1

DEPARTMENT OF HEALTH AND HUMAN SERVICES

FOOD AND DRUG ADMINISTRATION

CENTER FOR DRUG EVALUATION AND RESEARCH

JOINT MEETING OF

THE ARTHRITIS ADVISORY COMMITTEE AND

THE DRUG SAFETY AND RISK MANAGEMENT

ADVISORY COMMITTEE

VOLUME I

Wednesday, February 16, 2005

8:00 a.m.

Hilton Gaithersburg

620 Perry Parkway

Gaithersburg, Maryland

P A R T I C I P A N T S

Alastair J.J. Wood, M.D., Chair

Arthritis Advisory Committee:

Allan Gibofsky, M.D., J.D.

Joan M. Bathon, M.D.

Dennis W. Boulware, M.D.

John J. Cush, M.D.

Gary Stuart Hoffman, M.D.

Norman T. Ilowite, M.D.

Susan M. Manzi, M.D., M.P.H.

Drug Safety and Risk Management Advisory Committee:

Peter A. Gross, M.D.

Stephanie Y. Crawford, Ph.D., M.P.H.

Ruth S. Day, Ph.D.

Curt D. Furberg, M.D., Ph.D.

Jacqueline S. Gardner, Ph.D., M.P.H.

Eric S. Holmboe, M.D.

Arthur A. Levin, M.P.H., Consumer Representative

Louis A. Morris, Ph.D.

Richard Platt, M.D., M.Sc.

Robyn S. Shapiro, J.D.

Annette Stemhagen, Dr.PH. Industry Representative

FDA Consultants (Voting):

Steven Abramson, M.D.

Ralph B. D'Agostino, Ph.D.

Robert H. Dworkin, Ph.D.

Janet Elashoff, Ph.D.

John T. Farrar, M.D.

Leona M. Malone, L.C.S.W., Patient Representative

Thomas Fleming, Ph.D.

Charles H. Hennekens, M.D.

Steven Nissen, M.D.

Emil Paganini, M.D., FACP, FRCP

Steven L. Shafer, M.D.

Alastair J.J. Wood, M.D., Chair

P A R T I C I P A N T S (Continued)

National Institutes of Health Participants

(Voting):

Richard O. Cannon, III, M.D.

Michael J. Domanski, M.D.

Lawrence Friedman, M.D.

FDA Consultants (Non-Voting):

Byron Cryer, M.D. (Speaker and Discussant)

Milton Packer, M.D. (Speaker only)

Guest Speakers (Non-Voting):

Garret A. FitzGerald, M.D.

Ernest Hawk, M.D., M.P.H.

Bernard Levin, M.D.

Constantine Lyketsos, M.D., M.H.S.

FDA Participants:

Jonca Bull, M.D.

David Graham, M.D., M.P.H.

Brian Harvey, M.D.

Sharon Hertz, M.D.

John Jenkins, M.D., F.C.C.P.

Sandy Kweder, M.D.

Robert O'Neill, Ph.D.

Joel Schiffenbauer, M.D.

Paul Seligman, M.D.

Robert Temple, M.D.

Anne Trontell, M.D., M.P.H.

Lourdes Villalba, M.D.

James Witter, M.D., Ph.D.

Steven Galson, M.D.

Kimberly Littleton Topper, M.S., Executive

Secretary

C O N T E N T S

Call to Order:

Alastair J. Wood, M.D., Chair 6

Conflict of Interest Statement:

Kimberly Littleton Topper, M.S., 13

Welcome:

Steven Galson, M.D., MPH 16

Regulatory History

Jonca Bull, M.D. 24

Gastrointestinal Effects of NSAIDs and COX-2

Specific Inhibitors

Byron Cryer, M.D., 30

Mechanism Based Adverse Cardiovascular Events and

Specific Inhibitors of COX-2

Garret FitzGerald, 80

Committe Questions to Speakers 112

Sponsor Presentation: Vioxx (Rofecoxib),

Peter S. Kim, M.D. 130

Ned S. Braunstein, M.D. 131

FDA Presentation: Vioxx (Rofecoxib),

Lourdes Villalba, M.D., 227

Committee Questions to the Speakers 263

Sponsor Presentation: Celebrex (Celecoxib),

Joseph M. Feczko, M.D. 293

Cardiac Safety and Risk/Benefit Assessment of

Celecoxib

Kenneth M. Verburg, Ph.D. 295

FDA Presentation: COX-2 CV Safety: Celecoxib,

James Witter, M.D., Ph.D., 373

NIH and Investigator Presentation: Celecoxib in

Adenoma Prevention Trials: The APC Trial

(Prevention of Sporadic Colorectal Adenomas with

Celecoxib)

Ernest Hawk, M.D. 402

C O N T E N T S (Continued)

NIH Investigator Presentation: The PreSAP Trial

(Prevention of Colorectal Sporadic Adenomatous

Polyps)

Bernard Levin, M.D. 422

Committee Questions to Speakers 427

Sponsor Presentation: Cardiovascular Safety and

Risk/Benefit Assessment of Valdecoxib and

Parecoxib

Kenneth M. Verburg, Ph.D. 443

Concluding Comments

Joseph M. Feczko, M.D. 465

FDA Presentation: COX-2 CV Safety:

Valdecoxib-Parecoxib,

James Witter, M.D., Ph.D. 493

Bayer and Roche Joint Presentation on Naproxen,

Leonard M. Baum, R.Ph. 509

Safety Data

Martin H. Huber, M.D. 517

Committee Questions to Speakers 527

P R O C E E D I N G S

Call to Order

DR. WOOD: Let's get started. For those

of you who missed the memo, this is the committee

to discuss the safety and efficacy of COX-2

inhibitors. It is worth perhaps just giving some

thought to why we are here. We are here to

evaluate the relative efficacy and risk of these

drugs, and to decide whether the benefits from

these drugs outweigh the risk, in contrast to

whether the risks outweigh the benefits.

It is probably also worth just saying what

we are not here for. We are not here to delegate

blame or revisit the past. We are here to look

into the future and determine what we should do in

the future. It is important I think for everybody

to remember that as we move through the

discussions.

I guess the first thing to do is let

people at this enormous table introduce themselves.

Let's start down in this corner with John.

DR. JENKINS: Good morning. I am John

Jenkins. I am Director of the Office of New Drugs

in the Center for Drug Evaluation at FDA.

DR. O'NEILL: I am Bob O'Neill. I am the

Director of the Office of Biostatistics in CDER.

DR. BULL: Good morning. I am Jonca Bull,

the Director of the Office of Drug Evaluation V, in

the Office of New Drugs.

DR. GALSON: I am Steven Galson, the

Acting Director of CDER.

DR. TRONTELL: Anne Trontell, Deputy

Director of the Office of Drug Safety.

DR. SHAFER: Steve Shafer. I am not the

director of anything. I am a Professor of

Anesthesia at Stanford and Biopharmaceutical

Science at UCSF.

DR. HENNEKENS: Charlie Hennekens at the

University of Miami School of Medicine and Florida

Atlantic University.

DR. FRIEDMAN: Larry Friedman, from the

National Heart, Lung and Blood Institute.

DR. PAGANINI: Emil Paganini, a

nephrologist out of the Cleveland Clinic.

MS. SHAPIRO: Robyn Shapiro, I direct the

Center for of Bioethics of the Medical College of

Wisconsin. I am a Professor of Bioethics there and

I chair the Health Law Practice Group at Michael,

Best and Friedreich.

DR. CANNON: I am Richard Cannon. I am

Clinical Director of the Division of Intramural

Research, NHBLI, National Institutes of Health.

DR. MORRIS: Lou Morris, President, Lou

Morris and Associates.

DR. D'AGOSTINO: Ralph D'Agostino,

biostatistician from Boston University and the

Framingham Study.

DR. ILOWITE: Norm Ilowite, Schneider

Children's Hospital and Rheumatology at Albert

Einstein College of Medicine.

MR. LEVIN: Arthur Levin, Director of the

Center for Clinical Consumers and consumer

representative on the Drug Safety Committee.

MS. MALONE: I am Leona Malone. I am a

licensed clinical social worker and I am here as a

patient representative for the Arthritis Committee,

and I have struggled with rheumatoid arthritis and

osteoarthritis for 35 years.

DR. BATHON: Joan Bathon, Johns Hopkins

University, Department of Medicine, Division of

Rheumatology.

DR. CUSH: I am Jack Cush. I am a

rheumatologist from Presbyterian Hospital, Dallas.

DR. GIBOFSKY: Allan Gibofsky, Professor

of Medicine and Public Health, Cornell University;

Adjunct Professor of Law at Fordham University; and

I am Chair of the Arthritis Advisory Committee.

MS. TOPPER: Kimberly Topper, with the

FDA. I am the Executive Secretary for the

Committee.

DR. GROSS: I am Peter Gross. I am

Professor of Medicine and Community Health in New

Jersey Medical School; Chair of Medicine,

Hackensack University Medical Center; and I chair

the Drug Safety and Risk Management Advisory

Committee.

DR. HOLMBOE: I am Eric Holmboe, Vice

President for Evaluation Research at the American

Board of Internal Medicine.

DR. FARRAR: I am John Farrar. I am a

neurologist and epidemiologist at the Center for

Clinical Epidemiology and Biostatistics at the

University of Pennsylvania.

DR. MANZI: I am Susan Manzi. I am a

rheumatologist from the University of Pittsburgh

Medical Center, and with an appointment in

epidemiology at the Graduate School of Public

Health.

DR. HOFFMAN: I am Gary Hoffman. I am

Professor and Chairman of Rheumatic and Immunologic

Diseases at the Cleveland Clinic.

DR. DWORKIN: Hi. I am Bob Dworkin. I am

Professor of Anesthesiology and Neurology at the

University of Rochester School of Medicine.

DR. BOULWARE: I am Dennis Boulware,

Professor of Medicine, and rheumatologist at the

University of Alabama at Birmingham, and member of

the Arthritis Advisory Committee.

DR. DOMANSKI: I am Mike Domanski. I am a

cardiologist. I head the Clinical Trials Group at

the National Heart, Lung and Blood Institute.

DR. FLEMING: Thomas Fleming, Chair of

Biostatistics, University of Washington.

DR. FURBERG: Curt Furberg, Professor of

Public Health Sciences, Wake Forest University. I

am a member of the Drug Safety and Risk Management

Advisory Committee.

DR. DAY: Ruth Day, Duke University,

Director of the Medical Cognition Lab, and a member

of the Drug Safety Committee.

DR. PLATT: I am Richard Platt. I am

Professor and Chair of the Harvard Medical School,

Harvard Pilgrim Healthcare Department, Ambulatory

Care and Prevention. I am principal investigator

of one of the HHRQ centers for education and

research in therapeutics. I am a member of the

Drug Safety Committee.

DR. GARDNER: I am Jacqueline Gardner,

University of Washington School of Pharmacy and

Pharmaceutical Outcomes Research Program. I am on

the Drug Safety and Risk Management Committee.

DR. ELASHOFF: Janet Elashoff,

Biostatistics, Cedars-Sinai and UCLA.

DR. NISSEN: I am Steve Nissen. I am the

Medical Director of Cleveland Clinic Cardiovascular

Coordinating Center. I am a cardiologist, and I am

the Chair of the Cardiorenal Advisory Panel for the

FDA.

DR. ABRAMSON: Steve Abramson, I am

Chairman of Rheumatology at NYU and the Hospital

for Joint Diseases.

DR. CRYER: I am Byron Cryer. I am a

gastroenterologist from the University of Texas

Southwestern Medical School in Dallas, and the

Dallas VA Medical Center. My role here today is as

an FDA consultant to this group and as a member of

the Gastrointestinal Drugs Advisory Committee.

DR. STEMHAGEN: I am Annette Stemhagen. I

am an epidemiologist with Covance and I am the

industry representative to the Drug Safety and Risk

Management Committee.

DR. WOOD: I am Alastair Wood. I am the

Associate Dean at Vanderbilt and Professor of

Medicine and Professor of Pharmacology.

Now we will have the "reading of the

lesson" from Kimberly Topper.

Conflict of Interest Statement

MS. TOPPER: The following announcement

addresses the issue of conflict of interest with

respect to this meeting, and is made part of the

record to preclude even the appearance of such.

Based on the agenda, it has been determined that

the topics of today's meeting are issues of broad

applicability and there are no products being

approved. Unlike issues before a committee in

which a particular product is discussed, issues of

broader applicability include many industrial

sponsors and academic institutions.

All special government employees have been

screened for their financial interests as they may

apply to the general topics at hand. To determine

if any conflict of interests existed, the agency

has reviewed the agenda and all relevant financial

interests reported by the meeting participants.

The Food and Drug Administration has granted

general matters waivers to the special government

employees participating in the meeting who require

a waiver under Title 18 United States Code, Section

208. A copy of the waiver statements may be

obtained by submitting a written request to the

agency's Freedom of Information Office, Room 12A-30

of the Parklawn Building.

Because general topics impact so many

entities, it is not practical to recite all

potential conflicts of interest as they apply to

each member, consultant and guest speaker. FDA

acknowledges that there may be potential conflicts

of interest but, because of the general nature of

the discussions before the committee, these

potential conflicts are mitigated.

Further, during today's session Dr.

Bernard Levin will be presenting data on the

prevention of colorectal sporadic adenomatous

polyps trial, the PreSAP trial, a Pfizer-sponsored

clinical trial. We would like to note for the

record that Dr. Levin is attending this meeting as

a consultant to Pfizer.

With respect to FDA's invited industry

representative, we would also like to disclose that

Dr. Annette Stemhagen is participating in this

meeting as a non-voting industry representative,

acting on behalf of regulated industry. Dr.

Stemhagen's role on this committee is to represent

industry interests in general and not one

particular company. Dr. Stemhagen is the Vice

President of Strategic Development Services for

Covance Periapproval Services, Inc.

In the event that the discussions involve

any other products or firms not already on the

agenda for which FDA participants have a financial

interest, the participant's involvement and their

exclusion will be noted for the record.

With respect to all other participants, we

ask in the interest of fairness that they address

any current or previous financial involvement with

any firm whose product they may wish to comment

upon. Thank you.

DR. WOOD: For those of you still

standing, there are apparently seats in the

overflow room. Let's go right on to the first

speaker, who is Steve Galson. Steve?

Welcome

DR. GALSON: Thank you. I want to welcome

everyone and thanks in particular to our Chair, Dr.

Alastair Wood, committee members, special guests,

members of the public and FDA staff who have really

done a tremendous job in putting together a

particularly and unusually complex meeting.

We have some special guests today that I

want to point out. We have representatives from

the drug regulatory authorities of the member

countries of the European Union and six separate

countries--Canada, Japan, Singapore, Australia,

Switzerland and Mexico, and I really want to

welcome them. Thank you for being with us. We

also have several guests from congressional staff

offices and we are very pleased that they are with

us as well to learn about this important issue.

There is really an unprecedented level of

international attention to one of our advisory

committees today, and we are very proud that this

is taking place and we think it represents a new

level of collaboration and discussion around the

world about an emerging public health issue.

Many millions of people all over the world

are taking the products that we are discussing.

Indeed, they depend on them for a range of

conditions from the mild to the severe and

life-threatening. We must keep the interests and

health of these patients front and center in these

deliberations.

I wouldn't be complete in this

introduction if I didn't acknowledge the

controversy surrounding these products,

particularly over the last year. I want to

emphasize that we are anxious to hear all points of

views from the advisory committee and, of course,

from agency staff. It goes without saying that all

FDA staff are free to make any presentation without

fear of any retaliation. I don't want anyone

sitting around this table to be shy.

Also, we look forward to hearing a wide

range of views from the more than 50 members of the

public who are going to be making brief statements

later in the meeting. I want to remind the public

that all members of this committee have been

carefully screened for conflicts of interest and we

have used the same standards in this process that

we have used for other committees and similar

meetings.

A few comments about the challenging

risk/benefit balance that the agency must achieve

in making its regulatory decisions: Although you

have all heard strong opinions in the media and

medical literature about safety issues related to

the drugs we are discussing, our job and, indeed,

your job is to assess any safety concerns when

balanced by the benefit of these products. We

cannot lose sight of the reduced morbidity, pain

and suffering achieved by the products that are

under discussion and the real impact on people that

changes in the regulatory status may entail.

You will be assessing the risk/benefit

balance of these products this week in the midst of

a changing information environment and this

represents a particular challenge. We are aware of

at least a half dozen ongoing meta-analyses and

huge population-based studies, in addition to

several of the studies you will hear about this

week for which data analysis continues as we speak.

Although we have a full three days, the time really

isn't long enough to hear details about every

single ongoing, or incomplete, or unreviewed study

of which we are aware. Leaving them out of the

agenda has absolutely nothing to do with wanting to

keep information from you and everything to do with

allowing you to focus so that you have time to get

to our critical advisory questions.

We must be very cautious about

interpreting data for regulatory decision-making

that has not been thoroughly vetted and peer

reviewed, and even more cautious about interpreting

data of preliminary studies that are not even

complete. You will be hearing about some data in

these categories and I would remind you to exercise

caution in their interpretation.

As scientists, we have all seen examples

of ongoing studies whose findings have changed as

analysis is in the final stages, or examples where

inadvertent errors have led to misclassification in

epidemiologic studies, or when data that comes in

at the end of the data gathering stage influences

results. In today's 24-hour news environment, it

is difficult to not react to these incomplete

reports but we must go back to the basics of

relying on sound science and use the peer review

system to strengthen findings before utilizing them

to make regulatory decisions.

Lastly on the risk/benefit balance, as you

members know but it is sometimes difficult for us

to convey to the public, our job at FDA and your

job in the advisory group is to balance risks and

benefits on a population basis for the nation as a

whole. This is very different from the

risk/benefit assessment physicians do with

individual patients where specific risks of the

medications, family history, a patient's risk

tolerance and other factors must be taken into

consideration. A drug may, based on the weight of

evidence, have a positive benefit/risk balance for

the population leading to approval, yet, cause

grievous harm in a specific subset of individuals.

We say over and over again that all drugs have

risks, but when a person you know suffers an

adverse event the faulty assumption is sometimes

made that we must have made a mistake in the

approval.

I would also like to mention an unusual

feature of many of the data from the trials you

will be hearing over the next few days. The data

on safety of these drugs is, as I have mentioned,

unusually complex and represents the fact that

clinical trial methodology to look at

cardiovascular effects as adverse events has

changed dramatically. When discussions began about

cardiovascular safety of NSAIDs there was no

standard methodology by which cardiovascular

adverse events were confirmed or categorized.

Analyses vary by trial. Confirmatory processes

vary by trial. Only after the VIGOR trial did the

methods of establishing confirmatory processes and

standardization become better established. Of

course, in population-based cohorts and case

control studies case reporting and confirmation is

both rudimentary and completely inconsistent

between studies.

In addition, as you know already, unlike

drugs designed to treat cardiovascular disease,

these trials have not been designed to do a full

cardiovascular assessment. So, major pieces of

information that you might like to have are simply

not available. So, in many ways we are forced to

compare apples to oranges in these trials and

studies, and when you are not doing that you are

trying to draw conclusions based on insufficient

information, making your task even harder.

In spite of all the ambiguity, work in

progress, changing standards and questions, we ask

you for the miraculous job of crystal clarity in

your responses to our questions. We know this is

tough on such challenging scientific and

controversial issues, and we are enormously

grateful to you because we know that you all are up

to this challenge. The agency will act rapidly

within the next few weeks to act on the

recommendations you communicate to us over the next

few days.

I would like to quickly go to the agenda.

Today through midday tomorrow you will hear from

sponsor companies, FDA staff and NIH researchers

about data on both approved and unapproved COX-2

selective and non-selective products. Tomorrow

afternoon we have 54 members of the public

registered to speak. On Friday you will hear about

important methodological issues in interpretation

of these studies, and then we will move on to the

questions.

Again, thank you and on behalf of the FDA

I wish you the very best of luck on this important

endeavor. Thanks, Dr. Wood.

DR. WOOD: Thanks a lot. Two additional

people have joined the cast of thousands that we

have at the table, and perhaps it would be worth

having them introduce themselves. Bob, you go

first.

DR. TEMPLE: I am Bob Temple. I am

Director of the Office of Medical Policy.

DR. WOOD: Stephanie?

DR. CRAWFORD: Thank you, Mr. Chair.

Stephanie Crawford--good morning--University of

Illinois at Chicago, College of Pharmacy; member of

the Drug Safety and Risk Management Advisory

Committee.

DR. SELIGMAN: Good morning. This is Paul

Seligman. I am the Director of the Office of

Pharmacoepidemiology and Statistical Science.

DR. WOOD: Is there anyone else I didn't

notice arrive? No? Then, let's move on to the

next speaker. Jonca?

Regulatory History

DR. BULL: Good morning. Again, I would

like to extend a warm welcome to the members of the

committee and to extend and acknowledge a

particular thanks to our staff at FDA, specifically

Dr. Villalba, Dr. Witter, Dr. Schiffenbauer from

our team, our statistical staff, and colleagues in

the Office of Drug Safety who have put in countless

hours in preparation for this meeting.

The NSAID class is one that probably

everybody in this room has a product in their

medicine cabinet that is a member. It is a large

class of marketed products for both OTC and

prescription indication use. It is a wide range of

products with varying risk/benefit profiles. Their

approved indications are for short-term use such as

dysmenorrhea and acute pain; chronic use for

osteoarthritis, rheumatoid arthritis, familial

adenomatous polyposis in the example of Celebrex.

So, clearly, we have drugs that for everyone, from

the young female with cramps to the senior citizen

with arthritic pain, have importance and clearly

there is a need for them in the marketplace. There

are other proposed uses that are known to be under

investigation, and you will hear about studies in

the setting of Alzheimer's disease, as well as

sporadic polyp prevention.

I would like to briefly review some of the

regulatory history for these products, going back

to December of 1986 when there was a public

advisory committee meeting that discussed the GI

paragraph and databases were discussed at that

time.

This was followed in 1995 where revisions

for the NSAID class label were discussed, as well

as a subsequent advisory committee in 1998 when the

new science of the COX-2s were discussed and their

potential enhanced safety for GI benefit.

In December of 1998 an advisory committee

was held to discuss the data for Celebrex, followed

in December of 1998 when that drug was approved

first in this new class of products. In April of

1999 an advisory committee was held for Vioxx,

followed by its approval in May of 1999. We held

another advisory committee meeting in 2001 which

discussed the large outcome studies which sponsors

had undertaken to further evaluate how clinically

meaningful the data from endoscopic studies was in

order to further evaluate the enhanced GI safety

claim.

This time line has several points I would

like to bring to your attention. The first IND for

these products came in 1994 so we are dealing with

a relatively short time line, given that this is

year 2005, in drug development, marketing and an

evolving picture for safety.

The products below the time line are the

ones that have been approved, and I would like to

bring your attention to those above the line,

Arcoxia, Prexige, the IV formulation of Bextra

which have not been approved in the United States

due to insufficient safety data.

The COX-2 agents--are they different? In

what way? When we look at risk to benefit, how do

these agents differ from the traditional NSAIDs?

Can a clinically meaningful benefit for GI safety

and less risk, that is for CV risk, renal risk,

hepatic risk, allergy--can that be characterized?

What additional study is needed to better

understand the science of COX-2 inhibition?

When we think in terms of labeling risk

management, what risk management options are

appropriate in this settings, ranging from

potential withdrawal of the product to labeling

changes?

Certainly there are lessons learned for

drug development. I cite a quote at the end of an

article by Dr. Temple and Marty Himmel, in JAMA in

May, 2002, and I think the statement is quite a

relevant one to our deliberation, that no

improvements in drug development can completely

eliminate the risk of unexpected events.

Looking at large NDA databases is helpful

but continued monitoring is essential to assess

evolving risk profiles for new products.

Certainly, the impact of aggressive marketing must

be taken into account for these unknowns of drug

safety.

Dr. Galson has already gone through the

schedule for the meeting. I will just briefly

allude to our framework for this deliberation.

Following me, Dr. Byron Cryer will be discussing

the gastrointestinal effects of the NSAIDs and

COX-2 specific inhibitors; followed by Dr. Garret

FitzGerald on mechanisms for cardiovascular risk

from inhibition of COX-2s. This will be followed

by a presentation by Merck and the FDA presentation

by Dr. Lourdes Villalba.

This afternoon you will hear from Pfizer

and their review of cardiovascular safety and

risk/benefit assessment of celecoxib, followed by

the FDA presentation by Dr. James Witter. There

will be a presentation then on the NIH-sponsored

colon polyp prevention trials, with subsequent

presentations by Pfizer on valdecoxib and

parecoxib, and an FDA presentation on valdecoxib.

This will be followed by Bayer and Roche discussing

naproxen.

Tomorrow you will hear about the

epidemiologic studies, followed in the afternoon by

the open public hearing and committee discussion.

Day three in the morning will focus on the

Alzheimer's prevention trials. The ADAPT trial

will be discussed that morning by Dr. Constantine

Lyketsos; followed by a presentation by Dr. Milton

Packer on interpretation of cardiovascular events;

a presentation by Dr. Robert Temple on clinical

trial design and patient safety, future directions

for COX-2 selective agents; and a presentation by

Dr. Robert O'Neill on issues in projecting

increased risk of cardiovascular events to the

exposed population. Dr. Sharon Hertz will then

present a summary of the meeting presentations

prior to the afternoon discussion of our questions.

Again, our thanks to the committee members

for taking time from their extraordinarily busy

schedules for this important meeting as we reach

another milestone in the regulatory history of

these products.

DR. WOOD: Thanks very much. Let's just

go straight on to the next speaker, who is Dr.

Byron Cryer who is going to talk on the GI effects.

Dr. Cryer?

Gastrointestinal Effects of NSAIDs

and COX-2 Specific Inhibitors

DR. CRYER: Thank you. For the purposes

of full disclosure, I would first like it to be

noted that I have been invited to give this

presentation by the Analgesic and Anti-Inflammatory

Division of the FDA. I do have relationships with

sponsors of products being mentioned in today's

presentation, however, I am not being paid for my

participation in this meeting nor for my

presentation today.

For those of you not familiar with me, I

am a gastroenterologist and I am thrilled that the

FDA has been begun this meeting with the focus on

this subject because many of us have forgotten that

the initial reason for the development of the class

of the COX-2 specific inhibitors was entirely

because of the gastrointestinal effects of the

non-steroidal anti-inflammatory drugs and, for that

reason, I think it is very appropriate that we have

this review of the gastrointestinal effects of

NSAIDs and what the data say from the GI

perspective about the gastrointestinal effects of

COX-2 specific inhibitors.

From the perspective of the NSAIDs risk,

listed here are several of the known risks

associated with the non-steroidal anti-inflammatory

drugs, the gastrointestinal risks, the cardiorenal

risks and the anti-platelet concerns. Among these,

as the group knows, the adverse concerns of

greatest risk historically were the

gastrointestinal effects that present with features

such as ulcers, perforations, bleeding, obstruction

strictures and many other interesting

manifestations. Over the last several years, added

to this list and a focus of this meeting are

cardiovascular concerns of the non-steroidal

anti-inflammatory drugs but my perspective are the

issues listed at the top, the gastrointestinal

effects.

When looking more extensively at what the

specific gastrointestinal effects of NSAIDs are, we

have learned that NSAIDs have effects throughout

the GI tract. The upper gastrointestinal effects

are the most pronounced but there are some very

interesting effects that we see throughout the GI

tract, such as in the small intestine and colon.

In recent years we have had an increasing focus on

lower gastrointestinal effects of NSAIDs, a very

interesting phenomenon. Several have been assessed

by endoscopic means but there has been a lot of

discussion as to what are the clinically relevant

untoward major events that might happen in the

lower gastrointestinal tract. While this is

debated with respect to the prevalence of lower GI

effects, these effects are likely somewhere in the

range of 10-20 percent of total gastrointestinal

effects that happen within the GI tract

attributable to NSAIDs. Clearly, the major effects

of NSAIDs in the GI tract are in the upper

gastrointestinal tract, such as ulcers more

commonly in the stomach and the duodenum, and

concerns such as gastrointestinal bleeding,

perforations and obstructions. So, that is really

the focus upon which the strategies were developed

to increase NSAID safety within the

gastrointestinal tract.

With respect to the epidemiology of ulcer

disease in general, some very interesting phenomena

have been observed which have persisted into recent

years. But the overall summary of the phenomenon

that I would like to focus your attention to is

that while in recent years the overall incidence of

uncomplicated ulcers, both gastric and duodenal,

has been markedly declining in the U.S. and

worldwide, very interestingly, the incidence of

complications, specifically gastrointestinal

bleeding, has not declined in similar proportions

and, in fact, has persisted or increased. This

phenomenon, in particular the bleeding, has been

felt to be a manifestation of the effects of the

non-steroidal anti-inflammatory drugs within the GI

tract.

This problem presents itself clearly with

respect to morbidity and, unfortunately, mortality

and several hundreds of thousands of

hospitalizations. The costs have been debated.

The actual quantified amount of mortality in the

U.S. is also a number that is debated. The 16,500

estimate is probably an overestimate. But the

bottom line is that NSAIDs are clearly associated

with morbidity, mortality and costs in this country

as well as worldwide, and this is has been the

issue that has led to the discussions of the need

for increasing gastrointestinal safety for NSAIDs.

So, the various ways in which these

assessments have been done has ranged from studies

which we have seen over the years that have been

short-term evaluations of physiologic or

pharmacologic effects on healthy volunteers to the

more relevant studies of the gastrointestinal

effects of these drugs in arthritis patients.

These studies have ranged from long-term endoscopy

studies to a fewer number but very important

studies that have assessed clinical events such as

symptomatic ulcers, GI bleeding, perforation and

obstruction.

Over the years there has been extensive

discussion as to the relevance of the endoscopy

studies and how the endoscopic observations with

NSAIDs might relate to the outcome studies. One of

the criticisms of the endoscopic studies is that

the endoscopic lesions are numerous. They are

mostly only known from endoscopies that are done as

a part of a scheduled study and they are

asymptomatic. However, what we have learned from

comparing the numerous endoscopic studies to

observations that have been seen in the outcome

studies is that the relative proportions in terms

of outcomes seen in endoscopic studies tend to be

predictive of what one would expect to see in an

outcome study. So, we have come full circle then

in our understanding of the role of endoscopic

studies and, at least in the gastroenterology

community, we now feel that there is some

substantial value in endoscopic studies and that

they are predictive of what one might expect to see

in outcome trials.

