A spatial analysis of county-level variation in hospitalization rates for low back problems in North Carolina

Hospitalization rates for low back problems vary widely. In previous non-spatial analyses, population-level socioeconomic and health resource characteristics have explained little of the variation in rates. This study examines geographic variation in hospitalization rates for low back problems while controlling for spatial dependence in the data. County-level surgical and medical hospitalization rates were calculated using North Carolina hospital (USA) discharge data from 1990–92. Non-spatial and spatial regression models were estimated using socioeconomic and health resource predictors. Both surgical and medical rates varied significantly among the 100 counties. Non-spatial models explained 62% of variation in log-transformed surgical rates and 66% of variation in log-transformed medical rates; however, residuals showed significant spatial dependence. Spatial lag models were therefore applied. Using simple contiguity spatial weights, surgery rates increased with higher percent urban population, primary care physician density, and discharge rate for other causes, and decreased with higher percent college graduates, percent disabled, occupied hospital bed density, and unoccupied hospital bed density. There was a nonlinear relationship between surgery rates and percent employed in heavy lifting/transportation industries. Medical rates increased with higher other-cause discharge rate and with MRI/CT scanner availability, and decreased with higher percent urban population, percent nonwhite population, percent in heavy lifting/transportation industries, and unoccupied hospital bed density. The results show that population-level socioeconomic and health resource characteristics are important determinants of variation in low back hospitalization rates. Independent of these variables, a separate spatial process produces geographic clustering of high-rate counties. Spatial effects are important and should be considered in small area analyses.