Methods for Comparing Correlations Involving Left-censored Longitudinal Viral Load Measurements

• Activity Number: 175
• Type: Contributed
• Date/Time: Tuesday, August 13, 2002 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology*
• Author(s): Lorie Benning*+ and Robert Lyles and Stephen Gange
• Affiliation(s): Johns Hopkins Bloomberg School of Public Health and Rollins School of Public Health of Emory University and Johns Hopkins Bloomberg School of Public Health
• Keywords: correlation ; censoring ; repeated series longitudinal data

Abstract

Compartmentalization of HIV is of interest for studies of transmission and effectiveness of antiviral treatment. Correlating viral load measurements from two specimen types is a way to evaluate the association of HIV levels in different compartments. For transmission studies, it is of interest to know if the correlation differs among groups with varying levels of sexual activity. For treatment studies, it is of interest to know if correlation differs between treated and untreated individuals. Furthermore, in the context of repeated series longitudinal data, it is of interest to know if correlation changes over time. Because viral load measurements will be censored at the lower detection limit of the assay, standard correlation estimates from multivariate normal models will be biased. Maximum likelihood methods have been described to correlate two viral load measurments with known detection limits from the same time point (Lyles, et al., 2001). We discuss two extensions of these methods, to compare correlation between viral loads in the systemic and female genital tract compartments between two or more groups and to test if the correlation changes over several time points.