New Approaches to Mediation Analysis using Causal Inference Methods

描述（由申请人提供）：本申请涉及广泛的挑战领域（05）比较有效性研究和特定挑战主题，05- aa -101：现有临床数据集的创新分析和05- aa -102：酒精治疗研究的适应性设计和以人为中心的数据分析。行为变化的机制模型本质上描述了因果过程；然而，行为学文献中的大多数分析依赖于Baron和Kenny（1986）基于回归的方法，这种方法通常不能用于推断因果关系。由于缺乏对模型的充分理解和缺乏现成的软件，研究人员可能不愿意使用因果模型。本项目旨在弥合因果模型的理论与实践之间的差距，以评估行为干预研究中的调解，特别侧重于酒精滥用的干预。我们使用Robins和Greenland（1992）以及Pearl（2001）描述的直接和间接影响的因果模型。拟议研究的主要贡献将是发展、应用、评价和传播统计方法，使这些模型与观测数据相拟合。拟议的研究将开发统计方法，以发现针对酒精滥用的行为干预的途径和机制。研究计划的一个主要成果是开发、测试和传播用于实现模型的适当软件。对机制途径的了解可以更深入地理解某些干预措施是如何以及为什么起作用的，并为基于个人特征定制干预措施打开了大门。

DESCRIPTION (provided by applicant): This application addresses broad Challenge Area (05) Comparative Effectiveness Research and specific Challenge Topics, 05-AA-101: Innovative Analysis of Existing Clinical Datasets and 05-AA-102: Adaptive Designs and Person-Centered Data Analysis for Alcohol Treatment Research. Abstract Mechanistic models of behavior change inherently describe causal processes; however, most analyses in the behavioral literature rely on the Baron and Kenny (1986) regression-based approach, which generally cannot be used to infer causal effects. Researchers may be reluctant to use causal models due to lack of full understanding of the models and lack of readily available software. This project is intended to bridge the gap between theory and practice of causal modeling for assessment of mediation in behavioral intervention studies, with specific focus on interventions for alcohol abuse. We use causal models of direct and indirect effects described by Robins and Greenland (1992) and Pearl (2001). The primary contributions of the proposed research will be development, application, evaluation and dissemination of statistical methods for fitting these models to observed data. The proposed research will develop statistical approaches for discovering pathways and mechanisms of behavioral interventions targeted at alcohol abuse. A major outcome of the research program is development, testing and dissemination of appropriate software for implementing the models. Knowledge of mechanistic pathways allows deeper understanding of how and why certain interventions work, and opens the door to customizing interventions based on person-specific characteristics.