Further Improvements in CHD Risk Prediction for Women

Coronary heart disease (CHD) is the leading cause of death for women and men in the United States. Because half of first major coronary events occur in asymptomatic individuals,¹ clinicians who want to implement appropriate primary prevention therapy must be able to accurately identify “at risk” individuals. The third National Cholesterol Education Program (NCEP) Adult Treatment Panel III (ATP-III) guidelines² recommend that all adults should undergo an office-based assessment to evaluate risk of a CHD event based on the Framingham risk score. This prediction algorithm incorporates age, levels of total cholesterol, and high-density lipoprotein cholesterol (HDL-C), smoking status, and systolic blood pressure to estimate a 10-year risk for developing a myocardial infarction or death due to CHD.²

Three levels of risk (low, intermediate, and high) are identified. The 2001 NCEP ATP-III guidelines define intermediate risk as a 10% to 20% risk of a nonfatal myocardial infarction or CHD death over the next 10 years, whereas the 2003 American College of Cardiology Bethesda Conference on Atherosclerosis Imaging suggested that this intermediate-risk group should be reclassified as those at 6% to 20% risk.³ The ATP-III guidelines set thresholds for lipid treatment based on the Framingham risk score–determined 10-year CHD risk.² However, subsequent studies have suggested the ATP-III risk prediction model misclassifies both clinical⁴ and subclinical^{5 - 7} CHD risk in asymptomatic postmenopausal women.

In addition, the ATP-III version of the Framingham risk score predicts the risk of future hard CHD events, such as myocardial infarction and CHD death, but not soft events, such as stroke, angina, or coronary revascularization, even if revascularizations were performed to treat acute coronary syndrome. Yet, women are more likely to experience a soft CHD event compared with men. Furthermore, the Framingham risk score predicts CHD risk, but total cardiovascular disease risk should be the more important outcome for prediction and preventive strategies given that other forms of atherosclerotic cardiovascular disease now predominate.

The study by Ridker et al⁸ in this issue of JAMA provides a timely contribution to the cardiovascular risk prediction literature. Nearly 25 000 healthy US women 45 years and older were followed up for a median of 10.2 years for incident CHD and stroke. The authors evaluated 35 risk factors in a random two-thirds sample of the cohort and then assessed the validity of the new risk algorithm in the remaining women. Using this new algorithm, the Reynolds Risk Score, women were classified into higher- or lower-risk categories with improved accuracy compared with the currently used risk prediction model.

Importantly, this new risk prediction algorithm is simple. Only 2 new variables—parental family history of premature CHD and high-sensitivity C-reactive protein (hsCRP)—were added to the variables in the ATP-III risk score in the most “parsimonious” model reported by Ridker et al. Both of these risk predictors have been validated in other populations to be independently associated with an elevated CHD risk.^{9 - 13} An additional advantage of the Reynolds model over the ATP-III model is the prediction of total cardiovascular disease events (including stroke and coronary revascularization) in addition to hard CHD end points.

These findings raise several critical questions. First, what is the potential impact of this new model in terms of changing risk prediction category, altering the low-density lipoprotein cholesterol (LDL-C) and non–HDL-C goals, and influencing the choice of whether to treat with life-long aspirin in an individual patient? Among 100 000 intermediate-risk women (5%-20% risk of a major cardiovascular disease event, and with an initial LDL-C goal of <130 mg/dL [<3.367 mmol/L]), the new model reclassifies 5400 (5.4%) women to high-risk (LDL-C goals of <100 mg/dL [<2.58 mmol/L] and an optional goal of <70 mg/dL [<1.81 mmol/L]). On the other hand, 13 400 (13.4%) women were reclassified as very low-risk (<5% risk of a major cardiovascular disease event and with an LDL-C goal of <160 mg/dL [<4.14 mmol/L]). Overall, approximately 20% of women will have different lipid treatment goals based on the Reynolds model than recommended by the ATP-III guidelines (when considering 5%-20% 10-year risk as intermediate risk).

