a perspective on the regulation of the evaluation of new antithrombotic drugs

A Perspective on the Regulation of theEvaluation of New Antithrombotic Drugs

Robert M. Califf, MD

We have entered a period of great possibility for im-proving clinical outcomes for patients with arterialthrombotic disorders using new antithrombotic agents.A variety of new therapies are now in clinical trials,while several others have just received regulatory ap-proval for use in clinical practice. Because of the increas-ing complexity of the clinical environment and improve-ments in our ability to manipulate biology for potentialtherapeutic benefit, a host of new therapeutic agents willbe available for testing in the next decade, and it willbecome increasingly difficult to determine when a truebenefit has occurred. In this therapeutic field, regulatoryapproval requires a substantial level of evidence fromrandomized controlled trials. “Superiority” trials are rel-atively easy to understand, but placebo-controlled trials

are becoming increasingly difficult to justify when atleast one effective therapy already exists for a disease.Proving superiority over an active control can be difficultand may require a large sample size. Positive-controltrials are aimed at equivalence or noninferiority in aneffort to prove that a significant difference in clinicaloutcomes can be excluded. Another approach to dem-onstrating that the experimental therapy is superior to aputative placebo has been called “sufficiency.” A reviewof recent regulatory decisions demonstrates the need forcarefully constructed clinical trial programs that takeinto account previous experience, interactions, and thecurrent therapeutic environment. Q1998 by ExcerptaMedica, Inc.

Am J Cardiol 1998;82:25P–35P

We are in an exciting era in which multiple ap-proaches to the modification of the coagulation

system for therapeutic benefit are possible. Althoughaspirin recently celebrated its 100th anniversary as atherapeutic agent,1 only very recently has the molec-ular basis for alternatives that could improve uponaspirin been known. Similarly, heparin was discov-ered in 1917, but not until the recent deciphering ofthe coagulation pathways were multiple agonists andantagonists of critical steps leading to thrombin gen-eration, thrombin activation, and fibrin generation andcross-linking possible. The other commonly used an-ticoagulant, warfarin, has been known for some timeto have a definable therapeutic window, but only inthe past several years have the tests for warfarin beenstandardized and the appropriate therapeutic rangesbeen determined.2 Now multiple oral thrombin inhib-itors and antiplatelet agents are being evaluated inmulticenter trials.

Figure 1 illustrates a commonly held view of thecritical steps of coagulation and some of the majortherapeutic agents that are available either in practiceor in clinical trials.3 The number of possible newtherapeutic approaches at any level of the coagulationsystem is simultaneously exciting and frightening. Onthe one hand, never has the potential to improvedisease outcome been so evident; on the other hand,the task ahead overwhelms our current capability forconducting clinical investigation to determine risk andbenefit. At least one agonist and one antagonist foreach of the major therapeutic targets (i.e., protein C,protein S, tissue factor, Tenase, prothrombinase,thrombin, fibrin, von Willebrand factor, glycoprotein

[GP] IIb/IIIa receptors) already have been developed.Each of these targets has multiple feedback loopsconnecting it with other targets. In terms of sorting outrisk and benefit for each, the number of possiblepermutations or combinations exceeds our currentcomprehension.

An example is thrombin, which is a central elementof the coagulation pathway. In addition to its directrole in converting fibrinogen to fibrin, thrombin hasmany secondary effects. It is a direct agonist of plate-let aggregation, and it also enhances the production ofFactor Xa. Conversely, it enhances the activity ofprotein S, thereby exerting an antithrombotic effect.Which of these mechanisms predominates in a givensituation remains a mystery to be unraveled by exper-imental studies and clinical trials.

GENERAL PRINCIPLES OFTHERAPEUTIC EVALUATION

Patients and healthcare providers are interested inpaying for therapies when they provide a tangiblehealth benefit. More experience has been gained in thequantitative evaluation of the health benefits and risksof clinical therapeutics over the past several decadesthan in our entire history. Many therapies that wereassumed to provide a health benefit based on concep-tual models were found to be useless or dangerous.Although exceptions can always be found, a set ofgeneral principles emerges that can guide the meth-odology needed for evaluation of clinical therapeutics(Table I).

Therapeutic benefits are generally modest. It is rareto find an intervention with a treatment effect on aclinical outcome that is.25% on a relative scale. Thisprinciple leads to the inevitable conclusion that weneed to study large numbers of patients. For example,acute myocardial infarction (MI) or acute heart failurecauses mortality in 10% of patients during the first 30

From the Division of Cardiology, Department of Medicine, Duke Uni-versity Medical Center and Health System, Durham, North Carolina.

Address for reprints: Robert M. Califf, MD, Duke Clinical ResearchInstitute, P.O. Box 17969, Durham, North Carolina 27715.

25P©1998 by Excerpta Medica, Inc. 0002-9149/98/$19.00All rights reserved. PII S0002-9149(98)00661-4

days. To demonstrate beyond doubt the effectivenessof a proposed therapy that would decrease the risk ofdeath by at least 25%, a sample size of 6,000–10,000patients would be required. A smaller sample sizewould have at least a 20% chance of erroneouslyconcluding that the treatment was ineffective in aclinical trial.

Unintended targets are common. As therapies havebeen developed based on pathophysiologic models, ithas been relatively straightforward to demonstratewhether the relevant physiologic target is being favor-ably influenced. However, a number of promisingtherapies have been discarded because unexpectedtoxicities occurred during their systemic administra-tion. While these toxicities occur rarely, they can beserious, such as the drug interactions recently ob-served with mibefradil,4 the unexpected biologic ef-fects on cardiac valves with phen-fen (phentermine-fenfluramine),5 and the increased risk of intracranial

hemorrhage with more potent fibrinolytic agents com-pared with less potent agents.6–8

Qualitative treatment interactions are uncommon.Therapies that improve outcome for patients with agiven diagnosis usually are effective in all subgroups,given proper dose adjustment for factors affectingmetabolism. When treatment effect reversals occur,they tend to reveal a fundamental issue in the biologyof the treatment. The magnitude of the treatment ef-fect on a relative scale tends to be constant, whichmeans that the absolute benefit varies as a function ofthe underlying risk.

Quantitative treatment interactions are common. Itis uncommon for a treatment to be beneficial in menbut not women or in overweight patients but notunderweight patients. The combination of infrequentqualitative interactions and frequent quantitative inter-actions means that trials can be designed withoutoverly extensive data collection and that an observedtreatment effect will be generalizable to most patientswith the disease. Furthermore, the treatment effectwill usually be greatest in the sickest patients.

