Comparative accuracy of cardiovascular risk prediction methods in patients with diabetes mellitus

OBJECTIVE: To compare the accuracy of cardiovascular risk prediction methods based on equations derived from the Framingham Heart Study in a cohort of patients with diabetes mellitus. RESEARCH DESIGN AND METHODS: Risk factor data was collected prospectively from 906 patients with diabetes mellitus. Absolute cardiovascular risks were calculated using the Framingham equation, and estimated with the currently available Framingham-based risk tables and charts. The sensitivity, specificity, positive and negative predictive values of the tables and charts to assess cardiovascular risk were assessed using calculation of risk from the full Framingham equation as the reference method. RESULTS: In all, 146 subjects (16.1%) had calculated 10-year coronary heart disease (CHD) risks > or = 30%, and 585 (64.6%) had risks > or = 15%. For identification of those at 10-year CHD risk > or = 30%, the original Sheffield tables had a sensitivity of 43% (95% confidence intervals (CI) 19.9-61.7%) and specificity of 94% (CI 90.8-96.7%). Modifications of the Sheffield tables improve sensitivity (95% CI 93.9-97%) but reduce specificity (90% CI 85.6-95.7%). The Joint British Guidelines' charts have a moderate sensitivity (69.5% CI 51.8-81.9%) and high specificity (99.7% CI 98.9-100%). For identification of individuals at a 10-year CHD risk > or = 27%, the Framingham categorical tables had a sensitivity of 95% (CI 91.6-97.8%), but a specificity of only 83% (95% CI 79.1-85.5%). CONCLUSIONS: The Joint British charts appear to have the best performance in a cohort of patients with diabetes mellitus, however, calculation of CHD/CVD (cardiovascular disease) risks with personal or laboratory computers using the full Framingham equation remains the most accurate way to assess cardiovascular risk in a primary prevention setting.