Impact of IDU on HIV viral load after treatment initiation taking into account left-censoring and nonignorable dropout

In HIV clinical trials and longitudinal cohort studies, injection drug users (IDU) are an important part of the study population. On highly active antiretroviral therapy (HAART), HIV viral loads (a biomarker for HIV disease progression) in HIV-seropositive patients' plasma can be suppressed and maintained below the limits of detection for a prolonged period of time. However, there is evidence that IDUs do not benefit equally from HAART compared with nonIDUs. In statistical analyses, treating HIV viral load as a continuous outcome can result in bias due to the detection limits of the virologic assay. In addition, dropout is common in HIV clinical trials and longitudinal cohorts. Traditional mixed models ignores dropout and likely result in bias and loss of power. We propose a model which utilizes a likelihood function modified for both left-censoring and right-censoring in HIV viral load measurements, while incorporating a B-spline varying-coefficient approach to accommodate nonignorable dropout. This model will be used to estimate the impact of IDU on HIV disease progression. Analysis results using data from a longitudinal cohort study of acute and recent infections will be presented.