The interactions among agents affecting differentcomponents of the same biologic system are unpre-dictable and require testing. The relation between ab-ciximab and heparin is an excellent example. When

FIGURE 1. Interaction between antiplatelet and antithrombin agents and the process of thrombosis. ADP 5 adenosine diphosphate;ATIII 5 antithrombin III; Epi 5 epinephrine; GP Ib receptor 5 glycoprotein Ib receptor; GP IIb/IIIa receptor 5 glycoprotein IIb/IIIa re-ceptor; GP IIb/IIIa inhibitors 5 glycoprotein IIb/IIIa inhibitors; LMWH 5 low molecular-weight heparin; TxA2 5 thromboxane A2;VWF 5 von Willebrand factor. (Reprinted with permission from Curr Opin Cardiol.3)

TABLE I Principles of Evaluation of Clinical Therapeutics

• Treatment effects usually modest• Unintended targets common• Qualitative interactions uncommon• Quantitative interactions common• Interactions among therapies unpredictable

26P THE AMERICAN JOURNAL OF CARDIOLOGYT VOL. 82 (8B) OCTOBER 22, 1998

the optimal doses of abciximab and heparin are givenindividually, significant reduction in ischemic eventscan be demonstrated, but an excess of bleeding oc-curs.9 When the dose of heparin was incrementallydecreased, many expected both the bleeding and thereduction in events to decrease; instead, the bleedingwas decreased and the prevention of ischemic eventswas actually improved.10 Mixing 2 therapies that havedefined therapeutic ranges when used alone must involveredefining the therapeutic range for the combination.

Consideration of these issues has led to a regula-tory philosophy that is based on measuring clinicaloutcomes while carefully evaluating interactionsamong antithrombotic drugs. Clinical outcomes aredefined as tangible health benefits to the patient. Eval-uation of interactions involves, at the least, substantialsafety data. In the case of mibefradil, substantial in-teractions with compounds metabolized via the cyto-chrome P450 system posed the major problem to safeclinical use. This high hurdle of not only showingindividual clinical benefit, but also fitting safely into acomplex environment, presents a challenge to the clin-ical research community.

INDICATIONS FOR CURRENTANTITHROMBOTIC AGENTS

The development of new therapeutics is complexand expensive; recent cost estimates for introductionof a new pharmaceutical agent average $500 million.These estimates incorporate the cost of all of theagents that failed during development, not just the costof the agent that eventually was marketed. Nonethe-less, there is no question that the process by whichtherapeutic agents receive approval for marketing andadvertising is a major economic, public health, andclinical practice issue. Many potentially beneficialtherapies do not make it into practice because ofuncertainty about the ability to recover the cost ofdevelopment or to meet the regulatory hurdles.

Obtaining a labeled indication to allow marketingof a drug requires considerable effort and expense.Predominantly for this reason, the classic agents, as-pirin, heparin, and warfarin, are not approved formany indications for which they are routinely used.Table II demonstrates their current indications accord-ing to labeling approved by the US Food and DrugAdministration (FDA).

Aspirin is indicated for a variety of situations, butit is important to realize that some of these indicationswere approved just in the past several years due to theefforts of an ad hoc academic group. Using the Anti-platelet Trialists data,11 these academicians built thecase for the use of aspirin and presented it to the FDAfor approval in the absence of a primary industrysponsor.

Heparin is also quite interesting in this regard.Heparin appears to be effective in decreasing the riskof death and MI during the initial hospitalization fornon-ST-segment elevation acute coronary syndrome,and it decreases the risk of thrombotic complicationsassociated with percutaneous intervention. Unfortu-nately, no definitive clinical trials that would meetregulatory criteria for these indications have beencompleted. Due to the absence of commercial spon-sorship with sufficient financial incentive to pursuesuch trials, heparin is not labeled for these importantclinical indications in the United States.

Warfarin has been shown to decrease the risk ofrecurrent events after MI and after prosthetic valvereplacement. In addition, it decreases the risk of strokein patients with atrial fibrillation. It has received FDAapproval for these indications.

THE REGULATORY REQUIREMENTSFDA regulation: The legislation governing FDA

mandates has evolved through a series of nationalcatastrophes. The creation of the FDA in 1938 wasdeemed necessary as a result of nearly 100 deathsfrom contamination of a pharmaceutical with elixir ofsulfanilamide in pharmacies. Because of the obviousimplications to the public health of unregulated dis-tribution of unsafe food and pharmaceuticals, the FDAwas initially charged with assuring the safety of foodand drugs. When the Kefauver-Harris Act was passedin 1962, the FDA was required to assure not only thesafety but also the effectiveness of pharmaceuticals.More recent legislation, the FDA Modernization Actof 1998, has focused on clarifying this mandate andmaking the review process more rapid and more effi-cient.

Many practitioners and much of the public believethat the FDA has responsibility for assuring not onlythat drugs and devices are safe and effective, but alsothat newly approved agents are better than the current

TABLE II Approved Indications by the United States Food and Drug Administration

PCINon–ST-SegmentElevation ACS

ST-SegmentElevation AMI

AtrialFibrillation

SecondaryPrevention

Aspirin u u u — uHeparin

Unfractionated — — — — —Low-molecular weight — u — — —

Warfarin — — — u uGP IIb/IIIa inhibitors

Abciximab u* u* — — —Eptifibatide u u — — —Tirofiban u* u* — — —

* In setting of unstable angina with planned PCI. ACS 5 acute coronary syndromes; AMI 5 acute myocardialinfarction; GP 5 glycoprotein; PCI 5 percutaneous coronary intervention.

A SYMPOSIUM: ANTITHROMBOTIC THERAPY IN CAD 27P

standard. In fact, however, the FDA does not have amandate to require the types of active-control trialsthat would answer comparative questions. Rather, theagency must focus on the issue of whether a proposedtherapeutic agent is superior to nothing; determining therelative advantages of one beneficial approach over an-other has been considered a matter for clinical practice.

Types of endpoints: Because the goal of therapeuticintervention is believed to be a desire to improve thehealth of the patient, the legislative mandate to theFDA has generally been interpreted to mean that abenefit must be shown in a tangible human outcome.Acceptable outcomes can be simply stated as pro-longed survival, absence of critical events perceivednegatively by the patient (e.g., stroke or MI), andimprovement in how the patient feels and functions.Although avoidance of cost is also a tangible issue thatis recognized as important to the patient (and increas-ingly so in the United States), the FDA is not chargedwith evaluating cost effectiveness. Thus, the criticalelement of outcome measurement for registration canbe summarized by the phrase “live longer, feel better,and/or avoid unpleasant experiences.”

Particularly for therapies evaluated by the Cardio-renal Advisory Panel, the use of surrogate endpoints isavoided, except in the cases of blood-pressure lower-ing for hypertension and low-density lipoprotein(LDL) lowering for lipid-lowering agents. Recently,exceptions to this approach have been questioned withthe recall of mibefradil shortly after it was marketedfor the treatment of hypertension and angina. Mibe-fradil was shown to improve treadmill exercise time inpatients with angina and to lower blood pressure inpatients with hypertension. When released into thecomplex environment of clinical practice, however,drug interactions and toxicities surfaced. Perhaps amore real-life trial with clinical outcomes would haveuncovered these issues before marketing.

