In this paper, we examine the clinical and economic challenges that face developers of and payers for personalized drugs and companion diagnostics. We review and summarize clinical, regulatory and reimbursement issues with respect to eight, high profile personalized medicines and their companion diagnostics. Subsequently, we determine Medicare parts B and D reimbursement of the eight drugs from publicly available databases. Finally, we utilize surveys—each tailored to three key stakeholders; payers, drug and diagnostic developers, and pharmacogenomic expert analysts—to assess reimbursement of diagnostics, analyze the role that different kinds of evidence have in informing prescribing and reimbursement decisions, as well as the specific clinical, regulatory and economic challenges that confront pharmacogenomics as it moves forward. We found that Medicare beneficiary access to physician-administered (Medicare part B) drugs is relatively unfettered, with a fixed patient co-insurance percentage of 20%. More reimbursement restrictions are placed on self-administered (Medicare part D) drugs, which translates into higher and more variable cost sharing, more use of prior authorization and quantity limits. There is a lack of comprehensive reimbursement of companion diagnostics, even in cases in which the diagnostic is on the label and recommended or required by the Food and Drug Administration. Lack of evidence linking diagnostic tests to health outcomes has caused payers to be skeptical about the clinical usefulness of tests. Expert analysts foresee moderate growth in post-hoc development of companion diagnostics to personalize already approved drugs, and limited growth in the concurrent co-development of companion diagnostics and personalized medicines. Lack of clinically useful diagnostics as well as an evidence gap in terms of knowledge of drug and diagnostic clinical effectiveness appear to be hindering growth in personalized medicine. An increase in comparative effectiveness research may help to close the evidence gap.
Pharmacogenomics explores the ways in which genetic variations can be used to predict whether an individual patient will benefit from a drug, have a bad response or no response at all.1 Accordingly, therapies may be tailored to certain genetic characteristics of individual patients or sub-populations, drawing on data gathered from a variety of sources, including tests for biomarkers.2, 3 Knowledge of genetic variance can guide the selection of appropriate drugs or dosing tailored to an individual's specific circumstances. This, in turn, may reduce the chance of adverse events, maximize the probability of better health outcomes and diminish costs.4, 5, 6
In some instances, tests select which patients should or should not take a particular medication. For example, tests are used in conjunction with the breast cancer biologic trastuzumab to detect patients whose tumors overexpress HER2 protein.7, 8 In other cases, tests are used to predict the probability of adverse events associated with use of a particular drug.9 For example, there is a test that links hypersensitivity reactions to the human immunodeficiency virus/acquired immune deficiency syndrome drug abacavir to a specific genotype. Also, there are tests that suggest ways to modify dosing in patients with an innately poor ability to metabolize a certain drug. In the case of warfarin, for instance, tests detect variation in the way individuals metabolize the blood-thinning agent. This could help optimize dosing.
Some personalized medicines are developed pharmacogenomically, concurrently with companion diagnostics. For example, trastuzumab was co-developed with a diagnostic test, which was Food and Drug Administration (FDA)-approved and recommended before prescribing. However, most marketed personalized medicines were not co-developed.10, 11, 12 Rather, diagnostics are often developed post hoc as a way of personalizing a drug, for example, abacavir, irinotecan and warfarin.13
Ideally, a companion diagnostic, which is intended to inform the use of a drug in development, will be studied in parallel with that drug, in phases I or II. Here, clinical studies allow for the evaluation of the drug's safety and efficacy, and concurrent verification of the biomarker's clinical utility in guiding drug use and patient selection. If the sponsor is ‘reasonably sure’ that only patients having a specific biomarker status will respond to the drug in development, the sponsor may use biomarker status as a selection criterion. In this case, approval of the drug is ‘integrally linked to availability and use of the diagnostic at the time of its approval.’14 On the other hand, where the therapy's risk/benefit ratio is acceptable in the entire target population, parallel development may not be seen as necessary.14
Despite the success stories, pharmacogenomics has had limited impact on clinical practice to date. The list of FDA-approved companion diagnostics is still relatively short, perhaps a dozen or so.15 Approximately 10% of currently marketed drugs have pharmacogenomic information on the label, that is, mention the influence of genetic variation on drug response or safety.16 To date, testing is only required for trastuzumab, cetuximab and panitumumab.16
There are significant clinical, financial and ethical barriers to the successful implementation of pharmacogenomics. Scientifically, the process of biomarker discovery and validation has been disappointingly slow. Additionally, regulatory and reimbursement issues have been problematic, particularly with respect to companion diagnostics, but also drugs that lack clinically effective diagnostics. To illustrate, gefitinib only works in about 10% of patients with advanced non-small cell lung cancer.17 In June 2009, FDA partially withdrew the drug, no longer allowing its prescription to new non-small cell lung cancer patients, as no useful epidermal growth factor receptor biomarker test is (yet) commercialized in the United States to identify positive responders.
