Evaluation of the Framingham risk score in the European Prospective Investigation of Cancer-Norfolk cohort: does adding glycated hemoglobin improve the prediction of coronary heart disease events?

Description

Given the importance of the Framingham risk score in estimating a patient’s 10-year coronary heart disease (CHD) risk in primary and secondary care settings, the predictive value of this risk scoring system needs to be further studied and validated in other populations throughout the world. Simmons et al. therefore evaluated the Framingham risk score in a UK population-based prospective study, the European prospective investigations into cancer and nutrition (EPIC)-Norfolk study. They were also interested in investigating whether the addition of HbA1c to the Framingham risk score would improve CHD risk prediction on top of the Framingham risk score alone. In this analysis, 10,295 participants were followed for an average of 8.5 years, during which time 430 men and 250 women were admitted to the hospital or died with CHD as an underlying cause. Model discrimination was compared using area under the receiver operating characteristic (ROC) curves. The area under the ROC curve for the Framingham risk score (with variables with coefficients taken from the EPIC-Norfolk data) was 0.72 in men and 0.80 in women. When HbA1c was added to the Framingham risk score, the area under the ROC curves was 0.73 in men and 0.80 in women, suggesting that HbA1c does not add much to the CHD risk associated with the Framingham risk score. There was a trend towards net reclassification improvement in men but not in women. The authors suggested that their report will help fuel the debate regarding CHD risk algorithms and also called for other studies on the topic in other cohorts to further define the predictive value of HbA1c in individuals with or without diabetes. Pencina and D’Agostino were invited to comment on the results of this EPIC-Norfolk investigation. In their introduction, they reviewed the importance and the place of the latest statistical models used to assess the significance of biomarkers on top of well-established algorithms, e.g., the area under the ROC curve (or c-statistic), the net reclassification improvement, and the integrated discrimination improvement. After highlighting the pros and cons of the investigation by Simmons et al., they drew attention to the importance of statistical methods used in studies that test the predictability and the place of a novel biomarker in terms of CHD risk prediction.