Structural Models to Bridge Observational and Randomized Evidence: HAART and AIDS

描述（由申请人提供）：拟议的研究解决了从观察（即，非随机化）数据。拟议研究的总体目标是使用多个先前存在的NIAID资助的数据源来开发一种新的分析方法，以纠正年龄段观察分析的信息偏差，从而提供高效抗逆转录病毒治疗（HAART）对艾滋病或死亡时间的平均因果效应的精确估计。高效抗逆转录病毒疗法对艾滋病或死亡时间的影响是从不同研究设计的不同已发表结果中估计因果关系的一个范例。具体来说，观察性多中心艾滋病队列研究的结果与艾滋病临床试验组的320项随机临床试验结果不同。前一项研究由于非平凡的错误分类而受到偏倚，而后一项研究由于非平凡的不依从性而受到偏倚。日历周期和随机化将用作工具变量和分析方法（即，结构模型）将被应用和调整，以估计HAART对艾滋病或死亡时间的平均因果效应。将这些方法应用于观察性队列研究中的误分类问题，代表了结构嵌套模型的一种新应用。这项研究的具体目标是：（1）使用多中心AIDS队列研究的观察数据，开发一种结构嵌套模型方法，以估计HAART与联合治疗对至AIDS或死亡时间的错误分类校正平均因果效应，（2）使用现有的结构模型方法来估计合规性-HAART与联合治疗对AIDS或死亡时间的校正平均因果效应，使用AIDS临床试验组-320项随机试验数据，以及（3）比较目标1和目标2得出的估计风险比，并进行敏感性分析和蒙特卡罗模拟，以评估建模假设的影响，并帮助确定已发表结果中观察到的差异的原因。拟议的研究对生物医学科学具有相关性和高度重要性，因为它（a）将提供一种新的方法来完善现有的年龄段分析，以估计治疗或暴露的平均因果效应，（B）利用随机和观察证据之间的宝贵差异来阐明科学方法之间的联系，及（c）为爱滋病成本效益及医疗计划分析提供所需资料。鉴于最近观察性证据和随机证据之间的差异，拟议的研究是及时的（例如，绝经后激素治疗和心血管疾病）。拟议的研究具有超越艾滋病毒/艾滋病社区的意义，因为结果可能被用作其他生物医学领域的框架，这些领域在观察和随机证据之间存在断裂。拟议的研究对生物医学科学具有相关性和高度重要性，因为它（a）将提供一种新的方法来完善现有的年龄段分析，以估计治疗或暴露的平均因果效应，（B）利用随机和观察证据之间的宝贵差异来阐明科学方法之间的联系，及（c）为爱滋病成本效益及医疗计划分析提供所需资料。鉴于最近观察性证据和随机证据之间的差异，拟议的研究是及时的（例如，绝经后激素治疗和心血管疾病）。拟议的研究具有超越艾滋病毒/艾滋病社区的意义，因为结果可能被用作其他生物医学领域的框架，这些领域在观察和随机证据之间存在断裂。

DESCRIPTION (provided by applicant): The proposed research addresses the paramount and prevalent scientific problem of estimating causal effects of treatments or exposures from observational (i.e., nonrandomized) data. The overall goal of the proposed research is to use multiple preexisting NIAID-funded data sources to develop a novel analytic method to correct age-period observational analyses for information bias and thereby provide a refined estimate of the average causal effect of highly active antiretroviral therapy (HAART) on time to AIDS or death. The effect of HAART on time to AIDS or death is an exemplar case of estimating causal effects from disparate published results in different study designs. Specifically, results from the observational Multicenter AIDS Cohort Study differ from the AIDS Clinical Trial Group - 320 randomized clinical trial results. The former study was subject to bias due to non-trival misclassification, while the latter study was subject to bias due to non-trival noncompliance. Calendar period and randomization will be used as instrumental variables and analytic methods (i.e., structural models) will be applied and adapted to estimate the average causal effect of HAART on time to AIDS or death. Adapting these methods to the problem of misclassification in the observational cohort study setting represents a novel application of structural nested models. The specific aims of this research are to: (1) develop a structural nested model approach to estimate the misclassification-corrected average causal effect of HAART versus combination therapy on time to AIDS or death using the Multicenter AIDS Cohort Study observational data, (2) use existing structural model approaches to estimate the compliance-corrected average causal effect of HAART versus combination therapy on time to AIDS or death using the AIDS Clinical Trials Group-320 randomized trial data, and (3) compare the estimated hazard ratios resulting from aims 1 and 2 and conduct sensitivity analyses and Monte Carlo simulations that will evaluate the impact of modeling assumptions and assist in determining the cause for the observed discrepancy in published results. The proposed research is relevant and highly significant for biomedical science because it (a) will provide a novel method to refine existing age-period analyses to yield estimates of the average causal effect of a treatment or exposure, (b) capitalizes on a valuable discrepancy between randomized and observational evidence to illuminate connections between scientific approaches, and (c) will provide necessary inputs for HIV cost-effectiveness and healthcare planning analyses. The proposed research is timely in light of recent discrepancies between observational and randomized evidence (e.g., postmenopausal hormone treatment and cardiovascular disease). The proposed research has significance beyond the HIV/AIDS community, as results may be used as a framework in other biomedical fields suffering from a fracture between observational and randomized evidence. The proposed research is relevant and highly significant for biomedical science because it (a) will provide a novel method to refine existing age-period analyses to yield estimates of the average causal effect of a treatment or exposure, (b) capitalizes on a valuable discrepancy between randomized and observational evidence to illuminate connections between scientific approaches, and (c) will provide necessary inputs for HIV cost-effectiveness and healthcare planning analyses. The proposed research is timely in light of recent discrepancies between observational and randomized evidence (e.g., postmenopausal hormone treatment and cardiovascular disease). The proposed research has significance beyond the HIV/AIDS community, as results may be used as a framework in other biomedical fields suffering from a fracture between observational and randomized evidence.