Bounds for direct and indirect effects with application to perinatal epidemiology

描述（由申请人提供）：健康科学研究人员经常感兴趣的是检查一种治疗对某些结果的影响在多大程度上由中间变量介导。经常使用各种统计技术来进行这种中介分析。然而，为了解释这种统计分析的结果的因果关系，也就是说，作为一个特定的处理的直接和间接的影响，必须对混淆作出某些强有力的假设。在中介分析中需要的未测量的混杂假设比在研究治疗的总效果时需要的假设更强。为了估计受控的直接效应，不仅必须没有治疗-结果关系的不可测量混杂因素，而且必须没有中介-结果关系的不可测量混杂因素。此外，为了将总效应分解为直接效应和间接效应，还必须假设治疗-中介关系中没有未测量的混杂因素，并且治疗的后果不会混淆中介-结果关系。在许多观察性研究中，这些非不可测量的混杂假设并不成立。本研究的目的是在识别所需的非测量混杂假设失败时，获得并应用观察性研究中直接和间接影响的界限结果。进展将主要通过对未测量的混杂变量与治疗、中介和感兴趣的结果之间的关系施加单调性假设来取得。研究结果将应用于围产期流行病学中的一个问题。在过去的20年里，美国对产前护理的利用有所增加；然而，早产率也有所上升；剖腹产率有所上升；婴儿死亡率下降。对这些趋势的一种可能解释是，产前护理允许早期发现妊娠问题，适当的反应是早产剖腹产或药物引产。这将导致更大比例的早产，但最终婴儿会更健康。这项研究将审查产前护理对婴儿死亡率的影响在多大程度上是由医学诱发的早产介导的。这种分析所需的非未测量混杂假设不太可能在国家卫生统计中心将分析的出生证明和婴儿死亡率相关文件中得到满足，因此方法学研究中得出的结果将用于确定产前护理对婴儿死亡率的影响部分的界限，这部分影响是由医学诱导的早产引起的。

DESCRIPTION (provided by applicant): It is often of interest to investigators in the health sciences to examine the extent to which the effect of a treatment on some outcome is mediated by an intermediate variable. Various statistical techniques are often employed to carry out this mediation analysis. However, in order to interpret the results of such statistical analysis causally, that is, as direct and indirect effects of a particular treatment, certain strong assumptions must be made about confounding. The nounmeasured- confounding assumptions required in mediation analysis are stronger than those required in studies of the total effect of a treatment. In order to estimate controlled direct effects there must not only be no unmeasured confounders of the treatment-outcome relationship but there also must be no unmeasured confounders of the mediator-outcome relationship. Furthermore, for the decomposition of a total effect into a direct and an indirect effect, one must also assume that there are no unmeasured confounders of the treatment-mediator relationship and that there is no consequence of treatment that confounds the mediator-outcome relationship. In many observational studies these no-unmeasured-confounding assumptions will not hold. The purpose of this research is to obtain and apply results concerning bounds for direct and indirect effects in observational studies when the no-unmeasured-confounding assumptions required for the identification fail. Progress will be made principally by imposing monotonicity assumptions on the relationship between the unmeasured confounding variables and the treatment, mediator and outcome of interest. The results will be applied to a problem in perinatal epidemiology. Over the last twenty, utilization of prenatal care has increased in the United States; preterm birth rates, however, have also increased; c-section rates have increased; infant mortality has decreased. One possible explanation for these trends is that prenatal care allows for the early detection of pregnancy problems for which the appropriate response is preterm c-section or medically-induced labor. This would result in a greater proportion of preterm births but ultimately healthier infants. The research will examine the extent to which the effect of prenatal care on infant mortality is mediated by medically induced preterm birth. The no-unmeasured-confounding assumptions required for such an analysis are unlikely met in the National Center for Health Statistics Linked Birth Certificate and Infant Mortality files that will be analyzed and thus the results derived in the methodological research will be employed to determine bounds for the portion of the effect of prenatal care on infant mortality that is mediated by medically-induced preterm birth. PUBLIC HEALTH RELEVANCE: The statistical methodology developed in the proposed research will be useful in studies assessing the extent to which the effect of one variable on another is mediated by some intermediate. The methods will be employed to determine the extent to which prenatal care affects infant mortality by allowing for the early detection of pregnancy problems for which the appropriate response is a preterm c-section or induced labor.