An important objective of biomedical studies of HIV/AIDS is to estimate the effect of factors on HIV/AIDS outcomes. Although there has been a rich list of methods of analysis that are appropriate if such factors are assumed controlled, there is important need for designs and methods to validly estimate the effect of, and to maximize the benefit from factors that cannot be directly controlled. In this proposal, we will develop designs and methods to better estimate and to maximize the benefit that uncontrolled factors have on a population. The proposed methods will build on preliminary work we have conducted using the framework of """"""""principal stratification"""""""". The success of that work increases the potential impact of this proposal. The proposed methods are developed for two broad aims, and are motivated by needle exchange programs in the US, HIV treatment administration in East Africa, and HIV vaccine efficacy trials.
(Aim 1) Develop designs to maximize the treatment benefit for participants in studies that control the location of sites that offer treatments, but do not directly control who gets treatment or who provides outcomes. Such a situation arises with sites offering needle exchange programs in the US, and with sites operating HIV treatment programs in the developing world. We have previously developed methods to estimate the treatment effect on a population given a design. We now propose new methods to find designs that maximize both, the treatment effect on a population, as well as the information needed to periodically monitor that effect. Our methods are motivated by and will be applied to designate placements that maximize the benefit of the Baltimore Needle Exchange Program, and of the PEPFAR HIV programs in East Africa.
(Aim 2) Develop statistical methods to better evaluate the effect of a treatment on outcomes which are defined only in an uncontrolled subset of participants selected after randomization. Such a situation arises in HIV vaccine trials when assessing the effect of a vaccine on outcomes (e.g., viral load) that are defined only on the subsets who are infected with HIV post-randomization. Vaccines that are infected with HIV can be different from those who are infected without the putative effect of vaccine on infection risk. We propose more valid methods to estimate the effects of HIV vaccines on longitudinal post-infection outcomes. Our methods are motivated and will be applied to the first cell-mediated immunity HIV vaccine trial (Step trial), and to the Mashi trial on mother-to-child HIV transmission.
Showing the most recent 10 out of 18 publications
