Spatiotemporal prediction of scrub typhus incidence and environmental risk factors in Republic of Korea: a Bayesian hierarchical approach.

BACKGROUND

Scrub typhus is a vector-borne infectious disease that is a major public health concern in Republic of Korea. Environmental factors are associated with disease transmission. However, comprehensive predictive models incorporating multiple environmental determinants while accounting for spatial and temporal correlations remain limited. This study aimed to characterize the spatiotemporal patterns of scrub typhus incidence and develop a Bayesian spatiotemporal prediction model incorporating multiple environmental covariates across 229 administrative units in the Republic of Korea from 2015 to 2024.

METHODS

We developed a comprehensive analytical framework using national surveillance data from 229 administrative units collected from January 2015 to December 2024. Seasonal-trend decomposition, hotspot analysis, and negative binomial regression were employed to characterize spatiotemporal patterns to inform the Bayesian spatiotemporal modelling framework. Bayesian prediction modelling using an Integrated Nested Laplace Approximation was implemented to analyse environmental covariates, including temperature, relative humidity, normalized difference vegetation index (NDVI), elevation, and cropland ratio, while accounting for spatial clustering and temporal autocorrelation.

RESULTS

Spatiotemporal analyses revealed distinct seasonal dynamics, with November incidence rates 91-fold higher than those in February [incidence rate ratio (IRR): 91.00, 95% confidence interval (CI) 66.50-125.00], and significant geographic clustering in southern provinces. The Bayesian model identified significant positive associations between disease risk and temperature [IRR: 1.02 per 1 °C, 95% credible interval (CrI): 1.01-1.04], NDVI (IRR: 1.07 per 0.1-unit, 95% CrI: 1.03-1.11), and elevation (IRR: 1.32 per 100 m, 95% CrI: 1.15-1.51). Relative humidity showed contrasting effects depending on the lag period, with protective effects at a 1-month lag (IRR: 0.96, 95% CrI: 0.93-0.999) but increased risk at a 2-month lag (IRR: 1.16, 95% CrI: 1.12-1.21).

CONCLUSIONS

This comprehensive framework successfully captured spatiotemporal disease patterns and provided quantitative risk assessment capabilities for evidence-based public health planning and targeted preventive strategies.