Chronic obstructive pulmonary disease (COPD) is a highly heterogeneous disease with limited adequate treatments. Biomarkers—which may relate to disease susceptibility, diagnosis, prognosis, or treatment response—are ideally suited to dissecting such a complex disease and form a critical component of the precision medicine paradigm. Not all potential candidates, however, make good biomarkers. To date, only plasma fibrinogen has been approved by regulatory bodies as a biomarker of exacerbation risk for clinical trial enrichment. This review outlines some of the challenges of biomarker research in COPD and highlights novel and promising biomarker candidates.
Chronic obstructive pulmonary disease (COPD) is a heterogeneous disease, and better targeted therapy is warranted.
Biomarkers may be categorized by their primary role or purpose: susceptibility, diagnostic, prognostic, or therapeutic.
Biomarkers that predict a therapeutic response may be used to guide treatment and form part of the precision medicine paradigm.
Plasma fibrinogen, which predicts high risk of acute exacerbations, currently is the only biomarker approved by the Food and Drug Administration for use in COPD.
There are any number of associations between potential biomarkers and COPD outcomes, but few meet the characteristics of a good biomarker.
Chronic obstructive pulmonary disease (COPD) is a progressive disease of the lung characterized by irreversible airflow obstruction, persistent airway inflammation, and recurrent acute exacerbations. There is considerable heterogeneity in the susceptibility, inflammatory profiles, clinical presentations, long-term trajectories, and treatment responses among people with COPD. It, therefore, is not surprising that a 1-size-fits-all approach to managing COPD has led to limited progress in modifying the natural history of this disease. Biomarkers may have a role in dissecting this heterogeneity, from aiding diagnosis and early disease detection through to risk assessment and targeted treatment. This review outlines the characteristics of useful biomarkers while highlighting some of the challenges of biomarker research in a complex disease, such as COPD.
The move toward precision medicine
Randomized controlled trials, which sit at the pinnacle of the evidence-based medicine pyramid, aim to minimize the effects of interindividual variability by randomly allocating large numbers of people to 2 or more groups distinguishable only by the intervention they receive (eg, treatment or placebo). The reported trial outcomes, therefore, represent the average response to an intervention among an average group of patients. Considerable heterogeneity of treatment response, however, is likely to exist within the allocated groups. In order to maximize benefit and minimize harm for the individual, a more targeted approach clearly is necessary.
The National Institutes of Health (NIH) define precision medicine as “an approach to disease prevention and treatment that takes into account individual variability in genes, environment and lifestyle.” Under the precision medicine paradigm, heterogeneity is not minimized or ignored but instead is harnessed or embraced in order to provide a more targeted approach to patient care. Such an approach may maximize not only the benefit:risk ratio (which is an implicit part of precision medicine) but also the benefit:cost ratio. Targeting treatment toward patients who will gain the most benefit is necessary if health care systems are to be sustainable. Biomarkers, which help facilitate better-targeted therapy, will play a crucial role in the transition to a precision medicine future ( Fig. 1 ).
What is (and what is not) a biomarker?
A Broad Definition of Biomarkers
The NIH defines 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.” This broad definition is attractive because it does not prescribe what form a biomarker must take. That is, a biomarker is not restricted to molecules or cells but can be any measurable quality—as long as it meets the criteria of being objective and is demonstrably related to biological processes. Biomarkers of relevance to the respiratory system, therefore, may include cellular, molecular, or microbiological measurements from biological samples (blood, sputum, and bronchoalveolar lavage); physiologic measurements (arterial blood gases, pulse oximetry, and lung function tests); medical imaging (computed tomography [CT] and magnetic resonance imaging [MRI]); and clinical prediction rules.
