Hickey Graeme L, Parvu Valentin, Zhang Yongqiang, Cooper Charles K, Wan Ying
Becton, Dickinson and Company, Franklin Lakes, New Jersey, USA.
Thrive, EXACT Sciences Corp, Cambridge, Massachusetts, USA.
J Biopharm Stat. 2023 Jan 2;33(1):77-89. doi: 10.1080/10543406.2022.2080690. Epub 2022 Jun 1.
Clinical studies are generally required to characterize the accuracy of new diagnostic tests. In some cases, historical data are available from a predicate device, which is directly relevant to the new test. If this data can be appropriately incorporated into the new test study design, there is an opportunity to reduce the sample size and trial duration for the new test. One approach to achieve this is the Bayesian power prior method, which allows for the historical information to be down-weighted via a power parameter. We propose a dynamic method to calculate the power parameter based on first comparing the data between the historical and new data sources using a one-sided comparison, and second mapping the comparison probability through a scaled-Weibull discount function to tune the effective sample size borrowed. This pragmatic and conservative approach is embedded in an adaptive trial framework allowing for the trial to stop early for success. An example is presented for a new test developed to detect Methicillin-resistant present in the nasal carriage.
临床研究通常需要对新诊断测试的准确性进行表征。在某些情况下，可以从与新测试直接相关的同类设备获得历史数据。如果这些数据能够适当地纳入新测试的研究设计中，就有机会减少新测试的样本量和试验持续时间。实现这一目标的一种方法是贝叶斯功效先验方法，该方法允许通过一个功效参数对历史信息进行加权。我们提出了一种动态方法来计算功效参数，首先使用单侧比较来比较历史数据源和新数据源之间的数据，其次通过缩放威布尔折扣函数映射比较概率，以调整借用的有效样本量。这种实用且保守的方法被嵌入到一个适应性试验框架中，允许试验因成功而提前终止。文中给出了一个用于检测鼻腔携带的耐甲氧西林金黄色葡萄球菌的新测试的示例。