Even with FDA-approved biomarker tests that appear to be clinically useful lingering questions persist regarding their linkage to health outcomes.18 A case in point is warfarin, which illustrates the acute translational gap that exists between knowledge and application.19 Tests suggest patients deficient in a certain enzyme activity (CYP2C9) may require a lower warfarin dose or more frequent monitoring, and may be at risk for bleeding episodes.20 Since 2007, FDA has been recommending genotyping for all patients being prescribed warfarin. In spite of this, in April 2009, the Centers for Medicare and Medicaid Services (CMS) decided not to routinely pay for genetic tests intended to help doctors determine the best dose of warfarin.21, 22 CMS claimed there is not enough evidence to prove that use of the tests improves patients’ health. As a result, CMS stated that tests would only be paid for if patients enrolled in a specially designed post-marketing clinical trial, under its so-called coverage with evidence development program. In this context, additional post-marketing data will be entered into specially designed registries. Over time, CMS policymakers will evaluate registry data to revisit coverage decisions on the warfarin tests.
This paper's main objective is to improve our understanding of the clinical and economic challenges facing developers of and payers for personalized drugs and companion diagnostics, on the basis of evidence gathered from publicly available databases and elicited directly from pharmacogenomics experts, payers and drug/diagnostic sponsors. Section ‘Materials and methods’ outlines the methods used in this paper. The results are presented in the next section. We conclude by way of a policy discussion in the last section.
Materials and methods
In a Medline search, we used the keywords ‘pharmacogenomics’ and ‘personalized medicine’ to retrieve peer-reviewed articles and to identify 22 lead investigators in peer-reviewed publications as independent pharmacogenomics experts, unaffiliated with either the biopharmaceutical or health insurance industry. Based on this search, we also randomly selected eight personalized medicines. We used the Thomson Micromedex DrugPoints database to find information on all eight drugs.23 Additionally, we obtained all relevant cost-effectiveness analyses (CEAs) contained in the CEA Registry—a publically available database of over 2000 different cost-utility analyses published in English language peer-reviewed journals from 1976 through September 2009—for each drug by searching under the drug's generic name.24 The CEA Registry is maintained by the Center for the Evaluation of Value and Risk in Health, at Tufts Medical Center, Boston, MA, USA.
Health plan coverage review
For self-administered drugs—part D—we gathered information on reimbursement by 25 leading Medicare prescription drug plans and pharmacy benefit managers.25 We used the Medicare beneficiary population as a benchmark because (a) formulary data are publicly available CMS Medicare Prescription Drug Plan Finder;26 and (b) given the disease profiles associated with the eight personalized medicines, there is disproportionately high utilization of all eight drugs and diagnostics by Medicare beneficiaries. Per plan, we examined the formulary with the highest Medicare beneficiary enrollment: generally a four-tier formulary. We determined coverage for all eight drugs, drug price per defined daily dose or treatment cycle, patient cost sharing (co-payment tiers, co-insurance percentages) and conditions of reimbursement, including prior authorization, step therapy, quantity limits and indication restrictions. Finally, for physician-administered drugs, we conducted a Medicare part B National Local Coverage Determination (NCD and LCD) review.27 We obtained NCD and LCD information through the publically available Medicare Coverage Database.
We first sent questionnaires to pharmacogenomic experts. In all, 14 of the 22 lead investigators we selected agreed to be surveyed. Subsequently, we designed a separate survey for payers and drug/diagnostic developers, respectively. We sent the payer survey to 21 leading Medicare part D plans and pharmacy benefit managers, for which we conducted a coverage review. Twelve responded. We sent the drug and diagnostic sponsor survey to all eight drug and accompanying diagnostic manufacturers (eight drug companies and six diagnostic firms). Four drug developers responded; two diagnostics developers responded.
