Biomarkers, Better Than Conventional Risk Factors?
Abstract & Commentary
By Jonathan Abrams, MD, Professor of Medicine, Division of Cardiology, University of New Mexico, Albuquerque. Dr. Abrams serves on the speaker's bureau for Merck, Pfizer, and Parke-Davis.
Source: Wang TJ, et al. Multiple Biomarkers for the Prediction of First Major Cardiovascular Events and Death. N Eng J Med. 2006;355:25:2631-2639.
In recent years there has been considerable interest in the utilization of biomarkers to assist in the prediction of risk for cardiovascular events, including mortality. The most widely known marker is high-sensitivity C-reactive protein (hsCRP). Two peptides have been utilized extensively for evaluation of left ventricular function, as well as the characterization of congestive heart failure; these include B-type natriuretic peptide (BNP) and N-terminal pro-atrial natriuretic peptide (NT-ANP). A major area of interest in the biomarker field is to find markers that would help predict outcomes in addition to the classic risk factors, such as of nicotine, hypertension, diabetes, and hyperlipidemia.
Investigators from the Framingham Heart Study and other databases, including institutions in the NHLBI, sought to evaluate the efficacy of multiple biomarkers in predicting subsequent major vascular events and in particular, a first major cardiovascular event or death. The patient population was derived from participants of the 6th examination cycleof the Framingham Heart Offspring Study (1995-1998). Ten biomarkers were chosen, all of which have been associated with major CV events and have biologic plausibility. In addition to hs-CRP, BNP, and NT-ANP, other putative markers included plasma aldosterone, plasma renin, fibrinogen, plasminogen-activator inhibitor type 1 (PAI-1), D-dimer, homocysteine (HC), and urinary albumin-to-creatinine ratio. Each of these markers represents various aspects of vascular phenomena, such as thrombosis, inflammation,neuro-hormonal activity, endothelial function, fibrinolytic function, and glomerular endothelial function. Sophisticated and detailed use of multiple statistical approaches were employed; these will not be outlined in this review, but include multivariable proportional-hazards modeling "to examine the association of biomarker levels with the risks of death and major cardiovascular events." Participants were categorized by quintiles of a multi-marker score; the lowest 2 quintiles represented low risk; the highest, high risk, and the 3rd and 4th represented intermediate risk. Kaplan-Meier probability curves were constructed. In addition, hazard ratios were determined, adjusted for age and conventional risk factors. Any prior major CV event resulted in exclusion from the study. The C statistic was utilized "to classify risk," and is defined as "the probability of concordance among persons who can be compared." Receiver-operating characteristics were plotted for models with or without biomarkers. Secondary analyses included association of biomarkers with outcomes adjusted for age, gender, and lipid status. Of note, angina, claudication, or prior TIA were considered to be non-major CV events. After screening participants in the 6th examination of the Framingham Heart Offspring Study, 3209 individuals were selected to constitute the study sample. Mean age at the time of enrollment was 59 + 10 years. Follow-up was up to 10 years, median of 7.4 years, during which time 207 of the 3,209 participants died (6%). Multiple analyses of biomarker panel associations were made. In backward-elimination modeling, 5 biomarkers were selected as predictors of death, including CRP, NT-ANP, HC, plasma renin, and D-dimer. The final model utilized BNP, CRP, urinary albumin-to-creatinine ratio, HC, and renin.
Results: BNP and PAI-1 were predictive of future CV events; in the final model, BNP and urinary albuminto-creatinine ratio were included with hazard ratios of approximately 1.2-1.25. Kaplan Meier curves utilizing cumulative probability of death and major CV events demonstrated a marked elevation of risk in patients with high multi-marker scores, when compared to those with low or intermediate scores. Persons with high multi-marker scores had a 4 times greater risk for death and a 2 times greater risk of CV events than those in the low marker scores, p < 0.001 and p = 0.002. C statistic ranged from 0.68 to 0.79. ROC curves were constructed utilizing conventional risk factors with and without bio-markers. Adjustment for use of statins, aspirin, and other medications did not alter the findings. PAI-1 was statistically associated with outcomes.
The authors conclude that "the most informative biomarkers for predicting death were…BNP, CRP, HC, renin, and the urinary albumin-to-creatinine ratio." The most useful biomarkers for predicting major CV events included BNP and urinary albumin-to-creatinine ratio. Importantly, they state, "nonetheless, the use of multiple biomarkers added only moderately to the overall prediction of risk based conventional CV risk factors." Prediction of subsequent risk was affected in part by the overlap in the distribution of biomarker levels in individuals with and without CV disease. Furthermore, the authors stress that conventional risk factors are quite effective in predicting risk. The authors point out that the major biomarkers listed in the analysis have been individually shown to be predictors of death or CV events in single biomarker investigations, although PAI-1 data has been limited. CRP was predictive of death but not a major CV event, when other biomarkers were accounted for, while the risk of HS-CRP for CV events was 1.3-1.5 in other studies. The authors suggest that their data supports BNP and the urinary albumin-to-creatinine ratio as being better predictors of global CV risk than CRP. They conclude that their data supports only a moderate ability for biomarkers to predict subsequent death or major CV events when conventional risk factors are also considered. The authors suggest that biomarker assessment might be most useful in patients at intermediate risk, whereby accurate risk status might affect how aggressive one should be for treating cholesterol, hypertension, or diabetes. They also comment that the Framingham Heart Offspring population sample reflects unselected individuals with varying CV risk. They conclude that biomarkers are associated with the risk of death and major CV events, but only moderately, so when used in conjunction with conventional risk factor assessment. They posit whether new biomarkers may improve the situation in the future.
This is a rather sobering report, suggesting that targeted use of known selected biomarkers, all which have a pathophysiologic link to CV disease, may not be a particularly useful strategy. In particular, CRP turned out to only have modest predictive accuracy, consonant with other studies. While individuals with multiple biomarkers have a considerably increased risk of death in this report, they do not represent the majority of patients, and many subjects overlap with respect to biomarker prevalence without increasing predictability. Other approaches for examining prospective assessment of normal individuals with respect to future CV risk, such as carotid IMT, coronary calcium scores, or routine stress testing of healthy populations, have not fared as well as anticipated. The Framingham risk assessment instrument, while probably not widely used, has yet to be improved upon by other risk-screening approaches. Furthermore, it is well established that the large majority of individuals who develop CV disease can be predicted by standard risk factors, such as diabetes, hypertension, dyslipidemia, obesity, smoking, and inactivity. An editorial by James Ware, PhD, a biostatistician at the Harvard School of Public Health, discusses the role of biostatistics in risk prognostication, and in particular some of the results from the Wang study. He states that "the proposed biomarker score adds little to the sensitivity and specificity of a prognostic test for death within 5 years." He posits that the approach utilized in the study is of limited value in individual subjects, whereas it does contribute to the proportional-hazards model for predicting death from any cause. He concludes, "how difficult it is to achieve effective risk stratification with respect to multifactorial disease processes. Much work remains to be done before biomarkers of the type the authors consider here can provide a basis for the prognostic evaluation of the individual patient."
While this study adds to the disappointment of biomarker utilization in healthy individuals to predict subsequent risk, it would appear to clear the air, and emphasizes the importance of adherence to proven risk factors that are so well known to everyone, but too often ignored in patient care.