Statistical inferences: The current FDA legislationcalls for the demonstration of safety and efficacybased on adequate and well-controlled trials. Thisstatement has been understood to mean that, except inrare circumstances,.1 clinical outcome trial is re-quired. If 2 independent trials each demonstrate asignificant clinical benefit with p,0.05, this result isapproximately equal toa 5 0.052, or 0.0025; sincethe trials are looking for a result on the positive tail ofboth studies, this is equivalent to p5 0.00125(0.00253 0.5). Thus, the estimated goal for FDAapproval in the United States is to demonstrate that atreatment has a beneficial effect to the extent hypoth-esized before the start of the study, with a,125 per100,000 chance that the effect could be less than washypothesized before the start of the study due tochance. Importantly, the Clopidogrel vs Aspirin inPatients at Risk of Ischaemic Events (CAPRIE)12 datadid not meet this standard, and the labeling indicatesthat clopidogrel has not been proved to be superior toaspirin. Rather, the basis for approval for the market-ing of clopidogrel was the demonstration that it issuperior to a putative placebo (see below).

The samea value can be achieved, of course, with

a single trial of slightly larger size than the 2 inde-pendent trials combined. Several recent programs, in-cluding that of clopidogrel, achieved approval basedon a single large outcome trial. Often a strategy ofindependent, complementary trials in Europe andUnited States is taken. In the development of bivaliru-din, Biogen elected to perform a single trial but tostratify the randomization into 2 segments, thus pro-ducing independent statistical replication within thestructure of a single trial.13

When mortality is the primary endpoint, the ex-tremea required for softer clinical endpoints may notbe necessary. If a mortality reduction were demon-strated with a larger p value (i.e., 0.01–0.04), it wouldpose a major ethical dilemma to require additionalplacebo-controlled trials. This reasoning has been ap-plied to a broader array of endpoints representingirreversible damage to the patient. Thus, although theprimary endpoint of the Evaluation of c7E3 for Pre-vention of Ischemic Complications (EPIC)9 trial wasthe composite of death, MI, or emergency revascular-ization, abciximab was approved based on a cleardemonstration of reduction in death and MI as irre-versible endpoints within the single trial.

Timing of endpoints: The timing of the primaryendpoint measurement has been a point of contro-versy. Particularly for intravenous antithromboticagents, the duration of infusion is relatively brief (onthe order of days). Based on the mechanism of action,these compounds would be expected to be effectivewhile they are infused but to have little effect after thattime. In a hypothetical clinical trial of acute coronarysyndromes, when the therapy produces a 33% reduc-tion in the primary endpoint in the first 3 days, sub-stantial power exists to demonstrate benefit with amodest sample size (Figure 2). However, assumingthat no rebound occurs and that the treatment has noeffect, either beneficial or detrimental, beyond 3 days,events accrue randomly in both the treatment andcontrol groups. Although the total number of eventsincreases, the p value also increases, while the numberof events prevented per 100 patients treated remainsconstant between 3 days and 30 days.

These figures present a dilemma in designing clin-ical trials and interpreting health benefits. From theperspective of maximizing the chance of proving thatthe treatment has an effect, the 3-day endpoint isclearly superior. In contrast, a treatment that improvesthe patient for 3 days but has no value only 27 dayslater is not a significant health advance; indeed, onecould argue that improvement until the patient is re-leased from hospital followed by increased risk justafter the patient is discharged to home would bedetrimental. An alternative is to make the early timepoint the primary endpoint, with designation of the30-day or longer endpoint as a check to ensure that thetreatment effect is not reversed. This approach is prob-lematic because there is little experience with statisti-cal tests designed to prove that the curves are parallelin follow-up, and the random chance of convergenceis substantial. The sample size to prove that curvesremain parallel in follow-up can be larger than the


sample size to simply evaluate the endpoint at the laterpoint in time.

The usual shape of curves for thrombotic eventsposes another interesting issue for consideration.Since most events and most treatment benefits occurearly, the p value will be smaller when statistical testsare used that give more weight to early prevention ofevents. For long-term therapeutic evaluations, this ap-proach is reasonable, since patients generally wouldprefer immediate benefit to deferred benefit; increasedvalue of early benefit accounts for the practice ofdiscounting in cost-effectiveness analysis.14 However,as discussed above, it is not clear that prevention of anevent on the second day after admission for an acutecoronary syndrome is really of much benefit comparedwith prevention of an event on day 7. In the PlateletGlycoprotein IIb-IIIa in Unstable Angina: ReceptorSuppression Using Integrilin Therapy (PURSUIT)15

trial, the more conservative chi-square test was usedfor evaluation of the primary endpoint at 30 days offollow-up, yielding p5 0.042, while the exact samedata yielded p5 0.032 when the log-rank test wasused.

Population selection: Many therapeutic develop-ment programs have taken the tack of designing piv-otal clinical trials (those used in FDA applications todemonstrate efficacy) as controlled laboratory exper-iments by excluding patients with comorbidities orthose at high risk. When the primary endpoint of astudy is a physiologic measure, exclusion of con-founding factors makes sense. The primary outcomemeasure will be available and extraneous sources ofvariability are decreased. However, when the criticalclinical trials have excluded patients with confoundingfactors and comorbidities and true clinical outcomeshave not been measured, products have been marketedwithout clear knowledge of what the effects will be inpractice.

Another commonly used tactic is to focus on anarrow population in which the treatment is mostlikely to be effective. This method maximizes thechance of a positive study and an approvable new drugapplication. Unfortunately, many of the most impor-tant issues in risk–benefit trade-offs occur in the el-derly, patients with comorbidities, and those receivingmultiple other medications. In ischemic heart disease,the highest event rates are in older patients who oftenhave multiple other medical problems and impairedrenal function. The elderly thus have more to gainfrom effective therapy, but they also generally have ahigher risk of bleeding.16 As recently demonstratedwith the recall of mibefradil, drug interactions andlethal adverse outcomes may not be evident until aheterogeneous group of patients is treated.

What trials can constitute 2 trials for regulatory ap-proval? To guard the public health from random find-ings mistaken for truly positive results, it is critical toreplicate clinical trial results. When clinical outcomesare used for determining therapeutic benefit, however,the cost can be substantial. Responding to the concernabout needing to mount 2 independent studies for eachclinical manifestation of arterial disease, a guidance

document from the Cardiorenal Division of the FDAallows the use of 2 trials in overlapping indicationsthat share a common pathophysiology. The specificexample used is the overlap between unstable anginaand percutaneous intervention. Thus, a trial in patientswith non-ST-segment elevation acute coronary syn-dromes and a trial in elective percutaneous interven-tion could be used to support an application for ap-proval in both areas if the results were consistent andconvincing. Without such an approach, the require-ment for adequate evidence from trials could be inter-preted to mean that approval in each manifestation(acute coronary syndromes, percutaneous interven-tion, ST-segment elevation MI) would require 2 inde-pendent full-scale trials.

