Automatically Creating and Updating Meta-Studies of Randomized Controlled Trials

 描述(由申请者提供)：“元研究”(或“元分析”)收集并分析同一主题的许多研究，以了解是否有有意义的、全面的结果。元研究可以支持(或驳斥)干预措施，刺激新的研究，并导致新的临床指南。然而，构建元研究是一个搜索文献、汇编结果和进行统计分析的时间密集型过程。由于所需的时间投入，许多主题未被探索，并且许多元研究没有与最新发布的结果保持同步。最后，一些(未知的)偏见，通过元研究中的主观选择，可能会影响结果。我们的长期目标是尽可能地使元研究过程自动化。这将减少主观偏见；增加证据的传播，特别是对受到较少关注的疾病和干预措施；并允许随着新结果的发布自动更新元研究。我们提出了一个计算机系统，它使用统计机器学习来收集和分组专注于类似干预和结果的研究；从文本中提取必要的结果；并使用标准的荟萃分析技术分析结果。最终的输出将在一个类似电子表格的Web界面中呈现，用户可以在其中探索甚至更改数据和元分析。我们的团队独特地将机器学习方面的技术专业知识与发表关于炎症性肠病(IBD)的元研究的领导能力相结合，IBD是我们第一阶段可行性研究的重点疾病。因此，我们在技术上和能力上都是合格的 以确保这些技术产生有效和准确的元研究。我们的第一阶段结果将定义这项新任务的当前最先进水平。此外，尽管我们最初将重点放在IBD上，但我们的第一阶段结果将证明我们的方法可以推广到其他疾病，最终适用于任何干预措施和任何疾病。我们第一阶段结果显示的可行性将激励我们第二阶段的努力，我们将专注于显著改进方法，产生对所有医学文献的广泛覆盖，并产生人类质量的元研究。我们注意到，到第一阶段结束时，我们应该有一个可行的端到端原型，专注于IBD，我们可以开始将其推向市场。鉴于第二阶段的重点是改善覆盖疾病的技术、用户体验和范围，最终产品应该会使我们的目标市场显著受益。

 DESCRIPTION (provided by applicant): A "meta-study" (or "meta-analysis") collects and analyzes many studies on the same topic to understand if there is a meaningful, overall result. Meta-studies can support (or refute) interventions, spur new investigations, and lead to novel clinical guidelines. However, constructing meta-studies is a time intensive process of searching the literature, compiling the results, and performing the statistical analysis. Due to the time commitment that is required, many topics are unexplored, and many meta-studies are not kept up-to-date with the latest published results. Finally, a number of (unknown) biases, via subjective choices during the meta-study, may influence the results. Our long-term goal is to automate, as much as possible, the meta-study process. This should decrease subjective bias; increase the dissemination of evidence, especially for diseases and interventions that receive less attention; and allow for the automatic updating of meta-studies as new results are published. We propose a computer system that uses statistical machine learning to gather and group studies focused on similar interventions and outcomes; extract the necessary results from the text; and analyze the results using standard meta-analysis techniques. The final output will be presented in a spreadsheet-like Web-interface where users can explore and even change the data and meta-analyses. Our team uniquely blends technical expertise in machine learning with leadership in publishing meta-studies about Inflammatory Bowel Disease (IBD), our disease of focus for our Phase I feasibility study. We are therefore qualified technically and able to ensure that the techniques generate valid and accurate meta-studies. Our Phase I results will define the current state-of-the-art for this novel task. Further, although we will initially focus n IBD, our Phase I results will demonstrate that our approach can generalize to other diseases, eventually applying to any intervention and any disease. The feasibility shown by our Phase I results will motivate our Phase II effort where we will focus on dramatically improving the approach, yielding broad coverage of all medical literature and generating human-quality meta-studies. We note that by the end of Phase I we should have a viable end-to-end prototype, focused on IBD, which we can begin taking to market. The final product should significantly benefit our target markets given the Phase II emphasis to improve the technology, user experience, and scope of covered diseases.