Improved analysis of experiments and observational studies in HIV

摘要 更可靠和准确的卫生知识是更好的卫生政策和行动的基石。确实有 艾滋病毒中的棘手问题，可以通过新的量化工具更好地解决。实验结果 和观察性艾滋病毒研究可以通过开发和使用 统计学、流行病学、因果推论和人工智能接口的新方法。一个 半参数统计理论、因果模型和集成机器学习的创新组合 为艾滋病毒研究取得更好的结果提供了一个独特的机会。在这项工作中，我们提出了新的估计 风险(或生存)函数。这些新的估计器提高了准确性，适应了竞争事件，允许 将影响推广到特定的感兴趣人群，结合滋扰功能的机器学习 放松对模型形式的假设，并允许敏感度分析量化不受控制的影响 偏见。具体目标是开发、测试和传播这些新的估计器。这些目标是 1)评估在这个庞大的美国队列中所有原因和特定原因死亡率的长期治疗历史 2)使用治疗方法估计每个方案参数的观察性模拟 比较整合酶抑制剂治疗下的复合终点的决策设计比较 在北美艾滋病队列研究和设计合作中进行基于Eefavirenz的治疗； 3)估计TDF-FTC与ABC-3TC ARM的每协议参数；以及4)估计每协议参数。 己酸17α-羟孕酮与面膜安慰剂对早产风险的方案参数 赞比亚艾滋病毒携带者孕妇出生。组建的团队具有领域领先的专业知识 流行病学、统计学和艾滋病毒医学。科学产品将包括出版物和研讨会 介绍新的方法方法和出现的新的实质性发现 将建议的方法应用于具体目标中确定的每一个问题。我们还将生产 公开提供的R包和SAS宏，以实现建议的估计器。

ABSTRACT More robust and accurate health knowledge is a cornerstone of better health policy and action. There are tough questions in HIV that can be addressed better with new quantitative tools. Results from experimental and observational HIV studies can be made better and more policy-relevant through development and use of new methods at the interface of statistics, epidemiology, causal inference, and artificial intelligence. An innovative combination of semiparametric statistical theory, causal models, and ensemble machine learning provides a unique opportunity for better results from HIV studies. In this work, we propose new estimators of the risk (or survival) function. These new estimators improve accuracy, accommodate competing events, allow effects to be generalized to specific populations of interest, incorporate machine learning of nuisance functions to relax assumptions about model form, and allow sensitivity analyses to quantify the impact of uncontrolled biases. The specific aims are vehicles to develop, test, and disseminate these new estimators. These aims are to 1) estimate the long-term treated history of all-cause and cause-specific mortality in this large US cohort of women with HIV; 2) estimate the observational analog of the per-protocol parameter using a treatment decision design to compare composite endpoints under an integrase-inhibitor-based treatment compared to an efavirenz-based treatment in the North American AIDS Cohort Collaboration on Research and Design; 3) estimate the per-protocol parameter for TDF-FTC versus ABC-3TC arms; and 4) estimate the per- protocol parameter for 17 alpha-hydroxyprogesterone caproate versus masked placebo on risk of preterm birth in Zambian HIV+ pregnant women. The assembled team features field-leading expertise in epidemiology, statistics, and HIV medicine. Scientific products will include publications and workshop presentations describing new methodological approaches and new substantive findings that emerge after applying the proposed methods to each of the problems identified in the specific aims. We will also produce publicly available R packages and SAS macros to implement the proposed estimators.