Table 1 enumerates key characteristics of the eight case study drugs. First, we list the broad therapeutic category to which each drug belongs. Six are anti-neoplastics, one an anti-retroviral and another an anti-coagulant. Second, we tally the numbers of labeled indications per drug, as well as report whether a companion diagnostic was co-developed. Six of the eight drugs have multiple approved indications. Only one—trastuzumab—has a co-developed, FDA-approved companion diagnostic. FDA-approved diagnostics for six of the others were developed post hoc. One drug, gefitinib, does not have an FDA-approved companion diagnostic. Third, for each drug, we report off-label uses cited in Medicare-recognized compendia: American Hospital Formulary Service Drug Information, Thomson Micromedex's DrugPoints, National Comprehensive Cancer Network Drugs and Biologicals Compendium and Elsevier Gold Standard's Clinical Pharmacology Compendium. Three out of eight have no officially recognized off-label indications, while four of the five with off-label indications have multiple, officially recognized off-label uses. Fourth, we enumerate numbers of peer-reviewed clinical- and cost-effectiveness studies cited in the CEA registry, accompanied by the range of cost-per-Quality Adjusted Life Year (QALY) estimates referred to in the studies, as well as whether or not these studies included companion diagnostics in their calculations. Here, cost-per-QALY refers to the incremental cost of treatment with the drug in question per QALY gained. On average, about 50% of the studies included the clinical- and cost-effectiveness impact of companion diagnostics. Across the eight drugs, there was significant variance in terms of numbers of studies per drug, inter- and intra-drug cost-per-QALY estimates. Trastuzumab, imatinib and warfarin led the pack in terms of numbers of clinical- and cost-effectiveness studies.
Table 2 lists biomarkers associated with each drug, as well as the relationship between the biomarker and the drug's indication. In the first column, we mention whether FDA had ‘required,’ ‘recommended,’ or given advice in ‘information only’ form on the label regarding the companion diagnostic. Until recently, FDA categorized diagnostic testing as such. ‘Required’ was employed where only patients with a certain biomarker are likely to respond to the therapy. ‘Recommended’ was applied where the diagnostic provides information regarding adverse events associated with the therapy, but is not used in patient selection. ‘Information only’ labeling generally has been used where the diagnostic guides dosing.
Medicare part B coverage of the three physician-administered drugs in our sample—trastuzumab, irinotecan, cetuximab—is virtually automatic following FDA approval. There is a fixed rate of 20% co-insurance on all Medicare part B drugs. Many Medicare beneficiaries have MediGap or other forms of supplemental insurance to reduce levels of cost sharing below 20%. Currently, once FDA approves any physician-administered drug for marketing, it is generally covered for its indications by part B, unless a local contractor issues a non-coverage decision or CMS issues an NCD of non-coverage. An NCD is a nationwide determination by CMS of whether Medicare will pay for an item or service. An LCD refers to a determination by a fiscal intermediary or a carrier under Medicare part B on whether or not a particular item or service is covered. Restrictions on part B drug coverage may appear in LCD and NCDs. However, we found no NCD for any of the eight drugs, or their companion diagnostics. Furthermore, a search for LCDs referring to one or more of the drugs in our study turned up three; one for trastuzumab, one for irinotecan and one for cetuximab. Our review of Medicare part B LCDs for trastuzumab and irinotecan reveals that both drugs were covered without any caveats for both on- and off-label indications. Although a genetic test was mentioned in the trastuzumab LCD, neither LCD discussed reimbursement of genetic tests. On the other hand, an LCD on cetuximab recommended against off-label use, and implied that use of a genetic test to distinguish responders would be mandatory before reimbursement. No mention, however, of the test's reimbursement was given, in any of the LCDs. The specific language in the LCD for cetuximab typifies use of the ‘medical necessity’ clause as it relates to patients with or without the K-RAS mutation: ‘Medicare considers cetuximab medically necessary for treatment of beneficiaries with metastatic colorectal cancer with appropriate analysis of the K-RAS mutation. Use of cetuximab is not recommended for the treatment of colorectal cancer patients with this mutation.’
