ABSTRACT

Signal detection methods have been used extensively in post-market surveillance to identify elevated risks of adverse events. However, these statistical methods have not been widely used in detecting AE signals for medical devices. In this paper, we focused on the use of a likelihood ratio test (LRT)-based method in identifying adverse event (AE) signals associated with left ventricular assist devices (LVADs) using Medical Device Reporting (MDR) data. Among 110,927 adverse event entries identified in MDR data for LVADs, the LRT method detected 18 AE signals which included seven bleeding-related AEs such as hemolysis, thrombosis, hematuria, thrombus, blood loss, and hemorrhage. The LRT method was also applied to longitudinal data from 2007 to 2019 where a monotone alpha-spending function was used to ensure the control of type I error at each look and overall for trend analysis. Furthermore, the LRT method was compared to proportional reporting ratios (PRRs), Bayesian confidence propagation neural network (BCPNN), and simplified Bayes methods and found to be the most conservative method when examining the total number of detected signals, given its ability to control type-I error and the false discovery rate.

摘要

信号检测方法已广泛用于上市后监测，以识别不良事件的高风险。然而，这些统计方法并没有被广泛用于检测医疗设备的AE信号。在本文中，我们专注于使用基于似然比检验 (LRT) 的方法来识别与使用医疗器械报告 (MDR) 数据的左心室辅助装置 (LVAD) 相关的不良事件 (AE) 信号。在 LVAD 的 MDR 数据中确定的 110,927 个不良事件条目中，LRT 方法检测到 18 个 AE 信号，其中包括 7 个与出血相关的 AE，如溶血、血栓形成、血尿、血栓、失血和出血。LRT 方法还应用于 2007 年至 2019 年的纵向数据，其中使用单调 alpha 支出函数来确保控制每种外观和整体的 I 型错误以进行趋势分析。此外，LRT 方法与比例报告比 (PRR)、贝叶斯置信传播神经网络 (BCPNN) 和简化的贝叶斯方法进行了比较，发现它是检查检测到的信号总数时最保守的方法，因为它具有控制能力I 型错误和错误发现率。