Personalized cancer medicine—advances and socio-economic challenges

Introduction

From the revolutionary success of imatinib,¹ to the follow-on promise of compounds such as crizotinib² and vemurafenib,³ a strong rationale exists for the principles of targeted therapy. Yet, even with the improved response rates demonstrated with such agents, it is clear that the therapeutic activity of a molecular-targeted drug is only achievable under certain biological conditions.⁴ The very modest benefit seen in the majority of patients administered targeted drugs, such as bevacizumab, is a stark reminder that the association between target activity and tumorigenic potential varies greatly across a patient population.⁵ Current clinical trial design compounds this problem, with large cohorts used to fulfill the needs of statistical significance rather than generating a deeper understanding of the biologically sensitive contexts within a trial population. What emerges is a system of evidentiary medicine that leaves patients more dependent on the rules of chance than those of biological fact. It is hardly surprising that with most therapies failing to improve outcomes in patients with metastatic disease, the war on cancer has contributed to a financially battered health-care system and a growing cancer patient population for whom traditional clinical practice often offers little but a lottery of hope.

The growing discordance between the enormous costs and marginal clinical benefit associated with many anticancer blockbusters has led to increasing concerns.^6,7 Most stakeholders agree that urgent improvements in cost-to-benefit ratios are required; we strongly believe advances can only be realized through widespread adoption of truly targeted approaches to patient treatment. Thus, the definition of ‘personalized medicine’ that we have applied throughout this article is restricted to the process of identifying molecular subtypes of individuals or disease that are uniquely susceptible to a particular treatment. Using this truly personalized strategy, therapeutic interventions can be applied to those who will likely benefit, sparing expense and toxic side effects for those who will likely not benefit. Of course, the challenges of bringing molecularly targeted patient treatment to clinical reality are as diverse as the opportunities presented. Key stakeholders hold vested and often juxtaposed interests, making the challenge even greater. In this Perspectives article, we examine some of these issues and ask how financial and moral necessities are considered while achieving the widespread clinical implementation of targeted treatment to patients.

Targeted drugs versus therapies

Although the concept of targeted therapy has been embraced with considerable excitement, an unbiased appraisal of clinical response rates associated with drugs such as bevacizumab and cetuximab leads us to question the broad-scale validity of the term ‘targeted therapy’.⁸ Although targeted drugs act to specifically inhibit the activity of one or more oncoproteins, there is considerable uncertainty as to whether this inhibition will result in a therapeutic effect. This ambiguity derives in part from the heterogeneity of molecular drivers underlying a disease. Thus, to achieve targeted therapy, the molecular-targeted agent must be administered to patients where the intended target is a validated protagonist of disease progression in that tumor. For example, the BCR–ABL fusion protein is a ‘driver’ of disease in chronic myeloid leukemia, which explains why the tyrosine kinase (BCR–ABL) inhibitor imatinib is an effective treatment in patients with this disease.^9,10 In an analogous manner, the activity of crizotinib is currently restricted to non-small-cell lung cancer (NSCLC) tumors possessing an EML4–ALK fusion¹¹ and the activity of vemurafenib is observed in melanomas possessing a V600E substituted BRAF protein.¹² However, early results suggest that the presence of the V600E BRAF substitution in colon cancer might be a poor predictor of vemurafenib response in this disease,¹³ emphasizing that targeted-drug activity is context specific. Moreover, examples exist where promising anticancer-drug targets act as tumor suppressor proteins in certain tumor subtypes. For example, Ephrin type-B receptor 4, which is a focus of current drug discovery and development efforts, has been shown to act as an oncogene in bladder cancer¹⁴ and as a tumor-suppressor gene in colorectal cancer.¹⁵ The use of targeted drugs without sufficient knowledge of the molecular context of the target in a particular patient might result in poor survival or severe adverse drug reactions, and for the health-care system it can lead to considerable financial waste and associated societal effects.

