Accounting for confounding bias and heterogeneity in comparative effectiveness

描述（由申请人提供）：人们对比较有效性研究越来越感兴趣，这是一种告知临床医生、支付方和政策制定者不同治疗方法的相对有效性的方式，目的是最大化患者的利益和支付方的价值。尽管随机对照试验（rct）目前被用来提供新疗法有效性的信息，但它们往往不是比较有效性的数据来源。随机对照试验通常招募精心挑选的人群，这些人群不能代表所有使用新疗法的个体，并且通常只将新疗法与安慰剂进行比较，而不是与其他常用疗法进行比较。其他数据来源，包括由专业协会开发的临床登记和大型管理数据库，提供了一个机会来调查各种临床环境中治疗的有效性和安全性，而不是用于随机对照试验。然而，随着这些观察数据的出现，治疗中不受控制的选择带来了混淆偏倚的问题。因此，需要改进的方法来解决目前可用的比较有效性数据的局限性，包括随机对照试验中纳入的非代表性人群，缺乏随机对照试验中常用治疗之间的比较，以及观察性研究中的混杂偏倚。这些问题在医学的各个领域都很常见。本提案旨在开发解决这些问题的新颖和可推广的方法。具体来说，我们将开发新的方法来结合随机试验、注册表和/或基于索赔的数据（利用随机对照试验和观察数据的优势）；扩展工具变量分析的最新技术；并开发新的联立方程模型来解释对假设不那么敏感的混淆，而不是目前使用的方法。在此过程中，我们将把这些方法应用于三个重要的临床实例：双相情感障碍患者的精神合并症治疗，现有精神药物治疗的重新配方，以及腹主动脉瘤（AAA）的手术修复。

DESCRIPTION (provided by applicant): There is increased interest in comparative effectiveness research as a way of informing clinicians, payers, and policymakers about the relative effectiveness of different treatments with the goal of maximizing benefit to patients and value to payers. Although, randomized controlled trials (RCTs) are currently relied upon to provide information on efficacy of new treatments they are often a poor source of data on comparative effectiveness. RCTs typically enroll carefully selected populations that are not representative of all individuals using the new treatments and often compare a new treatment only with placebo rather than with other commonly-used treatments for the same conditions. Other data sources, including clinical registries developed by professional societies and large administrative databases, provide an opportunity to investigate treatment efficacy and safety in a variety of clinical settings beyond those used for RCTs. However, with these observational data comes the problem of confounding bias from uncontrolled selection into treatment. Thus, improved methods are needed to address the limitations of currently available comparative effectiveness data, including non-representative populations enrolled in RCTs, lack of comparisons between commonly-used treatments in RCTs, and confounding bias in observational studies. Such problems are common across a variety of areas of medicine. This proposal seeks to develop novel and generalizable methods for addressing these problems. Specifically, we will develop novel approaches for combining data from randomized trials, registries and/or claims-based data (taking advantage of the strengths of both RCT and observational data); extend the latest techniques for instrumental variable analysis; and develop novel simultaneous equation models to account for confounding that are less sensitive to assumptions than currently-used methods. In so doing, we will apply these methods to three important clinical examples: treatments for bipolar disorder for patients with psychiatric comorbidity, reformulations of existing psychiatric drug treatments, and the surgical repair for abdominal aortic aneurysm (AAA). PUBLIC HEALTH RELEVANCE: New methods are developed to address the limitations of currently available comparative effectiveness analysis, including non-representative populations enrolled in randomized controlled trials, lack of comparisons between commonly-used treatments in RCTs, and confounding bias in observational studies. Such problems are common across a variety of areas of medicine including mental health, surgery, cancer, and medical devices. The proposed methods will help to better inform clinicians, payers, and policymakers about the relative effectiveness of different treatments with the goal of maximizing benefit to patients and value to payers.