|
Linking Pharmacogenetics-Based Diagnostics And Drugs For Personalized Medicine
Progress toward personalized medicine in the five years following the sequencing of the human genome has been slower than many expected. We focus on two potential factors that might be important in explaining this disappointing progress: the limitations of genetic prediction and the lack of appropriate economic incentives. Clinical application of DNA-based and other biomarkers is likely to succeed only on a case-by-case basis, depending on such factors as information content of the biomarker, accuracy of current assessment methods, and effectiveness of available interventions. Both strong intellectual property and value-based, flexible pricing systems will be important in making personalized medicine a reality.
FIVE YEARS HAVE ELAPSED SINCE THE initial sequencing of the human genome, and the number of new pharmacogenetics applications—that is, those whose drug response varies across individuals because of genetic differences—can be counted on two hands. The recent report from the Royal Society cautions: "Pharmacogenetics is unlikely to revolutionize or personalize medical practice in the immediate future."1 Progress has been slower than many proponents of "personalized medicine" had hoped or predicted. Robert Califf has argued that achieving this promise will require a major overhaul of the U.S. clinical research enterprise as well as substantial educational efforts.2 In this paper we explore the potential roles of two factors that could help explain why progress has been slower than expected: (1) the limitations of genetic prediction, and (2) the lack of appropriate economic incentives. We explore both factors individually and also examine their interrelationship in product development and commercialization. We argue that overcoming these challenges could be key to making personalized medicine a reality.
Genetics in health care today. To date, there are very few pharmacogenetics-based targeted therapies. Trasutuzmab (Herceptin, Genentech) is the most prominent example of a linked test and drug for treating breast cancer, although it is not a true genetic test in that "somatic" DNA changes in the tumor are measured (as opposed to "germline" mutations that are present in all cells and are passed to offspring). Although national estimates are not available, it seems clear that most of the volume and recent growth in genetic tests have almost exclusively been for newly identified mutations for rare, "classically" inherited diseases. Examples of these tests include carrier screening for cystic fibrosis mutations (recommended for all pregnant women since 2001), Tay-Sachs screening, and newborn screening for disorders such as phenylketonuria. A few available tests address common diseases, but in limited subsets of patients. This is exemplified by BRCA1 and 2 mutation testing for increased breast and ovarian cancer susceptibility, affecting 5 percent or less of breast cancer cases. These rare, highly penetrant alleles (that is, gene variants associated with a high frequency of disease occurrence, or "phenotypic expression") for common conditions are, in essence, monogenic forms of the disease. They do not, however, account fully for the inherited predisposition for disease; in contrast, most genetic contributions to complex disorders are assumed to be through small, additive effects of multiple gene variants. Even rare monogenic diseases can show variation in the severity or form of presentation and manifestation, but they are still typically distinct in their "gestalt." By their very nature, they are clearly heritable and causally linked to mutations in DNA. They represent discrete traits that may be multifaceted in presentation but are clearly distinct as a phenotype. Their heritable nature does not abrogate the importance of environment, which can play an equally important role in the determination of a mutation carrier’s phenotype and health. For example, biotinidase deficiency is a rare inherited disorder that if left untreated can result in seizures, developmental delay, eczema, and hearing loss. Problems, however, can be prevented with a change in "environment"—that is, a daily supplement of prescription biotin. With highly heritable discrete traits, sufficient variation can be explained by genotype such that testing can be a reliable predictor of phenotype. On the other hand, for common disorders that are multifactorial and continuously distributed, the genotype-phenotype relationship is much more complex and difficult to predict. Genetics in health care tomorrow. Because of incomplete penetrance and the complexity of genotype-environment interactions in common traits, disease genes have proved—not surprisingly—to be much more difficult to find and validate than some predicted with the availability of the human genome sequence. While providing biological clues, they are often poor predictors of disease manifestation or response to drug therapies. We read weekly about scientists finding genes for all types of human illnesses and for explaining variations in treatment responses for those illnesses. Such studies are often small, and substantiating those findings in the broader population often proves difficult. Witness the recent work on COL1A1 Sp1 polymorphisms and risk for osteoporosis and fracture. When this variant was first discovered, it was believed to explain sufficient variation in risk for osteoporosis to be a plausible candidate for risk assessment testing. A large study has now cast doubt that Sp1 contributes enough to the variation to support stand-alone testing.3 Instead, it appears that genetic testing for the Sp1 variant will need to be part of a test that assesses variants in several genes. Further, to increase the fraction of risk that can be assigned to an individual, this genetic information will need to be used in conjunction with standard indicators of risk before genotype can be a clinically useful addition. This example illustrates the fact that the level of predictive value of genotype will vary depending on the type of trait—discrete or continuous—and the proportion of the variation observed in the trait contributed by genes. Clearly, considerable research beyond initial genetic hypotheses is necessary to