Categories of Biomarkers
Rather than grouping biomarkers by the type or source of the measurement, it is more useful to consider them based on their purpose or utility ( Fig. 2 , Table 1 ). This provides a useful framework for identifying gaps in knowledge in any given disease and for structuring biomarker research.
|Established Biomarkers or Most Viable Candidates||Associated but Less Viable Candidates||Promising Novel Candidates|
|Susceptibility biomarkers||Indicate the likelihood of developing a disease in individuals who do not yet show pathologic changes or clinical signs/symptoms||Aggressive surveillance |
Targeted preventative strategies
Enrichment of cohort studies/clinical trials of preventative interventions
|Diagnostic biomarkers||Assists diagnosis in individuals who already have pathology or clinical signs/symptoms of the disease||Surrogate tests to avoid complex/invasive diagnostic tests (eg, surgical bioposy) |
Increasing diagnostic certainty
Early-stage disease detection
Surrogate for disease severity
Monitoring disease progression over time
|Prognostic biomarkers||Indicate the likelihood of a particular event, outcome or trajectory in a person who already has the disease||Identify individuals at high risk of adverse outcomes—initiate aggressive treatment |
Enrichment of clinical trials with high-risk subjects
|Therapeutic biomarkers||Identify individuals in whom a biological response has occurred after a treatment/intervention (response biomarker) or those who are likely to respond favorably/unfavorably to a treatment (predictive biomarker)||Determine the success of treatment |
Targeted therapy—maximize benefit:harm ratio
What Makes a Good Biomarker?
Within a broad conceptual framework of biomarkers, where virtually any objectively measured characteristic could meet the definition, it is pertinent to consider why some biomarkers might be considered useful, whereas many are not.
Biological plausibility—there should be a strong, consistent, and independent relationship between the biomarker and the disease, either its pathology/physiology or its associated clinical outcomes. Although there need not be evidence of a direct, causal link between a biomarker and the disease, it should at least be plausible that the biomarker is reflecting the disease pathophysiology.
Test performance—biomarker measurements may take many forms, but the test should be reliable and repeatable, with high sensitivity and specificity for the outcome.
Confounders—the biomarker association with disease or treatment outcomes should be free from confounding influences unrelated to the disease itself. Known confounders (eg, active cigarette smoking) may be adjusted for in statistical models.
Treatment outcomes—for a response biomarker, the measured change in the biomarker should be related directly to a change in clinical state. For a predictive biomarker, the predicted response should be a measurable and clinically relevant outcome, taking into account the minimal clinically important difference.
Simplicity—in many cases, biomarkers may act as surrogates of pathology and thus avoid invasive diagnostic testing. A biomarker may not be useful in practice, however, if its measurement is excessively complicated or equally invasive. An ideal biomarker would be easy to obtain, measure, and interpret and would carry some meaning to a clinician.
The current (and future) landscape of biomarkers in chronic obstructive lung disease
Only 1 biomarker—fibrinogen—so far has been approved under the US Food and Drug Administration (FDA) Biomarker Qualification Program. This program applies rigorous standards to the evaluation of biomarker candidates for a specific context of use. Fibrinogen, therefore, is discussed on its own. For the remainder of this section, other biomarker candidates are discussed, including why they may or may not be considered useful. Our aim is to provide examples that illustrate some of the more novel – and challenging – aspects of biomarker development in COPD, without necessarily endorsing individual biomarker candidates.
Plasma Fibrinogen: The Early Qualifier
Fibrinogen is a plasma glycoprotein that is an essential part of the blood coagulation cascade and is an acute phase reactant. One of the first reported associations between plasma fibrinogen and COPD was by Alessandri and colleagues, who found that plasma levels were increased in COPD patients, independent of smoking. Subsequently, large general population studies showed that higher plasma fibrinogen was associated with lower percentage of predicted forced expiratory volume in 1 second (FEV 1 ), faster rate of FEV 1 decline over time, and incident COPD. Given the nonspecific nature of increased plasma fibrinogen, however, it was unlikely to ever be established as a diagnostic biomarker capable of identifying at-risk individuals within a general population.
Plasma fibrinogen is more useful as a prognostic biomarker. It is increased in COPD patients with a high rate of moderate and severe exacerbations, and high levels are associated with increased risk of death. High plasma fibrinogen, therefore, is considered predictive of these clinically relevant outcomes, although the effect size is small in comparison to that of a history of prior exacerbation, which remains the strongest predictor of exacerbation risk. In contrast to the findings in the general population, it has not been reliably associated with FEV 1 decline in COPD cohorts.
