doi: 10.1161/CIRCULATIONAHA.104.482570
(Circulation. 2006;113:2335-2362.)
© 2006 American Heart Association, Inc.
Basic Science for Clinicians |
Biomarkers of Cardiovascular Disease
Molecular Basis and Practical Considerations
From the National Heart, Lung, and Blood Institute’s Framingham Heart Study, and the Cardiology Section and Department of Preventive Medicine and Epidemiology, Boston University School of Medicine, Boston, Mass.
Correspondence to Ramachandran S. Vasan, MD, Framingham Heart Study, 73 Mount Wayte Ave, Suite 2, Framingham, MA 01702-5803. E-mail vasan{at}bu.edu
Key Words: atherosclerosis • epidemiology • genetics • genomics • proteins
 |
Introduction |
|---|
 Top Introduction What Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
Cardiovascular diseases (CVD) are the leading cause of morbidity and mortality in the United States.1 Primary prevention and secondary prevention of CVD are public health priorities.2 Substantial data indicate that CVD is a life course disease that begins with the evolution of risk factors that in turn contribute to the development of subclinical atherosclerosis.3,4 Subclinical disease culminates in overt CVD.5,6 The onset of CVD itself portends an adverse prognosis with greater risk of recurrent events, morbidity, and mortality.7,8 It is also increasingly clear that although clinical assessment is the keystone of patient management, such evaluation has its limitations.9–12 Clinicians have used additional tools to aid clinical assessment and to enhance their ability to identify the "vulnerable" patient at risk for CVD, as suggested by a recent National Institutes of Health (NIH) panel.13,14 Biomarkers are one such tool to better identify high-risk individuals, to diagnose disease conditions promptly and accurately, and to effectively prognosticate and treat patients with disease. This review provides an overview of the molecular basis of biomarker discovery and selection and the practical considerations that are a prerequisite to their clinical use.
|
What Is a Biomarker? Definition and Types |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
The term biomarker (biological marker) was introduced in 1989 as a Medical Subject Heading (MeSH) term: "measurable and quantifiable biological parameters (eg, specific enzyme concentration, specific hormone concentration, specific gene phenotype distribution in a population, presence of biological substances) which serve as indices for health- and physiology-related assessments, such as disease risk, psychiatric disorders, environmental exposure and its effects, disease diagnosis, metabolic processes, substance abuse, pregnancy, cell line development, epidemiologic studies, etc." In 2001, an NIH working group standardized the definition of a biomarker as "a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention" and defined types of biomarkers (Table 1).15
|
A biomarker may be measured on a biosample (as a blood, urine, or tissue test), it may be a recording obtained from a person (blood pressure, ECG, or Holter), or it may be an imaging test (echocardiogram or CT scan).
Biomarkers can indicate a variety of health or disease characteristics, including the level or type of exposure to an environmental factor, genetic susceptibility, genetic responses to exposures, markers of subclinical or clinical disease, or indicators of response to therapy. Thus, a simplistic way to think of biomarkers is as indicators of disease trait (risk factor or risk marker), disease state (preclinical or clinical), or disease rate (progression).16 Accordingly, biomarkers can be classified as antecedent biomarkers (identifying the risk of developing an illness), screening biomarkers (screening for subclinical disease), diagnostic biomarkers (recognizing overt disease), staging biomarkers (categorizing disease severity), or prognostic biomarkers (predicting future disease course, including recurrence and response to therapy, and monitoring efficacy of therapy).15
Biomarkers may also serve as surrogate end points (Table 1).15 Although there is limited consensus on this issue, a surrogate end point is one that can be used as an outcome in clinical trials to evaluate safety and effectiveness of therapies in lieu of measurement of the true outcome of interest. The underlying principle is that alterations in the surrogate end point track closely with changes in the outcome of interest.17–19 Surrogate end points have the advantage that they may be gathered in a shorter time frame and with less expense than end points such as morbidity and mortality, which require large clinical trials for evaluation. Additional values of surrogate end points include the fact that they are closer to the exposure/intervention of interest and may be easier to relate causally than more distant clinical events. An important disadvantage of surrogate end points is that if clinical outcome of interest is influenced by numerous factors (in addition to the surrogate end point), residual confounding may reduce the validity of the surrogate end point. It has been suggested that the validity of a surrogate end point is greater if it can explain at least 50% of the effect of an exposure or intervention on the outcome of interest.20
|
Characteristics of an Ideal Biomarker |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
The overall expectation of a CVD biomarker is to enhance the ability of the clinician to optimally manage the patient. For instance, in a person with chronic or atypical chest pain, a biomarker (eg, treadmill stress test or dobutamine stress echocardiogram) may be expected to facilitate the identification of patients with chest pain of an ischemic etiology (angina). In a patient presenting to the emergency department with acute severe chest pain (suspected acute coronary syndrome), a biomarker may help to differentiate patients with an acute myocardial infarction (MI) from those with unstable angina (eg, troponin I or T), acute pulmonary embolism (eg, D-dimer or ventilation perfusion scan), or an aortic dissection (eg, transesophageal echocardiogram) in a timely fashion to facilitate targeted management. In a patient with an established acute MI, a biomarker may be able to assess the likelihood of the following: a therapeutic response (eg, ECG ST-segment elevation indicating need for thrombolysis); the extent of myocardial damage (eg, troponin); the severity of underlying coronary disease (eg, coronary angiography); the degree of left ventricular dysfunction (eg, echocardiography); the risk of future recurrences (eg, exercise stress test); and progression to heart failure (eg, B-type natriuretic peptide [BNP]).
