Abstract 10030: Validation of Framingham and Reynolds Cardiovascular Risk Prediction Models in the Women's Health Initiative
Models predicting risk of total cardiovascular events in women have not been compared in external validation data. We examined the fit of the ATP-III models for CHD, Framingham risk score for CVD, and the Reynolds risk score for CVD in a diverse cohort of American women.
METHODS We measured traditional risk factors and novel biomarkers of CVD in a case-cohort sample of the Women's Health Initiative Observational Cohort, comprised of 1722 cases of major CVD (752 MIs, 754 ischemic strokes, and 216 other CVD deaths) and a random subcohort sample of 1994 women, all without prior CVD. We estimated the predicted risk according to the published ATP-III CHD model, the Framingham CVD model, and the Reynolds CVD risk score. Models were recalibrated to allow for differing outcome definitions. Model fit was assessed through calibration plots and the Hosmer-Lemeshow test, the c-statistic, and reclassification measures. Estimates were reweighted to reflect overall cohort frequencies. Fit was examined both including and excluding those with diabetes.
RESULTS Calibration appeared adequate for the Reynolds risk score, but the ATP-III model over-estimated risk for CHD endpoints and the Framingham CVD model over-estimated risk for major CVD. After recalibration, the Reynolds risk score demonstrated moderately improved fit over the ATP-III model through a higher c-statistic (0.766 vs. 0.758, p=0.04), positive NRI (4.7%, p=0.02) and positive IDI (2.8%, p<0.0001) both overall and after excluding diabetics (NRI=3.3%, p=0.04; IDI=1.3%, p<0.0001). The Framingham CVD model demonstrated worse fit for major CVD than either the Reynolds risk score (NRI=-14.9, p<0.0001; IDI=-3.5%, p<0.0001) or ATP-III model (NRI=-8.3%, p<0.0001; IDI=-0.7, p<0.0001). Results were similar among Caucasian and African-American women separately.
CONCLUSION The Reynolds CVD risk score was better calibrated for major CVD than the ATP-III CHD model and the Framingham CVD model in this external validation cohort. After recalibration, the Reynolds risk score showed modest improvement in fit over the other two models.
- © 2011 by American Heart Association, Inc.