A Bayesian semi-parametric model for colorectal cancer incidences

Area of Science:

• Biostatistics
• Epidemiology
• Statistical modeling

Background:

• Colorectal cancer incidence is influenced by demographic, spatial, and temporal factors.
• Understanding these interactions is crucial for effective public health strategies.
• Existing models may not fully capture the complex interplay of these effects.

Purpose of the Study:

• To propose a Bayesian semi-parametric model to simultaneously analyze demographic, spatial, and temporal effects on colorectal cancer incidence.
• To develop an extension of multivariate conditionally autoregressive (CAR) processes for a spatial-temporal setting.
• To investigate the interaction between age, gender, county, and time in colorectal cancer patterns.

Main Methods:

• A novel Gaussian demographic spatial temporal CAR (DSTCAR) process was developed.
• The model utilizes a Kronecker product for the precision matrix, incorporating CAR priors for spatial effects.
• A pth order intrinsic autoregressive (IAR(p)) prior was used for non-parametric temporal trends, and a Wishart prior for demographic effects.

Main Results:

• Significant spatial correlation in colorectal cancer incidence was observed specifically in the 50-59 age group.
• Strong correlations were found between males and females in their 50s and 60s.
• Bayes factor testing indicated that gender correlation is a significant factor that should not be ignored.

Conclusions:

• The proposed DSTCAR model effectively captures the complex interactions between demographic, spatial, and temporal factors in colorectal cancer incidence.
• Gender plays a significant role in colorectal cancer correlations, particularly for individuals in their 50s and 60s.
• The model provides valuable insights for targeted interventions and future research in colorectal cancer epidemiology.