Abstract:
Discovering all drug and vaccine side effects during the development process is impos-sible. This dissertation aims to propose a framework in exploring spatiotemporal adverse event surveillance models by identifying adverse effects, which co-locate together and is associated with FDA approved drugs or vaccines using spatial statistics and spatial science. This study aims to find statistically significant spatio-temporal clusters among co-occurring adverse effects. We use data obtained from the FDA’s Adverse Event Reporting System (FAERS) and Vaccine Adverse Event Reporting System (VAERS) to explore the spatio- temporal distribution of combinations of adverse effects using two methods: • Frequent Itemset Mining - to mine the most frequent sets of adverse events. • Latent Dirichlet allocation (LDA) -to mine the most frequent group of topics related to adverse effects. To assess the similarity of sets of adverse events or topics between spatial regions, we employ textual comparison algorithms. We apply an agglomerative hierarchical clustering approach to find clusters of regions that exhibit similar adverse events or topics. Finally, we explore the resulting clusters to discover spatial autocorrelation patterns using Global and Local Moran’s I measure of spatial autocorrelation. Our approach can be applied to any product where after consumption or application results in adverse events, to study if spatially localized side-effects that may justify further investigation.