Our formulary review examined the formulary with the highest Medicare beneficiary enrollment for 21 leading Medicare part D plans, with a total of 30 million covered lives, or approximately 80% of Medicare beneficiaries enrolled in part D. Formulary placement of self-administered drugs—part D—is qualitatively different from physician-administered drugs (Figure 1a). Coverage and conditions of reimbursement of the five self-administered (Medicare part D) drugs are at the discretion of each payer, subject to federal oversight. This contrasts with Medicare part B coverage of the three physician-administered drugs, where coverage is virtually automatic following FDA approval, with patient co-insurance equal to 20%. Gefitinib is the only self-administered drug not covered by some plans. It is the exception to the rule, as all other drugs are covered by all plans. Non-coverage of gefitinib may be due in part to unresolved regulatory difficulties regarding the companion diagnostic.
Warfarin and abacavir are the only two multi-source drugs. Warfarin is universally covered in the lowest cost share tier, while abacavir is also universally covered, with 95% of plans designating abacavir in their middle tier(s), and 5% in their highest tier. In this small sample, generic availability appears to correspond to a lower tier placement, as well as few restrictions. At the same time, the higher the price of a drug, the more restrictions there are, and the higher the cost sharing.
Step therapy was not used for any of the drugs. Nor did plans employ indication restrictions. Prior authorization and quantity limits were not used for warfarin and abacavir, but were used by some plans for the three biologics (Figure 1b).
There is little easily accessible data about payer reimbursement decision-making processes regarding personalized medicines and companion diagnostics. Our survey attempted to pry open that black box. For example, because information on reimbursement of diagnostics is not publicly available, we had to elicit coverage specifics from our survey of payers.
We found that 83% of the 12 payer respondents believe they should be permitted to limit reimbursement of certain drugs to patients whose test results indicate they are more likely to benefit. However, while payers’ intention may be to reimburse differentially, only a minority instituted explicit differential reimbursement depending on test results, and only a handful of respondents required documentation of tests, even with drugs for which the test is required by FDA before prescribing. Some payers did not reimburse the test even when it was recommended (Figure 2).
Many payers expressed doubts about the clinical usefulness of many of the companion diagnostics. More importantly, even if payers viewed a diagnostic as clinically useful, some regarded the conclusiveness of test evidence as inadequate. That is, in certain cases, there appears to be a discrepancy between (a) diagnostic test accuracy, and (b) conclusive evidence establishing a link between the diagnostic test and health outcomes. In other words, it may not matter to payers that a test accurately identifies a sub-population that has a particular genetic mutation, if it does not lead to improved health outcomes (Figure 3).
The virtual unanimity across payers is striking with regard to the degree to which criteria are strongly considered in payer pharmacoeconomic evaluations of diagnostic tests, including clinical utility, health outcomes and scientific establishment of a link between test results and drug response (outcomes). On the other hand, the cost of tests is not a strongly considered factor. It is also noteworthy that medication compliance and off-label use are not important considerations (Figure 4).
In all, 13 out of 14 experts agreed with the statement that pharmacogenomics at its most basic level is ‘the use of genetic information to guide drug development and drug therapy.’ In total, 11 of 14 experts attributed the most important barriers to future development of personalized medicine to science; while 3 believed economics, and reimbursement in particular, presented the most formidable obstacles.
In terms of the pharmacogenomics pipeline and personalization of drug development through genetic testing, experts foresee moderate growth in the coming 5 years, particularly in post-hoc development of companion diagnostics (Figure 5).
Across all eight drugs, reimbursement, specifically formulary placement and patient cost sharing, is seen as a significant barrier to developers and patients, particularly for the higher cost biologics. With the exception of gefitinib, experts generally did not think there were significant regulatory approval barriers, nor were a drug's comparative clinical effectiveness or cost of diagnostic test seen as barriers from the developer vantage point. Across all eight drugs, only two significant barriers to developers were cited; these included reimbursement and lack of evidence supporting an effective diagnostic (Figure 6).
All 14 experts considered gefitinib's lack of clinical effectiveness to be a very significant barrier to payers, while all but 1 considered lack of evidence supporting gefitinib's diagnostic to be a very significant barrier. Irinotecan and cetuximab followed in terms of perceived lack of evidential support, which was viewed by experts as a barrier to payers.
All experts saw scientific establishment of a link between genotype and drug response as a very significant criterion in clinical and cost-effectiveness evaluations of tests. Furthermore, most experts view the relationship between genetic variation and response variability as having significant impact on diffusion, as is the potential to address unmet need. Perhaps surprisingly, off-label and supplemental indications were not seen as having significant impact on diffusion.
