JAMA Netw Open. 2023 Sep 5;6(9):e2333618. doi: 10.1001/jamanetworkopen.2023.33618.
ABSTRACT
IMPORTANCE: Breast cancer mortality is complex and traditional approaches that seek to identify determinants of mortality assume that their effects on mortality are stationary across geographic space and scales.
OBJECTIVE: To identify geographic variation in the associations of population demographics, environmental, lifestyle, and health care access with breast cancer mortality at the US county-level.
DESIGN, SETTING, AND PARTICIPANTS: This geospatial cross-sectional study used data from the Surveillance, Epidemiology, and End Results (SEER) database on adult female patients with breast cancer. Statistical and spatial analysis was completed using adjusted mortality rates from 2015 to 2019 for 2176 counties in the US. Data were analyzed July 2022.
EXPOSURES: County-level population demographics, environmental, lifestyle, and health care access variables were obtained from open data sources.
MAIN OUTCOMES AND MEASURES: Model coefficients describing the association between 18 variables and age-adjusted breast cancer mortality rate. Compared with a multivariable linear regression (OLS), multiscale geographically weighted regression (MGWR) relaxed the assumption of spatial stationarity and allowed for the magnitude, direction, and significance of coefficients to change across geographic space.
RESULTS: Both OLS and MGWR models agreed that county-level age-adjusted breast cancer mortality rates were significantly positively associated with obesity (OLS: β, 1.21; 95% CI, 0.88 to 1.54; mean [SD] MGWR: β, 0.72 [0.02]) and negatively associated with proportion of adults screened via mammograms (OLS: β, -1.27; 95% CI, -1.70 to -0.84; mean [SD] MGWR: β, -1.07 [0.16]). Furthermore, the MGWR model revealed that these 2 determinants were associated with a stationary effect on mortality across the US. However, the MGWR model provided important insights on other county-level factors differentially associated with breast cancer mortality across the US. Both models agreed that smoking (OLS: β, -0.65; 95% CI, -0.98 to -0.32; mean [SD] MGWR: β, -0.75 [0.92]), food environment index (OLS: β, -1.35; 95% CI, -1.72 to -0.98; mean [SD] MGWR: β, -1.69 [0.70]), exercise opportunities (OLS: β, -0.56; 95% CI, -0.91 to -0.21; mean [SD] MGWR: β, -0.59 [0.81]), racial segregation (OLS: β, -0.60; 95% CI, -0.89 to -0.31; mean [SD] MGWR: β, -0.47 [0.41]), mental health care physician ratio (OLS: β, -0.93; 95% CI, -1.44 to -0.42; mean [SD] MGWR: β, -0.48 [0.92]), and primary care physician ratio (OLS: β, -1.46; 95% CI, -2.13 to -0.79; mean [SD] MGWR: β, -1.06 [0.57]) were negatively associated with breast cancer mortality, and that light pollution was positively associated (OLS: β, 0.48; 95% CI, 0.24 to 0.72; mean [SD] MGWR: β, 0.27 [0.04]). But in the MGWR model, the magnitude of effect sizes and significance varied across geographical regions. Inversely, the OLS model found that disability was not a significant variable for breast cancer mortality, yet the MGWR model found that it was significantly positively associated in some geographical locations.
CONCLUSIONS AND RELEVANCE: This cross-sectional study found that not all social determinants associated with breast cancer mortality are spatially stationary and provides spatially explicit insights for public health practitioners to guide geographically targeted interventions.
PMID:37707814 | DOI:10.1001/jamanetworkopen.2023.33618
