PREVALENCE AND INCIDENCE OF COLORECTAL CANCER IN RURAL SASKATCHEWAN: AN APPLICATION OF GENERALIZED ESTIMATING EQUATIONS (GEE) AND SURVIVAL ANALYSIS
Although agricultural activity is recognized to be associated with colorectal cancer (CRC) prevalence and incidence, little is known regarding the true prevalence and incidence or CRC risk factors related to farming. This study determines the prevalence of CRC both at baseline and follow-up among farm and non-farm rural residents, estimates the incidence of CRC, and further identifies risk factors for CRC incidence and prevalence. Data from the Saskatchewan Rural Health Study (SRHS) was collected in 2009 and a four-year follow-up in 2014 through completed questionnaires from 8,261 individuals (level 1) nested within 4,624 households (level 2) at baseline and 4,867individuals within 2,797 households at follow-up. A modified version of Dillman’s methods for mail and telephone surveys was used to maximize response rates both at baseline (42%) and follow-up (63%). The study sample consist of 5,599 individuals at baseline and 3,933 at follow who were 50 years or older. Hierarchy in data was accounted for using the generalized estimating equations (GEE) approach using the exchangeable covariance structure and the within-cluster correlations due to the longitudinal design were also accounted for using the PROC GENMOD robust variance estimation. Multilevel marginal logistic regression models based on GEE were fitted to determine risk factors for CRC. To determine risk factors for the incidence of CRC, the Cox proportional hazards (PH) regression model was used. The prevalence of CRC decreased over time among rural farm residents (baseline: 3.1%; follow-up: 1.3%, p<0.05), however increased among rural non-farm residents (baseline: 1.4%; follow-up: 2.0%, p>0.05). Individuals who spent their first year of life in a farm had higher risk of developing CRC that their counterparts who did not OR = 1.64, p<0.05). A similar relationship was observed for individuals who ever lived on a farm at one point in life times (OR = 1.29, p<0.05). Occupational exposure to grain dust and radiation were significant (p<0.05) determinants of longitudinal changes in CRC prevalence after controlling for important confounders. The crude incidence rate of 1.98 per 1,000 person-years (i.e. 27/13,632 total time under observation and at risk), resulting in a cumulative incidence of 0.8% during the study period (2010 - 2014). The results show that quadrant, use of natural gas, living on a farm in one’s first year of life, exposure to oil/gas well fumes and radiation, increasing age, increasing BMI, and female gender predict CRC incidence. We conclude that prevalence, longitudinal changes in the CRC prevalence among farming and non-farming rural residents as well as the incidence of CRC appear to depend on a complex combination of individual and contextual factors.
Colorectal cancer, Prevalence, Incidence, Rural, Saskatchewan, Longitudinal changes, Generalized Estimating Equations, Survival analysis
Master of Science (M.Sc.)
School of Public Health