Division of Biostatistics, Office of Surveillance and Biometrics, Center for Devices and Radiological Health, Food and Drug Administration, 10903, New Hampshire Avenue, Building 66, Room 2268, Silver Spring, MD, 20993, USA.
Ther Innov Regul Sci. 2020 Jan;54(1):211-219. doi: 10.1007/s43441-019-00047-9. Epub 2020 Jan 6.
The US Food and Drug Administration conducts on-site inspections and data audits through Bioresearch Monitoring program for assurance of the quality and integrity of data in the pre- and postapproval processes. It is important to inspect the study sites that are different compared with other sites in clinical studies and identify the problems related to those sites. Usually one cannot inspect all the sites in a clinical study because of limited resources, and statistical tools are needed to help in selecting sites for inspection.
We propose two technical approaches, namely Fisher combination approach and likelihood ratio test (LRT) approach, for site selection, with each approach integrating the information obtained from a P value matrix. The proposed approaches produce site rankings, and the sites with highest rankings may be selected for inspection.
The application of the approaches is demonstrated through a hypothetical data set reflecting the pattern of the real data in a premarket approval submission for a diagnostic device. The proposed methods are shown, through extensive simulations, to control false discovery rate, while maintaining good sensitivity.
The proposed approaches will be useful for site selection process. However, limitations exist when only using the statistical approaches proposed here. In practice, investigators will select the site for inspection by considering the outputs from the statistical approaches along with other important factors. Future research topic is discussed to facilitate practical application of the approaches.
美国食品和药物管理局通过生物研究监测计划进行现场检查和数据审计，以确保在批准前和批准后过程中数据的质量和完整性。检查临床试验中与其他站点不同的研究站点并识别与这些站点相关的问题非常重要。由于资源有限，通常无法检查临床研究中的所有站点，因此需要统计工具来帮助选择要检查的站点。
我们提出了两种技术方法，即 Fisher 组合方法和似然比检验（LRT）方法，用于站点选择，每种方法都整合了从 P 值矩阵中获得的信息。所提出的方法会生成站点排名，排名最高的站点可能会被选中进行检查。
通过反映医疗器械上市前批准提交中真实数据模式的假设数据集来演示方法的应用。通过广泛的模拟，所提出的方法被证明可以控制假发现率，同时保持良好的灵敏度。
所提出的方法将有助于选择站点。但是，仅使用此处提出的统计方法存在局限性。在实践中，调查人员将通过考虑统计方法的输出以及其他重要因素来选择要检查的站点。讨论了未来的研究课题，以促进这些方法的实际应用。
