Classification of patient characteristics associated with reported adverse drug events to neuraminidase inhibitors: an applicability study of latent class analysis in pharmacovigilance.

Abstract

Signal detection in reports of spontaneous adverse drug reactions is useful in pharmacovigilance, but does not adequately consider potential confounding factors such as patient background information contained in the report data. Multiple indicators should be considered when generating safety hypotheses. The aim of this study was to evaluate whether latent class analysis (LCA) can complement conventional methods in pharmacovigilance. We conducted LCA of 2732 reports of adverse drug reactions involving four widely used anti-influenza neuraminidase inhibitors in the Japanese Adverse Drug Event Report (JADER) database covering April 2004 to June 2020. LCA classifies the target population into multiple clusters based on response probability. The same data was subjected to multivariate logistic regression using an adjusted reporting odds ratio. LCA grouped the target population into three classes. Cluster 1 (46.4%) contained patients who developed adverse events other than neuropsychiatric events; these events were specific to adult females. Cluster 2 (28.7%) contained patients who developed abnormal behavior; these events were specific to underage males. Cluster 3 (24.8%) contained patients who developed adverse neuropsychiatric events other than abnormal behavior, such as hallucinations and convulsion; these events were specific to minors. Logistic regression of adverse events for which a signal was detected identified factors similar to those found in LCA. LCA classified adverse events in JADER with similar incidence tendencies into the same cluster. The results included signals identified by conventional logistic regression, suggesting that LCA may be useful as a complementary tool for generating drug safety hypotheses.