Criteria for clinical use of biomarkers

Any of the dozens of proposed biomarkers should fulfill at least three specific criteria to be clinically useful for CHD risk prediction.

It is no secret that the traditional risk factors for coronary heart disease, as exemplified by the Framingham Risk Score, do not adequately identify the one-in-three women and one-in-two men who will eventually suffer major coronary heart disease events.

By design, the FRS forecasts events only 10 years into the future. If we rely primarily on the four most tried-and-true CV risk factors — hypertension, smoking, hypercholesterolemia and diabetes — we find that more than 60% of patients who ultimately develop CHD have either none or one of these risk factors.

Preventive cardiologists have directed an enormous amount of energy for decades trying to improve risk prediction by identifying novel risk factors. Because of their potential ease of use, biochemical markers, or “biomarkers,” that can be measured from serum have received a great deal of attention. Several dozen biomarkers whose levels correlate with risk for CHD have been identified, with varied roles in lipid biology, inflammation, endothelial function, clotting and other biological processes.

How can clinicians make sense of this proliferation of emerging biomarkers, all vying to be introduced into clinical practice?

By applying three strict criteria, we can discriminate among these biomarkers and choose the ones that will contribute meaningfully to patient management:

Roger Blumenthal

• Does the biomarker offer information that is not already provided by the traditional risk factors in the form of a multivariate predictive model such as the FRS?
• Does it improve existing risk prediction models enough to change patient management?
• Does it add prognostic information to other candidate biomarkers?

We propose that unless the answers to these questions are all yes, a candidate biomarker is not ready for widespread use.

Nontraditional risk factors

Numerous studies have shown that increasing levels of high-sensitivity C-reactive protein (hsCRP), the most extensively studied emerging biomarker, correlate with increasing risk for CHD. Yet increasing hsCRP levels are also highly correlated with obesity and metabolic syndrome, begging the question of whether hsCRP is mostly a way of looking at various components of metabolic syndrome.

However, after careful adjustment for contributions by traditional risk factors, hsCRP levels continue to be strongly correlated with CHD risk, albeit at reduced levels of relative risk. This finding has been reproduced in multiple studies, yielding FRS-adjusted risk ratios typically around 2.0 (ie, a doubling of risk with high hsCRP levels compared with low hsCRP levels). Thus, in most studies, hsCRP appears to integrate both data from traditional risk factors and data that have not previously been assessed by the FRS.

For a biomarker to be widely accepted, multiple studies are needed in different populations showing significant risk ratios after adjustment for traditional risk factors. This would establish the biomarker as a legitimate nontraditional risk factor that captures some aspect of CHD risk not already accounted for in the standard approach to atherosclerotic vascular disease.

Strengthening risk prediction

Even if a biomarker is established to be a nontraditional risk factor, as described above, this is not enough for its routine use in clinical practice. It must be demonstrated in carefully designed studies that the biomarker improves the predictive power of a traditional risk model such as the FRS. Furthermore, this improvement should be significant enough to change the management of a substantial number of patients: to prescribe or withhold aspirin and lipid-lowering therapy, to set more aggressive lipid-lowering goals (eg, LDL <70 mg/dL vs. <100 mg/dL vs. <130 mg/dL, and non-HDL <100 mg/dL vs. <130 mg/dL vs. <160 mg/dL), and to more aggressively encourage lifestyle changes, perhaps with the help of structured exercise and weight-loss programs.

In the Women’s Health Study (WHS) population, hsCRP did appreciably improve the FRS model, particularly with regard to low- and intermediate-risk patients as defined by the traditional FRS (5% to 10% and 10% to 20% 10-year CHD risk, respectively). The predicted risks for CHD among subgroups of patients match up much better with the actual incidence of CHD in these subgroups when hsCRP is taken into account.

According to Adult Treatment Panel III risk categories based on National Cholesterol Education Program beta-coefficients rather than those derived from the WHS, among women originally classified as having 5% to less than 10% and 10% to 20% 10-year greater 10-year risk, 38% and 42%, respectively, were reclassified in the WHS model that included CRP to more accurately reflect what actually happens to the patients. Thus, the decision whether to begin lifelong aspirin and lipid-lowering therapy becomes much simpler.

There is a plausible argument to be made for selectively checking hsCRP in certain low- and intermediate-risk FRS patients, to better risk stratify them and help decide on the appropriate pharmacologic strategy.

Is this enough to warrant the endorsement of hsCRP by guidelines committees? Debate will undoubtedly continue on what degree of incremental predictive value is deemed sufficient to recommend hsCRP for routine use. Much has been made of the fact that hsCRP as well as the other emerging biomarkers do not appreciably improve the receiver operating characteristic curve, or c-statistic, of risk prediction models such as the FRS. The c-statistic is one quantitative measure of how well a model can separate those individuals at risk from those not at risk.

Of the traditional risk factors, only age and diabetes significantly improve the c-statistic. Dyslipidemia and BP often do not improve the c-statistic, despite their clear importance for risk prediction and their unequivocal causal relationships to CHD. Some would argue that it is unreasonable to hold nontraditional risk factors to a higher standard than traditional risk factors in this regard, ie, to hold that hsCRP’s inability to improve the c-statistic invalidates its use in risk prediction. Others maintain that the burden of proof is on the new biomarkers, not the conventional risk factors that constitute the current risk prediction models like the FRS.

