The real-world analysis of tinnitus-related adverse events across systemic medications: a pharmacovigilance study based on the FAERS and WHO-VigiAccess databases

Abstract Background Tinnitus is a prevalent and often persistent symptom that imposes a substantial psychological and functional burden. Beyond disease-related mechanisms, accumulating real-world observations suggest that tinnitus may also emerge as a manifestation of drug-related adverse events. However, systematic, cross-database evidence characterizing tinnitus-associated drug safety signals remains limited. Methods This study identified 21 systemic medications previously reported in association with tinnitus and retrieved post-marketing reports from the FDA Adverse Event Reporting System (FAERS) (2004–2025) and the WHO VigiAccess database. After data cleaning and standardization, disproportionality analyses were performed using four complementary algorithms (reporting odds ratio, proportional reporting ratio, Bayesian Confidence Propagation Neural Network, and Multi-item Gamma–Poisson Shrinker). Drug-event pairs meeting predefined thresholds were considered signals. We further conducted sex-stratified analyses, quantified the relative contribution of age, sex, and body weight to adverse events using an XGBoost model, and examined time-to-onset patterns with Weibull shape parameter modeling. Results Across databases, tinnitus-related adverse event signals were not randomly distributed but instead converged within specific system organ classes and formed recognizable clusters at the Preferred Term level, particularly within neuropsychiatric and musculoskeletal domains. Most reports originated from adult and older adult populations, with females being more frequently represented. Tinnitus appeared as a recurrent positive signal across multiple therapeutic categories, and a considerable proportion of signals overlapped between FAERS and VigiAccess, supporting robustness while revealing complementary differences between the databases. After stratification, several signals persisted across sexes, whereas age explained the greatest proportion of variability compared with sex and body weight. Moreover, time-to-onset analyses consistently suggested an early-failure risk pattern, in which adverse events occurred predominantly shortly after treatment initiation and then declined. Conclusion Integrating two large pharmacovigilance databases, we mapped consistent risk signals for tinnitus across multiple medications. These findings support closer monitoring during early treatment and offer real-world evidence to guide safer prescribing and future mechanistic research.