By: Richard S. Schifreen

There is nothing new about the concept of using diagnostic testing to guide therapy. Diabetics have measured their urine or blood sugar for decades to determine the proper dose of insulin. Pathologists have long used phenotypic histological and immunochemical markers to classify tumors and provide guidance as to prognosis and the most effective therapeutic options. Liver enzymes are routinely profiled to identify patients with adverse reactions to statin therapy. These are only a handful of examples; there are many others in routine use.

Why are companion diagnostics different? Testing for over-expression of the Her2 neu oncogene in order to select candidates for Herceptin therapy, approved in 1998, is considered the first example of a companion diagnostic, i.e., a diagnostic test developed with a new therapeutic agent to enable use of that agent. Patients with breast cancer tumors that over-express the target oncogene are considered strong candidates for therapy since the monoclonal antibody in Herceptin is targeted toward that antigen. Decades ago, breast cancer biopsies were tested for estrogen receptors to identify patients most likely to benefit from therapy with the anti-estrogen drug Tamoxifen. There appears to be little difference between how the two assays are used—so why is one considered a companion diagnostic and not the other?

The term companion diagnostic may be defined differently based on the perspective (application, regulatory status, and nationality) of the user. The term is sometimes used to refer to any diagnostic employed to select or modify use of a therapeutic agent. My definition for this essay reflects the FDA approval process in the United States. In this context, I define a companion diagnostic as a new diagnostic test developed in combination with a new therapeutic where the diagnostic test is essential to select those patients that should receive or continue to receive the drug. The drug and diagnostic are tested and approved together, and one would not be introduced without the other. In FDA nomenclature, the therapeutic and companion diagnostic are a “combination product” as defined in 21 CFR 3.2(e). This definition excludes a new drug where the labeling suggests monitoring with an old test, such as a new statin requiring monitoring of liver enzymes, or recommendations that patients receiving an old drug such as warfarin be tested for polymorphisms in its metabolic pathways. A very recent example of existing testing being applied to an established drug is the FDA-mandated “black box” warning on the blockbuster drug Plavix regarding its lack of efficacy in poor metabolizers.1 Although FDA does not mandate testing for CYP2C19, the implication is clear. It will be interesting to see if this warning prompts physicians to prescribe expensive genetic testing for this polymorphism, or if they will avoid this expense by basing dosage on testing of platelet function, or if they will switch their patients to similar drugs with a different metabolic profile.

 

There are two paradigms for companion diagnostic tests: (1) predictive testing to identify patients likely to benefit from the therapeutic, and (2) testing designed to monitor patients over time to identify adverse effects before they become clinically significant. The small number of approved companion diagnostics, i.e., combination products, currently available fall into the first category. They include testing to select patients most likely to benefit from monoclonal antibody–based cancer therapies. The paradigm is simple: select patients whose tumors express the antigen targeted by the antibody therapeutic. Examples include testing for over-expression of the Her2 neu oncogene (to identify candidates for Herceptin therapy), EGFR (epidermal growth factor receptor) expression as a determinant for therapy with Erbitux, and testing for the presence of the Philadelphia Chromosome for treatment with Sprycel. An exciting new diagnostic application might be using high-
sensitivity CRP to identify patients with normal lipid levels that could benefit from statin therapy.2 Predictive testing might also be based on genetics; for example, testing for CYP2C9 and VKROC1 (Vitamin K epoxide reductase) polymorphisms to assist in determining dosing for warfarin. However, genetic methods cannot respond to environment or other factors, such as disease or the impact of co-administered drugs.

The development of combination diagnostic tests designed to monitor patients over time is less likely due to the requirements for prospective clinical trials that will be discussed below. Nevertheless, this testing model is common in clinical practice. An example is management of patients with diabetes by regular monitoring of blood sugar, or, more recently, HbA1C.

 Another approach to monitoring long-term therapy is therapeutic drug monitoring (TDM). This approach is used for drugs that have a narrow therapeutic window; i.e., the dosage of drug necessary to achieve a therapeutic effect is close to that which may induce toxicity. TDM may include the measurement of free drug, metabolites, or both. Ratio analysis of the drug versus metabolites can identify patients who are slow versus fast metabolizers and further facilitate selection of suitable drugs and dosage. TDM is useful for monitoring a patient on therapy but can only be performed after the patient has received sufficient doses of the drug to reach a steady-state concentration. In that sense, it is reactive rather than predictive. However, TDM is sensitive to all factors that might affect the level of active drug and metabolites in an individual—genetics, environment, drug interactions, other disease, etc. Theoretically, predictive testing and TDM are complementary, although they are not often used together due to the cost of using two methodologies.

