Researchers found differences in the variable importance of cancer mortality risk factors depending on an individual’s geographical region.
Cancer mortality and risk factors may vary by geography, according to researchers who published their study findings in JAMA Network Open. However, high importance risk factors were not consistently prevalent in certain geographical areas.
“Our study further affirms that socioeconomic status variables, such as income, education, and female-headed households, are associated with cancer mortality as suggested in previous studies,” wrote the authors of the report.
Cancer mortality and risk factors are also disproportionality affected by certain geographic area. To understand the relationship between these factors, geographically weighted regression (GWR) models are used, although the models are not able to fully evaluate nonlinear relationships.
However, the conventional random forest (RF) algorithm can evaluate these nonlinear relationships. The newly developed geographical random forest (GRF) model can further look at spatially varying associations between cancer mortality and risk factors and nonlinear associations.
During the cross-sectional study, investigators aimed to use conventional RF to find national and regional scaled with the local and county level scaled GRF to identify the variable importance (VI) measure of risk factors associated with cancer mortality among different US counties. Taking data from the National Center for Health Statistics, investigators identified 7,179,201 people who died of cancer between 2008 and 2019 using the conventional RF model to assess cancer mortality risk factors associated with populations on the regional and national scale. Then, using the GRF model, the team identified cancer mortality risk factors at the county level.
Based on results from the GRF model, the top 6 risk factors associated with cancer mortality included household income, Supplemental Nutrition Assistance Program (SNAP) benefits, smoking, high school degree, female-headed households, and adult obesity. The VI of the latter 5 risk factors were associated with the greatest cancer mortality in the South, while most counties that identified adult obesity as having high or medium-high VI include those in Colorado, Florida, the Northeast, and Mississippi.
“This work suggests that risk factor importance may be a preferable paradigm for selecting cancer control interventions compared with risk factor prevalence,” wrote the authors of the report.
The team further observed that comorbidities may worsen a risk factor; because of this, the researchers explained that further studies should be done on prevalence versus severity of a risk factor in a certain geography.
However, researchers did not account for cancer type, stage at diagnosis in cancer mortality, or cancer incidence, which they noted do limit the results of the study. Additionally, the study was further limited because the team only looked at large counties with diverse population characteristics. GRF may have been biased against coastal areas as well, due to edge effects. Finally, the study did not analyze the cause of cancer mortality.
“Practitioners and policy makers should consider tailored interventions in reducing cancer mortality based on not only the prevalence but also the importance of the place-specific risk factors,” wrote the authors in the report.
Dong, Weichuan, Bensken, Wyatt, Kim, Uriel, et al. Variation in and Factors Associated With US County-Level Cancer Mortality, 2008-2019. September 9, 2022. JAMA Netw Open. 2022;5(9):e2230925. doi:10.1001/jamanetworkopen.2022.30925