This exploratory analysis reveals that relationships between lung cancer mortality and various environmental and socioeconomic factors are not constant across space. The results of the GWR indicate that, aside from smoking, there are other explanatory variables significantly associated with lung cancer mortality. The proportion of rented homes, proportion of non-white population, median income and radon risk were each significantly associated with county-level lung cancer mortality rates, relationships that were spatially nonstationary across the eastern United States.
The variance in parameter coefficients was a factor of its geographical context. In other words, proximal counties tended to have similar coefficient values, in contrast to those further away. This is perhaps the result of political influence of neighbouring counties, and the fact that environmental factors are not limited by political boundaries. County radon risk had the greatest effect on lung cancer mortality in Florida, where one in five homes tested has elevated indoor radon levels above the EPA’s target of 4 pCi/L (Florida Department of Health, n.d.). The median household income of a county had the greatest effect in the central Appalachian region, northern Florida and southern Illinois. As revealed by the hot spot analysis, the Appalachian region and northern Florida have statistically significant clusters of low income counties. This aligns with the findings of several other studies that suggest poverty increases the likelihood of lung cancer incidence and mortality (Halverson & Bischak, 2008; Jamal et al., 2018; Lengerich et al., 2005; Sidorchuk et al., 2009). The proportion of a county population classified as a racial or ethnic minority (“non-white”) had greatest influence on lung cancer mortality in Florida, Mississippi, Alabama and Tennessee. Counties in Mississippi and Alabama have a high proportion of African Americans, whereas the vast majority of the population in Florida and Tennessee counties are white. This indicates that race is differentially associated with lung cancer mortality rates across the eastern United States. The proportion of rented households was most associated with counties in Florida, Wisconsin, Illinois and Mississippi where the potential exposure to secondhand smoke may have been greater (CDC, 2015).
The Appalachian region (which in this analysis included Tennessee, Kentucky and West Virginia) exhibited statistically significant clustering of counties with high lung cancer mortality rates. As revealed by the grouping analysis, the Appalachian region can be largely characterized as having very high smoking rates, high radon risk, with percentage of renters, proportion of non-whites and median incomes near the global average of the study area. It should be noted, however, that this region exhibited statistically significant clusters of low median household income in our initial hot spot analysis.
Similar statistically significant clustering of counties with high lung cancer mortality rates were observed in Florida. This region was particularly interesting because, despite being a statistically significant cluster of high lung cancer mortality, it was not a hot spot for smoking rates. As previously mentioned, it exhibited relatively strong associations with radon risk, race, and rental housing. This suggests that smoking may influence Florida’s lung cancer mortality rate to a lesser extent relative to the rest of the eastern USA, and that these socioeconomic and environmental factors must be strongly considered in state-wide and local policy and planning.
The variation in parameter estimates from the GWR suggests the need to apply a geographical approach to other lung cancer studies that have been previously conducted with global (linear) spatial models. In OLS Model 1, 48% of county-level lung cancer mortality was explained by race, radon risk, income and rental housing. In OLS Model 2, 54% of county-level lung cancer mortality was explained by smoking prevalence, race, educational attainment and income. However, at an individual county level, the explanatory percentage (local R-squared value) ranged from 19% to 72%. Our results suggest that local contexts, policies and programs, attributes of the built and natural environment, and occupational practices are associated with lung cancer mortality and that their extent varies across the eastern USA. For instance, the Central Appalachian region is highly involved in the coal industry. This likely influences both economic status (i.e. income), as well as occupational and environmental exposure to atmospheric particulate matter. This context is different from the northeastern USA, where counties are typically more urbanized and less involved in resource extraction. There is a significant association between lung cancer mortality and non-white populations, income, rental housing and radon risk. Moreover, because these relationships exhibit non-stationarity, the need for localized context-specific action is key to prevent lung cancer and subsequent mortality.