1

DEPARTMENT OF HEALTH AND HUMAN SERVICES

FOOD AND DRUG ADMINISTRATION

CENTER FOR DRUG EVALUATION AND RESEARCH

ADVISORY COMMITTEE FOR PHARMACEUTICAL SCIENCE

CLINICAL PHARMACOLOGY SUBCOMMITTEE

Thursday, November 4, 2004

8:05 a.m.

Hilton Washington, D.C. North

620 Perry Parkway

Gaithersburg, Maryland

C O N T E N T S

PAGE

Call to Order 3

Conflict of Interest Statement 3

Tribute to Dr. Lewis Sheiner 5

Introduction to the Topic, Background

and Project Plan 20

Framing the Issues: What Needs to be

Done and How? 32

What are Industry's Expectations of the

Project and Process? 67

Opportunities, Challenges and Some Ways

Forward: How can Academia-Industry-

Government Collaborations Facilitate the

Development of Biomarkers and Surrogates? 90

Committee Discussions and Recommendations 116

Summary of Recommendations 146

P R O C E E D I N G S

[8:03 a.m.]

DR. VENITZ: Good morning, everyone. For

the second day of the Clinical Pharmacology

Subcommittee meeting, we have half a day agenda for

today. And I would like to point out that we don't

have anybody signed up right now for the open

hearing. If anyone in the audience wants to do

that, please contact Ms. Scharen as soon as

possible so we can lock you in.

The first order of business is to review

the conflict of interest, and Ms. Scharen is going

to do that for us.

MS. SCHAREN: Good morning.

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

meetings are issues of broad applicability, and

there are no products being approved.

Unlike issues before a subcommittee in

which a particular product is discussed, issues of

broader applicability involve 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 interest

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 matter waivers

to the special government employees participating

in this 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 12A30 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 this subcommittee, these

potential conflicts are mitigated.

With respect to FDA's invited industry

representative, we would like to disclose that Dr.

Paul Fachler and Mr. Gerald Migliaccio are

participating in this meeting as nonvoting industry

representatives acting on behalf of regulated

industry. Dr. Fachler's and Migliaccio's role at

this meeting is to represent industry interests in

general and not any one particular company. Dr.

Fachler is employed by Teva Pharmaceuticals, USA,

and Mr. Migliaccio is employed by Pfizer.

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 participants' involvement and their

exclusion will be noted for the record. With

respect to all other participants, we ask in the

interests 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. VENITZ: Thank you, Hilda.

Before we proceed with the scientific

agenda, we will pay a tribute to one of the seminal

members of this Committee, who passed away earlier

this year, Dr. Lew Sheiner, and Dr. Lesko and Dr.

Blaschke will pay tribute to his contributions in

clinical pharmacology.

DR. LESKO: Thank you and good morning,

everyone. Welcome back. We had a long day

yesterday filled with a lot of heavy duty

intellectual discussions, and it's nice to see you

all back and I think refreshed.

Anyway, we would like to pause at this

moment and remember our colleague, Dr. Lewis

Sheiner, who was what I would call a founding

member of the Clinical Pharmacology Subcommittee.

I remember inviting him to join the Committee a

couple of years ago, and he said to me I'll only

come if it's going to be intellectually

stimulating. And after each meeting, I would ask

him was that intellectually stimulating? And he

would say yes, and he came back to every meeting.

Dr. Sheiner, as everyone knows, and Jurgen

mentioned, passed away unexpectedly in April of

this year, and Lewis, we all know, was many things

to many people. He had an important role as a

member of the CPSC. He provided us with an

extraordinary dimension of opinions on many

different subject matters, always challenging us to

dig deeper into our intellect.

He was great as a member of this

Committee. He focused on solutions, and he didn't

dwell on the problems very much. I remember last

November, and many of you do, too; we were

discussing the end of phase two-A meeting, and I

think we spent about three or four hours of

discussion, and I still remember his question,

which came at the end of that discussion, and I

think it exemplified his way of spicing up a

Committee meeting. He said Larry, it sounds like a

good idea somehow, but I'm not sure exactly why.

I think that was his way of challenging us

to think clearly and fully about what we were

proposing at this meeting. And I think the topic

that we will discuss later this morning would have

been very near and dear to his heart. So I know

that I speak for many of you, members and audience

alike, all of us at FDA, when I say that it would

be an understatement of the highest proportion to

state that Lewis is sorely missed today.

I have invited Dr. Terry Blaschke, who was

a close friend and colleague of Dr. Sheiner to pay

him a tribute on all of our behalf.

DR. BLASCHKE: Well, thanks, Larry. This

actually is a harder talk to give than the one I'm

going to give later this morning.

Larry did ask me to pay a tribute to

Lewis, and I think we really did lose a visionary

leader in drug development in April. Lewis died

shortly after receiving the Oscar B. Hunter Award

of the American Society for Clinical Pharmacology

and Therapeutics, which is really one of the

premier awards in clinical pharmacology, and I

think Lewis was very pleased to get that award. I

had the pleasure of introducing him for that award.

Many of the people, of course, in this

room, not just on the Committee but in the

audience, knew Lewis and had an opportunity to

interact with him, and I think if you had that, you

really knew what a wonderful person, enthusiastic

and exciting as Larry has just expressed.

But one of the things that he really did

want to do and did do, I think, not only in this

Committee but elsewhere was really get involved in

improving the process of drug development. And one

of the things I'd like to do during the next few

minutes is really talk about some of those concepts

that he championed and I think have become very

important in the whole field of clinical

pharmacology and drug development.

But I'll start out with a little bit of a

background about Lewis, for those of you who don't

perhaps know some of his background. He was born

in New York City, and in fact, it took many years

for him to evolve his California-like approach to

discussions like this. Those of you who knew him

early in his career probably remember that he could

be pretty acerbic as a critic of presentations and

so forth, and certainly, as he grew older, he

became much more of a mellow individual when it

came to his discussions.

Lewis received his bachelor's degree from

Cornell University, his medical degree from Albert

Einstein. He was then an intern and a first-year

resident at Columbia Presbyterian Hospital in New

York City. He then, as many of us did in that era,

go to the NIH, where he was a research associate at

the National Institute of Mental Health.

There, Lewis actually published two papers

in the Journal of Biological Chemistry, and I think

but for a change that I'll tell you about in a

moment, he might have been a molecular biologist or

a molecular pharmacologist. He had planned to

return to Columbia University Medical Center to

finish his residency training and called down to

the chair of medicine when he was about to complete

his tour of duty down at the NIH and was told that

he should have called earlier; that basically, they

weren't ready to take him back.

So instead of returning to Columbia, he

joined the NIH Division of Computer Research and

Technology, where he, I think, had his first

exposure to computers in medicine and to modeling

and possibly a simulation at that time, but the SAM

program. This actually led to his first

publication, which had to do with the

computer-aided long-term anticoagulant therapy,

which was published in 1969 in Computers and

Biomedical Research.

After completing that additional two years

at the NIH, Lewis came to Stanford, where he

completed his medical residency and then went to

UCSF as a clinical pharmacology fellow, joining the

faculty there in 1972, and spending the rest of his

career there, where he was professor of laboratory

medicine and biopharmaceutical sciences.

Of course, Lewis is widely recognized as a

pioneer in the field of pharmacometrics, and his

career at UCSF really focused on the mathematical

and statistical methods applied to the problems of

clinical pharmacology. During the early part of

his career, Lewis was involved in the whole area of

therapeutic drug monitoring, which was then

becoming established at many hospitals through the

country.

Through Ken Melman, Lewis was introduced

to Bar Rosenberg, a brilliant statistician at

Berkeley, and this really represented another

pivotal point in Lewis' career and really marking

his entrance into the field of the world of

statistics. And this particular paper, again,

published in 1972 in Computers in Biomedical

Research, represented this first paper, actually, I

think it was the second paper along with Bar

Rosenberg in which the focus on individual

pharmacokinetics and computer-aided drug dosing was

first published.

Now, this introduction to Bar and interest

in computer-aided modeling of drug therapy led to

this paper, actually, two papers: a paper

published in 1973 in the New England Journal of

Medicine on computer-assisted digoxin therapy and

then this paper with our colleague, Carl Peck,

Lewis Sheiner, Bar Rosenberg and Ken Melman again

that appeared in the Annals of Internal Medicine.

This work really, I think, led, as it

inevitably would, to Lewis' interest in developing

methods for predicting pharmacokinetics of drugs in

individuals using sparse data sets; in other words,

using just a few drug concentrations obtained

during the patient's hospital stay, and I think as

a result of that, together with his colleague

Stewart Beal, Lewis developed and applied the

NONMEM program, which I think is probably most

associated with Lewis' work, and I think most of

you are familiar with NONMEM as a Bayesian

forecasting tool incorporating population

pharmacokinetic information to predict

pharmacokinetics.

This novel program and novel approach has

really led to greatly-enhanced predictions for

dosing regimens, for patients in clinical settings

allowing for individualization of drug therapy and,

of course, I think NONMEM really became the

standard in the industry and at the FDA for

characterizing population pharmacokinetic data

acquired during clinical drug studies, and, in

fact, I think really greatly expanded the entire

field of population PK over the last decade or two.

Lewis then moved from forecasting of

pharmacokinetics to, I think, another very

important area, again, with our colleague, Don

Stanski, in thinking about pharmacokinetic and

pharmacodynamic modeling. Lewis had a very keen

sense of clinical pharmacology, and he really

pioneered these new methods to simultaneously

analyze pharmacokinetic and pharmacodynamic data,

leading to the concept of the effect compartment.

I'm showing that basically with this slide.

This, I think, is the typical slide that

one would see in many different presentations, both

of Lewis and others. This has really become, I

think, the way in which PK/PD data is handled by

many individuals. As with NONMEM, this worked with

his pharmacodynamic PK/PD modeling that has really

become a standard both for industry and for the FDA

in analyzing drug response data.

Lewis' overall goal all along was to

improve patient care by individualization of dosing

regimens. And the work that he did really enabled

this to be done in a number of different

therapeutic areas. Lewis worked, as many of you

know, with anesthetic and analgesic drugs, much of

which was done in collaboration with Don and Don's

colleagues; worked with me and many others in

antiretroviral therapy and antiretroviral drugs and

in many other therapeutic areas with many

collaborators.

As I mentioned at the beginning, much of

Lewis' work was really focused on improving the

science of drug development by optimizing clinical

trial designs, and his vision was to develop

methods that allowed more efficient and more

informative clinical trials, optimizing dosage

recommendations and optimizing therapy. And one of

the things which he did, again, with his colleague

Nick Holford was, again, really to focus on

understanding the dose-effect relationship and

along, again, with Stewart Beal and Nancy Samble of

UCSF, I think this was one of the classic papers of

study designs that could be used for dose ranging,

particularly in phase two studies, and I've seen

this particular study quoted many times at meetings

and in the literature.

And Lewis would always say that this was

one of his signature slides. If you didn't see

this slide, you knew it wasn't Lewis talking. This

was his whole concept of a response surface, with

benefit-risk response surface, and he had many

variants of this slide, but this, I think was one

of his, as he said, signature slides and favorite

slides.

Now, Lewis really, as I mentioned at the

beginning, developed an intense interest in

statistics. And this led him, really, to question

the traditional approaches to data analysis in

clinical trials and this whole concept of--did I

pass one slide here?--well, I'll come to that

slide. This is a little bit out of order. But in

any event, he really got very interested in looking

at the whole issue of statistical approaches to

analysis of clinical trials, and this review that

was written just a couple of years ago in the

British Journal of Clinical Pharmacology was one

example; another example was this paper written by

Nicholas Johnson and Lewis just a couple of years

ago in Clinical Pharmacology and Therapeutics, and

he had begun to work very closely with a number of

statisticians, including Marie in the audience here

and other statisticians at Harvard really asking

questions about the analysis of clinical trials.

Now, I think perhaps his most important

contribution overall was his paper published in

1997 on the concept of the learn-confirm paradigm

of drug development. And I've heard this

particular paper and this particular concept quoted

again and again as I've talked with people in the

pharmaceutical industry and so forth, and I think

this really does represent a major contribution

that Lewis made to the whole thinking of how one

develops drugs, and I'm going to come back to that;

I won't talk much about that right now, but I'm

actually going to come back to that later on this

morning in my own presentation.

Lewis was obviously very interested in the

whole area of drug development and in the role of

pharmacokinetic and pharmacodynamic modeling in

drug development and published this review in 2000

in the Annual Review of Pharmacology and

Toxicology, which I think was--again, it's a

highly-cited paper, one that really gives an

excellent overview along with Jean-Louis Stymer, of

the role of modeling in the whole drug development

process.

Now, I'll mention to go on a little bit

about Lewis' specific service on FDA advisory

committees and committees such as this one. Since

1987, Lewis had been an expert consultant to the

FDA Center for Drug Evaluation and Research and had

participated in many meetings. He was, and this

will become important later on again, a member of

the Anti-Viral Drugs Advisory Committee from 1991

to 1994, and as you'll see in my presentation

later, this was a very critical time in that field

of antiretroviral drugs.

He was very involved in the whole area of

bioequivalents and was a member of an expert panel

on the guidance in population PK/PD as well as this

expert panel on individual and population

bioequivalents at CDER. As well, he was a member,

as one might expect, of the exposure response

guidance panel of CDER, and finally, as Larry has

already mentioned, a member of the Clinical

Pharmacology Subcommittee, in fact, a founding

member of the Clinical Pharmacology Subcommittee.

Lewis' substantial influence on the

science of drug development has, I think, been very

well apparent and documented, and those of us who

knew him will remember him for his passion for this

whole subject, his intellectual curiosity, as Larry

has mentioned; his warmth and engaging personality.

He had a great impact on the people he trained and

the people he collaborated with, even those of us

or those of you who had more limited interactions

with him.

He really established deep and lasting

relationships with his fellows, friends and a broad

spectrum of scientific and business associates. He

spawned several generations of

quantitatively-oriented clinical pharmacologists

worldwide, not only through his research but also

for his commitment to research and training, which

included a number of, I think, world-renowned

courses in pharmacokinetics and in NONMEM and

modeling, working in many cases with his friend

Malcolm Rowland and his colleague, Les Bennett, at

UCSF.

This is just a list of the many people

that Lewis trained. You can glance up at this

list. You probably see many people that you know

on this list, people who are very influential and

very important in the field of drug, clinical

pharmacology and drug development. This picture

was taken in 1992 at a 60th birthday celebration

that was held for Lewis. You see him down there in

the lower left-hand part of the slide. There were

probably about 100 people. Kathy was very

responsible for helping organize this meeting,

Kathy and Les Bennett, and I think it really

represents the kind of loyalty and so forth that

Lewis was able to generate.

Lewis served as president of the American

Society for Clinical Pharmacology and Therapeutics.

He authored more than 200 books and chapters; was

on the list of most-cited authors in the area of

pharmacology through ISI; had many honors and

awards, including an honorary doctorate from

Uppsala University; the Hunter Award that I

mentioned, the Rawls Palmer Award that I mentioned

from ASCPT and an honorary fellowship from the

American College of Clinical Pharmacology.

Lewis lectured widely throughout the world

as well as being involved in committees such as

this one, and as Larry said, he certainly will be

sorely missed. And I thought these two final

pictures of Lewis really represented Lewis at his

best: one in Amsterdam and one in Switzerland.

Thanks.

[Applause.]

DR. VENITZ: Thank you, Dr. Blaschke.

Our first agenda item as far as the

scientific agenda is concerned is a discussion of

surrogate markers, and Dr. Lesko is going to

introduce the topic.

DR. LESKO: Thank you, Terry, very much

for the thoughtful comments, and I'm sure Lewis is

looking down smiling and saying I told you so.

I'm here at this point to introduce the

last topic of this meeting, which we call the

transition of biomarkers to surrogate endpoints.

It's somewhat of a difficult introduction to make

because of the broad nature of biomarkers and

because of what's gone before, namely, a large

number of discussions, many of them passionate,

about the topic of biomarkers and surrogate

endpoints.

My colleague, Don Stanski, urged me to be

visionary, and being visionary is not something

that comes naturally to me, so it's difficult to be

visionary. So I looked for inspiration. And I

looked for inspiration to the movie that I was

watching on Sunday with my grandson, Nemo, and

there's a point in the movie where these two fish,

who you probably recognize, come around the corner

of a coral reef and come face to face with a

menacing shark, and they say something like oh, no,

not him again.