THE SPECIAL PROBLEMS OFPOSITIVE-CONTROL TRIALS

When an effective therapy exists for a disease, thedevelopment of a new therapy requires the decision ofwhether to attempt to prove that the new therapy isbetter than the standard, equivalent to the standard, orcan be added to the standard with incremental benefit.Each approach has certain advantages and disadvan-tages. In general, positive-control trials have someinteresting properties that continue to require carefulthought as more studies are done.

Superiority: If there is a belief that the new therapyis truly better, a superiority design should be used.Proving the superiority of a therapy is straightforwardin terms of study design. The clinical trial programshould be configured to demonstrate a clinicallymeaningful benefit of the new therapy. This approachis familiar to clinicians, and the size of the studydepends on the measure of clinical benefit that wouldbe considered worthwhile; that is, if the trial werepositive, clinical practice would change. This measureis termed the “minimally important difference”(MID).

A special regulatory issue emerges when there is a“standard of care” that has not been demonstrated tobe effective. One might argue that such a situationwould automatically lead to a randomization scheduleof the new agent, the standard agent, or placebo. Inmany circumstances, however, randomization to pla-cebo is not tenable because both practitioners andpatients are convinced about the efficacy of the stan-dard agent despite the absence of high-level evidence.

In essence, the hypothesis test in a superiority trialis designed to exclude the null hypothesis (i.e., that thetreatments do not lead to different outcomes) with alevel of certainty of 95–99%. In reality, the boundariesfor an operational definition of the same outcomesbased on the presumed MID are established. If a newfibrinolytic agent was shown to produce an improve-ment of 0.001 in mortality (i.e., 1 patient per 1,000treated), practice would not change unless the newagent was cheaper, more convenient, or also had apositive effect on a nonfatal outcome. The generallyaccepted MID of fibrinolytic therapy for MI is 1 lifesaved per 100 patients treated.17–19

An alternative design to head-to-head comparisons


is the add-on strategy in which standard care is givenand the new agent or placebo is added. This has theadvantages of not taking standard care away from thepatient and of allowing uncomplicated trial adminis-tration because the patient is on either placebo orcontrol. In head-to-head comparisons, either patientsmust be exposed to a double-dummy technique inwhich they take active therapy and a placebo for thealternative therapy or the study must be done in anonblinded manner.

The add-on strategy has been used successfully inheart failure, but it is often difficult to employ in theevaluation of antithrombotic therapy because of theunpredictable interactions of therapies with anticoag-ulant activity. Figure 3 contrasts the approach takenby the OPUS study of orbofiban, an orally active GPIIb/IIIa inhibitor and the first Sibrafiban versus aspirinto Yield Maximum Protection from ischemic Heartevents pOst acute coroNary sYndromes (SYMPHONY)study of sibrafiban, another oral GP IIb/IIIa agent. Inthe OPUS study, orbofiban or placebo was added ontop of aspirin, while in SYMPHONY sibrafiban wascompared with aspirin.

Figure 4 graphically depicts the approach to supe-riority studies and the determination of p values.

Equivalence and noninferiority: Equivalence is amuch more difficult concept to grasp because it is theconverse of superiority testing. Failing to find evi-dence of a difference is not the same as finding evi-dence that no difference exists. To demonstrate thatone treatment is equivalent to another, the MID shouldfirst be determined. What is the minimum differencein outcomes that would cause clinicians and/or pa-tients to choose one treatment over another? Thisfigure, of course, depends not only on the difference inefficacy but also on the difference in cost and sideeffects. As discussed above for fibrinolytic agents inST-segment elevation MI, this difference has consis-tently been approximately 1% in mortality, the MIDsought by the International Study of Infarct Survival(ISIS)-3, Gruppo Italiano per lo Studio della Strep-tochinasi nell’ Infarcto Miocardico (GISSI)-3, and theGlobal Use of Strategies to Open Occluded CoronaryArteries (GUSTO)-I trials.17–19

The statistical framework for equivalence studies isalso the converse of superiority. The goal of the hy-pothesis test is to refute the hypothesis that the treat-ments lead to different outcomes by at least the margin ofthe MID. Technically, absolute equivalence cannot beproved; 2 treatments cannot be proved to be the same.

If the MID is small, equivalence trials require verylarge sample sizes. Table III shows the sample sizes toexclude a 0.5%, 0.75%, and 1% difference with fi-brinolytic therapy.

An intermediate approach has been called the non-inferiority approach. Increasingly, molecular engi-neering will allow the development of therapies thatmay be more convenient, may have fewer side effects,or may be less expensive to manufacture. The clinicalbenefit of these therapies may be less than the MIDbut on the positive side of equivalence. When inves-tigators believe that the treatment is superior but not

likely to exceed the MID, a noninferiority trial can beperformed with a much smaller sample size than anequivalence trial would require to prove that the 2treatments are identical with respect to their effect onthe primary endpoint.

Several interesting features enter into the design ofan equivalence or noninferiority trial. Compliancewith the intended therapies and accuracy in data col-lection and event determination are critical. Whencompliance with the therapeutic regimen is less thanideal in a superiority trial, the power of the study todetect a difference is diminished; thus, the study par-ticipants have every incentive to maintain a high rate

FIGURE 2. The results of 2 hypothetical clinical trials in both ofwhich the primary endpoint is decreased 33%. In Trial A, theendpoint occurs in 30% of the control group (Pc) and 20% of thetreated group (Pt), and requires a relatively small sample size(approximately 250 patients in each arm) to have 80% power todetect the difference. In Trial B, the endpoint occurs in 9% of thecontrol group (Pc) and 6% of the treated group (Pt), an identical33% reduction, but it requires a considerably larger sample size(approximately 1,250 patients in each arm) to have 80% powerto detect the difference. In Trial C the endpoint occurs in 33% ofthe control group and 30% of the treated group, or 3% absolutereduction. The required sample size is much larger.

FIGURE 3. A comparison of the design of the OPUS andSYMPHONY trials. In the OPUS trial, all patients are treated withaspirin and 1 arm receives placebo. In the SYMPHONY trial,there is no background treatment with aspirin, and aspirin isincluded in 1 of the 3 treatment arms.


of compliance. The opposite is true in an equivalencetrial; if neither the experimental nor the control grouptook any of the prescribed treatment, the outcomeswould be equivalent in the 2 groups. Therefore, slop-piness in a clinical trial enhances the probability ofdeclaring that no differences exist. Similarly, any im-precision in the measurement of outcome would tendto increase the likelihood that the 2 treatments lookedthe same, even if they were different.

The preferred method of analysis is different in anequivalence trial compared with a superiority trial. Ina superiority trial, intention-to-treat analysis is pre-ferred, because without it the statistical inferencesbased on randomization lose their validity. Of course,to the extent that compliance drops and crossoversoccur, the power of the study drops as a square of therate of loss of patients treated as randomized. In anequivalence study, intention-to-treat is not conserva-tive. The higher the rate of noncompliance, dropouts,and drop-ins, the greater the chance that the study willshow no differences, even if a true difference in effi-cacy exists.

