Cross-National Pharmacovigilance Data Pooling Accelerates Drug Safety Signal Detection: Evidence from Change-Point Analysis of Australia's DAEN and the US FAERS

Introduction: Current pharmacovigilance signal detection relies on cross-sectional disproportionality analysis, which does not examine when signals emerge or whether earlier detection is possible. No study has systematically applied modern change-point detection (CPD) methods to any pharmacovigilance database or tested whether cross-national data aggregation accelerates temporal signal detection. Objective: To determine whether CPD identifies emerging drug safety signals earlier than standard disproportionality analysis, and to evaluate whether pooling data across national pharmacovigilance databases improves detection timing. Methods: Three CPD algorithms -- Pruned Exact Linear Time (PELT), Cumulative Sum Control Charts (CUSUM), and Bayesian Online Change Point Detection (BOCPD) -- with ensemble voting were applied to quarterly drug-specific adverse event reporting proportions in Australia's Database of Adverse Event Notifications (DAEN; 664,747 reports, 2004-2025). A rolling Proportional Reporting Ratio (PRR) served as the baseline comparator. Detection performance was validated against 32 Therapeutic Goods Administration (TGA) regulatory actions, 40 curated reference associations, and international OMOP/EU-ADR reference standards (111 confirmed positive controls, 87 testable negative controls). Parallel quarterly time series were constructed in the US FDA Adverse Event Reporting System (FAERS) for all 32 TGA validation pairs. Detection timing was compared across databases and on pooled DAEN+FAERS series. The study was pre-registered on the Open Science Framework prior to confirmatory analyses. Results: CPD detected signals a median of 12.0 quarters (3 years) before TGA regulatory action (20/23 confirmed detections preceded regulatory action, 87%). For genuine signal emergence -- where a drug was established in the DAEN before the adverse event appeared -- CPD detected signals a median of 5.0 quarters earlier than rolling PRR (Wilcoxon p=0.003). CPD and PRR were complementary: CPD outperformed PRR for established drugs developing new adverse events, while PRR was faster for new drugs with rapidly accumulating disproportionality. Overall sensitivity was 74% (Tier A) and 95% (Tier B) with 89% specificity. Prospective simulation confirmed that detection timing was maintained under realistic real-time monitoring conditions (90% concordance with retrospective results). In cross-database comparison, FAERS detected signals a median of 9.0 quarters earlier than DAEN (p=0.0001), reflecting its 10-1000-fold larger reporting volume. Pooling DAEN and FAERS data accelerated detection by a median of 8.0 quarters compared to DAEN alone (p<0.0001), with the greatest benefit for weak signals (15.0 vs 4.0 quarters gained for below- vs above-median EBGM, respectively). The detection gap between CPD signal emergence and regulatory action widened significantly over the study period (Spearman rho=0.78, p<0.0001), indicating that detectable signals increasingly precede regulatory response. Conclusions: Change-point detection provides a complementary approach to disproportionality analysis for pharmacovigilance, with the greatest advantage for identifying new adverse events in established drugs. Cross-national data pooling substantially accelerates signal detection, with the largest benefit for weak emerging signals that are hardest to detect in individual national databases. These findings support integrating CPD methods and cross-database temporal monitoring into regulatory pharmacovigilance operations.