And I thought about that, and I called

this the biomarker fear factor, because we've

talked about biomarkers endlessly for the last 10

or 12 years, and one might be apt to say oh, no,

not that again.

We've talked over the years in workshops

and symposia on the validation of biomarkers as

surrogate endpoints, and again, this is a topic

that ignites a lot of discussion and a lot of

debate, very much passionate debate, with the sides

taking shape.

I happened to look in the Internet, using

Google as a search, and I said I wonder what's

going on in biomarker workshops these days. And I

was able to pull up without a lot of trouble

biomarker symposia that are taking place all over

the world, from France to the Netherlands to South

America, and including Baltimore this weekend,

where there's a biomarker workshop that precedes

the ACPS meeting.

So a lot has gone before, and I'd like to

begin with definitions. These are definitions that

came from the FDA/NIH 1999 workshop, and you'll

probably see these occasionally throughout our

morning just to set the stage as to what we're

talking about in biomarkers and biological markers

and surrogate endpoints, and you can see that we're

talking about characteristics that are measured or

evaluated as indicators of a whole variety of

things, from normal disease processes to

pharmacological responses to drugs. And a

surrogate endpoint is a subset of biomarkers that's

intended to substitute for a clinical endpoint.

The problem that we have, I believe, with

biomarkers is that the pace of biomarker discovery

keeps increasing at a remarkable pace, with

measurable improvements In the biomarker discovery

area but not necessarily measurable improvements in

predicting the success of drug development. There

was an article yesterday in the New England Journal

of Medicine about the genetic basis for Parkinson's

Disease, and this type of discovery is so

ubiquitous these days that the genetic basis of

this disease or that disease is sure to spur the

discovery of biological markers that are going to

play a major role in drug development and in

patient monitoring.

But the past focus of biomarkers and maybe

even the emphasis or overemphasis has been on

biomarkers as surrogates, and despite the last 14

or 15 years of debate and discussion, there have

been relatively few successes of biomarkers being,

quote, validated as surrogate endpoints. We've had

discussions of conditions that favor or not favor

surrogate status for biomarker endpoint, things

like the pathophysiology characteristics. We

discussed these in our exposure response guidance

that came out in April of 2003, and if you go back

and read that now, it is not very explicit on

either how you develop a surrogate endpoint or what

the criteria is to specify one as such.

There's been a subtle resistance, I think,

stemming from the past failures of biomarkers as

surrogate endpoints to consider their development

further. And in some ways, there's been a

paralysis in development of this field related to

the statistical rigor that's been associated with

the biomarker to surrogate pathway.

Furthermore, much of the discussion of

surrogates has been fragmented into individual

therapeutic areas as opposed to an integrated

overview of the entire process. And finally,

there's been many workshops that I think have set

unreasonable expectations for biomarkers and

surrogates.

But putting surrogates aside, I think we

need to refocus again and enhance the integration

and use of biomarkers over the entire course of

drug development as a natural path to the surrogate

endpoint goal.

So with biomarkers, I think a lot has

happened, but it does raise the question about how

things can be improved. For example, have we been

settling for less in the biomarker area? We think

biomarkers are extremely relevant to efficacy and

safety, aside from them being surrogates or not.

We don't need surrogate markers to gain the full

impact of biomarkers. Just in the past couple of

months, we've had many examples of this, and only

using one of those, the Iressa story. EGFR

mutations and tumor tissues have been reported to

predict a response in eight of nine so-called

responders.

Another question is can we more fully work

up biomarkers from discovery to clinical outcomes

than we currently do? One of the goals of

biomarker development is to begin to reduce, over

the course of time from discovery to clinical

outcome, the uncertainty in what I'll call that

gray zone between preclinical biomarker discovery

and phase three clinical outcomes. By bridging

those two areas, by bridging them in a clinical

pharmacology/biostatistical context, it would seem

that the process would more naturally lead to

acceptable surrogate endpoints, instead of thinking

of it as a one-step process of going from a

biomarker to a surrogate endpoint.

You're all familiar, I believe, with the

critical path. It's a call to action. The

critical path calls for a collaboration between

academic, industry, patient groups to work with FDA

to help identify opportunities, to modernize the

tools for speeding and making drug development more

efficient and more successful.

The biomarker vision is expressed in that

document. It talks about adopting a new biomarker

or surrogate endpoint for effectiveness that can

drive clinical development, and it gives an example

of the well-known case of CD4 and viral load that

were used as surrogate markers for anti-HIV drug

approvals in the early nineties and from that point

forward.

It talks about the biomarker challenge:

additional biomarkers, which we can think of as

quantitative measures of biological effects that

really link mechanism of action, i.e., preclinical

biomarkers and clinical effectiveness or outcomes,

and additional surrogate endpoints are needed to

guide product development.

So the document, I think, has laid out the

problem. It's laid out a vision. It's laid out a

challenge. And the question that we're here to

sort of begin to discuss is what do we do next.

And what we do next is very important, I think. We

need a new construct. We need to break the pattern

of the past. I think we need to go down a

different path, with two objectives in mind.

The first objective: can we achieve a

general, agreeable conceptual framework to

continuously reduce the uncertainty associated with

biomarkers over the course of the entire drug

development process: what is that systematic path?

Can we define it in a general way that is not

disease-specific, that is not biomarker-specific

but can be applied to many therapeutic areas?

We're seeing with genomics an increase in

disease progression knowledge. We're seeing that

there's benefit from systematically aggregating

knowledge using modeling and simulation,

quantitative methods. We've seen that there are

increasing ways of establishing the predictive

nature of biomarkers. We talked about some of that

yesterday when we visited the markers associated

with predicting irinotecan toxicity. And there's a

lot of initiatives that relate to the standards for

biomarker performance. So taken together, these

individual initiatives, I think, bode well for a

general conceptual framework.

The second goal of this initiative would

be to better articulate the standards or

specifications to validate and accept biomarkers

for their intended use, including surrogates for

registration and any extension of those surrogates

for additional applications, for example, in other

drug classes. So it's a twofold goal that I think

we want to strive for in the context of this

initiative.

Now, we're not starting from scratch with

this initiative. The agency has taken steps and

intends to take many steps that move us along this

path, and many of these are hinted at in the

critical path. We've already implemented the end

of phase 2-A meeting, and we plan to have a

guidance out in 2005. We've invested in resources

at the FDA and are developing a new branch of

pharmacometrics to focus on quantitative methods in

the IND period.

We've begun to develop drug-disease state

models, disease progression models in several

therapeutic areas. We've articulated, and Dr.

Stanski has articulated in front of the Science

Board, a very clear stepwise framework for

model-based drug development. We intend to conduct

an inventory of surrogate markers and look at the

evidence, whether it's epidemiological,

pathophysiology, therapeutic or other supporting

evidence, that allowed them to become surrogate

markers, so that we can learn from our current

situation.

We intend to establish an FDA working

group on this topic, with the goal of moving those

two objectives that I mentioned forward. The

working group itself will explore the development

of a potential guidance on biomarkers. And we've

initiated this discussion with the Clin Pharm

Subcommittee today.

The critical path document and some of the

presentations today will also reflect upon an

express goal to develop a new form of

FDA-industry-academic collaborations for critical

path opportunities, and some of these are being

discussed as we meet today.

From the industry side, steps taken or to

be taken, I can't really speak to that. But there

are many other examples of consortia of

collaborations that have been successful. And I'm

going to use one of them. There's another one I

could have used; it's in the current issue of

Nature Reviews Oncology that talks about a vision

for the development of biomarkers in oncology drug

development.

But this is one that comes from industry,

and it was provided to me by Chris Webster, who is

associated with the PhRMA Biomarker Working Group,

and it was very appealing as a model for a

consortium, and it's the Semiconductor Research

Corporation. Very briefly, this is a nonprofit,

precompetitive academic-industry-government

consortium, which is now about 20 years old.

You'll notice some parallels between this

and drug development. It was formed in the 1982

time period because of a concern about decline in

the semiconductor industry. It was geared towards,

as an industry, reliance on huge payoffs from

individual successes and isolated research across

the industry in individual companies. There was a

noted reduction in R&D funding with a limited

success in new semiconductor technology and a shift

towards short-term R&D as opposed to an investment

in long-term successes. There was a talent crisis

at the time, and there were many different

technology challenges.

The consortium came together, with

industry, academia and government, to really lead

the industry's long-term research efforts, advance

problem solving technology, integrate university

research capability across the country and now

internationally and serve as a hub, as a catalyst

for a large global network of collaborative sites

that were charged with developing technology that

would enable the semiconductor industry. They

developed a central vision and implemented an

action plan.

It wouldn't take a lot of imagination to

see the parallels to what could be possibly the

case for the biomarker situation, and whether we

call it a biomarker consortium or a biomarker

institute, it would have at its heart the same

goals that this Semiconductor Research Corporation

had.

So the goals for the Committee and the

strategies to move forward today: we have no yes

or no questions. We have no preconceived plan as

to how we're going to move forward. We have some

general ideas. And what we're here today to

discuss is to hear your input on the science of

biomarkers, the data that would be necessary,

opportunities in this field, obstacles, whether

they be culture, process, impediments, and also,

any thoughts you have on collaborations. What

we're looking for is your input and help to define

a new path forward for biomarkers and surrogate

endpoints.

You're going to hear three presentations

that I think will set the stage for the discussion.

Dr. Woodcock will start off and frame the issues as

one of the principle authors of the critical path

and one of the visionaries for this field. We'll

hear from Dr. Wagner an industry perspective, and

Dr. Wagner will represent the PhRMA Biomarker

Working Group, and he has, again, been working with

the others on a very thoughtful position paper, and

we'll hear some of the principles of that today;

and then, finally, we'll hear an academic

perspective from Dr. Blaschke, who has lived

through over a decade of the biomarker surrogate

endpoint progression, starting with the AIDS

epidemic back in the early nineties and reflect on

that and advise us on some thoughts about moving

forward.

As I say, the discussion today, we'll be

listening to very carefully. What we hope to

develop is a foundation for a national critical

path opportunity, which the agency will begin

identifying in terms of a priority near the end of

this year. We realize that this project on

biomarkers is going to be a very ambitious one.

We're very optimistic. And of course, like any

initiative that FDA undertakes, there's always that

specter of progress dependent upon its funding, its

sustained commitment and dedicated staff for such a

project.

So we're not overpromising anything, but

we would like to begin and move forward on this

path, and I'll start by introducing Dr. Woodcock.

DR. WOODCOCK: Good morning, everyone.

I'm really delighted to be able to be here and

begin this discussion about moving the field of

biomarkers in drug development forward.

I've named my talk a framework for

biomarker and surrogate endpoint use in drug

development, because that's really what we're, I

think, discussing here, but obviously, it has much

broader implications if we're able to move this

forward. And I'm going to address those as well.

First, I'm going to cover--Larry already

went over the current definitions. I think there

are some self-imposed limitations in the current

definitions, and therefore, I'm going to present

them again and talk about them. Second, I want to

talk about overall the limitations, I think, of our

current conceptual and developmental framework and

the reasons which are multiple why we're not moving

forward more rapidly in this area, and by we, I

mean the biomedical research community overall.

And finally, I want to talk about what potential we

have for moving towards robust use of biomarkers in

drug development and then toward regulatory

acceptance of surrogate endpoints, which would

follow on after the robust use in drug development.

Now, in the late nineties, NIH put

together a definitions working group of which I was

a member and some other folks in this room were to

develop some terms and definitions about biomarkers

and surrogate endpoints and to have an overall

conceptual model. There had been a lot of thinking

that had gone into the field about how these

interact. And this was an offshoot of the

consensus conference that was held on this topic,

and this was published in a paper.

The definition the working group had for

biomarkers was that it is a characteristic that is

objectively measured and evaluated as an indicator

of normal biologic process, pathogenic process or

pharmacologic response to a therapeutic

intervention. And I don't have any quarrel with

this definition, this one.

And this is ubiquitous, I think widely

used and accepted, although there might be a few

modifications you could make on this, but in

our--in FDA's draft pharmacogenomics guidance that

we published last year, in order to set up this

structure for regulatory filing or not of

pharmacogenomic information, we had to go further

and define the pharmacogenomic tests as either

possible, probable or known valid biomarkers,

because this type of definition, then, determined

whether or not there would be a required regulatory

filing under the law.

And these categories were sorted based on

available scientific information on the marker and

how much confidence you would have the marker

actually represented some real outcome or real

information. And we got a lot of comments on that

to the docket for this guidance, saying that we

needed more specificity on these categories and to

define them more clearly, and we will very soon

issue the final pharmacogenomics guidance, but I

don't know if it's going to shed a whole lot more

light on these biomarker definitions. As Larry

said, that's something we need to take up in this

larger context. So those are some of the extant

definitions out there of biomarkers.

Now, the group put forth a definition of

clinical endpoint, all right? And that is a

characteristic or variable that reflects how a

patient feels, functions or survives. And this

kind of is the crux of the conceptual problem I had

with this whole area. Note, you should note, and

this is my editorial comment, except for survival,

all these outcome measures or variables involve

some kind of intermediary measurement. It's really

not possible to know how someone else functions or

survives; we can only measure it in some--I mean,

or feels.

And I think we can all agree with that.

We have some kind of measurement that we interpose

between that person and the numbers, and we somehow

quantify how they feel based on some kind of

measurement.

Now, you can disagree about this, and we

should talk about this later, because this is very

important, I think. But anyway, that's a clinical

endpoint. And those are given in the scheme of

things some kind of fundamental reality.

Now, surrogate endpoint, in contrast, is

defined as a biomarker that's intended to

substitute for these clinical endpoints. And the

surrogate is expected to predict clinical benefit

based on various scientific, you know, studies that

have been done. And there is a feeling about a

surrogate, and this is something that we need to

develop more. It actually was presented by Dr.

Rowland at the biomarkers meeting, but there is an

issue about how proximal or distal the surrogate is

to the actual clinical outcome that you're trying

to describe or quantitate and say a blood measure

might be quite far away or might be very close, and

that might be based on mechanistic pathway

proximity or it might be based on a sort of

clinical face validity, so there are a number of

different axes on these surrogate endpoints, and

I'm going to discuss that a little bit more in a

minute.

This is the definition that was put forth

by the working group, and there wasn't a lot of

dispute about this definition. Now, as we all

know, biomarkers are used in clinical medicine.

They're not simply used in drug development. And

that is kind of the larger issue here. They're

used in diagnosis, as a tool for staging disease,

an indicator of disease status and to predict and

monitor clinical response.

And I see Rick Pazdur today, who's the

head of our oncology group. He knows very well,

often, the clinical use gets well ahead of the drug

development use. And that's because the clinical

use may be based on, you know, there's less--you

can simply adopt a biomarker and use it without

having an organized set of data and evidence that

you base that adoption on. So sometimes,

biomarkers will be taken up and used in clinical

medicine, at the same time not being used for their

corollary use in regulation or in drug development.

But because biomarkers are critical to

clinical medicine, to the diagnostic tests of the

future, there's more at stake here in this

discussion, in this overall initiative that we're

having than just efficient drug development, and

this can't be stressed enough, especially to the

outside stakeholders. Biomarkers really are the

foundation of evidence-based medicine, because it

is those types of tests that determine who should

be treated, how they should be treated, and what

they should be treated with.

And so, those quantitative measurements,

diagnosis should go before treatment, and yet, for

many of our treatments, we have very few

discriminatory markers that we apply. Absent new

markers, our advances in targeting therapy, either

in the traditional ways, which would be according

to drug metabolism and other standard markers, or

in new ways will be limited, and to the extent that

we can't or don't adopt these markers and use them

in drug development, treatment will remain

empirical.

So it's imperative for good medicine as

well as cost-effective medicine that biomarker

development be accelerated along with the

development of new therapeutics.

Now, here, just to get people's minds

around this, many of you in the room are experts in

this, but many may not. According to the NIH

definition I just talked about in biomarkers, these

types of measurements would be considered

biomarkers of different kinds. So it isn't just a

blood test. It can be all sorts of imaging

technologies or bone densitometry, all sorts of

things. Even an APGAR score is a kind of

biomarker. It's a way of quantifying certain

observations on a newborn.