Figure 5 demonstrates the possible outcomes ofequivalence trials with differing results. It is possibleto conclude that a treatment is statistically inferior toan alternative therapy yet falls within the equivalencerange. In many cases it is hoped that the point estimateis superior to the comparative therapy, thus leading tothe comfortable conclusion that the treatment is atleast as good as the standard treatment. Using confi-dence intervals when presenting data prevents display-ing data as portraying equivalence when they actuallydemonstrate a lack of evidence of a difference. In

study 3, there is no difference in the point estimate,but the confidence limits extend far beyond the MID.

Concern exists about several general issues inequivalence margins. If the MID is set too broadly, aseries of equivalency trials could push the generaltherapy of a disease in the wrong direction (Figure 6).Even with narrowly defined equivalency margins,some risk exists that several consecutive equivalencytrials could result in regression of effective therapies.

A second concern is that the positive-control groupmay be no better than placebo would have been in theparticular trial. If a treatment is superior to placebo ina controlled trial, the confidence limits of the treat-ment effect exclude the possibility of no effect butinclude the possibility that in another sample, due tochance, the difference could be much smaller than thepoint estimate of the original study that found a dif-ference. If by chance the treatment behaves in a lesspositive fashion in a subsequent positive-control trial, the

FIGURE 4. Graphical presentation of superiority trial results. InStudy A, the experimental treatment (Tx) is better than control treat-ment (p <0.05), and 95% confidence intervals do not cross at a riskratio of 1.0. The point estimate for the treatment benefit does notexceed the minimally important difference (MID) and therefore thestudy may not indicate a clinically relevant difference. In Study B,the point estimate for the treatment is consistent with the same ef-fect as in Study A, but the smaller sample size leads to wider confi-dence intervals and more uncertainty. The result is also compatiblewith a detrimental effect of the experimental treatment since the95% confidence intervals cross at a risk ratio 1.0. In Study C, theoutcome is both highly statistically significant and excludes the pos-sibility that the benefit is less than the MID.

FIGURE 5. Graphical representation of equivalence trial results.In Study A, the 2 treatments are definitely equivalent since the95% confidence intervals for the treatment difference do not in-clude the minimally important difference (MID). In Study B, thepoint estimate is that no difference is present in the treatmenteffect, but the confidence intervals do not exclude the MID.Therefore, the study has not excluded the possibility that a clini-cally important difference could exist in outcomes as a functionof which treatment the patient receives. In Study C, the experi-mental treatment is worse than the control treatment, but thedifference falls within the MID. Therefore, the treatments are sta-tistically different, but the difference between them is not clini-cally important, and therefore they are clinically equivalent.

TABLE III Hypothetical Sample Sizes for Equivalence Trials

Baseline Rate

Minimally Important Difference

0.5% 0.75% 1.0%

7% 54,700 24,300 13,7008% 61,900 27,400 15,5009% 68,900 30,600 17,200

10% 75,900 33,600 18,900


experimental therapy could meet criteria for equivalencewhen indeed it is no better than placebo (Figure 7).

Sufficiency: Sufficiency is an even more complexissue. The FDA mandate is to ensure that pharmaceu-tical therapies are safe and effective, not comparedwith alternatives but compared with no therapy. Asmore effective therapies become available, the envi-ronment for testing new agents becomes more com-plex, and placebo-controlled trials become impossible.Furthermore, the size of equivalence trials can beprohibitive. One approach to overcoming this problemhas been to compare the experimental therapy out-come in a positive-control trial with the expectedoutcome of a putative placebo. Figure 8 demonstratesthis issue.

Many conceptual problems must be consideredwith regard to sufficiency. All of the problems withequivalence trials remain problems with the suffi-ciency design, but in addition, the problem of under-standing the outcome of a putative placebo must beovercome. Requirements to calculate the putative pla-cebo outcomes are given in Table IV. A systematicoverview must be available with substantial certaintythat the active control is superior to placebo. Theresults of the overview must be homogeneous. Theexpected outcome of the conservatively or placebo-treated control group must be stable over time, or if ithas been declining over time (as have most eventrates), the relative treatment effect in the overviewmust be constant. Basically, the construct is dependenton the acceptance of a historical control group. How-ever, given the large sample sizes often needed fortrial equivalence trials, there may be no reasonablealternative.

RECENT LESSONS ON COMPLEXITYOF THE THERAPEUTICENVIRONMENT

Reteplase: Reteplase is a recombinant molecule inwhich deletions have been engineered to give it alonger half-life, thus making it available for double-bolus administration. After very early feasibility trials,a series of trials evaluating coronary perfusion wereperformed. These trials indicated that a higher rate ofThrombolysis in Myocardial Infarction (TIMI) grade3 flow was achieved with reteplase at 90 minutes afterinitiation of therapy than with alteplase.20,21 Thus, itwas anticipated that a mortality benefit would belikely for reteplase compared with alteplase, and thesuperiority of reteplase to streptokinase also was as-sumed based on inferences from comparative perfu-sion data for alteplase compared with streptokinase.

To achieve licensing, a single pivotal trial wasconducted. The International Joint Efficacy Compari-son of Thrombolytics (INJECT)22 trial compared re-teplase with streptokinase and found a mortality ben-efit of 0.51% (21.74, 0.73) at 35 days. Although thisdifference was not statistically significant, it was ac-cepted as excluding the possibility that reteplase wasequal to placebo. This inference was based on the sys-tematic overview from the Fibrinolytic Therapy Trialistsdemonstrating a consistent benefit of streptokinase over

placebo.23 If the historical control is accepted, the IN-JECT data demonstrate that reteplase conserves at leasthalf the benefit of streptokinase over placebo, becausethe lower 90% confidence interval for the differencebetween reteplase and streptokinase does not overlapwith the upper confidence interval for the differencebetween streptokinase and placebo. The increase in in-tracranial hemorrhage with reteplase was similar to thatseen with alteplase compared with streptokinase. Based

FIGURE 6. Illustration of several consecutive equivalency trials, eachone showing slightly less treatment benefit than the previous oneuntil Trial 4, which demonstrates no treatment benefit. The treat-ment in Trial 1 is superior to placebo. The treatment in Trial 2 isequivalent to the treatment in Trial 1. The treatment in Trial 3 isequivalent to the treatment in Trial 2 but not better than placebo.

