Selective and Future Ignorability in Causal Inference

描述（由申请人提供）：在观察性研究中，大多数对因果推理的尝试都是基于假设治疗分配是可以忽略的。这种假设通常是随意做出的，而且往往是不可信的，部分原因是没有关于混杂因素的充分信息。在最近的工作中，我们已经形式化了可忽略性假设的变体，我们称之为选择性可忽略性和未来可忽略性，这可以更正确地代表许多研究中得到的情况。在选择性可忽略性下，在已知的数据子集中获得条件独立性；在未来可忽略性下，治疗和潜在结果的独立性有条件地取决于测量的协变量历史和未来潜在结果的组合。当获得选择性和/或未来可忽略性条件，而标准可忽略性条件没有时，我们已经开发了比标准方法更合适的初始推理方法。在这个项目中，我们将扩展我们对选择性和未来可忽略性假设的研究。为此，我们将1)扩展我们在制定这些假设方面的工作，特别是考虑估计治疗对重复测量和失效时间结果组合的影响；2)研究如何在这些假设下使g估计成为一个实用的估计选项；3)开发和研究选择性和未来可忽略性下g估计的替代方案；为此，我们将考虑最大似然、目标最大似然和贝叶斯方法；4)使用选择性可忽略性假设来开发更可靠的推理方法，当对兴趣的处理被错误测量时；5)利用美国肾脏数据系统的数据，利用所开发的方法估计使用促红细胞生成素对接受慢性血液透析的终末期肾病患者的红细胞压积水平和死亡率的影响。这是最近引起极大兴趣的一个话题。因此，该项目将开发在更合理的假设下使用的有用的推理方法，并将这些方法应用于一个重要的实际问题。

DESCRIPTION (provided by applicant): Most attempts at causal inference in observational studies are based on assumptions that treatment assignment is ignorable. Such assumptions are usually made casually and are often implausible, in part because adequate information on confounders is not available. In recent work, we have formalized variants of ignorability assumptions, which we term selective and future ignorability, which can more correctly represent the situation obtaining in many studies. Under selective ignorability, conditional independence obtains in a known subset of the data; under future ignorability, independence of treatment and potential outcomes holds conditionally on a combination of measured covariate history and future potential outcomes. We have developed initial approaches to inference which are more appropriate than standard methods when selective and/or future ignorability conditions obtain but standard ignorability conditions do not. In this project, we will extend our work on selective and future ignorability assumptions. To this end, we will 1) Extend our work in formulating these assumptions, in particular in considering estimating the effects of a treatment on a combination of repeated measures and failure-time outcomes; 2) Investigate ways to make G-estimation a practical estimation option under these assumptions; 3) Develop and investigate alternatives to G-estimation under selective and future ignorability; for this, we will consider maximum likelihood, targeted maximum likelihood, and Bayesian approaches; 4) Use selective ignorability assumptions to develop more reliable methods of inference when the treatment of interest is mismeasured; and 5) Use the methods developed to estimate the effect of erythropoietin use on hematocrit levels and mortality among subjects with end-stage renal disease receiving chronic hemodialysis, using data from the United States Renal Data System. This is a topic that has recently generated substantial interest. Thus, the project will develop useful inferential methods for use under more reasonable assumptions, and will apply those methods to an important practical problem. PUBLIC HEALTH RELEVANCE: The project will develop and evaluate new assumptions for controlling confounding in longitudinal studies and new methods for estimating treatment effects under those assumptions. The project will use those methods in estimating the effect of erythropoietin on hematocrit and on mortality using data from the United States Renal Data System.