C-Reactive Protein

C-reactive protein is a nonspecific marker of inflammation. C-reactive protein was initially tested for association with CVD as investigators increasingly appreciated the role played by inflammation in the pathogenesis of atherosclerosis. In several studies, researchers have reported associations between hsCRP level and incidental CHD, stroke, or both.

In view of the consistent association, Ridker and associates evaluated the value of hsCRP in risk prediction in a number of analyses. They first examined the value of hsCRP level when added to variables used in the Framingham risk score (age, total cholesterol level, high-density lipoprotein cholesterol [HDL-C] level, smoking, and blood pressure) in the Women’s Health Study. In a cohort of 15,048 women aged 45 and older, 390 women had incident CVD events (116 myocardial infarctions, 217 coronary revascularization procedures, 65 deaths from cardiovascular causes, and 100 ischemic strokes) in an average follow-up period of 10 years. Although adding hsCRP level to a risk prediction model based on Framingham variables only marginally improved the area under the ROC curve (to 0.815, in comparison with 0.813 for the model without hsCRP level), other tests of discrimination, such as the Bayes information criterion, suggested that a model that included hsCRP level would be better. According to model calibration tested with the Hosmer-Lemeshow goodness-of-fit test, the model with hsCRP level was a better fit when expected and observed events were compared. Of the individuals predicted to have a 5% to 20% risk over 10 years, about 20% were reclassified after the addition of hsCRP level.

Ridker and colleagues then investigated whether risk prediction could be improved with the inclusion of several novel markers (e.g., levels of hsCRP, hemoglobin A1c, homocysteine, soluble intercellular adhesion molecule–1, apolipoproteins) that had been identified since the Framingham risk score had been described. They divided the Women’s Health Study cohort into a model derivation cohort ( n = 16,400) and a model validation cohort ( n = 8158). The variables that resulted in the best fitting model included age, hemoglobin A1c in subjects with diabetes, current smoking, lipoprotein(a) levels (if apolipoprotein B level ≥ 100 mg/dL), apolipoprotein B level, apolipoprotein A-I level, parental history of myocardial infarction (at age <60 years), and natural logarithms of systolic blood pressure and hsCRP level. Ridker and colleagues then simplified this model for clinical use by substituting levels of total cholesterol and HDL-C for levels of apolipoproteins B-100 and A-I and eliminating the measurement of lipoprotein(a) level ( Table 5-2 ). This Reynolds risk score, which differed from the Framingham risk score mainly in its use of hsCRP level and parental history of myocardial infarction, was found to have better model discrimination and calibration and reclassified 40% to 50% of individuals in the intermediate-risk group into higher risk or lower risk categories. However, no patient was reclassified from the low-risk group (<5% CHD risk over 10 years) to the high-risk group (>20% CHD risk over 10 years) or vice versa; this suggests that prior probability of disease should be considered in determining for whom additional testing is recommended.

The Reynolds risk score was subsequently described in men as well: in comparison with a traditional model, the Reynolds risk score reclassified 18% of subjects in the Physicians Health Study II, including 20% of subjects at intermediate risk, and was associated with a better model fit and discrimination. In addition, the Reynolds risk score was associated with an NRI of 5.3% and a clinical NRI of 14.2%. Other analyses have also suggested that the NRI for adding hsCRP level is approximately 5% to 7%. However, in a case-control study of individuals in the European Prospective Investigation into Cancer and Nutrition (EPIC)–Norfolk study, the NRI for adding hsCRP level was 12.0%.

More recently, a strategy of treating individuals with elevated hsCRP levels was studied in the Justification for the Use of Statins in Primary Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER). Individuals with low-density lipoprotein cholesterol (LDL-C) levels lower than 130 mg/dL and hsCRP levels of 2 mg/L or higher were treated with rosuvastatin; treatment with this drug was associated with a 44% relative risk reduction in major adverse cardiovascular events, and the trial was discontinued early because of clear benefit. Yang and coworkers analyzed data on participants in the Atherosclerosis Risk in Communities (ARIC) study according to the entry criteria for JUPITER; their findings suggested that elevated hsCRP level confers high risk regardless of LDL-C levels (either <130 mg/dL or ≥130 mg/dL) and after various traditional risk factors are taken into account.

