Health data and environmental data are commonly collected at different levels of aggregation. A persistent challenge of using a spatial regression model to link these data is that their associations can vary as a function of aggregation. This results into ecological fallacy if association at one aggregation level is used for inferencing at another level. We address this challenge by presenting a hierarchically adaptable spatial regression model. In essence, the model extends the spatially varying coefficient model to allow the response to be count data at larger aggregation levels than that of the covariates. A Bayesian hierarchical approach is used for inferencing the model parameters. Robust inference and optimal prediction over geographical space and at different spatial aggregation levels are studied by simulated data sets. The spatial associations at different spatial supports are largely different, but can be efficiently inferred when prior knowledge of the associations is available. The model is applied to study hand, foot and mouth disease (HFMD) in Da Nang city, Viet Nam. Decrease in vegetated areas corresponds with elevated HFMD risks. A study to the identifiability of the parameters shows a strong need for a highly informative prior distribution. We conclude that the model is robust to the underlying aggregation levels of the calibrating data for association inference and it is ready for application in health geography.