Personalised Medicine

Changing Business Model

Bruce Quinn,  Senior Health Policy Specialist, Foley Hoag LLP, USA

Personalised medicine, in which sophisticated diagnostics guide drug choice, dosing, and patient appropriateness, challenges older business models for both pharma and diagnostic tests. In particular, the diagnostic tests are original inventions and often require substantial original clinical research, which may or may not be intermingled with development costs for the drug.

It has become clear that pharmaceutical development in the first decades of this century will be dominated by the “personalised medicine” concept. Personalised medicine encompasses the development and marketing of molecular tests which help to target the use of pharmaceuticals and biologicals in ways that maximise their effectiveness. Of course, lab tests have always guided diagnostics: for high glucose, a physician will diagnose diabetes and prescribe insulin, and so a lab test was paired with a drug. But personalised medicine, as the term is used, almost always involves carving out new subsets of patients with a particular form of disease, such as Herceptin-sensitive breast cancer.

The interactions between the development of complex tests, drug trials, regulatory approval, and drug marketing will cause substantial shifts in the way healthcare is delivered. One obvious business issue (a smaller market for a more specific drug) has dominated thinking about how personalised medicine will develop, how it will impact the pharmaceutical industry, and indeed, whether it is “good” or “bad” for pharma. Because that topic has been discussed so frequently, the present article will focus on other ways in which a new era of interactions between lab tests and drugs will stress existing business models or require the development of new ones.

We emphasise that there is no single model for clinical trials which incorporates genetic or other complex molecular information. There are at least three basic clinical models (combinations of these models are also possible) and they are as follows:

Evidence-based medicine and diagnostic tests
Throughout the developed countries, government-based and private insurance systems are showing rising interest in the rigorous application of evidence-based medicine to control costs, by reducing unnecessary or low-yield interventions. Clinical trial experts and payer decision-makers are quite familiar with outcomes analysis of therapeutic trials. In general, control and treatment groups are matched and randomised. To the degree possible, subjects and evaluators are masked to the interventions offered. Outcomes and adverse events are statistically compared. Controversies over the choice of clinical endpoints, or the use of clinical versus surrogate endpoints, are debated but gradually move towards a consensus. For example, is a negative Prostate Serum Antigen (PSA) at two years an adequate marker of effective radiation treatment for prostate cancer, or should the therapy be considered experimental until ten-year data are in hand? Even when regulators in different countries differ in their decisions, the basic issues for therapeutic trails, such as the choice of endpoints and the weighing of risks and benefits, are familiar.

The tools for assessment of diagnostic tests are very different, and the usual terms of art like sensitivity and specificity, familiar from college statistics, fall far short when the actual decision point for regulators and payers rests on the clinical utility of a test. For example, “sensitivity” can mean the chemical threshold of a test—does it measure down to 0.1 ng/ml of PSA? Sensitivity and specificity are usually used in statistical sense, describing the performance of the test in known populations of cases and controls. The clinical accuracy of the test depends on the base rate of positive and negative cases in the population at hand, which allows prediction of true and false positives and negatives. Sometimes, other already known characteristics of the patient will be calculated together with test results to give a more sophisticated prediction for the test in the patient at hand (Bayesian statistics). All of these statistics become much more complicated for diagnostic tests that have a spectrum of results, rather than just positive or negative results. Meanwhile, specificity and sensitivity lose much of their meaning for genetic tests that answer whether the patient does or doesn’t have the gene in question; the accuracy of the test is essentially 100 per cent. But even here, because of untested mutations or interactions with other untested genes with superimposed functions, the clinical variance accounted for by the gene may be much less than 100 per cent.

When the results of the test are proposed for clinical practice, different evaluators will differ as to whether the test is well-enough established for clinical use. Often, evaluators will differ sharply on whether enough is known about the test in practice to recommend the test to be used routinely.

