Using a novel Arizona Medicaid data set, we model Valley fever (VF) cases in the Phoenix Metropolitan Area as a spatial point process during six month intervals between 2013-2023. We estimate the intensity function of VF cases, observed at residential locations of patients, as a function of environmental covariates available at high spatial resolutions: exposure to various land cover development categorizes, exposure to changes in land cover type, Normalized Difference Vegetation Index (NDVI), change in NDVI, and a Habitat Suitability Index (HSI) for the Coccidioides fungus. All covariate data were spatially smoothed using a Gaussian kernel with a 1-kilometer bandwidth to capture local variation around residential addresses of patients. Models incorporating environmental covariates perform best in 21 of 22 semesters. We show that HSI measured at the residential (point) level is useful for understanding VF risk. Furthermore, we find that NDVI is often negatively associated with the intensity of VF cases. Despite improved model fit when compared to our simplest model, we show that even our best, most complex model underestimates the intensity in portions of the study region with lower population density. Our results provide initial evidence that spatially localized environmental information is useful for explaining the intensity of VF cases among Medicaid patients in the study window.
Ginos et al. (Sun,) studied this question.