Biomarkers are measurable properties of a patient believed to be predictive of a particular clinical status. This can include the susceptibility to an illness or the likelihood of benefit or adverse reaction to a given treatment. Biomarkers can thus be powerful tools to guide diagnosis and treatment. They are also potentially transformative for the translational research enterprise. Biomarkers, such as reductions in lipid levels for cardiovascular disease or progression-free survival for cancer, can be employed as endpoints in clinical trials to evaluate investigational drugs. Since biomarker endpoints can typically be measured more quickly than “hard” endpoints like patient survival, when all goes well, such uses of biomarkers can reduce the time and cost needed to assess the efficacy of new treatments relative to assessments based on actual clinical endpoints.
However, to be most useful in clinical practice and translational research, biomarkers require validation — that is, formal assessments showing that measuring the biomarker is a reliable indicator of the patient's clinical status or outcome. Patients can be harmed if clinical decisions are made on the basis of a biomarker that has not been validated. For example, at one point, patients who experienced heart attacks were treated with certain kinds of anti-arrhythmic medications if their heart monitors demonstrated excessive premature beats. The presumption underlying this treatment strategy was that those extra beats were associated with deadly arrhythmias. But although anti-arrhythmic medications did an excellent job preventing those extra beats from appearing on electrocardiograms, subsequent prospective randomized trials showed that patients treated with the drugs actually had greater risk of cardiovascular death.
Using unvalidated biomarkers as surrogate measures in clinical trials can also be problematic. For example, the FDA has approved some new treatments for patients with Type 2 diabetes on the basis of the drugs' abilities to cause a reduction in an aggregate measure of blood sugar (hemoglobin A1C). While this has seemed plausible according to conventional wisdom, the relationship between reducing hemoglobin A1C and prevention of the real damage of diabetes — kidney failure, heart attack, blindness — is not always well-defined. That is, a drug may lead to statistically significant reductions in hemoglobin A1C, but perhaps because of its mechanism of blood sugar lowering, also lead to no improvement or even worse clinical outcomes for diabetes patients. In one notable case, rosiglitazone (Avandia), at one point the bestselling diabetes treatment in the world, was found to increase the risk of adverse cardiovascular outcomes such as heart attack even as it lowered patients' hemoglobin A1C.Reference Nissen and Wolski1
These experiences arise in part because the science of biomarker validation is complicated.Reference Hey2 There are many different modalities for biomarkers (e.g., molecular, histologic, physiologic, imaging); biomarkers can play many different kinds of functional roles (e.g., diagnosis, prognosis, surrogate endpoints in clinical trials); the data and analysis required to properly validate a biomarker varies from role to role; and showing that a biomarker is valid (or invalid) for one role does not mean that is also valid (or invalid) in another role. Biomarkers thus represent something of a double-edged sword in medicine: They can be invaluable tools to reduce the material and epistemic costs of decision-making, but realizing these benefits may also require a scientifically complex (and costly) validation process.
Since biomarkers are becoming a more prevalent and fundamental part of drug development and medical practice, we organized a symposium to discuss potential approaches to validating biomarkers using innovative data sources and methodologies. In this introduction, we will lay out why biomarker validation is a distinctly complex enterprise and discuss how the complexity poses an obstacle to the use of biomarkers to enhance patient care. The following articles will then discuss potential approaches to validation that address this complexity.
Since biomarkers are becoming a more prevalent and fundamental part of drug development and medical practice, we organized a symposium to discuss potential approaches to validating biomarkers using modern data sources and methodologies. In this introduction, we will lay out why biomarker validation is a distinctly complex enterprise and discuss how the complexity poses an obstacle to the use of biomarkers to enhance patient care. The following articles will then discuss potential approaches to validation that address this complexity.
Biomarker Terminology
As we noted above, there are many different modalities and functional roles for biomarkers. Using biomarkers appropriately demands a clear understanding of these functions, how they differ, and what evidence is needed to demonstrate validity for a particular function. In 2015, the U.S. Food and Drug Administration (FDA) and National Institutes of Health (NIH) produced a document — “BEST (Biomarkers, EndpointS, and other Tools) Resource” — to clarify some of the terminology used in the field.3 As the authors of that monograph observed, ambiguous communication around biomarkers (e.g., about what they are, what they do, and how they should be used) is not only a barrier to good science, it can also have serious consequences for patients, payers, and public health.
The BEST Resource document provides a glossary that could help standardize the ways that scientists, regulators, clinicians, and other stakeholders communicate about biomarkers. In particular, the document defines seven different functional roles for biomarkers—they can be used for (1) diagnosis, (2) monitoring patient status, (3) assessing pharmacodynamic/responses to treatments, (4) predicting which treatments are likely to benefit a particular patient, (5) determining patient prognosis, (6) detecting safety signals, and (7) determining patient susceptibility or risk of developing a particular disease. Table 1 summarizes the definitions for these functions and provides an example of each. Although the BEST Resource does not classify surrogate measures in clinical trials as a possible functional role for biomarkers in the same way as these seven, it does discuss this use of biomarkers, and provides a number of specific examples, in its glossary. Therefore, we have added surrogate measures to Table 1 as well.
