Cyclooxygenase 2 controversy: Thoughts on the process of discovery, development, and assessment of benefit and risk of therapies for chronic immunoinflammatory diseases of uncertain autoimmune etiology


  • Daniel E. Furst,

    Corresponding author
    1. David Geffen School of Medicine, University of California, Los Angeles
    • Carl Pearson Professor of Medicine, David Geffen School of Medicine, 1000 Veteran Avenue, Rehabilitation Building, Room 32-59, Los Angeles, CA 90095
    Search for more papers by this author
    • Drs. Furst and Khanna contributed equally to this article.

  • Dinesh Khanna,

    1. University of Cincinnati Medical Center, Institute for the Study of Health, University of Cincinnati, and Veterans Affairs Medical Center, Cincinnati, Ohio
    Search for more papers by this author
    • Drs. Furst and Khanna contributed equally to this article.

  • Harold E. Paulus

    1. David Geffen School of Medicine, University of California, Los Angeles
    Search for more papers by this author

The association of selective cyclooxygenase 2 (COX-2) inhibitors with thrombotic events, especially cardiovascular events, has been exhaustively reviewed in a recent editorial in Arthritis & Rheumatism (1). Differences in the mechanism of action, pharmacokinetics, central nervous system penetration, and doses and durations of therapy have been posited to explain the higher incidence of these events in highly selective COX-2 inhibitors compared with most other nonsteroidal antiinflammatory drugs (1–8). Given the perception that the present drug approval process does not adequately define the rare cardiovascular events associated with selective COX-2 inhibitors, we would like to begin a discussion of a way forward. We view this editorial as a starting point and hope it generates further discussion and refinement.

Present Drug Development

In the drug development process, an important function of early phase II clinical trials is to sort out the conditions for which the new agent is and is not likely to be beneficial. Patient selection and trial design should accomplish this task as efficiently as possible, both to minimize the number of patients exposed to the unknown risks of the new agent and to limit the investment needed to determine that an agent is not likely to be useful in a particular condition. During phase II, clear-cut negative results are more helpful than uncertain positive findings, because clear negative results help discontinue development for a given indication and direct attention and investment into similar early phase II investigations of other potential disease indications or other potential agents. Composite outcome measures such as the American College of Rheumatology 20% response criteria (9) and disease activity score have encouraged the conduct of early “go/no go” clinical trials in patients with rheumatoid arthritis. If the proof-of-concept, dose ranging, and efficacy (phase II) studies are positive, efficacy and an appropriate dose range for a potent biologic product (or any medication) should be determined in phase II/III studies with a relatively small number of subjects, perhaps 300–500 in randomized controlled clinical trials.

Establishing safety is much more difficult, particularly for long-term administration to patients with a life-long chronic disease. Common adverse events are usually detected during preapproval clinical trials, but less common or rare serious events are unlikely to be recognized until after approval when tens of thousands or hundreds of thousands of patients have used and misused the agent in the myriad circumstances of ordinary clinical practice. If the adverse event is also common in the general population, a significant relationship to the drug may not be evident until large placebo-controlled studies are conducted in relatively healthy subjects. This is clearly illustrated by the selective COX-2 inhibitors for which a small increased relative risk of cardiovascular events became convincing only after a large placebo-controlled colon polyp study (10) and a yet unpublished Alzheimer's prevention study, both of which were of several years' duration and conducted in relatively healthy subjects.

The drug development and approval process is subject to conflicting tensions: prospective beneficiaries of a new agent want it to be available quickly because the agent may prolong their life or improve its quality, whereas regulatory agencies and public health policy makers want to be assured that even very uncommon or rare adverse events are recognized and quantified, often requiring a prolonged approval process. The former tensions are illustrated by the successful lobbying of acquired immune deficiency syndrome (AIDS) activists for the accelerated release of developing AIDS drugs at a time when there were no available treatments to prevent the early death of patients with AIDS. The latter tensions are illustrated by the Vioxx situation in which a modest increase in the absolute risk of a cardiovascular event from 0.78 events per 100 person-years of treatment with placebo to 1.5 events per 100 person-years of treatment with rofecoxib during >18 months (11) led to its withdrawal and the replacement of manufacturer-sponsored advertisements extolling the benefits of Vioxx with attorney-sponsored advertisements seeking Vioxx users who might be represented in lawsuits.