Regardless of the purpose for its use, a new biomarker will be of clinical value only if it is accurate, it is reproducibly obtained in a standardized fashion, it is acceptable to the patient, it is easy to interpret by clinicians, it has high sensitivity and high specificity for the outcome it is expected to identify, it explains a reasonable proportion of the outcome independent of established predictors consistently in multiple studies, and there are data to suggest that knowledge of biomarker levels changes management (Table 2
).21 Table 2 displays the desirable properties of biomarkers overall and of biomarkers of screening, diagnosis, and prognosis.22–29
|
|
The desirable properties of biomarkers vary with their intended use.30 For screening biomarkers, features such as high sensitivity, specificity, and predictive values, large likelihood ratios (discussed below), and low costs are important. For diagnostic markers of acute cardiac disease (such as acute MI), in addition to the aforementioned characteristics, rapid sustained elevation, high tissue specificity (indicating myocardial origin), release proportional to disease extent, and assay features conducive to point-of-care testing are critical.22 For biomarkers monitoring disease progression or response to therapy, features such as sensitivity or specificity are less important because the patient serves as his or her own control (baseline values are compared with follow-up values). Narrow intraindividual variation and tracking with disease outcome or therapy are critical. Costs may be less important for prognostic markers because only people with disease are tested. Some biomarkers (eg, exercise stress test) may be used for both diagnostic and prognostic purposes. Establishing the prognostic utility of a biomarker is more challenging because it requires a larger sample and a prospective design, whereas demonstrating its value as a diagnostic test requires a smaller sample and a cross-sectional design.31
Regardless of the intended use, it is important to remember that biomarkers that do not change disease management cannot affect patient outcome and therefore are unlikely to be cost-effective (judged in terms of quality-adjusted life-years gained). Typically, for a biomarker to change management, it is important to have evidence that risk reduction strategies should vary with biomarker levels, and a biomarker-guided approach translates into better patient outcomes compared with a management scheme (usually the current standard of care) without biomarker levels. It also means that biomarker levels should be directly or indirectly modifiable by therapy on the basis of evidence from prospective clinical trials. Biomarkers that do not result in medical intervention may still serve other useful purposes, such as the reassurance value of a negative exercise test in an asymptomatic airline pilot.32 There are other examples of psychological benefits accruing from negative biomarker test results such as testing for genetic susceptibility for cancer33 or Alzheimer disease.34 In other situations, biomarkers may serve as research tools by providing insights into disease mechanisms.
|
Defining Abnormal Biomarker Values |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
Defining abnormal values is a critical step before the clinical use of a biomarker.30 It is important to characterize the distribution of the markers in people in the community and in patient samples on whom the biomarker will be tested. Thus, variation in levels with age, sex, ethnicity, and prevalent disease and the relations of biomarkers to known risk factors must be characterized.35
At least 3 potential approaches exist for defining abnormal biomarker levels (Figure 1). Reference limits are generated with the use of cross-sectional analyses of a reference sample (usually a healthy sample free of the disease of interest), and an arbitrary percentile cutpoint (typically the 95th or 97.5th percentile) is chosen to define abnormality.36–38 The reference range is the interval between the minimum and the maximum reference values. Approximately 200 individuals are required within each category for the formulation of reference limits for subgroups (eg, defined by age and sex).39 Cutpoints that define abnormality are typically the lower and the upper bounds of the 95% reference interval (between the lower 2.5th percentile and upper 97.5th percentile), but they may vary on the basis of the intent. The reference interval may be moved up or down according to the tradeoff between the implications (medical, ethical, social, psychological, and economic) of false-negative and false-positive results, ie, the consequences of missing disease, the availability and efficacy of treatment for people with abnormal values, and the costs associated with follow-up of abnormal results. For instance, the 99th percentile value has been used to define an abnormal troponin or creatine kinase–MB value; values exceeding this limit would indicate the presence of myocardial necrosis and an acute MI.40,41 When less specific markers of myocardial necrosis are used, a higher threshold may be used; for example, if total creatine kinase is used for the diagnosis of acute MI, a value twice the upper reference limit is recommended.40
|
Several issues must be considered when reference limits are interpreted. First, a select proportion of "normal" individuals will exceed the reference limits on the basis of the percentile chosen. Second, values that lie within statistically defined reference limits may not indicate health in a given individual, especially when the person comes from a group inherently different from the one used to derive the reference limits. Third, a change in values within the reference range may indicate pathology. Accordingly, delta limits have been formulated to evaluate the change in biomarker values within an individual (in response to disease or therapy) relative to the physiological intraindividual fluctuation of values. Fourth, a value within the reference range may not necessarily be desirable, especially when the prevalence of undesirable values of a biomarker in the population is high. Thus, abnormal blood pressure or cholesterol values are not defined on the basis of the distributions of these risk factors in the community; rather, "desirable" levels are defined (see below).
