1

DEPARTMENT OF HEALTH AND HUMAN
SERVICES

FOOD AND DRUG
ADMINISTRATION

CENTER FOR DRUG EVALUATION AND
RESEARCH

ADVISORY COMMITTEE FOR
PHARMACEUTICAL SCIENCE

Advisors and Consultants Staff
Conference Room

2

PARTICIPANTS

Arthur H. Kibbe, Ph.D., Chair

Hilda F. Scharen, M.S., Executive
Secretary

MEMBERS:

Gerald P. Migliaccio, Industry
Representative

Marvin C. Meyer, Ph.D.

Patrick P. DeLuca, Ph.D.

Charles Cooney, Ph.D.

Melvin V. Koch, Ph.D.

Cynthia R.D. Selassie, Ph.D.

Nozer Singpurwalla, Ph.D.

Jurgen Venitz, M.D., Ph.D.

Marc Swadener, Ed.D., Consumer
Representative

SPECIAL GOVERNMENT EMPLOYEES:

Paul H. Fackler, Ph.D.

Gordon Amidon, Ph.D., M.A.

Judy Boehlert, Ph.D.

Leslie Benet

Charles DiLiberti

Laszlo Endrenyi

FDA:

Gary
Buehler, R.Ph.

Ajaz
Hussain, Ph.D.

Helen Winkle

Lawrence Yu, Ph.D.

3

C O N T E N T S

Call to Order, Arthur
Kibbe, Ph.D. 4

Conflict of Interest Statement,

Hilda Scharen, M.S. 4

Bioequivalence of Highly Variable Drugs,

Why Bioequivalence of Highly Variable
Drugs is an

Issue,

Charles DiLiberti, M.S. 11

Highly Variable Drugs: Sources of Variability,

Gordon L. Amidon, Ph.D. 35

Clinical Implications of Highly Variable

Dr. leslie Benet 60

Bioequivalence Methods for Highly
Variable Drugs,

Laszlo Endrenyi, Ph.D. 81

Bioequivalence of Highly Variable Drugs
Case

Studies,

Barbara Davit, Ph.D. 99

FDA
Perspectives, Sam Haidar, Ph.D. 125

Bioequivalence of Highly Variable Drugs
Q&A,

Dale Conner, Pharm.D. 130

Bioinequivalence: Concept and Definition,

Statistical Demonstrations of
Bioinequivalence,

Donald Schuirmann, M.S. 182

Update--Topical Bioequivalence,

Establishing Bioequivalence of Topical

Dermatological

Products, Robert Lionberger,
Ph.D. 225

Future Topics--Nanotechnology, Nakissa
Sadrieh,

Ph.D.
257

Conclusions and Summary Remarks, Ajaz
Hussain,

Ph.D. 270

4

1 P R O C E E D I N G S

2 Call to Order

3
DR. KIBBE: By the clock on the
wall, I

4
think we are at

5
electronics are working so we will be in good

6
shape. We need to start off with
the reading of

7 the
conflict of interest statement.

8 Conflict of Interest
Statement

9 MS. SCHAREN: Good morning.
I am Hilda

10
Scharen. I am the executive
secretary for the

11
Advisory Committee for Pharmaceutical Science and I

13
interest statement for the committee.

14
The following announcement addresses the

15
issue of conflict of interest with respect to this

16
meeting and is made a part of the record to

17
preclude even the appearance of such at this

18
meeting.

19
Based on the agenda, it has
been

20
determined that the topics of today's meetings are

21
issues of broad applicability and there are no

22
products being approved at this meeting.
Unlike

23
issues before a committee in which a particular

24
product is discussed, issues of broader

25
applicability involve many industrial sponsors and

5

1
academic institutions. All
special government

2
employees have been screened for their financial

3
interests as they may apply to the general topics

4 at
hand.

5
To determine if any conflict of interest

6
existed, the agency has reviewed the agenda and all

7
relevant financial interests reported by the

8
meeting participants. The Food
and Drug

9
Administration has granted general matters waivers

10 to
the special government employees participating

11 in
this meeting who require a waiver under Title

12
XVIII,

13
A copy of the waiver statements may be

14
obtained by submitting a written request to the

15
agency's Freedom of Information Office, Room 12A-15

16 of
the

17
Because general topics impact so many

18
entities, it is not prudent to recite all potential

19
conflicts of interest as they may apply to each

20
member and consultant and guest speaker.
FDA

21
acknowledges that there may be potential conflicts

22 of
interest but, because of the general nature of

23 the
discussion before the committee, these

24
potential conflicts are mitigated.

25
With respect to FDA's invited industry

6

1
representative, we would like to disclose that

2
Gerald Migliaccio is participating in this meeting

3 as
an industry representative, acting on behalf of

4
regulated industry. Mr.
Migliaccio is employed by

5 Pfizer. Dr. Paul Fackler is participating in this

6
meeting as an acting industry representative. Dr.

7
Fackler is employed by Teva Pharmaceuticals U.S.A.

8
In the event that the discussions involve

9 any
other products or firms, not already on the

10
agenda, for which FDA participants have a financial

11
interest, the participants' involvement and their

12
exclusion will be noted for the record.
With

13
respect to all other participants, we ask in the

14 interest
of fairness that they address any current

15 or
previous financial involvement with any firm

16
whose product they may wish to comment upon. Thank

17
you.

18
DR. KIBBE: Thank you, Hilda. Just so

19
that our audience knows who all is here, I would

20
like to ask everybody to introduce themselves and

21
give their affiliation. We will
start with Dr. Yu.

22

23
DR. YU:

24
Science, Office of Generic Drugs, Office of

25
Pharmaceutical Science, CDER, FDA.

7

1
DR. BUEHLER: Gary Buehler,
Director,

2
Office of Generic Drugs, Office of Pharmaceutical

3
Science, CDER.

4
DR. HUSSAIN: Ajaz Hussain, Deputy

5
Director, Office of Pharmaceutical Science, CDER.

6
MS. WINKLE: Helen Winkle,
Director,

7
Office of Pharmaceutical Science, CDER.

8
DR. AMIDON: Gordon Amidon,
University of

9 Michigan.

10 DR. VENITZ: Jurgen Venitz, Virginia

11 Commonwealth University.

12 DR. SELASSIE: Cynthia Selassie, Pomona

13 College.

14 DR. BOEHLERT: Judy Boehlert, and I have

15 my own pharmaceutical consulting business.

16
DR. SWADENER: Marc Swadener,
consumer

17
representative, retired from University of

18

19
DR. KIBBE: I am Art Kibbe and I
am

20
Professor of Pharmaceutical Sciences at Wilkes

21
University.

22
DR. MEYER: Marvin Meyer, formerly

23

24

25
DR. SINGPURWALLA: Nozer
Singpurwalla,

8

1

2
DR. KOCH: Mel Koch, the Director
for the

3
Center for Process Analytical Chemistry at the

4

5
DR. COONEY: Charles Cooney,
Professor of

6
Chemical and Biochemical Engineering at MIT.

7
DR. DELUCA: Pat DeLuca,
University of

8

9
MR. MIGLIACCIO: Gerry Migliaccio,
Pfizer.

10
DR. FACKLER: Paul Fackler,
industry

11
representative, Teva Pharmaceuticals.

12
DR. KIBBE: Thank you. We are going to

13
start this morning and Dr. Yu will set us up for

14 our
discussion.

15
Bioequivalence of Highly Variable Drugs

16
DR. YU: Good morning. My slides I guess

17 are
in a different file so I will give my

18
introduction without the slides.

19
Dr. Kibbe, Chair of the FDA Advisory

20
Committee for Pharmaceutical Science, members of

21 the
FDA Advisory Committee for Pharmaceutical

22
Science, distinguished speakers, distinguished

23
guests and distinguished audience, I am

24
Yu. I am Director for Science,
Office of Generic

25
Drugs, Office of Pharmaceutical Science, CDER, FDA.

9

2 and
privilege to introduce to you the first topic

3 of
bioequivalence, bioequivalence of highly

4
variable drugs. The objectives of
this discussion

5 are
to explore and define bioequivalence issues of

6
highly variable drugs, to discuss and to debate

7
potential approaches in resolving them,

8
specifically the pros and cons of the solutions and

9 the
benefits and limitations of these potential

10
approaches.

11
The bioequivalence issues of highly

12
variable drugs have been discussed in many

13
conferences and meetings nationally and

14
internationally. The issue is
obvious because of

15 the
high variability of the drugs or drug products

16
that require a large number of subjects or

17
volunteers in order to pass the confidence interval

18 of
80-125 percent. Despite many, many
discussions,

19
despite many, many publications in scientific

20
literature, to date there is no consensus and no

21
solutions have ever been reached.
In fact, there

22 is
no regulatory definition with respect to the

23
high variability drugs or drug products.
So, there

24 are
various approaches in resolving this in the

25
scientific literature, for example, expansion of

10

1 the
bioequivalence limits; for example, using

2
scaling approaches.

3
We have invited a panel of distinguished

4
speakers this morning to discuss this issue related

5 to
the bioequivalence of highly variable drugs from

6
various perspectives, from practical difficulties

7 of
bioequivalence of highly variable issues, from

8
mechanistic understanding of what causes the high

9
variability of drug or drug products, from

10
understanding of different approaches to resolve

11
understanding of clinical implications why high

12
variability drugs are safer, from case studies and,

13
finally, from regulatory options.

14
At the end of these presentations you will

15 be
asked to discuss or address the following

16
questions. First, what is
actually the definition

17 for
highly variable drugs or drug products?

18
Second, with respect to expansion of

19
bioequivalence limits, what additional information

20
should we gather in order to answer this question?

21 We
also ask you to comment on scaling approaches.

22
With this introduction, I want to turn the

23
podium over to our first speaker, Charlie

24
DiLiberti. Charlie?

25
Why Bioequivalence of Highly Variable Drugs

11

1 is an Issue

2
MR. DILIBERTI: Thank you, Dr.
Yu. Before

3 I
start I need to disclose the potential conflict

4 of
interest in that I am employed by Barr and I am

5
also a shareholder and option holder in the firm.

6
Also, before I get into the actual

7
discussion I would like to say that in the context

8 of
preparing this presentation I had numerous

9
discussions with many of my colleagues in the

10
industry and, based on the feedback that i got from

11
them, it seems to me that the views that I am about

12 to
portray in my presentation are quite widely held

13 in
the industry.

14
[Slide]

15
With that, let's start off with the

16
definition of highly variable drugs.
Oftentimes,

17
highly variable drugs are defined in the context of

18
within-subject variability in terms of a

19
bioequivalence study. I would
like to take it one

20
step further and look at variability within the

21
patient and what does this high level of

22
variability mean to an individual patient taking

23 the
drugs.

24 Commonly, the often used definition
of

25
highly variable drugs is those drugs whose

12

1
intra-subject or, as I characterize it here as

2
intra-patient, coefficient of variation, or CV, is

3
approximately 30 percent or more.
I will use that

4 as
my guideline for the rest of this presentation.

5
[Slide]

6
What are the current criteria?
Just very

7
briefly, for bioequivalence they involve a

8
comparison between test and reference product,

9
involving the natural log transformation of the

10
data. The current criteria are
that the 90 percent

11
confidence intervals around the geometric mean

12
test/reference ratios have to fall entirely within

13 the
range of 80-125 percent.

14
These criteria really apply to all drugs

15
here, in the

16
variability of the drugs. These
criteria do have

17 other
implications. For example, they can be
used

18 by
innovator and, for that matter, generic firms to

19
justify a substantial formulation change so it is

20 not
just in the context of approving a generic.

21
[Slide]

22
This really speaks to the
crux of the

23
issue with highly variable drugs in that it

24
portrays the number of subjects that you would have

25 to
plan on using in a two-way crossover

13

1
bioequivalence study given a particular

2
intra-subject CV. You can see
that for very low CV

3
drugs the number of subjects required is fairly

4
small and quite manageable from a practical

5
standpoint but, as the CV increases, you can see

6
that the number of subjects required can increase

7 to
quite large numbers, possibly in the hundreds.

8
[Slide]

9
Why do we possibly need alternative

10
criteria for highly variable drugs?
Well, first of

11
all, we have an ethical mandate to minimize human

12
experimentation. Second of all,
the prohibitive

13
size of some bioequivalence studies for some highly

14
variable drugs impacts on the availability of a

15 generic
version of that drug, which may mean that

16 in
the absence of a generic many Americans can't

17
afford the reference product so they may go either

18
untreated or they may be subdividing their doses

19
contrary to the prescription.

20
Also, changing criteria will reduce the

21
number of participants in the BE studies and I

22
think it can't be done without compromising the

23
safety and efficacy of the product. Also, there is

24
experience elsewhere in the world with criteria

25
other than 80-125 percent.

14

1
[Slide]

2
This slide shows some of the

3
bioequivalence criteria in other countries and

4
regions in the world. These are
not specific to

5
highly variable drugs and in many cases they don't

6
apply necessarily to all drugs.
That is why I have

7
"most drugs" or "some drugs" listed here. But,

8
certainly, there is experience with certain drugs

9 in
these different regions with confidence

10
intervals that are either wider than 80-125 or, in

11 the
case of

12
confidence interval criterion, just a point

13
estimate criterion.

14
[Slide]

15
What types of drugs are highly variable?

16
Well, the types of drugs really cut across all

17
therapeutic classes and include both new and older

18
products. The potential savings
to American

19
consumers could possibly be in the billions of

20
dollars if generics are approved.
In saying this,

21 I
want to be clear that the bioequivalence issues

22 for
many of these drugs are not the only barriers

23 to
getting a generic. In some cases there
might be

24
patent issues or formulation issues as well, but

25
still the bioequivalence issues do represent some

15

1
sort of a barrier.