Now, with respect to what we see in these

types of trials, when one looks endoscopically

there is a range of findings in people who are

taking high doses of NSAIDs. In greater than 90

percent, if one were to look, we would see this

phenomenon of NSAID gastropathy, which is this

constellation of erosions and hemorrhages but it is

mostly asymptomatic, mostly not clinically

relevant.

With respect to incidences of asymptomatic

endoscopic ulcers, gastric ulcers happen two to

three times more commonly than the duodenal ulcers,

with the ranges that are shown on the slide.

Again, these lesions are mostly asymptomatic and

don't progress in the majority of individuals to

clinically untoward gastrointestinal events.

What these things look like--this is an

endoscopic photograph of gastropathy demonstrating

the constellation of hemorrhages and erosions that,

again, are going to be mostly asymptomatic, ranging

to a picture, shown here, of an endoscopic ulcer

seen in the antrum of the stomach of an NSAID user.

The more clinically concerning endpoint,

that being clinically significant ulcers, occurs

with the non-selective NSAIDs on average about 2

percent, with a range of about 1-4 percent. This

range and this mean are important numbers as

benchmarks to remember because they will become

relevant as we discuss some of the outcome studies

that have been conducted with the COX-2 specific

inhibitors.

Having reviewed what the risks are, I

would now like to move the discussion to what our

strategies have been to reduce the risk of the

gastrointestinal complications with NSAIDs. It is

a simple strategy and most experts will recommend

identifying the patient population who might be at

risk and this is based upon identification of risk

factors. Then, once having identified susceptible

populations for risk, one employs strategies that

would reduce risk, such as either the use of

gastroprotective drugs or the use of safer NSAIDs,

and the category of safer NSAIDs clearly involves

the subclass of the COX-2 specific inhibitors.

With regard to identification of risk

factors, a risk factor not commonly mentioned is

the NSAIDs themselves. NSAIDs clearly provide risk

for gastrointestinal effects. Shown here are

various NSAIDs available by class and by

prescription in the United States. As you can see,

they have been divided into traditional NSAIDs,

non-salicylates; aspirin related, salicylate-based

compounds; and then COX-2 inhibitors which are

currently available, in development or previously

available in the U.S.

With regard to identifying patient

characteristics which may suggest risk, these have

been extensively studied and they are listed here,

things such as increasing age and the threshold age

is widely debated but one category that has been

suggested would be those greater than 65, let's

say. Clearly history of GI ulceration; having had

a complication; concomitant drugs such as

corticosteroids or anticoagulants; cardiovascular

disease, interestingly, such as CHF; and this issue

of multiple NSAIDs all increase the risk.

Of this list that the group is very

familiar with, the one that has probably not been

as widely appreciated and one which has been

highlighted from some of the outcome trials of the

COX-2 specific inhibitors is this issue of multiple

NSAIDs, and it is a risk factor that presents

itself in the context of a patient profile, a

patient who takes prescribed NSAIDs along with

either low doses of aspirin of over-the-counter

NSAIDs. Since we know that the risk for

NSAID-related gastrointestinal events is related to

dose, what one accomplishes in this group of

multiple NSAIDs is essentially to increase the

overall dose of NSAIDs delivered.

With regard to the strategies after having

identified the susceptible population, the first

category essentially is that of co-therapeutic

gastroprotection. As alluded to a minute ago, it

would be desirable to use the lowest effective dose

of an NSAID. Then really the two prevailing

gastroprotective or co-therapy strategies that we

have are the use of either misoprostol or proton

pump inhibitors.

Several studies have been done in either

of these categories. I will just highlight for

purposes of discussion two outcome trials that I

think nicely demonstrate the effectiveness of these

strategies. With regard to misoprostol, the most

widely quoted study was the outcome trial, the

MUCOSA trial in which misoprostol was given to

patients who were chronically taking NSAIDs over 6

months and were demonstrated to be associated with

a 40 percent or less reduction in gastrointestinal

complications.

From the perspective of the PPI outcome

trials, there have been fewer evaluations but there

have been, in fact, some evaluations for clinically

relevant outcomes for PPIs, this being one example

of a trial which was actually not intended in its

design to evaluate outcomes of a proton pump

inhibitor in patients taking NSAIDs but,

nevertheless, provided us with some insight into

the potential effects from the perspective of

gastrointestinal outcomes.

This was a trial that was designed with

the question in mind of whether or not H. pylori

eradication prior to starting an NSAID would be an

effective therapy or not for the reduction

potentially of NSAID-related bleeds. So, in this

group of H. pylori infected NSAID users, half of

them were treated for their H. pylori infections

prior to being started on an NSAID and acted as a

control. The other half were given a proton pump

inhibitor. In this specific instance omeprazole.

What was observed, very interestingly, at

the end of 6 months is that in this instance there

was a 76 percent reduction in the subsequent

incidence of upper gastrointestinal bleeding in the

group that had received the proton pump inhibitor

approach.

From the perspective of the safer NSAIDs,

this is a story that is also well known. Its focus

today is really to look at specifically the COX-2

specific inhibitors shown on the far right. The

concept has been widely discussed and is arguably

somewhat simplistic, but for the sake of today's

discussion, as the group knows, it is highlighted

by the observation that there are 2 COX isoforms

available, COX-2 and COX-1, and that COX-1 is the

isoform which is primarily responsible for the

protective prostaglandins in the stomach which

typically protect against injury. Once inhibited

by non-selective NSAIDs, the prostaglandin products

produced by COX-1 lead to an increased

susceptibility for injury. The concept at least

for COX-2 specific NSAIDs in that they have limited

inhibitory effects on COX-1 is that they would

likely not inhibit prostaglandins, likely not be

associated with ulcers, and likely be associated

with a reduction in clinically significant

gastrointestinal untoward events with NSAIDs.

Having said that, there have been a few

gastrointestinal outcome trials that have been

designed to evaluation whether or not the COX-2

inhibitors would meet this objective or not. Shown

here are two of the outcome trials with rofecoxib

and celecoxib.

As the group knows, there has also

recently been another completed outcome trial with

lumiracoxib. In general, the outcome trials have

compared COX-2 specific inhibitors at higher than

usual therapeutic doses for osteoarthritis to

non-selective NSAIDs and evaluated the clinically

significant events on average over a year. The

major difference of importance between the outcome

trials with celecoxib and rofecoxib was the

inclusion or exclusion of low doses of aspirin. We

know that low doses of aspirin are ulcerogenic. In

the CLASS trial 21 percent of patients took low

doses of aspirin, 325 mg/day or less, and none of

the patients in the rofecoxib experience were

taking low doses of aspirin.

The principal gastrointestinal

observations from the CLASS trial are, as shown

here in this figure, taken from the publication in

the JAMA, which represents the 6-month data point

from this year-long trial. In the top panel are

all the patients who were evaluated in the trial

who were taking either celecoxib or one of the

non-selective NSAIDs, ibuprofen or diclofenac. As

you note, there was a numeric but not statistically

significant reduction in ulcer complications in the

overall group, remembering that 21 percent of the

patients in the CLASS trial were taking low doses

of aspirin and that some of the ulcer effects were

related to the effects of aspirin.

So, to get a better concept of the effects

of a COX inhibitor compared to non-selective

NSAIDs, the middle panel looks exclusively at the

patients in this 6-month evaluation of the CLASS

trial who were not taking aspirin, just celecoxib,

ibuprofen or diclofenac. As you observe in this

middle panel, there were statistically significant

reductions associated for GI outcomes with

celecoxib when compared to traditional NSAIDs in

the absence of aspirin at 6 months.

However, for those of you who were here

four years ago this month at the long-term safety

evaluations of the FDA, the entire CLASS trial data

set was evaluated with respect to gastrointestinal

complications. When compared to either ibuprofen

or diclofenac alone or combined, with respect to

complications there were not statistically

significant gastrointestinal reductions in events

associated, as you can see, with celecoxib.

With regard to the VIGOR trial, just to

refresh the group's memory, this was clearly

exclusively an evaluation of rofecoxib versus

naproxen. There was no low dose aspirin. Their

observations were straightforward in with respect

to either primary or secondary event being

confirmed upper GI events or complicated events.

There was a statistically significant reduction

associated with rofecoxib compared to naproxen.

As I have mentioned, there has also been a

similar in design outcome study with lumiracoxib.

The variable observations between these outcomes

trials have led to extensive debate in the medical

and scientific communities as to why one might have

observed differences with respect to

gastrointestinal endpoints between the outcome

trials of COX-2 specific inhibitors.

While I don't have time to get into the

nuances and specifics of that debate, one point

that I would like to bring to the group's attention

that I do think is worthwhile reviewing is that, to

the extent that there were differences between the

observations in the outcome trials, these

differences may have had more to do with

differences in ulcerogenic effects with the

traditional NSAID comparators such as naproxen,

ibuprofen and diclofenac than they may have had to

do with differences with respect to ulcerogenic

effects between rofecoxib and celecoxib.

The point to be highlighted is that the

non-selective NSAIDs differ with regard to their

ulcerogenic effects and that the delta, the

difference observed between a COX-2 inhibitor and a

non-selective NSAID will matter, and it will be

based upon the choice of comparator being used. I

am not here to speak about cardiovascular effects.

Dr. Garret FitzGerald will talk about

cardiovascular issues in the talk to follow. But I

would like to point out that this concept of

differences in COX-1 effects of non-selective

NSAIDs is also applicable when we turn to a

discussion of considerations of potential

differences in cardiovascular observations between

the trials of COX-2 inhibitors.

Having pointed out the data with the COX-2

specific inhibitors, I would like to mention that

there are other potential approaches, and I would

like to turn the discussion to a consideration, as

shown on the bottom, of potentially older, safer

NSAIDs that may be associated with gastrointestinal

safety, agents such as the non-acetylated

salicylates, nabumetone, diclofenac and etodolac.

I mention this because--these are not

gastrointestinal events, this is a reflection of in

vitro evaluations of COX-1 versus COX-2 selectivity

of various NSAIDs. On the left, in the green, are

NSAIDs which have increasing in vitro COX-1

selectivity and are going in the negative

direction; on the right, is increasing COX-2

selectivity. When one evaluates COX-2 selectivity

in vitro, there is a group of NSAIDs which fall

within this mid-range category of what I would call

moderately COX-2 selective, and this COX-2

selectivity of agents such as meloxicam or etodolac

may be predictive of what one might see in outcome

trials.

Taking etodolac as an example, when it was

evaluated with respect to gastrointestinal outcomes

compared to a non-selective NSAID such as naproxen,

shown in the upper panel, there was a statistically

significant, greater than 50 percent, reduction in

gastrointestinal outcomes associated with an agent

such as etodolac. So, this leads me to conclude,

over here in this group of category for COX-2

specific inhibitors, that there are agents which

have COX-2 selective activity which had not been

widely appreciated historically.

Since aspirin was such and important

phenomenon in outcome trials, I think it is

relevant to review the gastrointestinal effects of

low doses of aspirin. This has been looked at

mostly from an epidemiologic perspective, and

trials such as this have tended to show a

dose-response relationship. Although not

statistically significant in this case, clearly

lower doses, at least numerically, of aspirin such

as 75 mg were associated with a lower rate of

clinically relevant gastrointestinal bleeding than

higher doses such as 300 mg. In this instance, at

least numerically from 75 to 300 mg, the odds ratio

of clinically relevant upper gastrointestinal bleed

doubled.

Because of the risk associated with very

low doses of aspirin such as 75 mg, doses of

aspirin that have been quite low, such as 10 mg,

have been evaluated in human studies to assess the

question of whether or not there would be any daily

orally administered dose of aspirin which would be

without gastrointestinal effects.

When measured by use of an intermediate

marker that would be of COX inhibition or

measurement of gastrointestinal prostaglandins,

daily doses of aspirin given out to 3 months, as

low as 10 mg, were associated with as great of a

reduction of gastrointestinal COX as seen with 320,

and gastric ulcerations were observed with a dose

of aspirin that was as low as 10 mg, suggesting

that there is likely not a dose of aspirin that

would be effective that would be daily administered

that would be without gastrointestinal risk.

Another commonly asked question would be

the potential benefit of an enteric coating or

buffered preparation of aspirin. When assessed in

this cohort from the Framingham trial of patients

who were taking various formulations of low dose

aspirin, as one sees that there was no appreciable

reduction in gastrointestinal bleeding associated

with either enteric coating of aspirin or buffered

aspirin when compared to plain, non-enteric,

non-buffered aspirin preparations.

Coming back to the risk factor which I

mentioned had been not widely appreciated, the risk

factor of multiple NSAID use, that is, combining

low dose aspirin with a non-selective NSAID or

COX-2 specific inhibitor, I think it is valuable to

appreciate for a moment the actual risk, numerical

risk, contributed by the addition of aspirin to

another prescribed NSAID.

From this population study in Denmark, it

was apparent that when one combines the use of low

dose aspirin and a non-selective NSAID the risk of

having a clinically significant bleed, upper

gastrointestinal bleed, more than doubled, such

that several people would feel that the risk of a

6-fold increase in the combination of a

non-selective NSAID plus aspirin is sufficiently

high that this population of users would need to be

further risk reduced.

These are data with non-selective NSAIDs.

The data with respect to COX-2 specific inhibitors

have come primarily from a few sources. In this

previous figure in which we saw earlier the 6-month

data from the CLASS trial we stopped with the

middle panel and had events in individuals taking

celecoxib or non-selective NSAIDs in the absence of

aspirin.

But when one looks at the bottom panel,

rates of events, complications or symptomatic

ulcers and ulcer complications in individuals who

were taking one of these agents in the face of low

doses of aspirin, it is clear that the use of low

dose aspirin in the face of a COX-2 specific

inhibitor markedly increased the rates of

gastrointestinal events.

But a point that I would like you to focus

your attention on is the actual incidence of events

in the patients who were taking either aspirin in

combination with a COX inhibitor or non-selective

NSAID. You will remember that the problem that led

to really the focus and development of classes of

safer NSAIDs is an incidence of ulcer complications

of 1-4 percent in the population that takes

non-selective NSAIDs. When one looks at the

incidence of events that occurs annualized in

patients who take aspirin, at least derived from

the data in the CLASS trial, it is clear that the

incidence that was observed of 2-6 percent is

higher than the original problem.

So, I would like to summarize with respect

to the effects of low dose aspirin that low dose

aspirin clearly increases the risk and mitigates

the potential gastrointestinal beneficial effects

of a COX-2 specific inhibitor. These observations

have been seen in other experiences with regard to

the total lack of outcome data which I previously

showed you, where we stopped on the top panel.

When looking at the observations in patients taking

low doses of aspirin, the beneficial effects of

total lack disappear.

In endoscopic trials recently we have also

seen this effect of aspirin in this trial over 12

weeks in which either aspirin was given alone or in

combination with rofecoxib and compared to

ibuprofen. Focusing on the rofecoxib plus aspirin

comparison, rofecoxib plus aspirin users have a

similar, equivalent incidence of endoscopic

ulcerations to non-selective NSAIDs such as

ibuprofen. So, the short conceptual way of

summarizing this is a COX-2 specific inhibitor plus

aspirin equals the effects of a non-selective

traditional NSAID.

The gastrointestinal discussion that we

have had so far has pointed out some of the

potential gastrointestinal effect benefits of a

safer class of agents such as a COX-2 specific

inhibitor. Clearly, the gastrointestinal benefit

does not exist in the face of aspirin and what we

have recently learned is that the gastrointestinal

benefit derived from a class of safer agents in the

GI tract might be mitigated by adverse events in

other areas, and other areas for consideration for

this week's meeting are potential cardiovascular

effects.

Given the limitations of COX-2 specific

inhibitors and low dose aspirin users or when there

may be potential cardiovascular concerns, one

question that we have been asked to address would

be in a potential world of no COX-2 specific

inhibitors would we return to the problem of

several gastrointestinal bleeds, hospitalizations

or mortality?

Well, this brings us back to the question

of what might be the other approaches to accomplish

the objective of reductions in GI events. We have

discussed some of the older, safer NSAIDs. There

are NSAIDs in development such as nitric oxide

NSAIDs or phosphatidylcholine NSAIDs, the effects

of which we are unsure of now and they are

currently being evaluated. But the other

prevailing strategy to accomplish this objective

would be the consideration of a non-selective NSAID

plus co-therapy with either a proton pump inhibitor

or misoprostol.

Data in support of the proton inhibitor

approach have been looked at in several trials, one

example of which is shown here, endoscopic

ulceration in NSAID users receiving co-therapy with

either placebo, a proton pump inhibitor or

misoprostol. What the data pretty consistently say

is that proton pump inhibitors have similar ability

to misoprostol to prevent recurrent ulceration in

NSAID users.

Given that there are two prevailing

approaches to accomplishing GI safety, either COX-2

specific inhibitor alone or a non-selective NSAID

plus a PPI, an important question which has

presented itself for evaluation has been how might

these two approaches compare directly and this is

an important question to consider when considering

the alternatives to having a world potentially in

which there might not be COX-2 specific inhibitors

available. Could GI safety be accomplished?

Well, this question has been asked at

least in two trials or similar design in which high

risk NSAID users--high risk being defined as people

who previously had a history of bleeding ulcers.

Once the ulcers were healed, they were then placed

on either of the combination of non-selective NSAID

plus a proton pump inhibitor or a COX-2 specific

inhibitor, and then were followed for 6 months for

rates of recurrent gastrointestinal bleeding. The

results of one of these trials has been fully

published in a peer reviewed journals, shown here.

The two endpoints being looked at--on the

right are outcomes such as upper gastrointestinal

bleeding; on the left are the results of endoscopic

ulceration. Either of these endpoints tells us

that the approach of a non-selective NSAID plus a

PPI appears comparable to the COX-2 specific

inhibitor approach for achieving the objective of

reductions in GI safety. However, two important

points that I would like to point out to the group

are, one, we have endoscopy on the left and

outcomes, GI bleeding, on the right. Again, the

endoscopic ulcerations that are seen in the trials

generally predict what one would see in an outcomes

study but, more importantly, if one looks at the

actual rates of events which occurred, on the

right, 5 percent and 6 percent with either approach

in a group of individuals at high risk, meaning

they previously had a history of gastrointestinal

bleed, it is clear that either approach, either

NSAID plus PPI or COX-2 specific inhibitor, is

sufficiently adequate to reduce the rates of events

back to a comfortable range. The rates of events

seen here in a high risk population are similar to

the initial problem for which these approaches were

developed.

In conclusion I have several observations.

The untoward gastrointestinal effects of NSAIDs, as

we know, cause considerable morbidity, mortality

and cost. Secondly, COX-2 specific inhibitors were

developed principally to achieve a reduction in

NSAID gastrointestinal toxicity. That was a very

desirable objective to be reached. But very

interestingly, as we just reviewed, this objective

has been partially reached. It seems that the risk

reduction may not be achieved to the extent that we

would have liked in patients who are at high risk

for gastrointestinal bleeding, and the reason this

is important is that that is clinically the target

group of interest for risk reduction.

Paradoxically, I did not mention that if

one looks at subgroup analyses of outcome studies

it appears that people who are at lower baseline

gastrointestinal risk do have a benefit from

receiving a COX-2 specific inhibitor. However, the

low risk group has a low prevalence of this problem

of NSAID-related gastrointestinal events in the

population.

So COX-2 inhibitors, it appears, have been

widely used by patients who are not at high risk

for GI effects, and we have reviewed over the last

several minutes that there are some limitations

with COX inhibitors. In my opinion, there is no

great clinical need for COX-2 specific inhibitors

in patients who are at baseline at low GI risk. It

is also clear that there is no GI benefit in

patients who are concurrently taking aspirin. We

are here to discuss the possibility that

cardiovascular concerns may exist for some groups

of patients.

So, the strategies to reduce the

gastrointestinal effects of NSAIDs should focus on

patients at greatest risk. Just to reiterate, the

patients at greatest risk may not be sufficiently

risk reduced by either of the prevailing strategies

which we currently have available clinically. For

such patients, COX-2 specific inhibitors may be an

attractive option but it looks like the target

group of interest may not have the anticipated

benefit.

For patients who are taking low dose

aspirin or, if cardiovascular concerns were to

exist, we have been asked to consider that if there

were a world without COX-2 specific inhibitors how

might we accomplish this objective, and it is clear

that there are other strategies available that may

lead to a reduction in NSAID GI effects. Thank you

very much.

DR. WOOD: Thank you very much. Byron,

could you just stay there in case there are

specific questions for you while the slides are up?

I have one. Could you put up slide 4 again? That

shows data through 1990.

DR. CRYER: Yes.

DR. WOOD: What surprised me is Jim Freis

has updated that data through 2000, and that

dramatically changes what that slide looks like.

In fact, he found a 67 percent decline since 1990

in complicated ulcers, the vast majority of which

occurred actually before COX-2 specific inhibitors

went on the market. So, I am interested, first of

all, in why you chose to present 15-year old data

when there is new data out there that contradicts

that, and whether you would like to comment on his

publications from which this data came as well.

DR. CRYER: Sure. It is correct that

there are newer data available that have

demonstrated a reduction in gastrointestinal bleeds

on a population basis. On the other hand, it is

also very true that this problem of

gastrointestinal bleeding with NSAIDs continues to

be a significant problem despite its more recent

decline. But, more importantly, he also

highlighted a very important observation which is

that the declines in gastrointestinal bleeding that

have been seen in populations preceded the

introduction of COX-2 specific inhibitors, and

there are some data sets to suggest, at least in

the U.S., that hospitalizations for

gastrointestinal bleeding since the introduction of

COX-2 specific inhibitors have not markedly

declined compared to hospitalizations prior to

their introduction.

DR. WOOD: Right. So, most of the 67

percent decline occurred before these drugs went to

the market, and that 67 percent occurs from the

points on your slide here.

DR. CRYER: Point well taken.

DR. WOOD: And one other point of

clarification I guess, the data you showed from

CLASS, was that data from the predefined endpoint

of the study at 18 months or the 6-month analysis

that was published?

DR. CRYER: Just for sake of review, I

have pointed out both time-dependent endpoints.

The endpoint that was published and shown here, in

the JAMA, was the predefined 6-month data and the

endpoints that are shown here represent an

evaluation of the entire data set. There are

clearly differences in the conclusions about the

effects of celecoxib which varied by time and

varied by whether one evaluates the data at 6

months or evaluates the entire data set.

DR. WOOD: Just remind us, at 18 months

what did the data set show?

DR. CRYER: At 13 months the data, with

respect to complications, indicate that there was

no statistically significant reduction in upper

gastrointestinal complications associated with

celecoxib, at a dose of 400 twice daily, when

compared to either diclofenac or ibuprofen

individually or when compared to both of them

together. I will point out for the sake of fair

balance that this data does include the 21 percent

of individuals who were taking low doses of

aspirin.

DR. WOOD: Other questions from the

committee? Dr. Nissen?

DR. NISSEN: Yes, this 1-4 percent rate, I

am interested in understanding the time-dependent

hazard. If a patient is put on a non-selective

NSAID and, let's say, for the first year has no GI

events, is the risk in the second and third and

fourth years the same as it is in the first year?

In other words, once you know that a patient is

tolerating an NSAID are they no longer at high

risk?

DR. CRYER: There are a few answers,

sub-answers to that question. It is a complicated

discussion. What is clear that risk persists, that

even in the individual who did not develop a

complication in year one, that individual continues

to have risk in subsequent years--two, three, four,

etc. There are data sets that suggest that the

period of highest susceptibility, highest risk is

within the first three months of administration.

Having said that, there are other data sets to the

contrary. This incidence of gastrointestinal

events that are time-dependent in individuals has

been difficult to assess primarily based upon a

concept of selection of susceptible individuals.

People drop out because of other reasons such as

dyspepsia. So, it is difficult to get a firm

estimate on that. But it is clear, in summary,

that the risk after one year or after any period of

time is always persistent as long as the NSAID

exposure is present.

DR. NISSEN: Two more quick questions. I

didn't see any analysis of COX-2 plus low dose

aspirin versus a non-selective NSAID plus low dose

aspirin. The reason I am asking that is that, as a

cardiologist, in my patients who are taking

conventional NSAIDs, if they need aspirin for

cardiovascular prophylaxis I give them aspirin.

So, the question is are there any studies looking

at NSAID plus aspirin versus COX-2 specific

inhibitor plus aspirin?

DR. CRYER: Well, the CLASS trial

addressed that question in a subpopulation of

individuals which was under-powered statistically

to give a definitive answer to that question. That

is an ongoing debate within the medical

communities. I will say, however, that while the

debate continues what is clear is that with either

approach COX-2 specific inhibitor plus aspirin or

non-selective inhibitor plus aspirin the ensuing

rates of gastrointestinal events are too high for

us to feel comfortable that we have risk-reduced

those patients sufficiently.

DR. NISSEN: And a final question,

symptoms of dyspepsia are obviously one of the

issues as well, and I want to make sure I

understand what fraction of the population, let's

say an osteoarthritis population, simply cannot

tolerate NSAIDs because of GI discomfort. Do we

have data on that?

DR. CRYER: Sure. A couple of comments

about dyspepsia which I didn't mention, NSAID

dyspepsia is common. Its prevalence varies

depending on how dyspepsia has been defined in

trials, and because there have been variable

definitions of dyspepsia, its reported rates have

varied anywhere from 10-30 percent of NSAID users,

but it is clearly more common than complications.

In the patient who has dyspepsia, the

presence of dyspepsia is not predictive of the

patient who might have risk. In most of these

studies dyspepsia, in my way of thinking, is

considered more of a nuisance issue that can be

controlled symptomatically with acid reduction

rather than something that presents significant

gastrointestinal concern.

DR. WOOD: Dr. Gibofsky?

DR. GIBOFSKY: You commented extensively

on the upper GI risk but in your second slide you

correctly pointed out that there are problems with

traditional medications affecting the structures of

the GI tract below the ligament of triads. Could

you comment somewhat on the data comparing the

effect of COX-2 specific inhibitors versus

traditional non-steroidals with or without proton

pump inhibitor protection on the lower GI tract?

DR. CRYER: There have been fewer data

sets which have assessed the lower gastrointestinal

events with NSAIDs. A few comments on the types of

studies that have been done, there have been

studies using pill endoscopy which have indicated

that lesions, endoscopic ulcers and erosions occur

in the lower gastrointestinal tract contributed to

by non-selective NSAIDs, an effect which can be

reduced by a COX-2 specific inhibitor, an effect

which is not reduced by the co-therapy approach of

adding a PPI to a non-selective NSAID. I am

speaking of the lower gastrointestinal effects.

Having said that, again similar to the

endoscopic ulcer story, these endoscopically

detected lesions in the lower gastrointestinal

tract probably have very limited clinical

relevance. When lower gastrointestinal clinically

significant events have been assessed from the

prospective trials, the one noted most commonly in

the literature is an assessment of the VIGOR trial

looking at the effects of rofecoxib compared to

naproxen, in which case a 40-50 percent reduction

was seen in lower gastrointestinal events with

rofecoxib compared to naproxen, again to reiterate,

a reduction which would not be expected to be

observed with the proton pump inhibitor approach.

Having said that, in that assessment of

the rofecoxib experience there was an inclusion in

the definition of lower GI events of individuals

who had had reductions in hemoglobin and hematocrit

and who did not otherwise have clinically apparent

gastrointestinal bleeding.

Probably the best assessment in terms of

the risk of lower gastrointestinal events on NSAIDs

comes from population-based observational studies.

While there is variance in that estimate, it looks

like the lower gastrointestinal events probably

contribute 10-20 percent of clinically relevant

events when compared to all GI events that might

happen on NSAIDs.

DR. GIBOFSKY: One last quick point, would

your recognize that there might well be a

population of patients whom you would stratify as

low GI risk who, nevertheless because of either

intolerance, as the last speaker asked, or lack of

efficacy to traditional non-steroidals, would be

candidates for another class of agents?

DR. CRYER: Sure. Their NSAID dyspepsia

is a common phenomenon. I will say that when

dyspepsia has been carefully evaluated in the

prospective trials of COX-2 specific inhibitors in

general there tends to be a reduction in the rates

of dyspepsia associated with the COX-2 specific

inhibitors. However, when one evaluates the

absolute reduction in rates of dyspepsia in the

trials it generally tends to be a few percentage

points. Finally, some of the other strategies that

were mentioned to accomplish risk reduction, for

reduction in GI events in patients on NSAIDs, also

accomplished reductions in dyspepsia in patients

who might experience NSAID-related dyspepsia.

DR. WOOD: Dr. Cush?

DR. CUSH: Byron, two time questions.

One, is there a time point at which peptic

ulcerations and bleeds plateau over time in NSAID

users or COX-2 users? Second, what is the longest

data set that we have as far as the use of a COX-2

agent in a clinical trial where observation is

carried out? Do we have two-year data; five-year

data?

DR. CRYER: Right. There does appear to

be some plateau-ing of the effect. The data sets

do suggest that after long-term exposure the rates

of events with longer-term exposure are not as

great as rates of events with initial exposure to

NSAIDs but, again, that may be attributable to the

phenomenon of dropping out of susceptibles. The

second portion of your question, Jack, was?

DR. CUSH: What is the longest data set we

have on COX-2 agents?

DR. CRYER: Well, when one looks at the

trials, the prospectively defined outcome

trials--we have CLASS, TARGET, VIGOR--there are

periods of observation out to 13 months. Having

said that, we certainly have longer periods of

observations of COX-2 specific inhibitors for

trials in which the specific outcome of interest

was defined for an endpoint that was other than

upper GI bleeding, so specific polyp reduction,

Alzheimer's disease, other trials that we certainly

will hear about over the course of the next few

days, many of which have gone out to periods as

much as 3 years.

