Prediction of age-specific cancer mortality by using multinomial time series model

Abstract: Cancer is the largest cause of death in Korea, and its proportion is increasing. Meanwhile, the cancer mortality rates vary over time as well as age. With the increased life expectancy in Korea, the proportion of the elderly age among cancer deaths has increased over time, while that of the young age has decreased. To reflect the proportions of the categories with such dynamic structures of age and time, a multinomial time series model can be used as a prediction model. However, there is a difficulty in estimating the parameters through the Markov Chain Monte Carlo (MCMC) method when some cell counts are very small relative to others, such as the number of deaths from cancer of young age group. In order to predict the age-specific cancer mortality by reflecting its dynamic structure and by overcoming estimation problems in MCMC, a power transformation is adopted as a link function of multinomial time series model instead of a logit link function, and forecasts the age-specific cancer mortality of male in Korea by 2040 using the proposed method.

Keywords: multinomial time series model, MCMC, link function, power transformation.