2
What are some examples? This is a
very

3
brief list and the list can go on and on but just

4 to
give you some kind of representative examples of

5
drugs that cut across many therapeutic areas, some

6 of
which are on-patent, some off-patent, just to

7
give a flavor.

8
[Slide]

9
Another issue is that as of last year we

10 now
have to meet confidence interval criteria for

11 fed
bioequivalence studies. So now the
variability

12
under the fed state is of concern.
There is

13
generally very little information available on the

14
variability of drugs in the fed state, and we have

15
found that some drugs do show more variability

16
under fed conditions than under fasting conditions,

17
leading to the potential for bioequivalence

18
failures because they may be under-powered. What I

19 am
trying to get across here is that because of the

20
lack of information on many drugs under fed

21
conditions, there may in fact be many more highly

22
variable drugs than we are led to believe.

23
[Slide]

24
Why aren't the current criteria

25
appropriate for some highly variable drugs? Well,

16

1 I
will start off by saying that the current

2
criteria are, I believe, appropriate for drugs with

3 low
to moderate variability because the

4
dose-to-dose variability that a patient would

5
experience is comparable and consistent with the

6
width of the criteria.

7
However, in the case of highly variable

8
drugs this is not true where the dose-to-dose

9
variability experienced by a patient may often be

10
much larger than the width of the criteria. I will

11
illustrate this point later on with some graphs.

12
Highly variable drugs are oftentimes wide

13
therapeutic index drugs. In other
words, they have

14
shallow response curves and wide safety margins. I

15
want to qualify this statement by saying when I say

16
highly variable drugs, highly variable in a patient

17
with respect to the parameter that is variable. If

18 a
patient experiences high variability, that means

19
that the drug is safe and effective despite this

20
wide variability in the patient.
Therefore, I

21
believe that modifying bioequivalence criteria on

22
highly variable drugs to reduce the number of

23
participants in bioequivalence studies could be

24
accomplished while still maintaining safety and

25
efficacy assurance.

17

1
[Slide]

2
Different highly variable drugs may

3
require different approaches. One
size may not fit

4
all. As we can see from the
earlier power graphs

5
that I had plotted, obviously the number of

6
subjects required for a drug with, say, 30 percent

7
coefficient of variation is very different from the

8
number of subjects required for a drug with, say,

9 70
percent intra-subject CV. And, there are
other

10
considerations that we have to take into account.

11
[Slide]

12
Probably one of the more important

13
considerations is whether the drug accumulates in a

14
patient at steady state. Let's
first take the case

15 of
a drug that does not experience significant

16
accumulation to steady state in a patient. These

17 are
typically short half-life drugs, in other

18
words, short half-life with respect to the dosing

19
interval. Here are some
examples. We could

20
possibly consider some sort of modification to the

21
criteria for both AUC and Cmax because an actual

22 patient
would experience significant dose-to-dose

23
variability for both Cmax and AUC because neither

24 is
smoothed out at steady state. Therefore,
the

25
drug could be considered to exhibit a wide

18

1
dose-to-dose variation in blood levels irrespective

2 of
chronic dosing.

3
The same sort of logic could potentially

4
apply to a highly variable drug that is not dosed

5
chronically. One particular
application of the

6
scenario of a relatively short half-life drug that

7
does not undergo accumulation might be the case of

8 a
parent drug with a short half-life and high

9
variability where there is also a metabolite that

10
needs to be measured which has a much longer

11
half-life and low variability. I
could easily

12
envision the case where the confidence interval

13
criteria are somehow modified to accommodate the

14
higher variability of the parent drug but, in the

15
same compound, the current 80-125 criteria could be

16
applied to the metabolite.

17
[Slide]

18
Now let's look at the case of accumulation

19 to
steady state. Typically, this is a case
where a

20 drug is used chronically and with a half-life
long

21
relative to the dosing interval so there is some

22
accumulation going on. Here are a
few examples.

23
In this case, because the accumulation

24
process will tend to reduce the fluctuation in AUC

25 and
Cmax, both at steady state, actually in

19

1
essence, the drug to a patient may not really be

2
highly variable because the variability may be

3 small at steady state. However, the Cmax and AUC I

4
think need to be looked at in a different light.

5 At
steady state the test/reference ratio for two

6
drugs, assuming linear accumulation, will be about

7 the
same as the test/reference ratio that we see in

8 a
single dose study because the accumulation

9
process preserves that test/reference ratio.

10
However, for Cmax, generally speaking, the

11
test/reference ratio that we see at single dose

12 conditions
will be the most extreme and the

13
test/reference ratio observed upon accumulation to

14
steady state will go closer and closer to unity,

15
one. So, that is why I think we
potentially need

16 to
consider these two cases differently in the case

17 of
a drug that accumulates.

18
[Slide]

19
The other possibility with drugs subject

20 to
accumulation is to actually conduct the steady

21
state study but this has all sorts of practical

22 limitations
for some drugs, including toxicity.

23
[Slide]

24
What I have tried to do in this graph is

25 to
get some sense of the magnitude of day-to-day

20

1
fluctuations in a pharmacokinetic parameter--I have

2
plotted this as if it were Cmax but it could

3
equally apply to AUC--in the case of a drug that

4
does not undergo accumulation.

5
What is plotted here, in orange, is

6
simulated data representing the sequential

7
day-to-day Cmax's that might be seen in a given

8
patient taking a single drug over the course of 30

9
days where the drug has a true mean of 100 percent.

10 In
fact, the sample mean here for this set of 30

11
data points is 100 and is the geometric mean, and

12 the
CV of this data set is 10 percent. So,
you can

13 see
that the drug is fairly well controlled within

14 a
fairly narrow range. Just as a yardstick
for

15
variability, I have plotted the bioequivalence

16
limits, the 80 percent limit and the 125 percent

17
limit. I want to make it clear
these limits do not

18
apply to individual day-to-day values, but I am

19
just plotting them here to give some sense of

20
scaling.

21
What I have plotted here, in the green, is

22 a
different formulation, formulation B of the same

23
drug that has a mean here of 125.
So, it is a 25

24
percent higher mean than this. CV
is still 10

25
percent. So, this could be seen
to represent the

21

1
magnitude of change that one would expect upon

2
switching a patient from one formulation to a

3
second formulation with a higher mean.
You can see

4
that there is some degree of overlap between the

5
second formulation and the first but, just

6
eyeballing this, it is not too hard to see that

7
there is visually some discernible shift in the

8
overall levels.

9
[Slide]

10
Let's see what the case looks like for a

11
drug with 30 percent intra-subject CV.
You can see

12
here that there are many more excursions on a

13
single formulation outside the range of 80-125

14
percent. Overall, there is much
more overlap

15
between formulation B and formulation A despite the

16
fact that these two formulations differ by 25

17
percent.

18
[Slide]

19
Let's increase the variability one notch

20
further to 50 percent CV, and we can see even more

21
day-to-day excursions in Cmax for a patient on a

22
given formulation, many of them outside 80-125.

23 You
can see now that the overlap between

24
formulation B and formulation A, again a 25 percent

25
difference here, is almost not discernible at all

22

1 to
the eye.

2
[Slide]

3
Finally, let's turn it up one notch

4
further to 70 percent intra-subject CV.
With a

5
drug that is this variable you end up, while on a

6
single formulation with no switch involved, with a

7
range of Cmax values that could be as far as a

8
5-10-fold range day-to-day. So,
there are wide

9
swings in the Cmax's achieved for a given subject.

10
In light of this, suppose that this is a

11
reference drug that is already approved by the

12
agency and known to be safe and effective, that

13
safety and efficacy is true in spite of the wide

14
variability from day-to-day so, therefore, the drug

15
cannot have a narrow therapeutic index and must

16
necessarily have a relatively wide therapeutic

17
index if it is safe and effective despite such wide

18
variation.

19
Also, you can see that the switch-over

20
product, formulation B, again a 25 percent higher

21
mean, is virtually indistinguishable now from the

22
range of blood levels that you see with formulation

23 A.

24
I think that the criteria, which are still

25
plotted here, 80-125 percent, need to be

23

1
commensurate with the degree of overlap that we are

2
trying to achieve between formulations.
Even

3
though these are the criteria, I would like to

4
point out that in order to pass the criteria the

5
actual observed mean in a bioequivalence study

6
generally has to be in a very narrow range, maybe 5

7 or
10 percent deviant from 100. Outside of
that,

8
your chances of passing a bioequivalence study on a

9
very variable drug are very, very poor.

10
[Slide]

11
There are certain special considerations

12
that we need to take into account in the discussion

13 of
highly variable drugs, one of which is where

14
parallel studies are conducted for long half-life

15
drugs.

16
Oftentimes you can't do a crossover study

17
because the wash-out period would be too long.

18
Powering parallel studies depends on between

19
subject variability rather than within subject

20
variability. Between subject
variability is often

21
large, necessitating large bioequivalence studies

22
just as with highly variable drugs.
However, the

23
high between subject variability does not

24
necessarily imply high within subject variability.

25
Instead, it could be due to inter-individual

24

1
differences in absorption, metabolism, etc. So,

2
these drugs, from a clinical perspective, may not

3
really be highly variable but we are still faced

4
with the powering problems in terms of conducting

5 bio
studies. In these cases, generally
speaking,

6
multiple dose studies are not feasible, and we

7
might consider some sort of alternative criteria

8 for
such studies.

9
[Slide]

10 A second issue that arises and is
directly

11
related to the issue of highly variable drugs is

12 the
issue of pooling data from multiple dosing

13
groups. Because of the large
number of subjects

14
often required for highly variable drugs,

15
oftentimes you have to split up dosing into

16
multiple dosing groups.

17
Currently, the FDA requires a statistical

18
test for the poolability of the data from these

19
multiple dosing groups and the test is a measure of

20 the
significance of the group by treatment

21
interaction terms in the analysis of variance. If

22
this interaction term is statistically significant,

23
then you are not permitted to pool the data from

24 the
multiple dosing groups. The consequence
of

25
this is that each group is then evaluated on its

25

1 own
merit and, because each group is generally

2
considerably smaller than the total pool of

3
subjects, each group will be grossly under-powered

4 to
achieve bioequivalence and, therefore, if you do

5
have a statistically significant interaction term,

6
overall you are likely to have failed the criteria.

7
This procedure results in
discarding and

8
having to repeat about 5 percent of studies based

9 on
random chance alone, even if there is no genuine

10
underlying effect. The concern
here I think is

11
that even if there were some sort of underlying

12
explanation for the statistical significance of the

13
interaction term, for example differences in

14
demographics among the dosing groups, I believe

15
that there is no reason not to use the data from

16 all
the dosing groups because had they been dosed

17
together in a single group it would be perfectly

18
usable and we wouldn't be having this discussion.

19
[Slide]

20
Conclusions--while the current

21
bioequivalence acceptance criteria I believe are

22
appropriate for drugs with ordinary variability, I

23
believe that they may not be appropriate for some

24
highly variable drugs.

25
Current bioequivalence acceptance criteria

26

1
make it difficult or impossible to develop generics

2 in
some cases, which has the public health issue of

3
effectively denying treatment to many patients

4
because of affordability issues.

5
I believe that practical, scientifically

6
sound alternative bioequivalence acceptance

7
criteria could be implemented for highly variable

8
drugs to reduce the bioequivalence study size while

9
still maintaining assurance of safety and efficacy.

10
Different approaches may be needed for

11
different types of drugs depending on accumulation

12
following multiple dosing, and also depending on

13 the
variability of the drug. And, other
related

14
situations, i.e., the issue of parallel studies and

15
multiple dosing groups should also be considered in

16
conjunction with any changes to acceptance criteria

17 for
highly variable drugs. Thank you.

18
DR. KIBBE: Does anybody on the
panel have

19
questions for our presenter to clarify information?

20
Nozer?

21
DR. SINGPURWALLA: Certainly, I
do. I

22
have four questions and five comments.
Do I have

23
time?

24
DR. KIBBE: You have until
everybody

25
leaves to go to the airport!

27

1
DR. SINGPURWALLA: The first
question is a

2
question of clarification. What
is Cmax? when

3
somebody puts C and a max I think of the maximum.

4
MR. DILIBERTI: That represents
the

5
maximum because concentration achieved within a

6
given patient or subject over the course--

7
DR. SINGPURWALLA: So, it is
maximum blood

8
concentration?

9
MR. DILIBERTI: Yes, it is maximum
blood

10
concentration.

11
DR. SINGPURWALLA: Thank you. What is

12
AUC?

13
MR. DILIBERTI: Area under the
curve,

14
which is generally taken to be a measure of the

15
extent of absorption.

16
DR. SINGPURWALLA: The third
question is

17 why
did you take natural logs?

18
MR. DILIBERTI: It is conventional
in the

19
analysis of bioequivalence data to do a log

20
transformation. This is already
established as

21 standard--

22 DR. SINGPURWALLA: Log transformation of

23 the whole data or just the maximum?

24
MR. DILIBERTI: You would log
transform

25
each of the individual Cmax's and then follow that

28

1 by
appropriate analysis of variance. The
same log

2
transformation also applies to the individual AUCs

3
prior to analysis of variance.

4
DR. SINGPURWALLA: Well, I can see
doing a

5 log
transformation of all the data to get

6
approximate normality if the distribution is log

7
normal.

8
MR. DILIBERTI: Yes, that is true.

9
DR. SINGPURWALLA: Just taking the log of

10 the
maximum--I don't know. By geometric
mean, you

11
mean product divided by--what do you exactly mean?

