Whitehead John, Horby Peter
Department of Mathematics and Statistics, Lancaster University, Lancaster, United Kingdom.
Centre for Tropical Medicine and Global Health, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom.
PLoS Negl Trop Dis. 2017 Mar 9;11(3):e0005439. doi: 10.1371/journal.pntd.0005439. eCollection 2017 Mar.
Conducting clinical trials to assess experimental treatments for potentially pandemic infectious diseases is challenging. Since many outbreaks of infectious diseases last only six to eight weeks, there is a need for trial designs that can be implemented rapidly in the face of uncertainty. Outbreaks are sudden and unpredictable and so it is essential that as much planning as possible takes place in advance. Statistical aspects of such trial designs should be evaluated and discussed in readiness for implementation.
METHODOLOGY/PRINCIPAL FINDINGS: This paper proposes a generic ordinal sequential trial design (GOST) for a randomised clinical trial comparing an experimental treatment for an emerging infectious disease with standard care. The design is intended as an off-the-shelf, ready-to-use robust and flexible option. The primary endpoint is a categorisation of patient outcome according to an ordinal scale. A sequential approach is adopted, stopping as soon as it is clear that the experimental treatment has an advantage or that sufficient advantage is unlikely to be detected. The properties of the design are evaluated using large-sample theory and verified for moderate sized samples using simulation. The trial is powered to detect a generic clinically relevant difference: namely an odds ratio of 2 for better rather than worse outcomes. Total sample sizes (across both treatments) of between 150 and 300 patients prove to be adequate in many cases, but the precise value depends on both the magnitude of the treatment advantage and the nature of the ordinal scale. An advantage of the approach is that any erroneous assumptions made at the design stage about the proportion of patients falling into each outcome category have little effect on the error probabilities of the study, although they can lead to inaccurate forecasts of sample size.
CONCLUSIONS/SIGNIFICANCE: It is important and feasible to pre-determine many of the statistical aspects of an efficient trial design in advance of a disease outbreak. The design can then be tailored to the specific disease under study once its nature is better understood.
开展临床试验以评估针对潜在大流行传染病的实验性治疗方法具有挑战性。由于许多传染病暴发仅持续六至八周，因此需要能够在面对不确定性时迅速实施的试验设计。疫情暴发是突然且不可预测的，所以尽可能提前进行规划至关重要。此类试验设计的统计学方面应进行评估和讨论，以便为实施做好准备。
方法/主要发现：本文提出了一种通用序贯试验设计（GOST），用于将针对新发传染病的实验性治疗与标准治疗进行比较的随机临床试验。该设计旨在作为一种现成的、随时可用的稳健且灵活的选择。主要终点是根据序贯量表对患者结局进行分类。采用序贯方法，一旦明确实验性治疗具有优势或不太可能检测到足够的优势，试验即停止。使用大样本理论评估该设计的特性，并通过模拟对中等规模样本进行验证。该试验旨在检测一种通用的临床相关差异：即较好而非较差结局的优势比为2。在许多情况下，两种治疗的总样本量在150至300名患者之间被证明是足够的，但具体数值取决于治疗优势的大小和序贯量表的性质。该方法的一个优点是，尽管设计阶段对落入每个结局类别的患者比例所做的任何错误假设可能会导致样本量预测不准确，但对研究的错误概率影响很小。
结论/意义：在疾病暴发之前预先确定高效试验设计的许多统计学方面是重要且可行的。一旦更好地了解所研究特定疾病的性质，就可以对该设计进行调整。
