Author + information
- ↵⁎Reprint requests and correspondence:
Dr. Robert M. Califf, Duke Translational Medicine Institute, Duke University Medical Center, DUMC Campus Box 3850, Durham, North Carolina 27710
- biological markers
- surrogate end point
- acute coronary syndrome
- brain natriuretic peptide
- B-type natriuretic peptide
In this issue of the Journal, Morrow et al. (1) from the TIMI Group, provide a provocative update of the MERLIN–TIMI 36 (Metabolic Efficiency With Ranolazine for Less Ischemia in Non-ST Elevation Acute Coronary–Thrombolysis In Myocardial Infarction 36) trial. Their report raises the possibility that a single measurement of plasma concentrations of B-type natriuretic peptide (BNP), taken early in the hospital course of patients with acute coronary syndrome (ACS), not only can identify a cohort at higher risk of poor clinical outcomes, but may also differentiate patients more likely to experience a reduction in recurrent cardiac events with ranolazine from those who would not benefit from ranolazine.
Studies such as this represent an attempt to break a path into the relatively uncharted territory of “personalized medicine” (2). Although the term has different meanings for different people, most “sales pitches” for personalized medicine allude to the incorporation of biomarkers into clinical research and practice, a practice that promises to distinguish those who will benefit from a proposed treatment from those who will not (or who may even be harmed). Because this practice divides populations into strata, we prefer the term stratified medicine, in which the personalized component revolves around the inclusion of the patient's desires and values when making choices about medical care.
A biomarker is defined 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 (3). Biomarkers have multiple uses in the arenas of research and practice. In the development of medical technologies, biomarkers can be used to identify preclinical models and human subjects with a pathophysiology of interest and to assess the effects of an intervention on both on- and off-target biological systems. In clinical practice, a biomarker may be used to diagnose a medical problem, serve as a tool for staging disease, or provide an indicator of differential prognosis (4). As noted earlier, biomarkers become especially powerful when they allow the identification and stratification of patients into subpopulations who will derive more or less benefit from an intervention (5). Under such circumstances, biomarkers can be used to predict or monitor response to a treatment.
A particularly cherished pipe dream of many involved in the development of medical therapeutics is one in which numerous biomarkers can be pressed into service as surrogate end points. Despite well-written treatises underscoring the shortcomings of this concept (6), it has proven difficult to convey the notion that only a very few biomarkers will actually be established as useful and valid surrogates for therapeutic benefit. Patients and their families care about clinical end points, which are measures of how a person feels, functions, or survives. A biomarker, however, approaches true surrogate status only when a change in the biomarker (the putative surrogate) attributable to an intervention approximates a complete prediction of the change in the clinical end point, a circumstance that rarely, if ever, occurs. Further, a treatment ultimately should be recommended on the basis of the net balance of benefits and risks in light of all possible clinical outcomes, but a perfect biomarker for a particular clinical end point cannot possibly be informative about all other outcomes of interest.
Additionally, the promise that deploying biomarkers in clinical trials would reduce the number of subjects needed has proven illusory. It is true that individual trials may be conducted with smaller numbers of subjects if the target population can be identified before the experiment. For example, if a biomarker is thought to predict a major differential benefit from treatment, then by preferentially enrolling patients on the basis of levels of that biomarker, such benefit may be shown in a smaller study population. Several important caveats apply, however. First, the benefit can be claimed specifically for biomarker-positive patients only if biomarker-negative patients are also studied, so that the contrast in benefit-risk balance is proven. Second, when both biomarker-positive and biomarker-negative patients are studied, the difference in treatment effect is typically not “all or nothing.” We are then obliged to conduct a series of studies to determine whether a particular subpopulation (or multiple subpopulations) exhibits a benefit-risk balance that justifies giving or withholding treatment on a more continuous scale. Finally, identifying a subpopulation that derives differential benefit implies that those who do not benefit warrant further evaluation, if effective therapies are to be developed for that group. When all these caveats are taken into account, the total number of subjects needed for clinical trials is likely not only to increase, but perhaps to increase by as much as a log order!
The developmental pathway for biomarkers intended for use in clinical research or practice is equally complex. Some have proposed that we should apply a structured conceptual framework to the development of useful biomarkers. For instance, one notion that has proved enticing is that, as is the case in drug development, researchers could move from a fundamental discovery science mode to development of biomarkers in nonhuman models, then into early-phase human studies (which might include both intensive small-scale and epidemiological studies), eventually to be followed by confirmatory studies.