Paradoxes of clinical trial ethics
A distinctive problem with diagnostic tests, as opposed to therapies, occurs when retrospective data analysis suggests that a clinical use of the test is likely but not certain to be valid—say, 70, 80 or 90 per cent likely. An example occurred when retrospective studies of clinical trials with EGFR-blocking monoclonals found them to be ineffective if a downstream gene, KRAS, was constitutively activated due to a mutation. Because the results of the retrospective study are incompatible with clinical equipoise, it is impossible to conduct a clinical trial where cancer patients with KRAS-activated tumours are randomised to receive EGFR therapy, a therapy which is very likely to be ineffective for such tumours. However, the state of information at this point may be criticised due to the fact that retrospective trials are fraught with confounding variables and a critic can quickly cite many retrospective conclusions that were invalidated by prospective controlled trials. Payers may question whether the cost of the molecular test should be covered, if its clinical use is uncertain. Although the resulting debate over levels of evidence may seem arcane, this scenario is fairly common with diagnostic tests.

Development risk for complex diagnostics
At present, relatively few pharmaceutical or biotechnology firms have the in-house capabilities to develop innovative molecular diagnostics de novo and carry them through commercialisation. Therefore, when tests are used as early as clinical trials, an outsourcing contract or a more sophisticated contractual partnership exists between a test developer and the drug developer. Contractual issues can become very complex and go beyond the scope of this article. But as one example, the freestanding test developer must develop a commercialisation-ready test by the beginning of Phase II. This is because regulators will be very sensitive to technical variations between the test used in clinical trials and the test that will be commercially available after drug launch. Therefore, most of the developer’s sunk costs occur early, although the drug candidate has 90 per cent risk of failure that any drug has at the beginning of Phase II trials. The drug manufacturer will be concerned about lock-in to a contractual relationship with the test manufacturer, while the test manufacturer will worry about a competitor who could produce a rival test after the risky and costly proof-of-concept stage has been passed. Intellectual property on the diagnostic test alone may be more difficult to defend, even in the short term, than the pharma’s core patents on the molecular structure of the companion drug or biological.

Regulatory challenges for complex diagnostics
Complex diagnostics raise a number of regulatory challenges. For example, several tests at the forefront of personalised medicine are too complex to be packaged as kits. An early example is the Trofile test (Monogram Biosciences), a gateway test to the use of a new-generation HIV anti-viral, maraviroc (Selzentry). The test requires gene-splicing steps and is currently run at one centralised and standardised laboratory. In the United States, this category of test is called a “laboratory-developed test” or LDT. In other cases, the lack of a “gold standard” test or variability of testing between laboratories has raised questions about the accuracy of diagnostics even for the prototypic personalised medicine test, the Her-2-neu test (See Fitzgibbons PL et al., Arch Pathol Lab Med 2006, 130:1440-5, and references therein).

Another regulatory challenge which may have seemed like science fiction only a few years ago is the prospect of prescribing cancer drugs based on oncogene characteristics rather than gross tumour type. Today, clinical trials for cancer drugs are categorised by the type of cancer: small cell lung cancer, adenocarcinoma of the pancreas, renal cell carcinoma, and so on. However, we are already seeing targeted cancer drugs which are effective in ways that are completely unforeseen by legacy histologic classifications of tumours (Both chronic myelogenous leukemia and gastrointestinal stromal cell tumours respond to imatinib, if they express the bcr-abl oncogene translocation). Although there are some highly effective cancer / chemotherapy regimens, the impact of chemotherapy on many cancers is notoriously limited (e.g. 5-10 per cent of patients benefit, or alternatively, the number-needed-to-treat is 10 to 20 or more.) Therefore, the hurdle rate for results matching tumours to chemotherapy based on molecular expression panels (e.g. tumour X in a case that also carries the bcr-abl mutation) should be equivalent efficacy. But all regulatory conventions, and payer guidelines for chemotherapy coverage, are based on the legacy approach to classifying tumours histologically and studying them in drug-specific trials.