Drug cost versus benefit

Over the past decade, the poorly directed use of anticancer drugs has contributed to the emergence of many worrying trends. For example, spending on cancer medications has risen by 14% annually over the past few years.¹⁶ In 1995, the only widely used cancer drug that cost more than $2,500 per month was paclitaxel.¹⁷ By contrast, many recently approved targeted drugs have entered the market priced at many times this cost, with more than 90% of the anticancer agents approved by the FDA in the past 4 years costing more than $20,000 for a 12-week course (Figure 1).¹⁸ With insurers bearing the brunt of these costs in the USA, their response has been to off load the expense onto patients; therefore, out-of-pocket spending has increased sharply, especially for low-income patients who are estimated to spend about 27% of their yearly income on medical costs,¹⁹ no doubt convincing many to avoid or discontinue treatment. It has been estimated that 62% of all personal bankruptcies in the USA are attributable to medical costs and, among those, cancer is often the illness that initiates the financial burden.²⁰ Moreover, given that evidence exists linking biobehavioral stress with tumor progression,²¹ it is clear that many patients may endure a vicious cycle of decaying wealth and health. With ASCO issuing a guidance statement affirming “the critical role of oncologists in addressing cost of care,”²² oncologists have become increasingly vociferous. However, their primary concern lies not with drug prices per se, but rather with what many perceive as a more-worrying pattern of diminishing therapeutic returns, where increases in price substantially outweigh the often modest benefit to the patient.^6,7

Antibody drugs provide particularly illustrative examples, for example, treatment with the antibody bevacizumab costs more than $50,000 per year (note the median household income in the USA was $52,029 in 2008),²³ with costs rising to about $100,000 per year when drug administration and routine monitoring are also considered.⁵ Although early clinical studies showed enormous promise for this drug,²⁴ later-stage development produced a number of disappointing results (for example in patients with prostate cancer²⁵ and pancreatic cancer^26,27) along with modest successes in terms of overall survival in other cancers—typically measured in weeks (Table 1). Given the enormous unmet medical need, the FDA has approved bevacizumab for treatment of colorectal cancer,^28,29 glioblastoma,³⁰ NSCLC,^31,32,33 and renal-cell carcinoma.^34,35 Approval for the treatment of relapsed or progressive glioblastoma was granted by the FDA on the basis of results from the phase II BRAIN study, and the large unmet medical need associated with this most-aggressive form of malignant brain cancer.³⁰ This approval came in the absence of large-scale phase III data and other data demonstrating whether or not the drug improves disease-related symptoms or survival in previously treated patients. It is noteworthy that the European Committee for Medicinal Products for Human Use (CHMP) issued a negative opinion regarding marketing approval based on the same data. Their main objection was the lack of a comparator arm without bevacizumab in the BRAIN study, and the fact that the CHMP tends to base its approval decisions on phase III studies only.³⁶ Most controversial of all, however, was the approval of bevacizumab for treatment of metastatic breast cancer in 2008;³⁷ this decision came counter to a recommendation by the Oncology Drugs Advisory Committee (ODAC), which expressed concern about the lack of overall survival benefit and significant risk of fatal adverse events.³⁸ Subsequent data from the AVADO³⁹ and RIBBON-1⁴⁰ trials have validated these concerns and a follow-up meeting in 2010 saw the ODAC vote 12 to one in favor of removing the metastatic breast cancer indication from the drug label of bevacizumab⁴¹—and the FDA is adhering to the advice.