Now, as opposed to use in medicine,

biomarkers are also used in drug development in a

decision making capacity to try to assess and

evaluate the performance of candidate treatments.

Where we have very good biomarkers, we can have

extremely efficient drug development, because the

performance of candidate therapies can be assessed

in animal models. And by the time we get into

humans, we have a very good idea of the

performance, a very good predictive idea of the

performance of the treatment.

The biomarkers can also be used to bridge

animal and human pharmacology and pharmacologic

effects of therapies by doing proof of mechanism.

And again, I'm stressing here the early acquisition

of information about the safety and effectiveness

of the therapy and bridging the animal knowledge

and the human knowledge.

There are safety biomarkers, and most of

those are 50 years old. I will tell you that the

markers we're using in the animal safety evaluation

in general and the human safety evaluation are

truly venerable, and they're tried and true, okay,

but they do not incorporate modern knowledge there.

They're largely empirically based, and they have

reasonable predictive value for major organ system

failure and not very good predictive value for

mechanistic understanding of the safety problem or

predicting more rare types of safety outcomes in

the same organ system. So there are problems with

that.

But the biomarkers, to the extent we have

them, can be used to evaluate human safety and

early development; hopefully predict safety

performance of drugs early.

And right now, we use serum chemistries.

We don't use cell surface protein expression very

much. That would be a target for the drug

intervention. Sometimes, that's used. Drug

pharmacokinetics over the last 15 years due to Dr.

Sheiner's efforts and many others, many in this

room, these types of measurements have become much

more standardized within drug development and have

tremendously contributed to our understanding of

drugs.

Serum transaminases and other safety

markers have been used forever. Genomic expression

profiles are used very, very rarely right now, and

imaging is, in specific fields, such as

neuropsychiatric disorders is being used widely,

the biomarker of imaging, but its utility is still

not clear, I think is fair.

In later drug development, this is where

the rubber really starts hitting the road as far as

cost of patient and so forth, and the stakes start

really rising. If you have good biomarkers to do

your dose-response work and develop optimal

regimens, it's extremely helpful before getting in

phase three to have a very good idea. Safety

markers to determine dose-response for toxicity, we

aren't as good there and determine the role, if

any, on differences in metabolism on the above

dose-response, and this isn't done as widely--is

that fair, Larry, to say--as it probably would be

optimal to do, for a variety of reasons.

Now, here's where we start getting some

probability areas for dispute or discussion.

Biomarkers used in later clinical development: I

would--psychometric testing or psychometric scales

or whatever are used as clinical outcome measures

in trials of psychiatric disorders. I would argue

to you that's as much of a surrogate as an HIV

viral copy number.

It's just we're used to this, so we don't

think of it as a surrogate. We've used it a lot,

and we're comfortable with it. But we don't know

that it represents a cure or a mitigation,

necessarily, in an individual patient. A lot of

work has gone on, and I think we have great

confidence that the testing and outcome measures

that are done for psychiatric diseases actually

reflect efficacy of the drug and have tremendous

utility in the approval of psychiatric drugs;

however, I don't think people recognize that this

is as much of a surrogate as many other types of

surrogate markers that have been discussed.

Pain scale is another thing: I mean, you

can't feel another person's scale of pain. We have

constructed different measures, metrics, and they

have been run through the psychometric testing

algorithm to look at their construct validity and

so forth and so on, and we know their performance

pretty well. But they are surrogates for actual

pain.

Imaging can be done; culture status is

obviously a very important marker, not necessarily

a surrogate for antimicrobials; pulmonary function

tests, serum chemistries, electrocardiogram. And I

think what's striking about many of these is they

are very traditional. They've been used in

clinical medicine a very long time.

Now, what about surrogate endpoints that

substitute for the clinical outcome measure? Well,

obviously, there are surrogates for efficacy that

can be used to assess whether a drug has clinically

significant efficacy, and there are surrogates for

safety. And basically, our entire drug development

program and the exposure of patients is, in some

way, a surrogate for the real world safety, because

that's what we're really concerned about is how the

drug will perform when it's marketed and out there

in the real world as far as safety goes, so the

entire development program and the patient exposure

experience and the way we look at that is used to

predict safety.

Right now, known surrogate endpoints and

points that are used include blood pressure,

interocular pressure for glaucoma, hemoglobin A1C;

as I've already said, psychometric testing; tumor

shrinkage for cancer, and there's criteria,

performance criteria around all of these. For

rheumatoid arthritis, the clinical endpoints used

in trials are the American College of Rheumatology

criteria that were worked through by the

rheumatologists with great effort, and then, pain

scales are used for pain.

Now, what I want to turn to after giving

sort of an introduction is what I consider

limitations of the current conceptual and

developmental framework for biomarkers and

surrogate markers. And the reason I want to do

this is because I think we have to start there in

rethinking, as Larry said, if we're going to put a

consortium together, if we're going to try and work

on new biomarkers, we all have to be on the same

page conceptually about what we're trying to

accomplish and what are the issues.

I think most people would agree that

biomarkers represent a bridge in many cases between

a mechanistic understanding that has been gained in

preclinical development or in actual basic science,

and what is largely now the empirical clinical

evaluation, and the goal is to bring the

mechanistic understanding more forward into the

clinical evaluation to make it more predictable,

both on safety and effectiveness. And the

hypothesis is we can use biomarkers to do that if

we understand their performance adequately.

Now, because of history, we didn't have

the science in the past, and as regulators and the

regulatory system has been focused on empirical

clinical testing. And there are tremendous

limitations to that, but that is the best we have

had. And that has really, though, we have that

historical momentum that is continuing to skew our

approach to the clinical, the human evaluation of

drugs to sort of an all-empirical.

And what do I mean by that? Well, you

just expose them, and you see what happens. You

randomize people, and then, you count whatever you

count at the end of the day, and that's basically

empirical drug development, and that's one of the

reasons it's so expensive and timely and risky, is

because there's a tremendous amount of failure in

this approach, and we don't gain as much knowledge.

This is not a highly informative approach, either.

And of course, the FDA is constantly criticized for

drugs that are on the market postmarketing that we

don't have as much information as would be

desirable about those drugs.

I think all of us in this room know,

nevertheless, how expensive, time consuming and

what incredible effort current clinical drug

development is, but this is contrasted with the

fact that at the end of it, we don't know that

much. And we should have a discussion about this

afterwards. That's true. We really don't know

that much at the end of current drug development.

And as a result of this being skewed

toward a more empirical approach, the early

mechanistic clinical evaluation has often been

lacking. And I think Larry can speak to that, our

end of phase 2-A meetings are speaking to that.

There really hasn't been that focus. And this

isn't to blame anyone; the reason we haven't

focused on that in the past is we have not had the

tools to do this, and the question is is now the

time where we are developing these tools, and

should we put a lot of effort into this to develop

those tools, and do we have enough scientific

knowledge to actually make the process a lot more

predictable? And I would say the answer is yes.

But I would say as a result of the

history, the business model for biomarker

development is lacking. There was just an article

in Biosentry magazine about this, I think, last

week, about companies who have been trying to get

into the biomarker business, and they say there's

really not that much interest or a model for how

they can move forward and develop these biomarkers

and have them used in drug development. And we've

heard this; I have heard this ubiquitously over the

past six months as I have been going around talking

to people about the critical path.

So a consequence of this that anyone can

easily observe looking at the literature is there

has been no rigorous pursuit of the evidence that

would be needed to qualify a marker, really

assemble the evidence on its performance or to

assemble that evidence at a level where you get

regulatory approval of that marker. That doesn't

happen that much, and there are a tremendous number

of markers out there, and we know very little about

their performance in a rigorous way. And the

exploration of their clinical relevance is

generally ad hoc; it's pursued in an academic

manner.

However, I think there's an urgent need to

overcome these obstacles I have just discussed. We

have new opportunities to link biomarker

development to the drug development process,

particularly with a newer genomic proteomic imaging

and other types of markers that have been developed

and with the kind of quantitative modeling that we

can now do.

This requires, though, a clear regulatory

framework, a signal to be sent from the regulators,

I think, of what kind of technical evaluation is

required. And within our pharmacogenomics effort,

we're getting a lot of questions. I think that's

probably one of the major questions that is sent to

us, which is what kind of information has to be

sent to the agency at different stages of

development?

But the need also is to develop some new

business models that are viable, because someone

has to develop these tests: either the drug

developers, device developers, someone has to

develop these tests. They can't just be an

academic tool if we're going to use them in this

manner.

Now, I'd like to turn to surrogate

endpoints. And I gave a definition previously

about surrogate endpoints, how they stand in for

clinical outcomes or clinical endpoints. As most

of you know, the current model for use of a

surrogate endpoint is based largely on

cardiovascular and HIV experiences in the 1990s and

sort of the analysis that went on around those

experiences.

The cardiac arrhythmia suppression trial

that was performed in, I think, sometime in the

1990s was done because of widespread use of

antiarrhythmic agents to suppress the ventricular

premature beats post-MI based on the hypothesis

that that would decrease the incidence of sudden

death in that population, because they're at risk

for sudden death, and the surrogate there was the

suppression of VPBs.

What happened when the arms of the trial

were unblinded is the mortality was increased in

the treatment arms of this trial. And that was

quite a shock to folks, probably akin to what

happened when they unblinded or they looked at the

postmenopausal estrogen treatment a year or so ago

and found that myocardial infarction was increased

in the treated arms.

This caused some--the cast outcome caused

a lot of skepticism, particularly in the

cardiovascular community, about our ability to rely

on surrogates. This is despite the fact that there

was a fair amount of evidence, I think, if you're

sort of impartial about this, a fair amount of

evidence that certain types of antiarrhythmic

agents can cause sudden death as well as certain

kind of antidepressant agents and everything that

have certain electrocardiographic properties and so

forth.

Nevertheless, this cast outcome was a real

shock. It kind of cast a pall over the adoption of

surrogate area. And the whole discussion about

this effort and everything can be seen in the

reference I have here by Bob Temple, who wrote up

in the midnineties some of the experiences that FDA

had encountered around surrogates.

Now, then, we had the HIV epidemic in the

nineties, late eighties, nineties, and there was

again discussion, there was discussion of the use

of surrogate endpoints in this disorder; first,

CD-4 counts, which were obviously not really on the

mechanistic chain as much as some other endpoints,

and as a result of this whole discussion, some

rigorous statistical criteria for assessing the

correlation of the candidate surrogate with the

clinical outcome were published have a reference

here they're called the Prentiss criteria, and it

really called upon a surrogate to really encompass

all the qualities of the clinical outcome, so you

wouldn't learn any new information, basically, if

you substituted the clinical outcome for the

surrogate.

This is probably impossible, and no

surrogate endpoint that is currently adopted has

met these criteria. But again, this has caused

concern for people about what do you need to do,

and is this a reasonable criterion? It is a good

postulate of the problem, okay? And it frames the

problem very well, and there are a lot of other

articles which I could provide to people if you're

interested by statisticians, discussing various

performance characteristics of surrogates and the

way you can be misled about surrogates.

But nevertheless, the outcome of this was

that HIV RNA copy number was used as an early drug

development tool. It's now used as a surrogate

endpoint in trials, and it's used for clinical

monitoring and antiviral therapy. There is a lack

of complete correlation of this outcome measure

with clinical outcomes, but my point is this does

not compromise the utility of this measurement for

its use in drug development or in monitoring

patients. And the point is that all of our

measurements are uncertain; there is some

uncertainty and lack of full information associated

with any measurement you might make on any person.

So there has been successful development

of antiretrovirals and control of HIV infections,

despite the fact that this particular surrogate RNA

copy number is not perfect and certainly misses

certain parts of the outcome for any given drug.

But I want to move now to what I think is

a more fundamental problem and has been a block in

our discussion, and I alluded to this earlier, and

as I said, people may disagree with my assessment

of this, but as a clinician, I would say there is

no gold standard in clinical outcome measurement.

People always argue with this, and they say

survival. Survival is an absolute.

And I will tell you if you look at the

data, say, of John Wendburg and folks who developed

that about what people would choose, would they

choose longer life? Would they choose better

quality of life? There are many people who would

prefer to live a shorter amount of time if their

longer life would--if they would have to trade off

a very poor quality of life for that prolonged

life.

So any measurement does not always capture

all the domains of interest for a patient; even

survival. Now, I realize that's a strong

statement, but obviously, if you survive sepsis or

MI or something, you're left with no sequelae,

you'd much rather be alive, and in those cases,

that's a pretty good sequela.

But the generalizability of any single

outcome measure can also be limited by the trial

parameters. So we aren't really getting to full

truth in a trial, even with a survival endpoint.

As a rheumatologist, I'm very well aware of this

because the rheumatologic diseases generally do not

have a single dimension outcome, and capturing just

one, capturing simply pain or function or whatever

is not adequate for fully describing impact on the

disease.

And therefore, many clinical outcomes and

many diseases are multidimensional, and any single

outcome measure we use may miss domains of

interest. That doesn't mean we should throw up our

hands. We should simply be aware of this, that

there is no single gold standard that we're

comparing anything to.

In addition, and this is something the

Prentiss criteria were talking about, because they

were looking at survival, and survival can be

diminished, obviously, by harm as well as prolonged

by treatment effect, but in general, it's very

difficult to capture both benefit and harm within a

single measure. And we don't even attempt to do

that within drug development. We're assembling

information from a wide variety of sources, so that

the concept of ultimate clinical outcome is very

elusive. There's always a longer duration, say,

for chronic disease. You could always follow

people longer. The definition of what is ultimate

is very unclear.

And so, I think we need to move away from

the idea, and maybe I'm beating a dead horse here,

that there's one single piece of knowledge that

everything has to be correlated to. That's just

really not how human beings and disease are. And

knowledge about various dimensions can be acquired

outside of a biomarker or surrogate measurement.

We don't have to put all our weight on a single

surrogate measurement.

In addition, and this is becoming very

important in this, I hope, new world of more

individualization of therapy, the per patient view

of outcomes is very different than population mean

view of outcomes. If you are the person who

experience an adverse effect from a drug, the

surrogate means nothing to you, the efficacy

surrogate, because something really bad happened to

you.

And where we have the ability now to more

individualize therapy through biomarkers, either

through pharmacogenomic, genetic testing for

metabolism, enzyme metabolism, where more

sophisticated measures of determining who stands to

benefit from a therapy or who is at high risk for

an adverse event for a therapy, this becomes very

important. So newer and older biomarkers do

provide information at the individual level. And

that's very important.

For the reasons I've just gone over, then,

I think our conceptual model should view drug

development more as progressive reduction of

uncertainty about the effects or, if you're the

glass half full type, increasing the level of

confidence about the correlation between treatment

and outcomes, not a single, binary measurement of

the drug is effective, it isn't effective; there

are safety problems; there are not safety problems.

We have to be dealing with, in other words, a

multidimensional set of information, not a binary

decision.

Now, I recognize that the regulatory

decision has this binary quality about it. And I

think what I'm telling you is that you should

suppress the science into a binary box. That's not

the right way to go about this. The regulators

have to figure out when that evidence is enough

separate from the way the evidence is developed and

understood.

So no single measurement contributes all

knowledge, and even if we get to the, you know, the

star--what is that Star Trek, where they wave that

thing over people, and they get--they probably got

a series of measurements. They weren't just doing

one measurement with their magic wand. And

population mean findings may not be valid for any

given individual. And that's a very powerful

statement, I think, as far as the fact that this

anyone surrogate measure may not really predict for

a given person a correct outcome.

So in the future, I think we need to move

to more composite outcome measurements, more of a

multidimensional understanding. And I realize--I

mean, this is the Clinical Pharmacology

Subcommittee; I'm preaching to the converted here.

These folks have understood this for a very long

time. However, we need to move this into the

general understanding of drug development and

therapeutics.

This means probably in general, as we move

forward, we need to be looking at responder

analyses and so forth and looking at the data in a

more careful way rather than population mean

analysis. And we also need to be moving towards

individualized therapy.

Now, we would expect, and this is kind of

the quid pro quo here, with these evaluations, we

also are going to have to see a larger treatment

effect to provide some face validity here, if you

follow me. But you would expect that if we were

able to predict who is able to respond to drugs and

sort out who is at risk for adverse effects. We

should be seeing larger treatment effects, and in

fact, we are for some of these therapies as they're

moving forward.