Discrimination limits are also used to indicate abnormal biomarker values.42 Such limits are generated by evaluating the degree of overlap between patients with and without disease in cross-sectional studies.42 Discrimination limits trigger decisions (they are referred to as decision thresholds). The 99th percentile value of troponin for a reference sample is in essence a discrimination limit because it identifies the presence of a MI. The discrimination thresholds can be varied depending on the relative importance of missing disease versus that of misclassifying nondiseased individuals. For example, a plasma BNP value >100 pg/mL with the Biosite assay may trigger suspicion of heart failure in a dyspneic individual.43 The reference limits of the Biosite assay exceed this threshold in women aged >65 years (95th percentile is 120 pg/mL).44 A plasma value >200 pg/mL has been suggested as a threshold indicating heart failure.45
A third method is to define "undesirable" biomarker levels by relating values to the incidence of disease and seeking a threshold beyond which risk escalates. For instance, a desirable systolic blood pressure may be 
115 mm Hg because incidence of vascular disease increases continuously above this level.46 On a parallel note, low-density lipoprotein cholesterol levels 100 mg/dL are deemed to be optimal.47 For most CVD risk factors, there is a continuous gradient of risk across the range of risk factors, and a majority of individuals in a population could be classified as having undesirable levels. "Treatment" levels (especially for pharmacological treatment) of risk factors may therefore differ from undesirable levels, being defined by the risk factor thresholds for which there is good evidence (typically from large randomized controlled trials) that treatment for values above a limit does more benefit than harm. Often such treatment levels may be defined not only by the level of the specific risk factor being evaluated but by taking into consideration absolute risk of disease based on the values of several other risk factors.48 Thus, a blood pressure level of 140 (systolic) or 90 (diastolic) mm Hg or more indicates systemic hypertension.49 However, experts have argued that blood pressure levels above or below this threshold could be treated on the basis of the absolute risk of CVD events, which in turn depends on the concomitant burden of other risk factors.48 For other biomarkers, the choice of the optimal cutpoint defining abnormality remains to be described and may vary with the purpose. For instance, Framingham data indicate that a plasma BNP value exceeding the 80th percentile value in the cohort (20.0 pg/mL for men and 23.3 pg/mL for women; Shionogi assay) is associated with a 76% increased risk of CVD and a tripling of congestive heart failure hazard.50 These values are below the 95th percentile value of a healthy reference sample at any age (Shionogi assay).51 Therefore, cutpoints of plasma BNP that identify discrimination limits for a diagnosis of congestive heart failure may differ from the upper reference limit, which in turn may vary from desirable levels.
Once abnormal thresholds of markers are formulated by any of the 3 aforementioned methods, biomarker performance can be assessed with the use of principles outlined in the next section.
|
Evaluation of Biomarker Performance: General Principles |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
The exact yardstick for evaluating the performance of a biomarker varies on the basis of the intended use. Good-quality biomarker studies make an independent masked comparison of the performance of a given biomarker with a reference standard in an appropriate sample of consecutive patients that represents an adequate spectrum of the disease.25 In general, the performance of biomarkers is seldom as good in a second sample as in the sample in which they were initially assessed. Consequently, it is desirable that biomarkers be evaluated initially in a derivation or training set and then investigated in a validation or test set.52 Standards have been proposed for designing and reporting the results of studies evaluating the performance of biomarkers for diagnosis25 and for prognosis.26
The accuracy of a biomarker test is evaluated in terms of its sensitivity (detection of disease when disease is truly present, ie, identifying true-positives) and its specificity (recognition of absence of disease when disease is truly absent, ie, identifying true-negatives) at select cutpoints. Several CVD biomarkers are continuously distributed quantitative variables, although there are some notable exceptions (for example, gender, race, diabetes, hypertension, genotypes). It is therefore critical to evaluate the information content of a biomarker over a range of values, often with the use of receiver operating characteristic (ROC) curves.53–55 The ROC curves illustrate the tradeoff between sensitivity and specificity when biomarker levels are used clinically to identify disease. Each point on the ROC curve indicates the conditional probability of a positive test result from a random diseased individual exceeding that from a random nondiseased person.56 Likelihood ratios57 (LR) are calculated with the use of sensitivity and specificity data (Table 2
) and may be more helpful to clinicians by answering the question of interest: how likely are we to obtain a positive test result in someone with disease compared with someone without disease (LR+), and how likely are we to get a negative result in someone with disease compared with someone without disease (LR–)?
For example, if a biomarker is to be used to screen for an uncommon condition in asymptomatic people (eg, preclinical left ventricular systolic dysfunction), it should have high specificity because a "rule in" (or confirm diagnosis) strategy is more important in this situation (also called the SpIN rule27). Expressed in terms of LR, a test with a greater LR+ (typically >10) is needed; this is because the costs of mislabeling a healthy individual (predicting disease when health is likely) may outweigh the costs of missing a rare condition. Sometimes when multiple tests are considered for screening, they are obtained in series.58 When multiple tests are obtained in series and disease is considered positive when all tests are positive ("AND rule"), specificity is enhanced but sensitivity is diminished.23,59 For instance, Ng et al60 have proposed that a sequential strategy of checking people with an initial urine N-terminal pro-BNP (N-BNP) test followed by a plasma N-BNP test (in urine "positive" cases) may facilitate screening for asymptomatic left ventricular systolic dysfunction in the community by reducing the need for follow-up echocardiograms. When multiple tests are obtained in parallel and disease is considered to be present when any of the tests is positive ("OR rule"), sensitivity is increased at the cost of specificity.23,59 For a biomarker to be accepted as a routine screening test it is important to demonstrate that a strategy of measuring the biomarker improves patient outcomes relative to a conventional strategy that does not include the biomarker measurement, usually in the context of a randomized controlled clinical trial.61 Such clinical trials prove the effectiveness of screening and also provide valuable data for cost-effectiveness analyses.