12
MR. DILIBERTI: The geometric mean
is what

13
results from the log transformation.
You do the

14 log
transformation and conduct analysis of

15
variance. From the analysis of
variance you get a

16
least-squares mean on a log transformed variable.

17
When you back-transform that by exponentiating it

18 you
end up with, in essence, a geometric mean.

19
DR. SINGPURWALLA: Okay. Now we will go

20 to
comments. As somebody who is new to all
this

21 and
doesn't know, the thought that first comes to

22 my
mind is that this HVD, highly variable drug,

23
should really be looked at as a bivariate problem.

24 You
have two variables. One variable is the
extent

25 of
absorption and the other variable is the rate of

29

1
absorption. So, I would look at
it as a surface

2
because the following is possible, suppose you have

3 a
drug which has a low variability with respect to

4
absorption but high variability with respect to

5
extent of absorption, how do you classify it? So,

6
what we need is a better measure of classifying a

7
highly variable drug which is a bivariate measure.

8
That is the first comment.

9
You proposed, I think, abolishing the

10
confidence limit notion.

11
MR. DILIBERTI: No, I didn't. I am not

12
here to propose solutions to the problem; I am just

13
here to really identify what the concerns and

14
problems are.

15
DR. SINGPURWALLA: Okay, but do
you have

16 any
sense of what is an alternative?

17
MR. DILIBERTI: Various
alternatives have

18
been proposed, including reference scaling or some

19
fixed point scaling that is different from 80-125--

20
DR. SINGPURWALLA: But you are not
putting

21
those forward?

22
MR. DILIBERTI: I am not really
here to

23
discuss that.

24
DR. SINGPURWALLA: So, your basic
focus is

25
criticizing what is there but without an

30

1
alternative in mind?

2
MR. DILIBERTI: Right, I think
many of the

3
later speakers will address the issue of potential

4
solutions.

5
DR. SINGPURWALLA: Now, in these
charts

6
that you showed, how did you choose the particular

7
patient whose charts you were showing?

8
MR. DILIBERTI: It is simulated
data. It

9 is
log normally distributed random independent

10
variables. It is not patient
data. I am sorry, I

11
thought that that was clear. It
is entirely a

12
computer simulation just to give some sense of the

13
relative magnitude of the variability.

14
DR. SINGPURWALLA: Well, I didn't
get that

15 message.
I thought that was a real patient--

16
MR. DILIBERTI: No, no, no.

17
DR. SINGPURWALLA: --those data
you were

18
showing.

19
MR. DILIBERTI: No.

20
DR. SINGPURWALLA: But you don't
need to

21
show it because if it is simulated we can

22
appreciate it. The last point is
when you talked

23
about pooling the data between two groups, how is a

24
group defined? What constitutes a
group?

25
MR. DILIBERTI: By the day on
which dosing

31

1
occurs. For example, it may be
impractical to dose

2 100
patients or subjects in a clinic all on the

3
same day. So, you may have to
dose half of them

4
today and maybe the other half several weeks from

5
today.

6
DR. SINGPURWALLA: So the groups
are

7
random depending on who shows up.

8
MR. DILIBERTI: Essentially, yes.

9
DR. SINGPURWALLA: Suppose one
were to

10
think about forming these groups based on some

11
other, you know biological or--defining a group in

12 a
certain way, conceivably you could justify

13
pooling. This is completely
random.

14
MR. DILIBERTI: Right, and I
believe that

15 the
way that the groups are conventionally arranged

16 in
a typical bioequivalence study pooling may be

17
justified even if you do have a statistically

18
significant interaction term.

19
DR. SINGPURWALLA: See, what I am afraid

20 of
is that if you did this on some other day and

21 you
had the same policy of pooling at random you

22 may
see a completely different result in the sense

23
that the point you are making may not be made.

24
Well, thank you.

25
MR. DILIBERTI: Thank you.

32

1
DR. KIBBE: Anybody else? Go ahead.

2
DR. SELASSIE: You mentioned that

3 potential savings to patients are in the
billions

4 of
dollars if generics are approved. Can
you tell

5 me
or do you have an idea of what percentage would

6
actually be the lack of savings due to the fact

7
that there are no generics for each of these as

8
opposed to other patent issues?

9
MR. DILIBERTI: That is very
difficult to

10
assess because, for example, in looking at patents

11 you
need to look even beyond the "Orange Book."

12
Some of these formulations have patents that are

13 not
listed in the "Orange Book."
So, to compile

14
data like that would be a Herculean task. However,

15 I
do know from personal experience that the

16
difficulties in meeting bioequivalence criteria do,

17 in
fact, pose a very real barrier to the

18
development of some generics.

19
DR. MEYER: If I could give an
example, if

20
your wife is on premarine you know you insurance

21
co-pays $20.00, because there is no generic

22
currently available because of bioequivalence

23
issues, instead of $5.00.

24
MR. DILIBERTI: Right.

25
DR. MEYER: Since my light is on I
will

33

1 just add that I do agree with you about
pooling

2
data together. A clinical trial,
after all, has a

3
patient come in to a doctor's office; they take a

4
measurement. A week later another
patient comes in

5 and
now you have two groups, and you don't analyze

6
those separately. So, unless
there is really some

7
reason to think that two groups of 50 can't be put

8
together to make one group of 100, I think it is

9
silly not to put them together.

10
DR. KIBBE: Paul?

11
DR. FACKLER: If I could just make
a

12
couple of comments, one addressing the issue of AUC

13 and
Cmax, there are very few drugs where I think

14
Cmax is not highly variable but AUC is.
I would

15 say
that from our experience it is the other way

16
around.

17
DR. SINGPURWALLA: I am sorry, I
missed

18
that. You are saying that the two
are correlated.

19
DR. FACKLER: I am saying that
there are

20
very few examples of drugs that are highly variable

21 on
AUC but not highly variable at Cmax.
Generally

22 it
is the other way around, AUC is not as variable

23 as
Cmax.

24
DR. SINGPURWALLA: So, it makes my
point

25
that you may have a bivariate situation.

34

1
DR. FACKLER: Yes, absolutely.

2
DR. SINGPURWALLA: Thanks.

3
DR. FACKLER: One of the things I
wanted

4 to
ask Charlie was on the simulated data you

5
represented 80 percent and 125 percent.
I am

6
wondering did you happen to calculate the

7
confidence intervals for the simulated data sets to

8
show where the 90 percent confidence intervals

9 would
have resulted? Because I am certain they
are

10 far
beyond 80-125.

11
MR. DILIBERTI: That is
right. No, I did

12 not
go through that calculation.

13
DR. FACKLER: The last point I
wanted to

14
make was that on the graph of the number of

15
subjects needed to get to 80 percent power versus

16 the
variability, it is important to recognize that

17 80
percent power means that one out of five studies

18
under those conditions will fail to show

19 bioequivalence, or only four out of five
will. So,

20
even if a product is tested against itself with,

21 for
instance, 30 percent variability, using the

22
number of subjects in that particular graph one out

23 of
five studies will fail to show that the product

24
against itself is bioequivalent.

25
DR. KIBBE: Shall we move
along? I think,

35

1
Gordon, you are up.

2
Highly Variable Drugs: Sources of Variability

3
DR. AMIDON: I am going to talk
about

4
sources of variability and emphasize mechanisms of

5
absorption and focus on bioequivalence from an

6
absorption point of view. It is
the approach I

7 have
been taking for the past 10 to 15 years.

8
[Slide]

9
If you think about bioequivalence where we

10 are
comparing drug products, then the question of

11
bioequivalence is really a dissolution question.

12
Right, the same drug? So, we
should be looking at

13
mechanism and dissolution and processes that are

14
controlling absorption and develop our tests around

15
that mechanism, what is controlling the process.

16
Of course, plasma levels are the gold

17
standard. Our business is to
ensure that plasma

18
levels match the innovator product used in the

19
clinical testing. That is the
criterion, no

20
question about that; no argument about that. The

21
question is what test.

22
[Slide]

23
So, I want to show some of the factors.

24 We
tend to focus on bioequivalence from a plasma

25
level point of view over here. We
focus on the

36

1
plasma which is the gold standard.
But if

2
absorption is controlled by the dissolution

3
process, dissolution controls the presentation of

4
drug along the gastrointestinal tract and,

5
therefore, controls the rate and extent of

6
absorption. If the rate and
extent of absorption

7 is
the same, then the plasma levels will be the

8
same. So, in the question of
bioequivalence then

9 the
real scientific issue is how do we set a

10
dissolution standard? My position
may be a little

11
extreme because no one seems to want to think about

12
that very much but that is the reality of the

13
science.

14
[Slide]

15
So, I think if you have two drug products

16
that present the same concentration profile along

17 the
gastrointestinal tract, they will have the same

18
rate and extent of absorption and systemic

19
availability. You may want to
think about that,

20 the
same rate and extent of absorption implies the

21
same systemic availability. So,
we need to focus

22 on
product.

23
[Slide]

24
Some of the processes in the

25
gastrointestinal tract that can lead to the

37

1
variability--and I will just illustrate some of the

2
processes here--would be the gastric emptying,

3
intestinal transit, luminal concentration both of

4 pH
and surfactants, phospholipids, presence or

5
absence of food. When you think
about it, there

6 are
a lot of sources of variability just in the

7
gastrointestinal tract.

8
[Slide]

9
Systemic availability--what should our

10
testing ensure? It is the gold
standard, no

11
question about it. But the
question then is what

12 is
the best test? What is the best test to
ensure

13
plasma levels? And, when plasma
levels are

14
difficult to measure or, in the case of highly

15
variable drugs where it requires a lot of subjects,

16
then I think it really requires us to think what is

17 the
source of that variability and then what type

18 of
test might we set.

19
I would argue that if two highly variable

20 drug
products dissolve the same way in the

21
gastrointestinal tract they will be bioequivalent.

22 It
might require 100 subjects to show that.
I

23
think that is unnecessary. I
think you just do it

24
with a dissolution test and the answer will be far

25
simpler.

38

1
[Slide]

2
So, what are some of the physicochemical

3
factors? Clearly, particle size
and distribution;

4
wetting and solid-liquid contact; and, of course,

5 in
some cases chemical instability such as prodrugs

6 and
esterases and peptidases in the

7
gastrointestinal tract can lead to highly variable

8
absorption and, hence, systemic availability.

9 [Slide]

10
I just put one graph in here showing the

11
dependence here of dissolution time, ranging up to

12 30
hours, and gastrointestinal transit time as a

13
function of particle size. I
can't manipulate this

14 in
this presentation but the dissolution time

15
increases dramatically as the drug solubility

16
decreases. Particle size becomes
a critical factor

17 for
low solubility drugs. Of course,
everyone

18
realizes that but it is not particle size that we

19 put
into the formulation, it is the particle size

20
that comes out of the formulation in the

21
gastrointestinal tract. So, those
process

22
variables are important.

23
[Slide]

24
Some of the factors in the

25
gastrointestinal tract then are gastric emptying,

39

1
intestinal transit, position dependent permeability

2
along the gastrointestinal tract--duodenum,

3
jejunum, ileum and colon and, of course, intestinal

4
mucosal cell metabolism, and in particular CYP3A4

5
which is highly expressed and differentially

6
expressed along the gastrointestinal tract, and

7
potentially PGP expression along the

8
gastrointestinal tract.

9
[Slide]

10
To give you an example of variability in

11
gastric emptying rates, we can just look at the

12
light blue because that is administered with 200

13 ml,
the approximate glass of water that we use.
We

14
used 200 ml here because we did this before we got

15
involved in drug regulatory standards and realized

16
that a glass of water was the

17 the
standard in

18
that out, what is a glass of water in

19
with 200 ml you can see that the variability in

20
gastric emptying. Depending on
when you dose in

21 the
fasting state, it ranges from 5 minutes to

22
about 22 minutes. There is about
a 4-fold

23
variation in gastric emptying rate depending on

24
when you administer to a particular subject. This

25 is
because of the different contractual activities

40

1 in
the fasted state, shown here as phase 1, 2, 3

2 and
4.

3
[Slide]

4
Clearly, intestinal transit--again, this

5 is
a movie but I can't show it with this

6
presentation--transit through the gastrointestinal

7
tract where the drug is released in the duodenum.

8 It
has a very short transit time, maybe 10, 15

9
minutes through the duodenum, jejunum, ileum and

10
colon. The dissolution rate,
particularly of a low

11
permeability drug where the permeability appears to

12 be
the rate-determining step to absorption, the

13
permeability profile along the gastrointestinal

14
tract is very important.

15
[Slide]

16
There are about 10 L of fluid processed in

17 the
gastrointestinal tract per day, actually

18
depending on which book you read, 8 to 10. Of the

19 10
L that are processed, only about 2 L are

20
actually ingested as external.
The other 8 L are

21
ourselves. We are continually
secreting and

22
reabsorbing not only fluids but cells and proteins

23 and
other ions that are secreted into the intestine

24 so
there is a tremendous amount of variability and,

25 of
course, food has a large impact on that as well.

41

1 So,
that is a major factor that can be involved in

2 the
variability and dissolution and absorption in

3 the
gastrointestinal tract.

4
[Slide]

5 I show here just ranitidine, a low

6
permeability drug. This is animal
data. I don't

7
have human data and, in fact, it is very hard to

8 get
human data although there is some data

9
available. The duodenum, jejunum,
ileum--there is

10 a
significant difference in permeability.
So, you

11 can
envision a slowly dissolving ranitidine

12
product--I don't know if there are any, but

13
releasing in the ileum would have very poor

14
absorption. So, dissolution for a
low permeability

15
drug is probably more important because, in

16
general, the permeability in the upper part of the

17
gastrointestinal tract is more important or higher,

18 I
should say.

19
You know, we used to use language like

20
"rapidly but incompletely absorbed." You would see

21
that in the literature after analysis of

22
pharmacokinetic data and I would say how can that

23
be? It doesn't make sense to
me. If it is rapid

24 it
should be well absorbed. Right? Clearly, there

25 has
to be position-dependent permeability and the

42

1
absorption rate must decrease dramatically at some

2
point very quickly after the drug is administered.