In light of these findings, the FDA has approved plasma fibrinogen as an “enrichment biomarker for clinical trials” that can be used to select participants at high risk of frequent exacerbations and morality, thus increasing statistical power. For example, the FDA analysis suggests that enrichment based on a high plasma fibrinogen level could reduce the required sample size by 12%. The recommendation comes, however, with several caveats, including a specified threshold for fibrinogen-high participants (350 mg/dL), the use of a single type of assay (K-Assay, Kamiya Biomedical Company, Seattle, WA, USA), and that it should be used only as a complement for other clinical assessments of risk.
Potential Susceptibility Biomarkers for Chronic Obstructive Lung Disease
The recognition that up to one-third of all COPD occurs in never-smokers has heightened interest in the genetic susceptibility to COPD. Genomic markers of COPD susceptibility are attractive because the genome is fixed over a lifetime; it is not altered by environmental exposures known to be associated with COPD but does give rise to important gene-environment interactions. , The only consistent monogenic association with COPD is mutation in the SERPINA1 gene causing alpha-1 antitrypsin deficiency, which is the subject of a dedicated review in this issue of Clinics . The heterogeneous clinical expression of SERPINA1 mutations, however, limits its viability as a genetic susceptibility biomarker of COPD in general populations.
Hundreds of other genetic loci have been linked to adult lung function in genome-wide association studies (GWAS). Few of the individual loci have been replicated in independent cohorts, and there is considerable overlap with other lung diseases, including asthma and idiopathic pulmonary fibrosis. No single gene or polymorphism would qualify as a biomarker of COPD susceptibility due to both small effect size and a lack of specificity. In the largest GWAS for COPD to date (>400,000 individuals), a genetic risk score (GRS) combining 279 genetic variants associated with lung function was used to assess COPD risk. A GRS in the top decile conferred an odds ratio of 4.73 for COPD; overall, 54% of all COPD cases could be attributed to genetic architecture alone. The use of GRSs for predicting future disease risk is controversial, but the complex polygenic nature of COPD makes this a promising way forward for identifying at-risk individuals.
There is some evidence that variation in airway structure may be a useful biomarker of COPD risk. In the Multi-Ethnic Study of Atherosclerosis general population cohort, airway branch variants determined by thoracic CT were present in 26% of participants, and the most common variant was associated with greater respiratory symptoms, chronic bronchitis, and a 40% increased risk of COPD. In the same cohort, a GRS for COPD was associated with lung structure (luminal diameter, total small airway count, and lung density) independent of the effects of smoking. Adjustment for these lung structural findings attenuated the association between the GRS and lung function, suggesting that the genetic determination of COPD risk may be explained partially by variation in lung structure. How and why anatomic variation, present from the time of lung development in utero, predisposes to COPD later in life is a matter for speculation.
Potential Diagnostic Biomarkers
Spirometry is insensitive to the narrowing and loss of small airways, which occur early in the natural history of the disease. , A significant number of people with normal spirometry report symptoms, experience exacerbation-like events, and have abnormal ventilatory mechanics during exercise—representing a form of pre-COPD. Hence, there are a significant number of patients with mild or early-stage pathology who are not diagnosed by the current gold standard test. A useful diagnostic biomarker, therefore, would detect this pre-COPD, prior to the onset of spirometric abnormalities.
Physiologic tests other than spirometry may be promising diagnostic biomarkers for pre-COPD. For example, diffusing capacity for carbon monoxide (DLCO) may be abnormally low despite normal spirometry and thus represent a form of pre-COPD. In a large cohort of active smokers with normal spirometry, however, only one-quarter had reduced DLCO at baseline and, of these, only 22% went on to develop spirometrically defined COPD after approximately 4 years of follow-up, meaning it may not have sufficient sensitivity to be a useful diagnostic biomarker. Similar findings have been reported for an increased residual volume to total lung capacity (RV/TLC) ratio in the presence of normal spirometry, which confers a 30% increased risk of progressing to overt COPD.
