Advancing spatiotemporal individual-level modeling of infectious disease transmission dynamics

Abstract

This thesis advances spatiotemporal modeling of infectious disease transmission dynamics by addressing key limitations of the Geographically Dependent Individual-Level Model (GD-ILM) framework. These limitations include reinfection, seasonal transmission, and computational inefficiency for large-scale epidemics. This work is further supported by development of two R packages. In addition, spatiotemporal analyses of Gonorrhea data from Manitoba identify persistent hotspots for targeted interventions. First, the GD-ILM is extended within a Susceptible-Exposed-Infectious-Recovered-Susceptible framework to capture reinfection dynamics, enabling the model to estimate susceptibility to both primary infection and reinfection while preserving spatially explicit transmission patterns. Application to individual-level Tuberculosis data from Manitoba reveals significant regional and individual-level risk factors and produces fine-scale infection probability maps, providing actionable insights for public health intervention. Second, the GD-ILM is extended to incorporate seasonally varying transmission, capturing temporal fluctuations in infection risk due to environmental, behavioral, and pathogen-driven factors. Applied to Influenza data from Manitoba, the seasonal GD-ILM identifies high-risk regions and periods, demonstrating the importance of integrating temporal dynamics into spatial models, with simulations confirming the ability to recover spatiotemporal patterns. Parameter estimation in both extensions is performed using a likelihood-based Monte Carlo Expectation Conditional Maximization algorithm. Third, to overcome computational challenges posed by large-scale epidemics when using the GD-ILM framework, a stratified temporally-weighted Kernel Density Estimation-based Probability Proportional to Size sampling approach, combined with Stochastic Approximation Expectation Conditional Maximization, is developed, enabling efficient parameter estimation and real-time application, and is illustrated through simulations and COVID-19 data from Manitoba. Fourth, two R packages, "GDILM.SEIRS" and "SeasEpi", are developed and made publicly available on the Comprehensive R Archive Network (CRAN) to facilitate reproducibility and practical adoption of the proposed frameworks in future spatiotemporal individual-level modeling of infectious disease transmission dynamics. Finally, spatial, temporal, and spatiotemporal cluster detection analyses of Gonorrhea data from Manitoba identify persistent high-risk areas and time periods, guiding targeted public health interventions.