3
Presumably, that is the result of drug getting into

4 the
ileum or distal in the small intestine where

5
there is lower absorption.

6
[Slide]

7
PGP--this is some immunoquantitation

8 results on CYP3A4 showing the variation in
the

9
duodenum, ileum and colon, much less in the colon

10 so
that there is less metabolism, particularly if

11
there is a controlled release formulation releasing

12
drug in the colon and, of course, much more in the

13
liver. I don't know, maybe Leslie
is going to say

14
more about the metabolism source of variability,

15
maybe not. You are shaking your
head, no.

16
[Slide]

17
I am going to propose that we classify the

18
drugs, highly variable drugs using BCS.
Here is

19
what I think we would see. We
need to actually

20
look at particular drugs. In
fact, I would like to

21 see
a list of drugs perhaps based on the

22
variability of reference products, whatever we

23
could find today, develop a list of highly variable

24
drugs or that we think might be highly variable,

25 and
then look at their properties and decide what

43

1 are
the likely sources of variability.

2
Anyway, I know there are certain so-called

3
highly variable drugs that are Class I drugs. They

4
have to be low dose, low solubility drugs but they

5 are
soluble enough to dissolve in a glass of water.

6
That is our criteria at the present time. So, if

7
those drug products dissolve rapidly--if they do; I

8
don't know if they do, we should look at that and

9 it
is over; there is no issue. It is all
biologic

10
variability; nothing to do with the product

11
variability. Again, that is a
hypothesis.

12
Probably the majority of the drugs that

13 are
highly variable are in Class II where there is

14 low
solubility, potentially Class IV for some

15
higher molecular weight compounds.
There, the

16
solubility-dissolution metabolism interaction can

17 be
difficult to separate and that is where we would

18
need to look more carefully at the drug products to

19
determine whether it is the solubility and

20
dissolution variability or whether it is a

21
metabolism variability that is leading to the high

22
variability in plasma levels.

23
[Slide]

24
So, I think that the BCS classification

25 can
help focus on the source of the high

44

1
variability. Then, in the case of
rapid

2
dissolution of Class I and Class III drugs a

3 dissolution
standard may be enough. There may not

4 be
too many highly variable drugs because I think

5 the
majority would be the low solubility Class II

6 or
Class IV drugs and there I think metabolism

7
and/or dissolution can be the source of

8
variability. In the case of
metabolism, the

9
metabolism variation would be due to the

10
variability and dissolution and presentation along

11 the
gastrointestinal tract. So, again, it
comes

12
back to a dissolution issues.

13
In fact, I would propose that we look more

14
carefully at the highly variable drugs, the sources

15 of
variability, again asking the critical question

16
what is the best test? What is
the best test? I

17
will go back to the original implementation of BCS

18 in
the case of high solubility, high permeability,

19
rapidly dissolving drugs. Plasma
levels are

20
telling us nothing about the product differences.

21 It
is only telling us about gastric emptying

22
differences at the time of administration of

23
patients or subjects. So, again,
focusing on

24
dissolution and classification I think can help us

25
unravel and simplify some--maybe not all. Maybe

45

1 not
all of the highly variable drugs can be

2
simplified this way but I think some of them can be

3
simplified this way. For those
drugs that are

4
complicated, we just say they are complicated.

5 Take a drug like premarine. You have already

6
mentioned that, Marvin. I think
that premarine is

7 a
complicated drug. That is life; that is
the way

8 it
is. It is too complicated for us to
unravel

9
today because of the way we regulate drugs and

10
approve drugs. So. I am happy to answer any

11
questions by the committee.

12
DR. KIBBE:
Questions, folks? Jurgen?

13 DR. VENITZ:
I agree with you, I am very

14
much in favor of identifying sources of variability

15 and
what you are presenting are obvious sources of

16
variability, and it always bothers me when we talk

17
about highly variable drugs and they are defined

18
phenologically. All we are doing
is a clinical

19
study. We are measuring Cmax and
AUC and we find

20
that they vary a lot, and that is the end of it,

21 and
now let's change criteria to see whether they

22 can
fit bioequivalence. So, I agree with you
on

23
that.

24
What I won't agree with you, at least not

25
fully, is that it is all a dissolution issue. I

46

1
think you are ignoring, in my mind at least, the

2
effects that excipients may have that could be very

3
different between formulations so that may not have

4 an
impact on dissolution but may have an impact on

5 pH,
may have an impact on permeability and may have

6 an
impact on GI metabolism. Now, I don't
know

7
whether that is a significant problem or not but I

8
think it is more than dissolution that you are

9
looking at. It doesn't preclude
what you are

10
recommending, which is basically do dissolution

11
tests and find out if that is an issue and then see

12 how
that matches your in vivo data. That is
just a

13
comment.

14
DR. AMIDON: If we extend the
dissolution

15 to
dissolution of the excipient, that is, the

16
dissolution of the excipient and the drug, then I

17
think we would be okay; I think my statement would

18 be
okay.

19
DR. VENITZ: But if you have
products that

20
have different excipients, that is my point.

21
DR. AMIDON: Yes, okay.

22
DR. VENITZ: As you said, life is

23
complicated. Sometimes it works;
sometimes it

24
doesn't.

25
DR. AMIDON: Right. So, that is the

47

1
function of what is the source of the variability.

2
DR. VENITZ: Yes.

3
DR. KIBBE: Ajaz?

4
DR. HUSSAIN: I worked with Gordon
for

5
many years on developing the BCS guideline, and so

6
forth, and we actually did examine that very

7
question of excipients and their impact not only on

8 the
dissolution process but on permeability and

9
metabolism and it is a serious issue and I think we

10
learn more about transport every day.
Therefore,

11
clearly, I think when Gordon mentioned dissolution,

12 we
have discussed that so many times and we always

13
include that as a source of variability and that

14 has
to be considered.

15
But, Gordon, I wanted to push you in a

16
different direction. One of the
hesitations as we

17
developed the BCS guidance was the reliability of

18 the
in vitro dissolution test. We were not

19
confident that the current test really was good

20
enough to extend it to the slower releasing

21
products. So, that was the reason
we crafted

22
rapidly dissolving and said dissolution is not rate

23
limiting and, therefore, we can rely on the current

24
dissolution test to do that.

25
I think as we move forward here, I think

48

1
what we have done with the PAT initiative is to

2
sort of say, all right, let's really ask the

3
question what are the criteria variables, what are

4 the
root causes of this. So, go back to the
basics

5 as
to particle size, and so forth, and if you

6
really understand those relationships then you have

7 a
better link between your formulation and your

8
excipients; you have your process directed to the

9
clinical relevance. So, that is
the opportunity

10
that technology is offering us to do that without

11
having to do an artificial in vitro test where

12
questions keep continuing and increasing with

13
respect to the relevance of that in vitro test.

14
DR. AMIDON: I certainly obviously
agree,

15
Ajaz. We have talked about these
issues for many

16
years. I did use the word in vivo
dissolution.

17
There is a big step from in vitro to in vivo. I

18
don't think it is magic; it is just complicated and

19 I
think we can figure that out. I think we
can

20
determine for any particular drug what might be a

21
good representative dissolution test, and I might

22
call that a bioequivalence dissolution test rather

23
than a QC, quality control, dissolution test. But

24 you
are absolutely right. The issue is
really in

25
vivo dissolution and how do we capture that in some

49

1 in
vitro methodology. I don't think we have

2
thought about that very hard at all.
I am not sure

3
why. We use the term dissolution
very generically

4
when it should be much more specific.

5
DR. KIBBE: Les wants to comment
and then

6
Nozer. Can you make a comment,
Les, because you

7 are
not part of the committee?

8
DR. BENET: They said as a visitor
I can.

9 I
wanted to comment on BCS and what Jurgen brought

10 up
in terms of the excipients. When we
initiated

11 BCS
I was very strong concerning the potential for

12
excipients on Class I drugs and we have written the

13
rules to make sure that these excipients don't have

14 an
effect. In fact, I now recognize that
with

15
Class I drugs that is not a problem, that the

16
excipients won't be a problem in terms of affecting

17 at
least the transporters. But they will be
a

18
problem with Class III drugs.

19
So, so far I have been very opposed to

20
moving the Class III drugs because I can make a

21
Class III formulation that will pass dissolution,

22 any
dissolution, and fail. The reason is
that

23
Class III drugs need uptake transporters to get

24 absorbed
and, therefore, I can block an uptake

25
transporter in the gut with a substance that has no

50

1
dissolution criteria. So, I still
think we are a

2
little early in translating this dissolution

3
criteria beyond Class I, but I think we were

4
correct in Class I and the extra safeguards we put

5 in
actually really turn out not to be necessary.

6
DR. SINGPURWALLA: I like this
concept of

7
looking at the causes of variability.
I see this

8 as
a first step towards going to a Bayesian

9
alternative for the existing methodology that was

10
criticized by the first speaker.
But I do have a

11
question perhaps both for you and also for the

12
first speaker. Has anybody looked
at the

13
reliability of the testing instrument itself?

14
Because if the testing instrument itself shows a

15
large variability--if the instrument itself shows a

16
large variability then you don't know whether the

17
variability is coming from the instrument or from

18 the
particular drug or the combination of the

19
instrument, the drug and the patient.

20
DR. KIBBE: Anybody? Who wants to handle

21 that?

22
DR. VENITZ: I think by instrument
what

23 you
mean is the human being used in those studies.

24 Are
you talking about dissolution or are you

25
talking about in vivo?

51

1
DR. SINGPURWALLA: Both.

2
DR. VENITZ: Well, then let's talk
about

3 in
vivo and I will leave it up to you to talk about

4
dissolution. What you are looking
at is the Cmax's

5 and
the areas under the curves. They do not
only

6
depend upon absorption and dissolution; they depend

7 on
everything that happens after the drug gets in

8 the
body, which is something we are not interested

9
in. If that contributes
significantly to the

10
variability, then you are looking at primarily

11
variability and disposition which determines why we

12
have a highly variable drug, not because there is

13
variability in absorption. So,
your instrument

14
would be a very noisy instrument I think, to use

15
your lingo.

16
DR. SINGPURWALLA: Right. You have an

17
instrument by which you measure these things, like

18 a
thermometer. If your thermometer is
bad--

19
DR. VENITZ: I am saying that for
some

20
drugs it could well be that you have a very noisy

21
instrument and the noise is not related to what you

22 are
trying to measure.

23
DR. SINGPURWALLA: Exactly.

24
DR. KIBBE: Let me just take the

25
prerogative of the chair for half a second and then

52

1 I
will let you speak. It is very difficult
for us

2 to
understand the real noise level of the

3
instrument. The instrument is the
bioequivalency

4
test itself and the agency gets submissions with

5
bioequivalency tests that are passed.
The question

6 is
how many were done that failed before the one

7
that passed, and what was done to make that work?

8
I think if you go back and we got a bunch

9 of
data together, which we can't but it would be

10
interesting to look at, we would find that the

11
instrument is very crude and the reason we live

12
with it is that it is close to the clinical

13
therapeutic outcomes that we really want to measure

14 in
terms of steps away from that outcome.
What

15
Gordon is recommending is that we even eliminate

16 the
human from our decision-making process, which

17
brings us further away from the ultimate goal which

18 is
to know that it therapeutically equivalent, and

19 we
have to be sure that our predictor is going to

20
hold true. Those are the problems
I think that we

21 all
have been struggling with for 25 years.

22
DR. HUSSAIN: Now I have three
comments.

23
With respect to the instrument variability, I think

24 it
is a very important question. In the
case of

25
bioequivalence testing we try to minimize that and

53

1 try
to make it more precise and more accurate by

2
doing a crossover study. We test
the two products

3 in
the same patient in a crossover fashion.
So,

4
that is our attempt to minimize that.
The other

5
attempt that we had to minimize is to get a group

6 of
more similar individuals but we wanted to move

7
away from that in the general population because

8 the
crossover is a way to minimize that. I
also

9
pointed out with respect to variability the

10
dissolution test. I think as we
think about that,

11 we
need to address that.

12
But the point I think, going back to the

13 key
question, is what are the important questions

14
here? Dr. Kibbe's comment was, in
a sense,

15
bioequivalence. For therapeutic
equivalence our

16
approach is very simple. First
you need to be

17
pharmaceutically equivalent and then, if there is a

18
need, you do a bio study. For
example, for

19
pharmaceutical equivalence for solutions you don't

20
need a bio study. So
pharmaceutical equivalence,

21
bioequivalence and then therapeutic

22
equivalence--those come together to define that. I

23
could sort of generalize what Gordon has said, in a

24
sense if we understand our formulations, if we

25
understand our processes, if we understand the

54

1
mechanisms, pharmaceutical equivalence essentially

2 is
defining therapeutic equivalence.

3
DR. AMIDON: To come back to your
question

4
about the dissolution apparatus, there is a range

5 of
dissolution apparatus in the USP that are used

6 internationally,
and you can study many of the

7
variables that change in vivo by pH and surfactants

8 in
those apparatus. The apparatus
themselves have

9
been proven perhaps historically to be very

10
reliable, although you could argue maybe today that

11 we
could design a better apparatus but that is very

12
complicated because these things are used in many

13
companies internationally with defined procedures

14
that are approved by the regulatory agencies and

15
making change in an apparatus is a very complex

16
process.