determine if and when testing is clinically useful. This is of great relevance to therapeutic response prediction. It reminds us that continuously distributed, complex quantitative traits with moderate heritability are not simply explained or predicted by genotype. Even if all of the genetic contributions are known and measured, the ability to predict outcome based on genotype is naturally related to the heritability. Typically, heritability estimates for most complex traits that are amenable to study reveal estimates of 0.3 to 0.7 (the ratio of the genetic variance to total phenotypic variance in a population).4 It is assumed that drug response will often prove to be a continuous, complex trait. In some cases, drug behavior in the body will be a discrete trait, such as with some instances of metabolism, where, depending on the substance, there can be three distinct phenotypes: ultrarapid metabolizers (UMs), extensive metabolizers (EMs), and poor metabolizers (PMs). Mutations at one locus contribute significantly to the observed phenotype. For example, codeine is metabolized to morphine in part by CYP2D6. Thus, PMs receive little if any pain relief, but as a case example from the New England Journal of Medicine illustrates, UMs may experience life-threatening side effects.5 It should also be noted that in this particular case, numerous factors beyond CYP2D6 status contributed to the adverse event. What if a continuously distributed complex trait were involved? Consider a hypothetical drug proven to be efficacious but in only 30 percent of patients treated, and this response is continuous: Everyone shows some degree of response, but only those over a certain threshold (for example, the 30 percent) are considered to be truly benefiting from the drug. This is similar to the actual case of lowering cholesterol and the corresponding reduction in heart attack and stroke risk. The disease risk reduction must be sufficient to outweigh the risk and cost of treatment; thus, some threshold is determined for when patients are receiving medical value and when they are not benefiting. This threshold is not completely arbitrary, because it will relate to some measurable benefit, but it is also not a clear, definitive yes-no relationship. Thus, for a continuously distributed drug response, the heritability of the trait will influence the usefulness of knowing the genotype. If the trait is highly heritable (0.95), genotyping will offer high positive predictive value. But, as is more likely the case, if the response only shows moderate (0.5) heritability, testing will produce higher rates of false positives and false negatives, leading to a corresponding reduction in the predictive power of testing. In such cases, genotype alone is less likely to be a robust enough predictor to be clinically useful. This will, however, depend on what other predictors are known, how well they inform clinical management, and how well the combined use of these risk factors works in the clinic. Many tests used in risk assessment today are far from perfect yet still prove to be highly beneficial. It should be kept in mind, also, that knowledge of genetic contribution to disease or drug response can have considerable value in drug development—apart from direct clinical application—by informing about molecular events underpinning disease or response. Thus, although there may be disappointment in what the genome project has delivered so far, it has enabled the science considerably. It will, however, take much more research to elucidate specific genotype–phenotype relationships and clinically validate genotyping applications. Importance of information over analyte. As the benefits and limitations of genotype have become better recognized, there has already been a gradual and growing appreciation that it is not the biomarker per se—whether the analyte is genetic or not—that is important, but rather the type, amount, and quality of information it provides. Any new biomarkers enter a world in which many health risk determinants are known and applied in practice. To bring value to the health care system, a new test must lead to a change in current clinical behavior. The degree of improvement needed for a new predictive test to provide clinical utility—how it will perform medically in real-world practice—depends on several factors such as disease and therapeutic choice.6 These relationships—as typified above and in the recent example of epidermal growth factor receptor (EGFR) measurement in non–small cell lung cancer—are proving to be very complex.7 The initial reports of somatic EGFR mutations predicting tumor response to gefitinib (Iressa, AstraZeneca) led to considerable speculation that a simple test would be a highly accurate predictor of tumor response to EGFR-inhibitory agents. Further studies have not borne this out but instead have revealed a complex biological situation with numerous mutations/ markers interacting in tumors. Although there has been considerable focus on pharmacogenetics, specifically, and genomics, more generally, as the means to reduce uncertainty in medicine, it appears that the focus should be on finding and validating predictive biomarkers of high information content, be they genes, proteins, or visible differences. Other scientific issues. Recognition of the fact that genes or sets of genes may be weak predictors of drug response raises several related, important scientific questions. First, given the exponential increase in the available data, it seems relatively easy to find gene-variant and disease associations: In one review, more than 600 positive associations between genes and disease were reported; 166 had been studied more than 3 times, but only 6 had been consistently replicated.8 Weak predictability combined with our lack of understanding of the causal relationship between genes and drug response makes it difficult and costly to conduct appropriate validation studies. These studies are probably going to have to be large-scale, prospective studies that measure genetics and other biomarkers over time and follow up with patients for long-term outcomes. Producing new knowledge this way will be costly and time-consuming in the near future but could enable smarter, faster therapy and diagnostic design in the long term.