If a biomarker is used to diagnose a potentially life-threatening condition in a symptomatic patient (eg, acute MI in a patient with chest pain), it should have a high sensitivity because a "rule out" (exclude disease) strategy is critical in this setting (also called the SnOUT rule27). Expressed in terms of LR, a test with a lower LR– (typically <0.10) is needed; this is because the costs of missing disease (projecting health when disease is likely) outweigh the costs of any additional testing or a false diagnosis.
Appropriate use of biomarker results requires use of a Bayesian approach,62,63 ie, integrating pretest probabilities with biomarker test results (expressed as sensitivity/specificity or as LR) to estimate the posttest probability of disease. Predictive values use this concept to facilitate interpretation of test results, taking into consideration disease prevalence. Even for a test with high sensitivity and specificity, false-positive tests will outnumber true-positive tests when disease prevalence is very low, and false-negative tests will outnumber true-negative tests when disease prevalence is very high. Pretest probabilities for estimating predictive values may be generated on the basis of the published literature combined with clinical experience. A nomogram is available that uses the pretest probability of disease and the LR of a diagnostic test to generate posttest probability of the condition.64
Biomarkers (whether for screening, diagnosis, or prognosis) are also evaluated in terms of their discrimination and calibration65–67 capabilities. Discrimination refers to the ability of the biomarker (by itself or as part of a composite score) to distinguish "case" from "noncase" in cross-sectional studies or to differentiate "those who will develop disease" from "those who will not" in longitudinal investigations. Typically, the c statistic (or concordance index) is used as the metric of model discrimination and is equivalent to the area under the ROC curve. The c statistic is the probability that in 2 randomly paired individuals (one with and one without disease), a given test correctly identifies the one with disease. It is important to note that the c statistic is a metric of overall performance. It is possible for 2 tests to have the same c statistic, yet one biomarker may be superior to the other in terms of performance at select thresholds.
Calibration tells us how the ability of a biomarker (or a model incorporating the biomarker) to predict risk relates to the actual observed risk in subgroups of the population. The Hosmer-Lemeshow goodness-of-fit statistic is often used as an indicator of model calibration.68 For this purpose, the sample is divided into deciles of risk, and the observed number of events is compared with the expected number of events. Calibration is particularly important in counseling of individuals when the question of interest is the numeric probability of disease in a given patient (rather than how sick they are relative to other persons with disease).67 Thus, risk prediction algorithms have been developed that incorporate select biomarkers and enable clinicians to predict the absolute event rates of disease; examples include estimating the risk of coronary heart disease (CHD) given values of vascular risk factors,69 assessing the risk of death or stroke in patients with atrial fibrillation,70 and appraising the risk of death in patients with established heart failure.71 Models can be recalibrated if they uniformly underestimate or overestimate risk. For example, the Framingham CHD risk score overestimated risk in a Chinese cohort. A recalibration of the risk functions (with the use of mean values of risk factors and mean CHD incidence rates in the Chinese cohort) substantially improved CHD risk prediction.72
|
Evaluation of the Incremental Value of New Biomarkers and the Multimarker Concept |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
To evaluate the incremental value of a new biomarker, investigators must demonstrate the elevated risk of an outcome associated with higher levels of the new biomarker with adjustment for other established risk factors. These results are typically presented as hazards ratios (relative risk estimates from a Cox model) and a probability value test of significance of the marker in the multivariable models. It has been argued that such an interpretation of a new marker’s association with risk as a reflection of its prognostic value may be inappropriate because the "hazard ratio is dependent on the measurement scale of the marker, cutoff(s) used for the novel marker, and the manner in which established variables are modeled."73 In other words, a high hazard ratio for a marker in relation to a disease outcome does not necessarily indicate better predictive performance. Indeed, very strong associations of markers with disease are required for a given biomarker to have good discrimination properties.74 Even when a biomarker threshold is associated with very high odds of disease, it often will identity only a small proportion of people with disease if false-positive rates are to be kept low.74 For example, the relative risk for CHD mortality comparing the top with the lowest decile of the distribution of serum cholesterol was &3 in a large study, indicating a strong association.75 However, if one were to accept a serum cholesterol treatment cutpoint that yields only a 5% false-positive rate (the threshold often used for screening studies), only 12% of the people who would later die of CHD would be identified by that threshold.75 In other words, risk factors for disease may not necessarily make good screening tools.76 This is because, notwithstanding an association of a risk factor with disease, the distributions of the risk factor levels in people with and without disease may overlap substantially.76