Five out of six sponsor respondents considered the following statement as an accurate depiction of pharmacogenomics: ‘It is often assumed that pharmacogenomics proceeds from the observation that exposure to a drug generates a differential response, identifying the predictive marker for that response and then creating a diagnostic product that will be co-marketed with the drug.’ This appears inconsistent with the reality that co-development is still a relative rare occurrence.
Sponsors reported the importance of pharmacoeconomic evaluations on termination and pricing decisions. However, they did not view off-label and supplemental indications as having significant impact on diffusion.
Marketing approval is a necessary condition of access to personalized drugs and companion diagnostics, but not a sufficient one. Payers serve as the linchpin for market access; a key bridge between drug/diagnostic development and clinical adoption. They mediate the market for pharmacogenomic therapies through their reimbursement policies, which allow or restrict access.28 Payers want evidence that therapies add value; that is, are clinically (and cost) effective relative to existing therapies, whether with or without companion diagnostics.
If on a consistent basis payers could accurately identify patients who will respond to drug therapies then such treatments would become a better value proposition. That is, based on test results, payers could limit coverage of personalized drugs to subgroups of patients who are much more likely to benefit.29 However, as we have seen from the survey, this is where there is a potentially significant translational challenge, which has largely been unmet: numerous diagnostic tests have proven clinically useful as biomarker identifiers, yet some payers do not see it that way, at least not for all diagnostics. Furthermore, many payers believe tests, even clinically useful ones, lack conclusive evidence with respect to their impact on health outcomes. In other words, in certain instances, the T1 translational problem of science to development of drugs and tests may have been solved, but not the T2 translational issue of, where appropriate, moving to adoption of marketed therapies in clinical practice and reimbursement by payers.10
Further buttressing the argument that test evidence may not be good enough, or may not even matter in some instances, is the fact that merely a handful of payers in our survey institute explicit differential reimbursement depending on test results. And, only a minority of survey respondents require documentation of genetic testing, even with respect to drugs for which the test is recommended by FDA before prescribing. Moreover, we found that some payers do not reimburse the costs of tests when they are recommended. It would appear that tests only matter insofar as they impact outcomes by effectively personalizing the prescribing process, and not merely as a way of stratifying populations without prescribing implications.
We see that for Medicare beneficiaries, access to physician-administered personalized drugs is relatively unfettered, with a fixed co-insurance percentage for each drug, unrelated to clinical effectiveness evidence. However, this could change as more comparative effectiveness evidence becomes available, which may have a bearing on national decisions, such as NCDs for Medicare part B drugs. Self-administered (part D) drugs, on the other hand, are managed differently. And, as we have observed, access to personalized self-administered drugs varies considerably from plan to plan, often with significant utilization restrictions in place. It is unclear, however, how much of this variance is supported by clinical- and/or cost-effectiveness evidence. In fact, judging from our coverage analysis, variance appears more related to acquisition cost and generic availability than clinical- and/or cost-effectiveness evidence.
On the whole, there appears to be a dearth of consistently supportive evidence for both drug and diagnostic clinical- and cost-effectiveness.30 Furthermore, there is variance in terms of availability of clinical-effectiveness and CEAs. Some drugs, such as trastuzumab, have been seemingly overanalyzed. With others, such as gefitinib, experts have barely scratched the surface. Even in cases like trastuzumab where there is abundant data, surprisingly few CEAs show conclusive evidence as to whether it represents ‘good value’ to society.31, 32
And, as we have seen in our survey, there appears to be no discernible connection between clinical- and cost-effectiveness and patient cost-sharing. If we look at our sample of drugs there is an inconsistency in terms of availability of clinical-effectiveness and CEAs. The CER registry lists 12 analyses on imatinib, but none on gefitinib. Moreover, a widely cited literature review from 2004 turned up just 11 CEAs of pharmacogenomic interventions.33
On the regulatory front, FDA is attempting to resolve T1 translational challenges by coordinating and clarifying the process that manufacturers should follow, including delineating when a companion diagnostic must be approved before or concurrently with approval of the therapy. The motto has become ‘do not ask for approval of a new drug with a molecular indication unless it comes with a proven diagnostic for the targeted genetic anomaly.’34 Sponsors responding to our survey seem to believe in this motto. But, the motto is inconsistent with the reality that parallel development of therapies and diagnostics is still a relatively rare occurrence.