Inert gas washout techniques, such as the multiple breath nitrogen washout (MBNW) test, are particularly sensitive to changes in the lung periphery. MBNW measures ventilation heterogeneity (VH) arising from the uneven branching structure of the airways. Up to half of current smokers with normal spirometry show increased VH on MBNW which does not completely normalize after smoking cessation, suggesting irreversible pathology. Recent computer modeling experiments show that increased VH on MBNW testing can be explained purely by removal of acinar units, which simulate the loss of terminal bronchioles described by Hogg and colleagues more than 50 years ago. It, therefore, is highly likely that increased VH on MBNW is detecting some form of pre-COPD, but whether or not it represents a good diagnostic biomarker requires further clinicopathologic correlation.
Forced oscillometry, where low-amplitude pressure oscillations are applied to the lung via the mouth, also is sensitive to changes in the small airways. Forced oscillometry measurements are abnormal in approximately half of smokers with normal spirometry, but long-term evidence supporting this tool as a biomarker of pre-COPD currently is lacking.
Imaging biomarkers may prove useful for detecting changes that represent pre-COPD. Kirby and colleagues performed hyperpolarized helium-3 MRI on ex-smokers with normal spirometry and no emphysema on thoracic CT. Subjects with DLCO less than 75% predicted had MRI evidence of airspace enlargement (measured by the apparent diffusion coefficient [ADC]) suggesting subtle emphysematous changes. Importantly, the increased ADC correlated with both symptom scores and 6-minute walk test distance walked. In more recent work, the same investigators found that machine learning analysis of apparently normal thoracic CT scans could detect subtle changes that differentiated between normal and abnormal DLCO subjects with 80% accuracy. Further development of such findings as useful diagnostic biomarkers will require replication in larger cohorts with long-term follow-up.
Potential Prognostic Biomarkers
Lung function decline
Long-term observational data suggest that approximately half of people who develop COPD had normal FEV 1 prior to age of 40 years followed by rapid FEV 1 decline, whereas the remaining half had low (<80% predicted) FEV 1 prior to the age of 40 followed by steady/normal FEV 1 decline. Furthermore, genetic signals associated with cross-sectional lung function are not necessarily predictive of longitudinal change in lung function. This suggests that lung development may be just as relevant as subsequent lung function decline for the development of COPD and may explain why good biomarkers of FEV 1 decline in COPD so far have been elusive.
Several potential blood biomarkers of FEV 1 decline have been explored in longitudinal COPD cohorts. In the Evaluation of COPD Longitudinally to Identify Predictive Surrogate End-points (ECLIPSE) study, reduced serum level of the anti-inflammatory pneumoprotein club cell secretory protein 16 (CC-16) was associated independently with FEV 1 decline in participants with COPD. A similar finding was reported in the Lung Health Study (LHS). Reduced serum surfactant protein D (SP-D), another pneumoprotein with an important role in lung homeostasis and innate immunity, also was associated independently with FEV 1 decline in the LHS. The association between these proteins and FEV 1 decline remained significant when analyzed under the mendelian randomization framework, which suggests causal roles in COPD progression. , The individual contribution of these factors to the rate of lung function decline in COPD is small. Combining biomarkers may be more advantageous: in the COPDGene Study cohort, a combination of CC-16, fibrinogen, and soluble receptor for advanced glycation end products (sRAGE), however, was the best biomarker predictor of FEV 1 decline, but it explained only an additional 6% of the variance compared with clinical factors alone. By comparison, a predictive model based on clinical and physiologic factors (available as a Web-based calculator at resp.core.ubc.ca/ipress/FEV1Pred ) explained 88% of the variance in FEV 1 decline.
The strongest known predictor of future exacerbations is a prior history of exacerbations, and any potential biomarker of exacerbation risk would need to perform better than this simple clinical history. In order to further dissect the heterogeneity of exacerbation risk, Adibi and colleagues recently developed and externally validated a tool for predicting individual exacerbation risk based on routinely available clinical information. The model predicted exacerbation frequency of greater than or equal to 2 per year with an area under the receiver operating characteristic curve of 0.81, with close agreement between the predicted and observed exacerbation rates. Importantly, the model performed well even when limited to people with a prior exacerbation history. Although it may not be considered a biomarker in the traditional sense, a clinical prediction tool based on objectively measured factors may fulfill many of the criteria of a good biomarker and may be useful particularly for determining individualized exacerbation risk or enriching clinical trials.
