Evaluating the performance of the Framingham risk equations in a population with diabetes

AIMS: The Framingham risk equations are widely used to estimate risk of coronary heart disease (CHD). The purpose of this study was to evaluate the reliability of these equations in predicting CHD risk in people with diabetes and the reliability of using imputed mean HDL-cholesterol values. METHODS: Data describing the baseline characteristics of recognized CHD risk factors for 938 people aged 30-74 years were extracted from the Cardiff Diabetes Database. Data describing CHD events were available for up to 4 years following the baseline year (1996). Several mathematical techniques were used to assess the reliability of predictions provided by the Framingham equations in this population. RESULTS: Thirty-four percent of males and 25% of females who experienced CHD events had a predicted 10-year CHD risk >/= 30%. Seventy-five percent of males and 58% of females had a predicted 10-year CHD risk >/= 20%. Using imputed HDL-cholesterol values, 26% of males and 6% of females who later developed CHD events had a 10-year CHD risk >/= 30%. Using imputed HDL-cholesterol values, the CHD risk predicted by the Framingham equations consistently underestimated the actual risk of CHD events. However, refitting the Framingham risk equations to the Cardiff data resulted in only marginal improvements in discriminatory capabilities. CONCLUSIONS: The Framingham risk equations can be unreliable when applied to the diabetic population, tending to underestimate an individual's probability of progressing to CHD; the equations perform marginally better in women than in men. The use of imputed mean HDL-cholesterol values improved the reliability of the estimates of risk.