The 2009 evaluation of hsCRP level by the United States Preventive Services Task Force (USPSTF) concluded that there is strong evidence that hsCRP level is associated with incident CHD, moderate evidence that hsCRP level can help in risk stratification of the intermediate-risk group, but insufficient evidence that reducing hsCRP level can prevent CHD events. However, in its systematic review of nine “emerging” CHD risk factors, including hsCRP level, the USPSTF concluded that current evidence does not support the use of any of these factors in further risk stratification. Similarly, other investigators have questioned whether adding hsCRP level has any additional value in risk stratification. Part of the reason that these questions have been raised is the significant correlation of hsCRP level with traditional risk factors and its minimal effect on the area under the ROC curve.

In an analysis of National Health and Nutrition Examination Survey (NHANES) data, Miller and associates reported that hsCRP levels were rarely high (>3 mg/L) in the absence of traditional risk factors associated with CHD, occurring in 4.4% of men and 10.3% of women, and that elevations in hsCRP levels that were attributable to a borderline or abnormal CHD risk factor occurred in 78% of men and 67% of women. Epidemiologic studies such as the Framingham Heart Study and the ARIC study have also demonstrated that the effect of hsCRP level on improving the area under the ROC curve is minimal and not statistically significant. However, using area under the ROC curve as the only metric to evaluate value in risk stratification can be suboptimal, because the C-statistic is based solely on ranks and is not as sensitive as measures based on likelihood. In fact, several well-established risk factors such as LDL-C and HDL-C may add little to the area under the ROC curve when added to other traditional risk factors.

Our own impression of the available data is that hsCRP level can help identify higher risk individuals among those classified as having intermediate short-term (10-year) risk for CHD by traditional risk prediction algorithms. However, it is unclear whether hsCRP level is a risk marker or a risk factor; that is, it is unclear whether hsCRP level plays a role in the pathogenesis of atherosclerosis or adverse cardiovascular events, or whether it is merely a bystander marking other changes that lead to atherogenesis and adverse cardiovascular events. Genetic studies have identified several loci associated with hsCRP levels but not with CVD, which suggests that hsCRP level may be a risk marker. However, whether it is a risk marker or a risk factor should not affect the ability of hsCRP level to predict risk.

In summary, there is consensus that elevation in hsCRP level is associated with increased risk for CHD and stroke. In our opinion, a clinically relevant number of individuals are reclassified, and a prospective trial has shown that treatment of individuals who have elevated hsCRP levels, “normal” LDL-C levels, and intermediate CHD risk can reduce both CHD and stroke. In addition, an expert panel convened by the National Academy of Clinical Biochemistry concluded, on the basis of a thorough literature review for a number of emerging risk factors, that only hsCRP level met all the criteria for acceptance for risk assessment in primary prevention.

Lipoprotein-Associated Phospholipase A ₂

LpPLA ₂ level is another biomarker that has consistently been shown to be associated with both CHD and stroke. LpPLA ₂ , which is predominantly associated with LDL in the circulation, is thought to mediate its inflammatory effects through its action on oxidized phospholipids, releasing lysophosphatidylcholine and oxidized nonesterified fatty acids, both of which are capable of attracting monocytes to an atherosclerotic lesion and further induce the expression of adhesion molecules.

LpPLA ₂ level has been evaluated as a marker for improving risk prediction. In the ARIC study, LpPLA ₂ level was the only marker (of 19 markers studied, including hsCRP level) that significantly increased the area under the ROC curve (by 0.006) when added to traditional risk factors that included age, race, sex, total cholesterol level, HDL-C level, systolic blood pressure, antihypertensive medication use, smoking status, and diabetes. However, in a more recent report from the EPIC-Norfolk study in which several markers were examined for their ability to improve risk prediction when added to a Framingham risk score–based model, only hsCRP level improved the C-statistic significantly; LpPLA ₂ level had no significant effect. Addition of LpPLA ₂ level in this study resulted in an NRI of 1.7% and a clinical NRI of 8.8%, whereas adding hsCRP level was associated with an NRI of 12.0% and a clinical NRI of 28.4%. However, the model fit was better with LpPLA ₂ level than with hsCRP level.

