Methods/Statistics
A Two-Stage Bayesian Framework for Pathway-Aware Inference of Infectious Disease Risk Following Extreme Weather Events Brittany L. Morgan Bustamante* Brittany L. Morgan Bustamante Morgan Bustamante Morgan Bustamante Morgan Bustamante Morgan Bustamante Morgan Bustamante Morgan Bustamante Division of Environmental Health Sciences, University of California, Berkeley, Berkeley, CA 94720
Background: Extreme weather events can influence infectious disease risk through pathways involving exposure, built and social environments, and healthcare utilization. Standard regression approaches often estimate associations but treat pathways as independent. This limits pathway-aware inference—whether associations operate directly or indirectly through intermediate system components—and the ability to identify effective interventions.
Methods: We developed a two-stage Bayesian framework to evaluate the influence of extreme weather events on infectious disease outcomes. First, we modeled spatio-temporal disease counts using a Bayesian conditional Poisson model with hierarchical random effects, temporal smoothing, and distributed lag terms to estimate changes in disease incidence associated with extreme weather events while accounting for spatial heterogeneity and secular trends. Second, we applied Bayesian Network Analysis (BNA) to integrate posterior estimates of event-associated risk from the Poisson model with measures of flood exposure, social/built environment, population health, and healthcare utilization, allowing the BN to characterize relationships among system components after accounting for spatio-temporal patterns. We demonstrate the framework using U.S. surveillance and electronic health record data from 2016-2023 to evaluate flooding-aspergillosis associations.
Results: In the flooding-aspergillosis case study, conditional Poisson models indicated that moderate-severe flooding events were associated with increased incidence of aspergillosis, with elevated risk following exposure and a subsequent decline at longer lags (7-12 months). BNA further highlighted complex pathways through which flooding may influence aspergillosis risk.
Conclusions: This work advances a systems-oriented epidemiologic framework that combines conditional Poisson regression with BNA to evaluate infectious disease risk following extreme weather events. Together, this approach supports more pathway-aware inference by distinguishing direct and indirect associations operating through interconnected system components that are difficult to assess using single-equation models and is broadly applicable across environmental exposures and health outcomes.
