Spatiotemporal modelling of meteorological drivers and outbreak detection of Legionnaires disease in England

Abstract

Legionnaires disease is a severe respiratory illness caused by Legionella bacteria, with most cases occurring sporadically and environmental sources often unidentified. Effective outbreak detection requires understanding the spatiotemporal dynamics of sporadic cases and their environmental drivers. We developed a mechanistically informed spatiotemporal model integrating fine-scale spatial heterogeneity, multi-week meteorological influences, and extended temporal lags. The framework combines a negative binomial generalised additive model (GAM), a Besag-York-Mollie (BYM2) spatial component, and distributed lag nonlinear models (DLNMs) to capture nonlinear, delayed effects of temperature, dewpoint depression, precipitation, and cloud cover. These outputs generate a national daily index of weather-driven vulnerability, which is combined with hierarchical clustering to identify potential outbreaks. Across 2000-2019, our model improved outbreak detection in 15 of 20 years compared with the baseline UKHSA approach; in the remaining years performance was either equivalent (two years) or only slightly worse (three years, 0.98% reduction). Mean relative improvements were 6.0%, with a maximum of 14.4% in 2013. Improvements were consistent across months, and coarser 0.25-degree grid evaluations likely underestimate the models advantage at finer spatial scales. The analysis also clarified dual-stage Legionnaires disease dynamics, distinguishing environmental bacterial growth from the shorter infection window, and demonstrated the necessity of extended lags for accurate risk prediction. This framework provides a robust platform for targeted surveillance and predictive modelling, supporting evidence-based interventions and enhancing preparedness for sporadic Legionnaires disease under observed climatic conditions.