The translation of the basic science of pharmacogenetics and other "-omics" biomarkers as applied to drug development and clinical care is occurring in a complex legal, regulatory, and reimbursement environment. Just as it may be difficult to predict phenotype from genotype for a complex disease, it is also difficult to predict clinical use and commercial success of a new diagnostic and drug linked via pharmacogenetics. Understanding and appropriately shaping this environment is vital for encouraging biomarker research and personalized health care. Incentives for innovation: intellectual property and pricing. The potential to acquire intellectual property (IP) rights for new drugs or new diagnostics, including pharmacogenetics-based tests, can be a powerful incentive for their invention and development. Patents, in effect, grant temporary monopoly rights to inventors, resulting—in theory—in a short-run "welfare loss" (as higher prices reduce use) offset by a higher innovation rate in the long run. The impact of the patent system will vary depending on type of technology, sector, and business model. The role of patents in pharmaceutical business is well documented. In brief, statutory patent life now lasts twenty years, but because of the eight to twelve years it takes to test and prepare a new drug for market, effective patent life—the actual time of market exclusivity—is typically much shorter.9 Furthermore, it is common for competitor pharmaceutical companies to be preparing competing "follow-on" compounds—that is, different molecules that rely on the same mechanism of action. This can weaken monopoly power through competition, potentially benefiting patients who have greater choice and more treatment options. For the diagnostics business, historically, IP and competition have centered on platform (that is, the instruments used to measure analytes such as blood glucose), not content (or actual analyte measured), but this has been changing. A manufacturer with a widely used, well-established platform benefits when there is free access to validated content (analytes) to run on that platform. Conversely, to have an economic incentive, a discoverer of new biomarkers needs to be able to recoup research costs by profiting either from direct sales of testing services or by licensing to other laboratories or manufacturers, or both, who can put tests in the hands of service providers. In essence, platform providers and laboratories need content, and content discoverers need a distribution route to patients. IP is central to this dynamic tension. Further, depending on the characteristics of the analyte—novel or with well-established relationship to disease—and other factors, test development times can vary widely. Moreover, business development cycles can be shorter than for drugs. Thus, IP is just as important for diagnostics as for drugs, but the best role for patents depends on the view of the stakeholder: The platform manufacturer would prefer broader, low-cost access to a wide range of content (analytes), while those with IP on content want to capture the clinical and economic value of a novel biomarker. Laboratories want broad access to content and freedom to operate using either a platform-based test or in-house testing, whichever they prefer. How this tension is resolved has major implications for the linked diagnostic-drug paradigm: The monopoly power conferred by a novel and useful patent allows the patent holder to set prices above the level that would normally prevail under open competition. This difference can be thought of as the reward to the innovation. Strong IP protection, however, is not the only factor shaping pricing and reward. Public health insurance systems have adopted various reimbursement policies that affect the size and structure of this reward for both drugs and diagnostics. We distinguish between cost-based and value-based approaches. In more technical economic terms, a competitive market would tend to yield a price equal to long-term marginal cost, including an appropriate rate of return on investments. Health insurance reimbursement schemes that attempt set reimbursement, or establish or push prices to this level, could be defined as "cost-based" systems. And in the extreme, they may try to set prices closer to short-run marginal cost. Alternatively, thinking of an implicit societal demand function based on willingness to pay for improvements in mortality, morbidity, or other cost savings, one could define a system that pays closer to what the monopolist would charge to be "value-based." For example, it is generally understood that the U.K. National Health Service (NHS) applies an approximate threshold of £30,000 per quality-adjusted life year (QALY) gained in its coverage decisions. This sends a strong signal to drug and test manufacturers about what willingness to pay and "value" are and what they can charge for them. Pricing and reimbursement: pharmaceuticals. Initial