Statistical inference for estimating the incidence of cancer at the prefectural level in Japan

Abstract

Information of prefectural cancer incidence in Japan is an important resource to plan an effective regional cancer control program. In order to obtain accurate prefectural cancer incidence, we have to adjust for both of completeness of regional cancer registries and regional dispersion, simultaneously, Then, we developed a hierarchal Bayes model with a completeness adjustment by using the relation between MI ratio (mortality divided by incidence) and proportion of death certificate notified among incidence. Prefectural cancer incidence is estimated with the 95% credible interval, which suggests that the national MI ratio should be 0.386 if the registry were complete. Because cancer behavior has a variation depending on prefecture, such accurate incidence must contribute to plan and assess regional cancer control program.

Keywords: Bayesian analysis; population registry; statistics and numerical data.