Several blood biomarkers of long-term exacerbation risk also have been of interest:
Peripheral blood eosinophil count has been associated with increased exacerbation rate in some studies , and was found to have marginal additional predictive value in people experiencing frequent exacerbations in the ECLIPSE and COPDGene studies. The FDA, however, has rejected blood eosinophils as a predictive biomarker for frequent exacerbations, on the basis of heterogeneous study designs and eosinophil count thresholds.
An investigation of multiple blood biomarkers related to inflammation (inflammome) found that persistence of this inflammatory signature doubles the risk of exacerbations even after adjusting for prior exacerbation history. The same study showed, however, that the inflammome is not necessarily stable over time and it is unknown whether or not exacerbation risk changes along with inflammatory status.
Low serum level of immunoglobulin G (IgG) has been found to increase the risk of COPD exacerbations and hospitalizations by up to 40% and 92%, respectively, in a concentration-dependent manner. This is a promising finding that may identify a particularly at-risk group who may benefit from a specific therapy (IgG replacement), but the nonspecific nature of IgG deficiency and the likelihood of confounding factors mean serum IgG may not turn out to be a good biomarker.
There are few studies investigating biomarkers of short-term (ie, imminent) exacerbation risk. Such studies are difficult because they require constant monitoring or sampling of patients over a long period of time. One potential biomarker, however, is the variability in physiologic indices: Zimmermann and colleagues reported that the variability in forced oscillometry measurements, recorded daily through home telemonitoring, increased in the days prior to an acute exacerbation and was more sensitive than the change in symptoms. This is an intriguing use of technology to help predict exacerbations at an individual level, particularly because it may offer the chance for timely and personalized intervention.
The causes of mortality in COPD patients vary according to disease severity. Fletcher and Peto acknowledged the association between low FEV 1 and mortality in their seminal work, and the addition of other clinical factors in composite scores, such as body mass index, airflow obstruction, dyspnea, and exercise capacity (BODE) index increases the predictive power. The number of comorbidities also influences the likelihood of mortality in COPD. It is perhaps not surprising that many of the biomarkers investigated to date do not add much additional predictive value over and above that of clinical factors. In the COPDGene study, the best-performing biomarker (SP-D) explained only an additional 2% of the variance in mortality compared with 56% for clinical factors alone.
More personalized approaches that identify subpopulations at a high risk of mortality and who may benefit from a specific intervention would be more beneficial. For example, Leitao Filho and colleagues studied the lung microbiome of COPD patients hospitalized for acute exacerbations and found that the presence of Staphylococcus sp and the absence of Veillonella sp in sputum conferred a markedly increased risk of death within 1 year of follow-up. Whether or not altering the microbiome can influence its association with mortality remains to be seen.
Potential Therapeutic Biomarkers
Because an entire article in this issue of Clinics is dedicated to the treatment of COPD, a brief discussion of some of the more interesting predictive therapeutic biomarkers is presented here.
Inhaled bronchodilators are the mainstay of maintenance therapy in COPD. Although a large proportion of COPD patients significantly increase their FEV 1 after inhaled bronchodilator, the magnitude of change in FEV 1 correlates poorly with clinical outcomes and, therefore, is not considered a useful predictive therapeutic biomarker for these medications. This may reflect the insensitivity of the test rather than poor correlation between the physiology and clinical outcomes. Forced oscillometry may be more useful as a predictive therapeutic biomarker: increased respiratory system impedance is correlated with gas trapping and hyperinflation and may predict a more significant deflation response to inhaled bronchodilators. This physiologic change is more relevant to clinical outcomes, such as symptoms and exercise capacity. This novel use of forced oscillometry for guiding bronchodilator therapy is yet to be tested, however, in a clinical trial setting.
