Is The Age-Period-Cohort Model Well Suited to an Epidemic Context? The Case of the French BSE Epidemic

Abstract

Although Age-Period-Cohort (APC) models are routinely used in time-trend analyses of chronic diseases, few examples of their application to epidemic diseases are available. APC analyses of the French bovine spongiform encephalopathy (BSE) epidemic revealed an unexpected period effect, which was attributed to the design of the study, in connection with low BSE prevalence and a short surveillance period. The aim of our study was to evaluate the relevance of this hypothesis, the behaviour of the APC model in an epidemic context (e.g. the BSE epidemic) and the impact of including the period effect in an APC model on the estimate of birth cohort effects. We simulated datasets mimicking the French BSE epidemic and its variable pattern, as well as duration and the surveillance time period, and analysed them with a categorical APC model. Results showed a period effect in 44% of analysed datasets, although no period effect had been introduced in the data simulation process. This type of artefactual period effect was detected when a sudden change in cohort prevalence occurred over a short period of time. Additionally, in the context of BSE, including a period effect in the model may dramatically affect the estimation of cohort prevalence, depending on epidemic pattern and, as expected, duration and the surveillance time period according to the year in which highly infected birth cohorts are detected as BSE cases.