pricing for new brand-name prescription drugs in the United States and European Union (EU) can be characterized as somewhat value-based. The manufacturer sets prices to try to capture citizens’ willingness to pay in a given country, which is filtered (and possibly heavily distorted) through insurance and payment mechanisms and institutions. Since only about 30 percent of drugs recover more than the average cost of development, many drugs are not profitable in the long run in and of themselves, but they do contribute some amount to recovering pooled fixed costs.10 But since one cannot know in advance which ones will be these high-return "blockbusters," the mix of high- and low-return investments is necessary to sustain the enterprise as a whole. Besides follow-on compounds, a number of other factors can make it difficult for pharmaceuticals to achieve the optimal monopoly price. First, economic theory suggests that the optimal price should differ in different countries or local markets, but it is difficult to charge greatly different prices across countries of the EU or U.S. states because of "parallel trade."11 Second, in some markets, particularly in the EU and Australia, the price negotiation is with a central government that has considerable monopsony power. Third, and most important for this discussion, it is difficult to change a price once a drug is on the market—either for a new indication or for a new targeted subgroup. This is particularly true in the EU, where each country has its own "administered pricing" system but where the initial price is often linked to other countries by "reference" pricing. In the United States, manufacturers are free to raise a price over time, subject to countervailing competitive economic and political pressures. Pricing and reimbursement: diagnostics. The pricing and reimbursement model for diagnostics is fundamentally different from that for drugs in both the United States and the EU.12 Reimbursement systems for diagnostics might best be described as resource- or cost-based, rather than value-based. Furthermore, they are also generally administered pricing systems. In the United States, for example, when new tests enter the market, an effort is made to link them to the existing reimbursement of tests involving similar effort or cost: for example, techniques such as "crosswalking" and "gap filling" are used.13 Similar systems apply in France and Germany. All of this implies that the reimbursement—the reward for a new test—is not necessarily based on value added. One can speculate as to why drug and diagnostic pricing and reimbursement systems have evolved differently in developed economies. It could be a matter of economics, politics, historical accident, and other factors. Sorting out these reasons is beyond the scope of this paper, although they could be important for understanding the feasibility and viability of reform. For now, we take the current state as given and discuss its potential impact on incentives to develop pharmacogenetics-based tests. Pricing and incentives: pharmacogenetics-based tests. What is unique about using pharmacogenetics-based tests to target subgroups of patients? Fundamentally, from an economic perspective, pharmacogenetics-based tests do not differ from other tests, such as blood tests for high cholesterol or diabetes, that are used to identify which patients are the best candidates for treatment. They are trying to define the set of patients for whom the average risk-benefit ratio is favorable. Within that set, typically, only some patients will respond fully, and only a subset will suffer side effects. For seven of fourteen major drug classes, it has been estimated that 50 percent or less of patients respond.14 One great hope for pharmacogenetics-based tests is that using genes as predictors will help us find the subset of responders or rule out those patients suffering side effects, for whom the risk-benefit ratio is unfavorable. This ratio can also be interpreted in economic terms: Clearly, those patients experiencing higher levels of benefit in relation to risk are obtaining higher value. Thus, a case can be made that the price—the reward paid to the innovator—should be higher. Imagine, for example, a drug for which only 20 percent of patients received benefit, but we could not identify them ahead of time without a pharmacogenetics test. Paying for value would essentially average this benefit over 100 percent of patients. However, if a test were available to identify the 20 percent, we should be willing to pay on the order of five times more, since the total benefit in the population is the same. But if pricing and reimbursement systems for diagnostics and drugs are not flexible enough to reward this higher ratio of benefit to risk in a subgroup of patients, what is the incentive to find the subgroup and develop a test? In a previous paper we examined the multiple factors that come into play that could affect this