A basic problem in a lot of drug

development is the drugs don't work very well,

because they are--a lot of people who are exposed

don't stand to benefit from the drug and aren't

going to benefit. But our empirical method of drug

development causes these apparent, very small

treatment effects.

Now, what should we do? What do I think

we should do? And I think Larry laid out kind of

the spectrum of probabilities or possibilities

pretty well. What I would like to stress is

biomarkers have to be used to be accepted. We have

lots of surrogate measures that we use in clinical

trials and regulatory, I believe. I believe a lot

of the things we use are surrogate markers. We

just are so used to them, we don't think they're

surrogate markers.

But what part of understanding the

performance of these newer technologies is to use

them, to see how they move with treatment or how

they fail to move with successful intervention, to

see how they perform in various populations and

with a wide variety of drug interventions? With

that kind of knowledge, that's the kind of robust

knowledge we need, then, to have both regulatory

acceptance and, then, wider acceptance in clinical

medicine.

The barrier to this up to this point has

been the add-on costs, and there have been many

barriers, but a major barrier is the add-on cost in

clinical trials. And I've talked to the imagers

about this. Nobody wants to put an imaging arm in

the trial if it's experimental, because it's going

to cost a lot of money, and not only could it not

be used to support approval, but it might show

things that are new and unknown. And there is

concern that these biomarkers will, and they have,

actually, segregate out the people who are most

likely to respond and thus narrow the target

population intended for that investigational drug.

There's also concern that questions, new

information would be found by these biomarkers;

questions would be raised by the regulators, and

that would slow the regulatory acceptance and

approval of the therapy.

And, you know, we all have to get over

this together, because otherwise, the use of

biomarkers in trials will not occur. And that's

what has to happen for us all to start

understanding these.

Now, as Larry said, to bring all this

about is going to require some kind of

collaborative effort between government, academia

and industry and probably not just the

pharmaceutical industry but diagnostic side of

industry as well. And we're going to have to

focus.

So I just said this: the diagnostic and

imaging industry sector needs to be fully engaged

in this effort. So it's going to require a lot of

parties. And FDA must provide the regulatory

framework and some reassurance as we move forward

that individuals and firms are not going to be

punished for this, so to speak. And the

pharmacogenomic guidance that we published the

draft last year is an example of that. It provides

a space, an experimental space, where those tests

can be done without the fear of all these

regulatory consequences occurring and where the

information can be shared.

Now, development of new biomarkers, you

know, new biomarkers are going to revolutionize

probably both the development and use of

therapeutics and preventatives. But as I said, it

requires commercial development of the biomarker

technology. Academia's role, I think, is to

identify these technologies, put them forward and

assist in their evaluation. But they have to be

commercially developed, and we need regulatory

pathways for the pair, the therapeutic intervention

as well as the biomarker, and that's what we've

tried to lay out for pharmacogenomics, but there

are many other types of technologies that we also

need to have the same pathway made available.

Now, for surrogate endpoints, I think we

need further exploration and discussion of some of

the ideas that I put forth today, and this is sort

of the kickoff, but we're going to have to have

more discussions of this. I could be dead wrong.

I don't think so, but we need to talk about it. I

think we need to get rid of the idea of validation,

and Gerry Migliaccio is here, and we've gone

through this in the last two years for the GMP

initiative.

Validation is a term that, unfortunately,

often conveys an idea of much more assurance and

rigor than is actually attached to the activity,

and we need to use more descriptive terms that

everybody understands what is required or what the

activity actually is, so I think validation is a

bad word to use in this context, because it doesn't

convey any information.

And we may need to adopt new nomenclature

overall around surrogates or perhaps refine the

nomenclature. We need more emphasis on the fact

that our understanding of disease and disease

interventions is multidimensional. It's not a

single dimension. And I think we need greater

emphasis on safety biomarkers, because safety

problems, obviously, are very prominent in the

news. They're also a tremendous source of loss of

compounds within drug development; maybe compounds

that would be very good and for 99 percent of the

people would actually benefit them and their

disease.

So, we need to replace, I think, the idea

of validation with something about degree of

certainty or progressive reduction in uncertainty

or some concept like that that is more graded. The

problem with validation is it's, like, you're

validated; you're not validated. It isn't like

that. And we have to recognize and remember that

the usefulness of any surrogate will be disease,

context, and to some extent, intervention-specific.

And that's why one of the dimensions that needs to

be investigated for any surrogate is

generalizability across product classes, across

patient populations, across stages of disease or

what have you. That's why these have to be used in

trials. We can't just have them out there in

papers.

We need to develop a framework for

understanding the usefulness of a surrogate as

evidence, used as part of the evidence that's

submitted to the FDA for approval of a drug or

safety in a context-specific manner.

So in summary--it looks like I'm right on

time here--there's an important public health need,

I think we can all agree, but we need to get this

message out, so that people understand why this is

important. I don't think the general world

understands what's at stake here. There's a need

for the development of additional biomarkers to

target and monitor therapy.

To do this basically is going to require

that they be used in clinical trials during

development and postmarketing trials as well. The

business model, in other words, who is going to pay

for this, how this is going to happen, and the

regulatory path for such markers is not clear to

industry. And we need both clarification, in other

words, what is the path forward, the technical,

scientific path, as well as some probably stimulus

is needed as well in the economic sense.

There have been definitions. Larry and I

both alluded to those for these various terms. But

I think further development of the model is needed

to get it to a higher level of sophistication in

order to increase the use and utility of markers in

development and enable us all to talk to one

another and know what we're talking about. I think

this further development has to recognize the fact

that single measurements will rarely capture all

dimensions of the clinical outcome for any patient.

So I think that a multidimensional and

continuous model needs to replace the current model

that we're using for clinical effect, and that's

critical for the targeted therapy of the future,

because this will be multifactorial as far as for

any individual patient, whatever their metabolizer

status or whatever it might be that the state of

elaboration of various proteins, receptor proteins

on their tumor cells, whatever it might be, these

factors will influence their response to therapy,

and many of these factors will not be binary

themselves. You would not elaborate receptors on

your tumor or not; it's going to be a gradation.

FDA is considering development of these

concepts, as Larry said, as part of our critical

path initiative, and this initiative, if we take

this part up, would include a process for refining

the general framework as well as individual

projects on biomarker and surrogate marker endpoint

development, because at the end of the day, the

surrogates in particular are going to be, as I

said, disease specific.

So I look forward to the discussion, and I

hope that this will lead to really something

getting started in this area. Thank you.

DR. VENITZ: Thank you, Dr. Woodcock. Any

quick questions or comments by the Committee

members?

DR. SINGPURWALLA: Yes. I do have a

comment. First is I find myself agreeing with much

of what you say. Sometimes, I wonder if you're a

doctor or an engineer.

[Laughter.]

DR. SINGPURWALLA: But the problems you've

described are very isomorphic to the problem that

engineers have found, and I'll give you two

examples of what you said: one of your slides

talked about validation, and you said that you

shouldn't have something which is either validated

or not. There's got to be some degree of

uncertainty.

There is a body of knowledge called vague

sets or imprecise sets where the boundary of the

set is not well-defined, and you say there is a

certain degree of membership in that set. It goes

under an ugly name called fuzzy sets, which the

President sometimes uses.

[Laughter.]

DR. SINGPURWALLA: But I would strongly

encourage you to look into that literature.

Now, as far as the markers are concerned,

the problem again that you are facing is similar to

what engineers face with, say, aircraft structures.

The aircraft structure is degrading, and what they

see is a crack. And they monitor the crack; they

study the crack, and based on the growth of the

crack, they predict the performance of the

aircraft. So there is a large industry which looks

at that. You may want to take advantage of that.

And the correct way to model these things

is through stochastic processes, and these are

bivariate stochastic processes, and that would be

the direction in which you may want to go. One

process is observed; the other process is

unobserved. It's the unobserved process you're

interested in, and the observed process gives you a

clue. So at least I'm telling you that there is

some parallel paradigm that you may want to

consider. I strongly encourage you to look into

this.

Thank you.

DR. WOODCOCK: Thank you. I think what we

found in our recently-completed GMP initiative is

that bringing in the engineers and various

other--multidisciplinary look to some of these

problems we're facing provides tremendous power,

because people have faced these problems in other

fields.

DR. VENITZ: Any other comments or

questions?

[No response.]

DR. VENITZ: Then, thank you again.

And our next speaker is going to be Dr.

Wagner, and he's going to give us the industry

perspective on surrogate markers.

DR. WAGNER: Great. So, thanks very much

to the Committee for the invitation to and the

opportunity to discuss a little bit of the industry

perspective on biomarkers and surrogate endpoints.

And we've been giving quite a bit of thought to

this. I represent PhRMA in this case, and in

particular, the PhRMA Biomarker and Genomics

Working Groups, and my colleagues Steve Williams at

Pfizer and Chris Webster have been very large

co-conspirators in this particular effort. And I

represent, actually, a very large group that is

noted at the very end of the slide.

So I want to step through a couple of

different areas. I want to talk really about what

our objectives and focus is right now, a little bit

about biomarker nomenclature, which Dr. Woodcock

and Dr. Lesko have already covered to some extent,

and then talk about the idea of qualifying

biomarkers as surrogate endpoints and the idea of

it's not--very much along the lines of what Dr.

Woodcock said, it's not really a binary process;

it's actually a continuous process of increasing

certainty and then end with some thoughts, some of

our thoughts on collaboration.

So, there's not a laser pointer, I guess.

That's okay. The landscape, I think we all agree,

is one that Dr. Woodcock already highlighted, that

there's really a much more intense focus on

biomarkers as aids for decision making in drug

development and the regulatory evaluation of new

drugs. And our objectives within the PhRMA

Biomarker Working Group is really to work towards

an improved framework for regulatory decision

making, regulatory adoption of new biomarkers to

work towards a refined nomenclature that will

enhance the discussion and also to work on an

optimized business model for biomarker research;

again, something--these three things are really

very important necessities in moving biomarker

science and use in drug development along.

So our focus has been on the process, the

process to select suitable biomarkers for potential

regulatory purposes, to define what research is

needed for qualification and regulatory use, to

execute that research in a cost-effective manner

and to review the results and agree on whether a

particular biomarker meets the needs.

So I also would like to go back to the

FDA/NIH consensus conference in 1990--oh, thank you

very much--and I won't dwell on this, but before

that consensus conference, there really was, well,

there was a lack of consensus. There

was--biomarkers were--the term biomarkers were

bandied about in a very casual way, and there was

really no consensus on what folks were talking

about. And the real seminal contribution to that

FDA/NIH consensus conference was was this

definition that Dr. Lesko and Dr. Woodcock already

read--I won't repeat it--for biomarker and

surrogate endpoint?

And it's really served as the groundwork

for all the efforts that have come since then,

because there really was a far-reaching agreement.

We've done that; now, we can move on to some of the

refinements that are really necessary to the next

stage. And that's been part of the thinking over

the last five years or since that consensus

conference, and that's where we're going to go in

the future.

But I think that we all agree that--or at

least Dr. Lesko and Dr. Woodcock agree that the

biomarker and surrogate endpoint distinction is

really not optimal for use of biomarkers in drug

development, and there's a couple of guidances that

really highlight that. One is, as has already been

highlighted, that the exposure response guidance

really makes a distinction based on the evidentiary

status of biomarkers going from valid surrogates

for clinical benefit to really remote from a

clinical benefit endpoint.

And then, also, in the pharmacogenomic

data submission draft guidance, there's really a

further--that point is really drummed home even

further, that there is a further distinction based

on the evidentiary status of dividing biomarkers

into probable valid biomarkers and known valid

biomarkers, and that really leads into this idea of

qualifying biomarkers in a way that makes them fit

for the purpose that you intend to use them for.

I also don't like the term validation,

maybe for not quite the same reasons as Dr.

Woodcock, but I don't like the term clinical

validation, which is often used in the literature,

because this process, I believe, has just as much

to do with biology as it has to do with clinical

outcomes. In the FDA/NIH consensus conference, the

term evaluation was used for the process of

qualifying biomarkers. That's probably okay, too,

but we've settled on a term of qualification; it's

really distinct from validation and captures, we

believe, the idea of a graded process that leads to

the right purpose for the use of the biomarker.

So we have sort of a simple working

definition here, an evidentiary process that links

a biomarker both with biology and with clinical

endpoints. The purpose here, after all, is really

to provide reliable biomarker data that's both

scientific and clinically meaningful, and in the

context that it's being used in.

In these remarks, my focus is very much on

disease-related biomarkers that are intended as

indicators in one way or another of clinical

outcomes. There's, of course, a great deal of

interest in all sorts of other biomarkers,

particularly pharmacodynamic biomarkers or

mechanism-related biomarkers, but I think that the

need for the regulatory scrutiny on those sorts of

biomarkers is a little bit less than the

disease-related biomarkers, because really, you

know, the--how we approach the evidence to how hard

a particular therapy is hitting a target is a

little more clear-cut than some of the issues that

relate to qualifying a disease-related biomarker.

So my remarks are a bit more restricted to these

disease-related biomarkers.

And then, the last point I want to make on

this slide is that this fit for purpose biomarker

qualification really is a graded with the accent on

graded evidentiary process of linking the biomarker

with biology and clinical endpoints, and it depends

on the intended application. So this is the

universe of biomarkers that came out of the

consensus conference, biomarkers versus surrogate

endpoints, and I think we can agree that it could

be more useful to provide a little bit more

granularity.

And one proposal that we've been exploring

is to fill in this spectrum of biomarkers with

graded levels of evidence, stretching from

exploration through demonstration through

characterization and finally through surrogacy.

So, an exploration biomarker would be a biomarker

which is really a research and development tool. A

demonstration biomarker, then, would, in this

proposal, would correspond to a probable valid

biomarker, and a characterization biomarker would

correspond to a known biomarker, and surrogacy has

the same meaning: a surrogate endpoint, a

biomarker that can substitute for a clinical

endpoint.

So just to put a little bit more detail on

there, it is not a lot of detail, because these

really are draft concepts, but an exploration

biomarker, then, again, is a research and

development tool. It's not that there's no

evidence. We wouldn't use a biomarker that had no

evidence associated with it. There wouldn't be any

sense in it. But the evidence is largely

restricted to in vitro or clinical evidence, and

there really is no consistent information that

links with clinical outcomes in humans.

A demonstration biomarker, then, one step

up in evidence, again, corresponding to a probable

valid biomarker is something with adequate

preclinical sensitivity and specificity and some

links to clinical outcomes but not really

reproducibly demonstrated or reliably demonstrated

or robustly demonstrated. A characterization

biomarker, again, corresponds to a known valid

biomarker, and this is one, again, that has the

adequate preclinical data associated with it and is

more reproducibly linked with outcomes through one

or more adequately-controlled clinical studies.

And then, surrogacy is, again, has the

same meaning as the NIH consensus conference, a

biomarker that can substitute for a clinical

endpoint. And the evidence, the details of how

that biomarker becomes a surrogate endpoint are

still very much a matter lacking in consensus. You

know, some of the thoughts that we've talked about

are having an association and treatment effects

across studies or times to events within studies;

you know, there's other ways to couch the evidence

that leads to surrogacy, and as I said, there's by

no means any consensus there.

So just to give a little bit of a couple

examples of where various biomarkers would fit in

this kind of a scheme, exploration biomarkers

really are only limited by the imagination and the

state of the evidence that exists scientifically.

There's numerous examples. A demonstration

biomarker could be something like adiponectin,

which is a P-par gamma agonist biomarker.

Adiponectin levels increase at P-par gamma

treatment, and they're associated with insulin

sensitization, but the tie to insulin sensitization

is far from perfect. There's also intriguing

associations with cardiovascular outcomes with

adiponectin, but the level of evidence is far from

perfect.

So this is a biomarker that I would at

least put into the demonstration bucket: do we

need it as a surrogate endpoint? I don't know; but

it's a very intriguing biomarker, especially for

P-par gamma agents, and in particular, because its

response is very rapid as opposed to hemoglobin A1C

and some of the more traditional surrogate

endpoints in diabetes.

