Improving Methods for Dealing with Missing Data in Drug Use and Addiction Research: The Use of Later-Retrieval in Ecological Momentary Assessment

项目摘要/摘要 拟议的K01指导研究科学家发展奖将为杨曼舒博士做好准备 成为开发尖端实用统计方法的独立研究员，以解决 及时发布药物使用和成瘾(DUA)研究中的缺失数据。杨博士目前是一名助理 罗德岛大学定量健康心理学教授。概述的建议书构建 基于她在统计学和心理学方面的培训和研究经验，并将有助于她成为 致力于将最先进的方法和实质性的理论联系起来以提高对 DUA的病因和干预策略。DUA有无数的有害影响，并继续提出 美国的公共卫生问题。在过去的二十年里，生态瞬时评价(EMA)一直是 越来越多地用于帮助研究人员了解心理社会和背景因素对 在真实世界中和几乎实时地使用物质，以便能够进行更有效的干预 相应地发展起来。然而，伴随着使用EMA的机会而来的是一种重要的方法学 挑战：遗漏答复是不可避免的，往往是实质性的，而且在数据分析中没有得到适当的处理， 因此大大增加了研究人员得出不正确结论并发展为无效或 甚至是不安全的干预。目前的方法不能解决DUA中所有独特的方法学挑战 EMA研究由于其复杂的缺失数据模式和不可检验的缺失数据假设。论 另一方面，EMA带来了一个独特的机会来应对这些挑战，很快就会重新提示参与者 在他们错过了一次EMA调查以检索他们的数据后。这类后来检索到的数据很容易从 现有数据(例如，捕获前一天错过的DUA后果的早晨报告)或可以很容易地 添加到自动EMA系统，而无需更改EMA时间表。概述的提案包括 全面的导师和教学计划，支持杨博士的职业发展和进步 掌握DUA病因/干预、EMA设计、数据管理和分析、贝叶斯等方面的知识和技能 缺失数据分析和统计编程。具体地说，拟议研究的目的是 (1)使用所有可用数据(最初观察到的和后来检索到的)来表征丢失的数据机制 Dua EMA，(2)开发一种新的贝叶斯方法来处理丢失的数据并做出有效的推断 DUA病因学，(3)开发贝叶斯敏感性分析方法，以测试发现的稳健性 可能偏离缺失的数据假设。拟议的研究将调查丢失数据的问题 并使用NIDA资助的三个项目的经验EMA数据集和计算机开发分析方法- 以经济高效的方式模拟数据。从研究中发展出来的统计方法将会有很大的帮助 研究人员阐明了DUA的真实机制，并开发了量身定制的有效干预措施。

Project Summary/Abstract The proposed K01 Mentored Research Scientist Development Award will prepare Manshu Yang, Ph.D. to become an independent researcher in developing cutting-edge and practical statistical methodology to address the timely issue of missing data in drug use and addiction (DUA) research. Dr. Yang is currently an Assistant Professor of Quantitative Health Psychology at the University of Rhode Island. The outlined proposal builds upon her training and research experience in statistics and psychology and will facilitate her path to become a scholar dedicated to bridging state-of-the-art methodologies and substantive theory to advance knowledge of DUA etiology and intervention strategies. DUA has a myriad of deleterious impacts and continues to raise public health concerns in the US. In the past two decades, ecological momentary assessment (EMA) has been increasingly used to help researchers understand the influence of psychosocial and contextual factors on substance use in the real world and nearly in real time, so that more efficacious interventions can be developed accordingly. However, along with the opportunity of using EMA comes a significant methodological challenge: missing responses are inevitable, often substantial, and not properly handled in data analysis, hence significantly increasing researchers’ risks of reaching incorrect conclusions and developing ineffective or even unsafe interventions. Current methods cannot address all the unique methodological challenges in DUA EMA studies due to their complex missing data patterns and untestable missing data assumptions. On the other hand, EMA brings a unique opportunity to address these challenges by re-prompting participants shortly after they missed an EMA survey to retrieve their data. Such later-retrieved data are readily available from existing data (e.g., morning reports capturing missed DUA consequences in the prior day) or can be easily added to an automated EMA system without altering EMA schedule. The outlined proposal includes a comprehensive mentorship and didactic plan to support Dr. Yang’s career development and advance her knowledge and skills in DUA etiology/intervention, EMA design, data management, and analysis, Bayesian missing data analysis, and statistical programming. Specifically, the aims of the proposed research study are (1) to use all available data (initially observed and later retrieved) to characterize missing data mechanisms in DUA EMA, (2) to develop a novel Bayesian method for handling missing data and making valid inference on DUA etiology, (3) to develop a Bayesian sensitivity analysis method to test the robustness of findings to possible departures from missing data assumptions. The proposed study will investigate missing data issues and develop analysis methods using empirical EMA datasets from three NIDA-funded projects and computer- simulated data in a cost-effective way. Statistical methods developed from the study will greatly help researchers elucidate the real-world mechanisms of DUA and develop tailored, effective interventions.