DR. WOOD: Is there anyone else who has a

question that specifically addresses something on a

slide that the speaker could show again? If not,

we will come back to these questions and ask you,

Byron, if you would, to be available this

afternoon.

DR. CRYER: Yes.

DR. WOOD: Are there any questions that

somebody has specifically? Tom?

DR. FLEMING: Yes, could we go back to the

slide that showed the CLASS trial with the time to

complicated ulcer?

DR. CRYER: There were two. You can tell

me which one you are referring to, this or the

next?

DR. FLEMING: Both, this and the next.

Basically, here what you are showing us is that in

the presence of aspirin there doesn't seem to be a

reduction in the complicated ulcers although in

those that are not taking aspirin there is this

reduction of about two-thirds. If you go to the

next slide, that is at 6 months. Hence, we see at

6 months this reduction in the rate in the

celecoxib group that is driven by those patients

who are not on aspirin. But that effect, as you

noted, has disappeared out at a year.

I know that is making a lot of a single

data set but is this suggestive of the possibility

that, in response to Steve Nissen's question, there

could be a group that is more susceptible and what

you are doing, in the presence of aspirin, is

achieving not effect; in the absence of aspirin you

are achieving a delayed effect but, in essence, you

are going to have the same overall incidence by a

year even with the COX-2 specific inhibitor?

DR. CRYER: Sure, your point is that there

are likely subgroups of susceptibility for GI risk

on NSAIDs or on COX-2 specific inhibitors. But I

would say also that underlying that argument, which

I think is accurate, is the observation which

confounds the whole discussion, which I have

mentioned previously, which is that early on in any

of these trials you are going to remove the most

susceptible of the individuals and those who

actually persist in the trial tend to be the least

susceptible subpopulation.

DR. FLEMING: Indeed, but that is the

essence of what I am saying, and this would be

consistent then with the theory that if there is a

particular susceptible group, that group is going

to have a higher risk and it is, in fact, going to

have complicated ulcers. They just occur somewhat

sooner with the non-specific NSAIDs. The COX-2s

are not preventing that, they are just delaying the

time to the occurrence.

DR. CRYER: I think we are in agreement

there.

DR. WOOD: Richard?

DR. PLATT: To extend that, on slide 13

you list some risk factors for NSAID-associated GI

toxicity. Can you tell us how well those

discriminate low risk individuals from high risk

individuals? And, if they do, what fraction of the

population falls into low risk, medium risk, high

risk? And, quantitatively what are those risks?

DR. CRYER: That is a complicated question

but it is an important one. When people like

myself have shown these risks we commonly lead to

the assumption that these risk are numerically

equivalent, which they are not. There are certain

risk factors which clearly place one individual at

higher risk than others. The highest risk most

consistently seen in trials would be that of having

had a previous history of a gastrointestinal

bleeding ulcer. But not far behind that would be

the risk of taking an anticoagulant, such as

Coumadin, in association with a non-selective

NSAID. Age as a risk factor is a variable one.

Although we suggest in our discussions of this that

there may be a threshold of age below which one may

be not at risk and above which at risk for having

it. In fact, it is a continuum. In fact, the risk

contributed by age is about a 2 percent increase in

risk per decade of life, such that people who are

in their 80s are at very high risk, much higher

risk than people who are in their 40s.

With respect to your question of

quantifying the risk in a population, that is a

difficult issue because all of these risk factors

do not individually present themselves in any one

patient. The more risk factors one has--two risk

factors present greater risk than one; three

greater than two. I would say, having said that

and trying to give you a reasonable estimate, in my

opinion the percentage of NSAID users who would

likely be candidates for this is probably somewhere

on the order of 20-25 percent, depending on how one

assesses that. If one looks at an OA or RA

population and concludes that age in and of itself

is a risk factor, then you are close to 80 or 90

percent of the population that might be at risk

based upon that risk factor of age. So, it really

depends on which risk factor, and it really depends

on the quantitative contribution of the risk factor

being described. But, certainly, I would say the

one that most clearly and consistently has

presented itself as highest risk in the various

trials has been the risk factor of having had a

previous bleeding ulcer, and it is the one that I

would like to underscore which does not appear to

be sufficiently risk-reduced by either of the

strategies which we have available.

DR. WOOD: Any other questions that are so

burning that they have to be asked now and not in

the discussion? Ralph? Burning? And let's try

and make the answers as brief as we can.

DR. D'AGOSTINO: What are the consequences

of complicated ulcers in, say, the CLASS trial

where you do see this differential and this

catching up? Do they follow to see the

consequences of these ulcers? Were they different

over the time period?

DR. CRYER: I am sorry, I don't

understand.

DR. D'AGOSTINO: What are the

consequences? What happened to these subjects

after? Were they reversible, the ulcer? Does it

lead to mortality?

DR. CRYER: Right, what I assume is

driving your question is whether there are

differences in mortality--

DR. D'AGOSTINO: Well, morbidity,

mortality, what happens.

DR. CRYER: Well, clearly, morbid effects

are hospitalization and the complications of them

having a massive gastrointestinal bleed, which can

be several. The ultimate complication or

consequence of these morbid effects is mortality

and in these outcome trials there were no

differences in the level of mortality. With regard

to the various other consequences, most of them are

clearly going to be reversible after having

suffered a significant hospitalization.

DR. WOOD: Any other smoking questions?

Peter?

DR. GROSS: A question on the third to

last slide, on recurrent ulcer bleeding in high

risk patients, the so-called non-selective NSAIDs

selected diclofenac to compare with celecoxib.

DR. CRYER: Yes.

DR. GROSS: Diclofenac is roughly

comparable in COX-2 selectivity. Is that the right

drug to test with PPI to show that the PPI plus a

non-selective NSAID is comparable to a COX-2

inhibitor like celecoxib? Should they have picked

a non-selective NSAID that was less selective for

COX-2?

DR. CRYER: Sure. Your point is very well

taken and it is one which I tried to underscore

throughout the talk, which is that there are

clearly differences in the COX-1, i.e.,

ulcerogenic, effects of non-selective NSAIDs.

Diclofenac clearly is an agent which is associated

with a lower rate of gastrointestinal ulceration

and complications than non-selective NSAIDs. So,

in this evaluation of the comparison of diclofenac

plus omeprazole compared to celecoxib there is a

valid discussion that the results may have been

biased in favor of the diclofenac plus omeprazole

approach.

The reason I showed that is that that was

a fully published paper. There are, however, other

trials not yet fully peer reviewed, which have been

presented in the gastrointestinal community,

looking at other NSAIDs, such as naproxen plus a

proton pump inhibitor compared to the COX-2

specific inhibitor approach, and the results of

those observations again are comparable endpoints

between the two strategies.

DR. WOOD: I am going to move us on now

and we will come back after the next talk. Dr.

Cryer, we would like you to come back up if there

are questions at that time as well. The next

speaker is Dr. Garret FitzGerald. Garret?

Mechanism Based Adverse Cardiovascular Events

and Specific Inhibitors of COX-2

DR. FITZGERALD: Thank you, Dr. Wood. You

are, please, going to have to forgive me, I feel

quite nauseated; I have a touch of the flu and I

took a medicine to reduce my temperature, but I am

not prepared to tell you what it is!

(Laughter)

I would like to thank Dr. Wood and the FDA

and the committee for the opportunity to visit

Gaithersburg at this time of the year.

(Laughter)

When I boarded the Metro last night at

Union Station and began the apparently interminable

trip to the sylvan embrace of Shady Grove I thought

to myself it might be useful to try and summarize

for you a message that will derive from my talk.

The message is that, just as low dose aspirin

affords cardioprotection and a small but absolute

risk of serious GI bleeds, as you heard from Byron

just now, through inhibition of COX-1, so specific

inhibitors of cyclooxygenase-2 afford

gastroprotection and a small but absolute risk of

cardiovascular events. So, I have titled my talk

mechanism-based adverse cardiovascular events and

specific inhibitors of COX-2.

Well, as every lawyer and broker and

journalist knows, this is the cyclooxygenase

catalyzed pathway of arachidonic acid metabolism.

Arachidonic acid is mobilized for release from cell

membranes by activation of phospholipases and it is

subject to metabolism by two enzymes which we call

prostaglandin JH synthases 1 and 2 but which are

known more commonly as cyclooxygenases 1 and 2.

They give rise to a series of lipid products called

prostaglandins which activate receptors and have

very diverse biological effects.

One of the reasons we are here is that

this, although depicted in a very simplistic way,

is actually a quite complex system. To illustrate

that, I will just mention two of these lipid

products, prostaglandin E-2 and prostacyclin or

prostaglandin I-2. When formed by

cyclooxygenase-1, these two lipid products afford

gastroprotection, and our thinking is that the

common GI adverse events of typical non-steroidal

anti-inflammatory drugs reflect the inhibition of

COX-1-derived PGI-2 and PGE-2, thereby, exposing

people to gastroduodenal liability.

But it turns out that when the very same

lipids, prostacyclin and prostaglandin E-2, are

formed by cyclooxygenase-2 as opposed to

cyclooxygenase-1 they mediate pain and

inflammation. Indeed, it is the suppression of the

formation of these two prostaglandins by COX-2

inhibitors that retains the anti-inflammatory and

analgesic efficacy of traditional non-steroidal

anti-inflammatory drugs which inhibit the two

enzymes together.

But it turns out that these two

prostaglandins, prostaglandin I-2 and prostaglandin

E-2, formed by cyclooxygenase-2 also afford

cardioprotection which can manifest itself in

various ways, and suppression of that capability is

the cogent mechanism which explains the

cardiovascular hazard which has emerged.

Well, I am sure this audience well knows

that cyclooxygenase-2 inhibitors do not inhibit

platelet aggregation, a way that we look at

platelet activation in people that have been

administered drugs. This just illustrates the

absence of an effect at several doses of celecoxib

in healthy volunteers compared to the inhibition of

this signal by a mixed inhibitor at the time of

peak drug action. Of course, that reflects the

absence of cyclooxygenase-2. There should be a big

shade here on this Western Blot if it was present

but, unlike cyclooxygenase-1, which is there in

abundance, cyclooxygenase-2 is not present in

mature human platelets.

The wrinkle in all of this is that if you

look at two structurally distinct members of the

class of COX-2 inhibitors, the depression of the

formation of that protective lipid, prostacyclin,

as reflected by urinary excretion of its major

metabolite which, believe it or not, is the gold

standard of how you look at prostaglandin formation

in people--this depression is comparable on

specific inhibitors of COX-2 with the depression we

see with structurally distinct mixed inhibitors

like ibuprofen and indomethacin.

So, one might logically deduce from this

that even under physiological conditions, never

mind under conditions of pathology, a COX-2 might

be induced by cytokines for example. It is a

dominant source of prostacyclin. We hypothesized

at the time that that reflected a mechanism which

had been described in vitro by Topper and Jim

Broney and which is illustrated here, which is when

you subject endothelial cells to laminar shear

force, which mimics the effect of the blood stream

on the lining of blood vessels, you up-regulate the

COX-2.

Well, that raised a question rather than

answered a question even though it anteceded the

approval of the first of these drugs. The first

proof of principle that prostacyclin did actually

modulate cardiovascular function in vivo stems from

this study where we used mice lacking the

prostacyclin receptor, known as the IP, or the

thromboxane receptor, known as the TP, or both

together. Thromboxane is the lipid which is formed

by COX-1 in platelets and has harmful effects on

the heart and cardiovascular system, and

suppression of thromboxane reflects the

cardioprotection of low dose aspirin.

In these studies we looked at the response

to vascular injury in mice and we found that there

was a signal of increased proliferation in response

to vascular injury in the mice lacking the

prostacyclin receptor which accorded with its in

vitro properties.

Furthermore, when you injure the lining of

a blood vessels in a mouse, just as if you do it in

humans by performing an angioplasty, you get an

attendant increase in platelet activation which is

reflected by a time-dependent increase in excretion

of a major thromboxane metabolite. We were

interested to see that this signal was grossly

augmented in the absence of the prostacyclin

receptor, and that all of these reflections of the

phenotype could be rescued by co-incidental

deletion of the thromboxane receptor along with the

prostacyclin receptor.

Now, these studies were criticized as to

their relevance to the COX-2 inhibitor story mainly

because people said, well, you have taken away the

prostacyclin receptor but when we give the drugs,

although we suppress prostacyclin, we do it to a

substantial but incomplete degree, maybe 60-80

percent on average.

So, we performed these studies in another

model of induced thrombogenesis in mice where we

injured the vasculature in a free radical catalyzed

fashion. In these studies we looked at the effect

of a biochemically selective regimen of a COX-2

inhibitor, and we found that the response time to

the thrombogenic stimulus was significantly

accelerated. Furthermore, as opposed to looking at

the absence of both copies of the prostacyclin

receptor, we looked at the effect of deletion of

just one copy and we found a significant and

intermediate phenotype.

More recently we have devised a technique

which permits us to remove cyclooxygenase-2 from

particular cells. What I am showing here is the

removal of only one copy of cyclooxygenase-2 from

endothelial cells. As you can see, that also

accelerates the response to a thrombogenic

stimulus. So, these new studies are proof of

concept of precisely the mechanism that we

originally proposed.

Well, I think this is a point that we will

come back to. We have some scientific evidence

that there is a very non-linear relationship

between inhibition of the capacity of platelets to

make COX-1 derived thromboxane and inhibition of

thromboxane-dependent function, that is,

aggregation.

To get into the red zone for inhibition of

platelet function you certainly have to be in

excess of 95 percent inhibition of capacity, more

like up in the 98 percent range. Where we have

actually almost no experimental evidence is whether

there is a discordance between that and the

relationship between inhibition of prostacyclin and

inhibition of its protective cardiovascular

function. Perhaps the intermediate phenotype of

the prostacyclin receptor deleted mice losing one

copy of the gene may suggest that that is so.

So, we are back in the mouse model of

induced thrombosis. The reason I am showing you

this slide is that a theme that will recur and is

relevant to the clinical consideration is whether

inhibition of COX-1, along with inhibition of

COX-2, modulates the implications of inhibiting

COX-2.

So, in these studies we have looked at the

rescue from thrombosis induced by intravenously

administering arachidonic acid to mice at two

different doses in mice that either lack completely

COX-1 or in mice that lack 98 percent of the

capacity to make COX-1 derived thromboxane by

platelets. As you can see, these two genetically

modified mice behaved very similarly in terms of

the rescue from arachidonic acid induced thrombosis

or, indeed, the time to complete occlusion induced

by the thrombogenic stimulus I showed you in the

earlier slide. This accords with that

non-linearity of the relationship for COX-1 that I

showed you. You would expect that to be suppressed

in the 98 percent inhibited mice.

Now, that is all very well because it is

in mice. So, you would way, well, how would we

address this in terms of seeking a proof of concept

in people? Well, if you delete the prostacyclin

receptor mice don't fall over dead with thrombosis.

They are more responsive to thrombogenic stimuli.

So, if you wish to seek proof of concept in people,

you would move to a population that had hemostatic

activation and you would postulate that in such a

population you would detect a signal faster and in

a smaller study than might otherwise be the case.

Indeed, given the widespread recognition

that patients undergoing coronary-artery bypass

grafting exhibit hemostatic activation, and some

suggestion also that they may be a model of aspirin

resistance, it is perhaps unsurprising that we are

able to detect a clear signal of cardiovascular

hazard in two placebo-controlled trials in this

condition.

Now, when I think of people at risk of

thrombosis when one is considering where one goes

with these drugs, I tend to think of middle-aged or

elderly people who have suffered a myocardial

infarction or stroke. But I think it is important

to remember that risk of thrombosis can manifest

itself in susceptibility to this cardiovascular

hazard of these drugs in other populations.

This is a ventilation perfusion scan of a

23 year-old athlete who had been on the pill for 3

years, who went on a 6-hour car journey, having

been put on valdecoxib for the antecedent 8 days

and, at the end of the trip, developed left-sided

chest pain; was misdiagnosed and continued on

valdecoxib for another 10 days; had right-sided

pleuritic chest pain that led to this VQ scan.

This is purely an anecdote but it brings

to mind that individuals who have environmental

predisposition to thrombosis, with a relatively

small absolute risk such as being on the pill or

prolonged stasis or genetic predispositions like

Factor V Leiden, might be susceptible to a

geometric interaction of relatively low risk from

this class of drugs.

So, as far as thrombosis is concerned,

where does this take us? Well, first of all, we

have evidence that at least in vitro COX-2 can be

induced in endothelial cells and produce

prostacyclin. We have evidence that it constrains

platelet activation and thrombogenesis in vivo.

Suppression of prostacyclin does not cause

spontaneous thrombosis but augments the response to

thrombogenic stimuli in vivo. So, the hazard from

coxibs would be expected to be particularly evident

in those otherwise predisposed to thrombosis, and

we have evidence that this hazard is modulated by

inhibition of COX-1 in the appropriate zone.

Well, there has been a lot of talk, as we

all know, about mechanisms and one of the things I

have found really curious is the notion that

hypertension is a distinct mechanism. People get

hypertension on traditional non-steroidal

anti-inflammatory drugs as well as COX-2 inhibitors

for a reason. The reason is the same mechanism.

Illustrated here from studies in mice by Matt

Breyer and his colleagues is how inhibition of

COX-2, shown in red, will augment the pressor

response to an infused pressor like angiotensin-II.

Again, as in the setting of thrombosis, COX-1 is

not neutral. As you can see, if he uses a

selective inhibitor of COX-1 he attenuates the

response to angiotensin-II.

Now, these studies have been complemented

by congruent data with gene-deleted mice. They

raise the prospect that the incidence of

hypertension would reflect not only the degree of

inhibition of COX-2 but the selectivity with which

it is attained. Indeed, in this week's Archives we

have the first epidemiological evidence consistent

with that concept.

Now, the products of COX-2 that buffer the

response to pressor agents include prostacyclin and

PGE-2. Here we are looking at the effect on blood

pressure, of deletion of the prostacyclin receptor

and, as you can see, blood pressure is elevated and

the response to salt loading is increased. One

sees exactly the same phenotype deleting one of the

receptors for PGE-2.

So, as far as blood pressure is concerned,

suppression of COX-2 derived PGI-2 and PGE-2

increases blood pressure and augments the response

to hypertensive stimuli in mice. Deletion or

inhibition of COX-1 depresses the response to

vasoconstrictors in vivo so again we see COX-1

modulating the hazard from COX-2 inhibition.

Hypertension on NSAIDs would be expected to relate

to the inhibition of COX-2 and the selectivity with

which it is attained.

Let's think of a more chronically

unfolding cardiovascular hazard. These data

arbitration taken from Narumiya. They are looking

at the development of atherosclerosis in a

genetically prone mouse, and you can see that

deletion of the prostacyclin receptor accelerates

atherogenesis in male ApoeE-deficient mice. In

fact, the impact was most particularly marked at

initiation and early development of

atherosclerosis.

By contrast, deletion of the thromboxane

receptor does the complete reverse, and other

studies conducted by us and others have shown that

inhibition of COX-1 selectively or antagonism of

the thromboxane receptor will have the same effect

as deleting the thromboxane receptor, as shown

here.

So, as far as atherosclerosis is

concerned, we see this buffering capacity between

COX-1 and COX-2. Furthermore, we have shown

recently that in a different genetically proned

mouse model deletion of the prostacyclin receptor

and inhibition of COX-2 dependent formation of

prostacyclin is important in affording the

atheroprotection conferred by estrogen in female

mice.

So, here we see the atheroprotection in

terms of reduction of lesion development with

estrogen treatment in vasectomized mice being

dramatically reduced by deletion of the

prostacyclin receptor, which raises a whole new set

of questions about the use of these drugs in

premenopausal women.

So, as far as this other manifestation of

a cardiovascular hazard is concerned, initiation

and acceleration of early atherogenesis occurs in

response to deletion of the prostacyclin receptor.

I haven't gotten into mechanism but it fosters

platelet and neutrophil activation and vascular

interactions of these cells, and removes the

constraint on attendant oxidant stress.

Now, we know that hypertension, which is

also a consequence of inhibition of this pathway,

itself accelerates atherogenesis. So, one could

imagine that the direct and indirect effect could

converge to transform cardiovascular risk.

Finally, again COX-1 is playing a modulatory role.

There is a lot of speculation, which will

no doubt be addressed in this meeting, as to

whether in the APPROVe study we actually saw a

delayed appearance of augmented cardiovascular

risk. I think, for me, the answer is we are not so

sure but, if we did, this mechanism would explain

not only early events but also the delayed

emergence of cardiovascular phenotype.

The other thing that is often trotted out

is, well, but people on aspirin have had some of

these events. Well, of course, people on aspirin

also have myocardial infarctions. But I think it

is worthwhile remembering as we consider that

prostacyclin will buffer effects of thromboxane on

blood pressure, atherogenesis, hemostasis and,

indeed, cardiac damage, which I haven't gotten into

today. It acts as a general constraint on any

agonist that acts harmfully on these systems. So,

one would expect aspirin, in a perfect world, to

damp rather than abolish the signal.

So, I think, if you will pardon me just

for a moment to muse, one could relate the ability

to detect a signal, expressed here as maybe numbers

needed to treat or trial duration, as a function of

the underlying cardiovascular risk of the patients

involved. The higher the risk, the more you would

be able to detect it easily. The lower the risk,

it may require that you either perform a very large

study or go on for a very long time because we are

all mindful of the fact that clinical trials, even

randomized clinical trials, are very crude detector

systems for uncommon risk.

Additionally, other elements will impact

on this, including elements related to drug

exposure and the degree of selectivity that is

actually attained in vivo. So, I think in some of

the efforts to dismiss this idea of a class-based

effect some have lost sight of the fact that one

would expect not only the underlying substrate to

be relevant, but elements of drug exposure like

dose, duration of dosing, duration of drug action

and, indeed, concomitant therapy to be relevant to

the ability to detect a risk. So, one is looking

for a needle in the haystack and, to some extent,

when one finds the needle it doesn't really matter

how long it has been in the haystack.

So, let's consider the extreme phenotypes

of cardiovascular benefit and hazard in this

pathway. First of all, let's consider aspirin.

Here we have a sustained mechanism of action that

leads to complete and sustained inhibition of

COX-1. Even low dose aspirin inhibits prostacyclin

to a minor degree. But one would expect, and one

sees, a cardiovascular benefit from aspirin, at

least in the secondary prevention of stroke and

myocardial infarction.

In the case of COX-2 inhibitors one sees a

reversible inhibition of COX-2. One also sees

variable degrees of inhibition of COX-1 but,

because of that non-linearity that I mentioned to

you in the relationship, effectively this makes

these drugs selective for COX-2 because you have no

inhibition of COX-1 dependent platelet function.

That brings me to the last topic that I

would like to address, and that is what about the

traditional NSAIDs? Well, here is one way of

comparing aspirin to a prototypic NSAID, ibuprofen.

You take healthy volunteers, you administer them

low dose aspirin to stead-state efficacy, or

ibuprofen 3 times a day to a steady-state effect,

and you look at the offset of effect on enzyme

inhibition and inhibition of function.

With aspirin you see sustained inhibition

over the 24 hours after stopping the drug. As you

would expect, with stopping ibuprofen you see

offset of this reversible inhibitor on the enzyme.

From whatever I have told you about that

non-linearity in the relationship, you are not

surprised to see a steeper offset of inhibition of

function.

Well, of course, we have no randomized,

placebo-controlled trials of traditional NSAIDs.

We have various overviews of the epidemiological

experience, with all the limitations of that

approach and we can see that ibuprofen looks like

it is not really altering cardiovascular hazard.

There seems to be a sort of 10 percent or so

reduction with naproxen, particularly 500 mg twice

a day which was the most commonly used dosage in

these studies.

Now, this would be like a dilute aspirin

effect and, obviously, has relevance to the

interpretation of studies like VIGOR and some of

100

the experience with the etoricoxib that you will

hear about as to whether naproxen is actually

behaving like aspirin.

Well, I think actually the epidemiology is

entirely consistent with the clinical pharmacology

of naproxen. This elegant study was performed by

Patrignani. Again we are looking at the offset

action of aspirin and naproxen 500 mg per day

administered to steady state. We are looking at

inhibition of enzyme function, and we see with

aspirin exactly what we would have expected,

sustained inhibition. However, at the end of a

typical dosing interval for naproxen we see

heterogeneity of response. In fact, everybody is

at 95 percent or lower, suggesting that within the

dosing interval there is a variable degree of

cardioprotection afforded through this mechanism,

which would be consistent with the dilute aspirin

effect from the epidemiology.

This is a plot of the IC-50 for inhibition

of COX-2. This is inhibition of COX-1 in whole

human blood. As we move in this direction we are

getting more selective for COX-2. It brings us

back to a point that arose in Byron's study, and

that is that although there is a difference in

101

potency, celecoxib and diclofenac look remarkably

similar.

I would also remind you that naproxen,

bearing in mind the Aleve study fiasco, is on the

other side of the line, just like ibuprofen is, and

exhibits preference for inhibition of COX-1.

Well, you have had a nice job giving you a

full data set, demonstrating that actually in whole

human blood diclofenac and celecoxib are

superimposable. So, I would contend that through

various lines of evidence diclofenac is probably a

selective COX-2 inhibitor like Celebrex.

Consistent with that is a pharmacodynamic

interaction where we showed that prior occupancy of

the COX-1 site by a typical mixed inhibitor like

ibuprofen would block access of aspirin to its

102

target acetylation site. If we give aspirin and

ibuprofen chronically we actually see a pattern

that looks just like giving ibuprofen alone, an

onset of action and a steep offset of function.

However, if we substitute diclofenac for aspirin it

looks like giving aspirin alone, which is

consistent with the type of information you get

with a selective COX-2 inhibitor like rofecoxib or

celecoxib in this assay.

So, I think we can start thinking of

diclofenac as Celebrex with hepatic side effects.

It has the same selectivity in whole blood in

vitro. It has no pharmacodynamic interaction with

aspirin. It has no clinical interaction with

aspirin in the one epidemiological study which has

addressed this interaction with the two drugs.

Also, it is consistent with the superimposition of

the GI and cardiovascular events in the

retrospective look at CLASS in non-aspirin users.

So, I would suggest the two trials that

you will hear about, EDGE and the ongoing MEDAL,

are actually the first trials that are a comparison

103

within the class.

Well, let's come back to this

relationship. I would remind you that while we

have very strong evidence for this being true, we

have almost no evidence that this is true. The

conjecture of this discordance underlies the

argument for the fact that we have a problem with

selective COX-2 inhibitors but, you know something,

we have a problem with all of these drugs which

clearly obscures the message. We have no evidence

for that and you will hear people parsing in

meta-analyses naproxen versus non-naproxen NSAIDs.

Well, I don't think that is a legitimate

lumping of non-naproxen NSAIDs, which is really

diclofenac plus ibuprofen in most instances. I

think it is as legitimate to consider them all

individually as it is to consider naproxen

individually.

So, could there be a hazard from a

non-naproxen NSAID like ibuprofen where there is

coincident inhibition of COX-1 and COX-2 over

typical multiple dosing interval? If there is a

104

discordance in the relationship between inhibition

of enzyme function and inhibition of enzyme

product, then there might be a narrow part of the

dosing interval where there could be a potential

exposure to risk. But the likelihood of detecting

this notional risk would be much less than the

likelihood of detecting the clear evidence-based

risk of selective inhibitors of COX-2.

So, there is some suggestion that naproxen

achieves sustained platelet inhibition in some

individuals. I like to think of it as a dilute

aspirin. There is evidence that diclofenac is

Celebrex. There is evidence that ibuprofen may

undermine the benefit from aspirin, although that

is not yet answered one way or the other with a

controlled trial. And, I would say quite

forcefully there is no rationale for lumping

diclofenac and ibuprofen as non-naproxen NSAIDs in

meta-analyses and the like.

I am not sure when a canard becomes a dead

duck--

(Laughter)

--so I decided to dismiss some of the

things that I think are worth dismissing and call

them dead dragons. First of all, naproxen clearly

105

is not the full explanation of VIGOR.

Here is another one that needs to be

chopped down, hypertension is not a different

mechanism.

There are a lot of off-target fantasies

being touted around at the moment, strange chemical

interactions that haven't actually been shown to

occur in vivo yet but are postulated as the

explanation for a drug-related rather than a

class-based effect.

Oddly, we never heard any of this

conjecture when we were considering how all the

drugs in this class afforded relief from pain and

inflammation.

Here is another nice notion that makes

clinical pharmacologists squirm in their seat, it

is just a matter of reducing the dose. Well, there

is a lot of interindividual variability in response

to COX-2 inhibitors and we all have our own

106

dose-response curves. It has been an approach in

the past when a hazard emerges to suggest that in a

population sense one just cuts the dose--perhaps in

a population sense but it certainly does not

obviate the possibility of individual hazard.

Finally, if there ever was one, I think we

have certainly moved beyond the need for a trial of

a COX-2 inhibitor in patients with acute coronary

syndrome. Indeed, I feel that the evidence that

supports a trial in patients at high cardiovascular

risk to detect protection is scientific quite weak,

and in the face of an emergent hazard is ethically

questionable.

Indeed, in the case of mice if one

combines a thromboxane antagonist as a surrogate

for the suppression of thromboxane by low dose

aspirin with a COX-2 inhibitor, one doesn't see any

benefit in terms of atheroprotection, but what one

does see is the loss of the fibrous cap in the

combination and necrosis of the atherosclerotic

core, consistent with destabilization of the

plaque.

Finally, and you will be glad to know it

is finally, I would just like to mention a couple

of things relating to where we might go from here.

107

Well, I think clearly an easy thing to write down

and perhaps a more tricky thing to do is to exclude

patients at high intrinsic risk of thrombosis, and

you have heard my views on that. Dose reduction

alone is a simple message. It has a political and

legal appeal but in pharmacological terms it is

misleading.

I think we are likely to subject new drugs

that might be approved from this class to

significant hurdles before they are approved. It

seems logical to me that existing drugs in this

class should be subject to the same hurdles to

retain approval, particularly for extended dosing.

And, I think that frankly one should logically

restrict the duration of dosing until the

parameters of safety for extended dosing have been

established.