Now, a characterization biomarker that I

listed here is HDL cholesterol, and there's

really--there really is a great deal of clinical

data associating HDL cholesterol with clinical

outcomes, but there still is a lot of ambiguity

about what some of those data mean. Some of those

associations are still a little bit murky, and I

think most folks would agree that it doesn't fit

bar of a surrogate endpoint. And then, I listed

LDL cholesterol as an example of surrogacy.

So we would say that there's a number of

potential regulatory uses of qualified biomarkers

in different categories. There's probably--you

could make the argument that there may be less need

for regulatory scrutiny of exploration and

demonstration biomarkers, but we would contend that

there's at least some interest in focusing on how

to move the biomarkers through an evidentiary

scheme like this, and there's some potential roles

of at least a demonstration biomarker, for example,

as supporting evidence for primary clinical

outcome.

A characterization biomarker, some of the

regulatory uses that we would assert would include

in dose finding and possibly in secondary and

tertiary claims, and of course, surrogacy, as is

already talked about, one of the examples of a

surrogate endpoint would be in registration.

Now, there is a--this is a graded process

of increasing levels of certainty, increasing

levels of evidence. There's also really a life

cycle for biomarkers. So not only is there a

natural progression that you could imagine that

goes from exploration to demonstration to

characterization to surrogacy and then use in

general medical use decision making; but as Dr.

Woodcock pointed out, it also goes back here:

things that are in general medical use. Not

everything goes through this data stream. It

comes--many things are used in general medical use

come back and only then become adopted into the use

of in drug development.

Similarly, not all of these things work

out, and we have to accept that as we study

biomarkers, we're going to develop evidence that

impugns their use. And I only put the arrows in

this slide in these top two categories, but in

fact, at any point, a biomarker can fall out of

qualified use. And I think again, we have to

accept that this is a risk of using biomarkers.

There's been much talk about the CAS study

over the last 10 or so years in the biomarker field

and about how that's really an issue, but I would

submit that in drug development, we accept the

risks of withdrawing drugs from the marketplace,

and no one wants to have a drug withdrawal from the

marketplace, but we seem to have a reluctance to

accept the idea that something that we've agreed is

a qualified surrogate endpoint, we're going to

develop evidence that it's no longer a qualified

endpoint.

I would submit that it's a risky--the

whole drug development process is a risky

proposition, and we are going to develop in some

cases evidence that surrogate endpoints aren't

going to work out. And that is really a fact of

life in biology and medicine.

The last thing I wanted to point out in

terms of this line about qualification is that this

really isn't the only example of a graded

evidentiary process for qualifying biomarkers. A

number of years ago, the NCI Early Detection

Research Network had come up with this concept for

phases of discovery and validation of cancer

biomarkers, and they have five stages that go from

preclinical exploration, where promising directions

are identified, through retrospective longitudinal,

where a biomarker detects a preclinical disease,

and a screened positive rule can be defined all the

way through cancer control, where the impact of

screening and reducing the burden of disease on a

population is quantified.

So this is a somewhat similar schema to

the one that I presented, and I think that in

general, this idea of a graded evidentiary scheme

is a useful one. Of course, there was a number of

issues here, and I list only some of them. There's

many different schemes of biomarker nomenclature.

There's many different uses of biomarkers, and I

talked to some extent about that as it relates to

ranging from hypothesis generation to regulatory

decisions.

A particularly difficult issue with

biomarkers is the different technology platforms

for biomarker assays. So they range from

immunologic assays to expression profiling to

imaging to psychometric scales. It's very hard to

talk in a uniform way about biomarkers in general

when the range of the measurements is so wide. And

also, as highlighted by Dr. Woodcock, there's the

potential role for multiplexed biomarkers, but we

really haven't gotten the scientific work done on

how to put those into the right conceptual

framework yet. It's really a very nascent field,

one that's rapidly developing but still very much

in its infancy.

And I did talk a bit about the different

strategies for qualification. And I didn't really

talk very much about the assay validation side.

But there's equally important issues about how the

assays themselves are validated and then put into

wider use.

And the last issue here is that there is

an obvious need for collaboration in biomarker

development. And that's what I wanted to spend the

remainder of this talk on. So we would be the last

to suggest that a collaboration model is the

solution for all biomarkers. There's many, many

uses of biomarkers that don't need any

collaboration. But there are many instances:

imaging is one example, where the scope of the

project has become so large that a collaboration is

really--it's really the only way to move it

forward.

And there's many options for

collaboration. I listed some of them here. The

PhRMA-FDA-NIH or other academic governmental

collaboration, that's what we would really think of

as the ideal new independent entity with FDA

collaboration, PhRMA with FDA; without some of

these other folks, PhRMA as a consortium or the

status quo.

If we assume that a more wide-ranging

collaboration is desirable, it really comes down to

the question of how members of PhRMA can work with

FDA, other governmental agencies, academics and

develop qualified biomarkers in regulatory decision

making. How can we do that?

Well, we believe that there are really two

broad issues here. One of the issues is really

deciding what biomarkers to pursue; making a

development plan; executing the development plan;

and maybe even at the onset, putting things into

the right framework. And this is an issue, a group

of issues that benefits from the widest possible

cross-collaboration between groups.

The second group of issues is deciding

what data would really be necessary for the

qualification of a particular biomarker or

reviewing that data on a biomarker and advising

regulators on its acceptance. And this is

something that we view should be more independent

of industry involvement.

So we would submit that one way to do this

would be to have an executive consortium that would

involve industry, both PhRMA and biotech, as well

as diagnostics, devices, perhaps other areas; the

government, in particular, the FDA, NIH, and

academics.

Then, the other really important group

would be a review and acceptance group, and this

would primarily, in our view, fall on the shoulders

of the FDA. How that would flesh out is something

that could take various forms: a relevant review

division for each biomarker if applicable; a new

intercenter advisory group or a designated FDA

advisory committee. If it were an FDA advisory

committee, we really would recommend powering that

committee appropriately so that the issues could

really be worked on.

And then, in our proposal, form would

follow function, and these separate groups would

deal with each of these broad groups of issues, so

that the executive consortium would deal with the

group one issues, and the review and acceptance

group would deal with the group two issues.

And then, going back to the executive

consortium, the idea there is really not as the

developer of all biomarkers; the biomarker science

is a very, very large field, but to coordinate

aspects of biomarker research, allowing a wide

membership; ensuring that interested parties and

specific biomarkers are connected and brokering

syndicates, identifying gaps for qualification in

biomarkers and really providing a forum, a one-stop

shopping for sharing biomarker science and then

acting as an expert interlocutor with regulatory

agencies.

Now, we recognize that there's a large

number of issues, some of them very vexing, toward

adoption of a collaboration approach. There's both

incentives and disincentives to industry for

collaboration. We would submit that a major

incentive would be regulatory predictability and

process. The funding for such an enterprise is an

issue, and it could take various different forms:

intellectual property in this kind of a consortium

idea is an issue, as is antitrust, and governance

is a particular issue. The last thing that we

would want to suggest is to create a new, difficult

bureaucracy that makes things harder to do rather

than easier to do.

So, again, I represent a large number of

people that are working both within the PhRMA

context and some outside of that. And in

particular, I want to acknowledge the Biomarkers

Working Group within PhRMA. It's been in existence

for about a year as well as the Pharmacogenomics

Working Group.

DR. VENITZ: Thank you, Dr. Wagner.

Any quick questions by Committee members?

Yes, Hartmut?

DR. DERENDORF: That was a very nice

overview, and I like your proposal at the end, but

I'm a little skeptical if that really will be

embraced by all companies. There's a lot of

biological development going on in most companies

right now. And you could look at it from the other

side that it may be a competitive advantage to do

that, and why would companies be interested in

sharing that with competitors?

DR. WAGNER: That's in part--I agree with

you. That's in part why I emphasize that not all

biomarkers would really be ones that you would want

to put in a collaboration effort. But there are

many biomarker areas that really are

basically--have grown too complicated and large and

expensive for any one even big PhRMA company to

tackle on their own, let alone having, you know, 20

of these companies all working at cross-purposes.

The folks that have been working on these

biomarker efforts within PhRMA would submit that

there's at least a subset of biomarkers that we

could get general agreement that a collaboration

model would benefit, but I agree that it's not

something that is necessarily the case for all

biomarker research and development.

DR. WOODCOCK: Yes, I would submit,

although I recognize all the work that's going on,

that it has not necessarily been successful in

bringing about either, in particular, more

predictable drug development or regulatory adoption

of these biomarkers. Therefore, when we published

a critical path report, quite a few firms indicated

that they would be willing to share in the

precompetitive area, which is very much like that

semiconductor example that was given. There may be

different areas of precompetitive research where

only a critical mass of effort will produce the,

you know, the results that are needed.

DR. SADEE: Yes, I think such a broad

approach is really necessary, and for those of us

who do work on looking at biomarkers from a

genomics point of view or, let's say, expression

profiling or proteomics, what you find is that you

begin with 20,000 transcripts or proteins, and you

narrow it down to a few hundred, even a few dozen.

And for each application, for, let's say, cancer,

chemotherapy outcomes, you can identify maybe a

dozen genes or proteins that are predictive.

And the combination of those, you evaluate

the best ones; what you end up with is a panel of

biomarkers that each is just maybe slightly better

than the other. There is no demarcation point.

Some may be totally unrelated to the disease. And

so, that's also coming to you, but it's not a

binary thing. It's just a complete gradation. So

you get a panel of biomarkers that just declines in

validity. And so, if you want to validate it, you

have to have a cutoff point someplace. But you do

not know which ones are going to be most predictive

in most clinical situations.

And so, I think that's really the reason

why this biomarker field has exploded, and there

are no singular solutions, and that's why we need

this type of collaboration on a very broad basis.

DR. WAGNER: I agree. And you're also

very much highlighting some of the issues

surrounding the multiplexing of biomarkers, where

one biomarker isn't worth its salt in a particular

prediction; a group of a dozen or so can be put

together in a model where the aggregate is actually

pretty good.

DR. DERENDORF: In your classification, I

think one very important aspect is the

differentiation between first in class or fifth in

class, because obviously, with the first in class

with an unknown mechanism, no clinical data, it's

very difficult to validate a biomarker. It's

impossible, as a matter of fact. And I think that

is the challenge is that you can have so many

different scenarios, it's very difficult to put

them in a systematic one, two, three, four

classification. I think we need to keep that

flexibility an creativity in this field that we can

really go any way that suits the particular case.

DR. WAGNER: Yes, I agree we certainly

want to stay as flexible as possible, but your

point also speaks to the idea that across classes,

there is the possibility of biomarkers as well.

And in diabetes, hemoglobin A1C is a gold standard

example of a biomarker that is a surrogate endpoint

100

that's accepted across different classes of

therapeutic agents, and there has really been

acceptance that new agents that, that new molecular

entities that are being--are first in class are

compared on the same standards as agents that have

been in existence for years.

DR. STANSKI: Okay; thank you, Dr.

Woodcock mentioned two important pieces of this

problem. One of them is individualizing and

improving therapy for patients; a second piece is

how do you pay for it, and how do you generate

economic incentives? And if a consortium could be

created whereby, with the right aggregation of

expertise, which included engineers to help us

learn to aggregate complex information and even

using Dr. Sheiner's concepts of multidimensional

response surfaces, because that's really what it

involves, is that this group could then both foster

the development of the research and at some point

be able to make clear recommendations to funding

agencies of what to pay for in terms of CMS or

other agencies as to when some aggregation of

101

biomarkers has reached a critical point that allows

improved therapy as demonstrated by clinical trials

and has proper statistical validity and therefore

can improve treatment; therefore, we're willing to

pay for it. That could create an incentive to pool

the intellectual capital, because ultimately, it's

the funding gate that will allow the business model

for this kind of work.

DR. WAGNER: I couldn't agree with you

more about that particular point. The reason why

the semiconductor effort was needed and why it was

successful was they worked on standards that then

could drive the expansion of their business. It's

very much of an analogous situation here, where if

there is agreement on regulatory standards both

for--within drug development and in diagnostics,

that would have a real role in substantiating a

business model.

DR. VENITZ: Okay; thank you, Dr. Wagner.

Our last presenter for today is going to

be Dr. Blaschke, who's going to give us the

academic perspective.

102

DR. BLASCHKE: Thanks. Well, when Larry

invited me to speak this morning, he suggested that

one of the things that might be helpful would be to

go into a little bit more depth on the issue of the

surrogate endpoints for HIV. We can learn

something from past experiences, and I think that

there are some important lessons to be learned.

I will say that I am a surrogate. I'm a

surrogate for Lewis Sheiner this morning, and some

of the slides that you're going to see, in fact,

will be Lewis' slides. I think he would have had a

lot of important things to contribute to this

discussion.

I think this is an important concept

cartoon that if you can't read, I'll read it for

you. It says it may very well bring about

immortality, but it will take forever to test it.

And that's a real problem with a lot of the drugs

that we're using now for chronic diseases, and I'll

give you a little bit of an academic perspective.

I'll give you my perspective on the situation.

I've been working in the HIV/AIDS area for about 15

103

years; I've been through a lot of the things that

I'll show you on the next few slides, and there are

a number of people in the audience who have also

been involved in this that I'll acknowledge as I go

through this review.

And we've seen this slide before. This is

the challenge. We need more rapid clinical

development. That was certainly true in the area

of HIV, and you've seen this before. This was the

example that was presented in the critical path

document showing that the adoption of CD-4 cell

counts and measures of viral load really led to a

speedup in the approval of antiretroviral drugs,

and this did result as a cooperative effort

involving the FDA, a number of stakeholders,

academic and industry, as I'll show you as I go on.

So what I want to spend the first part of

this talk discussing is now surrogate endpoints

were used for approval of antiretroviral drugs for

HIV infection. And it's important to go through a

little bit of the history of this, because it's not

as simple as it would like to be. The first

104

approval, in fact, based on a surrogate marker

occurred in 1992, with a drug called DDC, a

nucleoside analogue, zalcitabine, from

Hockman-LaRoche, and I've highlighted a couple of

the features of a press release that came out at

the time of that approval, which was on June 19,

1992; DDC was approved.

As noted in this release, it was the first

drug approved since the FDA had announced its

accelerated approval process, and as noted in red

on the slide here, the process incorporates the use

of surrogate endpoints to determine efficacy, and

as you'll see later on, the process allowed for

approval to be withdrawn if further review

determines the therapy was to be ineffective, and

John mentioned that point in his presentation.

So 1992 was really the first time that the

HIV RNA and CD-4 cell count was used as a surrogate

for approval of DDC. And what were the factors

that really accelerated the acceptance of it? At

this point, it was just the CD-4 cell count for

approval of DDC. Well, obviously, it was the

105

urgent need for new therapied for this fatal

illness, and one of the things in the position

paper that PhRMA has generated is the environment

here was risk-tolerant. We really didn't have

alternative therapies for HIV. We knew it was an

illness that was a fatal illness, and there was an

urgent need for developing therapies.

There were strong patient advocacy groups,

and most of us lived through that experience back

in the early 1990s, late 1980s of these advocacy

groups that were really pushing very hard for the

development and the approval of new therapies. It

led to Congressional interest in this, and

importantly, it led to some changes in FDA

regulations that allowed surrogate-based approval

when a clinical endpoint was perhaps not what we

were looking for.

I think very importantly, it also

represented a willingness of the FDA to take risks

by requiring a phase four commitment, and I would

point out that Carl Peck, who I think is probably

still in the audience, who was head of CDER at the

106

time, was also the acting head of the Division of

Antiretroviral Drugs, and Carl was very forceful in

promoting the approval of drugs based on surrogate

endpoints, and you'll see a paper that I'll allude

to in just a moment that I think represented a very

important effort on the part of the Food and Drug

Administration to look at surrogate endpoints.

And as I mentioned earlier, it really

represented a collaboration among clinical

scientists and statisticians from academia,

industry, and the government, and it wasn't all

that well-organized, as I'll try to show you. It

happened, but it didn't happen in a terribly

organized fashion, but it was a very important

point in making this actually happen.

Now, this was the paper that I was

alluding to by Stella Machado, Mitchell Gail and

Susan Ellenberg. As you'll see from the

affiliations, this is really a collaboration

between the NCI as well as the FDA. Stella was

somebody that Carl had really asked to lead this

issue of using laboratory markers as surrogates for

107

those clinical endpoints in the evaluation of

treatment of HIV infection. You'll see this was

published in 1990, and as I said, the first

approval based on these surrogate endpoints

occurred in 1992. This was a very important effort

and a very active, very busy effort to look at this

whole question.

