|
Abstract:
|
Since the introduction of laboratory assays to quantify HIV viral load in biospecimens, one of the challenges associated with the analysis of such data has been left censoring due to lower detection limits. Considerable attention has been devoted to parametric methods to estimate means, variances, and crude correlations characterizing HIV positive populations while accounting for non-detectables. While longitudinal data analysis with multivariate left censored observations has also been addressed, approaches have primarily dealt with estimation of fixed effect parameters and variance components. This talk focuses on an extension of common likelihood-based methods to address the estimation of partial correlation coefficients when bivariate logged HIV RNA measurements are to be analyzed in the presence of covariates. Simulation results are presented to evaluate estimation and inferential properties. A motivating example using paired viral load measurements from two consecutive visits in the HIV Epidemiology Research (HER) Study serves to illustrate the methods, as well as to motivate a brief discussion of univariate model selection.
|