I mentioned interindividual variability,

and these are log scales but they illustrate

108

looking at inhibition of COX-2 either in the

typical ex vivo assay or by excretion of

prostacyclin metabolite or inhibition of COX-1,

that with this sort of display of the data to

highlight it, there is considerable interindividual

variability of response. This is no surprise. It

is true of all drugs.

But perhaps we can exploit the biochemical

variability, the physiological response variability

and, indeed, perhaps some genetic markers such as

these polymorphisms associated with metabolism of

drug or these polymorphisms in cyclooxygenase-1 to

try and identify those patients at emerging

cardiovascular risk before they culminate in

events. So, you might say that the future of these

drugs or the challenge to the future of these drugs

is that if their value--and I believe they have

value as a class--is to be harvested, then to

manage the risk we have to actually move to an

example of personalized medicine.

One would want to obviously restrict these

drugs in some way to people who really needed them,

109

for GI reasons. We need to determine whether risk

transformation actually occurs during chronic

dosing and, if so, whether we can detect it. And,

it is likely, because we have so few events in any

one trial, we can only do this by a combined

analysis across the class in relevant trials.

Then, obviously, we would have to validate

prospectively such an index of emergent risk in a

prospective trial.

So, I really thank you for your patience

and I would like to conclude. Selective inhibitors

of COX-2 depress prostacyclin without a concomitant

inhibition of

thromboxane-A2. This can result in an augmented

response to thrombotic and hypertensive stimuli and

acceleration of atherogenesis in mice. Indeed, the

terrible beauty of this unfolding drama is how

faithfully the emerging clinical information has

fitted the predictable science, and that should

reassure us in terms of the likelihood that the

science can predict a way to conserve the value of

these drugs while managing the risk.

An increase in MI and/or stroke has been

seen at last count, as of yesterday, in 5

placebo-controlled trials with 3 structurally

110

distinct COX-2 inhibitors. Given the bulk of

evidence, the mechanism-based evidence from mice

and people, the pharmacopeidemiology and this, it

seems to be that most rational people would accept

a class-based mechanism as they did for efficacy.

Finally, hazard would be expected to

relate at the individual level to the drug

selectivity attained in vivo, dose and duration of

exposure and to interindividual differences in drug

response. Thank you.

DR. WOOD: Thank you. Just before you sit

down, one thing you seemed to be saying is that we

should exclude patients at high risk. The point

estimate in the APPROVe trial for people with no

symptomatic history of heart disease is 1.6 so that

would be one way you would exclude people, I guess,

but the point estimate remains 1.6. Does that

bother you?

DR. FITZGERALD: No, as I alluded to, I

111

think the nature of the information we have in the

APPROVe trial so far remains to be played out.

Clearly, there was an attempt to exclude people at

high cardiovascular risk but we all know that

people who are at risk slip through any exclusion

criteria. So, one question is, is all that we are

seeing people who, for one reason or another, are

predisposed to thrombosis and they are the people

that are having events? Or, are we seeing people

who through atherogenesis transform their risk?

Or, are we seeing some combination of the two? I

don't think we know the answer to that.

DR. WOOD: We are running behind time so

we will call a break right now and give everybody a

moment or two to get out. Before we do that, Dr.

Galson wants to say some things and then, whenever

he is finished, we will take a break and we will

reconvene at 10:15. So, those of you who don't

want to hear what Dr. Galson has to say can get out

now and the rest--

DR. GALSON: No, no, just a very brief

announcement, and that is we have a space problem

112

in this facility. There are more people than we

have seats for. So, we have established a live

video feed in our advisors and consultants

conference room on the FDA campus at 5630 Fishers

Lane, designed for FDA employees only. So, FDA

employees who may be sitting in the public section,

I strongly urge you to please move to that area to

make more room for the public and, of course, you

will need your FDA ID badge to get into that space.

But it is ready now and if you could move at the

break, it would be great. Thanks.

DR. WOOD: Okay, we start promptly at

10:15.

(Brief recess)

Committee Questions to Speakers

DR. WOOD: Let's get started and get the

two previous speakers up for questions, Dr. Cryer

and Dr. FitzGerald. Yes, Susan?

DR. MANZI: I have a question for Dr.

FitzGerald. This is really in reference to your

suggestion that we exclude people with high

thrombotic potential. I think there is clearly

113

evidence that the natural aging process is

associated with less effective fibrinolytic system,

really increased thrombogenic potential with high

levels of fibrinogen, PI-1 platelet aggregation,

and considering that the elderly population is a

huge target for non-steroidals, would you consider

age as a risk?

DR. FITZGERALD: Well, I think, as you

indicate, lots of things happen as we get older

including the complexity of administering drugs and

it ultimately culminates in death. But I think the

issue of determining cardiovascular risk is

actually a very challenging one because it includes

continuous and discontinuous variables. It is easy

to say if you have had a heart attack or a stroke

you are statistically at greater risk of having

another one. It is harder to say that at an

individual level, somebody who hasn't had a heart

attack or a stroke has a cluster of variables that,

in the eyes of their physician, determines their

cardiovascular risk.

With some of the discontinuous variables

114

like some of the genetic mutations we can have an

attributable risk that we can measure but, again,

that can play geometrically into other small but

absolute risks. So, unfortunately, I think it is

where the art and science of medicine intersect.

DR. WOOD: Richard Cannon?

DR. CANNON: You asked my question.

DR. WOOD: Joan Bathon?

DR. BATHON: We know that patients with

rheumatoid arthritis and other inflammatory

conditions are at higher risk for developing acute

MIs and strokes, and these are the very patients

who are taking NSAIDs chronically. This is a big,

confounding problem in interpreting some of the

data and I am wondering if you have any thoughts.

The reigning theory is that there is more

atherosclerosis and RA due to vascular inflammation

but I am wondering if you have any thoughts about

whether the NSAIDs might be the sole contributor to

increased events in these folks.

DR. FITZGERALD: Right. As I indicated,

through a COX-2 inhibitory mechanism one would

115

anticipate that the clinical substrate of

underlying cardiovascular risk would be one of the

modulators of either individual hazard or the ease

of detecting hazard with this crude detector system

we call clinical trials.

As you know, the relative risk of heart

attack or stroke and RA is increased by about 50

percent on average compared to RA or no arthritis.

As a population that would be one of the

ingredients predisposing towards emergence of a

hazard. Of course, within that population there is

a very substantial interindividual variability

conditioned by many other factors that impinge on

cardiovascular risk. So, at the time when we were

naval gazing, looking at the contrast between CLASS

and VIGOR, amongst the many things that were

discussed was whether the preponderance of RA

patients in VIGOR versus the preponderance of OA

patients in CLASS may have been a factor. I think

it is reasonable to say it may have been a factor

but I don't think we can really take it beyond

conjecture in light of any current evidence that I

116

am aware of.

DR. WOOD: Garret, let's cut to the chase.

Is what you are saying--that was such a long

answer, I am not sure what it meant!

(Laughter)

Is what you are saying that you think that

COX-2 inhibitors have an effect here that the most

selective, so-called non-selective like diclofenac

and naproxen may also have an effect, and the

non-selective, non-steroidals do not have an

effect, or at least have not been shown to have an

effect? Is that your position? If it is not,

correct that.

DR. FITZGERALD: No, I think that is

pretty true.

DR. WOOD: So, that is what you wanted us

to take away from all the mice and stuff, is it?

(Laughter)

DR. FITZGERALD: You have such a way with

words!

DR. WOOD: Because I am a Scot.

(Laughter)

DR. FITZGERALD: You are very economical

with them.

DR. WOOD: Exactly.

117

DR. FITZGERALD: Unfortunately, reality is

conditioned by a lot of different factors. I think

one of the things, both in terms of benefit and

hazard, we have paid insufficient attention to is

variability in drug response between individuals,

and I think actually one of the things that has got

us to today is not paying enough attention to that.

But I think one of the ways out of the challenge

that faces us today if we are to conserve the value

is to exploit that variability in imaginative ways.

So, I think that that is a tractable issue.

DR. WOOD: Okay. Dr. Abramson?

DR. ABRAMSON: Yes, Garret, even though

you are under the weather I wanted to follow-up

with Dr. Wood's question and put you on the spot a

little bit. It is partly definitions because we

use the word NSAIDs which we elect by inhibiting

COX-2s. Based on your presentation, it is clearly

a continuum and there are highly selective drugs.

118

There is a cluster of five or six drugs, like

diclofenac, that are in vitro at least comparably

COX-2 selected. Then you have these very complex

stories of what one might call functional COX-2

selectivity, which is based on the fact that the

COX-1 inhibition may be more transient effectively

than a more prolonged COX-2, which would give you

imbalance. So, I guess the "put on the spot"

question is what do you define as the class? How

do you propose we should think about this continuum

and personalize medicine?

DR. FITZGERALD: I think you are right. I

would remind all of us that COX-2 inhibitors are

NSAIDs; they were never anything else. They are

NSAIDs that are selective for COX-2 and, as you are

rightly pointing out, this is a continuous variable

and within each drug, as I tried to point out,

there is the same continuous variable between

individuals. So, my 800 mg of Celebrex may be your

200 mg of Celebrex for example.

So, I think all I am trying to raise is

that there is clearly a mechanism which reflects

119

the selective inhibition of COX-2. That selective

inhibition of COX-2, in terms of hazard, is

modulated by COX-1 inhibition that occurs at the

same time if it is sufficient to inhibit platelet

activation for example. So, I can't simplify that

because I believe there is that complexity, but

within the class--and I am referring to the class

as the mechanism by which selective inhibition of

COX-2 is attained--I think there is clearly a

mechanism that explains everything that we have

seen.

At the individual level this issue of a

continuum comes into play because not only is there

a continuum in terms of drug action and the degree

of selectivity attained in an individual, but also

many other factors impinging on cardiovascular risk

that condition the emergence of that hazard at the

individual level.

DR. WOOD: Steve?

DR. NISSEN: Yes, I have two quick

questions. You know, I want to talk with you a

little bit more about this issue of dose

120

dependency. I want to make sure we didn't

misunderstand you. What you are saying I believe

is that there is sufficient overlap in the

biological effects that a low dose in one patient

may be equivalent to a high dose in another. But

you didn't mean to suggest that we don't see

evidence, as I think we do see from the trials,

that the higher the dose of the drug on a

population basis, the more we see--

DR. FITZGERALD: No, no, clearly there is

evidence of a dose-related effect in populations.

I am talking more at the individual level, that the

assurance to a population based on population type

evidence that all you need to do is reduce the dose

and you, as an individual, will be protected from

hazard is a false one.

DR. NISSEN: Yes, but it is quite relevant

obviously to our discussions on Friday because one

of the strategies to limit risk with this class of

drugs is to limit dose--

DR. FITZGERALD: Sure.

DR. NISSEN: --and it may not make the

121

hazard go away but it may make it smaller, and we

are going to have to explore that in some detail

before we finish.

DR. FITZGERALD: Well, I think that

distinction between reducing it as opposed to

making it go away and the distinction between

population hazard and individual hazard is an

important one. It is the reason that I raised that

particular point because I think that had not

received sufficient attention.

DR. NISSEN: The second question I have

is, you know, we have very few direct head-to-head

trials amongst the so-called COX-2 inhibitors, but

we do have for hypertension and there seemed to be

really pretty striking differences in the

hypertensive response between rofecoxib and

celecoxib. Would your point of view be that those

differences are strictly a matter of COX-2

selectivity of the two drugs, or do you think that

it is possible that there is some dissociation in

the hypertension response?

DR. FITZGERALD: I would make two points.

122

I would say, first of all, that in that particular

comparison, again on average, we would anticipate

that selectivity and duration of action would be

confounded and it would be impossible to really

segregate the two.

The second is that, in a sense you pressed

my button, I believe we have not performed the

studies in hypertension that let us address the key

questions that are on the table, and that is

standardizing for the degree of selectivity

attained or the degree of COX-2 inhibition attained

do drugs come apart? That question has been on the

table since the mouse studies of Breyer and

Kaufman, and perhaps the first signal of that is

the epidemiological overview analysis from

Australia. But, in fact, we have never performed a

study to address the hypothesis and I think it is

timely that we do.

DR. WOOD: I see Dr. Cryer. Did you want

to say something?

DR. CRYER: Dr. Cryer has a question.

DR. WOOD: Go ahead.

DR. CRYER: Garret, you clearly made the

point that diclofenac appears to have some COX-2

selectivity. In fact, I think you called it

123

celecoxib with hepatic side effects. You also made

the point that we should subject drugs already

approved to the same requirements. So, the

specific question I have for you is are you

suggesting that we should evaluate diclofenac as

well for its potential cardiac effects?

DR. FITZGERALD: Yes, I think there are

quite a few unanswered questions on the table. I

think clearly the diclofenac question is one of

them. I think there are other drugs that fall into

potentially the same situation, like meloxicam and

nimesulide which, again, based on the IC-50

comparisons look awfully similar to diclofenac and

Celebrex but we just don't have the information

even at a more fundamental level than outcome

studies. So, I think those questions are on the

table.

The reason I made the comparison between

retention of approval and gaining approval is that,

124

to me, if we do actually have to address some

questions to determine the parameters within which

drugs in this class can be administered safely and

that would be a hurdle that any new drug would be

required to overcome, in logic to me, it would be

sensible to apply the sam standard to the extended

dosing of drugs that already are on the market as a

condition of their retention of approval.

DR. WOOD: Dr. Shafer?

DR. SHAFER: Yes, this is the question we

just talked about briefly at the break, but as you

pointed out, low dose aspirin gives you 100 percent

inhibition of COX-1. One might think then that low

dose aspirin plus a COX-2 selective antagonist

might give you the same risks as a non-selective

NSAID. Yet, in all the studies where they had

aspirin present and they showed a CV risk, when

they stratified by aspirin, among aspirin users the

hazard didn't go away. Now, what did happen is

that some statistically significant hazards became

non-statistically significant hazards but the

actual magnitude of the hazard, at least as far as

125

I can tell in all the studies that I looked at,

didn't change. I am having trouble understanding

how that is consistent with the whole thing being

the COX-2 imbalance.

DR. FITZGERALD: Right. So, one important

missing dimension in your question is time. One of

the key ingredients of aspirin's ability to afford

cardioprotection is that while it inhibits COX-1

like a ibuprofen does, it does it molecularly in a

quite distinct fashion. This results in sustained

maximal inhibition throughout the dosing interval.

By contrast, in the typical non-steroidal you are

in the red zone for platelet inhibition transiently

in the dosing interval. Therefore, one would not

expect the combination of, say, ibuprofen with a

COX-2 inhibitor to be similar to aspirin with a

COX-2 inhibitor in terms of cardiac protection.

DR. SHAFER: Doesn't that head in the

opposite direction?

DR. FITZGERALD: In terms of which?

DR. SHAFER: The fact that the aspirin's

effect is sustained because, you know, it is

126

covalently bonded there--the fact that you are

having a sustained aspirin effect means that you

should absolutely--I mean, it would seem to me that

that would really try to make the COX-2s look--

DR. FITZGERALD: Well, I will come back to

what I said during my talk, and that is that I

think a real mistake is to think of this as a yin

and yang type of seesaw arrangement between

thromboxane and prostacyclin. We know that

prostacyclin acts as a general biological

constraint on anything that will activate

platelets, elevate blood pressure, accelerate

atherogenesis, and so on. So, a priori we would

expect that aspirin would damp rather than abolish

the signal.

Now, I would contend that, first of all,

we have never formally addressed this and, in terms

of the trials that have events, although we have

attempted to look at the relationship to aspirin

the numbers are so vanishingly small that it is

really conjecture. But one would expect a signal

to be damped. Indeed, from some of the

127

epidemiology that is sort of what we are seeing,

you know, a signal goes away at 25 mg of rofecoxib

if they are on aspirin but not at 50, that sort of

stuff. But I would be the first to agree that this

is really a crude stab at the issue that you are

trying to get at.

DR. WOOD: Yes, and these studies did not

stratify by aspirin use. They were post hoc

analyses in the majority of cases. Dr. D'Agostino?

DR. D'AGOSTINO: I would like to go back

to the question that was asked right after the

break about the age. If you tried to say, well,

the perfect way of doing this is to make sure that

people at high cardiovascular risk aren't going to

take the drug, then males over 60, for example, are

almost certain to be excluded. How realistic--

DR. FITZGERALD: Certainly I am not trying

to be dictatorial--

DR. D'AGOSTINO: No, no, your suggestion

is fine, it is just how do you implement it?

DR. FITZGERALD: Yes, so I think all one

can really hope to do is set the bar at some low

128

level and then signal it in a way that is explicit

and leave it to the patient-doctor relationship to

divine the individual behavior. I would love to

say there is a different way of doing it but, yes,

as we get older our cardiovascular risk goes up and

multiple other things. But that is where the

balance against value comes into play. As we get

older with get arthritis; as we get older we get

more GI bleeds on non-steroidals.

DR. WOOD: Okay, we got it. Let's not go

too far there. One more question from Dr.

Gibofsky.

DR. GIBOFSKY: Dr. FitzGerald, in response

to Dr. Nissen, I believe, you raised the notion and

asked us to think about population variation as a

factor in addition to individual variation. One of

the things that I am struggling with is exactly

that, and one of the concerns I have is to what

extent then can one extrapolate observations in

populations of patients who may have Alzheimer's

disease or who may have taken a drug for polyp

prevention to the population of patients who are

129

taking the drug for their arthritis?

DR. FITZGERALD: Well, I think in a way

this whole cathartic experience is a cardinal point

in the way that we look at drug development. You

know, we have talked about individualized medicine

for a long time and never really had to care, and

here is a situation where we actually do have to

care and it is at the forefront of how we may or

may not be able to find a way out of this. You are

absolutely right, there may be factors associated

with an incident disease which is under study which

modulates the importance or non-importance of the

signal; modulates the way that drugs are

metabolized; may be associated with genetic

variance that influence outcome as well.

DR. WOOD: Any other questions for the

last two speakers?

(No response)

In that case, let's move on to the

sponsor's presentation. I understand Dr. Kim is

going to present first.

Sponsor Presentation: Vioxx (Rofecoxib)

DR. KIM: Mr. Chairman, members of the

advisory committee and FDA and ladies and

gentlemen, my name is Peter Kim and I am President

130

of Merck Research Laboratories. My colleagues and

I welcome the opportunity to present information at

this advisory committee meeting, and I would like

to begin with just a few introductory comments.

As you will hear, to determine both its

risks and its benefits, Merck extensively studied

Vioxx before seeking regulatory approval to market

it, and we continued to conduct clinical trials

after the FDA approved Vioxx.

As Merck continued to monitor the

cardiovascular safety of Vioxx, we recognized the

value and interest in obtaining additional

cardiovascular safety data on this medicine. After

deliberations with numerous outside advisors, Merck

developed and discussed with FDA a plan to

prospectively analyze cardiovascular event rates

from 3 large placebo-controlled trials.

It was preliminary information from one of

these long-term trials, the APPROVe trial, that led

131

to Merck's decision to voluntarily withdraw Vioxx.

When Merck made the decision to voluntarily

withdraw Vioxx from the market, we stated that we

believed that it would have been possible to

continue to market Vioxx with labeling that would

incorporate the data from the APPROVe trial. We

concluded, however, based on the science available

at that time, that a voluntary withdrawal of the

medicine was the responsible course to take given

the availability of alternative therapies and the

questions raised by the data.

Since that time new cardiovascular safety

data for other COX-2 inhibitors have become

available and were reported on just this week in

the New England Journal of Medicine. We look

forward to hearing and seeing presentations of

these data and to hearing discussions and

interpretation of them during this advisory

committee meeting. Thank you, and now I would like

to turn the podium over to Dr. Ned Braunstein.

DR. BRAUNSTEIN: Good morning, Dr.

Chairman, members of the availability committee,

132

FDA, I am Dr. Ned Braunstein, Senior Director of

Merck Research Labs.

Millions of patients suffer with painful

arthritis and need effective therapies. The recent

data that have come to light on NSAIDs and

selective COX-2 inhibitors raise many questions.

Patients and physicians need information and

guidance on the use of these effective medicines

that we know are not without risk. We recognize

that the cardiovascular safety of the NSAID and

coxib classes is an important public health issue

and we welcome the opportunity to present this

advisory committee information that we believe will

help the FDA and the committee in their work in

developing recommendations in the best interest of

patients. To assist us today, we have

brought along as consultants Dr. Marc Hochberg from

the University of Maryland School of Medicine, Dr.

Marvin Konstam from Tufts University School of

Medicine, and Dr. Loren Laine from the University

of Southern California School of Medicine. They

are here to help answer your questions and

133

otherwise assist the committee.

Merck's objective today is to provide you

with data on rofecoxib and review how those data

affected our assessment of risk/benefit over time.

The presentation will focus on GI and

cardiovascular data or rofecoxib, starting with the

data in the original NDA and proceeding through the

voluntary withdrawal of Vioxx and the APPROVe data.

In talking about the data, I will try to

highlight some of the methodology we used to

obtain, adjudicate and analyze cardiovascular data,

and I will spend some time discussing the

considerations that went into designing a study of

cardiovascular outcomes with rofecoxib as the

information may be useful in considering similar

studies.

The presentation of data will end with a

presentation of new exploratory analyses that we

have performed and I will follow with a

risk/benefit assessment, the review the major

outstanding questions of the day, and the next

steps we are taking and/or propose.

I will start with an overview of the

issues we face today. Starting with the GI tract,

as you have already heard, NSAID gastropathy has

134

been the most common cause of drug-related

morbidity and mortality in industrialized nations.

The development of rofecoxib was based on the

desire to limit and reduce this problem.

You have also heard already about the

COX-2 hypothesis. I just want to emphasize two

points. First, all NSAIDs inhibit COX-2 in a

dose-dependent manner and selective COX-2

inhibitors do not inhibit COX-1 at clinical doses.

The rofecoxib develop program confirmed

the COX-2 hypothesis and demonstrated a reduction

in clinical upper GI events, that is, actual GI

outcomes with rofecoxib versus non-selective

NSAIDs. This was shown for rofecoxib in the VIGOR

study and, based on that, rofecoxib was the only

selective COX-2 inhibitor with a modified GI

warning. Since that time we have accrued

additional information that extend the GI benefit

of rofecoxib and have shown that the reduction in

135

clinical upper GI events is consistently seen with

rofecoxib versus diclofenac, ibuprofen and

naproxen.

Although rofecoxib is associated with a

reduced rate of upper GI events compared to these

NSAIDs, rofecoxib is not placebo. In addition to

the upper GI findings, we have also observed a

reduced incidence of lower GI events compared to

naproxen in VIGOR, So, although there remain some

unanswered questions, for example for aspirin

users, the GI benefit for rofecoxib is clear.

As we have also learned, there are

important cardiovascular findings with these drugs

and perhaps with the larger class of NSAIDs. In

1998 Merck had implemented an adjudication standard

operating procedure to methodically study the

cardiovascular effects of its COX-2 selective

inhibitor drugs in clinical trials. Clinical data

on thrombotic cardiovascular events with rofecoxib

show an increased risk of events relative to

placebo. This was seen in APPROVe with

long-standing use.

In contrast to the difference seen from

placebo, we have not observed a difference in

cardiovascular event rates between rofecoxib and

136

NSAIDs other than naproxen. Long-term data,

however, are limited. In contrast to what had been

observed versus the placebo, the increased risk

compared to naproxen appears after short-term use.

I think it is worth noting that similar

observations have now been made with other

selective COX-2 inhibitors. We believe that these

new data on rofecoxib and COX-2 inhibitors raise

several questions about these drugs important to

the public health.

First, based on the data available, how do

we currently assess the relative risks or benefits

of selective COX-2 agents? I cannot speak to the

data on all of these drugs but I can talk about

rofecoxib. Clearly, there are risks versus

placebo, and not just cardiovascular risks.

however, placebo is not a choice for patients with

chronic arthritis and pain who require chronic

NSAID therapy. For these patients the question is

137

the risk and benefit of selective COX-2 agents

versus non-selective NSAIDs. I will present data

on this question related to the GI and

cardiovascular safety of these drugs.

Second, can we identify factors associated

with the observed increased risk for thrombotic

cardiovascular events with these drugs? Although

we do not have definitive answers, I will present

the data that we have.

Finally, is the increased thrombotic

cardiovascular risk that we have observed with

rofecoxib indicative of a larger class effect of

COX-2 inhibitor? If so, how big is the class?

That is perhaps the central question of this

meeting. At present we do not know the long-term

cardiovascular effects of traditional NSAIDs.

Other than aspirin, these agents have not been

studied long term versus placebo. We believe that

long-term studies are needed and, in particular,

comparator studies between selective COX-2 agents

and non-selective agents to better understand the

relative risk/benefit profiles.

I will now turn to a presentation of the

data, and will do so chronologically as it

highlights the magnitude of data that were

138

ultimately needed to dine the long-term

cardiovascular risks of selective inhibitors. This

information may be useful regarding the development

of future COX-2 inhibitors and in informing this

committee on its decisions.

I would like to start by reviewing the

initial GI and cardiovascular data that were in the

new drug application. There were two main clinical

components of the GI safety program in the original

rofecoxib NDA, the GI endoscopy studies, which are

described in your background package, and a pooled

analysis of clinical upper GI events, shown here.

Investigator reports of suspected upper GI

perforations, ulcers or bleeds or PUBs were

adjudicated by an external committee of blinded

adjudicators, and the confirmed events formed the

basis of this prespecified analysis.

The Kaplan-Meier plot of the data is shown

on this slide. Throughout my presentation I will

139

be showing several of these so I would like to take

some time to walk through this first one. Time is

shown on the X axis, and below that the number of

patients remaining in the studies at the different

time points. Cumulative incidence is shown on the

Y axis and also shown are summary statistics,

relative risk confidence interval and a p value.

At the time of the original NDA a

significant difference was demonstrated between

rofecoxib and the combined NSAID comparators,

mostly data on ibuprofen and diclofenac. The

relative risk of 0.45 corresponded to a 55 percent

risk reduction with rofecoxib and, thus, we believe

that we had established a GI safety advantage over

these older NSAIDs.

These are the cardiovascular safety data

from the OA development program. Rates per 100

patient years of investigator reports or cardiac,

cerebrovascular and peripheral arterial and venous

serious thrombotic events were examined both in

aggregate, as shown on this slide, and also in

individually, as shown in your background package.

140

As you can see, then rates were similar for

rofecoxib compared to the NSAIDs diclofenac,

ibuprofen and nabumetone and for rofecoxib compared

to placebo.

These cardiovascular and GI data, along

with our other data, were submitted to FDA in 1998

as part of the new drug application for rofecoxib.

They were discussed at the April, 1999 Arthritis

Advisory Committee and the FDA concluded that there

was a favorable risk/benefit profile for rofecoxib,

and rofecoxib was approved in May of 1999.

Around that time we were completing our

Phase III osteoarthritis studies. The results of

studies that we were doing in collaboration with

Dr. Garret FitzGerald became available, and he has

already told you about those and the hypothesis

that selective COX-2 inhibitors could be

prothrombotic by inhibiting systemic prostacyclin

production without inhibiting thromboxane

production.

In addition to that hypothesis, there were

other hypotheses being discussed in the clinical

141

literature and in the basic science literature at

that time, including the possibility that NSAIDs,

through their effects on COX-1, might decrease the

risk of cardiovascular events. Another was that

perhaps by inhibiting COX-2 there may be a

beneficial effect by inhibiting the enzyme in

atherosclerotic plaques.

Merck recognized that it would be

important to continue to acquire cardiovascular

data with its selective COX-2 inhibitors. To

address these hypotheses, in 1998 Merck initiated a

vascular event adjudication standard operating

procedure to standardize the evaluation of

cardiovascular events in all of its COX-2 inhibitor

studies. Adjudication of events was based on

predefined criteria. Under the standard operating

procedure all source documentation on events was

collected and the data were then reviewed by

blinded, external adjudication committees. With

this procedure, over 92 percent of cases had

sufficient data for definitive determination and

adjudication. Thus, we can be confident in the

142

quality of the data. By eliminating questionable

events, we would amplify and improve the clarity of

any signal if present. The standard operating

procedure called for a pooled analysis of events

across all studies to improve the precision of what

would be obtained from individual studies.

In order to obtain more data on the effect

of rofecoxib on GI outcomes Merck initiated the

Vioxx GI Outcomes Research, or VIGOR, study in

January, 1999. GI events would be adjudicated

using the same approach as had been done for the

osteoarthritis studies. The cardiovascular events

in VIGOR fell under the new standard operating

procedure.

VIGOR was designed to definitely assess

the GI components of the COX-2 hypothesis. It was

conducted exclusively in rheumatoid arthritis

patients because Merck believed that a GI benefit

had already been established in osteoarthritis

patients. Rofecoxib of 50 mg, 2-4 times the

recommended chronic dose, was chosen to provide a

rigorous assessment of safety. We chose as a

143

comparator naproxen 500 mg twice a day to extend

the GI findings to an additional NSAID and because

that was the most commonly prescribed NSAID regimen

in rheumatoid arthritis. Patients using aspirin

were excluded to avoid COX-1 inhibition as this

could confound the ability to rigorously assess the

COX-2 hypothesis.

The primary endpoint was reduction

confirmed clinical upper GI events. There were 56

events on rofecoxib and 121 on naproxen. The time

to event curve separated early and they continued

to separate, and the relative risk of 0.46

corresponds to a 54 percent risk reduction with

rofecoxib. The p value, as you can see, was highly

significant. A similar GI benefit was seen with

confirmed complicated events, and in a post hoc

analysis for lower GI events.

A second finding in VIGOR was the

difference in the rates of thrombotic

cardiovascular events between the two treatment

groups. There was a relative risk of 2.4 for the

confirmed events, as shown here. The p value,

144

again, was highly significant.

Examination of the individual types of

events broken down by vascular bed, cardia,

cerebrovascular and peripheral shows that the

difference between treatment groups was largely

driven by the difference in myocardial infarction,

20 on rofecoxib and 4 on naproxen. Of note, there

were similar numbers of patients with strokes in

the two groups.