The next ARV class that was approved were

the protease inhibitors, and they were approved in

the mid-1990s, 1995. Saquinavir was first,

followed shortly thereafter by ritonavir and

indinavir about six months later, four to six

months later. And this is an important, again,

press release that occurred at the time of the

approval of saquinavir that was provided by David

Kessler, who said that the review of saquinavir is

the fastest approval of any AIDS drug so far and

demonstrates the FDA's flexibility in situations

when saving time can mean saving lives. When it

comes to AIDS and other life-threatening diseases,

we have learned to take greater risks in exchange

for greater potential health benefits. And I think

108

again, that's a very important concept that we have

to remember, especially in something like HIV.

Carl has talked about this subsequent to

that in presentations that he has made, and I think

it's important to highlight what this meant for the

development of these protease inhibitors that I

just mentioned; that for saquinavir and indinavir

and nelfinavir, you can see from the top line there

that the development of these compounds really was

very, very short compared to the usual development

times: five, three and less than three years in

clinical development; a relatively small number of

clinical trials that were required prior to the

submission of the NDA; relatively small numbers of

patients in those trials, about 1,000 patients in

each of the NDAs, and accelerated approval, as I

mentioned before, that was based on a surrogate

endpoint and a requirement for postapproval

clinical confirmation. So it really did make a

difference.

The result of using these surrogates for

the antiretroviral drugs meant the rapid approval

109

of new drugs to treat HIV. We now have over 20

antiretroviral drugs on the market; most of them

really have been proved in record time, both the

pre-NDA time frame as well as the, obviously, the

review time for these compounds has also been quite

rapid and quite short.

It's provided, I think, incentives for

companies to develop new drugs for HIV, because the

pathway to approval is really fairly

straightforward. It's now been embodied in an FDA

guidance for antiretroviral drugs. And I would

also say that, in fact, because these drugs are so

efficacious in the treatment of HIV, approval now

without the use of surrogates would, in fact,

neither be feasible nor ethical. It would take

years and tens of thousands of patients in order to

demonstrate efficacy using clinical endpoints for

HIV infection, so this has really been a remarkable

achievement in terms of the development of

surrogate markers.

But let's go back a little bit and look at

the process that actually occurred in qualifying

110

the use of these two surrogates, that is, the HIV

RNA, plasma, CD-4 cells and surrogates, because it

really didn't occur in, as I say, in a nice, simple

fashion.

Let me go back and talk about some general

principles, and then, we'll illustrate how those

principles were, in fact, applied in the use of the

surrogate endpoints for HIV. First, is that a

surrogate endpoints qualification has to begin with

a hypothesis about the pathogenesis of the disease.

It ends with the establishment of its applicability

by using clinical trials, and what happens in the

middle? The important thing is that we have to

have basic and clinical studies of pathogenesis.

We have to have markers that are discovered about

disease progression. We have to collect data from

both preclinical and early clinical studies. I

assert that we need to develop mechanistic and

semimechanistic models and avoid the use of only

empirical models and, again, collaboration and

sharing of information in order to qualify those

biomarkers as surrogate endpoints is certainly what

111

occurred.

And I'll go through these components

pretty quickly, because I think they're fairly

well-known to everybody. We know that HIV is

caused by an infectious agent. That needed to be

discovered. It was discovered and was, I think,

well-documented to be proven as the causative agent

of AIDS, and of course, what we really needed to

show was that suppression and prevention of HIV

replication would really alter the course of the

disease.

A lot of work was put into pathogenesis of

HIV. We learned an enormous amount in a very short

period of time about the nature of HIV replication

and its interaction between HIV and the immune

system. These were extensively studied in vitro,

in animal models, and in vivo. This was largely an

academic endeavor carried out within the NIH and at

a number of different academic centers; really, a

tremendous effort that occurred in order to make

this happen, and it led to a detailed understanding

of viral structure, replication mechanisms,

112

interaction of the virus with the CD-4 cells,

involvement of co-receptors and so forth, and this

was all extremely important in the development of

therapies for HIV, and it was largely carried out

that--development of antiretroviral drugs was

largely carried out, as one would expect, within

the pharmaceutical industry, although in this case,

there was significant collaboration that occurred

with the NIH and with academia, and I would note

the role of the NCI in the development of

zidovudine and in protease inhibitor development.

So this really was a very collaborative effort in

terms of pathogenesis as well as in drug discovery.

And then, we had the discovery of these

biomarkers that I will call the biomarkers of

disease progression, and these occurred, really,

because of the efforts of multiple groups, again,

mostly from the academic side who evaluated many

possible biomarkers of the progression of HIV to

AIDS. Along the way, there were a number of

putative biomarkers that were evaluated. P24

antigen was one of the first; then came CD4 cell

113

counts and a number of other measures that were

looked at very carefully to look at disease

progression, and this occurred, really, because of

the availability and the support of a number of

cohort studies, and I've just listed half a dozen

or so here.

There were many others, both large and

small, that contributed enormously to the

information on biomarkers and on disease

progression, and that required these important

steps that John also alluded to, which was the

validation of biomarker assays such as the CD4 cell

count, the HIV RNA assays, and then, the next

important step which occurred essentially in

parallel with many of these was the collection of

that biomarker data from interventional clinical

trials, and Janet alluded to that as well.

And then, subsequent to that was the

creation of mechanistic or semimechanistic models,

which incorporated those biomarkers to see what

interventions might do to those biomarkers and

ultimately then to the qualification of those

114

biomarkers as surrogate endpoints. And this was

one of the very important studies that occurred

relatively early on in terms of trying to

understand mechanistic models for HIV infection, a

study that was done by David Ho and Alan Perelson,

published in Nature in 1995, looking at the rapid

turnover of plasma virions and CD4 lymphocytes in

HIV-1 infection.

This was done in collaboration with Abbott

Pharmaceuticals; John Leonard at Abbott

Pharmaceuticals, and what these investigators were

able to demonstrate was sort of this

multicompartmental location of HIV replication, a

very important observation, a very important

finding in terms of understanding viral

replication, and that, then, because this was an

interventional study as well, then helped

understand the role of antiretroviral drugs in the

treatment of HIV infection.

But as I said, this really didn't occur in

a nice, linear process. So you start looking at

some of those dates that I've shown you; we

115

approved, or we, the FDA, approved the first drug

in 1992, but in fact, a lot of this work with

biomarker development and the evolution and

qualification of those biomarkers into surrogate

markers ultimately or surrogate endpoints, John,

ultimately leading to a guidance on this approval

of antiretroviral drugs really occurred in much,

much later than that first approval in 1992. So

just recognize that when you have a disease like

HIV, where there's a lot of pressure to get things

done, things will happen, and they often happen in

a--as I say, a nonlinear fashion.

And I'm using this to, just, again,

recognize that here in 1997, we have a nice review

of the approach to the validation of markers for

the use of HIV RNA in clinical trials that was

done, again, a collaboration between academic, FDA

and the NIH, and even more recently published in

2000, we have a surrogate marker collaborative

group talking about a meta-analysis of the use of

RNA and CD4s prognostic markers and surrogate

endpoints in AIDS.

116

So there's still a lot of active work

going on in this field to try to really understand,

again, from a mechanistic point of view and a

pathogenesis point of view how these markers can be

used to help us better understand the therapies of

HIV and, in fact, approval of drugs.

And I show this one slide not to--because

you can read it but because I really want you to

see what a large group of people were involved, for

example, in this HIV surrogate marker collaborative

group that published that paper that I just showed

on the previous screen. So listed up here are

actually 55 people as part of that collaborative

group as both international representation from

both industry and academia.

So these kinds of things really do require

a lot of input, a lot of data, and a lot of the

people involved in this were heavily involved in

generating the data that's been used to develop

these biomarkers and surrogates in HIV.

So, now, I'm going to turn around and put

my Sheiner hat on, and I'm going to talk a little

117

bit from an academic perspective about the general

principles of biomarker use and qualification. And

this was, again, one of the slides that perhaps

Lewis showed at one of these earlier meetings; I'm

not sure, but basically, the principles here is

that to establish causality, given an empirical

association, by supporting pharmacological activity

as a mechanism, not by ruling out other causes.

And so, the evidence that would support a

pharmacologic action is that the response

correlates with temporally-varying exposure; that

causal path biomarkers change in a mechanistically

compatible direction, rate and temporal sequence,

and we saw that when we looked at viral RNA and

CD-4 in the HIV area. And as Lewis pointed out,

learning trials and analyses are well-suited to

mechanistic interpretation of time-varying data,

and independent causal evidence is still required.

Causal evidence from the same randomized controlled

trial doesn't rule out some sort of transience or

interaction. So again, the key point that he was

making there is that causal path biomarkers need to

118

change temporally in a mechanistically compatible

direction, rate and sequence.

So what are causal path biomarkers? Well,

that's illustrated on this cartoon here, and we

begin with the pathology that influences the

physiology and ultimately the disease progression.

What is next incorporated into this concept is the

idea that we have an intervention, and here, an

area that both Lewis and I were interested in was

not just to incorporate the drug but in fact

incorporate drug exposure, which represented both

pharmacokinetics as well as patient adherence in

order to get better information, so the model that

was used for the intervention represented, again,

both individual differences in pharmacokinetics as

well as patient adherence.

The pharmacokinetics, of course, lead to

time bearing plasma concentrations, and then, what

we're looking for are biomarkers that change as a

result of the changes in exposure to the drug. And

of course, what is important in terms of really

then understanding whether a biomarker is, in fact,

119

something that we really want to continue to pursue

in more detail is to look to determine whether we

see the correct temporal sequence, which gives us

some confidence that there is a mechanistic

involvement of this biomarker in the physiology and

ultimately in the clinical fact and in the disease

itself.

So let me just go back and talk a moment

about causal path biomarkers as opposed to

biomarkers in general. So causal path biomarkers

are those that serve as indicators of the state or

activity of the mechanisms that connect the disease

to the clinical manifestations. They have to be

scientifically plausible based on our current

understanding of the disease itself, and that was

certainly true with HIV/AIDS.

As knowledge increases, the confidence in

the validity of the biomarker will increase,

especially when drugs in the same class or with the

same indication affect the same biomarker, and I

think this is an important principle. If we have a

biomarker, if we have a disease, and we have a

120

biomarker that's influenced by drugs of different

structures and different class, it really increases

our confidence that this particular biomarker

represents a causal path biomarker, one that's

important in the disease and in the disease

progress itself.

More biomarkers will be useful in

developing models of drug action, and again, causal

path biomarkers need not be surrogate markers when

they're used for drug development decisions or as

confirmatory evidence of efficacy. And I won't get

off on that tangent for awhile; as you know, Lewis

and Carl Peck have been very interested in the

concept of using causal biomarkers as confirmatory

evidence along with fewer clinical trials.

So the credibility of these causal path

biomarkers does depend on the state of scientific

knowledge of the disease mechanisms, consistency of

the association with a clinically-approvable

endpoint and the biomarker; proximity of the causal

path of the clinical endpoint. Obviously, the

closer that biomarker is to the endpoint of the

121

disease, that gives us more confidence in that

biomarker and then multiple biomarkers changing in

the correct temporal sequence, and again, this

alludes to the concept of having perhaps multiple

markers that may be important rather than just a

single marker, and again, similarity of the

biomarker exposure and the clinical exposure

response when both are studied together, and all

that came, as you saw, at the bottom of that slide

from a workshop that was held by CDDS a couple of

years ago, involving Carl and Lewis and Don Rubin

as well.

And this next couple of slides and tables

just was something that appeared in a paper that

was published from that conference by Carl, Don

Rubin and Lewis in Clinical Pharmacology and

Therapeutics about a year ago, just a table of

causal path biomarkers. Just highlight a few here

that are really already either biomarkers or

becoming close to being surrogate endpoints and a

few others on this second part of the table of,

again, biomarkers that might well be those that

122

could be qualified as surrogate endpoints.

So again, establishing pharmacological

causality is really what we're trying to do here,

and what it basically means is that if we start

with an empirical association that we get from

preclinical or clinical studies, we establish

causality by directly supporting pharmacologic

activity as the mechanism and not by ruling out

other causes. It's more demanding, in fact, than

empirical confirmation, and the evidence is this

establishing the credibility of those causal path

biomarkers.

Now, this is, again, a slide from Lewis

that demonstrates that one can, in fact, gather

information about biomarkers and causal biomarkers

during phase two and phase three trials; in

particular, of course, Lewis, as I mentioned

earlier this morning, emphasized the learning

elements of the phase three trials that can be

carried out by looking, for example, as you see

from the slide here, at those surrogate prognostic

covariates, serial biomarkers. PK compliance,

123

again, is emphasized here and then the use of

model-based analysis as part of the process of

analyzing not only phase two trials but also phase

three trials.

One of the important things which I think

Lewis contributed was his concept of learning while

confirming, and I think again, this is a concept

which I hope we will see more of in the whole drug

development process. The point that he wanted to

make here was that when we look at confirmatory

trials, which we usually think of as phase three

trials, we're talking about random assignment,

placebo controls, clinical endpoints, baseline

covariates, homogeneous patients and so forth, and

that's a typical outline of a design for a phase

three trial.

However, if we add some additional

measurements, pharmacokinetic measurements in phase

three, compliance in phase three, but importantly

for the purposes of this discussion, serial

biomarkers or other covariates that we can look at,

we may increase somewhat the work involved and the

124

number of patients involved, but in fact, what we

gain is considerable. And then, if we add to that

heterogeneous patients that we begin to look at,

this individual patient therapy is, as Janet also

mentioned, we begin to have some mechanism for

looking at responders and non-responders rather

than looking at a more homogeneous group.

And then, specifically, an area that Lewis

and I have both been interested in is the use of

multiple different doses and potentially even

individual dose escalation trials to try to really

understand the dose-response relationship. And the

point, again, to be made from this slide is that we

can do this in the context of a phase three trial.

It may produce some increase in the effort involved

in the trial; it may increase some of the time

involved in carrying out those trials, but the kind

of information that we gain from this sort of

approach can really be quite valuable.

So the other point that I think needs to

be made is the issue of when is a surrogate ready.

And I've sort of alluded to that already in terms

125

of the HIV problem, but I think we're all

comfortable with the idea that the empirical

certainty is not highly necessary for drug

development decisions.

In fact, we want pharmacologic activity,

and we want mechanistic activity for those drug

development decisions and for labeling, but I think

the most important one that I want to focus on here

is that when we have great potential benefit along

with a high prior presumption of a positive

risk-benefit ration and the excessive cost of

objective evidence, those are really the kinds of

areas in which we really need to go ahead and look

at the use of alternatives to clinical outcomes in

terms of evidence for approval.

And again, what Lewis talks about is that

confirmatory really should also include learning.

And this goes back even to an APS meeting back in

1998, in which Lewis described the sort of

situation in which empiricism needed to be balanced

with the use of causal models, drug regulation

demand, certainty and information; causal models

126

are inevitably uncertain but highly informative, so

when do we use this sort of model, and when do we

use, in fact, surrogate markers at an early stage,

when lesser certainty is permissible, as in

labeling of the drug so that we can use modeling

and simulation and so forth to improve our

knowledge about labeling, but importantly, about

safety and efficacy when there's great potential

for benefit or high prior presumption, and

basically, again, a plug for the use of modeling,

that modeling certainly can yield high certainty

when we have credible models and the correct

performance of some of these tests under the null

hypothesis, and that sort of gets into this other

area that I mentioned earlier this morning about

use of alternative statistical tests when one is

analyzing trial data.

So, again, just from an academic

perspective, what do I see as some of the next

steps that we need to take? This is actually a

slide that I took from Janet's presentation a

couple of weeks ago at the ACCP meeting and what

127

she said at that meeting about what we need in

biomarker development, data pooling, synthesis,

analysis, identification of what's known and not

known and gap analysis. We heard John talk about

that, identifying what studies are needed to fill

those gaps and then doing the work and not just

standing on our heels.

And as a final comment, I think that

basically, the public wants more therapies at

reasonable prices. I think we've heard that over

and over again, and the high cost of drug

development is something that I think all of us

believe could be improved by a number of approaches

that are part of the critical path document,

including the implementation of better surrogate

marker data or surrogate endpoint data.