When a new biomarker X is evaluated, it is important to remember that the question of interest is not whether X is a better predictor of disease than a previously known biomarker Y.77 Rather, the pertinent question is whether X improves the predictive accuracy of the best available model (representing the standard of care for that disease) that incorporates several known predictors of disease including Y.77 Thus, the relative added values of new biomarkers is best evaluated by estimating the increment to the c statistic compared with that from a model that incorporates other previously known predictors.73,77,78 For example, the Framingham CHD risk score may be thought of as a composite of several biomarkers with a c statistic (a metric of predictive accuracy) varying between 0.74 and 0.76, values considered consistent with a "fair" test performance.69 Few risk factors of interest in terms of CHD risk prediction can enhance the c statistic beyond that provided by the Framingham risk score78; for instance, whereas C-reactive protein (CRP) was associated with vascular risk in 2 separate studies,79,80 addition of CRP did not improve the predictive accuracy of a model incorporating established risk factors that represent the current standard of care. Another way to evaluate novel risk factors is to assess whether knowledge of a putative risk factor alters the probability of disease (eg, changes the risk category from low to intermediate risk) estimated with the use of the global CHD risk score such as to change the recommended target threshold for a modifiable risk factor (eg, change the target low-density lipoprotein cholesterol from 130 to 100 mg/dL).81,82
There is considerable interest in generating multimarker scores that use a composite of several biomarkers (measured in parallel) for the purpose of predicting disease risk and patient outcomes.83–91 The comparison of several putative biomarkers and the generation of multimarker scores must take several factors into consideration. First, comparisons of biomarkers measured on the same set of individuals must account for their inherent correlation (people with high values of one marker will likely have high values of another).92 Second, the incremental utility of adding a new biomarker to a known panel of biomarkers is often estimated by ROC analysis. It is important to realize that the ability of the biomarker to identify cases not captured by the usual sets of predictors (conditional or multi-ROC analysis) requires specification of thresholds for the usual set of markers, and the performance of the new marker is conditional on the choice of those cutpoints.93 Sometimes, a multivariate formulation of several markers can be generated with the use of techniques such as neural networks to increase diagnostic accuracy.94
Risk prediction equations that incorporate multiple markers are often used for CVD risk prediction.69,94–97 The challenges associated with the development and application of such risk scores have been reviewed elsewhere.98 Nonetheless, use of a global risk score based on assessment of multiple risk factors is critical because of their synergistic influences and the importance of targeting undesirable levels of several risk factors to maximize patient benefits.99 Risk scores formulated on the basis of a sample should be demonstrated to be reproducible in the same population (with the use of data resampling techniques such as bootstrapping).67 Additionally, to become a routine part of clinical practice, risk scores should be "transportable": geographically to diverse locations; to different ethnicities; to a wide spectrum of patients; or for predicting events over a different duration of follow-up compared with what was used to develop the score.67 Risk scores derived in one sample may need to be recalibrated when applied to a very different population.
Although it is generally believed that new biomarkers should add to the c statistic to be useful, there are exceptions to this rule.99 Novel biomarkers (eg, homocysteine) that are not incremental to known risk factors may be measured in select clinical situations99,100 such as in the following: asymptomatic individuals without obviously elevated conventional risk factors but with very strong family history of vascular disease; patients with premature vascular disease but no obvious risk factors; and patients with aggressive recurrent vascular disease in the face of well-controlled levels of conventional risk factors.
In the case of studies in which genetic biomarkers are used, there is a major concern about false-positive associations with disease (or phenotypes) resulting from numerous additional factors. A detailed discussion of the factors contributing to the lack of replication of several genetic association tests is beyond the scope of this review but includes true genetic heterogeneity across samples, publication bias, confounding by population structure, misclassification of outcomes, allelic heterogeneity, small sample sizes, and failure to account for multiple testing (including the possibility that findings are due to chance).101,102 Measures to address these issues have been proposed, including but not limited to considering pretest probabilities of associations and using false discovery rates (estimated by permutation or bootstrap methodology).103–105 Replication of findings in multiple independent samples remains the gold standard for genetics of complex diseases.106
|
Evaluation of Biomarker Performance: Laboratory Factors |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
The above discussion of biomarker performance assumes a perfect laboratory and limited biological variability. In practice, preanalytical, analytical, and postanalytical factors are important contributors to biomarker performance. The greater the "noise" introduced by these factors, the lesser the "signal-to-noise ratio" offered by a biomarker.
