序贯事件率监测
提出一种名为序贯事件率监测(SERM)的新方法,用于临床试验中事件率的连续监测。该方法基于序贯概率比检验(SPRT),采用Kim和Woodroofe(2003)由非线性更新定理推导的改进边界,并以一项大型III期临床试验的真实数据演示其快速评估潜力。
要点
- SERM基于SPRT与改进边界,决策边界可由简单公式直接获得。
- SERM可在盲态数据上运行,能与多种研究设计配合并保持研究完整性。
- 患者入组缓慢会延长得出结论所需时间,但不显著影响决策相关的I类或II类错误。
核心结论
- SERM为临床试验监测提供了一种稳健工具,有助于在不同临床环境中及时做出有依据的决策。
研究设计与样本
- 方法学文章,提出序贯事件率监测(SERM)方法,并以一项大型III期临床试验的真实数据作详细演示。
主要发现
- SERM基于序贯概率比检验(SPRT),采用Kim和Woodroofe(2003)由非线性更新定理推导的改进边界,是其理论工作在该领域的首次实际应用。
- SERM易于实施、能高效利用数据,并可广泛适用于各类试验。
- SERM在盲态数据上运行,可与多种研究设计配合使用并保持研究完整性。
- 患者入组缓慢会延长得出结论所需时间,但不显著影响决策相关的I类或II类错误。
局限性
- 患者入组缓慢会延长得出结论所需时间,但不显著影响决策相关的I类或II类错误。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Effective monitoring of event rates is essential for maintaining statistical power and study integrity in clinical trials, particularly when the primary endpoint involves time-to-event outcomes. We propose Sequential Event Rate Monitoring (SERM), a new and innovative approach for continuous monitoring of event rates. SERM leverages the Sequential Probability Ratio Test (SPRT) with improved boundaries derived from the nonlinear renewal theorem by Kim and Woodroofe (2003). This method represents the first practical implementation of their theoretical work in this area. SERM offers several tangible benefits, including ease of implementation, efficient use of data, and broad applicability to trials. Decision boundaries can be directly obtained from simple formula. A detailed illustration of the method using real-world data from a large Phase III clinical trial demonstrates its potential for rapid assessment. SERM operates on blinded data so it can be used in tandem with a broad range of study designs while preserving study integrity. Although slow patient accrual lengthens the time needed to reach a conclusion, it does not significantly affect type I or type II errors associated with the decision. This new method provides a robust tool for enhancing trial monitoring, enabling timely and informed decision-making in diverse clinical settings.
引用
- Kim DY, Han SM. Sequential Event Rate Monitoring. Stat Med. 2026. PMID: 41572588. doi:10.1002/sim.70359
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。