Instrumental variables estimation under a structural Cox model
Instrumental variables estimation under a structural Cox model
复制标题
DOI：
10.1093/biostatistics/kxx057
复制
发表时间：
2019-01-01
期刊：
影响因子：
2.1
通讯作者：
Vansteelandt, Stijn
中科院分区：
文献类型：
--
作者：
Martinussen, Torben;Sorensen, Ditte Norbo;Vansteelandt, Stijn
仪器变量（IV）分析是一种越来越受欢迎的工具，用于推断暴露对结果的影响，例如，流行病学中越来越多的IV应用就是证明。然而，大多数至事件发生时间终点的IV分析以启发式方法为主。更严格的建议要么避开了考克斯模型，要么在具有二分法风险暴露和工具的限制性背景下考虑该模型。本文的目的是重新考虑结构考克斯模型下的IV估计，允许任意的曝光和工具。我们提出了一类新的估计量，并推导出它们的渐近性质。该方法说明使用两个真实的数据应用程序，并使用模拟数据。
Instrumental variable (IV) analysis is an increasingly popular tool for inferring the effect of an exposure on an outcome, as witnessed by the growing number of IV applications in epidemiology, for instance. The majority of IV analyses of time-to-event endpoints are, however, dominated by heuristic approaches. More rigorous proposals have either sidestepped the Cox model, or considered it within a restrictive context with dichotomous exposure and instrument, amongst other limitations. The aim of this article is to reconsider IV estimation under a structural Cox model, allowing for arbitrary exposure and instruments. We propose a novel class of estimators and derive their asymptotic properties. The methodology is illustrated using two real data applications, and using simulated data.