Preanalytical variability refers to biological variability and stability over time,107,108 whereas analytical variability relates to the performance of the test in the laboratory. Low analytical variability is a fundamental requirement of all biomarkers (Table 3).109–114
|
Guidelines for maintaining quality control within laboratories have been proposed.110 Analytical variability means good accuracy and precision. Accuracy refers to the degree of agreement with a reference standard for the analyte and is quantified in terms of percent bias.114 Standardization of an assay means use of a reference measurement procedure and reference materials.114,115 International reference standards have been established for several biomarkers, including interleukin-6,116 interleukin-8,117 serum amyloid A protein,118 fibrinogen,119 and high-sensitivity CRP.120 Precision refers to consistent measurement on replicates114 and is quantified in terms of the coefficient of variation (continuous markers) or the kappa statistic (qualitative markers). Analytical standards have been proposed for several CVD biomarkers, including lipids,121–124 troponins,40,41 and high-sensitivity CRP.125,126
If analytical imprecision is greater than biological variability, samples should be assayed in replicate, and quality control procedures that improve assay methodology and/or operating procedures should be instituted. This is critical for biomarkers used for point-of-care testing because imprecision may be greater in this setting compared with standard laboratory measurements. If biological variability is greater than analytical imprecision, the patient should be sampled on >1 occasion to obtain a true estimate of a biomarker. Biological variability can also be reduced by instituting a standardized protocol for phlebotomy if applicable (such as the requirement of a fasting state, supine posture, or an early morning specimen). Quality control protocols to enhance analytical precision for imaging studies have been proposed.111 In the case of newer technologies such as genotyping and microarray (discussed in a subsequent section), the possibility of analytical error is of a different order of magnitude. Standards for detecting errors in genotyping112 and in microarrays113 have been proposed as well.
Postanalytical factors affecting biomarker performance include the processes of approval and transmission and the appropriate display of test results with the use of the laboratory’s information management systems. As noted above, the quality control requirements for biomarkers also may vary depending on the mode of delivery/use of the test: an automated platform from a centralized laboratory, a point-of-care testing device, or a device used for home monitoring of analytes. Point-of-care testing usually involves small benchtop analyzers or hand-held devices that facilitate rapid decision making, earlier treatment, reduced incidence of complications, quicker optimization of treatment, reduction in hospital stay, greater patient satisfaction, and economic benefits.127
|
Biomarker Discovery: Challenges and Approaches |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
The development of CVD biomarkers is challenging for several reasons. As summarized in a recent consensus document,13,14 the patient vulnerable to CVD is likely harboring a triad of abnormalities: vulnerable plaque, vulnerable blood, and vulnerable myocardium. In terms of developing biomarkers, 2 of these 3 components (vulnerable plaque and myocardium) are less directly accessible relative to the third (vulnerable blood). In the case of atherosclerotic cerebrovascular disease, biomarkers that may be elevated as a consequence of brain injury may not be detectable in large amounts in the peripheral circulation because of the blood-brain barrier. It is challenging to identify diagnostic biomarkers in the peripheral blood within 3 hours of stroke onset, the critical time window for thrombolytic therapy. Furthermore, biomarkers selected to reflect a clinical phenotype may be confounded by the inaccuracies in the characterization of the phenotype. Conversely, there may truly be a poor correlation between biomarkers and the clinical phenotype itself. For example, in the setting of an acute MI, a majority of culprit ruptured plaques occur in nonstenotic coronary lesions.128 In addition, the process of atherosclerotic CVD is inherently so complex that it would be simplistic to assume that a parsimonious set of biomarkers could capture most of the interindividual variation in propensity to develop CVD or its sequelae.
The aforementioned caveats notwithstanding, 3 parallel developments have revolutionized the field of biomarker discovery. First, the completion of the Human Genome Project129 and the HapMap Project130 and the development of microarrays, proteomics, and nanotechnology together provide new avenues for developing exceptionally informative biomarkers of CVD, including high-throughput, highly sensitive, functional assays. Second, the advances in bioinformatics coupled with cross-disciplinary collaborations (eg, of biologists, clinicians, chemists, computer scientists, physicists) have greatly enhanced our ability to retrieve, characterize, and analyze large amounts of data generated by the technological advances noted above. Third, there is increased recognition that diseases arise out of the dynamic dysregulation of several gene regulatory networks, proteins, and metabolic alterations, reflecting complex perturbations (genetic and environmental) of the "system."131,132 The expectation that single biomarkers can capture these intricate derangements and can unambiguously identify disease or that targeting single molecules or signaling pathways is adequate for treating complex pathology is simplistic. Rather, a "systems biology" approach that investigates multiple components of malfunctioning regulatory networks (referred to as multiparameter analysis of tens of hundreds of molecules) may provide better insights into disease diagnosis, prognosis, and treatment.131
The development of biomarkers in CVD can be thought of as consisting of 2 potential approaches: the first strategy is "knowledge-based" (deductive method), and the second one is more "unbiased" (inductive strategy). These 2 approaches are complementary rather than mutually exclusive. The knowledge-based strategy relies on a direct understanding of the biological processes that underlie the process of atherosclerosis and the evolution of its sequelae. It may consist of improving existing biomarkers to enhance their performance, or it may comprise designing assays for attractive new candidate markers informed by the biology of the disease process. The unbiased approach involves trolling through tens of thousands of molecules with the use of current technological advances to characterize the biomolecular profile of a stage of the disease.
|
Biomarker Discovery: Molecular Biology Tools |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
The systems biology tools applied to biomarker discovery investigate the hierarchical organizational of biological information: the gene itself, the mRNA that it produces, the protein coded by the mRNA, biomodules or networks, cells, organs, individuals, populations, and ecologies.131 Table 4 provides an overview of some key techniques used for identifying putative CVD biomarkers.133 Table 5 defines broadly these "Omics" tools.134–137 Table 6 provides information about some of the mathematical and molecular biological techniques within the "Omics" toolbox.134–140 In the section below, an overview of strategies used to analyze different components of this hierarchical sequence is presented.
