Geographical and socioeconomic inequalities in breast cancer mortality: a machine learning analysis in Brazil

Socioeconomic inequalities shape breast cancer mortality in complex ways, particularly in countries with substantial territorial heterogeneity, such as Brazil. This study investigated the spatial and nonlinear determinants of breast cancer mortality among Brazilian women from 2015 to 2019, using Immediate Geographic Regions (IGRs) as the units of analysis. We applied machine learning techniques, including Random Forest (RF), Geographical Random Forest (GRF), and Shapley-value interpretation, to estimate the global and local importance of 23 sociodemographic, socioeconomic, and health service-related variables. The RF model yielded a root mean squared error (RMSE) of 2.33 and an R² of 62.8 %, whereas the GRF produced an RMSE of 2.31 and an R² of 63.9 % and revealed marked spatial variation in predictor relevance. Variables reflecting regional socioeconomic development, particularly female-headed households, Gross Domestic Product (GDP) per capita, population density, completed secondary education, and the availability of specialist physicians, were notably important at the national and regional levels. The magnitude and direction of each variable's contribution did not necessarily correspond to its regional prevalence, underscoring the presence of nonlinear and spatially heterogeneous relationships between variables. Understanding these spatial, socioeconomic, demographic, and structural patterns is essential for informing health policy designs. Such insights support the development of territorially sensitive interventions that move beyond generalized national strategies toward approaches tailored to Brazil's diverse regional contexts.