Bibiza-Freiwald Eric, Vach Werner, Zapf Antonia
Department of Medical Statistics, SGS proderm GmbH, Schenefeld, Germany.
Institute of Medical Biometry and Epidemiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany.
Stat Med. 2025 Mar 15;44(6):e70016. doi: 10.1002/sim.70016.
In determining the accuracy of a new diagnostic test, often two steps are performed. In the first step, a case-control study is performed as an efficient but potentially biased design. In a second step, a population-based cohort study is performed as an unbiased but less efficient design. In order to accelerate diagnostic research, it has recently been suggested to combine the two designs in one seamless design. In this article, we present a more in-depth description of this idea. The seamless diagnostic accuracy study design is formally introduced by comparison with the traditional pathway, and the basic design decisions are discussed: A stopping rule and a stopping time. An appealing feature of the design is the possibility to ignore the seamless design in the final analysis, although part of the data is used already in an interim analysis. The justification for this strategy is provided by a large-scale simulation study. The simulation study suggests also that the risk of a loss of power due to using a seamless design can be limited by a reasonable choice of the futility boundaries, defining the stopping rule. We conclude that the seamless diagnostic accuracy study design seems to be ready to use. It promises to accelerate diagnostic research, in particular if population-based cohort studies can be started without great efforts and if the reference standard can be evaluated with little delay.
在确定一项新诊断测试的准确性时，通常会执行两个步骤。第一步，进行病例对照研究，这是一种高效但可能存在偏差的设计。第二步，进行基于人群的队列研究，这是一种无偏差但效率较低的设计。为了加速诊断研究，最近有人建议将这两种设计结合成一种无缝设计。在本文中，我们对这一想法进行了更深入的描述。通过与传统方法进行比较，正式引入了无缝诊断准确性研究设计，并讨论了基本的设计决策：停止规则和停止时间。该设计的一个吸引人的特点是，尽管部分数据已在中期分析中使用，但在最终分析中可以忽略无缝设计。大规模模拟研究为这一策略提供了依据。模拟研究还表明，通过合理选择定义停止规则的无效边界，可以限制因使用无缝设计而导致的效能损失风险。我们得出结论，无缝诊断准确性研究设计似乎已可投入使用。它有望加速诊断研究，特别是在基于人群的队列研究能够轻松启动且参考标准能够及时评估的情况下。