17
But, yes, we can study the various

18
variables in vivo and I think that a dissolution

19
test that included changes in pH and surfactant to

20
reflect what is happening in vivo is something we

21
should do. We don't do that; we
just do fixed pH

22 and
follow the dissolution as a function of time.

23 So,
I don't think we use our apparatus very

24
insightfully actually.

25
DR. KIBBE: I would argue that the
way we

55

1 use
dissolution is reliable but insensitive, and we

2
need to do a lot more to be able to make that

3
conversion. Anybody else?

4
DR. MEYER: Gordon, I listened to the PAT

5
stuff all day yesterday and what I got out of it is

6
that it is applicable to this so the idea of why do

7 we
have variability--right now we are proposing to

8
potentially change our release specifications

9
because our product is too variable and that is not

10
acceptable in the manufacturing arena.
You go back

11 and
figure out why it is too variable. I
wonder

12 how
much data is really available on if I gave

13
myself a rapidly absorbed drug once for the next

14
three weeks, what would my profiles look like? I

15
don't know that there is a lot of data that shows

16
reproducibility in a subject, unless it was the old

17
multiple dose studies where the drug was

18
essentially eliminated in 24 hours.

19
So, I think we need some more information.

20 I
don't know, maybe the agency does this, but when

21 the
innovator firms do special populations and they

22
find the elderly are different than the young, do

23
they have to then go further and explain is that

24
gastrointestinal pH, is it transit, is it

25
metabolism, what is the reason for it.
Because I

56

1
think then we can get some background information

2 on
source of variability.

3
Just to bounce off an idea which is

4
undoubtedly ludicrous, do we need in a sense to

5
prescreen some subjects so we have a calibrated man

6 or,
if you will, a USP man or woman that is then

7
allowed into the study so if they have less

8
variability they get into our study?
Could we do

9
that? One thing that really
troubles me is the

10
current policy, and I understand why it is and I

11
think I support it, of having different mechanisms

12 of
release tested against each other in a

13
bioequivalence study, an oral study versus a

14
particular dosage form.
Intestinal transit can

15
have a profound difference on those two so if you

16
have a uniform man, that uniform man may show them

17 to
be equal but if you throw in a vegetarian, that

18
vegetarian might show the oral tablet is excreted

19 in
four hours and the other person may take much

20
longer. So, just some support
really for the idea

21 of
knowing where the problems are; can we reduce

22
variability somehow; are subjects legitimately--is

23
that a viable approach?

24
DR. AMIDON: I don't know, I am
not sure I

25
would want to take on preselecting subjects because

57

1
what criteria are you going to use?
Normal in what

2
sense?

3
DR. MEYER: I am thinking more in
terms

4 of,
say, rapid metabolism or poor metabolism.
We

5 do
that now somewhat routinely.

6
DR. AMIDON: Right.

7
DR. MEYER: So, we might give a

8
panel--CROs now, they use the same subjects over

9 and
over again anyway. Let's characterize
them

10
first before they are allowed into subsequent

11
studies.

12
DR. KIBBE: Paul, go ahead.

13
DR. FACKLER: If I can just
comment on

14
that, we used to do bioequivalence studies in males

15
only and restricted their ages from 18 to 45, I

16
believe. The agency has recently
requested that BE

17
studies be done in a larger group of people, more

18
representative of the American population so we now

19
include females and we include the elderly, and it

20
just makes the variability problem that much worse.

21 I
mean, I agree completely that ideally if we would

22 get
15 people all exactly the same way, all exactly

23
with the same physical habits, generally with the

24
same diet, it would make BE studies easier to pass

25
because we have reduced the variability in the

58

1
subjects. But the agency has been
going, at least

2
recently, in the opposite direction, making these

3
products in particular less likely to pass against

4
themselves again.

5
DR. KIBBE: It is my impression,
and I am

6
sure the FDA people will correct me, that they are

7
trying to get two answers using one study, and that

8 is,
are the two formulations behaving the same,

9
should be their behavior independent of the

10
subjects studied, and are there variabilities

11 between
product-subject interactions that might be

12
significant in special populations.
I think it is

13
really hard to do that in one study, and that is

14 one
of the problems you are running into.
What I

15
think Gordon is suggesting is if we understood the

16
variables we might not have to use that blunt a

17
tool to estimate what will happen in the average

18
patient.

19
I would love to see us be able to do that.

20
There was a wonderful report done--Les will

21
remember because he is almost as old as I am--by

22 the
agency that looked at dissolution and tried to

23
correlate it with bioequivalency data that they had

24
almost twenty years ago and there was absolutely no

25 way
that dissolution predicted any of the results

59

1
that they got on those studies.
So, it is more

2
complicated than it first appears.

3
DR. AMIDON: I got involved in
this

4
process about that time, and my position is you

5
just did the wrong test.
Okay? That is the

6
problem. So, it is a matter of
refining the

7
dissolution test to make it more relevant to the

8
variables that we need to control to ensure

9
bioequivalence. We haven't done
enough of that.

10
DR. KIBBE: Ajaz, you have a
comment?

11
DR. HUSSAIN: The key aspect I
think is

12
that we need to keep the focus on asking the right

13
questions and if a bioequivalence study is only

14
for, you know, males 18 to 45, is that the right

15
question from the public health aspect because the

16
product is going to be used in all populations?

17 So,
you really have to go and look at the

18
fundamentals of what is a bioequivalence study. If

19 it
is just confidence interval criteria, then that

20 is
one aspect.

21
DR. SINGPURWALLA: Why not have a
separate

22 set
of drugs for different categories of people?

23
Like, you know, you have cholesterol drugs 20 mg,

24 10
mg and you specify your milligrams based on the

25
population.

60

1
DR. HUSSAIN: That is a major
aspect of

2
dose finding and then labeling that goes into the

3 new
drug development process itself. The

4
bioequivalence essentially has been a quality

5
assurance approach to making sure that a

6
pharmaceutically equal product has an in vivo rate

7 and
extent of absorption similar to the innovator.

8
That is one of the main reasons for doing the bio

9
study, to make sure that your assumptions and your

10 in
vitro methods are more reliable or at least

11
conform from that perspective.

12
DR. KIBBE: Thank you. Unless someone

13
else has a comment we will let you off the hook for

14 a
few minutes, and go to Dr. Benet who will

15
enlighten us.

16
Clinical Implications of Highly Variable Drugs

17
DR. BENET: I am older!

18
[Laughter]

19
Thank you. It is a pleasure to be
here.

20 I
think the last two times I have appeared before

21
this committee I stayed in my office but it is nice

22 to
be here in person, and I thank you for the

23
opportunity.

24

25
We have been discussing at an

61

1
international level, I was reminded as I heard

2 this,
for 15 years--we held our first sort of

3
consensus conference in 1989 to try to develop

4
standards for bioequivalence and we are still at

5 it.

6
[Slide]

7
This was said by the first speaker but

8 this
is a slide that is now maybe 12 years old, or

9 at
least parts of it. The current U.S.
Procrustean

10
bioequivalence guidelines: the manufacturer of the

11
test product must show using two one-sided tests

12
that a 90 percent confidence interval for the ratio

13 of
the mean response--usually the area under the

14
curve and Cmax--of its product to that of the

15
reference product is within the limits of 0.8 and

16
1.25 using log transformed data.
It is

17
Procrustean, and those of you who don't remember

18
your mythology, the Procrustes himself was a robber

19
that took people when they came through his gate

20 and
put them on his bed, the Procrustean bed.
If

21
they were too long he cut off their feet. If they

22
were too short he stretched them out until they fit

23 the
bed. And, that is exactly what we have,

24
Procrustean guidelines that say all drugs must fit

25 the
same criteria no matter what the issues are.

62

1
Now, BCS, biopharmaceutical classification

2
system, is non-Procrustean. It is
an advance and

3 the
obvious answer, Arthur, to why a study failed

4 in
looking at dissolution is that we didn't

5
understand the flawed classifications.
So, the

6
only time dissolution is going to have any

7
relevance to bioequivalence or bioavailability is

8 for
Class I and Class III drugs. Since we
looked

9 at
all drugs about 20 years ago, we were obviously

10
going to fail. So, we are making
some advances.

11 But
I strongly believe and have suggested over a

12
number of years that there need to be other

13
non-Procrustean advances and that is what I will

14
talk about today.

15
[Slide]

16
What are we trying to solve? What
are the

17
bioequivalence issues and what concerns patients

18 and
clinicians so that they have confidence in the

19
generic drugs that are approved by the regulatory

20
agencies so that they feel there are no questions

21
related to their therapeutic efficacy?

22
It doesn't help to tell them--and that is

23 a
true fact, it doesn't help to tell them that

24
there has never been a drug that passed the U.S.

25 FDA
bioequivalence issues that ever caused any

63

1
therapeutic problems in a prospective study. That

2
doesn't help them because they always say, well, it

3 is
the next drug and they have a lot of emphasis

4 out
there from people who would like them to

5
question the bioequivalence criteria.
So, this is

6
always in my mind, that one of the major issues

7 that we face is not necessarily scientific but
it

8 is
creating an environment where the American

9
public has confidence in the regulations that we

10 use
and the drugs that we say can go on the market.

11
But what we have done and what our

12
concerns are now with therapeutic index drugs, NTI,

13 we
need to have practitioners have assurance that

14
transferring a patient from one drug product to

15
another yields comparable safety and efficacy, and

16 a
few years ago we termed that switchability and we

17
developed or tried to develop a number of

18
statistical criteria to approach that.
The issues

19 we
are facing today are for a wide therapeutic

20
index, highly variable drugs which do not have to

21
study an excessive number of patients to prove that

22 two
equivalent products meet the preset one size

23
fits all statistical criteria.
So, these are the

24
issues I want to address and ask the committee to

25
take cognizance of.

64

1
[Slide]

2
Now, it was not obvious a few years ago

3 but
it is very obvious today that if you have a

4
narrow therapeutic index drug it is very easy to

5
pass the bioequivalence criteria, and that is

6
because narrow therapeutic index drugs, by

7
definition, must have small intra-subject

8
variability. If this were not
true for narrow

9
therapeutic index drugs, patients would routinely

10
experience cycles of toxicity and lack of efficacy,

11 and
therapeutic monitoring would be useless.
So,

12 in
fact, it is not an issue. Narrow
therapeutic

13
drugs we take care of and we do very well from a

14 scientific issue. We might not have the

15
confidence, and I will come back and address that.

16
[Slide]

17
Let's look at some narrow therapeutic

18
index drugs. They have high
inter-subject

19
variability and they have low intra-subject

20
variability. That is why we don't
have to worry;

21
when we get the patient to the right place, they

22
stay there. The question was are
they all Class I,

23
Class II. Theophylline is a Class
I drug. So,

24
there are drugs on this list that are Class I drugs

25
although most of them are Class II drugs.

65

1
Getting back to the reliability of the

2
instrument, I would just like to make a comment.

3
Look at the warfarin sodium intra-subject

4
variability. The clinical measure
that the

5
clinician uses to judge the status of the patient

6 in
terms of his blood thinning capability, the INH

7 measurement,
is significantly more variable. So,

8 in
fact, what the clinician does in testing if the

9
drug is working is more variable than the patient

10 is
going to experience from dose to dose in terms

11 of
the criteria for this particular drug.
So,

12
these are interesting questions.

13
[Slide]

14
Now, we tried to address this

15
switchability issue over a long period of time with

16 the
concept called individual bioequivalence, and I

17 chaired
the expert panel for about three years and

18
tried to address this issue. The
ideas about

19
individual bioequivalence were that we were going

20 to
get these promises, we would address the correct

21
question, switchability in a patient.
We would

22
consider the potential for subject by formulation

23
interaction. There would be
incentive for less

24
variable test products. Scaling
would be based on

25
variability of the reference product both for

66

1
highly variable drugs and for certain

2
agency-defined narrow therapeutic range drugs.

3
And, we would encourage the use of subjects more

4
representative of the general population.

5
In fact, none of that worked and we gave

6 up
on it. So, did it address the correct
question?

7
Well, the question was, was there even a question

8 and
was there any necessity for this at all, and

9
there is no evidence that the present regulations

10 are
inadequate and that we need to be more rigorous

11 in
our definition related to switchability.

12
[Slide]

13
Consider that the subject by formulation

14
interaction turned out to be an unintelligible

15
parameter from both the agency and the exterior

16
scientific community.

17
Incentive for less variable test products,

18
yes, but that could be solved by average

19
bioequivalence scaling and that is what at least I

20 am
here to talk about today.

21
Scaling based on variability of the

22
reference product both for highly variable drugs

23 and
for certain agency-defined narrow therapeutic

24
index drugs, again average bioequivalence with

25
scaling could solve this issue.

67

1
Encourage the use of subjects more

2
representative of the general population, that was

3 a
good hope but it completely failed in terms of

4 how
people designed their study. So, it
didn't

5
work.

6
[Slide]

7
I recognized in Lawrence's introduction

8
that the FDA doesn't have a definition for highly

9
variable drugs. This is the
consensus definition

10
that came out of a number of international

11
workshops, highly variable drugs should be those

12
when the intra-subject variability is equal or

13
greater than 30 percent. The idea
is that for wide

14
therapeutic index highly variable drugs we should

15 not
have to study an excessive number of patients

16 to
prove that two equivalent products meet this

17
preset one size fits all statistical criteria.

