Predicting CHD risk in France: a pooled analysis of the D.E.S.I.R., Three City, PRIME, and SU.VI.MAX studies

BACKGROUND: We aimed to develop and validate a simple coronary heart disease (CHD) risk algorithm applicable to asymptomatic men and women in France, and to compare its accuracy with that of the last published version of the Framingham risk function for cardiovascular disease. DESIGN: A pooled analysis of four French prospective general-population studies. METHODS: The baseline and follow-up data from D.E.S.I.R., PRIME, Three City, and SU.VI.MAX studies were used. The 10-year CHD risk was estimated by the Cox proportional hazards model with candidate variables including age, gender, body mass index, waist circumference, family history of coronary heart disease, smoking status, diabetes status, systolic blood pressure, and total and high-density lipoprotein (HDL) cholesterol. RESULTS: The study population included 22,256 subjects (61.4% men) aged (SD) 56.0 years (8.3) without a personal history of CHD at baseline. After a mean follow-up of 8.0 years (2.3), 788 first CHD events occurred, 726 in men and 62 in women. The final model included age, gender, age × gender interaction, current smoking status, diabetes status, systolic blood pressure, total and HDL cholesterol. Using this model, the number of predicted coronary events fitted that given by the 10-year Kaplan-Meier survival estimates within each decile of estimated risk (calibration). This model had fair discrimination: Harrell C-index, 0.7831 (95% CI: 0.7704-0.7957). For comparison, the recalibrated Framingham risk function had equivalent performances compared to the French risk equation. CONCLUSION: Our 10-year French CHD risk equation based on traditional risk factors performed at least as well as the recalibrated Framingham cardiovascular disease risk function.