This results in a smaller proportion of women eligible for aggressive preventive strategies, such as lifelong aspirin and lipid-lowering pharmacotherapy use. Still, nearly 80% of the intermediate-risk women will remain in this broad category (5%-20% CHD risk) if either the ATP-III risk prediction model or the model proposed by Ridker et al is used. Whether the approach of routinely incorporating family history and hsCRP into treatment decisions will result in lower cardiovascular morbidity in a cost-effective manner remains to be determined, but it does appear to result in more accurate risk assessment.

The second question is which of the 2 newly added variables in the Reynolds Risk Score has a greater influence on risk prediction? Table 6 in the study by Ridker et al details how different levels of hsCRP along with the presence or absence of parental family history of premature CHD affects the 10-year risk based on variables for a 50-year-old woman without diabetes and who smokes considered to be at intermediate risk by the ATP-III model (10-year CHD risk of 11.5%). From the example presented by the authors, a family history of premature CHD is at least as strong a factor affecting 10-year risk prediction. This supports recent observations of the strong association of family history with clinical and subclinical CHD.^{11 - 13} As for hsCRP, either very low or extremely high levels seem to influence the 10-year risk depending on the absence or presence of family history of premature CHD, respectively.

Very few white women aged 50 to 59 years have an hsCRP level of 0.5 mg/dL or less or higher than 10 mg/L,^{14 - 15} thus reducing the likelihood that hsCRP determination will make a major change in risk prediction in the majority of women considered in the CHD risk range of 10% to 20% with the current ATP-III model in this age group. However, hsCRP may have more utility in women with an estimated risk of less than 10%, which constitutes the majority of postmenopausal women without documented diabetes mellitus until age 70 years.^{3 ,16} Moreover, although this study population did not include large numbers of black and Hispanic women, who tend to have higher levels of hsCRP, selective measurement of hsCRP may result in a significant change in the 10-year risk levels.^{15 - 16}

Third, could the Reynolds risk prediction algorithm work for men? Both parental family history of premature CHD and hsCRP have been shown to predict CHD events in men, and it is likely that incorporating both markers will be as effective in accurately reclassifying CHD risk among men. However, hsCRP levels are higher in women than they are in men across all ethnic groups.^{15 - 16} Because men have a higher incidence of CHD¹ despite lower CRP levels,^{14 - 15} further studies are needed to explain the paradox of hsCRP and CHD outcomes observed between the sexes.

And fourth, what is the utility of other markers in risk prediction, such as coronary artery calcification scores or exercise testing measures, that were not assessed in this study? Considerable data indicate that coronary artery calcification scoring and stress-testing measures (exercise capacity, heart rate recovery) provide prognostic information above and beyond the Framingham risk score in the intermediate risk group.^{17 - 18} Both hsCRP and coronary artery calcification scoring complement each other in CHD risk prediction.^{19 - 20} Outcome data from cohort studies, such as the Multi-Ethnic Study of Atherosclerosis (MESA),²¹ may help to not only validate the currently proposed model in different racial/ethnic groups in both women and men but may further improve it by incorporating selective use of both biomarkers and imaging tests.

From a practical standpoint, both the Framingham risk score and Reynolds models only predict 10-year risk, whereas for women the issue is most often lifetime risk. Data from Framingham Heart Study indicate that a woman who is free of cardiovascular disease at age 50 years has a lifetime risk for cardiovascular disease events of 39% (which exceeds lifetime risk of breast cancer, lung cancer, and colorectal cancer combined).²² This highlights the limitation with current treatment algorithms based on a calculation of short-term risk alone. Lifetime CHD risk is high and any single risk factor left untreated will lead to atherosclerotic vascular disease.

Future studies using multiple risk prediction markers in conjunction with outcome data will improve the ability to develop more accurate risk-prediction tools. This approach will permit more effective identification of which asymptomatic adults need treatment with aspirin and lipid-lowering pharmacotherapy, as well as more intensive dietary and exercise interventions. Future multivariable models to predict a woman's long-term (20-30 years) risk of developing a major atherosclerotic vascular disease event are also needed. The Reynolds Risk Score is an important contribution to preventive cardiology and provides the framework for evaluating future emerging risk factors.