A Bayesian age-period-cohort analysis and projection of diabetes incidence in Hong Kong from 2007 to 2027

Abstract

Background: Our study primarily aimed to examine the effects of age (biological changes), time period (current populationwide factors), and birth cohort (generational factors) on incidence rate of diabetes and generate projections of trends in Hong Kong. Population: We identified all individuals who had a recorded diagnosis for diabetes from the electronic medical record database of Hospital Authority in Hong Kong from 1 January 2007 to 31 December 2017. Setting: We included individuals aged from 20 years old (14 age groups at 5-year intervals) with birth cohorts from 1922 to 1997. Funding Source: The authors received no financial support for the research, authorship, and/or publication of this article. Innovation/Methods: We applied a Bayesian age-period-cohort model which decomposed the age, period, and cohort effects on diabetes incidence trends separately for males and females. Based on past data from 2007 to 2017, we projected future trends for the next decade to 2027. Results: We identified 469,325 incident cases of diabetes from 2007 to 2017. As expected, we observed clear age effects on incidence in both sexes. Overall incidence rates increased among successive birth cohorts from 1980s to 1990s while period effects remained negligible. In the projection model for male, incidence rate increased with older age groups in a progressive manner. We noticed an increasing trend projection for the 30-39 age group and a decreasing trend projection for the 70-79 age group; while other age groups remained stable. For female, incidence rate also progressively increased with older age groups in a pattern similar as male. 20-29 age group had a stable projected trend. The next 30-39 and 40-49 age groups showed a small rising trend. While for the remaining age groups with patients above 50 years old, they all revealed declining trends in the projected years. Limitations: Age-period-cohort analyses described trends so we could only speculate the aetiologies but not the causal factors that lead to the incidence trends observed. With limited data covering from 2007 to 2017, our projection of future trends had large confidence intervals in incidence projection. Discussion: Age had the greatest impact on incidence rates among people with diabetes, while period had negligible impact. The increase in cohort effects from 1980s in both sexes could be due to the increasing prevalence of obesity for the millennials in Hong Kong. We could observe that the age of onset of diabetes had a decreasing projected trend. This could represent an actual earlier onset of diabetes at younger age or due to earlier detection of diabetes.