It is astounding that the estimated costs for keeping a single patient with breast cancer alive for 1 year (or 1 quality-adjusted life year [QALY]) using bevacizumab is almost $500,000.⁶ Similarly, Fojo et al.⁷ have estimated a cost of $800,000 for keeping a patient with NSCLC alive for 1 QALY using cetuximab.⁷ Extrapolating their analysis to the 550,000 Americans who die from cancer each year, they estimated a financial implication of $440 billion to extend the lives of these patients by a single year.⁷ Recent health economics modeling of bevacizumab use in ovarian cancer has also yielded stratospheric projections.⁴² Working with data from the GOG-0218 trial,⁴³ Cohn et al.⁴² examined the potential cost effectiveness of adding bevacizumab to first-line treatment of advanced-stage epithelial ovarian cancer. Their projections suggest a cost of $1,305,000 per patient for maintenance bevacizumab added to the triple therapy combination of paclitaxel, carboplatin and bevacizumab and an incremental cost-effectiveness ratio (compared with paclitaxel plus carboplatin) of $401,088 per progression-free year of life saved.⁴² Costs such as these have encouraged considerable debate in the field, with Hensley,⁴⁴ asking what paying $78.3 million for 3.8 months of progression-free survival (PFS) for 600 women means for other participants in the health-care system? Similar concerns were also raised by Munro and Niblock,⁴⁵ who were not only critical of the marginal overall survival benefit reported for treatment with the antibody trastuzumab in the ToGA study,⁴⁶ but they also questioned the societal implications of the estimated costs. After conservatively projecting that clinical use of trastuzumab in patients with HER2-positive advanced-stage gastric or gastro-esophageal junction cancer will cost about $90,000 per life year gained, they examined the yearly health expenditure for citizens in 24 countries (range $40–$5,500) and asked, “what is the justification for introducing a treatment that might enable one individual to live a couple of months longer, but will consume, for each person treated, the total yearly health expenditure for scores of their fellow citizens?”⁴⁵ This question becomes all the more relevant given the clinical reality that most advanced-stage cancer patients will receive no survival benefit and on the contrary may have to contend with serious adverse drug reactions.⁴²

These facts raise important concerns about the poorly directed use of multiple targeted drugs in a single patient. With the appearance of every new targeted treatment option, has come the natural attempt to combine drugs with distinct targets and mechanisms. In the treatment of metastatic colorectal cancer (mCRC), for example, the cost of the 5-fluorouracil–leucovorin combination is less than $100 for a 6-month course, but the addition of irinotecan or oxaliplatin costs an additional $30,000, adding bevacizumab to the combination requires an additional $24,000 and cetuximab a further $50,000.⁴⁷ Therefore, the aggregate cost for treatment of patients with mCRC using this multiagent approach lies somewhere around $150,000 to $200,000 to achieve an additional year of survival compared with treatment with 5-fluorouracil–leucovorin alone.⁴⁷ So, although the growing adoption of multiagent therapeutic strategies is to be welcomed in scenarios of proven clinical benefit, the associated financial costs demand a greater emphasis on the development of combination therapy linked to predictive biomarkers in the hope that this will improve the cost–effectiveness ratio.

Factoring adverse events

A critical issue that is poorly factored into the aforementioned financial estimates is the cost incurred owing to adverse drug responses. Not only are adverse drug responses a crucial determinant of the outcome of patients with cancer, they also considerably compound the financial outlay required to treat the patient. In support of this view, two of the top five drug expenditures by clinics in the USA in 2010 (January to September) were for drugs that are used to treat the adverse effects induced by antineoplastic agents (Table 2).⁴⁸ In the case of the GOG-0218 trial, the 3.8 months of additional PFS was not only associated with enormous direct costs, but also with a 23% risk for developing grade 2 hypertension, a 10% risk for grade 3 or 4 hypertension, and a 2.3% risk for grade 3 or worse gastrointestinal perforation, fistula formation or hemorrhage.⁴³ The costs associated with hospitalization for these adverse events are presumably enormous. Interestingly, a recent meta-analysis of 10,217 patients with a variety of advanced-stage solid tumors from 16 randomized controlled trials showed that bevacizumab in combination with chemotherapy or biological therapy was associated with increased treatment-related mortality compared with chemotherapy alone.⁴⁹ The overall incidence of fatal adverse events with bevacizumab plus chemotherapy was 2.5%, and was some 1.5-fold more frequent among these patients than patients receiving chemotherapy alone. The most common of these events were hemorrhage (23.5%), neutropenia (12.2%), and gastrointestinal tract perforation (7.1%).⁴⁹ Such effects are a stark reminder of the importance of developing appropriate treatment biomarkers to ensure that the optimal treatment is given to each patient, particularly in the context of highly expensive drugs that produce marginal survival benefits (Table 3).