18
This is because, by definition, again

19
highly variable approved drugs must have a wide

20
therapeutic index, otherwise there would have been

21
significant safety issues and lack of efficacy

22
during Phase III testing. In
fact, highly variable

23
drugs fall out; don't get to the market.
They fall

24 out
in Phase II because the company can't prove

25
that they work and they can't prove that they are

68

1
safe. So, we don't have highly
variable narrow

2
therapeutic index drugs. We only
have drugs that,

3
with this tremendous variability that we

4
potentially saw in the first speaker's slide, don't

5
have any problems. And, those
individual patients

6
having very high levels one time, low levels the

7
next time, high areas under the curve one time, low

8
areas under the curve the next time get through.

9 In
fact, for those highly variable drugs we don't

10
need to worry about the genetic differences in

11
their enzymes. It has already
been shown that,

12
yes, there are tremendous differences.
Somebody is

13
going to have very high levels because they lack

14 the
enzyme; somebody is going to have very low

15
levels but still they are safe and effective

16
because they are wide therapeutic index drugs.

17
[Slide]

18
But it makes it very difficult, as was

19
also pointed out by the first speaker, to get them

20 to be
bioequivalent and here is my champion or what

21 I
think is the champion from the data that I have

22
seen, and this is progesterone which I believe is

23 the
poster drug for highly variable variability.
A

24
repeat measures study of the innovator's product

25 was
carried out in 12 healthy post-menopausal

69

1
females and it yielded intra-subject variability in

2 an
AUC of 61 percent for the coefficient of

3
variation and intra-subject coefficient of

4
variation for Cmax of 98 percent.

5
If you did the calculations, it came out

6
that you needed 300 women just to meet the

7
statistical criteria and, in fact, this was not a

8
study that a generic company, or at least the

9
company interested in this, could afford to carry

10 out
because, for sure, we know that the way we

11
design the studies there is a chance, even if you

12 had
the right numbers, that one out of ten or one

13 out
of five studies would fail just on statistical

14
chance and you have carried out a study with 300

15
people in it to prove that this highly variable

16
drug is bioequivalent. This is
the issue that we

17 are
asking you to talk about today, and can we

18
solve this problem so that we don't have highly

19
variable, very safe, wide therapeutic index drugs

20 for
which we can't prove bioequivalence because of

21 the
inherent variability of the innovator product.

22
[Slide]

23
I appeared before this committee three and

24 a
half years ago to give the recommendations of the

25 FDA
expert panel on individual bioequivalence, and

70

1
these are some of the recommendations.
One that I

2
didn't put on here is that all generic drug studies

3
must be submitted to the agency, and I am very

4
pleased that that has happened and congratulations

5 to
the agency.

6
Our recommendations at that time were that

7
sponsors may see bioequivalence approval using

8
either average bioequivalence or individual

9
bioequivalence, and we recommended that the subject

10 by
formulation parameter be deleted since no one

11
knew what to do with it and we couldn't justify it

12
statistically.

13
We asked that scaling for average

14
bioequivalence be considered, that the agency and

15 the
statistical group go into this and it be

16
something to be followed up and presented to this

17
advisory committee at some time in the future.

18
We recommended at that time that if an IBE

19
study, individual bioequivalence study, was carried

20 out
and the test product fails you could not then

21
reanalyze with average bioequivalence because in

22
those days we said you had to pick one or the

23
other.

24
Here is something that we recommended that

25 I want to bring up again today because this
has to

71

1
do with confidence. We recommended the point

2
estimate criteria be added, and we added this not

3 on
any scientific basis that we are going to rule

4 out
products, we said that these criteria are

5
always met today and what we have is a conception

6 or
a view outside that it would be possible to have

7
products that differ by 25 percent, and that we

8
would be well served if we would say let's put a

9
point estimate criterion in addition to our

10
criteria--AUCs of at least plus/minus 15 for point

11
estimate criteria and Cmax plus/minus 20 percent no

12
matter what you do, and if you have narrow

13
therapeutic index drugs make it even smaller, make

14 the
point estimate plus/minus 10 percent for AUC

15 and
plus/minus 15 percent for Cmax.

16
[Slide]

17
So, what I am suggesting here today and

18
what I am recommending to the committee to do is

19 ask
the agency to develop methodology, and we are

20
going to hear some, to allow approval based on

21
weighting of average bioequivalence analytical for

22
highly variable drugs so that we can bring some

23
drugs to the market that can't be studied because

24 of
the progesterone example. Also, that the
point

25
estimate criteria be added to the criteria because,

72

1 in
fact, all products will pass these criteria at

2 the
present time and we won't be harmed, or we will

3
increase the confidence of those that say, you

4
know, you could have two products that differ by 50

5
percent because look at what the FDA criteria say.

6
Now, the FDA criteria, as they used to be

7
written two years ago, were easily misinterpreted

8 but
that also changed two years ago and now the

9
criteria are written in a way that no clinician can

10
understand them in the first place so they won't be

11
misinterpreted.

12
[Laughter]

13
They still say exactly the same thing but

14
they can't be misinterpreted to say you could have

15 two
products that differ by 50 percent. So,
these

16 are
my recommendations. Thank you for
listening to

17 me.

18
DR. KIBBE: Questions for Dr.
Benet?

19
DR. SINGPURWALLA: I have a
comment not

20
just to you but to everyone else.
This example of

21
highly variable drugs shows, to me, how the drug

22
industry is buried under the tombstone of

23
frequentist methods. Such methods
ignore clinical

24 and
biopharmaceutical knowledge, and it is bogged

25
down by its own weight.

73

1
DR. BENET: I disagree.

2
DR. SINGPURWALLA: Why?

3
DR. BENET: I think you are coming
to this

4
fresh and that is good, but what we are interested

5 in
is safety and efficacy, and in all cases

6
measures of safety and efficacy are more variable

7
than any pharmacokinetic measure.
What we are

8
really interested in, what the agency is interested

9 in
is safety and efficacy.

10
DR. SINGPURWALLA: Who said that
Bayesian

11
methods do not incorporate high variability? It is

12
these confidence intervals and these confidence

13
limits, and the comment you make is a failure to

14
understand Bayesian methods.

15
DR. BENET: I understand Bayesian
methods.

16
DR. SINGPURWALLA: No, you don't;
you

17
wouldn't say this.

18
DR. BENET: Well, I welcome the

19
committee's spending the time discussing this with

20 you
and if you adjourn I get to go home.

21
[Laughter]

22
DR. MEYER: I think I agree with

23
everything you have said and it embarrasses me no

24 end
to say that!

25 [Laughter]

74

1
Is there still going to be a perceived

2
problem when you have, let's say, a Cmax point

3
estimate of plus/minus 15 percent?
Isn't that

4 going to solicit illustrations of, well,
look, my

5
Cmax was 115 units and their Cmax was 85 and the

6
high and low can be switched in the marketplace?

7
DR. BENET: I think we are never
going to

8 get
around that. There are always going to
be

9
people who will take the present situation and use

10 it
to their marketing advantage. So, I
don't think

11 we
can get around that. You know, we have
the same

12
issues today. I am not sure that
everyone on the

13
committee is aware that in terms of BCS Class I,

14
where you don't have to do a clinical study--I

15
don't know of a generic company that has used that

16 for
exactly the reason you are bringing up, Marvin.

17
They would be afraid that someone will go out there

18 and
say this product has never been tested in

19
humans; it was approved on the basis of a

20
dissolution. You have confidence
in this product

21 so
that people that use BCS Class I at the present

22 time
are the innovators who use it when they have a

23
SUPAC change or something like that.
So, I think

24 we
are always going to face that, and I think what

25 we
need to do is just try to do the best job that

75

1 we
can in making it happen.

2
DR. KIBBE: Let me just ask about
an

3
application of one of your recommendations to your

4 own
example. If you use methodology that is

5
developed as a weighted average, how would that

6
play out with progesterone? In
other words, what

7
kind of numbers would we start to work with?

8
DR. BENET: I mean, I do agree
with

9
weighting to the variability of the innovator

10 product.
In other words, that would be the term in

11 the
denominator that you would weight. But
there

12 are
different statistical issues that have to be

13
addressed that I can't do so we need the expert

14
statisticians to tell us how to approach that. But

15
that is what I want. I would want
a weighting on

16 the
variability of the innovator product in terms

17 of
the coefficient of variation for Cmax as one

18
criterion and for AUC as another criterion.

19
DR. KIBBE: I have always found

20
intellectually attractive the concept of three ways

21
where we could look at variability and then compare

22 it
to the generic. Is that going to help us
get to

23 the
numbers that we need to make these kinds of

24
decisions?

25
DR. BENET: Well, there is going
to have

76

1 to
be some measure of intra-subject variability.

2 We
need to know that, and I have requested the

3
agency for many years to make this a requirement

4 for
new drugs, that a measure of intra-subject

5
variability in humans or even in patients be

6
included in the approval process and be included in

7 the
package insert. So, we do have to have
that

8
measure some place.

9
I am very encouraged, even though the

10
agency does not require that, that we are starting

11 to
see with many new products, when you look at

12
their package insert, measures of intra-subject

13
variability included because it is important

14
criteria and value that clinicians want to know.

15
What is the inherent pharmacokinetic variability so

16
that then I can say is the pharmacodynamic

17
variability more than this inherent pharmacokinetic

18
variability. If they don't know
the inherent

19
pharmacokinetic variability, then they have a tough

20
time making any decision about whether the change

21 in
efficacy is related to pharmacokinetics or to

22
real variability. So, somebody
has to do this,

23
Arthur, and I think that has to come out of what

24 you
recommend.

25
DR. MEYER: Les, you put a little
bit less

77

1
weight on Cmax than you do on AUC; there is a less

2
stringent requirement. Is that
because Cmax is

3
more variable because we don't measure it very

4
precisely, or is it because Cmax is less important

5 than
AUC? And, I would quarrel that we don't
have

6
enough data for the latter conclusion.

7
DR. BENET: Well, in some cases we
do but,

8 as
was initially discussed, it is confounded.
As

9 we
all know, Cmax is a very confounded measure and

10 the
agency and many academics have spent years and

11
years in trying to develop a new measure. None of

12
them turned out to be any better.
So, it is very

13
confounded and, as was stated, is always more

14
highly variable than AUC. I know
of no case.

15
DR. MEYER: But it is the only
measure we

16
have that has any component of rate in it.

17
DR. BENET: That is correct, but
it is

18
more variable.

19
DR. VENITZ: Les, I agree with
your

20
additional recommendation to put constraints on the

21
point estimates. You mentioned
one of the reasons

22
being that the public needs to be reassured that,

23
indeed, no matter whether it is unintelligible

24 regulation or not, we do have generics that
are

25
bioequivalent.

78

1
What I am personally not certain about is

2
whether I agree with the reference scaling--and,

3
again, we are going to have some more presentations

4 on
that--because you are now, in my mind,

5
aggregating variance and mean differences, and I am

6 not
sure whether one can offset the other.
In

7
other words, if you have a large mean difference,

8 can
that be offset by differences in variance?

9
When we had the discussion last time with IBE,

10
surprisingly there were drugs out there in the

11
database that the FDA provided us with that passed

12 IBE
but wouldn't have passed ABE, which I think was

13
counter-intuitive for most of us, at least on the

14
committee, in terms that we expected IBE to be much

15
more conservative than ABE and it didn't turn out

16
that way. So, I still personally
withhold judgment

17 on
the reference scaling but I am very much in

18
favor of putting in additional constraints.

19
DR. BENET: Let me just answer
that. I

20
think having the additional constraints solves part

21 of
the problem.

22
DR. VENITZ: Yes, that was the
reason why

23 I
think the committee at that time went along with

24
that because we were worried about the IBE not

25
being conservative enough. Right
now you are

79

1
basically breaking drugs down into two categories,

2
NTIs and non-NTIs, in terms of the criteria that

3 you
are going to use or that you are proposing to

4 be
used for BE assessment.

5
DR. BENET: Yes.

6
DR. VENITZ: Can you think of
additional

7
criteria along the lines that we heard Gordon talk

8
about, that if we understand where the variability

9
comes from we might use different criteria? In

10
other words, is NTI the only thing that we have in

11
some decision tree that decides which way we are

12
going to go?

13
DR. BENET: As I said, the NTI
statement

14
there has nothing to do with science because it is

15
easy to prove bioequivalence of NTI drugs. It just

16 has
to do with confidence. So, that is why I
made

17 it
lower, because it is easy to pass.

18
I definitely believe that as we progress

19 we
are going to have different criteria, and I

20
think BCS has a real potential for it.
I have a

21 big
list, my BCS list, and I looked to see what

22
drugs were there and that is why I made sure that

23
theophylline was a Class I drug.
I think as we

24
progress--and I presented to the agency last

25
November my newest concepts in terms of using BCS

80

1 or
some sort of variant of BCS to actually predict

2
drug disposition, and I think we are going to

3
progress a lot in the next few years.

4
DR. KIBBE: Nozer?

5
DR. SINGPURWALLA: Well, just a
general

6
comment. I was pleased to hear
you acknowledge

7
that newcomers can identify things like

8
confounding, but I also think that newcomers can

9
look at an old problem and come up with new methods

10 of
addressing that. Therefore, I urge you
to pay

11
more attention to alternate methods and not get

12
committed to an old, archaic notion of confidence

13
intervals. These have been
criticized in the

14
literature. And, what we see here
is repeated use

15 of
confidence limits, and the difficulty that

16
confidence limits poses both to the FDA and also to

17 the
drug industry in getting their drugs approved.

18 So,
I am going to urge you to start paying more

19
attention to alternatives and don't dismiss it.

20
DR. BENET: I don't dismiss it,
and my

21
colleague, Dr. Scheiner, has spent a lot of time

22
informing the committee and the agency of these

23
approaches and the Bayesian approach, and I think

24 we
are all well aware of it and do recognize it.

