Unraveling the drivers of leptospirosis risk in Thailand using machine learning

Leptospirosis poses a significant public health challenge in Thailand, driven by a complex mix of environmental and socioeconomic factors. This study develops an XGBoost machine learning model to predict leptospirosis outbreak risk at the provincial level in Thailand, integrating climatic, socioeconomic, and agricultural features. Using national surveillance data from 2007-2022, the model was trained to classify provinces as high or low risk based on the median incidence rate. The model’s predictive performance was validated for the years 2018–2022, spanning pre-COVID-19, COVID-19, and post-COVID-19 periods. SHapley Additive exPlanation (SHAP) analysis was employed to identify key predictive factors. The optimized XGBoost model achieved high predictive accuracy for the pre-pandemic (AUC = 0.937 with 95% CI: 0.878 – 0.976) and post-pandemic (AUC = 0.951 with 95% CI: 0.861 – 0.999) testing periods. SHAP analysis revealed rice production factors, household size, and specific climatic variables as the strongest predictors of leptospirosis risk. However, model performance declined during the COVID-19 pandemic (2020–2021), suggesting surveillance disruption and potential underreporting. This study demonstrates the utility of machine learning for predicting leptospirosis risk in Thailand and highlights the complex interplay of environmental and socioeconomic factors in driving outbreaks. The adaptable modeling framework provides a foundation for developing early warning systems and targeted interventions to reduce the burden of this neglected tropical disease.Author summary: Leptospirosis, a disease caused by Leptospira bacteria, poses a significant public health challenge in Thailand. The bacteria thrive in contaminated environments, particularly those associated with rice farming. In this study, we developed a machine learning model to predict the risk of leptospirosis outbreaks in Thailand based on climatic, socioeconomic, and agricultural factors. Our analysis revealed that rice production practices, household size, and specific climatic variables were the strongest predictors of leptospirosis risk. We also observed a reduction in model performance during the COVID-19 pandemic, suggesting surveillance disruptions and potential underreporting. These findings highlight and explain the complex interplay of environmental and socioeconomic factors in driving leptospirosis outbreaks. Our adaptable modeling framework provides a foundation for developing early warning systems and targeted interventions to reduce the burden of this often-overlooked tropical disease. Better understanding the factors that contribute to leptospirosis risk can guide responses to protecting vulnerable populations and improving public health outcomes in Thailand and beyond in times of socio-environmental changes.