Does rurality and region play an important role in cancer incidence in the U.S.?

Background: According to the CDC, the top four cancers in the U.S. by incidence rate are breast (BC), lung (LC), prostate (PC), and colorectal (CRC).  Some studies have shown disproportionate cancer incidence burden in rural areas (with the exception of BC which trends higher in urban areas), but the data are inconsistent. This may be attributable in part to the use of ‘rural’ as a generic term implying homogeneity of risk/protective factors across wide geographic spans such as all Surveillance, Epidemiology and End Results (SEER) registries. We sought to determine if cancer incidence significantly varied by region or rurality after adjustment for known risk factors. 

Methods: Counties in SEER 18 registries were classified by their Rural-Urban Continuum Code (RUCC; ranging from 1-9) and aggregated into urban (RUCC-u; 1-3), adjacent rural (RUCC-a; 4, 6, 8), and non-adjacent rural (RUCC-na; 5, 7, 9). The registries were aggregated into 3 Regions: North (IA, MI, CT, NJ), South (GA, LA, KY), and West (NM, UT, CA, WA, HI). Two-way ANCOVA was performed with Region and RUCC as factors with adjustment for rates of obesity, smoking, alcohol abuse, and inactivity obtained from the County Health Rankings (2013). Incidence data were from the years 2000-2011.

Results: RUCC has a significant effect on incidence rate on BC (114.9 RUCC-u, 108.4 RUCC-a, 108.8 RUCC-na, p=0.001) and PC (143.3 RUCC-u, 139.6 RUCC-a , 132.9 RUCC-na, p=0.009). CRC significantly varies by region (p<0.0001), and the effect of rurality significantly varies across Regions (50.1 North,  45.6 South, 40.6 West, p=0.0005). Both the South and West show stable CRC incidence rates across RUCC, but the North CRC incidence increases as rurality increases.  North RUCC-na is significantly higher than any other stratification (p<0.01). LC rates significantly vary across both Region and RUCC (p<0.0001 and p=0.0001, respectively), and the effect of rurality significantly varies across Region (p<0.0001). South LC incidence slightly increases with rurality, whereas North and West decrease with rurality.

Conclusion: The analysis shows that risk-adjusted cancer incidence varies significantly across Regions, which is not unexpected from the literature. However, we also found that rural cancer incidence significantly varied across otherwise-similar rural areas. The implication is that ‘rural’ is not a homogeneous classification, but likely characterized by regionally-relevant factors which must be included in data analysis and intervention development.