Additional exploratory analyses were

undertaken to better understand these

cardiovascular findings. I will focus on the types

of analyses that I will show later for APPROVe. In

VIGOR the use of 50 mg, a dose 2-4 times the

recommended approved chronic doses, was associated

with a higher incidence of hypertension adverse

experiences than with naproxen. In analyses

described in the background package the relative

risk of events was similar in patients with or

without increases in blood pressure during the

study. The relative risk of events was also

similar in patients with or without baseline risk

145

factors for cardiovascular risk.

Finally, multiple analyses were performed

to examine the patterns of risk and relative risk

over time, both by Merck and the FDA. Merck's

interpretation was that there was no significant

increase in relative risk over time for rofecoxib

versus naproxen. However, the FDA felt that a

change in relative risk over time could not be

excluded.

Because VIGOR did not have a placebo

control, we turned to other data from other studies

to better understand these results. Merck had

initiated a program to assess the ability of

rofecoxib to delay the onset of Alzheimer's disease

in patients with minimal cognitive impairment or to

slow the progression of Alzheimer's disease. In

these studies, rofecoxib 25 mg was compared to

placebo in an elderly population.

An initial review of the cardiovascular

data, in March, 2000 when the VIGOR results were

first learned, did not show an imbalance. In a

subsequent review, undertaken in September, 2000,

146

in advance of the VIGOR advisory committee, which I

will show you next, at that time there were over

2000 patients enrolled, with a median duration of

therapy of approximately one year.

The analyses at that time were based on

investigator-reported events since at that time few

had been adjudicated. Subsequent analyses that I

will show using the adjudicated data were

consistent with these initial analyses. Clearly,

there was no evidence to suggest an increased risk

with rofecoxib based on the aggregate endpoint

shown on this slide, or based on the analysis of

individual type of events such as myocardial

infarction or stroke shown in the background

package.

Consistent with the approach envisioned in

the adjudication SOP, we also performed a pooled

analysis of all the available cardiovascular data

to obtain more precise estimates of the relative

risk for rofecoxib versus each of the various

comparators. The pooled analyses include all

randomized, controlled trials from Phase IIb

147

through our Phase V postmarketing trials of 4 weeks

or longer duration that had been completed by

September, 2000 and also included the Alzheimer's

data that I just showed you.

Studies were included if there was a

placebo or an NSAID comparator. For the pooled

analysis we prespecified to use the anti-platelet

trial as collaboration combined endpoint of

myocardial infarction, stroke and vascular death.

There were several reasons for this choice. First,

the rofecoxib pooled analysis included data from

studies that antedated the adjudication SOP.

Investigator reports of the APTC endpoints had the

highest confirmation rates in the studies that were

adjudicated so restricting the analyses to these

events ensured consistency among the data. Second,

the APTC combined endpoint was a standard and would

allow comparison to other published reports. The

analysis pooled double-blind patient level data

stratified by disease. In September, 2000 there

were data from over 28,000 patients and over 14,000

patient-years of exposure.

In the analysis for the three data sets,

placebo, non-naproxen NSAIDs and naproxen

controlled data, a difference was only observed in

148

the naproxen data set. It was, therefore,

considered not appropriate to combine the three

data sets as this would tend to obscure the

difference from naproxen.

In our plots the triangle points to the

estimate of relative risk and the size of the

triangle is proportionate to the overall exposure.

The 95 percent confidence interval is shown as a

horizontal line, and the same information is

provided numerically along with the numbers of

events in each data set.

In the placebo and non-naproxen NSAID data

sets the data do not suggest an increased risk

standard rofecoxib. The data in the naproxen set

were largely driven by the VIGOR data and,

consistent with VIGOR, there was an increased risk

for rofecoxib compared to naproxen. The 95 percent

confidence interval did not cross 1, consistent

with the statistically significant difference.

Our conclusions: There was a clear

evidence for GI safety benefit of rofecoxib

compared to non-selective NSAIDs. Because the data

did not suggest increased risk of cardiovascular

events with rofecoxib compared to placebo or

non-naproxen NSAIDs, we believe that the weight of

149

the evidence was most consistent with naproxen

having provided a cardioprotective benefit in

VIGOR. Data to support that naproxen 1000 mg can

provide sustained anti-platelet effects, as well as

animal data with naproxen and clinical data on

agents with similar properties are all provided in

the background package. Subsequent data with other

selective COX-2 inhibitors would also show a

cardiovascular difference from naproxen while

having similar cardiovascular events with

non-naproxen NSAIDs.

The Arthritis Advisory Committee agreed

that the VIGOR study had shown a GI safety benefit

for rofecoxib compared to naproxen. With regard to

the cardiovascular data, they determined that the

results were inconclusive. They recommended that

150

both the GI and cardiovascular data be described in

the rofecoxib label. Those recommendations were,

indeed, reflected in the approved labeling. There

is now a modified GI warning acknowledging that the

risk of GI toxicity with rofecoxib 50 mg once daily

is significantly less than with naproxen 500 mg

twice daily.

There was a new cardiovascular precaution

which provided the cardiovascular results from

VIGOR and from the Alzheimer's disease studies

which concluded that the clinical significance of

the cardiovascular findings were unknown. The

specific precaution stated that caution should be

exercised when Vioxx is used in patients with a

medical history of ischemic heart disease.

Finally, because there were dose-related

trends and NSAID type adverse experiences with

rofecoxib 50 mg and no greater efficacy at 50 mg

compared to 25 mg, the new label further emphasized

that the chronic use of rofecoxib 50 mg was not

recommended.

I would like to turn now to the period

151

starting after we learned the results of VIGOR up

to the unblinding of APPROVe, and I will focus on

the unique information that Merck can provide to

this committee, information on our approach to the

design of a study of cardiovascular outcomes that

we implemented in 2002, and the final data from our

programs in arthritis and Alzheimer's disease that

were completed in this time frame. I will briefly

touch on data that others will be presenting or

have presented, such as epidemiology studies and

the ongoing preclinical work, and will end this

section of the presentation with our assessment of

the data available before APPROVe.

In considering outcome study designs, we

recognized two different approaches we could take.

Each had different merits and would answer

different questions. The first would be to perform

an NSAID-controlled study. This could involve

arthritis patients so we could study the patients

in whom the drug was indicated, knowing, however,

that a placebo control would not be appropriate in

a several-year study of patients who require

152

chronic NSAIDs, and the use of chronic NSAIDs over

several years was not appropriate in patients who

did not have that need.

The alternative was to do a study versus

placebo. Obviously, this would preclude the

ability to study patients with chronic arthritis.

So, the applicability of the finding to arthritis

patients would need to be inferred. Despite this

potential limitation, we decided for rofecoxib to

answer the question for difference from placebo.

I think it would be useful to discuss with

this committee how bit these studies need to be.

As we all know, it is easier to prove a difference

than to prove similarity. In order to exclude even

a 30 percent increased risk with 95 percent

confidence and with 90 percent power, you need data

on over 600 confirmed events. Based on anticipated

event rates and typical dropout rates on our

studies, this would require enrolling approximately

25,000 patients for a study design to run over

about 3 years. To exclude a 20 percent increased

risk you would need approximately 1300 events and

153

over 60,000 patients. To exclude a 10 percent risk

you would need approximately 4800 events and over

200,000 patients in the studies.

We considered several placebo-controlled

designs. One study in acute coronary syndrome was

rejected for a variety of reasons after extensive

discussions with our consultants. First, these

unstable patients are at particular risk for bad

outcomes associated with GI or renovascular effects

known to be present with rofecoxib, and considering

the unknown benefit this raised concerns.

Second, these patients would all need to

be taking aspirin and, as you recall, one of the

hypotheses at the time, and it still continues to

be a hypothesis, was that aspirin would abrogate

any increased cardiovascular risk of selective

COX-2 inhibition and, thus, a negative finding

would not have answered the question raised by

VIGOR.

However, the emerging data on possible

chemopreventative benefits of COX-2 inhibitors and

the extending database that we had of

154

chemoprevention studies with rofecoxib versus

placebo provided an alternative means to address

this question. In addition, these patients present

a broad spectrum of cardiovascular risk similar to

the arthritis patients in whom rofecoxib was being

used. Thus, it was decided to develop a study of

cardiovascular outcomes for rofecoxib based on a

combined analysis of placebo-controlled

chemoprevention studies.

The APPROVe study comparing rofecoxib 25

mg to placebo had already been initiated during

2000 and a second study, also comparing rofecoxib

to placebo, was initiated in 2002, VICTOR, a study

to assess reduction in colon cancer mortality. A

third study examining the ability of rofecoxib to

prevent prostate cancer in men at risk, the ViP

study, was initiated in 2003. Together, these

three studies would provide information on

thrombotic cardiovascular events in over 25,000

patients and targeted to enroll 20-30 percent of

patients on aspirin. The combined analysis had its

own protocol analysis plan and an external safety

155

monitoring board to monitor the cardiovascular

safety in the three combined studies.

The protocol for the combined outcome

study was finalized in October of 2002 and was

submitted to and discussed with the FDA and with

the regulatory agency in the United Kingdom.

Also during the 2000-2004 time frame final

data became available from our programs in

arthritis and Alzheimer's disease. As the data

became available, we performed updates to our

cardiovascular pooled analysis and, in 2003,

performed a final cardiovascular update. Also, in

2003 we updated our pooled analysis of upper GI

clinical events so I will show you now the final GI

and cardiovascular data from these programs.

Final GI data from the osteoarthritis and

rheumatoid arthritis programs were analyzed in

pooled analysis of clinical upper GI events using

the same approach to the data as in the initial

analysis I showed before, except now we had data

that extend up to 30 months of treatment. The

pooled analysis included all Phase IIb through

156

Phase V randomized clinical trials 4 weeks or

longer and excluded VIGOR as those data would

otherwise overwhelm the data in the pooled

analysis, and that is shown separately on this

slide.

As you can see, even excluding VIGOR, the

relative risk of a confirmed clinical upper GI

event for rofecoxib compared to the combined NSAIDs

was 0.36, a 64 percent reduction, and a similar

benefit could also be demonstrated for confirmed

complicated events.

In this final pooled analysis there was

sufficient data to assess whether the findings for

the combined NSAID groups were consistently

observed for each of the comparator NSAIDs,

diclofenac, ibuprofen and naproxen and, as you can

see, this was clearly the case.

I will turn now to the cardiovascular

data. This is the Kaplan-Meier plot of the final

data for the osteoarthritis Phase IIb/Phase III

studies for rofecoxib compared to the non-naproxen

NSAIDs. Over 30 months the curves are

157

indistinguishable from each other, although

starting around 18 months, as you can see, the

numbers of patients begin to drop off and the 95

percent confidence intervals begin to widen

consistent with the data becoming sparse.

This is the time to event plot for the

final cardiovascular data. For the Alzheimer's

disease studies, these are the confirmed events

from these studies. The average relative risk

across the Alzheimer's program was very close to 1.

However, in this data set there was a statistically

significant non-constant relative risk, with an

apparent decreased incidence for rofecoxib compared

to placebo for the first approximately 24 months of

the study and an apparent increased risk for

rofecoxib thereafter. however, as the overall

relative risk approximated 1 and as data in our

pooled analysis did not suggest this pattern of

changing relative risk in any of the data sets, the

data from Alzheimer's were interpreted to represent

variation about a mean and no difference between

the treatment groups.

I want to point out that there were 90

patients with confirmed cardiovascular thrombotic

events in the Alzheimer's disease data and there

158

have been over 70 in the osteoarthritis data set.

Thus, each of these data sets was large enough to

exclude the 2-fold increased cardiovascular risk

with rofecoxib that we had seen in VIGOR.

This is the final update to the pooled

analysis. The pooled analysis included data now

from 28 studies in over 32,000 patients and over

19,000 patient-years of exposure. Again, relative

risk for rofecoxib compared to placebo and

rofecoxib compared to non-naproxen NSAIDs

approximated 1. However, the relative risk

compared to naproxen continued to show a difference

with a 95 percent confidence interval excluding 1

and, thus, indicating statistical significance.

So, what was our assessment of the data in

2004 before we learned the results of APPROVe? The

data available in 2004 came from three sources,

observational epidemiology studies, preclinical

studies and randomized controlled trials. There

159

were 10 observational epidemiology studies, either

published or publicly presented, on the

cardiovascular risk with these drugs and an

increasing literature on preclinical models. These

are described in detail in the background package

and I will not go into these data as others will be

speaking to them.

With regard to these other studies, I will

just observe that the results were mixed and they

did not provide clarity on the cardiovascular risk

with rofecoxib or selective COX-2 inhibition. We

believe that clarity would best come from the

outcome study that we had initiated.

Also in this same time frame the TARGET

study results with lumiracoxib were published.

These were consistent with the pattern of overall

cardiovascular findings that with had observed with

rofecoxib, with cardiovascular event rates similar

to a non-naproxen NSAID, in that case ibuprofen,

but a cardiovascular event rate higher with

lumiracoxib than with naproxen. With rofecoxib we

had also observed similarity to placebo in the

160

Alzheimer studies. Thus, in assessing these

different data we place the greatest emphasis on

data from randomized clinical studies and, based on

these, the assessment was that the risk/benefit

profile remained favorable for rofecoxib. With

regard to any remaining questions our ongoing study

of cardiovascular outcomes would provide the

answers.

APPROVe was the first component of the

study on cardiovascular outcomes. It was

anticipated to complete in November of 2004.

However, on September 23 we received a call from

the administrative committee that they had accepted

a recommendation from the external safety

monitoring board to terminate treatment in the

study.

APPROVe studied rofecoxib 25 mg versus

placebo in approximately 2600 patients.

Stratification was by baseline aspirin use because

aspirin had been shown in previous studies to

reduce the incidence of colon polyps. There was a

3-year on drug treatment period and 1-year off-drug

161

period. Colonoscopies were performed at screening,

year 1, year 3 and there was a year 4 follow-up

after withdrawal of therapy to assess the

possibility of rebound. The primary endpoint was

the cumulative incidence of patients with

adenomatous polyps at year 3. The first patient

was screened in December of '99 and the first

patient was randomized in February, 2000.

Patients had to be 40 years or older and

have a histologically confirmed large bowel adenoma

at screening. Patients with a prior history of

thrombotic cardiac events could be enrolled if they

were more than a year post event; 2 years for a

cerebrovascular event. Patients were excluded if

they were medically unstable, for example, if they

had uncontrolled hypertension or angina or CHF at

rest.

The data that led the ESMB to terminate

the study early are the data on this slide. These

are the preliminary data from the ESMB September

meeting. In the final data, which are now

published on-line, there were two additional

162

events, one myocardial infarction in each treatment

group so the current curves look very similar.

Overall, there was an approximately two-fold

increase in risk with rofecoxib compared to

placebo. However, there was a statistically

significant change in relative risk over time.

Event rates were similar to placebo over the first

approximately 18 months, consistent with our

previous data. Starting after 18 months of

treatment the curves began to separate with the

difference becoming significant.

Looking at the types of events, you can

see that there were imbalances in myocardial

infarction, 20 versus 8 here or, in the final

numbers 21 versus 9, and imbalances in stroke, 11

versus 6. In addition to these findings, we also

observed differences from placebo in NSAID-like

renovascular effects, for example, edema,

congestive heart failure and hypertension.

After APPROVe our assessment of the risk

of cardiovascular thrombotic events with rofecoxib

had changed. APPROVe was the first study to show a

163

statistically significant increased risk of

cardiovascular thrombotic events with rofecoxib 25

mg versus placebo. Although the risk had been

similar to placebo for the first approximately 18

months, the risk in APPROVe began to diverge from

placebo starting after approximately 18 months.

The mechanism for this finding at that

time was uncertain. At the time, available

clinical data on other agents did not support a

class effect so we were left with a potentially

molecule-specific effect. As I previously

indicated, the administrative committee indicted

its recommendation to terminate study treatment to

us on September 23 and, on the basis of the data

Merck voluntarily withdrew Vioxx from the market on

September 30th.

APPROVe was the first clinical trial with

rofecoxib that showed an increased cardiovascular

risk versus placebo. At the time alternative

therapies were available without evidence of a

similar cardiovascular risk and, thus, Merck

believed that voluntary withdrawal best served the

164

interests of patients.

Since withdrawal of Vioxx we assiduously

worked to obtain the final data from APPROVe and

preliminary data from the other placebo-controlled

chemoprevention studies, VICTOR, the colon cancer

study, and ViP, the prostate cancer study. I will

start with the final analyses of the APPROVe data

and additional exploratory analyses that we

performed to identify possible relationships

between various risk factors with increased

relative risk.

I want to start by pointing out, however,

that we performed numerous post hot exploratory

analyses of the data to identify factors that might

predict patients with increased relative risk. We

looked at well over 10 different baseline risk

factors. We looked in multiple different analyses

and we also examined patients who were not taking

aspirin. We also examined over 40 analyses of

blood pressure. We analyzed these by one subgroup

factor at a time with tests for

treatment-by-subgroup interaction.

Given the large number of subgroups tested

and the post hoc nature, the data that I am about

to show you need to be regarded as hypothesis

165

generating and not definitive. So, let me start

with the analyses of risk factors other than blood

pressure.

This slide shows the relative risk for

rofecoxib versus placebo for different

cardiovascular risk factors. To conserve time, I

am only showing the few in which possible trends

were seen. Patients with what we called increased

risk are patients with two or more baseline

cardiovascular risk factors, or a history of

symptomatic atherosclerotic cardiovascular disease;

aspirin users in the study which we defined as

patients who used aspirin at least 50 percent of

the time on study and before an event; patients

with diabetes; and patients with a history of

atherosclerotic cardiovascular disease. However,

these four subgroups were not independent. The

events in the patients with a history of

atherosclerotic cardiovascular disease were also

166

included in the aspirin user and in the increased

risk subgroups and, in fact, were driving the

differences in these subgroups. So, what we have

are potentially two independent risk factors,

patients with a history of atherosclerotic

cardiovascular disease and patients with a history

of diabetes. For these two subgroups, the test for

treatment-by-subgroup interaction was borderline,

with a p value between 0.05 and 0.1. At this time

these observations can only be regarded, as I said,

as hypothesis generating.

We also looked at blood pressure in

APPROVe. Blood pressure was measured in this study

once per visit which occurred at 4-month intervals.

The blood pressure measurements, however, were not

standardized across sites for example with respect

to time of day or measurement technique. And,

blood pressure changes in APPROVe were typical of

what had been published for NSAIDs, between group

differences and the change from baseline and

systolic blood pressure of about 4 mm Hg systolic

and for diastolic about mm Hg. Baseline mean

167

systolic and diastolic blood pressure data from

population studies or from studies on the

cardiovascular effects of lowering blood pressure,

the change in mean systolic and diastolic blood

pressure we observed in APPROVe would not appear to

account for the magnitude of the cardiovascular

findings that we have observed. Nonetheless, we

performed numerous analyses to assess whether

associations could be identified between the blood

pressure and cardiovascular data.

Multiple blood pressure analyses are

described in your background package. Neither the

preliminary nor the final analyses identified

consistent patterns or consistent patient subgroups

or covariates associated with increased relative

risk. Variables assessed included baseline blood

pressure, change from blood pressure, on treatment

blood pressure and hypertension reported as an

adverse experience. The one subgroup of the many

we tested in which a trend was identified was in

patients with systolic blood pressure greater than

or equal to 160. However, other data sets, in

168

particular VIGOR and our placebo-controlled data

from the pooled analysis, did not show a similar

trend when assessed in this manner.

With the final data we also learned the

results of the efficacy endpoint. The primary

efficacy endpoint was the cumulative incidence of

patients with recurrent colon polyps over the

3-year treatment period. The primary approach to

the data was intention-to-treat, and the primary

population was patients at increased risk for

colorectal cancer based on baseline risk factors

such as histology and number of polyps. Rofecoxib

use was associated with a 24 percent reduction in

the risk of colon polyp recurrence, and the p value

was highly significant.

As I indicated earlier, the study of

cardiovascular outcomes was the pooled data from

APPROVe, ViP and Victor. We have preliminary data

from ViP and VICTOR and wanted to share those

preliminary data with you as well.

This slide shows a pooled analysis for the

primary endpoint that we had prespecified for the

169

cardiovascular outcome study confirmed thrombotic

cardiovascular events. Again, I want to emphasize

that VICTOR and ViP data are still preliminary.

There are still five cases that are pending

adjudication to which we remain blinded. For

VICTOR we have very limited information on overall

exposure and on patient demographics.

The study was conducted by Oxford and they

are working hard at getting the information to us.

Given the preliminary nature of the ViP and VICTOR

data, we are unable to draw at this time definitive

conclusions from these data.

Also with the data available, we can

provide a comprehensive perspective on mortality in

the rofecoxib clinical program. Shown is all-cause

mortality. This is a bit busy so let me orient

you. Rofecoxib is shown in yellow; NSAID

comparators are shown in blue; and placebo is shown

in white. The figure provides mortality rates per

100 patient-years and 95 percent confidence

intervals.

Compared to the NSAIDs, overall mortality

170

rates were similar for rofecoxib. In one instance,

the osteoarthritis Phase IIb/III studies, there

were significantly fewer deaths on rofecoxib than

the comparator but this was not reproduced in other

data sets. With respect to placebo, mortality

rates were similar between rofecoxib and placebo in

all the data sets except the Alzheimer's disease

where there was a significantly higher rate on

rofecoxib and the difference was statistically

significant.

We looked at this carefully. Although

some of the imbalance was due to a difference in

mortality due to thrombotic cardiovascular events,

the larger part of the difference was due to

trauma, poisoning and infections, causes that one

would not expect to be associated with an NSAID

type drug effect. So, we don't have an explanation

for this observation in the Alzheimer studies.

What do we believe the implications of

these data to be? As I alluded to earlier, we

believe that there are several public health

questions raised by the new data. The first is the

171

risk/benefit for selective COX-2 inhibitors

relative to standard of care in their established

indications. Rofecoxib has a GI safety profile

superior to ibuprofen, diclofenac and naproxen.

The cardiovascular profile is more

complex. Although there have been no differences

observed between rofecoxib, ibuprofen or

diclofenac, based on what we learned from APPROVe,

the type of long-term data needed to establish

similarity to these agents does not exist and at

the time we withdrew data for a class effect of

COX-2 inhibition was limited.

Amongst the non-selective agents, naproxen

100 mg appears to have the lowest risk of

thrombotic cardiovascular events, but also the

highest risk of upper GI clinical events. Can we

identify risk factors associated with increased

risk for thrombotic cardiovascular events with

these drugs? I have shown you our data to support

the effect of duration in our exploratory analyses

on patient demographics. More work needs to be

done to investigate the hypotheses raised by our

172

data. With regard to dose, our data cannot

definitively address this.

Finally, is the increased cardiovascular

risk that we observed with rofecoxib a class effect

of COX-2 inhibition? We believe that the data that

have been reported on celecoxib from the APTC study

and on valdecoxib and from the CABG study, together

with the APPROVe findings, strongly suggest an

effect of COX-2 inhibition on increasing

cardiovascular risk.

If the committee agrees that this is a

class effect, the next critical question will be

determining the size of the class. Traditional

NSAIDs such as ibuprofen and diclofenac have not

shown a different cardiovascular risk profile from

the selective COX-2 agents. However, those data

are limited beyond one year. We would argue that

long-term comparative studies of these agents are

needed to better assess the relative cardiovascular

risk.

Well, what do we think are next steps? At

Merck, we are continuing to analyze our clinical

173

data and we will be analyzing, for example, frozen

samples from patients to try to identify markers

that correlate with an increased relative risk of

cardiovascular events with COX-2 inhibition. In

addition, patients in APPROVe are being followed

off-drug as had been prespecified in the protocol,

and we are in the process of meeting with

consultants to further explore scientific

hypotheses for the findings.

We are also aware of efforts that are

under way to analyze data across the different

drugs, and we support those efforts.

Finally, comparative outcome studies, we

believe, are needed to determine the relative risk

amongst these agents in relevant populations. Dr.

Curtis will talk to you tomorrow about one such

study that we are conducting at Merck, the MEDAL

study. This is the largest study in arthritis

patients ever conducted and compares the long-term

safety of etoricoxib with that of diclofenac, the

most widely used traditional NSAID worldwide. We

believe MEDAL will provide the kind of information

174

needed to weigh the risk/benefits of these drugs

and improve the ability of physicians to make

recommendations for arthritis pain and treatment

that is in the best interest of their patients.

Thank you, Dr. Chairman, members of the committee,

FDA. I am available for your and the committee's

questions.

DR. WOOD: Great! Thanks very much. As I

am sure you would agree, the primary job of this

committee is to assess all the risks and benefits

that these drugs can produce, and we have certainly

been encouraged to do that by everybody who has

spoken so far.

That being the case, I was very surprised

not to see the Kaplan-Meier curve for pulmonary

edema. can you show us that from the APPROVe

study?

DR. BRAUNSTEIN: Certainly. That would be

slide 213.

DR. NISSEN: Yes, heart failure and

pulmonary edema would be helpful.

DR. BRAUNSTEIN: We certainly examined

175

that. You know, the question that has been on the

table--we believe the hypothesis we were exploring

was the incidence of thrombotic cardiovascular

events. Pulmonary edema is a mechanism-based

effect of selective COX-2 inhibition that has been

well appreciated and, in fact, is already described

in product labeling.

So, we did see an effect. This is in our

publication. As shown here, we saw an effect.

This is a combined endpoint of congestive heart

failure, pulmonary edema of cardia failure, so all

congestive heart failure type of events that we

observed in the study.

DR. WOOD: And this had a hazard ratio of

4.6 and a p value of less than 0.004. Right?

DR. BRAUNSTEIN: Yes.

DR. WOOD: So, I mean, it is important for

the committee--and this goes to all the speakers I

think, that if there are other hazards with a

hazard ratio of 4.6, that we see these as they are

presented so that we can make some cumulative

estimates of what the hazards are for these drugs.

176

Just because they are in the label does not mean we

shouldn't hear about them here, it seems to me.

The second question I have, which has

always worried me, is when you go back to the

original label change that you made, you know, when

you changed the label to say caution should be

exercised when Vioxx is used in patients with a

medical history of ischemic heart disease, as a

physician what am I supposed to do with that? Am I

supposed to say to patients take the drug slowly,

or swallow it with milk, or only take it with the

lights on? Tell me what I am supposed to do with

that information. I am not being facetious here

because as we go through this process we are going

to have to decide how we make whatever labeling

changes we make, if that is the decision we make,

and that doesn't seem to me to have been helpful.

But maybe you knew something that I didn't. So,

what did you intend me to do with that information?

DR. BRAUNSTEIN: At the time when we

conducted negotiations with discussions with FDA on

that labeling, there were no specific data that

177

showed a statistically significant increased risk

in one patient group or another. However, given

the uncertainty in the data, it was felt to be

prudent to recommend that caution be exercised in

that patient group if you are considering using the

drug.

What we meant by that was that you need to

carefully weigh the risks and benefits of the

different treatment options. We think that when

evaluating the options on patients it needs to be

done on an individual patient case-by-case basis.

Patients differ with respect to their

cardiovascular risks, with respect to their GI

risks, with respect to their history of allergies

and with respect to how they responded to these

different medications in the past, and all of that

information needs to be taken into consideration

when assessing and determining what type of therapy

should be used versus another. And, we felt that

one of the things that should be considered was

cardiovascular history, and that is what we meant

by that.

DR. WOOD: Okay. Other questions?

Stephanie?

DR. CRAWFORD: Thank you. I appreciate

178

the presentation. I heard both the speakers say

that the sponsor, Merck in this case, made the

decision to voluntarily withdraw rofecoxib in the

interest of public health although the drug could

have been continued on the market. When we look at

adverse events we desire to predict uncontrollable

events and control controllable events. The bottom

line question which is really important to me as we

consider these issues when we look in this case at

the issue of hazard of cardiovascular events is how

much is too much? In other words, how did the

sponsor come to the conclusion that the evidence

was so compelling as to take the step of

voluntarily removing the drug product from the

market?

DR. BRAUNSTEIN: Well, at the time when we

saw the increased risk compared to placebo there

were not data to allow us to conclude that this

could be a class effect, and we felt that there

179

were other options available to patients, including

therapies that adverse event not known to have this

increased cardiovascular risk. So, given those

options and alternatives, we felt that the

responsible action at the time was to withdraw

Vioxx.

DR. CRAWFORD: Excuse me, but I am asking

specifically what was that signal that was at the

level where, in the interest of caution or whatever

the mechanism was, you said this level is

unacceptable at this time based on the given

evidence?

DR. BRAUNSTEIN: Well, we saw overall a

two-fold increased risk and that was seen versus

placebo so it was something that we knew was

statistically significant. The magnitude of the

risk was on the order I think of one or two

percentage points, but still at the time the other

agents--it was a determination that amongst the

choices that patients had available to them there

were other agents that were not known to have this

risk and, given the ability for patients to have

180

alternatives that they could discuss with their

physicians, we felt that we should withdraw Vioxx

at that time.

DR. WOOD: Dr. Shafer?

DR. SHAFER: Two questions. I will make

them fast. Do any studies show improved analgesia

on Vioxx?

DR. BRAUNSTEIN: No. I mean, all of our

efficacy studies show very similar results at

comparable doses to NSAIDs.

DR. SHAFER: Okay. The other thing is can

you go to slide number 36?

DR. BRAUNSTEIN: Yes?

DR. SHAFER: I just can't help but notice,

but the upper bounds of the confidence intervals

for the first two groups encompass the mean of the

naproxen comparison. Does that give you pause in

justifying excluding naproxen as a separate

comparison group? If you take a look at the upper

bounds, they include the mean of naproxen which

might suggest that statistically those groups

really shouldn't be segregated as you have done.

DR. BRAUNSTEIN: Well, when you look at

this, if you were to combine all the data one would

not see a statistically significant difference. It

181

would tend to obscure the naproxen finding, and we

felt, given what we observed in VIGOR and what we

had observed all along the program, that that

wasn't the right way to go, especially given the

difference pharmacologically. I mean, in terms of

looking at the data we also were taking into

context what we understood about the pharmacology

of these agents and the ability for naproxen to

provide that kind of inhibition of COX-1 that Dr.

