Application of a likelihood ratio test based method for safety signal detection to left ventricular assist devices.

Effective post-market safety surveillance of medical devices is critical for public health. However, many current statistical methods for safety signal detection do not control for type I error when assessing multiple device and adverse event (AE) combinations. This can result in increased false signals, underscoring the need for more robust statistical methods. Moreover, the duration of medical device use can be an important factor to consider in safety surveillance. In this study, we adapted a likelihood ratio test (LRT) based method, which was initially developed and applied to drugs, to identify safety signals for left ventricular assist devices (LVAD). Among patients with chronic, advanced left ventricular failure, we analyzed AE data for HeartWare and HeartMate II patients during a two-year period and further incorporated person-years (henceforth exposure-time). The novel modified LRT and conventional Z-test with p-values adjusted by the Benjamini-Hochberg (BH) procedure were used to explore safety signals by comparing HeartWare and HeartMate II patients in the presence of multiple adverse events. Both methods identified greater incidence of stroke among HeartWare as compared to HeartMate II patients without exposure-time (p = .025 for LRT and p = .027 for Z-test with BH) and with exposure-time (p = .002 for LRT and p = .005 for Z-test with BH). By using improved statistical methods for safety signal detection, potential safety issues can be identified and addressed in a more timely manner to enhance public safety.