Climate extremes, WASH deficits, and infectious diseases in the Brazilian Amazon: Insights from explainable machine learning (2010-2022).

Infectious diseases transmitted through environmentally mediated faecal-oral pathways and climate-sensitive vector-borne routes remain major public health challenges in the Brazilian Amazon, where sanitation deficits interact with hydroclimatic variability. This study investigated how ENSO-related climate variability, sanitation conditions, and socioenvironmental factors jointly influence disease incidence across seven Amazonian states from 2010 to 2022. Epidemiological, climatic, and socioeconomic datasets were integrated using statistical models and explainable machine learning approaches. Disease incidence exhibited strong heterogeneity across transmission pathways. Climate was the dominant predictive domain for faecal-oral, contact-related, and arboviral diseases (mean SHAP up to 0.871), whereas spatial structure dominated protozoan vector-borne diseases and the combined model (up to 0.628). At the variable level, contributions were concentrated in a limited set of predictors, particularly state effects (up to 0.401) and key climatic variables, including temperature and lagged humidity and rainfall. Lagged climatic variables (t - 1) contributed substantially to model performance, indicating delayed responses to environmental variability. Generalized additive models showed high explanatory capacity, with explained deviance ranging from 68.2% to 94.9% across disease groups. Ensemble models outperformed conventional approaches, capturing nonlinear relationships among predictors. Overall, climate variability acted as a modulating factor within a broader system of structural vulnerability, rather than as an isolated driver. These findings support integrated strategies combining sanitation improvements, climate-informed early warning systems, and spatially targeted public health interventions.