Cancer Incidence Study: The Application of Bayesian Multi-stage Carcinogenesis Model

In the past fifty years, the multi-stage carcinogenesis model proposed by Armitage and Doll has been widely applied to the cancer incidence studies. It estimates different numbers of stages for various types of cancer. In this work, we build a Bayesian Armitage-Doll multistage model to fit the Surveillance, Epidemiology, and End Results (SEER) 1973-2006 incidence data. Based on preliminary studies and literature reviews, knowledge of number of stages is translated into the prior distribution. The choice of prior distribution is made with respect to the direction of inference, especially for the derivation of the marginal posterior. Non-informative and informative hyperprior are considered in the model. Metropolis-Hasting algorithm is used in the Monte Carlo Markov Chain (MCMC) sampling method. Bayesian inference including the point estimates, confidence interval and hypothesis testing is obtained. The robustness of Bayesian model is assessed through various model fitting methods. The result of Bayesian model shows that the posterior estimates derived from the combined information (prior and likelihood) result in greater precision as compared to the classical estimators. Bayesian model can be a powerful tool that complements classical multi-stage model to improve the accuracy of cancer incidence estimation, which is important in setting priorities for cancer education, prevention, and control.