Economic hurdles for complex diagnostics
Laboratory tests have traditionally been treated as commodities in the medical marketplace. In most countries, costs of laboratory tests are either bundled with episodes of care or paid at fixed rates based on the chemistry of the test (e.g. nucleic acid amplification, US$ 20; serum immunoassay, US$ 25; flow cytometry, US$ 50). These fixed fee schedules appear to be adequate for the development of new tests of the same type when the main parameter of the test is accuracy. Fixed fee schedules also encourage technological change to produce the same types of tests faster and less expensively. However, laboratory test reimbursement that is administratively locked to the marginal cost of the test’s chemistry is incompatible with significant clinical trials to develop and launch a novel type of test. In economic terms, this can be very inefficient, penny wise and pound foolish. For example, a hypothetical new test which could save the healthcare system US$ 100 million over a few years costs US$ 10 million in clinical trials to develop and needs to be sold at US$ 300 to cover risk, investment, and marginal cost; but the test will never exist if the reimbursement is locked at US$ 20 and the US$ 100 million will never be saved. New regulatory schemes for payer systems will need to adapt to some form of value-based reimbursement at least where the net outcome is to encourage cost-saving forms of investment.

One bright spot in the economic logic does occur if the healthcare system bundles a spectrum of interlocking costs together, such as the costs of cancer chemotherapy and chemotherapy diagnostics. Here, the test manufacturer could command value-based market prices in its direct transactions with chemotherapy centres, if the net outcome of test usage was ultimately cost-saving for the chemotherapy centre.

Challenges in marketing

Historically, there has been relatively little advertising or detailing of laboratory tests, probably because there were undifferentiated commodity products with low margins. Therefore, there is little apparatus in place to educate physicians about new molecular diagnostics, and older physicians may find the topic of molecular diagnostics quite confusing. One solution could be integration of electronic medical records and e-prescribing programmes with pop-up information on relevant diagnostic tests, or flags in the healthcare system that hold a prescription until a relevant diagnostic test has been performed.

In addition, diagnostic tests have a rare reverse-supply chain configuration. Instead of manufacturing a drug or device in a factory, shipping to a regional warehouse, and then a retail location (pharmacy or hospital), a diagnostic test requires shipping of the test from thousands of doctors’ offices or clinics backwards to a central laboratory. The logistics become quite formidable for very complex diagnostics that are only performed at one or a very few locations, and especially if the samples must be chilled or frozen at the point of collection and during shipping.

Moving forward
Speculation on the slow growth rate for personalised medicine focusses too much on the supposed reluctance of the pharmaceutical industry to investigate diagnostics which could carve down the market size of new drugs. When discussion stops there, the analysis has fallen into a mental trap, in which one credible answer is found quickly and this halts the search for alternative and perhaps more important explanations. In fact, in the past year, several CEOs at leading pharmaceutical firms have stated that personalised medicine—the pairing of diagnostics and drugs—has to be a core competence of their development strategy. Numerous additional barriers to development and commercialisation were described in this article, but they can be dealt with by good policy and appropriate innovation in the regulatory process. Only by elevating the addition problems into view will they become part of the dialogue on personalised medicine and part of our solution kit in moving the healthcare industry forward for more effective and accurate patient care.