Although such a linear approach is satisfying to the orderly mind, several major trends have turned this neatly compartmentalized world on its head. First, we have come to recognize that human biology cannot be reduced to a series of linear mechanisms; rather, it comprises a vast network of interwoven pathways and systems. This recognition has led to the concept of systems biology, which is built in part on the assumption that much future discovery will depend on mathematical constructs that elucidate the various relationships among multiple pathways, an enterprise that requires the study of higher-order systems (e.g., intact animals and people).
Second, modern technology now allows us to measure genes, gene expression, proteins, and metabolites in human populations in great detail using unbiased discovery platforms. Such unbiased analyses will inevitably lead to the identification of many genes, proteins, and metabolites, the exact biological functions of which are currently unknown. Indeed, when biomarkers are used for clinical prediction of prognosis or treatment response, it is not necessary to understand their underlying biology (although such understanding would always be preferable) because their association with disease, adverse prognosis, or response to therapy will provide clues about new therapies and biomarkers (7). In such an environment, the fluid interplay between classical mechanistic biology, systems discovery, and clinical application becomes crucial.
Third, we continue to struggle with the complexity of human decision making. A question that appears to be simple on the surface—if a patient has ACS and a high BNP level, should we use ranolazine?—becomes dauntingly complex when an array of related information about concomitant treatments, patient preferences, and heuristics comes into play (8). All of these complexities and uncertainties boil down to a binary decision: to treat or not to treat. The need to distill simple decision rules, however, must not dissuade us from elucidating the full complexity of biomarkers during the research phase.
Where does the current report by Morrow et al. (1) fall on this spectrum of thinking? The authors carefully collected samples in the context of a randomized clinical trial and analyzed BNP concentrations in light of an a priori hypothesis that ranolazine may have a specific effect on wall stress in the myocardium and that patients with higher wall stress as measured by BNP might benefit preferentially from treatment. The authors' previous work suggesting that BNP measurement identifies a high-risk population of ACS patients was confirmed. More exciting, however, was the finding (supported by a rigorous test for interaction between treatment assignment and BNP level), that ranolazine reduced the risk of poor outcomes in subjects with a BNP >80 pg/ml, but not in subjects with lower BNP concentrations. The authors appropriately note that the overall trial showed no significant reduction in cardiovascular events with ranolazine; therefore, technically, the test for a differential benefit as a function of BNP concentration could not be conducted while maintaining a measurable false-positive rate.
Although this report does not provide definitive evidence that ranolazine should be used in patients with ACS and elevated BNP levels, it nonetheless represents a carefully conducted study with findings that are sufficiently persuasive to warrant a prospective clinical trial. Concomitantly, several other issues raised by this report deserve consideration in the context of studies evaluating the role of biomarkers as clinical tools. First, the specific role of BNP measurement in the setting of ACS could best be defined by an exploration of the likelihood of events and treatment effect as a function of the continuous measurement of BNP, rather than settling on a binary cut point too early in the exploration phase. In this case, a BNP threshold level of 80 pg/ml may be optimal for discriminating risk of death or heart failure, although the authors did not display the raw or modeled data that would allow readers to judge the appropriateness of the cutoff. The optimal threshold for treatment benefit with ranolazine may also be different. In addition, given the biological complexity of ACS and related downstream clinical events, the appropriate place of BNP measurement in a “multimarker strategy” needs to be defined. Finally, the ultimate test of the utility of BNP, as for any biomarker, would be to randomly assign patients to have the biomarker measured or not, with subsequent pharmacological therapy with ranolazine determined by BNP levels in the biomarker arm and with clinical outcome as the dependent variable. The TIMI investigators have made significant strides toward the biomarker bonanza, but much work remains before we can reap the rewards.
Dr. Shah has received research funding through an unrestricted grant to Duke University from Medtronic Inc. Dr. Newby has received a research grant from AstraZeneca, GlaxoSmithKline, and Medicure; and consulting fees from AstraZeneca, Biosite Inc., CV Therapeutics, Eli Lilly and Co., Heartscapt Technologies, Mosby, Roche Diagnostics, Schering-Plough, and Scios.
↵⁎ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
- American College of Cardiology Foundation
- Morrow D.A.,
- Scirica B.M.,
- Sabatine M.S.,
- et al.
- Ginsburg G.S.,
- Donahue M.,
- Newby L.K.
- Califf R.M.
- Hlatky M.A.,
- Greenland P.,
- Arnett D.K.,
- et al.