In view of the strong association of LpPLA ₂ level with stroke (ischemic), Nambi and colleagues, using an analysis of a case–cohort random sample ( n = 949, of whom 183 had incident ischemic stroke) from the ARIC study, evaluated whether LpPLA ₂ level could improve stroke risk prediction. Nambi and colleagues classified individuals’ 5-year risk for stroke as low (<2%), intermediate (2% to 5%), or high (>5%) on the basis of a traditional risk factor model that included age, sex, race, current smoking, systolic blood pressure, LDL-C level, HDL-C level, diabetes, antihypertensive medication, and body mass index and then added hsCRP and LpPLA ₂ levels separately and together to the analysis. Overall, adding LpPLA ₂ level significantly improved the area under the ROC curve (from 0.732 to 0.752; 95% confidence interval [CI] for change in area under the ROC curve, 0.0028 to 0.0310), whereas adding hsCRP level did not significantly increase the area under the ROC curve (from 0.732 to 0.743; 95% CI for change in area under the ROC curve, −0.0005 to 0.0183). However, adding both LpPLA ₂ and hsCRP levels, as well as their interaction, resulted in the best improvement in the area under the ROC curve, which increased to 0.774 (95% CI for change in area under the ROC curve, 0.0182 to 0.0607). The addition of hsCRP level, LpPLA ₂ level, and their interaction reclassified 4%, 39%, and 34% of the individuals originally classified as being at low, intermediate, and high risk, respectively.

In summary, LpPLA ₂ level has not been as well studied as hsCRP level, especially with regard to improving risk prediction. Available data suggest that its ability to improve CHD risk prediction may be modest, but its ability to improve ischemic stroke risk prediction may be better. Additional studies are needed to examine whether pharmacologic treatment of patients who have elevated LpPLA ₂ levels can reduce CVD events. LpPLA ₂ level may be a risk factor, not only a risk marker, and a large outcomes trial is examining whether inhibition of LpPLA ₂ in patients at high risk can reduce CVD events. Further studies will be needed to evaluate and identify the role for LpPLA ₂ level in CVD risk stratification.

Amino-Terminal Pro–B-Type Natriuretic Peptide

B-type (or brain) natriuretic peptide (BNP) is a cardiac hormone secreted by cardiomyocytes in response to pressure and ventricular volume overload. The amino-terminal fragment of its prohormone (NT-proBNP), which has traditionally been thought of as a marker for congestive heart failure, has also been associated with both CHD and stroke. The contribution of NT-proBNP level in risk stratification was examined in the Rotterdam study, in which NT-proBNP level was analyzed with traditional risk factors to investigate its ability to predict 10-year risk of CVD. For a group of 5063 individuals older than 55 years and free of CHD, addition of NT-proBNP level to traditional risk factors significantly improved the C-statistic both in men (0.661 to 0.694; change in C-statistic, 0.033; 95% CI, 0.012 to 0.052) and in women (0.729 to 0.761; change in C-statistic, 0.032; 95% CI, 0.016 to 0.047) and resulted in an NRI of 9.2% (95% CI, 3.5% to 14.9%; P = 0.001) in men and 13.3% (95% CI 5.9% to 20.8%; P < 0.001) in women. In the Rancho Bernardo Study, increased NT-proBNP levels or detectable troponin T levels in asymptomatic elderly participants were associated with increased risk for CVD death and total mortality rate, and participants with elevations of both markers had even higher risk.

Other Markers

Several other markers also have associations with CVD; however, information regarding their use in CVD risk stratification is limited. In the analysis from the ARIC study noted previously, in which researchers examined the effect of adding various markers ( n = 19) to traditional risk factors, only LpPLA ₂ level improved the area under the ROC curve. Rana and associates investigated the effect of adding levels of hsCRP, myeloperoxidase, LpPLA ₂ , secretory phospholipase A ₂ group IIA (sPLA ₂ ), fibrinogen, paraoxonase, macrophage chemoattractant protein–1 (MCP-1), and adiponectin to analyses of CHD risk stratification. Overall, hsCRP level was the only marker that significantly improved the area under the ROC curve (to 0.65, from 0.59 for a Framingham risk score–based model; P = 0.005). Level of hsCRP was also associated with the best NRI and clinical NRI (12% and 28.4%, respectively), and sPLA ₂ level was the next best (6.4% and 16.3%, respectively). However, when model fit was examined, adding hsCRP or paraoxonase or MCP-1 level to the Framingham risk score was associated with lack of model fit, whereas the addition of the other markers was associated with a good model fit. In the intermediate-risk group, the greatest numbers of individuals were accurately reclassified with the addition of sPLA ₂ level, followed by levels of fibrinogen, LpPLA ₂ , adiponectin, and myeloperoxidase. In separate case-control analyses from the EPIC-Norfolk study, CHD risk was noted to increase across increasing quartiles of myeloperoxidase level.

Multiple Markers

Because many of these markers improve risk prediction marginally, efforts have been made to evaluate the value of a multimarker approach by combining several biomarkers. With many of these multimarker approaches, the researchers examined primarily the association of markers (in concert) with CHD/CVD but not their use in risk stratification (reviewed by Koenig ).

