Applying causal inference and Bayesian statistics to understanding vaccine safety signals using a simulation study

Abstract

Abstract Community perception of vaccine safety influences vaccine uptake. Our objective was to assess current vaccine safety monitoring by examining factors that may influence the availability of post-vaccination survey data, and thereby the specificity and sensitivity of existing signal detection methods. We used causal directed acyclic graphs (DAGs) and a Bayesian posterior predictive analysis (PPA) signal detection method to understand biological and behavioural factors which may influence signal detection. The DAGs informed the data simulated for scenarios in which these factors were varied. The influence of biological factors such as severity of adverse reactions and behavioural factors such as healthcare-seeking behaviour upon survey participation was found to drive signal detection. Where there was a low prevalence of moderate to severe reactions, false signals were detected when there was a strong influence of reaction severity on both survey participation and seeking medical attention. These findings provide implications for future vaccine safety monitoring.