A1264
Title: Signal detection in adverse drug reactions under fluctuating reporting rates
Authors: Tatsuhiko Anzai - Institute of Science Tokyo (Japan) [presenting]
Kunihiko Takahashi - Institute of Science Tokyo (Japan)
Abstract: Spontaneous adverse event reporting systems, such as the Japanese adverse drug event report database (JADER) and the FDA adverse event reporting system (FAERS), are central resources for pharmacovigilance. A standard approach for signal detection is the reporting odds ratio (ROR), based on disproportionality analysis in a two-by-two contingency table. Conventional methods assume that reporting rates remain stable across drugs and events, but this assumption is frequently violated. During the COVID-19 pandemic, substantial shifts in reporting behavior were observed, leading to spurious increases in disproportionality measures. A statistical framework is proposed that incorporates reporting rate variations into the signal detection process. The method extends the contingency table with four fluctuation parameters representing deviations in cell-specific reporting ratios. These parameters are estimated by minimizing a divergence between observed and predicted counts, with predictions derived from a regression model assuming stable reporting rates. A two-layer error structure accounts for uncertainty in both predicted frequencies and observed data. The proposed method is applied to real pharmacovigilance data, focusing on psychotropic drugs where reporting rates are particularly unstable.