25 It
is important to have fresh eyes and fresh views

81

1 of
these kinds of issues, but it is also important

2 to
recognize that the agency's criteria are safety

3 and
efficacy, and when we have criteria that have

4
never failed it is tough to say that we move beyond

5
that criteria to untested criteria in terms of this

6
particular issue. So, that is why
the agency must

7 be
very careful in the changes that they make.

8
DR. KIBBE: Thank you, Les. We have one

9
more speaker before the break.
Dr. Endrenyi,

10
welcome.

11
Bioequivalence Methods for Highly Variable Drugs

12
DR. ENDRENYI: Thank you.

13
[Slide]

14
This presentation was put together with

15
Laszlo Tothfalusi and I would like to acknowledge

16
that.

17
[Slide]

18
I would like to raise a number of

19
questions which I believe that this committee will

20
have to make recommendations about eventually that,

21
certainly, the agency ought to consider.
I would

22
like to go through the first part fairly quickly

23
because much of that has already been considered.

24 So,
we have the usual criterion of comparing two

25
formulations and the confidence limits for the

82

1
ratio of geometric means should be between 0.8 and

2
1.25. This has already been
stated.

3
[Slide]

4
It has also been stated that for highly

5
variable drugs this presents a problem because with

6
large variations it is very easy to hit that 0.8 to

7
1.25 and, therefore, many subjects may be needed in

8
order to satisfy that.

9
[Slide]

10
For the purpose of this presentation but

11 not
necessarily as the final word at all, the

12
coefficient of variation has been considered

13
exceeding 30 percent for highly variable drugs.

14
[Slide]

15
This slide would simply ask is there an

16
issue and this has already been asked and the

17
answer was probably yes. In this
case, two

18
formulations of isoptin are considered in the same

19
subject repeatedly, and two different occasions

20
different relationships between the two

21
formulations were obtained. So,
it looks as though

22 the
drug is not really bioequivalent with itself

23 and
that is a concern, but this has already been

24
demonstrated by Dr. DiLiberti.

25
[Slide]

83

1
This is perhaps more recent. This
was

2
obtained from Diane Potvin, from MDS, who

3
demonstrated that, indeed, things look reasonable

4 as
long as the intra-individual CV is up to about

5 70
percent but beyond that it is very difficult to

6
satisfy the criteria. There are
many, many studies

7
submitted that failed.

8
[Slide]

9 Then she went on, very kindly, to
look at

10
details of these highly variable drugs.
From this,

11 one
could conclude that there is a relationship

12
between the coefficient of variation and failure

13
rate, higher failure rate with higher coefficient

14 of
variation. Mind you, these are all
submitted

15
studies so this analysis is still biased because

16 the
company submitted them in the hope that they

17
would pass, so these are not all studies at all.

18 The second conclusion is that, indeed,

19
AUCs fail less frequently than Cmax's but they

20
still fail with a high frequency.
So, the

21
variation of AUCs should not be dismissed.

22
[Slide]

23
Study condition--perhaps I would omit this

24
almost entirely because it is considering single

25
dosing versus steady state. In
the U.S. this is a

84

1
non-issue because U.S. goes by single

2
administration even though it has been demonstrated

3 and
we know that frequently in steady state we get

4
lower variation--not frequently but not always.

5
[Slide]

6
This is a study showing that and in the

7
U.S. I think this is largely at the moment

8
irrelevant.

9
[Slide]

10
Study designs, which one to choose?
A 2 X

11 2
traditional or replicate design? It need
not be

12 a
4-period replicate design; it could be 3.

13
[Slide]

14
Now, the advantage of replicate designs

15
includes that one gets clear estimates of

16
within-subject variations.
Particularly the

17
concern would be to get a clear estimate of

18
within-subject variation for the reference product.

19 I
would note that this design is favored by K.K.

20
Midha who has worked long years and is certainly

21 one
of the foremost experts on the bioequivalence

22 of
highly variable drugs and drug products.
So,

23 his
voice ought to be respected.

24
Secondly, on the other hand, my concern is

25
that one can have a pooled criterion which could

85

1
have better properties, pooled criterion related to

2 the
test and reference products together.

3
There are issues that these replicate

4
design studies can be evaluated by various

5
procedures, and a question is whether these

6 procedures
would give the same results and,

7
therefore, would agencies be able to check how

8
those results would be calculated and were

9
calculated.

10
Another question arises, namely, is a test

11
comparing the variations of test and reference

12
products useful; is it needed?
Or, perhaps is an

13
estimate of these variations simply sufficient or

14 is
that needed?

15
[Slide]

16
Turning to the 2 X 2 crossovers, they are

17
simple; simple to execute, simple to evaluate. An

18
advantage is that there are many studies on file

19 and
they could be evaluated retrospectively.

20
Another comment is that the ratio of

21
within-subject variabilities could be estimated.

22
There are procedures that would permit this even

23
from 2 X 2 crossover studies. For
example, the

24
procedure suggested here by Guilbaud and Gould is

25 to
have for each subject the sum of the test and

86

1
reference response, AUC or Cmax in this case, and

2
then the difference of the two; plot them against

3
each other, have a linear regression and evaluate

4 the
slope, and then apply the slope in that fashion

5
which gives the ratio of the estimated variances.

6 So,
it would be possible to evaluate this ratio

7
from 2 X 2 crossovers. However,
features of this

8
procedure have not been studied and they ought to

9 be
evaluated.

10
[Slide]

11
Now, various possible methods of

12
evaluation, the usual procedure is unscaled average

13
bioequivalence with a criterion of 0.8 to 1.25 for

14 the
ratio of geometric means, the GMR. It is
also

15
possible to apply unscaled average bioequivalence

16
with expanded bioequivalence limits.
One way of

17
doing it is to present these bioequivalence limits.

18 It
has been shown that some jurisdictions do this.

19 For
example, the ratio of GMR could be between 0.75

20 and
1.33 or 0.7 to 1.43. This is one
possibility

21
which is practiced in some areas, or to expand the

22
bioequivalence limits flexibly depending on the

23
estimated variation. I shall talk
more about these

24
procedures.

25
Another approach is the scaled average

87

1
bioequivalence and, again, I shall refer to this

2 and
shall talk about this, and I also should

3
mention scaled individual bioequivalence for

4
comparisons only.

5
[Slide]

6
To talk about unscaled average

7
bioequivalence--these scissors are supposed to be

8
less than or equal signs so instead of scissors, it

9 is
less than or equal--the unscaled average, as we

10
have seen--this is a bit more formalized but, as

11 you
see here, the ratio of geometric means should

12 be
between, say, 0.8 or 1.25 or 0.75 and 1.33.

13
This is the same statement as saying that the

14
logarithmic bioequivalence limits should be plus

15 and
minus and in between is the difference of the

16
logarithmic means, and that is a useful way to look

17 at
it. Now, the procedure is simple but as
the 0.8

18 and
1.25 limits were arbitrary so would be any

19
other criteria.

20
But another concern is that whatever way

21 it
would be decided, if this is the way to go, then

22
0.75 to 1.33 is a partial solution because it may

23
help drugs with, say, 30, 40 percent intra-subject,

24
intra-individual variation but not those which have

25
higher variations and 50, 60 percent would still be

88

1 the
cut off.

2
[Slide]

3
Another approach would be to expand the

4
limits in proportion to the estimated variation.

5
This has been suggested by Boddy and coworkers.

6 So,
here there is a proportionality factor, and the

7
other factor is the estimated standard deviation,

8
intra-subject variation. This
procedure has the

9
advantage that the usual testing procedure can be

10
applied with some proviso. The
statistical power

11 is
independent of the variation and the statistical

12
power is higher, much higher than the unscaled

13
average bioequivalence with the usual criterion so

14 we
need fewer subjects.

15
On the other hand, the criterion is that

16 bioequivalence
limits, as shown there, are really

17
random variables because they include the estimated

18
standard deviation, estimated intra-subject

19
variation. So, the limit itself
is a variable.

20
Therefore, the two one-sided test procedure is not

21
quite correct, however, it is becoming

22
approximately correct with large samples.

23
[Slide]

24
Scaled average bioequivalence is very

25
similar to the previous one except that the S from

89

1 the
bioequivalence limits, here, came over to the

2
measure that we apply. So, it is
formally very

3
similar and we have developed and have recommended

4
procedures for setting the bioequivalence limits.

5
Again, the advantages are that the

6
statistical power is independent of the variation

7 and
with the same sample size is much higher than

8 the
unscaled average bioequivalence. I am
going to

9
demonstrate this. There is a
sensible

10
interpretation. The first
interpretation is very

11
similar to that applied with individual

12
bioequivalence, namely, the expected change to

13
switching is being compared with the expected

14
difference between replicate administrations and

15 one
can make sense of that.

16
A second interpretation is that the

17
standardized effect size is being applied which is

18 a
clinical interpretation. There are procedures
to

19
evaluate confidence limits. If it
is a 2 X 2

20
crossover, then non-central t-test can be applied,

21 or
there is a procedure recommended by Hyslop and

22 her
coworkers which is somewhat more involved but

23
still reasonable I think.

24
[Slide]

25
This is a demonstration comparing the

90

1
procedures and effectiveness of various approaches.

2
They include the scaled individual bioequivalence,

3
scaled average bioequivalence and unscaled average

4
bioequivalence. You see the
probability of

5
acceptance. These are results of
simulations. It

6
amounts to the probability of acceptance at various

7 distances
between the two means. The first thing

8 you
can see is that for individual bioequivalence

9 the
range is very wide. Ranges are much
narrower

10
with scaled average bioequivalence.
So, this wide

11
range raised the concern of Dr. Benet.
The second

12
observation is that scaled average bioequivalence

13 is,
indeed, much more powerful than unscaled

14
average bioequivalence. So, we
again need fewer

15
people.

16
[Slide]

17
What is the limiting variation for highly

18
variable drugs? This is obviously
a subject of

19
regulatory decision, as are the others.
The

20
procedure could be that we apply unscaled average

21
bioequivalence if the variation is less than the

22
cut-off measure and use some kind of a different

23
procedure appropriate for highly variable drugs if

24 the
variation is higher.

25
Perhaps I should go down here.
This is

91

1 the
same kind of mixed model that was suggested for

2
individual bioequivalence but, just as Dr. Benet

3
suggested, it is not reasonable that a sponsor

4
should play both ways. The
sponsor should declare

5 the
intention of using one procedure or the other

6 in
the protocol.

7
I wouldn't necessarily dismiss these other

8
possibilities. For example, K.
Midha recommends 25

9
percent. The outcome of those
probabilities that

10 you
have seen on the previous slide depend on how

11 you
set these limiting variations.
Obviously, 30

12
percent is stricter than 25 percent.
In all cases

13 you
and the agency will ask what is the practically

14
reasonable criterion that one can live with, the

15
agency can live with and the industry can live

16
with, and the public can live with.
So, don't

17
necessarily set everything on the 30 percent; do

18
consider what the effect of, say, 25 percent would

19 be.

20 [Slide]

21
Now, this method of the secondary

22
criterion has arisen in connection with the

23
features of individual bioequivalence.
So, we talk

24
about two approaches, that of individual

25
bioequivalence and today we are talking about

92

1
highly variable drugs. There are
two very

2
different concerns.

3
First of all, we have already seen that

4 for
highly variable drugs the potential variation

5 is
smaller than with individual bioequivalence.
In

6 the
case of individual bioequivalence the

7
deviations arose because the regulatory criterion

8 was
changed. A much more liberal regulatory

9 criterion
was introduced whereas in the case of

10
highly variable drugs it is a natural change of the

11
variability between the two means.
You know this

12
very well. With the usual kind of
drug the

13
variation between the means just fluctuates

14
slightly. Most of the differences
are probably

15
between the two means and are within the range of

16 10
percent. But with highly variable drugs
those

17
means also fluctuate much more.
So, to impose a

18
constraint of 10-15 percent on this natural

19
variation means that the natural fluctuation is

20
altered so the sources of the concern are very

21
different. Whereas in the case of
individual

22
bioequivalence you have to deal with the criterion,

23 here
you have to deal with the natural variation.

24
[Slide]

25
So, I would like to raise some caution.

93

1 In
addition, the imposition of the secondary

2 criterion has serious consequences. I present this

3
from my life earlier when I dealt with individual

4
bioequivalence because we had the results then; I

5
don't have many results for average bioequivalence.

6
But, again, here you have the results for

7
individual bioequivalence. This
is the probability

8
curve for the constrained criterion alone and this

9 is
then the application of the combined criterion.

10
The combined criterion is expected and

11
does always run below the two separate criteria.

12 But
when the GMR criterion is highly constricting,

13 as
in this case, then the combined criterion is

14
really a GMR criterion essentially and has nothing

15 to
do, or very little to do with the bioequivalence

16
criterion. So, if you were to
consider the

17
secondary criterion, then this slide suggests to do

18 it
with great caution and after serious

19
consideration.

20
[Slide]

21
Here are the questions again which I have

22
raised for the committee's consideration and for

23 the
agency's consideration. They certainly
suggest

24
that many of these issues require further

25
consideration and further investigation.

94

1
Originally I wanted to end with this loose and

2
compliant mode, however, I looked at the questions

3
being raised and, since after this I may have to

4
shut up, I would like to call attention to question

5
2(b) in which the application of scaling is

6
combined with the application of this secondary

7
criterion. I would like to call
your attention to

8 the
fact that these are two separate questions.