Major ethics issue, minor benefit

The reason often put forth to explain the modest overall survival benefit derived from molecular-targeted therapies include the confounding effects of post-progression treatments, early follow up and/or insufficient statistical power. Investigators involved in the AVAiL trial suggested that “PFS benefit did not translate into a significant overall survival benefit, possibly due to the high use of efficacious second-line therapies”.³³ However, inspection of post-study therapies shows conclusively that they are essentially identical in both trial arms.⁶ A plausible explanation is that therapeutically positive effects on survival are being masked by patients whose disease becomes worse upon exposure to the treatment. For example, evidence from clinical studies suggests that the addition of cetuximab to chemotherapy in the 40% of patients whose tumors harbored mutant KRAS, may have led to worse survival and a significant increase in serious toxic effects.^50,51 When one considers that tens of thousands of patients were treated with cetuximab who might never have responded in the first place, it is clear that before the advent of KRAS testing the drug posed considerable risk, worse quality of life and decreased survival for many of these patients. On the basis of this and similar evidence, one can only conclude that when a molecular-targeted drug is approved on the basis of minimal improvements in overall survival or PFS (typically in the absence of a predictive biomarker), we could be committing to shortening the lives of a considerable number of patients, in some cases almost as many patients as we are able to extend the lives of. Is this morally acceptable?

The marginal effects on overall survival must also be interpreted in the context of the tightly controlled randomized controlled trials from which these results derive. Trial patients are clinically more homogenous than patient populations in clinical practice, with their diverse comorbidities and other clinically mitigating factors. Although randomized controlled trials produce results that are statistically robust and easy to interpret, an important level of pragmatism and the ability to generalize is often lost. While comparative effectiveness research (Box 1) might provide important insights into the real-world value of such therapies,⁵² the future emphasis and associated investment must be biased towards the identification of clinically effective predictive biomarkers, starting at the point of clinical development. In cases such as cetuximab,⁵³ or bevacizumab,⁵⁴ where there is poor correlation between target status (for example, expression or mutation of the receptor) and response to the drug, a greater emphasis should be placed on identifying other gene predictors of tumor responsiveness and patient risk.

Moral imperative, clinical reality

The paucity of standard-treatment options for advanced-stage cancer patients emphasizes the urgent need for novel predictive biomarkers and associated therapies. Few would argue against the hypothesis that the earlier we can target a drug (or drugs) to the true molecular protagonists of a cancer, the greater the chance for improved patient outcomes. The inherent challenge is, however, highlighted by the diversity of aberrations in germline and tumor-based molecules that can impinge on the activity of a drug. With only 20 oncology drugs associated with FDA-approved predictive biomarkers⁵⁵—the result of years of research and investment—we are still far from the widespread adoption of personalized cancer medicine as a standard-of-care. Nevertheless, many important concepts (for example, oncogene addiction⁵⁶), technologies (such as next-generation sequencing⁵⁷) and projects⁵⁸ have been established during this time and we are currently witnessing a modestly rising trajectory of therapeutic return on investment.

We must aggressively realign our discovery efforts towards an enhanced assault on human disease systems, as opposed to model organism counterparts. For example, while there are enormous quantities of clinical data distributed throughout oncology practices, comprehensive cancer centers and pharmaceutical companies, such data are rarely accessible and are thus currently of limited benefit to the study of the disease. Yet, such data could provide us with an invaluable opportunity to study the relationship between a human phenotype and drug-induced protein perturbations within a patient system. Deciphering the molecular basis of clinical responses, through mapping of drug regimens to associated targets and pathways presents a unique opportunity to dissect disease systems in search of novel response biomarkers, drug targets and efficacious combination therapies. In addition, a significant proportion of such data also have associated tumor samples available, raising the exciting opportunity of aligning clinical profiles with matching molecular aberrations.