There is a double standard with respect to the evidence required to establish efficacy and evidence needed to consider a drug to be unsafe. Efficacy is accepted only after it has been confirmed to be statistically and clinically significant in several well-designed, adequately controlled randomized clinical trials. For example, usually 2–3 large (often 1,500–3,000 patients overall) 6–12-month studies are needed to demonstrate efficacy. Safety may be questioned based on a signal perceived by an insightful observer in a mass of uncontrolled, spontaneously reported anecdotes collected in a computer database. Even 1 or 2 serious adverse events may be influential if they occur in particularly vulnerable or sympathetic subjects who can be portrayed as victims in the lay press. Adverse events are generally reported in terms of relative risk, usually without reference to the absolute risk of the event occurring in untreated patients. A rare event that occurs in 1 in 10,000 untreated patients and is seen in 2 patients participating in a 500-patient clinical trial has an apparent but unacceptable relative risk of 40, although the apparent absolute risk of 1 in 250 may well be acceptable to a patient who expects to derive substantial benefit from the treatment, especially when one considers the degree of uncertainty surrounding both the 1 in 10,000 and the 2 in 500 risk estimates. In the rheumatic diseases, the ratio of risk to benefit must strongly favor benefit. Although a specific ratio is not possible across drugs, the ratio of risk to benefit that was found in the Vioxx Gastrointestinal Outcomes Research (VIGOR) trial (12) was 1:500 (8,000 patients in the trial with a difference in myocardial infarctions between naproxen and rofecoxib of 16), which was apparently an unacceptable risk-to-benefit ratio. It is possible, although difficult, for individual patients to weigh this sort of current benefit versus potential risk information. It is much more difficult for individuals who are not ill to judge potential benefit versus a potential rare but serious risk, e.g., with prophylactic vaccinations, use of COX-2 inhibitors to prevent colon polyps (future cancer), or to prevent possible Alzheimer's disease.

It is our opinion that new drugs or biologic agents should not be “launched” through widespread advertising. If successful, launching rapidly increases the number of persons exposed to the new product from a few thousand carefully selected subjects in clinical trials to tens of thousands or even hundreds of thousands of persons with all types of comorbidities, concurrent medications, and misinterpreted dosing regimens. This maximizes the probability of harm during the interval between the initial occurrence of and an appropriate response to previously undetected serious adverse events, relative overdosing, and drug–drug or drug–disease interactions.

Staged Approval for New Drugs

To better accommodate persons who desire earlier availability of needed new therapies and those who would delay approval until all risks have been documented, we endorse a system of controlled or staged approval for new drugs. We propose a very thorough phase II development program during which an efficacious dose range and the most common adverse events are established. This phase may need to explore the potential dose range more thoroughly than is usually done presently to clearly establish the lowest effective dose and the highest dose that still avoids frequent severe toxicity. The highest dose may not, of course, be carried forward but will establish the upper limit of the true dose range. At this point, in parallel with the usual phase III program, the drug should be approved for prescription to a subset of patients who genuinely need it because there are no effective standard treatments and/or because the patients have not benefited sufficiently or are intolerant of standard treatments, and to patients who wish to accept the potential risks associated with an incompletely evaluated new drug. Prescriptions would be filled through pharmacy outlets, but only after the patient has signed an informed consent acknowledging unknown risks because the drug has not yet been fully evaluated; has agreed to report to the pharmacist (or report through another mechanism) regarding the indication for the treatment, its relative benefit, and any serious side effects that required medical intervention, hospitalization, or discontinuation of the drug; and has given Health Insurance Portability and Accountability Act consent for medical record evaluation of serious adverse events. The pharmacist would collect this information whenever the prescription is refilled; if the prescription is not refilled, the pharmacist would have received consent to contact the patient by telephone and obtain the same information by brief telephone interview at the time when the original prescription should have been consumed. The sponsor would collect this information from the pharmacists for all patients who receive prescriptions for the new drug.

