018 Forecasting diabetes prevalence using a simple model: England and Wales 1993–2006
Background Current projections of diabetes prevalence are mostly based on demographic change. Explicitly including trends in obesity and other risk factors could improve the accuracy of the projections and assist in evaluating policy options for prevention.
Methods The model integrates population, obesity and smoking trends to estimate future diabetes prevalence. From three starting states (healthy, obese and smokers) the number of people with diabetes and deaths by diabetes status are estimated using a Markov approach. The transition probabilities and RR associated with risk factors were obtained from the literature, except for diabetes incidence that was estimated using DISMOD. For validation purposes, we developed a model for the England and Wales population (1993–2006), and compared model outputs with diabetes prevalence reported by the Health Survey for England (HSE) and the English Longitudinal Study of Ageing (ELSA).
Results The prevalence of diabetes mellitus in England and Wales in 1993 was 3% in men and 2% in women (HSE; adjusted for self reporting, 3.9% and 2.6% respectively) and increased to 6% and 4% (7.3% and 5.5%, adjusted) by 2006. Obesity prevalence almost doubled and smoking trends showed a more complex pattern. Comparisons with the HSE showed almost parallel trends, over a period of 13 years. Prevalence as estimated from the model was 7.3% for men and 5.7% for women for 2006 and 8.9% and 7.2% for 2012. The model tends to slightly overestimate prevalence but accuracy improved in later years. The estimated prevalence compared well with that reported in ELSA (Men: model: 9.9%, ELSA: 11.6%; women: 8.3% and 6.8%).
Conclusions The model provide a reasonably close estimate of diabetes prevalence for England over the 1993–2006 period, compared with contemporary independent prevalence surveys in the same population. Although the model seems to slightly overestimate prevalence, the observed and modelled trends are almost parallel. Further testing and validation in a range of populations would be desirable but the model appears to provide reasonably accurate estimates of diabetes prevalence that could be used by policymakers.