FIGURE 7. Graphical representation of noninferiority trial results.In Study A, the point estimate indicates that the experimentaltreatment could be better than the control, but the confidenceintervals do not exclude no difference. Yet, the confidence limitsexclude the possibility of a clinically important difference (mini-mally important difference [MID]) favoring the control treatment.Therefore, the study has not shown that the experimental treat-ment is better than control; it has shown that the experimentaltreatment is not inferior to the control within clinically reasonablebounds. In Study B, the point estimate is the same as Study A,but with a smaller study, the confidence intervals cannot excludethe possibility of a clinically meaningful difference favoring thecontrol treatment. In Study C, the point estimate excludes a clini-cally meaningful difference but leaves a good possibility that thenew treatment is taking a minor step backward.


on this information, reteplase was released for marketingwithout placebo-controlled clinical trials.

This finding was not, however, directly informativeabout the question of whether reteplase use would leadto similar outcomes compared with alteplase. There-fore, the GUSTO-III trial was designed to determine ifreteplase was superior to alteplase with regard to thesurrogate endpoint of TIMI grade 3 perfusion based

on 90-minute angiography. GUSTO-III failed to dem-onstrate superiority for reteplase.24

Clopidogrel: Clopidogrel is an antiplatelet agentthat appears to prevent thrombotic events predomi-nantly through the inhibition of adenosine diphos-phate-induced platelet aggregation. It is similar toticlopidine, which has already been shown to have abenefit compared with aspirin in selected settings.With very little phase II data, a large clinical trial(CAPRIE) was undertaken comparing clopidogrelwith aspirin for secondary prevention of cardiovascu-lar death, nonfatal MI, and stroke.12 Patients withrecent MI, recent stroke, or evidence of peripheralvascular disease were included. With a sample size of19,185, the trial results indicated that use of clopi-dogrel was associated with a relative reduction of8.7% in the primary endpoint (p5 0.04). Despite theenormous size of the trial, the Cardiorenal AdvisoryPanel of FDA concluded that CAPRIE did not meetthe usual regulatory criteria (p,0.00125) to provesuperiority over aspirin, especially considering thatwhen total mortality was included in the primaryendpoint, the difference no longer met nominal statis-tical significance. However, the result certainly metcriteria for equivalence, and when evaluated with theAntiplatelet Trialists Collaboration data,11 clopidogrelwas determined to be superior to the historical controlby several standard deviations. Confidence in the suf-ficiency argument was augmented by the absence ofmajor toxicities with clopidogrel, and the drug wasconclusively found to represent a good alternative forpatients who are allergic to aspirin.

Enoxaparin: Enoxaparin is a low-molecular-weightheparin that has been marketed for several years forthe treatment of deep venous thrombosis. Based on theconceptual belief that more proximal inhibition of thecoagulation system would not only inhibit thrombinactivity, but might also decrease thrombin generation,the Efficacy and Safety of Subcutaneous Enoxaparinin Non-Q-wave Coronary Events (ESSENCE)25 trialwas designed to evaluate enoxaparin compared withunfractionated heparin in patients with non-ST-seg-ment elevation acute coronary syndromes. The trialwas designed as a superiority study with 90% powerto detect an expected 25% reduction in the triplecomposite endpoint of death, nonfatal MI, and recur-rent ischemia at 14 days after treatment was started.Indeed, the result was a 16% reduction in this end-point, with no difference in death. The investigatorsobserved no difference in nonfatal MI, but an inde-pendent blinded events committee found a substantialdifference in favor of enoxaparin.

In addition to the positive trial results, 3 factorswere crucial in the deliberations of the CardiorenalAdvisory Committee culminating in their recommend-ing approval. First, low-molecular-weight heparin hasbeen used in hundreds of thousands of patients in thetreatment of venous thrombosis, and its safety is as-sumed to have been demonstrated. Second, trials withdalteparin, another low-molecular-weight heparin,have been published; although these results do notdemonstrate superiority of low-molecular-weight hep-

FIGURE 8. Line A depicts an odds ratio of 1.0 comparing the ex-perimental and the control treatment. Line B depicts the relative out-come with a putative placebo compared with the control agent inan active-control trial. The control agent has been compared withplacebo in several past trials with a consistent treatment effect. LineC provides an estimate of half of the benefit of the control agentcompared with the putative placebo. In Study A, the point estimateis multiple standard deviations from the behavior of the putativeplacebo. The confidence interval for the comparison of the experi-mental and control agents excludes the possibility that placebo isequivalent to the experimental agent. In Study B, the experimentaltreatment is not superior to the control treatment, but the 95% confi-dence interval for the active control comparison excludes the possi-bility that the experimental therapy conserves less than half of thebenefit of the active control versus the putative placebo. In Study C,although the point estimate indicates that the experimental therapyand the active-control therapy yield the same outcomes, the confi-dence intervals for the difference in outcomes fail to exclude thepossibility that the new treatment does not preserve half of the dif-ference between the active control and placebo. In Study D, al-though the experimental treatment is statistically inferior to the con-trol treatment, it preserves at least 50% of the benefit of the activecontrol versus placebo. Thus, if it has fewer side effects or is muchless expensive, it could still be of clinical value.

TABLE IV Establishing Sufficiency

• Active control must have been shown to be better than placebo• Multiple trials of active control showing homogeneous

treatment effect• Underlying mortality rates in disease must be stable• Baseline characteristics of population trial must be similar to

baseline characteristics of patients in overview• New treatment must preserve at least half the treatment benefit

of control agent versus putative placebo


arin, they do demonstrate that low-molecular-weightheparin is at least as good as unfractionated heparin,adding to the totality of data indicating that publicharm would not occur from approving enoxaparin.26,27

Finally, and crucially, enoxaparin was considered tobe superior to a putative placebo when the ESSENCEtrial was evaluated in conjunction with a systematicoverview of heparin. Interestingly, the evidence forheparin in the systematic overview did not even reachnominal statistical significance (p5 0.06),28 but de-spite this weakness, because heparin is a commonlyaccepted treatment of arterial thrombosis, the Cardio-renal Advisory Panel of the FDA voted for approval.

Eptifibatide and tirofiban: Eptifibatide and tirofibanare GP IIb/IIIa receptor inhibitors available for intra-venous administration. Although both compoundswere studied as add-ons to an aspirin regimen, theirclinical trial designs were substantially different (Ta-ble V). In the PURSUIT trial of eptifibatide, heparinwas also given as background therapy.15 With tirofi-ban, the Platelet Receptor Inhibition in Ischemic Syn-drome Management (PRISM) study evaluated theagent without heparin compared with heparin andaspirin, whereas the Platelet Receptor Inhibition inIschemic Stroke Syndrome Management in PatientsLimited by Unstable Signs and Symptoms (PRISMPLUS) Study evaluated 3 regimens: heparin and as-pirin; aspirin and tirofiban; heparin, aspirin, and tiro-fiban.29,30 In the PRISM PLUS study, the aspirin andtirofiban arm was discontinued early by the Data andSafety Monitoring Board because of a nominal excessof adverse outcomes. The PURSUIT study was alarge, relatively simple trial involving North andSouth America and Western and Eastern Europe, withlittle direction to the investigators about other therapyfor acute coronary syndromes; the trial was designedto evaluate eptifibatide as an add-on therapy to usualclinical practice. In contrast, PRISM PLUS was con-ducted only at sites doing angiography and had aprotocol-mandated angiogram 48 hours after random-ization. PRISM PLUS had intended to carry all 3 armsto completion, whereas PURSUIT had a planned in-terim analysis with a goal of dropping the low-dosearm if the bleeding rates were acceptable in the high-