There is great interest in peripheral blood eosinophil count as a predictive biomarker of response to inhaled corticosteroid (ICS) therapy. In a meta-analysis of clinical trials of triple therapy (inhaled long-acting beta-agonists and antimuscarinics, with or without ICS), Cazzola and colleagues found that the number needed to treat with ICS to prevent 1 exacerbation per year reduced from 38 to 8 when restricted to patients with blood eosinophil count greater than or equal to 300 cells/μL. A majority of clinical trials that have undergone (predominantly post hoc) subanalysis by blood eosinophil level have used a 2% cutoff to demonstrate a superior reduction in exacerbation rate with ICS therapy. There currently is no consensus, however, on what the appropriate threshold should be. This is part of the reason given by the FDA in their rejection of blood eosinophils as a predictive therapeutic biomarker.
The use of the acute phase reactant C-reactive protein (CRP) to guide antibiotic therapy for COPD exacerbations recently has been explored in both inpatient and community settings. In these studies, patients presenting with acute exacerbations were randomized to receive antibiotics based on an elevated serum CRP level or based on usual care (ie, patient symptom–driven antibiotic prescription). In both studies, the CRP-guided treatment strategy was associated with significantly reduced antibiotic use without significant differences in clinical outcomes or treatment failures. , Although it may not predict a superior therapeutic response, CRP may be useful as a biomarker that prevents unnecessary antibiotic prescription and maximizes the benefit:harm ratio, which is consistent with the precision medicine paradigm. One of the major challenges of developing a biomarker-guided strategy, however, is clinician acceptance or beliefs. For example, a similar strategy using serum procalcitonin was investigated for antibiotic treatment of lower respiratory tract infections but did not reduce antibiotic use compared with usual care. In this study, clinicians often broke with the treatment protocol and prescribed antibiotics despite a low procalcitonin level: acute COPD exacerbation was one of the most common reasons given by the physicians for ignoring the treatment protocol.
Biomarkers might be beneficial for guiding nonpharmacological therapies. The National Emphysema Treatment Trial showed that in patients undergoing lung volume reduction surgery, upper lobe–predominant emphysema on CT scan predicted a reduction in mortality compared with medical treatment in patients with poor exercise capacity. Similarly, heterogeneously distributed emphysema appears to predict the success of bronchoscopic lung volume reduction. Central to the success of this treatment is the presence of intact interlobar fissures, which minimize collateral ventilation between the target and adjacent lobes and allow the targeted lobe to collapse. , Assessment of fissure integrity can be performed by multiple methods, most commonly CT or endoscopic methods (eg, the commonly used Chartis system). An adjacent interlobar fissure that is greater than 90% complete is associated with treatment success in up to 65% of patients undergoing bronchial valve placement; fissures that are less than 90% complete confer a very low chance of treatment success. Preprocedural estimation of treatment success is critical when the intervention carries significant risk, and biomarkers may play an increasing role in determining the benefit:risk ratio for each individual.
There is intense interest in developing biomarkers in COPD, driven largely by the lack of progress in modifying clinical outcomes in this highly complex and heterogeneous disease. There are any number of associations between genetic, physiologic, biochemical, or radiological factors and COPD outcomes. Biomarker development is about taking these findings beyond mere association and exploring their potential roles in the context of personalized medicine. For example, a biomarker of COPD susceptibility, exacerbation risk, or mortality is of benefit only if it is used to identify a subpopulation for targeted intervention. Similarly, a predictive therapeutic biomarker is useful only if it reliably identifies patients who would gain clinically meaningful benefit or be saved from unnecessary harm. For these reasons, the authors believe the main role for COPD biomarkers in the foreseeable future will be in predicting treatment response and enriching clinical trials for high-risk patients. Biomarker research in COPD needs to move beyond discovery and instead focus on identifying and developing the most viable candidates. This undoubtedly will require ongoing collaboration and data sharing between the large, longitudinal COPD cohorts.
There were no direct financial sponsors for the submitted work. S. Milne reports personal fees from Novartis, Boehringer Ingelheim, and Menarini and nonfinancial support from Draeger Australia, outside the submitted work. D.D. Sin reports grants from Merck ; personal fees from Sanofi-Aventis, Regeneron, and Novartis; and grants and personal fees from Boehringer Ingelheim and AstraZeneca , outside the submitted work.