A Gee Approach for Estimating the Correlation between Left-Censored Variables

HIV (Human Immunodeficiency Virus) researchers are often concerned with the correlation between plasma viral loads and a covariate, or the correlation between viral load levels from two reservoirs or from two competing quantification assays. Due to the lower limit of detection (LOD) of such assays, plasma viral load measurements are subject to left-censoring. In this paper, we propose a generalized estimating equations (GEE) approach to estimate the correlation coefficient between two continuous variables, where one or both of them may be left-censored. We present simulation studies to evaluate point and interval estimates of the correlation and compare the GEE results with a maximum likelihood approach (Lyles 2001a, 2001b). We also conduct a simulation study to explore the robustness of GEE estimates to the normality assumption. We apply the proposed methods to two HIV viral load data sets from clinical studies conducted in Bangkok, Thailand. The proposed method can be easily extended to incorporate covariates