Bayesian estimation of participants' adherence in an HIV prevention trial using multiple data sources and pharmacokinetic models

Pre-Exposure Prophylaxis (PrEP) is an HIV prevention practice which uses antiretrovial drugs to keep HIV-negative individuals from infection with HIV. One of the antiretrovi ral drugs that is usually used for HIV treatment but has been widely adopted in PrEP is Truvada by Gilead. In the recent studies of using Truvada as PrEP, the results have been variable regarding to the efficacy of Truvada for HIV prevention. However, the adherence to pill-taking varies on a large scale across those studies, which can potentially underestimate the drug efficacy.;The HPTN 067 trial is an open-label, phase II HIV prevention trial, that studies the coverage and adherence associated with different PrEP pill-taking strategies. Motivated by the HPTN 067 trial, a statistical model has been developed to estimate the individual-level adherence, misreporting and pharmacokinetic (PK) parameters using self-reported adherence and drug concentrations in plasma. The goal of this thesis is to investigate the estimation method used in this model and evaluate/improve its performance in terms of computational efficiency and data with detection of limits. In addition, a new statistical model (two-biological-measure model) is developed to estimate the individual-level adherence, misreporting and PK parameters from two biological samples (i.e. plasma and PBMC).;Pharmacokinetic models are used to estimate the individual-level PK parameters. A one compartment IV bolus model is used to estimate the PK parameters for plasma. The two-biological-measure model incorporates a second biological measure (i.e. PBMC) with different half-life into the previously developed statistical model, using a one compartment first-order absorption model. The statistical models are analyzed using MCMC methods.;All the data are simulated in this study. The observed data include self-reported adherence, drug concentrations measured in plasma at multiple time points, drug concentration measured in PBMC at multiple time points and lower limits of detection of drug concentrations in plasma and PBMC.;In this study, a shorter look-back period is used to improve the computational efficiency and correct the sensitive estimation. Estimation accuracy is also improved using detection limits. The two-biological-measure model improves the accuracy in estimating adherence but has a non-identifiability issue which can be fixed using informative priors for the population PBMC absorption rate parameter. However, the subject-specific absorption rate parameters in PBMC still cannot be accurately estimated.;A two-biological-measure model adopting a multi-compartment system is also evaluated. Although the model still has non-identifiability issue, it gives good estimation of individual adherence probabilities and some PK parameters.