Accounting for Hidden Bias in Vaccine Studies: A Negative Control Framework

疫苗研究中隐藏偏差的解释：阴性对照框架

• 批准号: 10541905
• 负责人: Xu Shi
• 金额: $ 36.28万
• 依托单位: UNIVERSITY OF MICHIGAN AT ANN ARBOR
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-01-01 至 2024-11-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10541905
• 关键词: Accounting; Affect; Area; Benefits and Risks; Complex; Data; Data Set; Dimensions; Electronic Health Record; Ensure; Event; Failure; Gene Expression Profiling; Health Policy; Healthcare; Knowledge; Machine Learning; Methods; Modernization; Monitor; Nature; Observational Study; Outcome; Policy Analysis; Property; Public Health; Recording of previous events; Reproducibility; Research; Research Design; Research Personnel; Residual state; Risk; Statistical Methods; Techniques; Testing; Time; Tissue-Specific Gene Expression; Use Effectiveness; Vaccination; Vaccines; adverse event risk; design; effectiveness evaluation; effectiveness study; flexibility; high dimensionality; improved; innovation; interest; machine learning method; novel; pathogen; public health priorities; safety study; semiparametric; theories; tool; user friendly software; vaccine effectiveness; vaccine evaluation; vaccine safety; vaccine trial

Project Summary / Abstract The proposed research aims to develop novel causal inference methods to resolve unmeasured confounding bias known to plague vaccine effectiveness and safety studies by leveraging so-called negative control variables widely available in vaccine studies. A negative control outcome is a variable known not to be causally affected by the treatment of interest, while a negative control exposure is a variable known not to causally affect the outcome of interest. Both share a common confounding mechanism as the exposure-outcome pair of primary interest. Examples of negative controls abound in vaccine studies. Such known-null effects form the basis of falsifica- tion strategy to detect unmeasured confounding, however little is known about when and how negative controls can be used to resolve unmeasured confounding bias. We plan to develop principled negative control methods for identification and semiparametric estimation of causal effects in the presence of unmeasured confounding, incorporating modern highly adaptive machine learning methods. We also plan to develop negative control meth- ods to detect and quantify causal effects in complex longitudinal and survival settings critical to vaccine studies using routinely collected healthcare data. Finally we plan to apply the proposed methods to evaluate vaccine effectiveness using data collected from a pioneering test-negative design platform and to monitor vaccine safety using electronic health record data. Successful completion of the proposed research will equip investigators with paradigm-shifting methods to unlock the full potential of contemporary healthcare data, encourage investigators to routinely check for evidence of confounding bias, and ultimately improve the validity of scientific research.

项目总结/摘要 该研究旨在开发新的因果推理方法来解决不可测量的混杂问题 已知通过利用所谓的阴性对照变量来困扰疫苗有效性和安全性研究的偏差 在疫苗研究中广泛使用。阴性对照结果是已知不受以下因素因果影响的变量： 关注的治疗，而阴性对照暴露是已知不会对结果产生因果影响的变量 感兴趣两者都有一个共同的混淆机制，即主要关注的确定性-结果对。 阴性对照的例子在疫苗研究中比比皆是。这种已知的零效应构成了虚假的基础， 检测未测量混杂因素的方法，但对阴性对照的时间和方式知之甚少 可用于解决未测量的混杂偏倚。我们计划制定原则性的阴性对照方法 对于在存在未测量混杂因素的情况下因果效应的识别和半参数估计， 结合现代高度适应性的机器学习方法。我们还计划研发阴性对照冰毒- ods用于检测和量化对疫苗研究至关重要的复杂纵向和生存环境中的因果效应 使用常规收集的医疗数据。最后，我们计划将所提出的方法应用于疫苗的评价 使用从一个开创性的阴性检测设计平台收集的数据， 使用电子健康记录数据。成功完成拟议的研究将使研究人员具备 范式转换方法，以释放当代医疗保健数据的全部潜力，鼓励研究人员 定期检查混杂偏见的证据，并最终提高科学研究的有效性。