Unraveling social and environmental drivers of heat-related hospitalizations in the Netherlands through Random Forest analysis

Extreme heat has increasingly affected population health over recent decades, with rising occurrences of heat-related mortality and morbidity across different climate zones. The severity of these impacts, however, is not solely determined by ambient temperature; it is profoundly shaped by environmental and social factors such as demographic composition, living and labor conditions, income and education levels. These factors jointly determine vulnerability and adaptive capacity, translating social inequalities into disproportionate health impacts among specific population groups. This study aims to quantitatively characterize the interplay of these social and environmental factors in shaping differences in heat-related hospitalizations in the Netherlands. We focus on admissions due to cardiovascular, respiratory, and direct heat-exposure conditions such as dehydration, renal failure, and heat stroke. Using municipal-level data from Statistics Netherlands (CBS) and climate indicators over a five-year period, we applied Random Forest regressor and classifier algorithms to explore the relationships between heat-related morbidity and a wide set of socioeconomic and demographic variables. Through SHapley Additive exPlanations (SHAP), we interpret the relative importance and interaction effects of predictors while accounting for multicollinearity and nonlinear relationships, advancing over conventional linear models commonly used in vulnerability assessments. The results highlight dominant vulnerability patterns associated with age structure, marital status, labor participation, income, and social assistance, and differentiate linear, nonlinear and threshold effects across variables. The spatial character of the analysis allows the identification of municipalities where multiple vulnerability drivers converge, indicating local “hotspots” of heat-related risk. Our results demonstrate the value of machine learning approaches for uncovering complex, intersectional patterns of vulnerability to extreme heat. Beyond methodological advancement, this work provides actionable insights for spatially targeted adaptation planning and public health interventions. It underscores the urgency of integrating health, social, and climate data in national adaptation strategies to protect populations disproportionately affected by intensifying heat extremes.