I don't think the regulatory issues are

necessarily any longer a major impediment. I think

the regulations are in place to approve drugs on a

surrogate endpoint basis, so we don't need to have

a lot of new legislation in order to make this

happen.

128

I think what we're hearing this morning

and what we're hearing in general is that the FDA

is very willing to move forward with new

surrogates, that we don't need to think that

there's a resistance on the part of the FDA to do

this.

Substantial collaboration among academia,

industry, and regulatory bodies will be necessary,

and I think John spoke to that very nicely. All

I'd say about academia is that unlike the FDA and

unlike the industry, we are not organized.

[Laughter.]

DR. BLASCHKE: And when I talk about

academia, who knows what I mean?

[Laughter.]

DR. BLASCHKE: There are a lot of us out

there. But I think that there are mechanisms for

getting people to come together for this kind of

important activity.

And I think what I've tried to illustrate

is that this past history with antiretroviral drugs

for HIV indicates that such collaboration can occur

129

and that it benefits all of the constituencies.

And we've already heard that there are already a

number of meaningful collaborations underway and

that we really need to encourage and support these.

So I'll just finish with this: I think

the goal that we all have is not just another

proprietary bestseller but really to get through

some major breakthroughs, and I think that this

kind of approach that we're hearing about this

morning can help along that path. And I'll stop

here, and I think we'll be ready to open it up.

Thanks.

DR. VENITZ: Thank you, Dr. Blaschke.

Any quick questions before we take a break

and start the--

DR. SINGPURWALLA: I have a comment.

DR. VENITZ: Go ahead.

DR. SINGPURWALLA: I enjoyed your

mentioning of causality, but I wanted to draw your

attention to the fact that there is a body of

knowledge called probabilistic causality which your

colleague at Stanford, Supes, specializes in. And

130

there are different interpretations. There is

something called prima facie cause; genuine cause;

and a spurious cause.

I'm wondering--and a lot of information on

causality is rarely discussed in the literature,

the philosophic literature. And I'm wondering if

the drug community is looking at that particular

angle, and if it's not, I'm recommending it.

DR. BLASCHKE: Well, I'd go back to the

comment that Janet made to your earlier comment,

and that is I think that bringing together people

with different expertise and so forth really does

add to the value, and if there's a reason for

collaboration, it's just exactly that kind of

reason, that we can't all know everything, and

there are plenty of experts out there in various

disciplines that I think we need to bring to bear

on these questions.

And I don't know them all, and I think

that's the kind of input that we need to have.

DR. DERENDORF: Very nice presentation. I

agree with everything you said. I'd like to come

131

back to this definition or desire of a causal path

biomarker. Clearly, that's the most desirable

situation. But I don't think it should be a

prerequisite for biomarkers. There are many

examples where there is no causal or no apparent

causal relationship. Think about developing of

benzodiazepenes based on EEG as a surrogate or

fentanyl derivatives, as Don has done.

So it doesn't necessarily have to be a

causal path, and it can still be operative.

DR. BLASCHKE: Well, I think we start with

empiricism. And what the academics can often

contribute to this is to move that in the direction

of understanding the mechanism or the scientific

basis for the change, whatever it is, whether it's

a change in receptor, et cetera. I certainly don't

think it's a prerequisite, but it's something that

I think we do strive for is to really understand

how something works and why it works and the way it

works.

DR. DERENDORF: I think it has to be

reproducible and predictive. I think--

132

DR. BLASCHKE: Ultimately, absolutely.

DR. SINGPURWALLA: I think your point is

very well taken, and that's why I'm drawing

attention to Supes' book on causality, where he

does cite spurious cause as an empirically observed

phenomenon which may not be the real cause, but

that's the best you can do. So again--

DR. BLASCHKE: Point taken. I agree.

DR. VENITZ: Okay; then, let's take our

break. We'll reconvene at 11:00 and start a

general discussion of the topic.

[Recess.]

DR. VENITZ: Okay; before we start the

Committee discussion, I would like to ask Dr. Lesko

to kind of give us our charge, what kind of

feedback you would like to get by the Committee.

DR. LESKO: Okay; thank you, and I'll try

my best to lay out some structure for the

discussion.

A couple of--I mean, we've heard some very

interesting presentations this morning that I think

lay the groundwork and help us tee up what amounts

133

to a new initiative in the world of biomarkers and

surrogate endpoints. Some of the thoughts I had

with regard to the Committee discussion would be

knowing what you know from the presentations, what

are your thoughts on what FDA can do to assure that

we gain some momentum behind this project and move

it forward?

Let me continue with a few others that we

can keep on the table: what does the Committee

think industry can do to facilitate the proposal

that we've tried to lay out collectively here this

morning? And finally, what can academia do?

Another issue would be what didn't you

hear today in the area of the biomarkers? What was

missing from the presentations that may be on your

mind with regard to advancing this field in the way

that we've talked about?

Dr. Blaschke in his presentation

mentioned, in a sense, a means to an end, but the

means to the end was not a linear process in the

area of AIDS. It was a process that at the end

worked out. But the question would be, and maybe

134

some discussion can occur around this, is that the

way it's going to be? Is that the way it has to

be? Or can there be a more systematic way, if we

were to think of the problem of the AIDS again and

then think about how that could be moved forward?

Is it possible in the current environment to do

that in a systematic way?

We didn't talk about this too much in the

presentations, but there was the list of biomarkers

that was in one of the slide sets that came from

the CDDS workshop on biomarkers, and there were

many biomarkers there listed side-by-side with

clinical outcomes. And one of the thought I had is

does the Committee have any specific ideas on what

we would now call biomarkers that would be in close

proximity either in a causal way or even in an

empirical way to a clinical outcome, and what could

be done to close the gap between the biomarker and

the surrogate endpoint in terms of predicting

clinical outcome?

A couple of examples of what I mean: one

example would be bone mineral density; that is, a

135

causal path biomarker for fractures and reduction

in fracture rate. Bone mineral density is used as

an approvable endpoint for a claim of prevention of

osteoarthritis, but it is not used as an endpoint

for an indication of fracture rate reduction. So

there's a gap there. What kind of data would be

needed to move biomarkers in specific therapeutic

areas to further along towards the surrogate area,

and how could those sort of gaps be identified in

terms of what we know and how we might get the

additional data?

A couple other examples: gastric acid, a

causal state biomarker; can it be advanced with

additional data, data mining, new research to

become a surrogate endpoint for additional clinical

approvals. Third example, just to stimulate some

thinking, H pylori eradication and its usefulness

in terms of duodenal ulcer recurrence and things of

that sort.

So anyway, I'll just pause here. I think

there's a couple of things on the table that maybe

we can get some discussion going, and there is no

136

boundaries on the discussion. There's a lot of

possibilities, but I just wanted to throw out a few

things for the group to think about and to kick

around.

DR. VENITZ: Okay; any comments by the

group?

Jeff?

DR. BARRETT: Larry, I wanted to address,

you know, the point about the systematic approach

relative to maybe the convoluted path. One of the

things that struck me, and we talked about this

briefly, was a lot of the emphasis is focused on

the early stage discovery processes involving

biomarker identification and evolution through the

development process, but it strikes me that another

area of focus could be from the back end as far as

working with thought weeders relative to the basis

for an approval.

I think we seldom are in areas where it's

completely unknown what is going to constitute the

basis for an approval. So from the standpoint of

looking at those study designs, criteria both

137

statistical and clinical that constitute the basis

for an approval, what would those decision makers

at that stage like to see at the earlier stage to

show some level of association between a marker to

be named and that basis for an approval.

So, you know, perhaps there could be a

meeting in the middle of the biomarkers that get

advanced at early stages relative to what is

ultimately going to potentially be a surrogate

marker. So that was one thing that struck me. And

the other thing that I thought was an interesting

point was acceptance criteria on making

generalizations. We talk about empiricism a lot as

perhaps being a dirty word here, but I think the

exploratory nature of the biomarkers has to be

there at the early stages, and it's very rare that

a company will invest in studying a biomarker

without some justification or rationale, so I

simply feel that for the most part, that is in

place, but there has to be some criteria by which

we make those generalizations, when, it's okay,

when it's not.

138

So that kind of acceptance criteria on

generalizations will help you, I think,

differentiate compound-specific mechanism-related

biomarkers versus things that may be associated

with a class.

And then, I think the other point I wanted

to make was just to be able to differentiate

between the measurement detection issues relative

to the response measurement issues associated with

observational and exploratory versus a confirmatory

test. Those pieces, I think, really need to be

compartmentalized and focused on if we're going to

move forward.

DR. VENITZ: Comment that I had in my mind

the crux, as far as it relates to coming up with

surrogate markers is this mix of using empiric

evidence and mechanistic evidence in the right mix

to convince ourselves that we have either lots of

empirical evidence on the Prentiss criteria, which

means it's going to be very difficult to actually

do that short of doing clinical outcome studies; at

the same time, what is the level of evidence that

139

you need mechanistically to convince ourselves that

those biomarkers are related to the causal

pathophysiology in the disease?

So I think one of the things to focus on,

in my mind, at least, would be what evidence, what

burden of evidence do we put on mechanistic

information? Just like we classify right now in

clinical treatment, therapeutic treatments, the

evidence to support individual treatments? Let's

come up with criteria to assess what mechanistic

evidence do we need to argue that a biomarker is

more likely than not related to the causal path? I

don't think we have had that discussion, and it may

be a matter of just going through a couple of

examples.

We had a similar discussion last year when

we talked about the pediatric decision tree, where

one of the key questions is is the disease similar?

Well, what evidence do you need to support the

contention that the disease is similar in pediatric

and in adults?

And you're getting back to the same issue:

140

short of doing empiric studies, which means it's

very expensive and very long-term doing it, what

mechanistic studies, at what level, in vitro, in

vivo, animals, what have you, do you need to

support that hypothesis? So I think we really need

to think about how we evaluate mechanistic evidence

to support transition from biomarkers to surrogate

markers no matter what the ultimate qualification

would be like.

DR. STANSKI: Yes, I think that's a very

good point. Obviously, at some level, this is

going to be marker and intervention specific.

However, we could, I think, much more exploration

of the general principles on the mechanistic side

could be done to provide a general framework, and I

think that's what we were talking about earlier,

that perhaps we can engage in a discussion about

the general framework for doing this; maybe using

examples is a good idea. What do you actually

mean? And what level of evidence is acceptable

that something is on the causal chain?

There are so many variables that probably

141

even elucidating those variables would be helpful.

I was talking to Rick Pazdur at the break, and we

talked about, you know, for the serious and

life-threatening illnesses, because we have the

accelerated approval mechanism that was spoken

about earlier, then, the tolerable degree of

uncertainty is greater. You accept greater

uncertainty, because you can pull the drug back,

and you're expecting those confirmatory studies.

I think depending on your priors, the

priors that you have are extremely important in

this analysis. And, you know, whatever we did or

did not know about HIV, we were pretty sure it was

an infectious disease, and we have a very good

model about eradication or, you know, suppression

or microbes or viruses and the relationship to

disease progression in many infectious diseases.

And so, we had very strong priors about that doing

that would be successful in helping control HIV

disease.

And that's very different in each kind of

disease area we're talking about. But a general

142

discussion of that would be helpful.

Now, getting to the other end, which was

just raised by the previous comment, on the

acceptance end, the regulatory acceptance end, I

think we also need to write specific guidance,

because a surrogate doesn't stand alone. It has to

be embedded within a trial design. There have to

be quantitative limits on what success means as far

as the duration of the trial, the kind of

observations, the analytic validation that has to

go on for the particular measurement and so forth

and so on. So there are a lot of specific,

condition-specific things that could be talked

about at a disease-specific area as well.

DR. VENITZ: Wolfgang?

DR. SADEE: I think that maybe a

compilation of a few examples would be useful in

where it's becoming very clear what we need to do

and others that are not so clear. And so, one

example would be the growth factor receptors and

tarsin kinases that are increasingly targets for

cancer chemotherapy.

143

And so, you already have--you know about

the mechanism, the expression or the mutations in

these target genes are important. In many cases, y

you can inhibit these target genes, and nothing is

happening. And so, it becomes exceedingly

important to define the criteria by which we go

forward, and that's a whole class of compounds that

comes to the fore, and I think that would be a very

useful mechanism to set up a rational approach from

the beginning, because we are only looking at the

tip of the icebergs in terms of the types of

compounds coming along the line and which ones will

be useful, and with EGFR inhibitors, only 15

percent responds, and that's correlated to certain

mutations.

But maybe not always. And so, that's one

class that requires a clear set of guidelines that

one can use in order to take maximal advantage of

this over the next five years.

DR. VENITZ: Another comment relates to

the fact that you are advocating to find more

safety markers, which I think we all would agree

144

with, but a lot of safety issues are not

necessarily related to the primary mechanism of

action of the drug. So I think most of our

discussion so far has really focused around the

mechanism of the drug and the pathophysiology of

the disease, which may or may not be related to any

safety issues.

So I think there should be a separate

initiative, if you like, to look at potential

safety markers for hepatotoxicity, and things that

are very difficult to, at this stage at least, to

predict. So maybe we can get away from the true

and tried serum transaminases. So safety markers

to me is a different domain to look at, because it

does not relate to the mechanism of action of the

drug. It may or may not relate to the

pathophysiology of the disease.

DR. STANSKI: Yes, we agree with that, and

in fact, safety biomarkers, safety markers in

general have had a different evidentiary threshold

completely than what we're talking about for

evidence of clinical benefit. So it is really a

145

different game entirely and probably can be pursued

separately but probably is equally important.

DR. WATKINS: Just to expand on that, you

could imagine a treatment for osteoporosis that you

could show was effective in 20 people with the

right genotype, with the right surrogate marker.

But until the issue of safety and particularly

idiosyncratic reactions is solved, even if the FDA

were willing to allow that to go to some

postmarketing surveillance, you know,

aftermarketing, the medical-legal environment in

the United States, I think, would be a powerful

argument for the company to go ahead and study

thousands of people for a long period of time

anyway.

So all the advantage of the efficacy

surrogate markers would be lost until there is some

kind of an understanding or progress made in safety

biomarkers.

DR. MCLEOD: Sticking on the theme of

safety, safety does represent an area that all

three of the stakeholders that were mentioned have

146

commonality. And it's probably the only area where

there is commonality across all the companies. I

mean, if you're interested in cancer, you may not

care about bone disease and vice versa. There are

some large companies that try to do everything, but

many do not.

And so, it may be as a proof of principle

for pushing this concept forward that that would be

the right framework, if nothing else to try to

standardize things, because it's starting to happen

to a bit. We, in this Committee, have spent some

time on surrogate safety markers like QT

prolongation, et cetera. And there's some--but

there's also a lot of those areas that are very

different from company to company, and maybe they

want to stay that way. But it is one area of

commonality.

On the efficacy side, people usually care

about a small number of things, and that's going to

make it very hard to get people on the same page,

even just programmatically.

DR. VENITZ: Hartmut?

147

DR. DERENDORF: Well, I'm not so sure if

it's really a difference, at least not

conceptually. I think what we're trying to do with

the biomarkers, we're trying to find something that

is easy to measure to replace it with something

that's hard to measure and do it in a faster way to

predict what we would get if we do the hard thing.

So a good example for a safety biomarker

that fits in that mold is cortisone suppression for

inhaled corticosteroids is a great predictor for

long-term osteoporosis or growth retardation in

children, studies that would take years to do; you

can do it in a single dose study and have a pretty

good idea how that product will perform in

long-term use. So I think conceptually, it's the

same thing. The issues, obviously, are different.

DR. GIACOMINI: Yes, I just want to

amplify on the safety biomarkers, I think it's a

really good model for bringing together a

consortium of people from academia, FDA and

industry. First of all, if it's a rare adverse

event, it requires large populations, large

148

clinical populations. I think Paul is

participating in the drug-induced hepatotoxicity

NIH-sponsored network, right? And that's one that

requires a lot of people together, but this could

bring together industry, academia, and all of that

around safety biomarkers, so I just want to second

that.

I also want to say on the efficacy

biomarkers, one thing I think that FDA could do is

bring together people from different

disease-related or treatment-related groups to talk

about the issues in those particular

treatment-related groups, because I do feel that

the biomarkers in each group may be very different,

and it would be more conceptual to think about them

in group-by-group, disease-by-disease.