 

The fundamental barrier to adoption of a diagnostic test in a combination product is the potential for limiting sales of the new therapeutic to a limited target population. The positive argument is that a companion diagnostic may allow commercialization of a new drug that might not otherwise be approved, or “rescue” a drug that has demonstrated adverse affects in certain patient groups. Herceptin is used as the model where the drug was introduced based on a conservative model of selecting patients most likely to benefit, that ultimately led to blockbuster status as the target patient population expanded based on proof of efficacy. The counter argument is that the use of companion diagnostics will limit the market, especially as competitors introduce products that are more widely suitable and don’t require testing. The recent black-box warning for Plavix may validate this concern.

A second barrier is the complexity of determining and validating the clinical utility of combination diagnostic and therapeutic products. This is discussed in the draft Preliminary Drug-Diagnostic Co-Development Concept Paper issued by FDA in April 2005 (see sidebar) and discussed at the 4th U.S. FDA Workshop on Pharmacogenomics in Drug Development in December 2007.3 The scope of this document is limited to validation of single biomarker assays for a single drug and excludes testing used for determining dosage. FDA also emphasizes that this document is for discussion and not for regulatory implementation. One significant element of the concept paper is that it separately addresses issues related to analytical validation, clinical validation, and clinical utility.

Analytical test validation is similar to that performed for other in vitro diagnostics. The principles are the same whether the diagnostic test is developed in combination with or independently of the therapeutic. Validation must include pre-analytic variables such as sample collection, transport, and preparation as well as the analytical methodology. In some cases, additional issues will need to be considered related to the use of multiplexed assays (e.g., microarrays), instrumentation, and interpretative algorithms embedded in software. FDA and other standard-setting organizations have addressed many of these issues (see sidebar).

The basics of clinical validation are also well accepted in the diagnostics community. The purpose of these studies is to determine how accurately the diagnostic test reflects the clinical status of the patient. Statistical tools such as determination of predictive value, likelihood ratios, odds ratios, and the use of receiver-operator curves (ROC) may be used to characterize the clinical performance of assays. These tools are discussed in depth in the concept paper.

A thesis of the concept paper is that analytical and clinical validation should be performed prior to the clinical trials that will be designed to determine safety and efficacy; i.e., clinical utility. Understanding clinical utility also requires determining the potential harm that might be associated with misclassification of patients. This information will be critical in selecting trial participants and in developing the design of the clinical trials. It will be especially significant in developing the protocol for Phase III trials of diagnostic tests and therapeutics being developed in combination.

One of the most noteworthy discussions in the concept paper is the guidance regarding the use of retrospective versus prospective trial models to prove clinical utility. In many cases, diagnostic tests are validated through studies conducted on a single group of patients at single or closely related institutions. Very often, the studies are conducted retrospectively. Retrospective models are attractive in that samples can be selected to reflect a desired patient distribution, researchers can be certain that all of the supporting information they desire is available, the clinical outcome for each patient is already known, and the samples can be assembled and tested relatively quickly. Typically, researchers will take a portion of the patients to determine cut-off values and then apply these cut-offs to the rest of the subjects. Conclusions based on these studies may be adversely impacted by unknown bias specific to patients at that institution (e.g., local environmental factors or standards of care), unexpected bias related to pre-analytic or analytic methods as conducted at that time at that institution (e.g., tissue collection or processing), and unrecognized chance events that may impact data interpretation.

FDA recognized that retrospective studies are very useful in performing analytical and clinical validation and in designing the definitive clinical studies. However, the concept paper encourages the use of prospective study models to demonstrate the utility of the diagnostic test and therapeutic combination. In this model, the diagnostic test methodologies, including the cut-off values, are determined prior to the initiation of the clinical study. The laboratory conducts and interprets the assays independent of the sponsor using the protocols provided. The diagnostic test results may be used by clinicians according to the study protocol or blinded until the patient outcome is known. These study models are discussed in detail in the concept paper.

 

Ultimately, the value of any laboratory test is determined by its clinical utility. To illustrate these limitations, let’s consider two common examples. The first is a genetic test to assist in determining dosage of a drug by predicting the rate of its metabolism, and the second is a molecular assay to assess the likelihood that a tumor will respond to treatment with a new monoclonal antibody–based therapeutic. The first test determines the sequence of a small portion of a gene that codes for the predominant enzyme that metabolizes the drug. The clinical utility of the first test might be affected by the following non-analytic variables:

• Other drugs being administered to the patient or foods the patient eats may alter metabolism of the target drug.