|
|
|
Genetic Studies
Genetic biomarkers are variants in the DNA code that alone or in combination are associated with disease susceptibility, disease expression, and disease outcome, including therapeutic responses. Single nucleotide polymorphisms (SNPs; DNA sequence variation when a single nucleotide in the genome sequence is altered) have been evaluated extensively in relation to CVD. The 2 classic complementary approaches used for relating genetic sequence variation to CVD risk are the linkage approach and the association strategy.141
The linkage approach investigates families with a whole genome scan consisting of hundreds of anonymous markers to identify genetic loci that may be related to disease susceptibility. The linkage strategy identifies a segment of the genome (typically involving millions of bases of DNA) that segregates with disease. Fine mapping within these segments may lead to the identification of a gene related to disease susceptibility. To date, the linkage approach has been successful in detecting genes for single-gene disorders with large genetic effects. However, linkage studies have provided very modest yields for investigating complex traits like CVD.142
The association strategy evaluates the relation of genetic variants, typically in unrelated individuals, to the presence versus absence of disease or to variation in values of a quantitative trait.143 The scientific rationale behind association studies is that common genetic variants with modest effects contribute to the variation of complex disease in the population.141,144–146 Association studies have the ability to detect more modest genetic effects (relative to linkage). The recognition that groups of neighboring polymorphisms in the genome are highly correlated147 (in linkage disequilibrium, ie, inherited together and not easily separated by recombination) has led to the concept of tag SNPs, which can be used as proxies for most of the common genetic variants in a region of linkage dysequilibrium.148 The identification of tag SNPs is expected to greatly facilitate association studies because fewer markers need to be genotyped.149 The availability of dense SNP maps of the human genome has also fuelled interest in genomewide association analyses, studies that survey the whole genome for causal common genetic variants with the use of a dense set of SNPs.150–152 It is important to emphasize that use of SNP databases is challenged by constant updating, the need for SNP verification and/or primary resequencing (given sequencing errors and rare or population-specific variants), and variation in the linkage disequilibrium patterns across different populations that can influence the selection of tag SNPs.106
Both linkage153–160 and association161–171 studies have provided valuable insights into genetic markers with a role in pathogenesis of CVD. New putative susceptibility genes for CVD have been identified, including cytokine lymphotoxin-
(LTA, on 6p21.3 for MI), galectin-2 (LGALS2, an LTA-interacting protein on 22q12-q13 for MI), 5-lipoxygenase activating protein involved in synthesizing potent proinflammatory leukotrienes (ALOX5AP, on 13q12-q13 for MI and stroke), phosphodiesterase 4D (PDE4D, on 5q12 for ischemic stroke), and the myocyte enhancer factor 2 (MEF2) signaling pathway of vascular endothelium.172
Gene Expression
The availability of rapid, high-throughput analytical platforms has facilitated molecular phenotyping of disease states by analyzing the transcriptome. The global analysis of gene expression represents a paradigm shift from the traditional single-molecule approach to the evaluation of gene regulatory networks.173–175 The National Heart, Lung, and Blood Institute has launched a multicenter Program in Genomic Applications (http://www.nhlbi.nih.gov/resources/pga/) to advance functional genomic research related to heart, lung, blood, and sleep health and diseases.
Changes in mRNA expression of select genes in tissues can be evaluated by several techniques (such as Northern blotting, RNA differential display, RNase protection assay, and various polymerase chain reaction–based methods including real-time polymerase chain reaction). Quantitative assessments of mRNA expression on a genomewide basis can be accomplished with techniques such as the serial analysis of gene expression176 and DNA "microarrays"173; genes are grouped into expression clusters, and upregulated and downregulated clusters in disease states can be recognized. Expression analysis facilitates recognition of dysregulated gene clusters and the identification of candidate genes for association tests177 and may suggest therapeutic targets. The protein products of highly upregulated genes may be candidate biomarkers if they are secreted extracellularly.
