Implementing Sequential Analysis for Vaccine Safety Surveillance in Victoria, Australia

Abstract

This preprint presents an end-to-end implementation framework for MaxSPRT-style sequential analysis in vaccine safety surveillance, using the Surveillance of Adverse Events Following Vaccination in the Community (SAEFVIC) system in Victoria, Australia, as the operational setting. Sequential methods such as the maximised sequential probability ratio test (MaxSPRT) allow repeated monitoring of accumulating adverse event reports while controlling the Type I error rate. However, translating the statistical method into a working surveillance system requires a series of practical decisions — about data preprocessing, outcome coding, risk-window definition, age stratification, comparator-rate construction, reporting-delay adjustment, expected-count estimation, and alpha-spending design — that are rarely documented in full in published applications. This manuscript provides a reproducible, step-by-step workflow covering each of these decisions. A SAEFVIC-derived synthetic dataset is used to demonstrate every calculation stage, from observed event counts and exposure-length approximation through to likelihood-ratio computation, alpha-spending boundaries, and signal review. The framework is intended to support transparent and auditable implementation of sequential vaccine safety surveillance in spontaneous reporting systems, particularly when near real-time vaccination denominator data are limited. The companion repository (this Zenodo record) contains the full worked tutorial, including Appendix A (terminology and notation), Appendix B (preprocessing steps, implementation details, and environment setup), Appendix C (step-by-step calculation details), and Appendix D (additional variations of sequential analysis). The repository supplements the manuscript and provides pseudocode, worked examples, and synthetic data sufficient to reproduce all calculations shown in the paper. This work was funded by the Royal Children's Hospital Foundation and was conducted at the Centre for Health Analytics, Melbourne Children's Campus, Parkville, Victoria, Australia.