Modelling the Age-period-cohort Trend Surface

Abstract

Age-period-cohort analysis of incidence and/or mortality data has received much attention in the literature. Though several authors seem to offer solutions to the non-identifiability problem inherent in this kind of analysis, the constraints imposed lack sound biological bases and the interpretability of the parameters is in doubt. By introducing the axes of 'adjusted' age, period, and cohort variables in the age-period-cohort trend surface, the authors find that separate effects from these variables can be estimated and can be clearly interpreted. The constraint behind the procedure is also found to be mathematically simple and elegant. Since the method still lacks a biological foundation, it is best viewed as an adjunct to the graphical trend-surface analysis rather than as a solution to the non-identifiability problem. Previously published mortality data of prostate cancer is used to illustrate the methodology.