Identification and Inference for Longitudinal Causal Mediation Analysis in HIV

描述（由申请人提供）：抗逆转录病毒治疗（ART）的良好依从性被广泛认为是持续抑制人类免疫缺陷病毒（HIV）的关键因素，并且通常被认为是维持处方药物方案和最佳患者结局的先决条件。然而，在何种程度上坚持一个给定的选择的治疗（相对于另一种治疗），有助于病毒学失败是复杂的，仍然知之甚少，是一个紧迫的调解问题，在艾滋病毒研究。理解这个问题在资源贫乏的环境中特别重要，在这些环境中，ART方案选择有限，并且坚持终身多药每日给药具有挑战性但必要。在这种情况下，量化不同治疗之间的依从率差异导致病毒学失败率的差异程度，将告知通过增加某些ART依从率的计划而不是改善ART方案本身可以改善失败率的程度。这种坚持性干预措施在治疗结核病方面非常成功，并被认为在治疗艾滋病毒方面很重要。不幸的是，传统的中介分析方法假设没有未观察到的混杂的静态设置，因此不适合复杂的纵向数据的分析通常遇到的观察性研究的艾滋病毒。我们的计划是弥补这一不足，利用最近的发展，因果推理的基础上潜在的结果和图形模型，现代半参数理论。

DESCRIPTION (provided by applicant): Good adherence to anti-retroviral therapy (ART) is widely accepted as a key factor for sustained viral suppression of human immunodeficiency virus (HIV), and is often considered a prerequisite for maintenance on a prescribed drug regimen and optimal patient outcomes. However, the extent to which adherence to a given choice of therapy (relative to another therapy), contributes to virologic failure is complex and stll poorly understood, and is a pressing mediation question in HIV research. Understanding this issue is particularly important in resource poor settings where ART regimen options are limited and adherence to lifelong multi- drug daily dosing is challenging but necessary. In such settings, quantifying to what degree differential rates of virologic failure are due to difference in adherence rates between therapies would inform the extent to which failure rates could be improved by programs that increase adherence rates for certain ARTs, rather than improving ART regimens themselves. Such adherence interventions have been very successful in the treatment of tuberculosis and are considered important in the treatment of HIV. Unfortunately, traditional approaches to mediation analysis assume static settings with no unobserved confounding, and are thus not suited for analysis of complex longitudinal data typically encountered in observational studies of HIV. Our plan is to remedy this deficiency using recent developments in causal inference based on potential outcomes and graphical models, and modern semi-parametric theory.