and Vladimir Novitsky and Victor De Gruttola
• Companies: University of Connecticut Health Center and Florida International University and Harvard T.H. Chan School of Public Health and Harvard
• Keywords: HIV ; Clustering ; Longitudinal data ; Biomarker profile ; Missing data

Abstract

A key issue in understanding HIV pathogenesis and its implications for treatment is characterizing different patterns of HIV disease progression, as measured by HIV viral load and CD4 cell count. Longitudinal measures of these biomarkers provide a basis for such analyses, and most cohorts that provide such data are based on chronically infected individuals. We present clustering methods and inference for joint longitudinal biomarker profiles to investigate whether joint examination of the progression of the markers provides a different set of clusters than does investigation of the markers individually. Missing data commonly occur in longitudinal data, and some HIV biomarker values can be censored below a detection limit; these data limitations require special handling in the profile clustering context. Here we present methods for joint biomarker clustering, taking into account missingness and censoring due to detection limits, and apply the methods in a prevalent cohort study of HIV-1 subtype C infected treatment naive individuals in Botswana.