FitzGerald talked about. So, we thought that not

only were there differences in the clinical data

but there were differences in the pharmacology data

that supported keeping naproxen separate.

DR. WOOD: Dr. Gibofsky?

DR. GIBOFSKY: One of the stratifications

we are asked to do during the next three days is,

of course, the risk/benefit relationship. I am

wondering if you have calculated the risk/benefit

of cardiovascular thrombosis outcome versus the

182

benefit of cancer prevention in the population. I

can understand where the relative risk of 1.92 is.

I understand what it means when the relative risk

goes up above a certain number above 1.0 but, you

know, you can't go much below 1.0. So, have you

calculated to what extent your risk of

cardiovascular events is related to your protection

against cancer?

DR. BRAUNSTEIN: Well, we didn't actually

study cancer as an outcome. We were looking at

polyps which are precancerous lesions.

DR. GIBOFSKY: The same question

basically.

DR. BRAUNSTEIN: Well, even there, you

know, polyps are easily--there is a different

mechanism. There is an alternative therapy

available for the treatment of polyps. So, in

order to evaluate risks and benefits one has to

compare the risks and benefits of one treatment

option versus the risks and benefits of another

treatment option. In doing so, I think that this

wouldn't--

DR. GIBOFSKY: Well, let me ask it another

way then, if you did not see a cardiovascular

signal in APPROVe would you have concluded that the

183

reduction in risk in polyp formation was

efficacious?

DR. BRAUNSTEIN: We concluded that the

reduction in risk in polyp formation was

efficacious regardless of the cardiovascular

finding. Are you asking whether the overall

risk/benefit would have been favorable???

DR. GIBOFSKY: Yes.

DR. BRAUNSTEIN: That would be speculative

for me. We haven't looked at the data with that

specific question in mind. I think we would need

to take a look at all the patients that we looked

at in all the different subgroups to see if that

remained the case. You know, you saw some

congestive heart failure. We say NSAID type

typical effects that one would see in one of these

studies, not just cardiovascular risk but there was

a small increase in ulcers, not as much as one

would anticipate to see with a non-selective NSAID

184

but still present. There was a small increase in

other NSAID type effects like edema and

hypersensitivity. So, we haven't made a formal

risk/benefit assessment.

DR. GIBOFSKY: Just one last point, you

stressed the concept of their being other modes of

therapy available and so that factored into your

decision to take this agent off the market. But

there are other ways of treating polyps as well,

which leads me to question in that context the

rationale for the APPROVe study.

DR. BRAUNSTEIN: We thought this was an

interesting and important scientific question that

had been raised in the literature.

DR. WOOD: That sounds like a

retrospective question so I will let you off the

hook. Let's move on. Ralph?

DR. D'AGOSTINO: Two quick questions. In

slide 48 you, I think quite sensibly and again post

hoc, split out the cardiovascular risk and redid

the analysis. Now, if this were preplanned and I

got a result like that I would say that this is

185

great; this shows me that placebo is better no

matter what I do. I mean, the cardiovascular does

increase a bit but the placebo is still maintaining

itself even in individuals without cardiovascular

risk.

DR. BRAUNSTEIN: This slide shows the

relative risks in each of these groups. It is not

placebo and then rofecoxib.

DR. D'AGOSTINO: Well, it is all against

placebo.

DR. BRAUNSTEIN: It is all compared to

placebo, yes.

DR. D'AGOSTINO: Right, and placebo wins

everywhere. So, no matter if you have

cardiovascular risks or not, still placebo was

better. Am I misinterpreting this?

DR. BRAUNSTEIN: You know, in this we only

see trends for some subgroups and in others we

don't identify particular subgroup factors where

there is an important difference.

DR. D'AGOSTINO: Well, that is a subgroup

and it sort of indicates consistency to me. In

186

slide 42 there was consistency regardless of CV

risk. In slide 42, if I look at those numbers on

the bottom, I presume those are individuals

available. You are dropping about 100 individuals

after 12 months or so. Do we know anything about

the loss to follow-up on these individuals?

DR. BRAUNSTEIN: We did not see

differences, for example, in changes in

cardiovascular risk associated with patients who

discontinued--

DR. WOOD: Wait a minute, these are not

all patients who dropped out, are they?

DR. BRAUNSTEIN: These are all the

patients who remained in the study.

DR. WOOD: So, some of these patients may

not have advanced to the end of the study.

DR. D'AGOSTINO: Well, if you start at the

beginning--that is my question, I mean it is

randomized, right? So, there must have been about

a 50-50 break so you would think at each point you

would have approximately the same numbers in the

two groups.

DR. BRAUNSTEIN: Well, there is a

differential dropout due to adverse experiences for

example that one would normally see in an NSAID

187

trial against placebo.

DR. D'AGOSTINO: Well, why couldn't they

be followed for CV events? Why wasn't it like an

intent-to-treat analysis or something?

DR. BRAUNSTEIN: Yes, the way we had

prespecified the analysis was that all events were

determined up to 14 days after discontinuing

therapy. The only intention-to-treat analysis was

one done for mortality overall.

DR. WOOD: Dr. Nissen?

DR. NISSEN: Yes, I think Ralph's point is

very, very important. We need to see an

intent-to-treat analysis. You are telling me that

14 days after they dropped out of the study these

folks were not followed beyond that?

DR. BRAUNSTEIN: We are following patients

who are off-drug, who terminated treatment in the

study, and we don't have data yet on that.

DR. NISSEN: Because there are a lot more

188

people dropping out of the rofecoxib arm and the

question is why are they dropping out and what

happened to them. The signal here could be a lot

stronger than we see using this somewhat selective

analysis. I am used to an intent-to-treat

analysis, Ralph, for a trial like this and I am

confused as to why it was done in this way. You

know, a cardiovascular hazard, if this is a

pro-atherogenic therapy, is going to persist quite

a while after you stop the drug. So, I think we

really do need to see--I mean, to clear the air

here we have to see that intent-to-treat analysis.

I would track those people down and find out what

happened to them.

As a cardiologist, I obviously use a lot

of low dose aspirin so I am very familiar with the

low dose aspirin literature, and we see in low dose

aspirin perhaps up to a 20 percent reduction in

cardiovascular risk in individuals who are at risk.

So, what I am really confused about is that you

attributed what you found in VIGOR to the

beneficial effects of naproxen, but you are talking

189

about a 4- or 5-fold difference in myocardial

infarction rates and I just want to know how you

came to the conclusion that that amount of

difference could be explained by naproxen.

Naproxen would have to be a lot more effective than

aspirin. We know aspirin inhibits platelets as

well as anything else out there. So, how did you

guys arrive at that conclusion that it was naproxen

related?

DR. BRAUNSTEIN: Well, other than in

addition to the data that support that naproxen can

have this effect, and specifically with regard to

the magnitude that you are pointing out in

myocardial infarction, there were only 24 events in

VIGOR. The cardiovascular outcome studies that you

are referring to oftentimes have hundreds, if not

thousands, of events that they are assessing and

that allows one to very carefully and with

precision identify what the relative risk reduction

is. In VIGOR we had fairly wide confidence

intervals and, in addition, VIGOR studied

exclusively patients with rheumatoid arthritis.

190

These are patients with chronic inflammatory

disease, elevated C-reactive protein and in those

patients we know that the effect of aspirin is also

magnified. So, given those factors, we felt that

it was certainly compatible with an aspirin-like

effect.

DR. NISSEN: Again, I am not sure I buy

that. You know, post-MI patients have a very

elevated risk and the most we ever expect from

aspirin is perhaps a 20 percent reduction in

recurrent events. Even with dual platelet

antagonism with aspirin and clopidagrel we don't

get a whole lot more than that. So, this story

about naproxen, as I think Garret FitzGerald apply

discussed--it doesn't stand the test of any kind of

scientific rigor.

I guess the other question I wanted to

challenge you on is this comment that you made that

the blood pressure effects in APPROVe were

consistent with what is seen in other NSAIDs. I

hope many of you have had a chance to look at the

Archives manuscript that compares a meta-analysis

191

of blood pressure effects. It sure looks like

rofecoxib is an outlier here, showing a weighted

mean difference of about 5.5 mm Hg or almost 6 mm

Hg compared to NSAIDs which are substantially

smaller. Is it your position that rofecoxib does

not produce greater degrees of hypertension than

comparable NSAIDs?

DR. BRAUNSTEIN: Most of the studies that

are referenced in that analysis, unfortunately, are

confounded by dose. We think it is very important

when one looks at a pharmacologically mediated

effect, especially one that is known to have a

dose-dependent association, that the drugs be

assayed at doses that provide pharmacologically

equivalent degrees of inhibition of COX-2. For

example, for rofecoxib and celecoxib that would be

25 mg of rofecoxib and 200 mg twice a day of

celecoxib.

DR. NISSEN: Okay. I want to clear the

air on one more thing and, obviously, this drug has

been the subject of a great deal of public

attention and I think it would be a great

192

opportunity for you to explain, from your

perspective, why did it take 14 months, from

February of 2001 to April of 2002, for the label to

change? Were you fighting the FDA? Was there a

big battle over what the wording ought to be of the

label? I mean, it seems like 14 months is an

awfully long time after an advisory committee

meeting that recommended a warning to take for

agreement to be reached about what that warning

ought to say.

DR. BRAUNSTEIN: The advisory committee--

DR. WOOD: I think that is something

probably we should let him pass on--unless you want

to; go ahead.

DR. BRAUNSTEIN: No, no, no.

DR. WOOD: Go ahead.

DR. BRAUNSTEIN: After the advisory

committee there were a lot of discussions with FDA.

There were data requests from them which we

provided to them. We submitted at that same time

the NDA supplement for rheumatoid arthritis because

we felt it was important. As you know, VIGOR had

193

been conducted in rheumatoid arthritis patients at

50 mg and it was important to communicate to

physicians that the appropriate dose in those

patients was 25 mg. So, there was a lot of

information for the FDA to review. They also asked

for updated analyses of all our safety data. So,

they had a lot of work cut out ahead of them, and

we worked diligently with them to provide the

information, conduct the analyses that they

requested, and collaborated in that way to make

sure they had that information, and then we worked

assiduously to conclude a label. So, I don't

think, considering the wealth of information, that

the time frame is unusual.

DR. WOOD: And after 14 months, it was

"take the tablets slowly."

DR. BRAUNSTEIN: Well, after 14 months the

advice was that cardiovascular risk factors,

cardiovascular history should be taken into

account--

DR. WOOD: Well, that is not what it said.

It is most important to remember it didn't say you

194

shouldn't give it to people with cardiovascular

risk factors. It didn't say it shouldn't be given

to people who had had an MI or any other expletive

statement like that. It said caution should be

exercised in patients with history of heart

disease. That is quite different.

DR. BRAUNSTEIN: What I tried to say or at

least what I was trying to communicate was that the

risk/benefit assessment we felt needs to be done on

a patient by patient basis and, in addition to

taking GI risk into account, one should also take

cardiovascular risk into account given the

uncertainty of the data that was available at that

time and, as the label said, the clinical

significance of these cardiovascular findings were

unknown and that, therefore, the cardiovascular

information should be taken into account when

considering the use of rofecoxib.

DR. WOOD: Dr. Hennekens?

DR. HENNEKENS: I would be interested in

knowing the total number of deaths in the

randomized trials of rofecoxib against all other

195

comparators and then against placebo, non-naproxen

NSAIDs and naproxen.

DR. BRAUNSTEIN: You have that on your

slide. The numbers of deaths are underneath the

rows. I don't have the numbers at the top of my

head. We would have to do a quick tally. Also,

the only problem with looking at the numbers is

that the numbers themselves don't take into account

imbalances in exposure, which is why we showed them

as rates per 100 patient-years because it certainly

takes into account the differences in exposures.

Compared to the NSAIDs we did not see differences

in the rates, and compared to placebo we did not

see differences except, as I pointed out, in the

Alzheimer's disease study where there was a

statistically significant higher rate with

rofecoxib.

DR. WOOD: Dr. Cannon?

DR. CANNON: You mentioned in the VIGOR

and APPROVe clinical trials that the major driver

for the increased cardiovascular events on

rofecoxib was acute myocardial infarction. My

196

question is were these myocardial infarctions

apparently random events or was there any setting

in which they seemed to occur more frequently? For

example, in relationship to a procedure, including

a coronary interventional procedure, or surgery, or

were the myocardial infarctions random events? I

am thinking in terms of Dr. FitzGerald's

presentation and the recent valdecoxib experience

with bypass surgery.

DR. BRAUNSTEIN: We haven't identified any

kind of associations such as you are asking. But I

am not sure that we have specifically looked at the

question the way you are asking. So, I am not 100

percent sure.

DR. WOOD: Dr. Abramson?

DR. ABRAMSON: Yes, I guess one of the

surprises or unexpected findings in APPROVe was

that it took 18 months for these curves to separate

with rofecoxib. I was unaware of the heart failure

and pulmonary edema data until this morning. Often

fluid retention occurs early in the course of

putting people on NSAIDs. So, I am wondering could

197

you tell us more about when those heart failures

occurred over the course of time. Were they early

events, or was this also something that took some

time to appear in the population?

DR. BRAUNSTEIN: As one would expect from

an NSAID, fluid retention, heart failure were early

events. If you look at discontinuations for

example due to edema-related adverse experiences,

including heart failure, patients tended to

discontinue--if they were going to discontinue,

they discontinued early and then the two groups

continued in parallel. But, yes, it was an early

finding as you would expect.

DR. WOOD: Tom?

DR. FLEMING: Could you show us the curves

back from the VIGOR trial that looks at complicated

confirmed upper GI?

DR. BRAUNSTEIN: Complicated confirmed

upper GI?

DR. FLEMING: Correct.

DR. BRAUNSTEIN: We don't have those.

DR. FLEMING: You just quickly referred in

198

your presentation to the results being positive.

DR. BRAUNSTEIN: The results were that the

two curves showed the same V-like difference and

they continued to separate over time. I am just

looking here and apparently we don't have that

slide.

DR. FLEMING: You showed us the confirmed

upper GI and those cumulated to rates of 4.5

against 2.1. The data we have been provided

separately for the complicated confirmed upper GI

are 1.4 against 0.6. So, it is the same relative

risk but a much less frequent event.

DR. BRAUNSTEIN: Sure, yes, and those were

mostly GI bleeds.

DR. FLEMING: I was just curious to see a

pattern as to whether that is, in fact,

cumulatively increasing or more apparent early in

time.

Let me go on to the next point. That

reflects approximately numerically almost exactly

the same number of prevented cases of complicated

confirmed upper GI as there were excess numbers of

199

thrombotic cardiovascular SAEs. In essence, what

you have said is that the analgesia was comparable.

So, essentially what we are really looking at is

relative safety profiles and the goal here is to

reduce the upper GI. And, we are essentially

preventing an equal number of upper GI complicated

events for equivalent numbers of excess events in

the thrombolytic cardiovascular arena. Yet,

essentially I think you were saying the latter

didn't seem as established yet numerically it was

the same.

There were also in the trial excess

numbers of deaths of 22/15 and when you presented

the Alzheimer's data you gave us I think slide 35

that indicated that when you looked at the

Kaplan-Meier curves for confirmed thrombolytic

cardiovascular events that didn't seem to reinforce

the excess rates that you were seeing with VIGOR

and, yet, it did reinforce the excess mortality as

you have now circled back and reported at the end.

In 2003 the excess mortality is quite significant

but it was also significant in 2001. The latter

200

date is in Tab G, page 2 but the former data is in

Tab F, page 39 where excess mortality was

significant at 33/20 and the cardiovascular were 8

to 4. So, you were seeing from these two sources

excess mortality and you were seeing excess numbers

of thrombolytic events that were equivalent in

number to the number of prevented complicated

confirmed upper GI events. Am I correct on this

summary?

DR. BRAUNSTEIN: Well, no. There are a

couple of points I would disagree with. First, in

VIGOR the difference in mortality was not

statistically significant and also in terms of

looking at the causes of death, cardiovascular

mortality which is the difference we would see was

not different between the two groups. There were 7

on rofecoxib and 6 versus naproxen. So, I am not

sure--

DR. FLEMING: Well, I don't think we

disagreed. I am not talking about statistical

significance here. I am talking about what the

data are actually suggesting in what is available--

DR. BRAUNSTEIN: Well, I must say there is

a lot of data that you pointed out to me and--

DR. FLEMING: Well, just to summarize the

201

essence, while you have emphasized appropriately

the upper GI events being decreased, when you look

at the actual number prevented in complicated

confirmed upper GI it is numerically almost

identical to the number of excess thrombolytic

cardiovascular SAEs that were seen in VIGOR. You

also saw a numerical increase of a relative risk of

1.5 on mortality, which was also seen in the

Alzheimer's study which you were saying at the time

was contradicting the sense of concern related to

the overall thrombolytic excesses. And, what you

were seeing at the time, even back in 2001, was a

statistically significant excess in death rates

with a doubling in cardiovascular-related deaths.

DR. BRAUNSTEIN: Let me ask Dr. Reicin

because she perhaps has a better handle on it and I

am sort of getting lost in the mass of data that is

coming up.

DR. REICIN: I think there are two issues

202

that I think you brought up.

DR. WOOD: Sorry, just for the record, can

you identify yourself?

DR. REICIN: I am Dr. Alise Reicin, Vice

President of Merck Research Labs. In terms of

looking at VIGOR, I think you are correct. There

was excess in cardiovascular events on Vioxx and

there was a decrease in the complicated GI events

on naproxen.

DR. FLEMING: Which numerically were

almost identical.

DR. REICIN: And I think that that is also

fair to say. If you compare our data versus

diclofenac and ibuprofen at the time, there was no

difference in cardiovascular events. In fact,

numerically it was in favor of Vioxx and, yet,

there was a significant reduction in GI events.

So, that takes care of that. So, versus naproxen,

I think you are right, there was excess in CV,

lower in GI versus ibuprofen and diclofenac,

however, no evidence of an increase in CV and a

reduction in GI.

In terms of the mortality data that we had

at the time, we had a significant reduction in

mortality on Vioxx versus non-naproxen and the

203

NSAIDs that we had in our Phase III OA studies, and

at the time we actually did not make a lot of

those. We thought it was potentially by chance.

That was actually driven by CV mortality in the

non-naproxen group.

In VIGOR there was a numeric imbalance, 22

to 15 in deaths, but cardiovascular mortality was

similar. In terms of Alzheimer's I don't think

there was statistical significance back at the time

of VIGOR. There was a numeric imbalance. In terms

of cardiovascular I think the numbers were 8 versus

4. They were put in the label. So, pretty small

numbers. The rest of the difference that we saw

was due to things like poisoning, electrocution and

other things that we thought were no drug related.

DR. FLEMING: You are correct, it was 8

versus 4 in cardiovascular related deaths, but it

was statistically significant in total mortality at

that time as well. It was 33 against 20, with p

204

values reported, depending on the method, of 0.007

to 0.26.

Now, the final data are significant but

even the early data were significant and reflected

the level of excess mortality that VIGOR was

establishing but not in a significant fashion.

DR. REICIN: Again, we didn't see it

though in any of our other data sets. In fact, in

the early data sets statistically it went the other

way, non-naproxens had higher one. I think you can

see that in RA also there was no evidence of an

excess. In ADVANTAGE there was no evidence of an

excess. You see now in ViP and--

DR. FLEMING: But there was in ADVANTAGE.

There was an excess.

DR. REICIN: Not in overall mortality.

DR. FLEMING: Yes, in overall

mortality--oh, I am sorry, in Alzheimer's.

DR. WOOD: Tom, have you finished?

DR. FLEMING: Yes.

DR. WOOD: Dr. Shapiro?

DR. SHAPIRO: I guess I want to follow up

205

on a comment that you, Dr. Chair, made. I am still

concerned about the label change and how helpful or

not helpful it was, not only because it may not

have been as helpful as it might have been to

clinicians but also to patients in the informed

consent conversation. What else was made available

or should have been made available or could have

been made available to clinicians to make some

sense out of this, caution should be exercised when

Vioxx is used in patients with a medical history of

ischemic heart disease?

DR. BRAUNSTEIN: Were you addressing me or

the Chairman? Me? What we made available were the

data. I mean, I think that is the answer to the

question in terms of the labeling and in terms of

what we had published.

DR. SHAFER: So, the VIGOR and Alzheimer's

results were made available. You just weren't

going to analyze them to make any more definitive

statements at that time about what clinicians

should take away?

DR. BRAUNSTEIN: Well, by 2002 we were

206

also starting to implement our outcome study. We

thought the important message to clinicians was

that there is a GI benefit and there is also a

cardiovascular finding that we don't understand

given the differences between the two data sets.

It says the clinical significance was unknown and

that this information needs to be taken into

consideration when assessing the risks and benefits

of these drugs in individual patients. Individual

patients differ in terms of their risk profiles and

that decision on which drug to be used is best made

on a patient by patient basis.

DR. WOOD: Dr. Ilowite?

DR. ILOWITE: Rofecoxib was pulled from

the market approximately three weeks after its

approval in children with juvenile rheumatoid

arthritis. I have two quick questions. Were there

any cardiovascular events in any of the trials in

children?

DR. BRAUNSTEIN: No.

DR. ILOWITE: Second, did you give any

consideration to the fact that there were no other

207

COX-2 inhibitors, other than one NSAID that was

available as a liquid, before you made the decision

to pull it from the market?

DR. BRAUNSTEIN: The focus I think was on

the list we had seen versus placebo in the adult

patients. This kind of disease, cardiovascular

disease, is not very common in children and we

hadn't seen anything like that in our population.

DR. WOOD: Dr. Boulware?

DR. BOULWARE: I want to go back to the

previous question. What I heard was a discussion

about an offset between complicated GI events and

it sounds like non-fatal MIs. If I understood the

discussion back and forth here, they are roughly

comparable. Now, in patients requiring an NSAID,

and I am not talking about the APPROVe data here

but in patients requiring NSAID treatment if there

is roughly comparability of complicated

gastrointestinal events with non-fatal MIs, it

sounds like Merck's thinking was that the risk of a

non-fatal MI far outweighs, in a patient requiring

NSAID treatment, the risk of complicated GI events

208

and that that was what drove the decision.

The reason I am interested in this is that

obviously this meeting is entirely about how you

make a risk/benefit calculation. So, your thoughts

in September about this issue are I think helpful

to us in thinking about these risk/benefit issues.

DR. BRAUNSTEIN: I wouldn't put it exactly

the way you stated it, and that is because the

individual patients at risk for these problems

differ and there were alternative approaches for

patients with GI risk that were available at the

time. Now, we recognize that rofecoxib had met the

highest standard. Well, yes, it had met the

highest standard but there were alternatives

available and we did not have data on what one

could do for more studies. The data was unclear as

to the mechanism so we felt that given those

options, the withdrawal made the most sense.

DR. BOULWARE: Can I just make a little

follow-up comment? It sounds like you are trying

to have your cake and eat it too. On the one hand,

you would have liked to have said pre-September

209

that rofecoxib was the only COX-2 selective drug

that had demonstrated effect in reducing GI

toxicity. Now you are saying, after you pulled it

from the market, there are lots of other

alternatives that are almost just as good. I don't

really understand.

DR. BRAUNSTEIN: I couldn't say "almost

just." There haven't been head-to-head studies to

answer that latter part of your question. There

were alternatives. We did not know that there is a

class effect for cardiovascular.

DR. WOOD: Dr. Manzi?

DR. MANZI: This question actually may

better be answered by Dr. FitzGerald, I am not

sure--is he here?

DR. WOOD: Here he comes, just in time.

DR. MANZI: He eloquently pointed out that

there is clearly variability in individual dose

response with regard to COX-2 inhibition. Since we

are grappling with this issue of class effect

versus a specific drug effect, is it feasible or

helpful to look at the degree of COX-2 inhibition

210

in association with these events?

DR. WOOD: You are up, Garret. Just take

that microphone.

DR. FITZGERALD: I would say yes amongst

all those things.

DR. WOOD: Amongst all those things? I

don't understand.

DR. FITZGERALD: I mean one of the issues

that you would hypothesize is relevant to outcome

is the degree of selectivity attained in an

individual.

DR. WOOD: You mean amongst other things

related to the drug?

DR. FITZGERALD: Amongst other things

related to the drug and underlying--

DR. WOOD: Sure. Dr. Platt?

DR. PLATT: Compared to other NSAIDs, do I

understand properly that 98 out of 100 patients who

take the drug would have about the same outcome?

That is, the significant difference between the

regimens is approximately--2-fold means about a 2

percent absolute difference.

DR. BRAUNSTEIN: Which outcome are you

referring to?

DR. PLATT: To the GI outcomes.

211

DR. BRAUNSTEIN: There is a range. There

is a small range because it does seem that we have

a larger difference--you know, if you line them up

it is a little larger with naproxen and a little

smaller with diclofenac but I would say on average

it is about two-fold.

DR. PLATT: Right, but that 2-fold

translates into about two patients out of 100

having a different outcome than they would have if

they had taken the comparator. I am trying to get

at the question of whether we can identify those

two patients with greater certainty than just

treating everyone. And, I would ask the same

question about the cardiovascular complications.

That is, in this complicated business of risks and

benefits, can we do better than we have at guiding

both clinicians and their patients in having at

least semi-quantitative estimates of what the risks

will be and what the benefits will be so they can

212

make an informed judgment?

DR. BRAUNSTEIN: We know that from the

VIGOR results because we looked at patients with

different baseline risk for GI disease, and this is

something that is well understood, what the

different risk factors are for GI disease,

including things like prior history of a GI event,

and we saw the same 50 percent reduction across all

the different risk factors. In terms of

cardiovascular, we are still introduction he

process of trying to see if we can identify

particular risk factors that would correlate. So,

that is still an open question based on our data.

DR. PLATT: But saying 50 percent really

obscures the fact. Some people may have a baseline

risk of a serious GI event of 20 percent or 30

percent, in which case 2-fold is a very big

improvement for them--

DR. BRAUNSTEIN: Yes, of course.

DR. PLATT: If we knew enough we would

know that most people have effectively a zero risk.

So, there is very little benefit for them. Have

213

you put the data together in a way that helps us

identify the people who stand most to benefit and

the people who stand most at risk, and is it

possible that those are different groups?

DR. BRAUNSTEIN: Dr. Reicin can I think

provide more information on the VIGOR results

because she was involved extensively in the VIGOR

study.

DR. REICIN: Dr. Laine may come up to help

me if I don't remember something. We actually

published a paper on looking at specific subgroups

in the VIGOR study. What we found is very similar

to what Byron talked about during his discussion.

Patients with typical risk factors, age more than

65--do you want to add something?

DR. LAINE: I agree absolutely. The

reason I actually took these data and published

this paper with the VIGOR results is that I have

had the same idea. Relative risk isn't important

in practice; it is the absolute change, the number

needed to treat. So, we looked at that with

absolute incidence of number needed to treat and

214

for clinical events, for instance, if you had a

prior event you only have to treat ten people for

one additional event. But if you don't have a

prior event you have to treat, let's say, 60 or 70.

The same with age, if you are over 75 you only need

to treat ten people for one additional event. But

if you are under 65 you need to treat 50 or 60.

So, I agree absolutely that at least with the VIGOR

data, we stratified by these different clinical

risk factors that Byron showed earlier.

DR. WOOD: We have three more questions,

Dr. Shafer, Dr. Cush and then Dr. Temple.

DR. SHAFER: Two questions. Can you go to

slide 48?

DR. BRAUNSTEIN: That is the subgroups,

yes?

DR. SHAFER: Yes, is the one on various

subgroup analyses. Can we show the slides? Just

to highlight what the question is, in slide 48,

this is following on the comment by Dr. Nissen

regarding the aspirin use, what you show in the

APPROVe trial is that the risk factor for those

215

with aspirin on board, in fact, is 3.25 with a

confidence interval which is wide, as Dr.

FitzGerald has suggested it might be because of

small numbers, but it goes from 0.98 to 13.81.

Now, the hypothesis behind VIGOR and

interpreting VIGOR as an aspirin-like effect, was

that aspirin was going to confer safety. Doesn't

the data on slide 48 essentially disprove the

naproxen hypothesis in VIGOR?

DR. BRAUNSTEIN: No, there is no naproxen

in the study--

DR. SHAFER: Right, but the hypothesis was

that naproxen was acting like aspirin.

DR. BRAUNSTEIN: Yes.

DR. SHAFER: Yet, here in the presence of

aspirin to provide the safety, you are not seeing

benefit.

DR. BRAUNSTEIN: I would argue that the

mechanism for what we saw in VIGOR, which was a

very early difference between the two treatment

groups, is qualitatively very different than what

we see in APPROVe. So the mechanism for the

216

cardiovascular difference in the two studies is not

necessarily the same and, therefore, whatever

difference we are seeing here or not seeing with

aspirin doesn't really relate to what we saw in

VIGOR. I would also point out, as you have already

pointed out, there are wide subgroups. I think Dr.

Villalba has pointed out that when we looked at the

APTC endpoint, which was just myocardial

infarction, stroke and vascular death, the

difference actually seems to go away but, again,

there are very small numbers and we don't want to

over-interpret at this point what the data say.

DR. WOOD: But the major point here, just

to help you here, is that these people were not

randomized to aspirin. So, people who were on

aspirin were a different subset than the people who

were not on aspirin in terms of cardiovascular risk

and so on. So, it is not like naproxen.

DR. BRAUNSTEIN: Yes. Yes, of course.

Sure.

DR. WOOD: The one thing I would say while

you have that slide on there is that I think is

217

going to be important for us is that our job is not

to identify groups that are at particular risk,

Richard. Our job I think is to see if we can

identify patients who are at low risk--

DR. PLATT: Yes.

DR. WOOD: I am not arguing with you. I

am just making a generic point and it is not clear

to me that there is such a group identified there.

DR. PLATT: Well, it seems to me that

there will always be risk--

DR. WOOD: Right.

DR. PLATT: --the question is can we help

inform decisions that patients have to make?

DR. WOOD: Dr. Cush?

