Evaluating medical device adverse event signals using a likelihood ratio test method.

Area of Science:

• Medical device safety
• Pharmacovigilance
• Biostatistics

Background:

• Signal detection methods are crucial for post-market surveillance of adverse events (AEs).
• Statistical methods for AE signal detection are underutilized in medical device safety monitoring.
• Left ventricular assist devices (LVADs) require robust methods for identifying potential safety concerns.

Purpose of the Study:

• To evaluate the efficacy of a likelihood ratio test (LRT)-based method for detecting AE signals in LVADs.
• To apply LRT to Medical Device Reporting (MDR) data for identifying specific safety concerns.
• To compare the performance of LRT against other established signal detection methods.

Main Methods:

• Utilized a likelihood ratio test (LRT) for signal detection on 110,927 LVAD adverse event entries from MDR data.
• Applied LRT to longitudinal data (2007-2019) with an alpha-spending function for trend analysis and type I error control.
• Compared LRT performance against Proportional Reporting Ratios (PRRs), Bayesian Confidence Propagation Neural Network (BCPNN), and simplified Bayes methods.

Main Results:

• The LRT method identified 18 distinct AE signals from the LVAD MDR data.
• Seven of the detected signals were related to bleeding events, including hemolysis, thrombosis, and hemorrhage.
• LRT demonstrated the most conservative performance among the compared methods, effectively controlling type I error and false discovery rates.

Conclusions:

• The likelihood ratio test (LRT) is a valuable and conservative statistical method for detecting adverse event signals in medical device data, specifically for LVADs.
• LRT provides reliable identification of safety signals, including critical bleeding-related events, enhancing post-market surveillance.
• The LRT method's ability to control statistical errors makes it a robust tool for ensuring the safety and effectiveness of medical devices.