While efforts such as comparative effectiveness research might deliver important data that will assist immediate treatment and reimbursement needs, such investment still fails to deliver the fundamental determinants of drug response and risk. To achieve this, we must pursue a rebalancing of the funding ledger in favor of projects and technologies that encourage the widespread generation and integrative analysis of patient-specific clinical and molecular information for all cancer types. Efforts such as The Cancer Genome Atlas (TCGA) and the International Cancer Genome Consortium (ICGC)⁵⁸ represent positive, but tentative, steps in this direction. Moreover, while there is always a place for multiomics datasets in basic clinical research (such as gene and microRNA expression or gene methylation status), a more strategic use of funds would prioritize the generation of actionable clinical data. Three broad classes of such actionable predictive biomarkers, namely resistance biomarkers, response biomarkers and risk biomarkers—here captured under what we term the ‘3R principle’—can provide us with information about this actionable data (Figure 2). Resistance biomarkers help to define drugs for which the patient will most likely not respond, response biomarkers help identify patients for whom a drug may work, and risk biomarkers identify potential safety concerns such as suprapharmacological accumulation of an active drug species. Thus, while complete genome sequence information is much vaunted, a more-targeted appraisal of all genes and/or proteins that are associated with the pharmacokinetics and pharmacodynamics of a drug and its mode of action within a disease system might yield quicker, cheaper and more reliably interpretable results that can inform both trial design and clinical decision.

Genetic sequence data from clinical trial participants could be used prospectively to assess whether potential resistance mechanisms might exist or retrospectively to identify predictive biomarkers of response, resistance and risk—in our opinion, all these data points should be a compulsory part of future regulatory submissions. This change in the regulatory framework would allow us to build on both existing knowledge and the precedents set by innovative clinical studies such as the prospective and adaptively randomized BATTLE trial (Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination) in heavily pretreated patients with NSCLC.⁵⁹ Combining both molecularly guided treatment with an adaptive Bayesian design, BATTLE provides an exciting paradigm for knowledge-driven drug and biomarker development. Beyond clinical trials, medical insurance companies in the USA could also contribute to the endeavor by encouraging coverage with evidence development (CED) for promising new biomarkers and/or therapies. CED provides coverage for medical technologies deemed ‘experimental’ or ‘investigational’ and is contingent on the member’s participation in a payer-organized clinical research program. Using this approach, coverage for FDG–PET as a diagnostic tool for various cancers was expanded by the Centers for Medicare and Medicaid Services (CMS) as a result of positive evidence from the National Oncologic PET Registry.⁶⁰ Armed with such diverse sources of clinical data, the community could drive the discovery of novel molecular-guided treatments in an expedited fashion. Nevertheless, this still represents only one aspect of a potential solution.

As the catalog of candidate treatment biomarkers grows, so too does the challenge of prospectively validating them. And even when validated, further complex questions arise. Are the treatment implications of particular biomarkers extendible across indications? Does the medication history of a patient influence how we use such biomarker information? How do we address the influence of comorbidities and what influence might current medications have on biomarker fidelity and anticancer drug response? And, what is the best treatment option when numerous aberrations are encountered—can we design appropriate combinatorial treatments to offset the complex drug resistance effects? These are but a few of the complex questions that will face oncologists, and with the growing clinical application of next-generation molecular technologies, it seems clear that a molecular-data overload might quickly limit our ability to derive maximum clinical value from research investment. In silico technologies will, therefore, have a pivotal role in powering both knowledge-guided data generation and subsequent analytical processes. Efforts in system-specific data integration of text-based knowledge and omics-based data must remain a key focus in this endeavor,⁶¹ providing the framework for further integration and analysis of patient-specific clinical and molecular information. Once personalized in this manner, such models may provide the best foundation upon which to address the complex array of questions that face every treatment decision and clinical study.

Conclusions

We have reached a revolutionary turning point in the history of medicine. Technological advancements and conceptual foresight has advanced patient care, but at a considerable cost! Key to the long-term goal of cancer treatment is our understanding of the molecular compatibility between the mode of action of a drug and the driver mechanisms associated with tumor progression. Such knowledge provides the very foundation of factual clinical decision making and if those making the financial decisions regarding the availability of drugs could also integrate this into their evaluations, this could incentivize manufacturers to set prices commensurate with the actual health benefit delivered—as opposed to the level of patient need. The moral imperative for aggressively pursuing this path is enormous, as is the potential for new levels of patient survival and cure. As Hippocrates reminds us, physicians must have a leading role in guiding the diverse stakeholders along this revolutionary path: “Life is short, and the Art is long; the occasion fleeting; experience fallacious and judgment difficult. The physician must not only be prepared to do what is right himself, but also to make the patient, the attendants and externals cooperate”.⁶²