Urologist Density and County-Level Prostate Cancer Mortality: A Geospatial Analysis

Background: Previous cancer mortality studies have investigated the role physician density on mortality rate in specific types of cancer. Ordinary linear regression techniques assume that the relationship between independent variables and outcome variables is constant or stationary across the study region. This study examines the relationship between urologist density and prostate cancer mortality using geographically weighted regressions (GWR), which allows regression estimates to vary across the study region. Effect modification by geographic subregion would indicate that the relationship between predictor and outcome variables is not universal.

Methods: We used county-level prostate cancer mortality, incidence, and other covariates for the years 2006-2010. Mortality data came from the National Center for Health Statistics. Incidence data came from the State Cancer Profiles website of the National Cancer Institute. Covariate data was obtained from the Area Resource File. 1751 counties in the contiguous states were included in the study. Kansas, Minnesota, Ohio,  Virginia, and Washington were not included because of missing data of cancer incidence rates. 910 additional counties with too few cases of incidence or mortality were excluded. We used ordinary linear regression methods to compare against the results obtained using GWR. We examined whether collinearity problems exist in the GWR model using variance-decomposition proportions and condition indexes. We also examined the global spatial autocorrelation of variables and residuals using Moran’s I. We performed sensitivity analysis including only counties from north-eastern states with few holes of missing data in the map.

Results: The GWR model did not have collinearity problems based on diagnostic information. The ordinary linear regression found an overall negative association (-0.20, p = 0.018) between urologist density (per 100,000 people) and prostate cancer mortality (per 100,000 people), which is consistent with previous studies. The effect size varied from region to region, with the largest (most negative) coefficients reported in the Arkansas, Louisiana, Mississippi, West Virginia, and Appalachian Pennsylvania. Coefficients closest to zero were reported in the New England states and West Coast.  The sensitivity analysis also revealed geographically varying effects of urologist density on county-level prostate cancer mortality in north-eastern states.

Conclusions: Urologist density is more strongly associated with disparity in prostate cancer mortality rates in the southern states and northern Appalachian counties than in New England and West Coast states. The presence of geographically varying predictors of prostate cancer mortality suggests that a regional approach to reducing cancer disparities could be necessary. Targeted increases in the supply of urologists to areas with the highest coefficients could have more impact on prostate cancer mortality than with untargeted placement.