Coupling Results Data from ClinicalTrials.gov and Bibliographic Databases to Accelerate Evidence Synthesis

项目摘要 临床试验是循证医学的基础，但试验结果报告不完整， 经常迟到。据估计，多达一半的临床试验没有发表， 的已发表试验少报或误报结果。这种结果报告歪曲了证据 特别是在评估药物等干预措施的安全性时， 这可能会使患者处于不必要的风险之中。迫切需要新的方法来识别和 监测药物安全性数据。通过ClinicalTrials.gov提供的基础设施， （包括安全性结果）现在可用于越来越多的试验， 及时的时尚然而，在证据合成任务中获取和使用这些数据仍然有限。 ClinicalTrials.gov是全球最大的单一临床研究注册中心， 注册研究。在该网站注册的108，941项已完成的试验中，20%已上传结果数据， 共有785万人参加。ClinicalTrials.gov上报告的结果数据有可能填补空白 由于发表文章的延迟和偏见而造成的，并提供了更早和更完整的概述， 现有的审判证据。我们建议开发新的信息学方法的基础上的组合， 信息检索和机器学习方法，以方便访问和分析报告的试验结果 这个注册表。我们专注于2型糖尿病、肥胖和肿瘤的药物干预试验， 这项工作有三个具体目标：1）开发半自动化的试验筛选，用于识别和汇总试验 与临床干预相关; 2）从报告的结果中提取不良事件和安全性结局数据 3）进行验证研究，以评估不良事件的检测和半定量分析的性能。 安全性结局的自动荟萃分析。在这个项目中开发的方法将有助于及时，广泛- 大规模使用ClinicalTrials.gov上报告的试验结果， 及时提供药品安全数据。所有方法将公开提供，以支持不良事件 对药物干预措施进行监测和系统审查。

Project Summary Clinical trials are foundational to evidence-based medicine, but results reporting from trials is incomplete and frequently delayed. It is estimated that as many as half of clinical trials are not published and as many as half of published trials underreport or misreport outcomes. This type of results reporting distorts the evidence available to clinicians—particularly when it comes to assessing the safety of interventions like drugs and devices—and may place patients at unnecessary risk. There is a critical need for novel methods to identify and monitor drug safety data. Through the infrastructure provided by ClinicalTrials.gov, structured trial results (including safety findings) are now becoming available for an increasing number of trials in a comprehensive and timely fashion. However, access and use of these data in evidence synthesis tasks remain limited. ClinicalTrials.gov is the largest single registry for clinical studies worldwide and includes more than 260,000 registered studies. Of the 108,941 completed trials registered with the site, 20% have uploaded results data for a total of 7.85 million participants. Results data reported on ClinicalTrials.gov have the potential to fill gaps created by delays and biases in published articles and provide an earlier and more complete overview of available trial evidence. We propose to develop novel informatics approaches based on combinations of information retrieval and machine learning methods to facilitate access and analysis of trial results reported in this registry. Focusing on trials testing drug interventions in type 2 diabetes, obesity, and oncology, we perform this work in three specific aims: 1) Develop semi-automated trial screening for identifying and aggregating trials relevant to a clinical intervention; 2) Extract adverse event and safety outcomes data from results reported in the registry; and 3) Perform validation studies to assess detection of adverse events and performance of semi- automated meta-analyses of safety outcomes. Methods developed in this project will facilitate timely, broad- scale use of trial results reported on ClinicalTrials.gov in order to augment the availability of comprehensive and timely drug safety data. All methods will be made publicly available in order to support adverse event monitoring and systematic reviews of drug interventions.