dose arm.15,30 PURSUIT had a data analysis centerthat was independent of the sponsor, whereas PRISMand PRISM PLUS were conducted and analyzed bythe sponsor. Both studies had independent steeringcommittees and data and safety monitoring boards.PURSUIT evaluated all cases for potential events,including the screening of all cardiac enzymes; anindependent Clinical Events Committee reviewed.45% of cases and frequently overruled the investi-gators opinion about whether or not myocardial ne-crosis had occurred. PRISM and PRISM PLUS reliedon the investigator to identify potential events andthen had the Clinical Events Committee adjudicatethose possible events. The primary endpoint forPURSUIT was 30 days after randomization based on thebelief of the investigators that a benefit needed to persistat least that long to identify a worthwhile clinical benefit.In contrast, PRISM had a 48-hour primary endpoint withthe goal of proof of principle whereas PRISM PLUS hada 7-day primary endpoint with a secondary goal of dem-onstrating that the curves did not converge over the next30 days. Upon completion of the trials, the relativetreatment effects appeared quite divergent, but the dif-ferent design locations and analysis methods make truecomparison impossible.

A guidance document from the FDA CardiorenalDivision indicated that the 2-trial principle could bemet by designing 2 trials for indications addressing asimilar pathophysiology. Thus, the PURSUIT trial andthe Integrelin to Manage Platelet Aggregation to Com-bat Thrombosis (IMPACT)-II31 trial, which assessedefficacy and safety during percutaneous intervention,were interpreted as overlapping enough to allow ac-ceptance of the package of both trials for both indi-cations (non-ST-segment elevation acute coronarysyndromes and percutaneous intervention).15,31Eithertrial alone would not have been sufficient. Similarly,neither PRISM nor PRISM PLUS met the criteria forapproval of tirofiban without the package of both trialsin addition to the trends observed in the RandomizedEfficacy Study of Tirofiban for Outcomes and REst-enosis (RESTORE) trial,32 which evaluated tirofibanin percutaneous intervention.

Mibefradil: Although it is not an antithromboticagent, the recall of mibefradil shortly after it wasreleased makes it worthy of discussion. This drug is aT-channel calcium antagonist with promising animalstudies indicating its ability to decrease ischemia andto lower blood pressure. Mibefradil is metabolized inlarge part by the cytochrome P450 system. Pivotalstudies with mibefradil focused on carefully con-trolled scientific investigations in hypertension andangina. Patients with comorbidities or confoundingsituations were generally excluded. Short-term clini-cal trials nicely demonstrated that mibefradil lowersblood pressure and improves treadmill time in patientswith chronic angina. Pilot studies in heart failure pa-tients showed a nonsignificant excess of deaths inpatients treated with mibefradil.

At the time of the FDA review, a major heartfailure trial, Mortality Assessment in CongestiveHeart Failure (MACH)-1, had completed enrollment

TABLE V Comparison of PRISM PLUS and PURSUIT Trials’Designs

PRISM PLUS30

(Tirofiban)PURSUIT15

(Eptifibatide)

General design Scientific (narrowpopulation, prescribedtherapy)

Pragmatic (broadpopulation, mimicsclinical practice)

Restricted geography MulticontinentalIntention to study 3 treatment

armsIntention to drop a

treatment arm perprotocol

Use ofangiography

Mandatory As used in routinepractice

Endpoint Triple (death, MI, refractoryischemia)

Hard endpoint (death,MI)

Statistics Cox regression analysis x2 test

MI 5 myocardial infarction.


and was scheduled to be unblinded within the nextyear. The Cardiorenal Advisory Panel voted in a splitvote to approve marketing of mibefradil for the treat-ment of angina and hypertension. Shortly after thedrug was released, cases of sudden death were re-ported, and significant interactions with other com-pounds metabolized by the cytochrome P450 systemwere discovered. Based on these reports and the ab-sence of benefit demonstrated in the heart failuretrials, the sponsor withdrew the drug from the market.

SUMMARYThe discovery of multiple therapeutic agents that

are active at critical steps in the coagulation pathwayhas the potential to improve disease outcome; theimportance of designing and conducting appropriateclinical trials, however, is crucial to their safe andeffective use. Safety and efficacy have generally beendemonstrated by the results of adequate and well-controlled clinical trials. Importantly, trials can bedesigned to determine if a new therapy is better thanthe standard therapy, equivalent to the standard, or canbe added to the standard with incremental benefit. Thedesign chosen to evaluate a new therapy may have asignificant impact on timing of drug approval andapproved indications. Other issues integral to the de-sign of clinical trials evaluating antithrombotic drugsinclude the types of endpoints assessed and their tim-ing. Recent decisions regarding the evaluation andclinical use of new therapeutic agents, particularlycardiovascular agents, demonstrate the value of care-fully constructed clinical trials.

1. Jack DB. One hundred years of aspirin.Lancet1997;350:437–439.2. Hirsh J, Dalen JE, Deykin D, Poller L. Oral anticoagulants: mechanism ofaction, clinical effectiveness, and optimal therapeutic range.Chest 1992;102(suppl 4):312S–326S.3. Alexander JH, Harrington RA. Antiplatelet and antithrombin therapies in theacute coronary syndromes.Curr Opin Cardiol 1997;12:427–437.4. Po AL, Zhang WY. What lessons can be learnt from withdrawal of mibefradilfrom the market.Lancet1998;351:1829–1830.5. Connolly HM, Crary JL, McGoon MD, Hensrud DD, Edwards BS, EdwardsWD, Schaff HV. Valvular heart disease associated with fenfluramine-phenter-mine.N Engl J Med1997;337:581–588.6. Neuhaus KL, von Essen R, Tebbe U, Jessel A, Heinrichs H, Ma¨urer W, DoringW, Harmjanz D, Ko¨tter V, Kalhammer E, Simon H, Horacek T. Safety observa-tions from the pilot phase of the randomized r-Hirudin for Improvement ofThrombolysis (HIT-III) study. A study of the Arbeitsgemeinschaft LeitenderKardiologischer Krankenhausa¨rzte (ALKK). Circulation 1994;90:1638–1642.7. The Global Use of Strategies to Open Occluded Coronary Arteries (GUSTO)IIa Investigators. Randomized trial of intravenous heparin versus recombinanthirudin for acute coronary syndromes.Circulation 1994;90:1631–1637.8. Antman EM, for the TIMI 9A Investigators. Hirudin in acute myocardialinfarction: safety report from the Thrombolysis and Thrombin Inhibition inMyocardial Infarction (TIMI) 9A Trial.Circulation 1994;90:1624–1630.9. The EPIC Investigators. Use of a monoclonal antibody directed against theplatelet glycoprotein IIb/IIIa receptor in high-risk coronary angioplasty.N EnglJ Med1994;330:956–961.10. The EPILOG Investigators. Platelet glycoprotein IIb/IIIa receptor blockadeand low-dose heparin during percutaneous coronary revascularization.N EnglJ Med1997;336:1689–1696.11. Antiplatelet Trialists’ Collaboration. Collaborative overview of randomizedtrials of antiplatelet therapy–I: prevention of death, myocardial infarction, andstroke by prolonged antiplatelet therapy in various categories of patients.BrMed J1994;308:81–106.12. CAPRIE Steering Committee. A randomised, blinded trial of clopidogrelversus aspirin in patients at risk of ischaemic events (CAPRIE).Lancet1996;348:1329–1339.