Wang and coworkers assessed 10 biomarkers (levels of hsCRP, BNP, N-terminal pro–atrial natriuretic peptide, aldosterone, renin, fibrinogen, D-dimer, plasminogen-activator inhibitor type 1, and homocysteine, and the urinary albumin-to-creatinine ratio) in the Framingham Heart Study ( n = 3209) for their ability to predict major adverse cardiovascular events. BNP level (hazard ratio = 1.25) and urinary albumin-to-creatinine ratio (hazard ratio = 1.20) had the strongest association with major adverse cardiovascular events, and BNP level (hazard ratio = 1.40), hsCRP level (hazard ratio = 1.39), and urinary albumin-to-creatinine ratio (hazard ratio = 1.22) had the strongest association with death, but none of the markers affected the C-statistic significantly. The C-statistic for major cardiovascular events was 0.70 in a model that included age, sex, and the multimarker score; 0.76 in a model with age, sex, and conventional risk factors; and 0.77 in a model with all predictors.

Melander and associates evaluated the additional value of 6 biomarkers (levels of hsCRP, cystatin C, LpPLA ₂ , midregional proadrenomedullin [MR-proADM], midregional pro–atrial natriuretic peptide, and NT-proBNP) in 5067 participants without CVD from Malmö, Sweden (mean age, 58 years). After using a backwards elimination model to identify the best markers for prediction of CVD events ( n = 418) and CHD events ( n = 230) (median follow-up, 12.8 years), they reported that hsCRP and NT-proBNP levels best improved the C-statistic for prediction of CHD events (increase in C-statistic, 0.007; P = 0.04), whereas NT-BNP and MR-proADM levels best improved prediction of CVD events, although the improvement was not statistically significant (increase in C-statistic, 0.009; P = 0.08). Very few individuals were reclassified: 8% of the study population was reclassified for CVD risk prediction and 5% for CHD risk prediction. Similarly, improvements in NRI for CVD and CHD were nonsignificant, although improvements in clinical NRI were significant (7% and 15%, respectively, largely through reclassification to a lower risk category).

Multiple markers have also been studied in older individuals. In one study in individuals older than 85 years, traditional risk factors were poor predictors of cardiovascular mortality, and of the markers studied (levels of hsCRP, homocysteine, folic acid, and interleukin-6), homocysteine level was the best predictor of cardiovascular mortality (area under the ROC curve, 0.65; 95% CI, 0.55 to 0.75). On the other hand, Zethelius and associates reported significant improvement in prediction of CHD death in individuals older than 75 years with the use of biomarkers (levels of troponin I, NT-proBNP, cystatin C, and hsCRP); the C-statistic improved from 0.664 for traditional risk factors alone to 0.766 (difference, 0.102; 95% CI, 0.056 to 0.147) in the whole cohort and from 0.688 to 0.748 (difference, 0.059; 95% CI, 0.007 to 0.112) in subjects without CVD. The NRI for adding all the biomarkers was significant (26%, P = 0.005). Overall, this study was limited by the fact that only 136 subjects died from CVD.

Hence, even with the use of multiple markers, a consistent reliable set of markers has not been identified for CVD risk prediction. Of the novel markers studied, the addition of BNP level to hsCRP level appears the most reliable.

Advanced Lipoprotein Testing

Assessment of apolipoprotein B concentration and measurement of lipoprotein particle sizes with nuclear magnetic resonance (NMR) have been suggested as tests that may refine and improve risk prediction in comparison with cholesterol measures currently used clinically. Mora and colleagues examined the association of these tests with CVD and their ability to improve risk prediction in the Women’s Health Study, a study of healthy female health care professionals aged 45 years or older. Although both NMR lipid profile and apolipoprotein B concentration were associated with CVD after adjustment for nonlipid risk factors, the hazard ratios were similar to those for traditional lipid measures. The C-index was 0.784 for the model with nonlipid risk factors and ratio of total cholesterol to HDL-C levels, and it was not significantly different with the addition of LDL level measured by NMR (0.785) or apolipoprotein B level (0.786). NRI also did not show net improvement; in comparison with nonlipid risk factors and the total cholesterol–to–HDL-C ratio, NRI was 0% with NMR-measured LDL level and 1.9% with apolipoprotein B. This finding suggests that these novel lipid measures do not significantly enhance risk prediction in comparison with the traditional lipid measure of total cholesterol–to–HDL-C ratio. However, other studies in populations with higher baseline triglyceride values have demonstrated that apolipoprotein B level and other measures of LDL particle number provided additive prognostic value over LDL-C level.