Complementing biomarkers

With several dozen candidate biomarkers from which to choose, it is likely that many of these biomarkers provide similar prognostic information. Out of the several inflammatory biomarkers listed in the table, will any provide utility above and beyond that provided by hsCRP? If one biomarker comes into wide use, there will be no role for another biomarker that simply replicates the first biomarker without offering something new. Thus, every candidate biomarker that is proposed for clinical use will not only have to be differentiated from the traditional risk factors but also from other nontraditional risk factors (eg, hsCRP) that are in serious consideration for clinical use.

This highlights the need for studies demonstrating that a candidate biomarker provides additive information compared with other candidate biomarkers. In studies comparing the relative predictive qualities of hsCRP and lipoprotein-associated phospholipase A₂ for CHD, they were found to be complementary in terms of providing prognostic information. Patients with high levels of both biomarkers were at increased risk compared with patients with a high level of one of the biomarkers but a low level of the other biomarker. The latter group of patients was at increased risk compared with patients with low levels of both biomarkers. A similar synergism has been observed with hsCRP and fibrinogen. This reflects that hsCRP, Lp-PLA₂, and fibrinogen may represent different axes, ie, inflammation, plaque instability and coagulation, that independently contribute to CHD.

Choosing biomarkers

To be sure, there are a number of features we want for the ideal biomarker: good sensitivity and specificity, well-defined reference limits, an internationally standardized assay and low cost. But even with all of these features, to be a truly practical biomarker, it needs to pass the litmus test comprising the three criteria we have outlined. To date, hsCRP is the only emerging biomarker that arguably fulfills these criteria. It independently predicts CHD risk after adjustment for traditional risk factors, it improves the risk prediction of an established model (the FRS) enough to change the management of some low- to intermediate-risk patients and it appears to offer information distinct from at least two other proposed biomarkers. Although not yet in routine clinical use, hsCRP seems likely to be more widely adopted as a new risk factor in persons at moderate risk for CHD in whom it is unclear whether to add aspirin and how low to go in terms of LDL and non-HDL.

Which other biomarkers merit serious consideration at this time? By our criteria, Lp-PLA₂ and fibrinogen are the most promising candidates. Both have been demonstrated in numerous studies to qualify as nontraditional risk factors, yielding CHD risk ratios around two (comparing patients with high levels of the biomarker vs. low levels) after adjustment for traditional risk factors. Both have also been shown to provide additive risk assessment when combined with hsCRP, suggesting that they represent distinct pathogenic mechanisms contributing to CHD. What is not yet available are data to show that the addition of either Lp-PLA₂ or fibrinogen improves the predictive power of established models such as the FRS. At a minimum, studies to evaluate such integrative risk models will be needed before guidelines committees should endorse Lp-PLA₂ and fibrinogen.

What is not needed to qualify a biomarker for clinical use is a direct pathogenic mechanism by which the biomarker causes CHD. Although much effort has been devoted to understanding how hsCRP and Lp-PLA₂ might directly contribute to atherosclerosis, this is not pertinent to their use in risk prediction models. Although it is tempting to think of these biomarkers as pharmacological targets via which one might reduce the risk for CHD, there is not yet data to support this thinking. Nevertheless, it is not necessary to equate utility as a risk predictor with utility as a target for preventive therapy; after all, age is the strongest predictor of CHD, but nobody has yet found a way to make us younger. In addition, a high troponin is associated with an increased risk in people with chest pain; however, our current therapies are not directed at troponin specifically.

Kiran Musunuru, MD, PhD, and Jacob Abraham, MD, are fellows at the Johns Hopkins Ciccarone Preventive Cardiology Center. Roger S. Blumenthal, MD, is Associate Professor of Medicine and Director of the Johns Hopkins Ciccarone Preventive Cardiology Center and a member of the Cardiology Today Editorial Board.

For more information:

• Khot UN, Khot MB, Bajzer CT, et al. Prevalence of conventional risk factors in patients with coronary heart disease. JAMA. 2003;290:898-904.
• Cook NR, Buring JE, Ridker PM. The effect of including C-reactive protein in cardiovascular risk prediction models for women. Ann Intern Med. 2006;145:21-29.
• Ballantyne CM, Hoogeveen RC, Bang H, et al. Lipoprotein-associated phospholipase A₂, high-sensitivity C-reactive protein, and risk for incident coronary heart disease in middle-aged men and women in the Atherosclerosis Risk in Communities (ARIC) study. Circulation. 2004;109:837-842.
• Koenig W, Khuseyinova N, Löwel H, et al. Lipoprotein-associated phospholipase A₂ adds to risk prediction of incident coronary events by C-reactive protein in apparently healthy middle-aged men from the general population: Results from the 14-year follow-up of a large cohort from southern Germany. Circulation. 2004;110:1903-1908.
• Mora S, Rifai N, Buring JE, Ridker PM. Additive value of immunoassay-measured fibrinogen and high-sensitivity C-reactive protein levels for predicting incident cardiovascular events. Circulation. 2006;114:381-387.