• The rate of absorption, metabolism, or other critical pharmacokinetic variables may be affected by the patient’s overall health or environmental factors.

• There may be polymorphisms in other metabolic enzymes that result in their having a greater than anticipated activity toward the drug.

• There may be patient-specific changes in the drug receptors, signaling pathways or other cellular functions that impact the activity of the drug.

The second hypothetical test represents the most common model for companion diagnostics. New monoclonal antibody–based therapeutics offer the potential for treating cancer without the debilitating side effects associated with chemotherapy or radiation. However, these antibodies are only effective when there is a suitable antigen associated with the tumor for them to bind to. Theoretically, testing should be able to predict candidates for use of this therapy with a high level of certainty. Experience has shown that different analytical approaches, e.g., immunohistochemistry vs. in situ hybridization, do not always yield the same result. Analytic errors aside, some patients testing negative for Her2 expression have been proven to benefit from Herceptin therapy. The following factors might impact clinical utility:

• Heterogeneity of the primary tumor such that the biopsy specimen is not fully representative.

• Heterogeneity between the primary tumor and metastases.

• Individual differences that impact the transcription of the target oncogene.

• Differences in translation of the mRNA into protein that is the actual target of the therapy.

• Mutations or other factors that might impact regulation by non-coding RNA.

• Impact of unidentified cellular changes in the tumor that might impact efficacy of the therapy.

• Alternative undiscovered mechanisms for the drug to provide a therapeutic effect.

In this case, the clinical benefit is clear when the patient achieves remission following therapy. Harm is most strongly associated with failing to provide the therapy to a patient who might benefit. However, there may also be harm associated with treating a patient who doesn’t benefit and suffers side effects from the therapy. Financial issues may also be significant when there is a high cost to provide therapy to patients who don’t benefit. What level of certainty is required to justify denying patients with terminal illness the potential benefit of a therapy that might significantly impact the length and quality of their lives?

 

There is no question that diagnostic testing will continue to play an important role in selecting patients for therapy and monitoring therapeutic status over time. There is also no question that diagnostic tests will never be perfect but do improve outcomes when used properly. Scientists continue to develop new possibilities, such as the researchers at Stanford who have reported progress in identifying predictive markers for the response of multiple-sclerosis patients to beta interferon.4 The question is whether the pharmaceutical industry will perceive that the cost and limitations of developing therapeutics requiring companion diagnostics as combination products will be offset by the profits of introducing these new drugs.

Pharmocogenetic testing for warfarin is an interesting case in point. FDA-initiated labeling suggests that this testing is of benefit to patients. However, in August 2009, CMS decided that Medicare reimbursement for this testing will only be available to patients enrolled in randomized, controlled clinical studies.5 More recently, Robert Epstein, chief medical officer at the Medco Research Institute, in collaboration with the Mayo Clinic presented a controversial study at the March 2010 American College of Cardiology conference that showed genetic testing for polymorphisms related to warfarin treatment decreased hospitalizations by one-third. What is the message to companies considering development of a companion diagnostic product? The answer will be determined by the policies adopted in the United States and other countries as we struggle with the issues of providing cost-effective healthcare.

 

1. Mega JL, Close SL, Wiviott SD, et al., “Cytochrome p-450 polymorphisms and response to clopidogrel,” New England Journal of Medicine 2009 (4), 354-62. Epub Dec 22, 2008.

 

2. Cushman M, McClure LA, Lakoski SG, et al., “Eligibility for statin therapy by the JUPITER trial criteria and subsequent mortality,” The American Journal of Cardiology 2010 105(1):77-81. Epub Nov 14, 2009.

 

3. Hinman L, Spear B, Tsuchihashi Z, et al., “Drug-diagnostic codevelopment strategies: FDA and industry dialog at the 4th FDA/DIA/PhRMA/PWG/BIO Pharmacogenomics Workshop, Pharmacogenomics 2009 10(1), 127-136.

 

4. Axtell RC, de Jong BA, Boniface K, et al., “T helper type 1 and 17 cells determine efficacy of interferon-beta in multiple sclerosis and experimental encephalomyelitis,” Nature Medicine, Epub 28 March 2010.

 

5. Pub 100-03 Medicare National Coverage Determinations, Transmittal 111, Change Request 6715, December 18, 2009. Available online at http://www.cms.hhs.gov/Transmittals/downloads/R111NCD.pdf.

 

Richard S. Schifreen, PhD, is president and CEO, Platypus Technologies, LLC (Madison, WI). He can be reached at rschifreen@platypustech.com.