incentive in a given situation. These include (1) whether the drug is already on the market (and priced) before the test is developed; (2) the extent to which drug and diagnostic prices are flexible and are value- versus cost-based, (3) the competitiveness of the insurance market, and (4) the strength of the patent protection on the diagnostic versus the therapeutic.15 On the face of it, it is obvious that drug manufacturers could be considerably worse off if a pharmacogenetics-based predictive test is discovered after the drug is on the market, and that test developers will have much less of an incentive to develop a pharmacogenetics-based test if they cannot themselves capture a large share of the value created by identifying a responding subgroup. Hence, there will be limited incentives for both drug and diagnostic manufacturers to develop tests for drugs that are already on the market, although the disincentives are greater for drug manufacturers. This would argue for codeveloping test and drugs, coming to the market with the association fully proven. Other barriers to diagnostic-therapeutic linkage. In addition to the scientific and economic barriers highlighted above, two other factors deserve mention that might be inhibiting the development of pharmacogenetics-based test-drug combinations. One factor is the high cost of the basic research that is needed to validate genetic markers. Financing this remains a question: What should fall to the public sector, how much should the private sector contribute, and what is best done in partnership are under active debate. Second, approval and reimbursement for drugs in the United States and EU require greater levels of evidence than is customary for new diagnostic tests. Some would argue that the lower evidentiary requirements for regulatory approval of tests has discouraged the development of better clinical data and that payers practice "cost-based" reimbursement in part because of this lack of evidence on clinical and economic value. In contrast, both the Academy of Managed Care Pharmacy (AMCP) Format for Formulary Submissions guidelines in the United States and the requirements for submissions to the National Center for Clinical and Health Excellence (NICE) in the United Kingdom request economic models that synthesize the clinical and cost evidence to assess value added of new drugs. Similar standards and mechanisms do not exist for diagnostics, although the discussion is beginning.16 But asking for more evidence means raising the costs of developing—and particularly validating—new diagnostics. What are the incentives for diagnostic manufacturers if no additional rewards are forthcoming through the reimbursement system?
In our opinion, pharmacogenetics-based diagnostics and drugs are unlikely to be linked in large numbers unless these scientific and economic challenges can be met. Most importantly, until we have much greater knowledge of the actual predictive power of new molecular markers, useful applications will be much slower in coming than we would hope. This is likely to require a substantial increase in the level of public investment in both basic and translational research. We also hypothesize that value-based, flexible pricing systems—backed by strong, consistent intellectual property rules—are a necessary, but not sufficient, condition to achieve the promise of personalized medicine. It remains to be determined what the science can deliver with respect to valid clinical applications. Public policy, however, needs to focus on supporting the paradigm of more closely aligned biomarkers, diagnostics, and therapeutics and not focus on a single analyte or technology, such as genetic testing, if the full scientific potential is to be realized. Given the complexities of the biological systems we are attempting to parse and then manipulate, as well as the many years it typically takes to develop and test a new drug or validate a biomarker, this will still most likely take decades. Nonetheless, this new knowledge holds promise in the long run, and major medical progress can be achieved under economic incentives that foster innovation.
Lou Garrison (lgarrisn{at}u.washington.edu) is a professor of pharmaceutical outcomes research and policy in the Department of Pharmacy, University of Washington, in Seattle. M.J. Finley Austin is director of U.S. external science policy at F. Hoffmann-La Roche AG in Basel, Switzerland. The authors have benefited from discussions with Adrian Towse, David Veenstra, Chris Chamberlain, and James Creeden. Lou Garrison worked for Roche Pharmaceuticals from 1995 through May 2004 and has since been a consultant to Roche and other pharmaceutical and diagnostic companies. The views expressed in this paper are those of the authors and should not be attributed to their respective organizations.
This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||
Top
Challenges In The Underlying...
Challenges In The Underlying...