DR. CUSH: Dr. Braunstein, a few times you

mentioned that you made this decision based on the

signal that you found in the alternatives existing,

and not knowing if it is a class effect. If you

knew that this was a class effect would you have

made the same decision? And, knowing what your

COX-2 potency is, does that factor into that?

DR. BRAUNSTEIN: I couldn't go back and

218

speculate what decision we would have made based on

a different set of data.

DR. WOOD: I think that is a fair answer.

Let's move on to Dr. Cryer.

DR. CRYER: I would like to come back to a

consideration of the potential gastrointestinal

benefits of COX inhibitors and specifically Vioxx,

and I am going to use your slide 33 to help me with

my questions and comments.

You repeatedly made the point that Vioxx,

rofecoxib, was unique in its labeling with respect

to its gastrointestinal benefit and that was a

label revision that was largely derived from a

discussion of the data in the VIGOR trial in which

naproxen was the comparator.

I want to underscore that the conclusions

reached may be as much of a reflection of the

comparator as they could be a reflection of

properties intrinsic to the COX-2 specific

inhibitor. As I look at the pooled analyses from

the rofecoxib experience and specifically look at

diclofenac, it does not appear that the difference

219

in reduction compared to diclofenac is

statistically significantly different.

So, the question that I have for you is do

you think that the revisions in the label would

have been the same with respect to the GI

observations in VIGOR had diclofenac been the

comparator rather than naproxen?

DR. BRAUNSTEIN: In an adequately powered

study. I think the failure here in these confirmed

events, in order to have the confidence interval

narrow enough we would need enough power to do

that. In fact, when we looked at investigator

reports of these events, in all, including the

unconfirmed, we did have statistical significance.

So, I think that, yes, in an adequately powered

study we would show a difference from diclofenac.

DR. WOOD: Bob?

DR. TEMPLE: Actually, I wanted to pursue

something Dr. Shafer raised. The aspirin subgroup

is a baseline subset. People are probably

reasonably well randomized to whether they get--

DR. WOOD: They didn't get aspirin.

DR. TEMPLE: No, I know. They were

different populations from people who were on

aspirin but they are randomized to the two

220

treatments, and there is about a thousand of them.

From everything that I would have understood from

Dr. FitzGerald's talk, when you are on both aspirin

and rofecoxib you are not on a selective drug

anymore, or probably not because you have plenty of

COX-1 inhibition. But the hazard ratio there is

higher than the other people. I wonder whether

that is easily explained, or it could be explained

by blood pressure effects which, of course, aspirin

will not reverse. Because I think it needs some

kind of explanation.

DR. WOOD: So, is that addressed to

Garret?

DR. TEMPLE: Either.

DR. BRAUNSTEIN: With regard to aspirin

data, they are not robust enough. We are talking

about a total of 11 events, as I recall, in that

analysis for the APTC. There are not a lot of

events in that analysis.

DR. TEMPLE: There were 16.

DR. BRAUNSTEIN: Right, 16 events. There

are very wide confidence intervals, as you know.

So, I think it is difficult to draw specific

conclusions about aspirin. With regard to blood

pressure, as I indicated, when we looked at that

221

the blood pressure changes that we observed would

not appear to explain the magnitude of the

cardiovascular findings that we observed in

APPROVe.

DR. TEMPLE: One of the reasons to worry

is that people with underlying heart disease or

diabetes are probably more sensitive to blood

pressure effects. There is some evidence of that.

Anyway, just a thought.

DR. WOOD: Garret?

DR. FITZGERALD: I would just say one can

over-parse extraordinarily small amounts of data in

retrospect, and that there is enough flexibility in

what one would expect to see to account for that.

For example, we don't actually know if inhibition

of COX-1 has no impact on the blood pressure

222

response to a COX-2 inhibitor. In fact, from what

I showed you in mice, one would anticipate if one

actually designed a study to address that question

that the answer would be yes. So, I think that,

coupled with the fact that aspirin, even if one had

loads of data, would be expected to modulate rather

than abolish the hazard through this mechanism

really means that it is not an answered question

rather than an answered one.

DR. WOOD: Great! Well, let's stop at

this point and break for lunch. We will restart at

exactly one o'clock.

(Lunch recess.)

223

A F T E R N O O N P R O C E E D I N G S

DR. WOOD: Merck has a couple of slides

they wanted to show to address the blood pressure

issue that came up in the previous discussions.

So, let's go ahead and do that first quickly.

DR. REICIN: The first was in relation to

the issue about congestive heart failure, which is

a known side effect of all NSAIDs and COX-2

inhibitors and is reflected in their labeling.

Since the only data we showed was from APPROVe, we

had an expected difference from placebo but if you

look at this slide you see that in our OA database

the incidence of congestive heart failure was low,

and it was generally similar to ibuprofen. You can

see that it ranged from 0.1 to 0.4 percent on

rofecoxib; 0.4 percent on ibuprofen; and 0.8

percent on diclofenac--so, generally similar to the

NSAID comparators. I will acknowledge that there

was one epidemiologic study that suggested that the

rate was higher on rofecoxib.

DR. WOOD: But the data in the APPROVe

study are up to 1.5 in the rofecoxib group, and

224

this is for serious heart failure--congestive heart

failure, pulmonary edema. Right?

DR. REICIN: It was versus placebo. The

rate was higher in that study than we have seen in

other studies.

DR. WOOD: Right. But it is not a

question of whether it is against placebo or not.

The underlying rate is much higher.

DR. REICIN: The rate was higher in that

study. We didn't see it as high in our other

studies.

The other slide, 232--it was a question

about whether rofecoxib had effects on blood

pressure that were very different than the other

NSAIDs. This was a study done in elderly patients.

It was not an ambulatory blood pressure study but

blood pressure was measured in these patients 4

times a day. If you look, rofecoxib 25 mg was

compared to celecoxib 200 mg BID. That is the

highest recommended chronic dose for both of these

medications. The medications at that dose had

similar inhibition of COX-2. We compared it also

225

with naproxen 500 BID and placebo, and I think you

can see that the changes in systolic blood pressure

and diastolic blood pressure are similar among the

active treatments and greater than placebo.

DR. WOOD: Okay. Thanks very much. These

are helpful comments. Are there any questions

specifically and only on these two things? Steve?

DR. NISSEN: Could you show us the use of

antihypertensive agents in the two arms of APPROVe?

I would be interested in seeing if there was a

difference in use of antihypertensive drugs. I am

also interested--you know, these mean changes are

useful but it is also useful to know the fraction

of patients that had sustained increases of, say,

15 mm or more because that is the kind of level of

increase that would constitute a substantial risk.

So, I am interested in use of antihypertensive

drugs and I am interested in the number of people

who had greater than a 15 mm sustained increase in

each arm.

DR. NORGAN: Kevin Norgan, Merck. The use

of antihypertensive drugs in the APPROVe study, at

226

baseline it was approximately 30 percent. It was

30 percent in one treatment group and 29 percent in

the other treatment group. Then, during the course

of the study the numbers increased to approximately

40 percent in the rofecoxib group and approximately

35 percent in the placebo group. The actual

numbers are in the publication that is on the

Internet.

DR. NISSEN: And was that difference

statistically significant?

DR. NORGAN: I don't recall. I think it

was but we would have to check.

DR. WOOD: Then 25 patients dropped out

because of hypertension versus 7 in the placebo

group. Right? So, that should be added to the

number that actually ended up on antihypertensives

in the APPROVe study.

DR. NISSEN: What about the issue of the

15 mm or greater? Do you have any data on that?

Bob Temple, isn't that something you guys like to

look at in the FDA, the sort of 15 mm outlier

group?

DR. TEMPLE: I don't know. I think we

look at mean just as often.

DR. NISSEN: All right, but I would like

227

to know because I didn't see that.

DR. REICIN: We will get back to you later

with that data, Dr. Nissen.

DR. WOOD: That being the case, let's move

on to the next presentation which is from the FDA.

FDA Presentation: Vioxx (Rofecoxib)

DR. VILLALBA: Good afternoon. My name is

Lourdes Villalba and I am a medical officer in the

Division of Anti-Inflammatory, Analgesic and

Ophthalmic Drug Products. I have been the primary

reviewer for Vioxx since 1998 when the NDA was

originally submitted for approval.

DR. WOOD: You need to move closer to the

mike.

DR. VILLALBA: So the important thing, I

have been the primary reviewer for this since 1998,

since its original submission for approval for the

treatment of acute pain, dysmenorrhea and

osteoarthritis.

I am going to show you an overview of my

presentation. First of all, my goal is to show you

that we were not sleeping behind the wheel, that we

have been actively engaged in reviewing the

enormous amount of data that came to our division

throughout the years, and this has been a very

228

challenging application, a very complicated process

to review a lot of information that was not always

that clear to interpret.

The other issue is that I want to point to

some observations that may help you to think about

the best study designs. Everybody is talking about

future studies to clarify the question but the

issue is exactly what kind of studies we need; what

should be the comparator; how long, etc.

So, first of all, I am going to show you a

brief background with a chronology of events to

point out just some specific areas that I want you

to remember. Then I am going to give you an

overview of the Vioxx data sources we reviewed and

that will be presented also in a chronological

order. Then I am going to spend a few minutes

229

talking a little bit again about the different

classification of cardiovascular events. Then I

will go into the different Vioxx databases showing

cardiovascular safety in the way we saw it at the

time they were presented to us. Then a summary,

pointing out again to the challenges in

interpreting this data.

This is a busy slide and I apologize for

it but I want to point out a few areas here. The

NDA was originally submitted in 1994 and it was

approved in '99 after the data was presented at an

advisory committee meeting. Around the time of

approval there were all sorts of submissions of IND

investigational new drug applications for other

indications. One of them was submitted to the

Division of Neuropharm. Products for the evaluation

of the role of Vioxx in the prevention and

treatment of Alzheimer's disease. Another one was

submitted to the Division of Oncologic Drug

Products and that was approved, the adenomatous

polyps prevention trial that now led to the

withdrawal to the product. It was initially

230

submitted to the oncology division and then was

switched to GI. That is just a detail. This was

back in '99.

Then we had the results of VIGOR in the

year 2000. The advisory committee meeting of

February, 2001, and after that we reviewed a lot of

information and finally got the labeling changes in

April, 2002. Later on, in October of that year,

there was a submission of another study in the

Division of Reproductive Products to evaluation the

role of Vioxx in the prevention of prostate cancer.

About the same time Merck came to us with a

proposal for conducting a pooled analysis of some

of these studies, particularly APPROVe, the

prostate cancer prevention and another study that

was being conducted in Europe.

The Alzheimer's data was very important

data that I will go into detail later, but I want

to mention that preliminary data from these studies

that were placebo-controlled studies were initially

submitted in July, 2001. The final data from this

database was provided to us in March, 2004.

This is an overview of all the databases

we reviewed and this does not include APPROVe. As

you can see here in the first column, although it

231

says "indication" the indication refers to the

line. But here we have the treatments. We have

Vioxx at 3 doses, the 2 approved for chronic use

and also for acute pain, the 50 mg, as well as some

comparators, ibuprofen, diclofenac, nabumetone and

placebo. The original NDA did not have naproxen.

Then with have VIGOR which had Vioxx 50

and naproxen. We had other studies. Unfortunately

this slide is not the last one. There are missing

2 important marks, the 25 mg dose in study 102,

also known as ADVANTAGE study, and that was 25 mg

versus naproxen; then the rheumatoid arthritis

efficacy database that compared Vioxx 12.5, 25 and

50 to naproxen and placebo.

Then we had several studies, safety

outcome reports for various indications with

different comparators and also with placebo. I

want to point out here with placebo that in the NDA

database we had up to 18 weeks. Most of the

232

studies were 6-week studies but there were 18-week

placebo-controlled studies that were endoscopic

studies.

We have placebo here in the rheumatoid

arthritis efficacy data base, but the most

important data we had was here, in the Alzheimer's

studies that were long-term placebo-controlled in

an elderly population. We had data for at least 3

years. So, we put a lot of weight on this

information. We also had access to the adverse

event report system but, unfortunately, it is known

that it is not very helpful to look in this way

when we are talking about relatively prevalent

events such as cardiac events. Then we also have

literature, epidemiologic studies, re-analyses and

meta-analyses of data that had been published.

Before I show every database I want to

spend a few minutes on the cardiovascular endpoints

because there are many different ways of looking at

cardiovascular endpoints. At the FDA we routinely

look at all adverse events reported under that

category--cardiovascular deaths, discontinuation

233

due to cardiovascular adverse events, serious

cardiovascular adverse events--this is routine and

this is what we did in the NDA application. We did

the same as well with all the other organ systems.

So it is a very in-depth review.

At the time of VIGOR and all studies

subsequently, the sponsor used a standard operating

procedure that used a subset of cardiovascular

serious adverse events, a category of

cardiovascular and thrombotic adverse events, and

these were referred for adjudication to a blinded

adjudication committee. The committee of three

cardiologists would determine if the events were

confirmed or not confirmed. Another definition

that was used was if they were part of the APTC

definition or not.

So, these two ways of looking confirmed

cardiovascular/thrombotic and APTC are not a subset

of one analysis; they are complementary. The APTC

endpoint, the composite endpoint that looks at

cardiovascular and unknown cause of death is

non-fatal myocardial infarction and non-fatal

234

stroke. It includes ischemic and hemorrhagic

events, but does not include unstable angina,

transient ischemic attack and peripheral events.

These kind of events are included in this

definition up here. But this does not include

hemorrhagic events.

So, we looked at the data in all these

different ways. Again, at the NDA stage we looked

in this general way and we became more

sophisticated and looked in all ways, the original

one and the others.

This is very important. Let me show you

just an example from VIGOR so you can have a clear

idea of what I mean. You have different ways of

looking at it. If you look at all investigative

cardiovascular and thrombotic events you are going

to have more events. If you look at confirmed or

adjudicated events you still see the difference but

the number of events is small. The same with the

APTC. Therefore, this way of looking is more

specific because it looks at the hard endpoints.

Although it may be less sensitive, the other way of

235

looking--let's say we have here all cardiovascular

events submitted under that category, we would have

for VIGOR 600 events and 400 events.

So, in my presentation I am going to use

the APTC way of looking at it because it also makes

very clear if there were cardiovascular deaths or

not. This is a different way of presenting the

data that was already presented by the sponsor.

Here we have the NDA database. The NDA,

submitted in 1998, was very large and involved 5400

patients with substantial exposure in multiple dose

trials of 6 weeks, 6 months and up to a year

studies. Some of the 6-week studies had extensions

to 21 months.

I want to point out that this number is a

substantial number for an NDA. This is greater

than most NDAs. Although most of the COX-2

selective agents have had this kind of size of NDA,

but before that we used to approve products based

on much smaller data. These numbers are above

minimum requirements by the International

Conference on Harmonization Guidances.

In this database, looking the way I told

you, looking at all adverse events that were

cardiovascular adverse events and were potentially

236

thrombotic, serious and non-serious, this is what

we found. This was kind of a definition that I

made myself to look at these events. This has not

been validated but, in any case, this is what we

saw.

In the 6-week studies with Vioxx with all

doses we have 0.7-1.1 rate, and these are crude

rates here. There was 0.4 with ibuprofen; 0.2 with

placebo. Therefore, we said, okay, there may be

something here but if you look at the number of

patients exposed, they were different. There were

more patients exposed to the Vioxx doses. And we

didn't really know what to do with these

percentages. What dose it mean with an endpoint

that is not really well defined?

Then, in the 24-week studies that had

placebo, up to 18 weeks, the crude rates on Vioxx

at all doses were right between ibuprofen and

diclofenac. Ibuprofen was 0.5 percent; diclofenac,

237

2.0; and placebo was 0.8 which was also in between.

So, based on these data, based on the fact that

they looked similar to the other NSAIDs and that it

was a pretty large database, of course, not

designed specifically to address cardiovascular

issues, what we said was that there doesn't seem to

be a big problem here, however, we cannot rule out

that there could be something but this is not the

right database to address it. On the other hand,

this was 1998. There were theoretical concerns

regarding that inhibition of prostacyclin could

induce prothrombotic events. But based on these

data, there was not much to say about it. Also,

Celebrex had recently been approved, in December,

'98, and Celebrex had not shown anything either.

So, again, based on adequate evidence of

efficacy, safety for the intended uses and the

similarity to the comparators, this drug was

approved in May, 1999. If you look at the safety

profile, it was pretty similar to other NSAIDs.

Cardiovascular safety was between ibuprofen and

diclofenac. Hypertension, there was a very clear

238

dose response with the 50 mg being greater than the

12.5 and 25. Endoscopic data suggested that Vioxx

was better than ibuprofen and the liver suggested

that Vioxx was better than diclofenac. So, it was

an NSAID.

The sponsor wanted to pursue the claim

that this was a COX-2 selective agent; this needs

to be different. We really don't want to see the

GI warning template in our label. And, that was

even before the NDA was submitted. It was

discussed a long time before. If you really want

to have a substantial change in the GI label, then

you have to do large outcome studies or at least

one large outcome study. That is why we had VIGOR.

VIGOR was not a requirement. It was something that

the sponsor decided to do because they wanted to

distinguish themselves from the other NSAIDs.

We know the result of VIGOR. I am showing

here the APTC results. From now on I will use the

same format for all my slides. So, please bear with

me for this first slide. We have the APTC total

events in the first row; then cardiovascular

239

deaths; non-fatal myocardial infarction; non-fatal

stroke; and non-fatal hemorrhagic stroke. We have

the comparators here. N is the number of patients

randomized. Here, in the footnote, I have the

number of patients in patient-years of exposure. N

gives you the number of events. Rate is the rate

based on 100 patient-years of exposure, and the

relative rate is the overall rate for Vioxx as

compared to the comparator.

I think I don't need to spend too much

time describing VIGOR. It was a large study, 400

patients per arm; patients with rheumatoid

arthritis, 60 percent using corticosteroids, 40

percent using methotrexate and most were women, and

patients on low dose aspirin were not included in

this study.

This is what we found. There was a

difference in rate of cardiovascular/thrombotic

events or APTC events, and the different risk was

driven by the non-fatal myocardial infarction.

This number was statistically significant.

But if you look at cardiovascular deaths

240

there was no difference. Non-ischemic stroke,

there were no differences. So, this was the first

time when we saw the signal of Vioxx being

different from naproxen. This is the time to vent

plot that shows the cumulative incidence of events

over time. This is for the

cardiovascular/thrombotic events but it looks very

similar for the APTC events.

I did show this slide back in 2001 at the

advisory committee meeting that we had to discuss

VIGOR. You see that the curves start to separate

here, at 6 weeks, but this separation is more

marked after 8 months. So, if you look at the

overall hazard ratio it is 2.4, but after 8 months

that hazard ratio increases and is 4.0. There was

a lot of discussion with the sponsor about the

interpretation of this part of the curve and our

position was that there was increased risk after 8

months. Finally we got that into the 2002 label in

the form of a table that shows an analysis of

cumulative rate of events over time that shows that

the hazard ratio increases after 8 months.

In any case, that difference between

naproxen and Vioxx was driven by the non-fatal

myocardial infarctions. There wee 9 myocardial

241

infarctions during the last 3 or 4 months of the

study on Vioxx and there were none on naproxen.

The position of Merck was that this was the

cardioprotective effect of naproxen. However, we

did state clearly at the advisory committee back in

2001 that we were very skeptical about that

interpretation and that actually there was

biological plausibility for a prothrombotic effect

of Vioxx as well.

This is another study that was submitted

to us in June, 2000 along with VIGOR. That was

also presented to the advisory committee meeting in

2001, showing study 090 and 085. These were two

identically designed studies, placebo-controlled, 6

weeks in duration comparing 12.5 mg of Vioxx with

nabumetone and placebo. Of note, they have a 2:1

randomization. That means that the number of

patients in the active treatment groups was twice

the number of patients on placebo.

In this study, study 090, showed 3

non-fatal myocardial infarctions and 1 non-fatal

stroke on Vioxx, 1 on nabumetone and none on

placebo, and no cardiovascular deaths. Study 085

showed only 1 non-fatal myocardial infarction in

the Vioxx 12.5 mg and nothing else in any of the

242

other arms. Therefore, with this information, the

small number of events, the fact that study 085 did

not reproduce the findings in 090 which, to start

with, were very mild, we didn't know what to do

with this. This was discussed at the advisory

committee and there were not any meaningful

conclusions from these studies. Again, I want to

mention that there were twice the number of

patients in the Vioxx group as compared to placebo.

The conclusions of the advisory committee

were that Vioxx showed a superior GI safety profile

as compared to naproxen; that the cardiovascular

signal was of concern, however, given the study

design it was unclear how it would apply to other

populations, other doses, NSAIDs other than

naproxen in populations that were at high

243

cardiovascular risk because this trial had excluded

patients using low dose aspirin. And, that

labeling changes should reflect both benefits and

potential harms and that additional data were

needed to clarify these issues. There was no

recommendation for a specific trial to be

conducted, or specific design. That is why I think

it is important that today you actually give some

kind of firm recommendations and give us direction

as to which kind of studies you want to see.

We asked for more data and we got more

data. That came continuously right after the

advisory committee and at the end of February we

had the application for the rheumatoid arthritis

efficacy indication. It was relatively large.

There were 1500 patients on Vioxx and the active

comparator was naproxen. There were not other

comparators. There was also placebo here.

There were 5 studies. The endoscopic

studies were 12-week studies but the other studies

had a 12-week base study with re-randomization of

patients who were on the lower doses of Vioxx to

244

Vioxx 25 and 50. And placebo here, to 25, 50 or

naproxen. So, it was a pretty complex NDA to

review.

In any case, here we have the summary.

This follows a different format but this is the

summary of the results. In the first column you

have the treatment and the number of APTC events;

patient years at risk; and risk per 100

patient-years. As you can see here, it is clear

that Vioxx 25, 50 and 12.5 showed a greater risk

than naproxen and than placebo. However, if you

look at the number of patient-years at risk you

really cannot reach the conclusion that there is a

clear signal against placebo. What is clear is

that there is a signal against naproxen because

exposures to naproxen and Vioxx were closer. I

think that the risk with the 12.5 dose is 6.9 as

compared to 0.3. So, there is something wrong; you

have too small numbers to compare. If you were to

believe this, I mean, here naproxen had half the

risk of placebo.

So, the conclusion was that Vioxx 25 and

245

50, both doses, in a rheumatoid arthritis

population had a higher risk of

cardiovascular/thrombotic events as compared to

naproxen.

Then we had the ADVANTAGE study. I am

presenting you this data on one slide but this took

months to review, and we were not looking only at

cardiovascular safety; we also looked at GI, renal,

liver, everything, fractures, so anything that was

of a theoretical concern we were looking at. So,

this took months. Also, when you get the

information you get questions, get the responses

within one or two months so it is a long process

for each one of these studies.

That was for the RA. This was ADVANTAGE.

ADVANTAGE, or study 102, was submitted in March and

another piece in April, 2001. This was a 3-month

study in patients with osteoarthritis comparing

Vioxx 25 mg with naproxen. Approximately 2700

patients were randomized to each arm. Here you

have the patient-years of exposure, although I

don't like to use this number because this is a

246

3-month trial but, still, you have it there in case

you are interested. But if you look at the

numbers, the number of APTC events was about the

same. There was a signal for cardiovascular death

and non-fatal MIs. There were 9 events here and 1

here. However, there were 6 non-ischemic strokes

on naproxen and 1 on Vioxx.

So, again, the conclusion here is that

there is a signal. There is a signal for Vioxx as

compared to naproxen but we still didn't know what

the role of naproxen was here because it may have

some role but it wasn't clear what the extent of

that was. Based on epidemiologic data, the data

were conflicting. I think I would like to know if

someone knows exactly what is the role of naproxen

from all these findings.

Then we had several safety update reports

that came in July, 2001 that included studies in

the original NDA and that were follow-up from

patients who had been included in the original NDA.

There were also new studies, short-term studies and

long-term studies. The most important was the

247

study 083, the bone density study with Vioxx 25

versus ibuprofen. The most relevant data for us

was the Alzheimer's data that compared Vioxx 25

with placebo. That included 3 long-term studies.

There was also an updated meta-analysis of

cardiovascular events. The sponsor had presented a

meta-analysis in February, 2001 at the time of the

advisory committee initially and here there was an

update. Basically there was no difference in

confirmed or thrombotic APTC events. Actually, I

am talking about these other studies because the

meta-analysis was done with APTC events only.

Here is a description of the Alzheimer's

studies. There were 3 studies, 2 of them on

established Alzheimer's disease that had identical

design, 15 months in duration, placebo-controlled,

350 patients per arm, with age of at least 65 years

or older.

One of these studies has been completed

and showed no efficacy, and the median exposure in

this trial was 13 months. The second one being

conducted was stopped because the first one had not

248

shown efficacy. The median exposure was 6 months.

Then there was another study that was ongoing at

the time of the safety update. That was study 078

that was designed as a 2-year study and was

eventually extended to a 4-year study and had 730

patients per treatment arm. At the time of the

safety update report that we received in July, 2001

the exposure in this study was 18 months.

Regarding the population here, 60 percent of them

were male with a mean age of 75 years, and aspirin

was not allowed in this study initially but it was

then amended to allow low dose aspirin for those

patients who needed it. Approximately 7 percent of

patients were on low dose aspirin.

Here we have the results of that study.

Again, here you have the APTC events. I am sorry,

this is wrong. This should be 0.73 but still it is

below 1.0. If you look at total events you have 17

and 27. This doesn't look bad for Vioxx. If you

look at cardiovascular deaths, yes, there were 8

and 5, of which 3 were thrombotic and 1 was

hemorrhagic and the other was a ruptured aortic

249

aneurysm. There was twice the number of non-fatal

myocardial infarctions and 12 non-fatal ischemic

strokes.

So, based on these data, it was puzzling

that cardiovascular deaths tended to be against

Vioxx but, still, the number is relatively small.

You have 8 versus 5. Looking at the myocardial

infarction and stroke, there were more events on

placebo than on Vioxx. So, that is why I am saying

the interpretation of this data was very

challenging. How do we put together this

information for 14 months, because there was a

median of 14 months. Putting the 2 large studies

together, 091 and 078, had a median duration of 14

months and here we do not see the signal that we

saw with VIGOR.

Here is the table with the summary of the

meta-analysis that was conducted by the sponsor,

the updated meta-analysis comparing Vioxx all doses

with placebo events. I think this is the most

valuable part of this slide because the other one

is comparing non-naproxen NSAIDs and I would agree

250

that not all non-naproxen NSAIDs are the same. But

the total number, in any case, doesn't look bad for

Vioxx. The relative risks are below 1.0.

Here is what we had so far. In the NDA

database in '98 where we didn't look specifically

at APTC but, let me tell you because I forgot to

mention it before, there were 3 cardiovascular

deaths with the 12.5 mg dose of Vioxx. There were

no cardiovascular deaths with the 25 and 50 mg

doses, and there were 3 cardiovascular deaths with

diclofenac, and diclofenac had much lower exposure,

number of patients and time of exposure, as

compared to Vioxx. So, there were not signals in

the original NDA.

Then we had VIGOR that showed a signal in

APTC and non-fatal MIs. Then we had the rheumatoid

arthritis ADVANTAGE study that, as compared to

naproxen, showed trends. Again, this should be all

yellow and this, here, should be "no" because there

were no cardiovascular deaths in the rheumatoid

arthritis database.

Then we have the safety update reports

251

with the Alzheimer's studies that had 14-month,

placebo-controlled studies without difference in

MIs and strokes, but with that cardiovascular

trend.

After 2001, after the presentation at the

advisory committee meeting of 2001, there were

several epidemiologic studies and re-analysis of

the data that had been presented or published, and

meta-analysis but they showed conflicting results.

Basically we had to do our labeling changes. By

this time we were around October, November probably

of 2001. After negotiations with the sponsor, we

ended up in April of 2002 including a label that

for many of you may be very confusing or not

helpful, but that was the situation in which we

were at that time. We had conflicting data. So,

what we did, we put the result of VIGOR there. We

included two tables showing the cardiovascular

events over time, the list of cardiovascular events

by category. There was also some language in the

precautions section and the indications because the

rheumatoid arthritis indication was approved now,

252

after we reviewed all the data, not 6 months before

when we should have approved--not 6 months, we have

a 10-month clock to review efficacy supplements.

Anyway, they were not approved until we

had reviewed a substantial amount of data. There

was something also in the adverse reaction section

that pointed out to the risk of hypertension in

patients with rheumatoid arthritis with the lower

dose. Before we had something that referred to the

50 mg dose being worse than the 25 and 12.5 but in

rheumatoid arthritis patients the 25 mg dose also

showed to be worse than naproxen. There was also

some language in the dose and administration.

I am not going to go through all this,

don't worry, but I just want to point out that we

put a lot of information there and we said that we

didn't know how to interpret this data; that

prospectively designed studies have not been

conducted.

Following these label changes--again, I am

not going to insist on this but we also had

language regarding the 50 mg dose not being

253

recommended for chronic use.

In October, 2002 we had the proposal by

the sponsor to conduct a prospective analysis of

cardiovascular thrombotic events that I mentioned

earlier in the 3 long-term placebo-controlled

studies. One of them was ongoing already since

early 2000. Another one was being conducted or was

going to start soon in Oxford. They submitted the

prostate cancer prevention study at that time so it

had not even started. But all the 3 studies

together were going to provide approximately a

25,000 patient database that was placebo

controlled. the prostate cancer prevention studies

were planned to be up to 6 years in duration. So,

we had potentially a lot of information there.

We agreed with the concept of pooling

these studies and we specifically said it is

possible that these studies may address the

question we have, however, we cannot assure you

that if you don't show anything in this study you

are out of the woods So, that was a review issue.

Also, there were a lot of discussions regarding the

254

data analysis plan for these pooled analyses.

Here we have the result of the updated

data from the Alzheimer's studies. This was

submitted to us in March, '04. As you can see

here, the rate of APTC events still is not worse

than placebo. It is about the same. There are

more events on placebo but there was also longer

exposure if you look at patient-years of exposure.