Table 1 Eight Functional Categories of Biomarkers
Abbreviations: RNA, ribonucleic acid; BRCA 1/2, breast cancer genes 1 and 2; PARP, Poly (ADP-ribose) polymerase; QTc, corrected QT interval; APOE, Apolipoprotein E; HIV, human immunodeficiency virus.
This terminological framework underscores the point that a biomarker cannot be discussed without its clinical context. A biomarker is always measured for some particular clinical purpose, and clear communication requires that we are explicit about that purpose. Similarly, the BEST document helps communicate that a biomarker can have multiple functional roles. For example, a patient's hemoglobin A1C levels may be valid as a diagnostic biomarker; and differences in the effects of treatment on hemoglobin A1C in clinical trials may also be associated with differences in the effects of those same treatments on how patients feel, function, or survive — making the biomarker suitable as a surrogate measure in clinical trials. However, even for ostensibly valid uses of biomarkers, there can also be exceptions, as observed in the case of rosiglitazone's hemoglobin A1C-modifying power not correlating with improvements in patient function. This points to yet another complexity for the science of biomarkers because it shows that a biomarker's validity — even for a clearly defined context of use — is not absolute. Therefore, safe and reliable use of biomarkers requires constant scrutiny to ensure that the use reflects the evolving state of evidence. If exceptions are discovered for a generally valid use of a biomarker, then it is important for researchers and clinicians to update their understanding about how to use the biomarker.
Challenges for Aggregating and Evaluating Biomarker Data
In addition to the challenges arising from the terminological and epistemological complexity of biomarker science, there are also practical challenges to generating, tracking, and aggregating the evidence surrounding a biomarker. Ideally, a central repository for biomarker data would exist to allow researchers or health-technology assessment teams to determine valid and invalid applications. In the absence of such a tool, the majority of clinical trial investigators' and regulators' knowledge about how to use biomarkers currently relies on independent teams of researchers to conduct systematic reviews and meta-analyze the biomarker data reported in the medical literature. But conducting such validation exercises are limited by a number of factors.
First, reports of clinical trials are one of the primary sources for such meta-analyses. Aggregate results from trials that include biomarker outcomes (e.g., progression-free survival) or that stratify their results by biomarkers (e.g., BRCA 1/2 mutation-positive vs. mutation-negative status) can be informative for assessing the validity of surrogate measures and predictive biomarkers, respectively. However, evaluating the validity for the other functional categories of biomarkers typically requires patient-level data, which are rarely reported in the literature or made available by investigators or regulatory authorities.
Second, as valuable as clinical trials are for producing high-quality data, there are domains of medicine in which biomarkers play a central role, and yet there will never be enough clinical trials to validate all the potential biomarkers. For example, the field of precision oncology seeks to leverage data about genetic biomarkers to match patients to therapies that are more likely to be beneficial for the patient's particular genotype. However, with so many potential genotype-treatment options for patients, the high cost of clinical trials, and the limited pool of patients and research resources, it may very well be impossible to test and validate all of these predictive biomarkers using traditional clinical trials.
Third, even just trying to assess the validity of biomarkers using existing aggregate trial data faces obstacles from the time and resources required to conduct a systematic review and meta-analysis. High-quality systematic reviews can take teams of researchers a year or more to complete.Reference Ganann, Ciliska and Thomas4 Given that a publication of the systematic review may then take many months more to appear in print, and the fact that new, relevant data may be appearing in the literature all the time, the existing methods for analyzing biomarker data and disseminating information about safe and reliable biomarker use struggles to keep pace with the evolving state of evidence.
Finally, no research institution (public or private) is currently responsible for monitoring the evidence for or against particular biomarker uses.Reference Hey and Kesselheim5 As a consequence, there is no single body that has the authority to revoke the use of a biomarker in clinical practice if it has been shown to be invalid. Similarly, biomarker-driven research programsReference Hey, Franklin, Avorn and Kesselheim6 can continue to use biomarkers as surrogate measures even if the evidence shows that they are not good predictors of clinical outcomes.Reference Haslam, Hey, Gill and Prasad7
Conclusion
Despite these challenges and complexities, we believe that there is a promising path forward for biomarker validation efforts. Research institutions and stake-holders are increasingly recognizing the risks from relying on unvalidated biomarkers, and there are a number of new initiatives around data sharing and data generation that can help overcome these challenges. In the series of articles that follows in this symposium, a diverse group of authors will describe some of these initiatives — which include both public and private institutions — and put forth some new models for how the data from different institutions could be aggregated to provide a more robust framework for analyzing of the evidence supporting biomarkers.