High-throughput sequencing of randomly selected clones from human heart cDNA libraries has been used to generate a compendium of expressed sequence tags.178 A cDNA microarray called the CardioChip (containing 10 368 redundant and randomly selected sequenced expressed sequence tags) has been developed on the basis of human heart and arterial tissue cDNA libraries.179 Gene expression analyses have been performed on myocardial tissue to identify specific patterns in cardiac hypertrophy,180–182 MI,183,184 different forms of heart failure,185,186 and cardiac transplants.187 Such gene expression analysis may enable molecular profiling of patients with dilated cardiomyopathy, including the correlation of therapeutic responses with transcriptional changes.188 On a parallel note, the differential patterns of gene expression in ischemic and nonischemic heart failure subsets may have therapeutic relevance.189,190 Likewise, gene expression profiles of hypertrophic and dilated cardiomyopathy have been demonstrated to be different, thereby providing clues to molecular mechanisms underlying the conditions as well as identifying distinct biomarkers for each condition.191–193 DNA microarrays have also been applied to analyze molecular signatures of atherosclerotic lesions,194–196 vascular endothelial cells subjected to shear stress,197,198 and vascular smooth muscle cells.199 These investigations have provided valuable clues to genes implicated in atherosclerosis,194,195,200 plaque rupture,201 and vascular remodeling. Genomic techniques have also been extended to peripheral circulating blood cells (progenitor cells202 and blood cells203) to evaluate the effect of statins and to identify transcripts that are altered in coronary disease; these observations raise the exciting possibility of using more readily accessible tissues (blood cells) for genomic screening. The utility of expression profiles may be extended to predicting perioperative outcomes in patients undergoing cardiac surgery.204
Whereas global gene expression profiling offers a unique opportunity for molecular profiling of CVD with implications for diagnosis, prognostication, and treatment (including identifying disease subtypes) and for identifying new therapeutic targets, technical and conceptual challenges may limit its use. The technical limitations include the limited number of transcripts on available chips, the possibility of false-positives (emphasizing the need for confirmation of results with an independent approach such as real-time polymerase chain reaction), and the challenge of isolating cell types from heterogeneous cell populations in tissues.174,205 The availability of techniques such as laser capture microdissection has facilitated the isolation of cell populations, however.206 The conceptual challenges lie in the fact that there may be a poor correlation between mRNA expression and the proteome (because all transcripts may not be translated) and with protein function (due to inability to detect alternate splicing, posttranslational modification, subcellular localization, and interactions among proteins that can influence function).207 Additional gain- or loss-of-function studies are necessary for mechanistic interpretations.174,205
Proteomics
Proteomic approaches to the identification of disease biomarkers rely principally on the comparative analysis of protein expression in normal and disease tissues to identify aberrantly expressed proteins that may represent new biomarkers, analysis of secreted proteins (in cell lines and primary cultures), and direct serum protein profiling. Proteomics methodologies include assessment of protein expression (by Western blotting and enzyme-linked immunosorbent assay and by other antibody-based methods) and the isolation, identification, and quantification of proteins in biosamples with high-resolution 2-dimensional gel electrophoresis, high-performance liquid chromatography, surface chromatography by adsorbion of proteins to activated surfaces (matrix-assisted laser desorption–ionization technology), or via peptide ionization procedures and mass spectroscopy. Mass spectrometry can yield a comprehensive profile of peptides and proteins in biosamples without the need for initial protein separations, thereby facilitating biomarker identification with reduced sample requirements and a high throughput.
The parallel development of a human protein reference database (Human Proteome Organization; www.HUPO.org) has enabled the annotation identification of proteins detected in biosamples. The Human Proteome Organization initiative includes the mapping of proteomes in biological compartments such as the plasma, urine, brain, liver, and heart.208,209 Protein profiling with the use of multidimensional automated platforms is interfaced with database search tools to facilitate the rapid identification of constituent proteins. An important caveat in the use of proteomics is that biomarkers identified by such technology may not be consistent with those generated from mRNA expression profiling.
Proteomic databases of cardiac proteins have been constructed,210–212 and alterations of several cardiac proteins have been described in both experimental and human cardiomyopathies.213 For instance, the upregulation of ubiquitin carboxyl-terminal hydrolase in experimental214 and human215 cardiomyopathic tissues is consistent with the notion that inappropriate ubiquination and proteolysis of select cardiac proteins may play a role in promoting ventricular systolic dysfunction in human heart failure.213 Several programs of the NIH support proteomics technology development, and an NIH Roadmap initiative emphasizes the importance of studying dynamic systems.216 Advances in computational biology have facilitated computer-based sophisticated cellular and whole organ modeling of the various protein-protein interactions to reconstruct the physiological processes in the heart.217
Molecular Imaging
Noninvasive molecular imaging can enable clinicians to quantitatively identify the causative molecular constituents of disease in time and space. Molecular imaging will likely facilitate targeted therapy of CVD on the basis of the molecular elements delineated in diseased tissue.218 For example, newer targeted contrast agents are being developed for plaque characterization: "by identifying fibrin within plaque microfissures,219 adhesion or thrombogenic molecules expressed on endothelium of vulnerable plaques,220 matrix metalloproteinases in the cores of progressive lesions,221 or the early angiogenic expansion of the vasa vasorum that supports plaque development."222,223 Plaques that look morphologically similar (in terms of lipid core and fibrous cap) may be distinguished with techniques such as thermography,224 multicontrast MRI,219,225 and intravascular optical coherence tomography.226
|
Biomarker Development: The Processes From Discovery to Delivery |
|---|
|
TopIntroductionWhat Is a Biomarker?...Characteristics of an Ideal...Defining Abnormal Biomarker...Evaluation of Biomarker...Evaluation of the Incremental...Evaluation of Biomarker...Biomarker Discovery: Challenges...Biomarker Discovery: Molecular...Biomarker Development: The...Currently Available CVD...Cardiovascular Biomarkers:...ConclusionsReferences |
|---|
Figure 2 displays the various stages from the discovery of a biomarker in a laboratory with the use of the "Omics" technologies to development of an assay and finally to its delivery, ie, application in clinical practice.227 Briefly, the process begins with the identification of target biomarkers with the use of standardized technology platforms, followed by validation of the assays,228,229 statistical evaluation of biomarker distributions in reference samples and in those with disease, and assessment of the correlation between biomarker levels (or expression patterns of biomarkers) and clinical measurements that define disease status.227
|
