Abstract 15643: Validating, Updating, and Extending Long-Term Framingham Cardiovascular Disease Predictions: The Rotterdam Study

Abstract

Introduction: Risk scores that predict long-term risk are increasingly advocated to support cardiovascular disease (CVD) prevention. In this study, we aim to validate, update, and extend the Framingham prediction of long-term CVD risk by simultaneously analyzing coronary heart disease (CHD) and Stroke.

Method: Participants from the Rotterdam Study free of CHD and Stroke (N=6,004), were used to assess calibration and discrimination of the Framingham function that predicts 15-year cumulative incidence of CVD based on age, sex, systolic blood pressure, anti-hypertensive treatment, smoking, diabetes, total cholesterol, and HDL-cholesterol. We recalibrated and refitted the original model that predicts CVD as a combined endpoint of first non-fatal or fatal CHD or Stroke event. We subsequently estimated the cumulative incidences of CHD and stroke separately, and used the sum as an estimate for the cumulative incidence of total CVD. Calibration plots and c-statistics were used to evaluate the performance of the models.

Results: During 15 years of follow-up, 1,348 first CVD events occurred, of which 736 were CHD and 612 were Stroke. Calibration of the original Framingham model was found to be good in the low- to intermediate risk (15-yr CVD risk <=30%) categories (17.5% observed risk versus 16.6% expected risk) but poor in the higher risk (15-yr CVD risk >30%) categories (36.3% observed risk versus 44.1% expected risk). The c-statistic of the original model was 0.66 and increased to 0.69 after refitting. Using the sum of CHD and Stroke as an estimate for CVD revealed considerable heterogeneity with regard to the contribution of CHD and Stroke to the total cumulative incidence of CVD. This was visualized with an accompanying risk prediction tool.

Conclusion: The long-term Framingham CVD risk function callibrates well in the low- to intermediate risk categories in the Rotterdam Study. Extending the original model by predicting CHD and Stroke separately provides clinicians with additional information about the relative contribution of CHD and Stroke to the total individual risk of CVD. This may have important consequences for subsequent allocation of preventive treatment to individual patients.