Clustering Adverse Events of COVID-19 Vaccines across the United States
Ahmed Askar and Andreas Züfle
We study the similarity of adverse effects of COVID-19 vaccines across different states in the United States. We use data of 300,000 COVID-19 vaccine adverse event reports obtained from the Vaccine Adverse Event Reporting System (VAERS). We extract latent topics from the reported adverse events using a topic modeling approach based on Latent Dirichlet allocation (LDA). This approach allows us to represent each U.S state as a low-dimensional distribution over topics. Using Moran’s index of spatial autocorrelation we show that some of the topics of adverse events exhibit significant spatial autocorrelation, indicating that there exist spatial clusters of nearby states that exhibit similar adverse events. Using Anselin’s local indicator of spatial association we discover and report these clusters. Our results show that adverse events of COVID-19 vaccines vary across states which justifies further research to understand the underlying causality to better understand adverse effects and to reduce vaccine hesitancy.