Computational Methods, Resources, and Tools to Assess Transparency and Rigor of Randomized Clinical Trials

项目摘要/摘要 随机对照试验(RCT)是循证医学的基石，在 “证据金字塔”。当严格地设计、执行和报告时，它们提供了最有力的证据 关于治疗干预的有效性。然而，他们通常遭受各种类型的偏见(例如， 研究设计和实施中的选择偏差、磨损偏差)。在报告中，关键的方法特征，如 由于随机化和盲法经常被省略，因此很难评估试验的有效性和适用性 调查结果。遵守报告准则可以提高报告的透明度和完整性 生物医学研究。SPIRIT和COSORT指南帮助作者报告随机对照试验方案和结果 分别是出版物。尽管得到了许多高影响力的医学期刊的认可，但坚持这些 准则仍然不是最优的，可能是因为期刊缺乏强制执行和核实的方法，这 涉及大量的期刊工作人员或编辑时间。此外，透明的报告不会 保证方法的严密性。我们假设自然语言处理(NLP)方法 通过SPIRE/COSORT指南以及用于临床研究的术语和本体论资源 (A)通过确定研究报告中的关键研究特征并标明其缺失，提高遵从性；和(B) 通过提取粒度机支持自动化严谨评估和大规模方法学研究- 来自随机对照试验报告的可读方法信息。为了达到这些目标，我们的具体目标是： 目标1.创建文本分类模型以评估RCT报告一致性的透明度和完整性 精神和配偶准则。 目的2.发展信息提取方法以确定随机对照试验报告的方法学特征。 目标3.建立一个基于网络的合规工具，生成关于透明度和准则遵守情况的报告 RCT报告。 目标4.从已出版的区域风险评估文献中生成结构化的透明度报告，以分析方法和 报道质量。 拟议的研究将开发一套模型、资源和工具，以帮助临床利益相关者 在保持较高的报告标准、合成证据和促进开放的科学实践方面的研究。 它们将为整个科学生态系统的改进做出贡献，导致更好的临床护理和 健康政策。

Project Summary/Abstract Randomized controlled trials (RCTs) are a cornerstone of evidence-based medicine and are placed high in the “evidence pyramid”. When rigorously designed, conducted, and reported, they provide the most robust evidence on effectiveness of therapeutic interventions. However, they commonly suffer from various types of biases (e.g., selection bias, attrition bias) in study design and execution. In reporting, key methodological characteristics such as randomization and blinding are often omitted, making it difficult to assess the validity and applicability of trial findings. Adherence to reporting guidelines can improve transparency and completeness of reporting for biomedical studies. SPIRIT and CONSORT guidelines help authors report RCT protocols and results publications, respectively. Although endorsed by many high-impact medical journals, adherence to these guidelines remains suboptimal, possibly because journals lack methods for enforcement and verification, which involves a substantial amount of journal staff or editorial time. Furthermore, transparent reporting does not guarantee methodological rigor. We hypothesize that natural language processing (NLP) methods underpinned by SPIRIT/CONSORT guidelines as well as terminological and ontological resources for clinical research can (a) improve compliance by locating key study characteristics in RCT reports and flagging their absence, and (b) support automated rigor assessment and large-scale methodological research by extracting granular machine- readable methodological information from RCT reports. To achieve these goals, we specifically aim to: Aim 1. Create text classification models for assessing transparency and completeness of RCT reports consistent with SPIRIT and CONSORT guidelines. Aim 2. Develop information extraction methods to identify methodological characteristics in RCT reports. Aim 3. Build a web-based compliance tool that generates reports on transparency and guideline adherence of RCT reports. Aim 4. Generate structured transparency reports from published RCT literature for analysis of methodology and reporting quality. The proposed research will develop a set of models, resources, and tools that will assist stakeholders of clinical research in maintaining high reporting standards, synthesizing evidence, and promoting open science practices. They will contribute to improvements throughout the scientific ecosystem, leading to better clinical care and health policy.