Scholarlink

<p>The high cost and complexity of drug discovery and development make clinical trial outcomes critical for regulatory approval and patient care. However, limited availability of large-scale, high-quality trial outcome data hinders research in predictive modelling and evidence-based decision-making. Here we introduce the Clinical Trial Outcome (CTO) benchmark, a scalable, reproducible and large-scale dataset covering 125k drug and biologics trials. We integrate interpretations of trial publications, phase-wise tracking, news sentiment, stock price and trial-related metrics. We additionally manually annotated a subset of recent trials from 2020 to 2024. CTO labels align with previous expert annotations, achieving a 91% F1 score. Predictive models trained on CTO consistently outperform models trained on older, static data when evaluated year over year on recent trials, mitigating performance decay from data distribution shifts. Beyond simply providing more outcome labels, CTO creates a continuously updated benchmark that supports trial design by enabling models and analysts to evaluate how design decisions relate to success or failure over time, improving prediction, optimization and scenario planning for future studies. Beyond outcome prediction, the CTO benchmark can also serve as an open resource to support trial design and clinical development by informing end point selection, protocol refinement, enrolment feasibility and phase-transition decisions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A large-scale database for clinical trial outcomes and features

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Chufan Gao,
  • Jathurshan Pradeepkumar,
  • Trisha Das,
  • Shivashankar Thati,
  • Jimeng Sun

摘要

The high cost and complexity of drug discovery and development make clinical trial outcomes critical for regulatory approval and patient care. However, limited availability of large-scale, high-quality trial outcome data hinders research in predictive modelling and evidence-based decision-making. Here we introduce the Clinical Trial Outcome (CTO) benchmark, a scalable, reproducible and large-scale dataset covering 125k drug and biologics trials. We integrate interpretations of trial publications, phase-wise tracking, news sentiment, stock price and trial-related metrics. We additionally manually annotated a subset of recent trials from 2020 to 2024. CTO labels align with previous expert annotations, achieving a 91% F1 score. Predictive models trained on CTO consistently outperform models trained on older, static data when evaluated year over year on recent trials, mitigating performance decay from data distribution shifts. Beyond simply providing more outcome labels, CTO creates a continuously updated benchmark that supports trial design by enabling models and analysts to evaluate how design decisions relate to success or failure over time, improving prediction, optimization and scenario planning for future studies. Beyond outcome prediction, the CTO benchmark can also serve as an open resource to support trial design and clinical development by informing end point selection, protocol refinement, enrolment feasibility and phase-transition decisions.