13. Bittl JA, Strony J, Brinker JA, Ahmed WH, Meckel CR, Chaitman BR, Mara-ganore J, Deutsch E, Adelman B. Treatment with bivalirudin (Hirulog) as comparedwith heparin during coronary angioplasty for unstable or postinfarction angina.Hirulog Angioplasty Study Investigators.N Engl J Med1995;333:764–769.14. Mark DB. Medical economics and health policy issues for interventionalcardiology. In: Topol EJ, ed.Textbook of Interventional Cardiology. Philadel-phia: WB Saunders, 1994:1323–1353.15. The PURSUIT (Platelet Glycoprotein IIb/IIIa in Unstable Angina: ReceptorSuppression Using Integrilin Therapy) Trial Investigators. Inhibition of plateletglycoprotein IIb/IIIa with eptifibatide in patients with acute coronary syndromes.N Engl J Med1998;339:436–443.16. Berkowitz SD, Granger CB, Pieper KS, Lee KL, Gore JM, Simoons M,Armstrong PW, Topol EJ, Califf RM. Incidence and predictors of bleeding aftercontemporary thrombolytic therapy for myocardial infarction. The Global Utili-zation of Streptokinase and Tissue Plasminogen activator for occluded arteries(GUSTO) I Investigation.Circulation 1997;95:2508–2516.17. Gruppo Italiano per lo Studio della Sopravvivenza nell’Infarto Miocardico.GISSI-3: effects of lisinopril and transdermal glyceryl trinitrate singly andtogether on 6-week mortality and ventricular function after acute myocardialinfarction.Lancet1994;343:1115–1122.18. The GUSTO Investigators. An international randomized trial comparing fourthrombolytic strategies for acute myocardial infarction.N Engl J Med1993;329:673–682.19. ISIS-3 (Third International Study of Infarct Survival) Collaborative Group.ISIS-3: a randomised comparison of streptokinase vs tissue plasminogen activatorvs anistreplase and of aspirin plus heparin vs aspirin alone among 41,299 casesof suspected acute myocardial infarction.Lancet1992;339:753–770.20. Smalling RW, Bode C, Kalbfleisch J, Sen S, Limbourg P, Forycki F, HabibG, Feldman R, Hohnloser S, Seals A. More rapid, complete, and stable coronarythrombolysis with bolus administration of reteplase compared with alteplaseinfusion in acute myocardial infarction.Circulation 1995;91:2725–2732.21. Bode C, Smalling RW, Berg G, Burnett C, Lorch G, Kalbfleisch JM,Chernoff R, Christie LG, Feldman RL, Seals AA, Weaver WD. Randomizedcomparison of coronary thrombolysis achieved with double-bolus reteplase (re-combinant plasminogen activator) and front-loaded, accelerated alteplase (recom-binant tissue plasminogen activator) in patients with acute myocardial infarction.Circulation 1996;94:891–898.22. The International Joint Efficacy Comparison of Thrombolytics. Randomised,double-blind comparison of reteplase double-bolus administration with streptoki-nase in acute myocardial infarction (INJECT): trial to investigate equivalence.Lancet1995;346:329–336.23. Fibrinolytic Therapy Trialists (FTT) Collaborative Group. Indications forfibrinolytic therapy in suspected acute myocardial infarction: collaborative over-view of early mortality and major morbidity results from all randomised trials ofmore than 1000 patients.Lancet1994;343:311–322.24. The Global Use of Strategies to Open Occluded Coronary Arteries (GUSTO-III) Investigators. A comparison of reteplase with alteplase for acute myocardialinfarction.N Engl J Med1997;337:1118–1123.25. Cohen M, Demers C, Gurfinkel EP, Turpie AG, Fromell GJ, Goodman S,Langer A, Califf RM, Fox KA, Premmereur J, Bigonzi F. A comparison oflow-molecular-weight heparin with unfractionated heparin for unstable coronaryartery disease. Efficacy and Safety of Subcutaneous Enoxaparin in Non-Q-WaveCoronary Events Study Group.N Engl J Med1997;337:447–452.26. Klein W, Buchwald A, Hillis SE, Monrad S, Sanz G, Turpie AG, van derMeer J, Olaisson E, Undeland S, Ludwig K. Comparison of low-molecular-weight heparin with unfractionated heparin acutely and with placebo for 6 weeksin the management of unstable coronary artery disease. Fragmin in unstablecoronary artery disease study.Circulation 1997;96:61–68.27. The Fragmin During Instability in Coronary Artery Disease (FRISC) StudyGroup. Low-molecular-weight heparin during instability in coronary artery dis-ease.Lancet1996;347:561–568.28. Oler A, Whooley MA, Oler J, Grady D. Adding heparin to aspirin reduces theincidence of myocardial infarction and death in patients with unstable angina. Ameta-analysis.JAMA 1996;276:811–815.29. The Platelet Receptor Inhibition in Ischemic Syndrome Management(PRISM) Study Investigators. A comparison of aspirin plus tirofiban with aspirinplus heparin for unstable angina.N Engl J Med1998;338:1498–1505.30. The Platelet Receptor Inhibition in Ischemic Stroke Syndrome Managementin Patients Limited by Unstable Signs and Symptoms (PRISM-PLUS) Study1Investigators. Inhibition of the platelet glycoprotein IIb/IIIa receptor with tirofi-ban in unstable angina and non-Q-wave myocardial infarction.N Engl J Med1998;338:1488–1497.31. The IMPACT II Investigators. Randomised placebo-controlled trial of effectof eptifibatide on complications of percutaneous coronary intervention: IMPACTII (Integrilin to Minimise Platelet Aggregation and Coronary Thrombosis-II).Lancet1997;349:1422–1428.32. The RESTORE Investigators. Effects of platelet glycoprotein IIb/IIIa block-ade with tirofiban on adverse cardiac events in patients with unstable angina oracute myocardial infarction undergoing coronary angioplasty.Circulation 1997;96:1445–1453.


a perspective on the regulation of the evaluation of new antithrombotic drugs

Documents