Clinical models

  1. Drug targeted by clinical trial
    The already-classic example of this approach is the identification of Her-2/neu as marker for breast cancer patients who are likely to respond to trastuzumab (Herceptin). Alternately, we could classify Her-2 / neu-negative patients as ruled out for trastuzumab therapy. There are already several examples of this model in oncology, some of which emerged only after the drug's regulatory approval. For example, increasing evidence suggests that cancer monoclonals which target the EGFR receptor are ineffective in patients whose tumour has a mutation downstream of EGFR which tonically activates the KRAS gene regardless of whether the monoclonal neutralises the surface receptor for EGFR.
  2. Drug rescue by identification of adverse events
    Many drugs cause distinctive adverse events only in a minority of patients. To some extent, this is a truism: otherwise the drug candidate would not reach the market. A molecular cause for the uncommon adverse events might now be identified either before launch or during post-marketing surveillance. An example is identification of the HLA B*5701 genotype, which is associated with serious adverse events in response to abacavir (Ziagen). This test became a recommended diagnostic only several years after the drug's launch, and improved the acceptance of the drug. It also helped in avoiding the prescription of the drug in the subgroup of patients at risk.
  3. Third-party innovator differentiates members of a drug class
    A third-party innovator (or academic laboratory) identifies genes which allow choice of the drug in a drug category which is best suited to individual patients. For example, genes which optimise choice of statins could be identified and commercialised. Such a test, if developed, would allow the genetic profile of patients to be assessed so that a physician can prescribe the statin which is most likely to be effective. So far, we lack clinical examples of this model. The most likely reason is limited motive for this clinical trial investment by any single pharma or diagnostics company, since a pharma may lose market share and a diagnostics company may be unable to set a price high enough to repay the trial. Alternately, the a priori development risk may seem too high, since there is no certainty that a small gene panel could be found and commercialized that would classify statin patients effectively. We can use three categories to highlight the diversity of gene types that may contribute to personalised medicine. Like the three clinical models just discussed, the three gene categories shown here are not absolute, but they do illustrate the diverse biologies which are being studied to support personalised medicine.

Molecular models

  1. Enzymes of metabolism and transport
    Across many different drug classes, human beings are remarkably diverse in their metabolism and transport of drugs. For a given drug, a patient may be a typical metaboliser, slow metaboliser, or ultrametaboliser. In the United States, the FDA has remarked on pharmacogenetic information (usually related to metabolism) of over 100 drug labels (Frueh FW et al., Pharmacotherapy, 2008, 28:992-8). However, very few of those drugs have clear labelled instructions for changes in dosage or carry a direct recommendation that testing should be undertaken before prescription. Genes related to tamoxifen metabolism, warfarin metabolism and pharmacodynamics are currently being investigated for clinical utility.
  2. Drug target
    Molecular analysis of drug targets has found the quickest clinical application in chemotherapy. Examples include expression of Her-2/neu as a marker of response to trastuzumab (Herceptin), expression of the EGFR receptor or related downstream mutations as markers of response to cetuximab (Erbitux), and expression of the bcr-abl rearrangement with respect to imatinib's (Gleevec's) effectiveness in individual patients.
  3. Markers of adverse events (other than metabolism)
    Examples in this group include hypersensitivity reactions (the HLA B*5701 variant and reaction to abacavir) and idiosyncratic adverse events such as myopathy (an infrequent adverse event caused by statins) or rhabdomyolysis (a life-threatening but extremely uncommon event caused by statins). Ideally, early identification of vulnerable patients could allow clinical trials to proceed and the drug to be marketed, if always paired with a genetic test. On the other hand, for highly toxic but extremely rare genotypes, there are unwelcome economic issues such as number-needed-to-test to obtain a better outcome. For example, it is impractical to give a US$ 200 genetic test on 10,000 patients to identify one rare patient who would be vulnerable to a severe adverse event.
    The design and analysis of therapeutic trials is costly and complicated, and so is the conduct of diagnostic test development, but for different reasons. The business dynamics and regulatory requirements of both therapeutic trials and diagnostic test trials must be satisfied to bring a combination test and drug to market, while meeting prospective estimates for development risk and likely economic return. Putting all of these factors together may mean that the net risk and complexity is actually squared or cubed relative to a more routine therapeutic trial. Six ways in which potential difficulties manifest themselves include:
  • Evidence-based medicine
  • Paradoxes of clinical trial ethics
  • Development risk for complex diagnostics
  • Regulatory challenges for complex diagnostics
  • Economic hurdles for complex diagnostics
  • Challenges in marketing.

Author Bio

Bruce Quinn

Bruce Quinn is US physician executive in the law firm Foley Hoag LLP. A former medical school professor and strategy consultant with Accenture, from 2004-2008 he was the regional medical director for the Medicare program in California.