DR. VENITZ: Other comments?

DR. LESKO: Yes, just to throw out another

thought, and it actually somewhat relates to our

discussion yesterday of predictive tests in the

context of irinotecan. At some point in time,

we're going to have to come face-to-face with the

149

statistical issues that revolve around the

biomarker and the predictiveness of it. And

yesterday, when we were talking about a

pharmacogenetic test, we were talking about the

probabilistic nature of the test and attributes of

the test that convey its ability to predict

something. We talked about sensitivity,

specificity, predictive values, likelihood ratios,

et cetera.

And there seemed to be some common ground,

or at least we could probably, with more

discussion, reach a common ground on the

performance of a test that would be generally

acceptable. So it gets me around to the question:

is an approach or a framework that has been used

for the predictiveness of diagnostic screening or

other types of tests appropriate for biomarkers?

Or is the statistical sort of framework for what

we're talking about in place already, or are there

needs for new statistical models to deal with this

problem?

Dr. Woodcock mentioned the Prentiss

150

criteria. That was one model. But do we need to

be thinking about new statistical approaches, new

ways of expressing predictiveness of biomarkers, or

are we sort of satisfied with where we are on that,

and that may be for Marie and David.

DR. DAVIDIAN: Well, there is a lot of

work in the statistical literature; there has been,

in fact, recently, as we speak, in trying to sort

of refine the--the Prentiss criteria are, let's

face it, very stringent criteria, but they do lay

out the, I think, what's the key issue for a

surrogate, which is that you want the effect of the

treatment on the surrogate to--the effect of the

treatment on the clinical endpoint to be seen when

you paw the treatment, you know, through the

surrogate.

So, I mean, I think that is the key issue

there. Now, how you go about quantifying that and

characterizing that, I think, is what you're

talking about. How do you actually do that? And

there's been various proposals that are out there

to do so. I think to try to get a perfect

151

surrogate is impossible, as has already been

mentioned.

But I think in the context of this sort of

discussion here and bringing in mechanistic

considerations and so on, I think there would be

additional work to be done, and I think bringing

statisticians in from that point of view would be a

good thing. I mean, most of the work in the

statistical literature now, in fact, all of it is

totally empirical. It's trying to come up with

empirical models and ways of characterizing

surrogacy and based totally empirically.

So I think that's where the new work can

be done.

DR. JUSKO: The discussions this morning

were extremely good and very informative, and as a

member of this Committee, I very much encourage all

of the participants to continue evolving this area.

One thing that is admirable about what companies do

is when they screen drugs, they often use receptor

systems and animal studies, and eventually, they

get to a study commonly called proof of concept, a

152

phase 2-A type of study, where they then may try to

utilize a vast array of potential biomarkers to see

whether or not the drug has any activity that's in

concert with its basic mechanism of action that

they understand it to be. And then, many more

studies are pursued after that.

One thing that's frustrating to me in

academia is this huge vault of information

accumulated by companies in diverse areas,

including all of these kinds of biomarkers that

they've measured. The FDA may be aware of part of

it, but there's probably an immense amount of

information that's lost to the general scientific

public that could be better harvested if there was

some concerted activity through this type of

organization that's being proposed here.

So I just want to voice that degree of

frustration and encouragement towards collecting

some of this information in a more systematic

manner.

DR. SINGPURWALLA: I was going to respond

to your question. I think I've already said a few

153

things, and I'm just going to repeat them.

You talked about modeling and simulation

in one of your slides, MNS. That's the kind of

stuff you hear at the Pentagon all the time, and

that's good.

[Laughter.]

DR. SINGPURWALLA: I think one of the

things that you may consider in this context of

markers is the stochastic process models. You

don't want to look at them in a very traditional

statistical framework. You want to look at it in a

dynamic way. Markers evolve dynamically; diseases

evolve dynamically. They're correlated and what

kind of inference you should do and what kind of

confirmatory studies are needed is something that

needs to be researched and worked.

I also hear the word mechanistic models,

mechanistic considerations. I would hope that

you're looking carefully into Bayesian methods,

which combine both the knowledge of medicine and

whatever have you with empirical evidence and try

to put the two together.

154

And lastly, I would suggest that when you

have these panels of people looking at various

things, I would encourage you to go out of the

normal umbrella and look into other disciplines.

And I just don't have in mind engineers. I

strongly suggest you look into the philosophers.

They write a lot on causality; in fact, there are a

lot of books on causality written by philosophers.

I think also, you should look at ethicists

and people who look at moral issues. So I think

you should expand your umbrella of expertise to

include some other cultures and characters.

DR. BLASCHKE: I want to come back to a

question that you raised, Jurgen, and also a point

that Marie made. And that is maybe one of the

principles of surrogate endpoints and part of this

qualification process is that you have an

advantage, in fact, if there are multiple drugs to

treat the same condition. If you're getting the

same effect when you're using drugs, working

through what are believed or hypothesized to be

different mechanisms, yet at some point, their

155

effect on a surrogate is consistent and also then

consistent with a clinical outcome, it gives you a

lot more confidence that this surrogate is, in

fact, not an epiphenomenon of some sort but, in

fact, is a causal path marker that could be used as

a surrogate endpoint.

So perhaps when we're trying to think of

sort of general principles and so forth of things

that make a biomarker more likely to qualify as a

surrogate endpoint, I think the fact that it

could--and that could even work with new chemical

entities. I mean, even if it's a first in class.

I mean, somebody mentioned earlier that maybe it's

hard for a first in class compound to be approved

on the basis of a surrogate endpoint, but in fact,

no. If that surrogate has been proven for several

other drug classes, it may even be a stronger

evidence that this new drug about which maybe has a

new mechanism is ultimately working through that

same pathway to produce the beneficial effect in

the disease.

DR. STANSKI: Bill Jusko mentioned the

156

sequestering of information. I'd like to ask

people who work within the pharmaceutical industry

to what degree is this precompetitive knowledge and

prevention of sharing to do patent issues and

competitive advantage something that can be

overcome? Or is that just a reality of a

for-profit industry, or for the sake of moving this

concept forward and having more efficient drug

development, how can that barrier be broken?

DR. VENITZ: Would anybody care to

comment, or was this a rhetorical question?

DR. STANSKI: Well, someone in the

industry must think of this and to be able to

respond to it, I'd hope.

DR. VENITZ: Go ahead. Can you introduce

yourself?

MR. WEBSTER: I'm Chris Webster. I'm

director of regulatory strategy and intelligence

from Millennium, and I'm speaking for myself here.

I'm not speaking for the industry, but perhaps my

views are, because I've been involved in some of

the working groups, may be useful to you at this

157

point.

Obviously, everybody is very aware of the

topicality of this issue relating to the

publication of clinical trials, and there has been,

as you know, an initiative published by PhRMA to

put up clinical trial data in a public place for

patients and physicians and others to see it.

I think what you're talking about here is

something more far-reaching than that, and it's

not, I think, a--you know, this is not the first

time I think the industry has become aware of it.

I'll refer you, for example, to the comments of Dr.

Kalif at the Science Board last April, where he

again touched on this point, and so I think we are

aware of it.

I think that it's probably not impossible

to be done, but I think that there would need to be

some kind of really high level working group to

really look at very sensitive and difficult issues

related to intellectual property and ways in which

information could be perhaps shared in an anonymous

way, in a generic way so that it wasn't identified

158

with particular companies or particular drugs but

perhaps could be useful for the purposes of

scientific research.

And perhaps some degree of parallel to

that is the creation of voluntary data submissions

for pharmacogenomic data which, of course, was

published by the agency just about a year ago now,

and so perhaps, that might be to some extent a

model for this.

I think it's very difficult, though; I

don't want to project any illusions about this that

it would be easy, but I think perhaps it's a

conversation which the industry might be ready to

have. Thank you.

DR. LESKO: Yes, Chris, while you're

there, you did mention the voluntary genomic data

submission pathway that the agency created, which

was kind of a groundbreaker in many ways, and I

know you were part of that with the working group

and the workshop. SO, really, my question is do

you see a difference between a similar pathway for

nongenomic biomarkers as we set up for that

159

particular reason? We set it up for genomic

biomarkers, but is there any reason why it couldn't

be utilized for getting some of the information

that's sequestered in some of these areas to submit

to a group separate and apart as we've set up the

interdisciplinary pharmacogenomic review group to

do the evaluation of these and begin to synthesize,

really, a greater association with the clinical

outcomes and so on.

MR. WEBSTER: Yes, I think that's why I

suggested it could be a model, and personally, I,

myself, don't think that there is a qualitative

difference there. But I think that in the sense

that genomics is a new science, a new technology;

its application to drug development versus drug

discovery is something that is perhaps newer; and

also, the fact that there was kind of this safe

harbor concept around the submission of data, all

of those were, I think, if you like, material

facts.

Now, as I say, I think it perhaps is a

model which we could explore, and if, perhaps, in

160

the context of this morning's discussion, the

agency were to create some parallel to the IPRG but

which allowed companies to come in and discuss a

broader context of biomarker research with the

agency, and if that was part of the entire, if you

like, game plan, then, I think that might be a

lever to move this forward.

DR. VENITZ: Wolfgang?

DR. SADEE: There are actually companies

out there that make their business to compile vast

amounts of data of that very nature, for instance,

Iconics. And you not only have array data; you

have 500 assays available for the 500 common drugs

used, and so, that's a business model by itself.

And I would strongly suggest that we get this type

of folks involved in the process, because they have

already integrated much of the information one

would like to use, actually.

DR. BARRETT: Larry, I wanted to come back

to your initial question about the statistics. In

the discussion yesterday, when we got to look at

some parameters associated with sensitivity,

161

specificity, and predictive value, my comment to

your question was I don't think I've seen enough of

that across different therapeutic areas to where

you could make an assessment of that, and they seem

to be very reasonable and applied metrics.

The question I had is, you know, it would

seem to be a good example where you could use some

modeling and simulation to look at what would those

metrics look like if you had good association or

bad association, if you had a high prevalence rate

or low prevalence rate, as well as if the

pharmacokinetics were predictive of the biomarker

or not.

It would seem to be that you could look at

the performance of these characteristics almost

independent of their application to define whether

or not they were reasonable to look at. But to

answer your question, I don't think we've seen

enough of it in a standardized manner, which is,

again, part of the problem of having enough of a

data set to look at across therapeutic areas.

DR. DERENDORF: I liked the proposal that

162

we've heard many times this morning on

collaboration between industry, FDA and academia.

But I think there is a big problem coming our way,

and that is that we are not training enough

scientists in this field. There is a shrinkage of

clinical pharmacology programs, pharmacometrics

programs, a lack of funding in academia, and this

will be a problem. And I think industry really--I

feel it's in their own interest to maybe help

academia a little bit in establishing systems, how

we can provide the training. It's going to be a

problem otherwise.

DR. WATKINS: Sorry, just to bounce around

a little bit, but in the issue of getting companies

to cooperate and sharing data, I'm aware of one

initiative which is the International Life Sciences

Initiative that's been going on for several years

where participating companies are submitting

preclinical toxicity data and safety data in man in

a blinded fashion, creating a database to look at,

you know, markers of predictivity from animals into

man, so that there's at least one precedent for

163

that.

The other thing I thought I would just

mention is what Cathy brought up, which is the

drug-induced liver injury network as a potential

for collaboration with industry and the agency.

This is funded by the NIH and the NIDDK in

particular. And these five centers, which cover

about 12.5 million lives, are prospectively

enrolling into the study people who have clinically

significant toxicity due to any drug. And in

addition, they're getting genomic DNA and

immortalizing lymphocytes and getting serum and

liver wherever possible; we're also creating a--and

I'm chair of the steering committee--creating a

registry, and the people agree to be contacted up

to 20 years to undergo genotype/phenotype

correlation studies in focused clinical centers so

that, you know, that seems to me a very nice

potential model for industry to participate;

obviously, we'll be finding out things about their

drugs before they know them, and I'm sure we'd be

open to any kind of collaboration that could come

164

down the pipe.

DR. D'ARGENIO: Yes, this comment also has

to do with databases and biomarkers. One of the

real challenges in developing these causal paths

that are mechanistic-based biomarkers is

understanding them and disease progression. And

that is a real challenge, but there certainly are

data out there on just general models of disease

progression, at least one would think, in the

postmarket area, and those data would help inform,

you know, the relevance of biomarkers to follow

disease progression.

DR. CAPPARELLI: I think the last two

comments also focus back on the issue of looking at

the surrogate marker going backwards as well. You

know, one of the issues, even with the disease

state, this is a dynamic issue. You know, looking

at HIV as the example, working in pediatrics, the

surrogates don't work exactly the same.

And so, I think there will be sort of an

evolutionary process of understanding the

relationship, and that is a huge data mining and

165

iterative process of working that forward, so, you

know, the concept of looking at some key areas,

especially ones where the clinical endpoint takes

so long to develop, and we may have good

mechanistic reasons to think we have something that

occurs rapidly that we can measure.

And that was the other aspect of HIV, that

the whole research really showed that it wasn't

such a static disease that takes a long time, and

we can see the effect of drugs very rapidly, and

that time differential was, I think, extremely

important in bringing that forward from an industry

and academic standpoint to utilize these tools.

DR. VENITZ: Any other comments, perhaps

on the recommendations that Dr. Wagner talked about

with respect to setting up committee structures to

manage the process?

[No response.]

DR. VENITZ: Any other comments?

[No response.]

DR. VENITZ: Then, I guess, I'm looking at

you, Larry, as the final comment.

166

DR. LESKO: So, I guess that means it

brings us to the end of the road--

DR. VENITZ: Right.

DR. LESKO: --for this meeting, and the

closure is stated as a summary of recommendations,

and before I do that, I'd like to not be remiss in

acknowledging the people that helped put this

committee meeting together, and I'm specifically

referring to Hilda Scharen, who's sitting next to

Dr. Venitz; Karen Summers, who was behind me for

most of the meeting, I guess keeping me in line;

I'm not sure why, and Bob King, who has been

helpful in getting all these materials out to the

Committee and my colleague to the left, Peter Lee,

who did a lot of the coordination of it.

We didn't make it easy for this crowd this

time around. We really imposed upon their

administrative support, and I really appreciate

their flexibility in meeting deadlines and going

the extra mile to get everyone who participated

cleared appropriately and within the laws.

As far as the summary of recommendations

167

goes, I suppose the summary is really captured by

the voting that the Committee did on the yes and no

questions that we posed yesterday in particular,

and there really isn't much more to comment on

those questions, because I think they did speak for

themselves, although the discussion in between the

various questions were very useful to us in

illuminating the vagaries that we're dealing with

in some of these areas, in particular, the area of

transporters and multiple inhibitors.

What was particularly useful to us was

what I said yesterday: voting aside, the value of

this meeting, the added value of this meeting is

really in the areas that surround the discussion of

the issues. And the discussions in this Committee

meeting were very helpful to us in helping shape

our way of thinking about pharmacogenetics, drug

interactions and biomarkers, and I think that's why

we came here together.

I really enjoyed this meeting. It was

quite an interesting intellectual debate. The

members, even late last night until 6:00, were

168

fully engaged. I did miss the after-meeting

discussion last night, but I'm sure it was also

very intellectual, but you were willing to work

hard and late night, and I want to express my

thanks on my behalf, and as Dr. Woodcock had to

leave to go downtown, she asked me to express her

appreciation to the hard work that the Committee

did on her behalf as well.

Well, I think this meeting, we really teed

up some new issues and some challenging topics,

some of which, of course, haven't been resolved.

We didn't expect that: transporters, the

biomarkers, the surrogate endpoints, and I hope all

of you really look forward to further meetings,

where we hope to discuss these issues in more

details as our thoughts come together and as more

data become available.

So in closing, I would like to express my

thanks, thanks on behalf of the Clinical

Pharmacology team that worked to bring the topics

to you. Of course, all of the presenters and to

all of you for your time and public service and

169

providing us the intellectual firepower that we

need to resolve these issues. So have safe travels

home; thank you, and I'll turn it back to the

chair.

DR. VENITZ: I agree. I thank everybody

for participating; wish everybody a safe trip home,

and the meeting is adjourned.

[Whereupon, at 11:42 a.m., the meeting was

concluded.]

- - -