Spatial analysis of county-level determinants of overdose mortality in the United States using spatial machine learning.

In recent years, there has been a growing body of literature on identifying effective determinants for modeling the spatial variation of overdose rates, addressing this emerging public health concern globally. We compiled a range of widely recognized factors to examine spatial heterogeneity and its associations with overdose mortality using a non-linear geographically weighted random forest approach. The model outperformed conventional ones with (R = 0.83 and MAE = 0.26). We found that, on average, the population rate of Asians (12.8%) is the most important determinant of the model, followed by the population rate of African Americans (10.1%) and the rate of cost-burdened housing units (9.9%). Although the results indicate that climatic determinants have had a lesser impact on overdose mortality rates, locally, their importance is greater in central and eastern counties. The spatial analysis revealed that the significance of determinants varies greatly by location. These findings could inform the development of localized spatial models, enabling more efficient allocation of resources to control overdose mortality rates at the community level.