|| Checking for direct PDF access through Ovid
Current cardiovascular risk prediction models are largely derived from traditional risk factors. Recent studies have indicated that including novel biomarkers as risk factors in new algorithms could improve risk stratification.To construct and evaluate an algorithm, derived from both established and novel risk factors, for cardiovascular risk prediction in women.Participants were drawn from the Women's Health Study (WHS) cohort, which comprised female US health professionals aged 45 years or older who had no evidence of cardiovascular disease at study enrollment in September 1992. Baseline plasma samples were measured for lipoprotein(a), apolipoproteins A-I and B-100, total cholesterol, HDL and LDL cholesterol, high-sensitivity C-reactive protein (hsCRP), glycated hemoglobin, soluble intercellular adhesion molecule 1, creatinine, fibrinogen, and homocysteine. The algorithms were developed with two-thirds of participants and validated in the remaining third with measures of entropy, the Yates Slope, and the Brier score. Two models were created. Risk factors were selected for inclusion in Model A on the basis of minimization of a likelihood measure known as the Bayesian Information Criterion (BIC). Nine of a possible 35 factors were included in this model: age, current smoking, systolic blood pressure, hsCRP, apolipoprotein A-I, apolipoprotein B-100, lipoprotein(a), glycated hemoglobin, and parental history of myocardial infarction. Model B (the Reynolds Risk Score) was a simplified version of Model A and was identical except for the substitution of lipoprotein(a) and both apolipoproteins by total cholesterol and HDL cholesterol. Study participants assigned to one of four 10-year risk groups (<5%, 5−<10%, 10−<20%, and ≥20%), in accordance with the Adult Treatment Panel (ATP) III risk prediction model, were reclassified by use of Model A or B. Predicted event rates were then compared with actual event rates during the follow-up period (mean 10.2 years).Predicted and actual 10-year event rates for coronary revascularization, myocardial infarction, ischemic stroke, and cardiovascular death.A total of 24,558 women participated in the study - 16,400 in the derivation group and 8,158 in the validation group. The median age of participants was 52 years. During follow-up, 504 cardiovascular events occurred in the derivation group and 262 in the validation group. The BIC values for Models A and B were 9,039.4 and 9,067.5, respectively, and were both lower than the BIC values for the ATP III model (9,098.5) and the Framingham Risk Score (FRS; 9,161.2). In model A, 50% of women without diabetes at 5−<10% risk were reclassified to a higher risk category. In model B, 44% of women without diabetes at 10−<20% risk were reclassified to a lower risk category. Improved accuracy of classification, when compared with the ATP III model, was achieved for 99.72% and 99.07% of participants with Models A and B, respectively.The algorithms developed and validated in this study greatly improved the accuracy of cardiovascular risk prediction in women when compared with existing models.