Purpose: Currently, statistical (UKPDS and Framingham) and simulation (Archimedes) models are available to predict the risk of complications in type 2 diabetes patients. However, little is known about the comparative performance of these alternative models in the type 2 diabetes population. The goal of this study is to compare the 5-year and 10-year cardiovascular risk predictions from the alternative models with observed outcomes for patients with type 2 diabetes.

Methods: Detailed data from a diabetes-specific electronic management system, clinical electronic medical record, and administrative data sets were used to calculate Framingham and UKPDS risk equations for coronary heart disease (CHD) and stroke for 635 primary care patients with type 2 diabetes (1997-2006). In addition all patients were entered through the online application for the Archimedes model to estimate 5- and 10-year probability of fatal and non-fatal CHD and stroke. Model calibration was assessed using Hosmer-Lemeshow goodness-of-fit and model discrimination was assessed using the c-statistic. Predicted rates of CHD and stroke by each of models were compared to observed rates by gender and age-groups.

Results: The average age of the population was 49.7 years and 60.7% were males. The mean baseline 10-year CHD risks were 16.3%, 12.8%, and 16.8% for the UKPDS, Archimedes, and Framingham models, respectively. We observed 116 CHD and 85 stroke events during the 10-year follow-up and 61 CHD and 35 stroke events during the 5-year follow-up. All three models showed significant lack of calibration (p<0.001) and discrimination (c-statistic=0.68 UKPDS, 0.65 Framingham and 0.56 Archimedes) for the 10-year CHD risk. In addition, the models lacked calibration (p<0.001) and had poor discrimination (c-statistic=0.69 UKPDS, 0.67 Framingham and 0.66 Archimedes) for 10-year stroke risk. All 3 risk prediction methods generally under-predicted the risk of CHD and stroke. However, UKPDS over-estimated the probability of CHD and stroke events for the higher risk deciles and in the elderly (over age 65).

Conclusion: Currently available risk models for CHD and stroke poorly predict events for patients with type 2 diabetes. This limits the applicability of these models for clinical decision making. Future research needs to identify other measures that may improve the model performance.