Statistical Methods for Multi-jurisdictional Australian Vaccine Safety Investigations of Rare Adverse Events.

Background: In Australia, surveillance of adverse events following immunisation is primarily conducted by states and territories, with each jurisdiction only able to view and analyse reports originating from their own population. Distributed data models (aka federated data models) are a form of decentralised collaboration, with each site maintaining ownership of its data from end-to-end including data collection, storage and analysis. The primary benefit of this model is that it maintains independence and autonomy while enabling interdependence, collaboration and scalability.

Objective: We aimed to investigate statistical methods for a multi-jurisdictional collaboration when conducting a rigorous assessment of rare adverse events following immunisation at a national level.

Methods: Victoria and Western Australia have independently established routine data linkage for vaccine safety surveillance. A data collaboration model is proposed, whereby each jurisdiction can generate de-identified population-level data for adverse events following immunisation, using agreed case definitions and analytical methods. To demonstrate its utility, Victoria and Western Australia combined data from a self-controlled case series via a meta-analysis approach using aggregate data and a pooled approach using individual-level data to investigate the association between coronavirus disease 2019 vaccines and Guillain-Barré syndrome.

Results: There were 519 and 176 new Guillain-Barré syndrome International Classification of Diseases, Tenth Revision, Australian Modification coded admissions in Victoria and Western Australia, respectively, between 01/01/2020 and 31/12/2023. Combining data using a fixed-effect meta-analysis method (relative incidence: 2.64, 95% confidence interval 1.90, 3.66) and a pooled method (relative incidence: 2.45, 95% confidence interval 1.76, 3.41) confirmed the known increased incidence in the 42 days following a coronavirus disease 2019 Vaxzevria^® vaccination. Both methods resulted in a decreased standard error when compared with either state alone.

Conclusions: This project represents an ongoing successful collaboration between two Australian jurisdictions using data linkage to investigate rare adverse events following immunisation and inform accurate benefit-risk analyses. The decision to use meta-analysis and pooled analysis methods should be considered on a case-by-case basis and may depend on data-sharing agreements, the ease of pooling potentially discordant data variables and underlying population characteristics.