In what may foretell a wave of the future Pfizer has recently submitted an investigational drug called crizotinib to FDA, together with a diagnostic test being co-developed with Abbott to identify patients most likely to benefit from the drug. Crizotinib appears to be successful in preventing the growth of lung cancer tumors with a genetic mutation.35 And, in a promising sign for post-hoc development of diagnostics, last year European Medicines Agency (EMA) announced approval of a companion diagnostic for gefitinib. Subsequently, the clinical- and cost-effectiveness watchdog National Institute for Health and Clinical Excellence published a new guidance recommending gefitinib as first-line treatment for patients with locally advanced or metastatic non-small cell lung cancer if they test positive for the epidermal growth factor receptor mutation, with the newly approved diagnostic.
Analysts have also recommended regulatory initiatives to boost personalized medicine development. In a powerful appeal to establish better conditions for personalized R&D, Jenkins36 suggested creating incentives comparable to the Orphan Drug Act for personalized medicines that target relatively small populations. Although we think this may prove fruitful, we also believe it misses the larger point. It is not a lack of personalized medicines per se, but a lack of successfully personalized medicines, which in turn implies a need for co-development of more effective tests.
Compounding the evidential and regulatory issues is the fact that payers worry about the added costs, which result from testing many individuals to identify the relatively few that may benefit from a particular therapy, or change in dosing regimen. In this context, it is not that per unit tests are expensive—they are not, with some priced as little as $40, and very few over $300 per test—it is the budgetary consequences of reimbursing for every eligible member of the population.37
Should payers begin routinely paying for clinically effective genetic tests to guide the prescription of personalized drugs, then personalized medicine may reach a turning point.38 Further, should medical professional societies incorporate evidence-based testing in their clinical practice guidelines, this may facilitate institutionalization of personalized medicine. A robust example of this occurred in 2008 when human immunodeficiency virus treatment guidelines were revised to incorporate human leukocyte antigen-B*5701 screening into routine care for patients before initiating abacavir treatment.
However, in order to resolve the T2 problem and achieve appropriate clinical uptake as well as favorable reimbursement for genetic tests and targeted therapies, manufacturers will need to bring more and better clinical effectiveness evidence to the table.39 Where is the evidence going to come from?
As part of enactment of the Patient Affordability Act of 2010, comparative effectiveness research (CER) has received a huge infusion of federal funding, culminating in the establishment of the Patient-Centered Outcomes Research Institute.40 If implemented appropriately, CER can be structured to account for individual variability, accessing vast sources of data, and including more studies that incorporate companion diagnostics, correspondingly facilitating the translational process for molecular diagnostics.
Biomarker-linked reimbursement contracts: payers may make reimbursement conditional on results from biomarker tests.41 For example, payers have made reimbursement of trastuzumab contingent on a positive biomarker test that screens for human epidermal growth factor receptor-2-positive patients. Biomarkers can limit the size of a drug's eligible patient population, but can significantly enhance the product's clinical- and cost-effectiveness for the eligible population. A specific example of successful risk-sharing is the agreement between UnitedHealthcare and Genomic Health for the Oncotype Dx test used to predict recurrence of breast cancer.42 UnitedHealthcare agreed to reimburse the test while results were tracked to determine if the test is having the anticipated effect on actual clinical practice, in terms of stratifying patients into those who should and those who should not be on chemotherapy. If the number of women receiving chemotherapy exceeds an agreed-on threshold, in spite of tests indicating patients would not benefit from therapy, UnitedHealthcare would pay a lower price for the test.
For certain drugs and diagnostics with limited real-world exposure, payers may need to consider expansion of coverage with evidence development programs; cover drugs or diagnostics provided beneficiaries enroll in prospective data collection registries.43
In summary, the lack of clinically useful diagnostics may be hindering growth in personalized medicines. In particular, the lack of evidence linking diagnostic tests to health outcomes is causing payers to be skeptical about the clinical usefulness of tests. To improve prospects for personalized medicine, more evidence is needed, but also better integration and co-development of drugs and companion diagnostics.
About this article
Clinical Pharmacology & Therapeutics (2014)