Background: The Framingham Heart Study risk model has been used in the majority of cardiovascular risk management guidelines. Recently, a new model based on the SCORE system has been proposed. We compared both risk models with regard to their ability to predict cardiovascular mortality in the Netherlands.
Design: Cohort study.
Methods: In a Dutch cohort study of 39 719 persons, three properties of the risk models were investigated: discriminating ability (ranking persons in order of risks, expressed in area under the curve); calibrating ability (prediction of events compared with actual events expressed in goodness of fit); and the number of persons assigned to treatment according to the guideline.
Results: The discriminative ability of both models was similar: the area under the curve of Framingham was 0.86 and of SCORE 0.85. Calibration of both functions was inadequate. The goodness of fit of the SCORE model was 35 and of the Framingham model 64, whereas a goodness of fit less than 20 is considered acceptable. Using the Dutch guideline treatment threshold of 10% mortality risk, the SCORE risk function assigned 0.4% of the population to drug treatment where the Framingham function assigned 0.7%.
Conclusion: The findings of this study show that both the SCORE and the Framingham model function have a good discriminative ability but are insufficient in predicting absolute risks. SCORE assigned fewer participants to treatment than Framingham. If a new risk model is implemented in treatment guidelines, comparison with the model in use and evaluation of calibrating features is needed.