Supplementary materials accompanying this paper appear on-line.
Despite technological advances in efficiency enhancement of quantification assays, biomedical studies on HIV RNA collect viral load responses that are often subject to detection limits. Moreover, some related covariates such as CD4 cell count may be often measured with errors. Censored non-linear mixed-effects models are routinely used to analyze this type of data and are based on normality assumptions for the between-subject and within-subject random terms. However, derived inference may not be robust when the underlying normality assumptions are questionable, especially in presence of skewness and heavy tails. In this article, we address these issues simultaneously under a Bayesian paradigm through joint modeling of the response and covariate processes using an attractive class of skew-normal independent densities. The methodology is illustrated using a case study on longitudinal HIV viral loads. Diagnostics for outlier detection is immediate from the MCMC output. Both simulation and real data analysis reveal the advantage of the proposed models in providing robust inference under non-normality situations commonly encountered in HIV/AIDS or other clinical studies.
|Original language||English (US)|
|Number of pages||19|
|Journal||Journal of Agricultural, Biological, and Environmental Statistics|
|State||Published - Mar 1 2015|
- Case-deletion diagnostics
- Skew-normal independent distributions