Aims: Diet and physical activity interventions can prevent diabetes in those at high risk due to impaired glucose tolerance. We determined whether simple measures of physical activity and diet predicted incident diabetes and enhanced prediction by known risk factors including age, body mass index and family history.
Methods: This was a population-based prospective cohort study (EPIC-Norfolk). Participants aged 40-79 years (n = 25,633) attended a health check between 1993 and 1998 and completed diet and activity questionnaires. We assessed the association between simple behavioural indices of physical activity and diet derived from the questionnaires as well as known risk score variables with incident diabetes at follow-up (mean 4.6 years). We developed a new diabetes risk score incorporating simple behavioural indices in a randomly selected half of the EPIC dataset using forward step-wise multivariate logistic regression, and tested this score in the remaining half. We compared existing and new scores using receiver-operating characteristic (ROC) curves.
Results: There were 417 incident cases of diabetes during 115,137 years of follow-up. A simple physical activity index independently predicted risk of diabetes. Eating one or more daily portion of vegetables, fresh fruit and wholemeal bread was associated with reduced risk; whilst eating meat products was associated with increased risk. The area under the ROC curves for the new and original score was the same (76.3%).
Conclusions: Simple indices of diet and activity are feasible to collect, predict future diabetes risk and might enhance routine data collection in primary care. However, they do not improve the prediction of risk scores based on known risk factors.