Bayesian hierarchical modeling of colorectal and breast cancer data in Missouri

Abstract

Data on cancer in the United States is collected through cancer registries. The Missouri Cancer Registry and Research Center (MCR-ARC) maintains a statewide cancer surveillance system and participate in research in support of the prevention of cancer and the reduction of the cancer burden among Missouri residents. We applied Bayesian hierarchical models to colorectal cancer (CRC) and breast cancer related data collected by the MCR-ARC. In the first project, CRC incidence and mortality rates in Missouri were studied with emphasis on different groups of people categorized by age, gender and county at diagnosis. The incidence and mortality data were aggregated into different spatial regions due to data confidential requirements, which was identified as a misaligned-region problem in multivariate disease mapping literature. The Marginally and Conditionally CAR models were built to address the problem. Later on, colorectal cancer screening (CRCS) prevalences were analyzed due to its importance to the early detection of CRC. We applied small area estimation techniques to produce county-level CRCS prevalences from the state-level Behavioral Risk Factor Surveillance System (BRFSS) data. The last two projects focused on breast cancer related data. One is about breast cancer survival analysis in Missouri with emphasis on detecting the spatial variation of survival time among counties in Missouri, after accounting for the differences in demographic and cancer stages. The other one is studying the disparities of breast cancer treatment delay with respect to patient's race, age, stage of cancer, county at diagnosis and year of diagnosis.