There was no difference in cardiovascular deaths as

adjudicated by the committee. There were 14 and 14

non-fatal myocardial infarctions; 17 and 6

non-fatal strokes. The strokes were all in the

placebo group--sorry, not all. The point is that

here we don't see a signal on stroke; we don't see

a signal on MI. Death kind of is there, maybe or

maybe not, because if you look at the subset of

cause of death then you may argue that, okay, there

were more sudden deaths in Vioxx as compared to

placebo maybe but all together they looked about

the same.

So, this is what we had up till March.

Actually, when the APPROVe study was presented to

255

us this application was still under review. It is

still under review because we had requested

additional information so it takes time until it

comes to us and we can review that data again. So,

there are still many questions we have regarding

this database.

Let me show you the Kaplan-Meier curve

first. This is again the percent of patients with

events versus time. As you can see here, placebo

was about here up to almost 24 months and then they

completely overlap. But the confidence intervals

all along were very wide.

If you look at this table that I took from

the sponsor looking at relative risk over time,

again you see that after 18 months the risk was

higher on placebo as compared to Vioxx and after 18

months the risk switches and is higher on Vioxx

compared to placebo.

Again, if you look at adverse events with

an overall risk you have a number, but it is very

important to look at risk over time because down

here, after 36 months, it seems that Vioxx is

256

picking up. But, still, I mean the confidence

intervals are so wide we can't make any conclusion

out of this.

This is the total-cause mortality in the

Alzheimer's studies. As I think has been pointed

out before, there was a difference in total-cause

mortality in Vioxx versus placebo, but if you look

at the cause of death they were kind of not

clustered under one specific organ system. They

were all over. Also, this is the first time that

we had a placebo-controlled database of 3 years of

an NSAID or a COX-2 selective NSAID. So, it is

very hard to make any conclusion based on a

comparison of Vioxx with placebo when we do not

have any information on diclofenac or ibuprofen,

the same kind of data up to 3 years. Still, it is

of concern because, as I said, we are still

reviewing this application.

Then I want to mention the epidemiologic

studies because there were many epidemiologic

studies and re-analyses and meta-analyses.

Although I will mention that those meta-analyses

257

did not include substantial information that we had

access to. Unfortunately, we could not share that

with the world if they were not published in the

literature. But epidemiologic studies in general,

the ones looking at Vioxx and the ones looking at

naproxen--some of them were conflicting. What was

consistent was that there was increased

cardiovascular/thrombotic risk for Vioxx 50 and

that was in the label already. Actually, we have

said that for everyone with ischemic disease people

should be cautious. There was no clear evidence

with the 12.5 and 25 mg dose. Again, we had seen

the signal but as compared to naproxen, not to

placebo. And, there was conflicting evidence

regarding the cardioprotective effect of naproxen.

Out of 9 studies, 5 would say it is

cardioprotective and another 5 would say it is not,

or 4 would say it is not and 1 would say it

actually causes myocardial infarction.

So, I think that up to today I am not

clear as to what is the role of naproxen. I think

that it is possible, it is plausible that there is

258

a prothrombotic effect of Vioxx but that big effect

that we saw in VIGOR and in the other databases as

compared to naproxen--I think that naproxen does

have a role there too but that does not explain

everything, for sure.

In the meantime, during this time we were

awaiting the results of the long-term

placebo-controlled studies and then we had APPROVe.

This is data submitted to us in January, 2005. So,

I am not sure if they are exactly the same numbers

that the sponsor has shown because there was

another submission from October that is slightly

different. Anyway, the point is that here it is

very clear if you look at APTC events--for fatal MI

there was only 1. So, if you look at non-fatal MI,

there were 10 and 8. For ischemic stroke there is

also a signal, but not for hemorrhagic stroke. But

looking at cardiovascular deaths, there were 6 and

This is the time to event plot that you

already saw several times. I want to show you this

later. What I want to show you now is that up to

259

here what we had was a signal for Vioxx as compared

to naproxen. That is clear. But compared to

placebo, in the Alzheimer's data the only thing is

there was a trend for cardiovascular death. In

APPROVe it is completely opposite. You have a

negative effect on cardiovascular/thrombotic

events, non-fatal MIs, stroke, but not in

cardiovascular death. If you look throughout, this

is the first time where stroke appears as being a

problem with Vioxx.

This refers more to the second goal that I

had. Well, first of all, it was to show you that

you need to look at risk over time. The other

issue is what is the role of aspirin in these

studies in how it may affect different endpoints.

This is how it affects APTC endpoints. Don't even

look to the left side. There are too many numbers

here. The point is that the difference in

cardiovascular and thrombotic events or in APTC

events is driven by the non-aspirin users. In the

aspirin users the relative risk decreases,

particularly because there is an increase in the

260

patients in the comparator. I think that this has

to do with the kind of population that you want to

see in the studies. You would want to see patients

at high risk but not all patients at high risk

because I think that use of aspirin may make it

actually difficult to find a difference between

treatments. Anyway, we should have both high risk

an not high risk. These were not very high risk;

they were just patients that needed aspirin.

I am going to show you this slide just

quickly. I know that Dr. Temple is going to spend

more time talking about blood pressure. The

sponsor conducted several analyses of blood

pressure and I chose this one, which is a very

simplistic one but, still, I think it makes the

point that when you look at on-treatment

hypertension those who develop no hypertension

still had increased risk for Vioxx 25 compared to

placebo. The risk is very obvious here, that it

increases in patients with hypertension. This is

using the definition of patients who develop a

diastolic blood pressure of 100 or systolic blood

261

pressure of 160.

The point of this slide is that if we are

going to look at those patients with very high

blood pressure we are missing the boat here because

we need to look at those patients who have not as

bad hypertension. We need to look at those

patients who are within the range of 140/90 or

maybe even high normal blood pressure.

This is again a busy slide and I am not

going to walk through it, but just to make the

point that if you go through different databases

you have different numbers, all over, and the rate

of events in the Alzheimer's studies was higher,

particularly in placebo. Here in the Alzheimer's

study it was 2.07--I am sorry, I am going too fast.

Let me start again. You have VIGOR, Alzheimer's

database and APPROVe with Vioxx/naproxen;

Vioxx/placebo; Vioxx/placebo. Here with have APTC

events, myocardial infarction and total-cause

mortality. N is the number of events and this is

the patient-year rate in 100 patient-years of

exposure.

The point was that placebo here--the

patient-year rate is 2.07 while here it is 0.54 in

the APPROVe study. Naproxen here is in between in

262

the VIGOR study. The point is different

populations, different background rates in the

active treatment and also in the comparator

treatments. So, again, we need to define what kind

of population we want to have in these studies.

The other point is that if you look at

total-cause mortality, the one that looked bad was

Alzheimer's. There was no difference in

total-cause mortality in APPROVe. There was a mild

imbalance here in VIGOR and in the other databases

there was no difference in total-cause mortality.

So, I hope you understand how challenging

it was for us as we were reviewing this data.

There was a clear signal compared to naproxen that

was not consistent when compared to placebo. And,

we have no comparative data, particularly

cardiovascular safety data, for Vioxx and

non-naproxen NSAIDs or not a lot of data on

long-term placebo controlled with traditional

263

NSAIDs.

We still need to clarify the role of blood

pressure and what is the role of aspirin in

protecting for cardiovascular events. I think that

is it. I kind of said this while I was talking.

So, this is the end of my presentation.

Committee Questions to the Speakers

DR. WOOD: Thank you very much. Could you

go back three slides, and then I am sure Dr.

Fleming will want to ask you a question?

DR. VILLALBA: Which one?

DR. WOOD: The third last slide in the

handout. That one.

DR. VILLALBA: This one?

DR. WOOD: Yes. Am I right?

DR. FLEMING: You read my mind. I wanted

to follow-up on this because it is also a follow-up

to a question I asked this morning. Just to get a

sense of what the totality of the data is telling

us about whether there is an all-cause mortality

risk increase, and the two studies on the left

definitely strongly suggest that there is. In the

264

discussion this morning it was pointed out that

there are other sources of data that might

complicate the interpretation, the ADVANTAGE trial

being one of those. But if you look on sponsor

slide 54, which we won't go back to now, the other

studies are all very small relative to the numbers

of events. More than a half of the total deaths in

the meta-analysis of all the studies are from the

VIGOR study and the Alzheimer's studies and that is

where we are seeing the signal. The ADVANTAGE

study that we were told about that didn't show

significance still had one more death, and you said

in your presentation it was 4 versus 0 in the wrong

direction. There are 2 in the cardiovascular.

I guess my concern here is that when I

look at this it is on-drug, and I think it is

getting back to a question Ralph was asking earlier

today. All of these analyses, are we correct, are

only giving us that deaths that occurred

within--what?--30 days of being on drug?

DR. VILLALBA: Two weeks actually.

DR. D'AGOSTINO: Yes, I raised that

265

morning. I mean, why weren't these individuals

followed till the end of the study to find out

about mortality?

DR. VILLALBA: Well, actually that is a

good question to the sponsor because we know that

they were followed as much as they could do it, but

it was not mandatory. They tried to collect all

the data they could but it was actually--I would

prefer them to answer.

DR. WOOD: Well, let's not involve

motivation right now. Let's just keep going with

the facts. So, Tom, keep going.

DR. FLEMING: Well, that is the essence

that I wanted to get at. It was just to understand

that this is just on-drug and there is nothing else

you can provide us in terms of a true ITT? Is that

correct?

DR. VILLALBA: There were more deaths also

after but there was not a balanced exposure.

DR. WOOD: No, what he is asking is do you

have an intention-to-treat analysis?

DR. FLEMING: Correct.

DR. VILLALBA: No, I don't have it with

me. That is why I said this is still under review.

There is pending information.

266

DR. WOOD: But before we leave this slide

though, it is important to remember why we are

here. I mean, this is a drug whose indication is a

safety indication, and the reason to give the drug

was to reduce an adverse event which is always

thrown up as causing this terrible outcome,

although the outcome has improved substantially

over the last 10, 15 years.

It is certainly worrisome when a drug that

is supposed to produce a safety benefit, in fact,

is producing an increase in mortality, it seems to

me, and that is worthy of some discussion.

Certainly, an ITT analysis would have been

important.

DR. VILLALBA: Again, I completely agree.

We are concerned, but we don't know how other

NSAIDs would look here.

DR. WOOD: I understand.

DR. VILLALBA: We need to put it into

267

context.

DR. WOOD: That is what my teenaged kids

say as well. Curt?

DR. FURBERG: I was wondering whether you,

within the agency, considered the risk of heart

failure. I mean, when I look at the tables and in

your presentation you are using the term heart

arrest signal in a narrow sense. There is nothing

in your tables on heart failure. It is an issue.

As the Chairman found out a little bit earlier, in

the APPROVe study, a 4-fold increase in a long-term

trial. Do you have information from the Alzheimer

trials on heart failure? If you look at the

adverse effects of the drug, we shouldn't just

narrow it to heart attacks and stroke. Let's

broaden it to heart failure and make that part of

our evaluation.

DR. VILLALBA: Yes, I don't have slides

with me regarding congestive heart failure but,

again, we don't have the data for other NSAIDs.

That is the only thing that I can keep saying. But

there was more heart failure, for example, in VIGOR

268

clearly as compared to naproxen.

DR. WOOD: I sense that there is a

response coming from the sponsor. Do you want a

couple of minutes to think about that before you

get up? You can take a couple of minutes and we

will take another question, if you want. Take your

time; we won't forget you. Dr. Bathon?

DR. BATHON: I am a little confused about

the aspirin issue. On your slide 35 you showed a

decreased hazard ratio or relative risk for the

aspirin users compared to non-aspirin users. But

in Dr. Braunstein's presentation it was the

opposite. I realize that the outcomes were

measured a little bit differently.

DR. VILLALBA: That is a very good point.

These are APTC endpoints and the way that Dr.

Braunstein showed it was all

cardiovascular/thrombotic events that included also

peripheral events, unstable angina and TIA. So,

the point of this slide is precisely that when we

design a study that is going to address these

issues in the best possible way we need to choose

269

the right endpoint. And I don't know what that

endpoint is because if you look at all

cardiovascular events you may see more than if you

look only at APTC.

DR. WOOD: Dr. Shafer?

DR. SHAFER: I know it is always easier in

retrospect to try to make sense of things than

prospectively when you are looking at many possible

adverse outcomes and trying to figure out where to

focus one's attention. But if you could go back to

slide 23, what we see here, in slide 23, is a lot

of suggestions of danger signals. Dr. Braunstein

made an interesting point earlier when he said that

it would take about 30,000 patients to demonstrate

an increased risk, and yet we see danger signals in

very small studies of short duration. So, that has

obviously to be a cause for concern.

Then along comes the VIGOR trial. As I

understand, basically VIGOR had a 2-5X increase in

serious adverse cardiovascular events depending on

the endpoint you chose to look at. Now, there are

two possible interpretations of that. One

270

interpretation was that rofecoxib increased risk.

At the time you had this background worrisome

signal rate which was consistent with the

mechanisms that Dr. FitzGerald spoke about, and if

that were the true state of things, then

potentially millions of patients were being placed

at risk.

The converse choice is that Naprosyn

decreased risk. There were very weak data to

support that. As we heard from Dr. Nissen, the

effect was too large to be really explained by any

known effect of aspirin. And, the safety data that

were used to support the safety of rofecoxib was

far less than the 30,000 patients that would be

required to significantly show the difference. By

Dr. Braunstein's own statements, you know, it would

take far more patients to really statistically

significantly show that up.

What I first thought was the company and

the FDA chose to give pretty good credence to the

naproxen hypothesis. It sounds from the comments

today that that is still the position of Merck.

271

What would it have taken, what kind of data would

it have taken, given the results of the VIGOR trial

and the two alternative hypotheses, for the FDA at

that point in time to either put a black box

warning or perhaps even remove Vioxx from the

market? What kind of data would you have had to

have in addition to what you have?

DR. VILLALBA: I cannot answer that

question. What I can tell you is that this was as

compared to naproxen. We never bought the naproxen

theory, but we also did not have evidence that

Vioxx was worse than placebo or other NSAIDs.

DR. SHAFER: You have great evidence in

VIGOR though.

DR. VILLALBA: I completely agree but it

was naproxen, and I think the presentation tomorrow

with the epidemiologic data on naproxen will be

very informative about how confused we are until

today.

Regarding the signals, yes, those were

observed but that was after VIGOR, not before.

Again, we have that long-term, placebo-controlled

272

data in Alzheimer's patient elderly population that

had shown no difference in myocardial infarctions

or strokes. There was that signal of

cardiovascular death that, by the way, was put in

the label. But there were 8 versus 3 events and we

didn't know what to make of that.

DR. KONSTAM: Hi, there. I am Marv

Konstam. I am from Tufts University and I am here

with Merck as a consultant. In 2001 I was first

author on the overall pooled analysis for the

entire rofecoxib database so I just think I want to

speak to it, and the interpretation of VIGOR and

where the company I think was, and the world was,

at that point.

I think it is really difficult to look at

individual studies with very, very small numbers

and find signals, and one can draw all kinds of

conclusions from them; and there may be signals in

the other direction in some of the other small

studies.

So, what was done at that time was, you

know, there was a signal from the VIGOR study.

273

This finding was unexpected. It showed an adverse

effect on cardiovascular endpoints. Now, one thing

I want to stress about that is that of all of the

information that could be brought to bear, I think

the point estimate for the hazard ratio from that

is probably the least important to me. You know,

you are looking at very small numbers of events,

unexpected finding, wide confidence intervals. So,

I just want to point that out.

What was done at that point was that the

entire rofecoxib database to that point was

reviewed in a systematic way, and all of the data

were pooled. They were divided, as you heard,

between Naprosyn comparator, other NSAID comparator

but, most importantly, the placebo comparator.

Because VIGOR was an active controlled study and

none of us to this day know exactly to what extent

the result was contributed to by an adverse effect

of rofecoxib, a favorable effect of Naprosyn or a

combination. So, the most valuable data are the

placebo-controlled data. And, reviewing all of the

placebo-controlled data to that point, pooling all

274

of those data, there was 3000 patient-years of

follow-up, there was not a hint of an adverse

signal--not a hint of an adverse signal.

Now, granted, there were confidence

intervals around that signal so that is real. We

still didn't know, and I think we know a lot more

today thanks to the APPROVe study, but at that

point in time if you look at all of the

placebo-controlled data that existed there was not

a hint of a problem, which I think led me at that

time and I think led others at that time to say

this may be contributed to by a significant

beneficial effect of Naprosyn.

DR. WOOD: Just let me make sure I

understand. Are you saying that that is still your

position?

DR. KONSTAM: No, no. That was the

position at that time. One might then ask, okay,

what is different between the APPROVe data, and I

might say that I was on the data safety monitoring

board for APPROVe, and why is APPROVe different

than the pooled placebo-controlled at that time? I

275

think that is a really cogent question to ask and I

have asked myself that question.

I believe the difference now, in

retrospect, is exposure time. From APPROVe we see

no evidence of a hazard in the thrombotic events

through 18 months and then there is a separation.

The median follow-up in that pooled analysis that I

just referred to is relatively short. I don't know

what it was exactly but it was months. It

certainly wasn't the 9 months that was there in the

VIGOR study or the 2.4 years in the APPROVe study.

So, that is a substantial difference. There are

other differences, but to me that may be the

explanation for why the pooled analysis, back in

2001 and as it went forward, showed no problem but

APPROVe then came and did show a problem. I think

it probably was the exposure time.

DR. WOOD: But just to be absolutely

clear, you are not saying that you still believe

the VIGOR study was due to a totally protective

effect of naproxen, are you?

DR. KONSTAM: No, no, I am not.

DR. WOOD: Good. I just wanted to be

clear on that.

DR. SHAFER: While you are there, Dr.

276

Konstam, in terms of the relative risks of the two

possible choices--either rofecoxib increases risk

or Naprosyn decreases risk--was that part of your

thinking as well? What are the possible outcomes

of the two competing hypotheses? The truth is

probably somewhere in between.

DR. KONSTAM: Well, first of all, let me

just add one other point that I should have

mentioned. The other point about the VIGOR study

was the dose. So, there was a very high dose used

in VIGOR and there were lower doses in the pooled

analysis. APPROVe was 25 mg; an intermediate dose.

What was your question again? I am sorry.

DR. SHAFER: There are somewhat different

potential concerns with the conclusion that

rofecoxib increases risk as opposed to the

conclusion that Naprosyn decreases risk. Was that

part of your decision analysis at the time?

DR. KONSTAM: Yes, thinking back at that

277

time, there was no adverse signal from the

placebo-controlled data. I don't think, you know,

most people were completely satisfied with that.

If you look back at what the company did at that

point, first of all, there was a warning put on the

label and we can argue whether that was good enough

or not. But then we embarked on a large

placebo-controlled program with a prespecified

adjudication process for cardiovascular events, and

that is the process that led to the definitive

finding of APPROVe, even with a much smaller N than

they were planning to do so they had a much larger

program planned and we decided to stop APPROVe

because we saw it in APPROVe.

DR. WOOD: Let's go on. Dr. Nissen?

DR. DOMANSKI: We have heard a lot of

discussion about who know what, when, and we have

seen a tremendous amount of data presented, and in

the end this committee is going to have to make

some recommendation about what to do going forward.

I am very interested, if we could, in hearing from

each of the pharmaceutical manufacturers, as well

278

as everybody else of course, before they sort of go

away into the distance. I am very interested,

given the totality of data that are currently

available--not what you knew when or who should

have known what, how or when or who should have

done something else--I am very interested in what

you think ought to be done now going forward,

knowing what we know. What recommendation would

you make? What would you like to see come out of

this? Or, maybe what do you think we should see

come out of this?

DR. WOOD: Dr. Nissen?

DR. NISSEN: Yes, a quick comment and a

question. The comment is--and I think for people

in the audience who may not fully understand why we

are drilling down on this intent-to-treat aspect of

the analysis--that it may be that the individuals

who are dropping out of these trials because of

adverse events, that received the COX-2 inhibitor,

they may be pharmacogenomically more susceptible to

the adverse effects of COX-2 inhibitors. So, you

are taking out of the trial the people that are at

279

greatest risk. If you don't follow those people

you may not find that out.

This idea of censoring events after two

weeks--you know, I think we have to all learn

something from what happened here, and this is the

first time I really realized that that was the way

these studies were conducted. That was a mistake.

Once a patient is exposed to drug you ought to

follow him as long as you can because there may be

a persistence of risk and we learn something from

that. So, a lesson is learned. I think it is a

useful lesson to learn.

I guess the second question--and, you

know, you may or may not want to answer this but if

you had to do it all over again would you do it

differently?

DR. WOOD: Let's keep the tense in the

future tense. Let's not keep regurgitating that.

Bob, do you want to say something in the future

tense?

DR. TEMPLE: Yes. I just want to remind

people that intent-to-treat analyses are generally

280

loved by people because they are conservative

analyses. They ten to make effects go away. That

is why we like them. If you are worried about

informative censoring and other stuff like that--

DR. WOOD: But it tends to make efficacy

effects go away.

DR. TEMPLE: That is correct. They also

make time effects go away.

DR. WOOD: Not if you are dead.

(Laughter)

DR. TEMPLE: No, no, you have to count the

deaths. It is not that you shouldn't follow people

up but the analysis that includes all people long

after they are off the drug has a very high

likelihood, I believe, or not showing the effect of

the drug. You have to remember it is a

conservative analysis for looking for effects.

Before we get too enthusiastic about it, if I make

the effect look less when it really doesn't deserve

to look less--

DR. FLEMING: Could I just quickly add to

that? Historically we look at safety and often we

281

do truncate follow-up after two weeks or a month.

That is based on the premise that safety risks are

acute. If they are, in fact, acute, then you are

going to get a clear sense of what is going on with

the type of approach you are talking about.

Mortality effects, I would think, are much more

difficult to justify as being purely acute. There

is a basis to what you are saying. If you follow

everybody for a long time after they are off

therapy there could be some diluting.

Nevertheless, if you want an unbiased assessment of

the truth you need to do what Steve is talking

about, an ITT analysis, and then make your judgment

as you look at the hazard ratio over time.

DR. D'AGOSTINO: We are sitting here and

we don't know the answer. It may have washed it

away and it may not have.

DR. TEMPLE: I am not saying don't get the

analysis but, for example, our ordinary position in

an outcome study is that we want to see the

intent-to-treat analysis.

DR. WOOD: Let's hold this for the

282

discussion. Let's just keep focused on the

questions right now. Any further questions for the

speaker? I am not forgetting about you. Hang on

just a moment. Dr. Holmboe?

DR. HOLMBOE: I just had a question to the

speaker. Again, we are trying to give you some

advice and some guidance as to this. Given, as

Alastair said earlier, that this drug was really

evolved for a safety indication, therefore, being

compared to another class of drugs, in retrospect

learning that those comparisons were based on drugs

approved prior to new knowledge that has been

accumulated, such as presented by Dr. FitzGerald,

it would be helpful for me to hear what has the FDA

learned about the process or form? What can you

tell us that might help in the future when you are

faced with these sorts of things? For example, the

diclofenac is a perfect example, well, it turns out

that maybe it is not, you know, your

run-of-the-mill NSAID. A lot of what you presented

in the original data was, like, well, it was

between ibuprofen and diclofenac, therefore, we

283

determined it was probably okay. So, I would be

anxious to hear what you have learned since you

have been with this project now for seven years.

DR. VILLALBA: Well, actually we wanted to

have the recommendation from you to know how to

proceed now because we have close to 20 approved

NSAIDs so what do we do with them?

DR. WOOD: Ever the optimist, right! Dr.

Domanski?

DR. DOMANSKI: I guess before Merck gets

away I would still like to hear their view of where

we should go from here. I am really quite curious

about that. I understand about intention-to-treat.

We do clinical trials. But I would just like to

hear what their thoughts are.

DR. WOOD: Their thoughts on what?

DR. DOMANSKI: What their thoughts are on

where we should go from here.

DR. WOOD: I thought we were talking about

where we have been. I am happy to hear them on

where they should go. Do you have thoughts on

that, Bob?

DR. DOMANSKI: No, I am asking that of

Merck.

DR. WOOD: Oh, I am sorry.

284

DR. BRAUNSTEIN: I think we showed that on

our last slide. Can I see our last slide, 57? I

mean, for the short term what we are trying to do

is trying to better understand our data; trying to

better understand which patients were at increased

risk for the events that we observed in APPROVe

based on both the clinical data and also the

specimens that we have from these patients. We

also are working with various people to try and

explore different hypotheses for the data, and we

are collaborating with others who are looking at

the data across all the drugs in order to get a

better feel to see if we can understand when we

pool all the data because I don't think any one

data set that we have is powerful enough to address

these questions. So, hopefully, by pooling the

data we will be able to get a better feel for this.

The last is that we think we need to do comparative

outcome studies to better understand the relative

285

risks of the selective COX-2 agents with the

traditional NSAIDs. There are not long-term data

on the traditional NSAIDs to really establish what

their cardiovascular risk profile is, and we think

that the study that we are doing, for example the

MEDAL study is one such study in the right

direction.

DR. WOOD: Merck wanted to present some

other data. Right?

DR. REICIN: I think there was a question

about congestive heart failure in the Alzheimer

studies.

DR. WOOD: Right.

DR. REICIN: So, just put up slide for us

12-22 and then we will go to 12-28. I showed you

this slide just at the beginning and you noted that

in our 6-month population--so this is a shorter

population than either APPROVe or what I am going

to show you in Alzheimer's--the rates were quite

low and they were similar to the NSAIDs.

If you go now to 12-28, in the Alzheimer's

studies, in protocol 078 which was a 4-year study,

286

interestingly, the rate of congestive heart failure

was similar between the two groups, 2.2 percent on

rofecoxib 25 mg, 2.6 percent on placebo. In 091,

however, which was a one-year study the rate was a

little bit higher on rofecoxib, 3.2 percent versus

1.4 percent. I think these rates are more what you

would expect in an elderly population. The mean

age of this patient population was 75 years old.

DR. WOOD: Thanks.

DR. REICIN: One other thing, there was a

question about ITT mortality. In APPROVe we are

following patients in an ITT way for mortality.

That is still ongoing. To date, there were 3

thrombotic events in each treatment group following

that 14-day period.

DR. WOOD: Dr. Paganini?

DR. PAGANINI: I have a question on the

comparative data with other NSAIDs. Is there not a

post-approval period of time for drug review, and

from that post-approval Phase IV type studies can

you not draw anything from that to compare to?

DR. VILLALBA: Phase IV commitments are

287

made at the time of approval. If there were not

specific agreements between the FDA and the company

to conduct those studies we have no legal power to

mandate any kind of studies. So, some studies are

done basically pursuing different--I mean with

promotional, advertisement or whatever there are

many studies. But those are really not usually

large outcome studies. They are short studies with

small numbers of patients. I don't know if I

answered your question.

DR. PAGANINI: You did in a way. One of

the issues that I think we are going to have to

face is how do you compare these things, both

things that have already been approved and new, to

the same standards when they were approved back

then to current standards? Perhaps one of the ways

around that might be an approval comparison with

longer Phase IV commitments by companies to

follow-up on what is happening to that drug over

time. That way, you would have the ability to

compare a new to a similar in a similar population

of patients.

DR. VILLALBA: Absolutely. That is

something that we learned, yes.

DR. WOOD: Ralph?

288

DR. D'AGOSTINO: I am all for torturing

data and during Lent I always read Dante's

"Inferno."

(Laughter)

But shouldn't we be impressed with the

APPROVe study? You leave us with a table that

compares a lot of studies and you throw out some

obviously important questions, but shouldn't we

sort of look very seriously at the APPROVe study?

It was well designed--

DR. VILLALBA: Of course.

DR. D'AGOSTINO: --and shouldn't we sort

of diminish in our view some of the previous

studies?

DR. VILLALBA: The Alzheimer's studies, do

you mean? Now, yes. What I was saying is that

these are different populations and I do not have a

good explanation for why we didn't see the same in

an elderly population.

DR. D'AGOSTINO: Well, could it be that

the APPROVe study was going after a particular set

of outcomes and the others weren't, and it was more

retrospective?

DR. VILLALBA: No, because in the

Alzheimer's studies they also used the same

289

standard operating procedures to adjudicate the

event.

DR. D'AGOSTINO: But do they have the same

ascertainment? You know, in designing a

placebo-controlled study where you go after

something retrospectively, looking at that and

trying to say the ascertainment might have been the

same.

DR. VILLALBA: You are completely right.

That is possible but that is a question to the

sponsor, if the ascertainment could have been

different in the Alzheimer's studies.

DR. WOOD: Dr. Hennekens? Actually, I

would like to ask Marvin a question. Marvin, the

APPROVe study was scheduled to terminate at about 6

weeks after the early termination on the basis of

290

the board's recommendation that you were on. As I

recall, the numbers of events were 45 and 25 at

that time. So, was the board unanimous in its

decision to terminate, and was the basis clearly

related to that particular endpoint?

DR. KONSTAM: Yes. Yes, that is exactly

right. I would say that the reason, if I might say

why we recommended termination--the reason we

recommended termination is that we felt at that

point in time that we had a definitive piece of

information that wasn't going to change. The

reason we recommended termination was that we felt

the patients in the APPROVe study were not aware of

this and had not been consented to this adverse

effect. So, in our judgment, you know, from an

ethical viewpoint if you were going to continue you

would have to go back and re-consent them and that

certainly wasn't practical at that point in time.

So, that is the specific reason we recommended

termination.

DR. WOOD: But you told them that caution

should be exercised in patients with heart failure.

291

Right?

DR. VILLALBA: May I say something?

DR. WOOD: Sure.

DR. VILLALBA: I don't want to leave you

with the impression that we think or I think that

APPROVe is not important. I just want to show you

how puzzled we were with all the data. So, until

APPROVe we didn't have a firm reason to really take

a regulatory action that was different from what we

had done up to that time.

DR. WOOD: Ralph again?

DR. D'AGOSTINO: In terms of the

Alzheimer's study, do you have information on the

all-cause mortality? I forget what you said. Do

you have anything about CVD, cardiovascular

mortality when off drugs?

DR. WOOD: Let's take that under

advisement unless you have it right there. Do you?

No? All right, we will get back to that. Any