Abstract
INTRODUCTION: Despite statewide progress and continuous HIV prevention efforts in Texas, HIV diagnosis at a late stage of infection persists. Diagnosis delay differs in magnitude and spatial distribution. We examined the local spatial relationships of late HIV diagnosis with a selection of variables in an area of Texas that includes large metropolises and high HIV morbidity. METHODS: We compared regression modeling approaches to study the associations between the regional percentage of late HIV diagnosis from 2011 through 2015, regional measures of poverty, lack of health insurance (uninsurance), educational attainment, unemployment, and the average regional distance from residence to an HIV testing site: global ordinary least squares linear regression, spatial error model, geographically weighted regression, and multiscale geographically weighted regression (MGWR). Cartographic representation of the local R(2), coefficient estimates, and their t values assisted in the interpretation of results. RESULTS: The MGWR model resulted in a better fit and identified education and uninsurance as globally fixed predictors, whereas the relationships between late HIV diagnosis and poverty, unemployment, and distance varied spatially. The model performed better in rural areas and in suburban areas of the largest cities than in urban areas. CONCLUSION: The MGWR results provided local estimates of associations. The results highlight the importance of focusing on a local context. Modeling at the local scale is particularly useful for characterizing relationships between explanatory and dependent variables when the relationships vary spatially. In the context of HIV prevention, relationships that are of local relevance can inform local policy and complement routine screening in clinical settings.