This process will be associated with increased costs that would have to be included as part of the investment in drug development by the sponsor. However, because patients would be expected to pay for the provisionally approved drug, pricing could be structured so the pharmacist would be reimbursed for expenses to dispense the drug. The sponsor would cover data collection, transmission, and analysis costs in the same manner as for phase II and III clinical trials. The additional costs for this undertaking would be offset by avoiding at least some postmarketing studies (because there would be a much better understanding of these drugs' safety profiles than there is presently). Serious adverse events would require medical records review and documentation by the sponsor's medical staff or a clinical research organization. In this way, one would have an accurate denominator of the number of patients exposed to the new drug, as well as an enumeration of serious adverse events and a general sense of the drug's effectiveness in selected patients who accepted the unknown risks of the new drug in return for access to its possible benefits. Therefore, the profile of both common and rare serious adverse events should be established after use for perhaps 25,000 or 50,000 patient-years using this simple data collection procedure. At that point, the conditions for appropriate use of the drug should be evident, and the drug would be approved for general prescription and promotion.

For example, one estimate of the cost of the Therapeutic Arthritis Research and Gastrointestinal Event Trial (TARGET) (13), which assessed the cardiovascular and gastrointestinal safety of lumiracoxib (Novartis, East Hanover, NJ), is approximately $80 million. The multinational study enrolled 18,324 subjects for 1 year. We estimate that the study cost was approximately $3,000 per patient plus a 33% in-house cost to the company for supplies and monitoring, etc. This conservative estimate does not include the cost of distributing the drug and incidental costs such as institutional review board fees at individual sites. Although there are no actual published data on our proposed approach, we provide a conservative estimated cost to conduct the parallel study design we propose above. A 2-year, parallel study involving 25,000–50,000 subjects will hopefully be able to provide a reasonable estimate of rare but serious side effects. The cost of this approach will be approximately half that of the abovementioned TARGET study ($12 million + $20 million = $32 million) if we assume the following costs: 1) a cost of $2,000 per pharmacist to start up the study, with ∼10,000 pharmacists required to conduct the proposed study (∼$20 million); 2) the cost of manufacturing or distribution of the drugs is the same as the TARGET study; 3) the in-house costs are twice that of the TARGET study because there are many more sites and followup is twice as long, although the followup of the TARGET study costs much less per site, thus ∼66% of the abovementioned $20 million or approximately $12.2 million; 4) patients pay for their usual copay for the investigational agent (conservatively $25 per 3 months × 2 years = $200), which will go for reimbursement to the pharmacists. Although this approach has the potential to require an additional up-front cost during phase III, it will surely nearly obviate the need for a very costly phase IV study and will ensure a much better understanding of the drug when it is actually marketed. This approach need not be mandatory, of course, and might not be the best for smaller markets, where a more traditional, sequential route may be adequate. This approach will also help avoid the problems that occurred during the postmarketing experience with COX-2 inhibitors. Further consideration of this proposal is, of course necessary.

This staged/parallel approval will help determine the true incidence of adverse events because the exact number of adverse events and the total number of subjects treated will be recorded. This is different from the current phase IV drug development program where reporting to the Adverse Events Reporting System in the US is largely voluntary and the incidence of the adverse events is likely to be underestimated. Further, because these data will be consolidated within the regulatory agencies, data from one drug can be used as a control group for comparison with results from other drugs; this would obviously be most appropriate when examining drugs of the same class, but it will even be useful as a measure of background incidence when drugs are not of the same class.

As with the current voluntary postmarketing surveillance registry, the lack of a concurrent control group might make it difficult to distinguish serious treatment-related adverse events from similar incidental events that may occur during the course of ordinary life, and additional studies may be needed to definitively establish cause and effect relationships.

Staged approval of new pharmaceutical agents has several advantages. Patients who urgently need a drug could be treated several years sooner than with the usual approval process. The true incidence of rare adverse events could be determined with a real denominator number of patients exposed and numerator number of events that occurred. The patients treated would be representative of all of the usual comorbidities, polypharmacy, and dosing problems of real-life patients. Similarly, effectiveness of the drug in clinical use could be determined. The expertise of conscientious, professional pharmacists could be used to generate the data using the pharmacists' existing computer systems, and part of the cost of these large simple trials could be included in the price of the drug. This limited-approval phase would not replace the usual phase III studies, which should continue in parallel. Approval for general prescription will probably occur at approximately the same time as with the current system, but with much greater knowledge about its safety and effectiveness.