9
Both of them ought to be studied further but, to my

10
mind, the restriction criterion is much more

11
controversial and requires thorough exploration for

12 its
need as well as for its application. So,
I

13
would recommend a separation of those questions.

14
Also, I have a question about reference

15
scaling. I would certainly like
to be an advocate

16 for
scaling, but whether the scaling ought to be

17
reference scaling I would like again to be a

18
subject for study. Thank you.

19
DR. KIBBE: Thank you. Questions?

20
Jurgen?

21
DR. VENITZ: I have a question
about your

22
first simulation slide where you compare the IBE to

23 the
ABE and scaled ABE. My question basically
is

24
that you are assuming for the purposes of

25
simulation that the COVs for both test and

95

1
reference are the same, 40 percent.
Is that

2
correct?

3 DR. ENDRENYI: Yes.

4
DR. VENITZ: What would happen if
you had

5
differences in COVs between test and reference? In

6
other words, let's assume that the test product has

7
much less intra-individual variability than the

8
reference, how would that affect your curves?

9
DR. ENDRENYI: It does affect the
curves,

10 but
mainly the curve of the individual

11
bioequivalence. It affects little
the average

12
bioequivalence curve.

13 DR. VENITZ: What about the scaled average

14 BE?

15
DR. ENDRENYI: The same. But that is an

16
artifact in a way because here we consider the

17
scaling by reference product so we didn't

18
have--these were 4-period studies.
Your question

19 is
relevant if you consider the 2-period studies.

20
DR. VENITZ: Right, right.

21
DR. ENDRENYI: Which we haven't
done, but

22
that is an interesting question.
It would be worth

23
investigating.

24
DR. VENITZ: So, the answer that
you are

25
using then is the reference variation.

96

1
DR. ENDRENYI: That is right.

2
DR. VENITZ: So, you are assuming
that you

3
know but you wouldn't necessary do a 2 X 2--

4
DR. ENDRENYI: No, the estimated

5
reference.

6
DR. VENITZ: So, you could get
that from a

7 2 X
2 design?

8
DR. ENDRENYI: Well, it is a
different

9
interpretation. Yes, we could but
it has to be

10
validated whether it works or not.
We haven't done

11
that.

12
DR. KIBBE: Anybody else? Ajaz, do I see

13 you
leaning forward? No? Go ahead.

14
DR. SINGPURWALLA: I just have a
technical

15
comment. Somewhere in your slides
you had a

16
restricted maximum likelihood.
Right?

17
DR. ENDRENYI: Yes, as a possible

18
procedure.

19
DR. SINGPURWALLA: As a possible

20
procedure?

21
DR. ENDRENYI: Yes.

22
DR. SINGPURWALLA: Well, this is a

23
technical comment, the maximum likelihood is

24
advocated because of its asymptotic properties in

25 the
sense that it converges to the center.
You

97

1
know, you get the central limit theorem.
When you

2
restrict your maximum there is no assurance that

3 you
converge, the central limit theorem.

4
Therefore, the value of that process cannot be

5
really evaluated. I don't know
what impact all

6
that has on the proposals you have made but I just

7
want to caution you.

8
DR. ENDRENYI: You are absolutely
right,

9 but
the point I think was that in the case of

10
replicate design probably the procedure of

11
evaluation would have to be defined very clearly

12 and
very strictly, otherwise one can go in all

13
different directions and that will be another task

14 if
the agency goes that way.

15
DR. KIBBE: Go ahead.

16
DR. BENET: Just a quick follow-up
on

17
Laszlo's comment, I think it would be worthwhile if

18 the
agency went back and looked at the content

19
uniformity criteria and published two sets of data.

20 I
think it would be worthwhile to go back and look

21 at
the bioequivalence data and look and see how

22
often it falls within certain criteria.
You have a

23 big
database and it would be nice to see what those

24
numbers were, and I think that would be useful for

25 the
committee on the secondary criteria.

98

1
DR. YU: Actually, you will see
that in

2 the
last talk. Sam is going to talk about
data.

3
DR. KIBBE: It is always good to
have data

4
when we are having a discussion.
No one else?

5
Marv?

6
DR. MEYER: This is probably a

7
statistically ignorant question but under the

8
scaled condition, however you want to scale it, is

9 it
possible to have a product with a scale

10
confidence limit that was, say, 60-90?
If so, then

11
let's say the ratio would be somewhere around 75

12
percent and that wouldn't be acceptable.
So,

13
without the point estimate constraint you have a

14
potential for allowing 60-90 approved and 120-140

15 to
be approved.

16
DR. ENDRENYI: No--

17
DR. MEYER: Two different studies?

18
DR. ENDRENYI: In two different

19
studies--within each study it should be one and I

20
wouldn't envision between study variation and I

21
don't--I doubt it very much.

22
DR. MEYER: Even if the test product only

23
released 70 percent of its dose and the innovator

24
released 100 percent of its dose the true ratio

25
would be 0.7 and you wouldn't know that; you would

99

1 be
looking for 1.0. It is not possible?

2
DR. ENDRENYI: No, I think if it
is

3
inter-study variation, then with the low variation

4
drugs each of them would be between 0.8 and 1.25

5 but
the two in comparison with each other could be

6
quite different. That is equally
possible but it

7 is
not likely. If it is the same reference

8
product, then it is not possible.

9
DR. KIBBE: I see no other
questions.

10
Thank you very much. We will take
our break now.

11 We
will be back at 10:52.

12
[Brief recess]

13
DR. KIBBE: We have open public
hearing at

14 one
o'clock but there are no presentations to be

15
made at that time so what we will be able to do is

16
modify our schedule to try to get everything done

17 and
get back on schedule. I know there is a
lot of

18
interest in what we are talking about so we might

19
allow our speakers a little extra time and some

20
questions and answers to go a little further. I

21 see
our next speaker is at the podium, ready to go,

22
Barbara Davit.

23 Bioequivalence of Highly
Variable

24 Drugs Case Studies

25 DR. DAVIT: I am pleased to be able to

100

1
respond to one of the questions that Les raised, in

2
that we do have a survey of some of the data that

3 has
been submitted to the Division of

4
Bioequivalence.

5
[Slide]

6
When Dale and I were talking about putting

7
this presentation together for the advisory

8
committee, one of the things we thought we would

9
consider is looking at what has been submitted to

10 the
Division of Bioequivalence and to answer the

11
question of whether highly variability is a

12
significant issue in these bioequivalence studies

13 in
ANDA submissions.

14
By looking at these data and focusing on

15
some case studies, we thought also we could maybe

16
answer the questions in a limited number of cases

17 of
what is contributing to the variability or what

18 are
some of the sources of this variability.

19
[Slide]

20
So, what we were trying to do is see if

21
there is a significant problem with highly variable

22
drugs, and I would like to mention, first of all,

23
that this obviously represents a biased sample

24
because we receive predominantly studies that have

25
passed the 90 percent confidence interval criteria.

101

1 So
we obviously don't see the big picture like

2
people from industry would be seeing.
We don't see

3
what percentage that is of the total number of

4
drugs in a company's pipeline for example.

5
But of the submissions we saw, which are

6
passing studies, what percentage were for highly

7
variable drugs? Did these studies
involve

8
enrolling a large number of subjects because that

9 has
been one of the issues that has been raised

10
today, the large number of subjects that might be

11
necessary to show bioequivalence for these generic

12
products of highly variable drugs?
Also, how

13
narrow and wide are these 90 percent confidence

14
intervals? That goes along with
how many subjects

15 are
necessary for a passing study.

16
[Slide]

17
We collected data from all the

18
bioequivalence studies that were submitted to the

19
Division of Bioequivalence in 2003.
We used the

20
root mean square error as an estimate of

21
intra-subject variability. I
realize this is just

22 a
rough estimate and it is not a pure estimate of

23 the
intra-subject variability but, unfortunately,

24
most of the studies that we had to look at were

25
two-way crossover studies so the best estimate that

102

1 we
could get of the intra-subject variability was

2 the
root mean square error.

3
We defined a highly variable drug as one

4
with a root mean square error which is greater than

5
0.3, representing 30 percent intra-subject

6
variability. The data that I am
going to present

7 is
only solid oral dosage forms, and I would like

8 to
point out that all the studies that I am going

9 to
be presenting passed our 90 percent confidence

10
interval criteria, but that is because for the most

11
part we don't receive submissions of studies where

12 the
product did not pass bioequivalence criteria.

13
[Slide]

14
First of all from 2003, this was a total

15 of
212 in vivo bioequivalence studies. Of
these

16
212, looking at only those studies in which the

17
root mean square error of AUC or Cmax was greater

18
than 0.3, in 15.5 percent of these studies, AUC or

19
Cmax, was greater than 0.3. In
other words, in

20
about 15 percent of our studies the drug would

21
qualify as having highly variable characteristics.

22
Most of this was due to Cmax and this has been

23
discussed today. So, in about 13
percent of the

24
total only Cmax was highly variable.
There were no

25
studies in which only AUC was highly variable. But

103

1
there were 5 studies in which both AUC and Cmax

2
were highly variable, and this was 2.5 percent of

3 the
total.

4
[Slide]

5
This goes along with the previous slide

6 and
it just shows the number of studies in which we

7 saw
a root mean square error of a particular value

8 for
Cmax. There is an error in this
particular

9
slide in your handout but this is the correct

10
slide. Really, obviously, for
most of the Cmax

11
values the root mean square error is below 0.3. I

12 have
a line here representing 0.3. I think I
said

13
earlier that 15 percent of all the studies, 15.5

14
percent of all the studies that came in had a root

15
mean square error for Cmax of greater than 0.3.

16
[Slide]

17
This is for AUC. Of course, the AUC is a

18 lot
less variable than Cmax. Really, for the
most

19
part the root mean square errors were hovering

20
around 0.1 to 0.15, so quite a bit less variability

21 in
AUC than Cmax.

22
[Slide]

23
One of the questions that we wanted to ask

24 was
what is contributing to this variability.

25
Since for a lot of products we look at

104

1
bioequivalence studies in fasted subjects as well

2 as
fed subjects, we wanted to see what impact was

3
having on variability. I
mentioned 33 studies.

4
This represented a total of 24 of the ANDAs that

5
were submitted and reviewed in 2003.
Of these,

6
both AUC or Cmax were highly variable in both the

7 fed
and fasted studies. In 8 of these the

8
pharmacokinetic parameters were highly variable in

9
only the fed study, and for 7 the PK parameters

10
were highly variable in only the fasted study. But

11
this is a little bit skewed too because we have

12
submissions, for whatever reason, which contain

13
only a fed study and submissions that contain only

14 a
fasted study--not a lot but it does happen.

15 [Slide]

16
This shows some of our data. I
think

17
these are all the Cmax values from the 212 studies

18 I
was talking about in which Cmax was variable in

19
only the fed study and not the fasting study. So,

20 this
would suggest, of course, that we are seeing

21
variability because of food effects.
I am not

22
giving the names of the drugs but I have

23
illustrated them by class.

24
There is a variety of reasons I think for

25 the
variability. Some of these are
prodrugs. We

105

1
have a number of angiotensin converting enzyme

2
inhibitors and most of these are prodrugs.

3
Generally the parent is present at low

4
concentrations so this could contribute to the

5
variability. A number of these
drugs also are

6
highly metabolized and this would contribute to the

7
variability. But, in this case,
obviously there

8 was
a food effect. The variability was
observed in

9 the
fed state, not in the fasting state. In
these

10
studies too the number of subjects ranged from

11
about 27 to 51 I guess, so all over the place in

12
terms of numbers of subjects.

13
[Slide]

14
It is pretty unusual to only see a highly

15
variable Cmax in the fasting study and not the fed

16
study, and this occurred in only two cases last

17
year. These were both angiotensin
converting

18
enzyme inhibitors, both prodrugs.
For one of them

19 the
bioequivalence was based on measuring the

20
parent. For the other one the
company could not

21
measure the parent despite I guess a number of

22
attempts. This is actually true
for pretty much

23
everyone who has worked with this particular drug.

24 So,
the bioequivalence here is only based on the

25
metabolite. But that is quite
rare. In the vast

106

1
majority of submissions that we have the

2
bioequivalence is based on the parent.

3
[Slide]

4
This table shows the Cmax data where Cmax

5 was
highly variably in both fed and fasted studies.

6 So,
for this drug product obviously there will be

7
highly variable regardless of whether it is the fed

8
study or the fasted BE study.
This was six drug

9
products, various drug classes, various reasons for

10
variability; some prodrugs, some highly metabolized

11
drugs; some drugs that undergo extensive first-pass

12
metabolism. The number of
subjects varied from I

13
guess 18 to 57.

14
[Slide]

15
Finally, this table is for two-way

16
crossover studies and shows the data for which both

17 AUC
and Cmax were highly variable, and this was for

18
four drug products. For the one
that I have shown

19 in
yellow, for this particular product both AUC and

20
Cmax met the highly variable criteria in both the

21 fed
and the fasting state. For the other
drugs

22
there was high variability in the fed but not

23
necessarily the fasted, or Cmax and not necessarily

24
AUC. So, this was four drugs that
fell in this

25
class. The number of subjects that
the companies

107

1
used varied from 26 to 62.

2
[Slide]

3
In trying to explore some of the sources

4 for
this variability, we wanted to compare the

5
intra-subject variability for the test versus the

6
reference product. We don't see
very many

7
replicate design studies anymore.
In this

8
particular class of drugs we only had two

9
submissions last year so these are the data from

10 the
two submissions.

11
These data are a good sign because what

12
they show is that the variability, based on the

13
root mean square error, was comparable for the test

14 and
the reference product for both of these drug

15
products. That is obviously what
we are looking

16 for
because we want to see people achieve a generic

17
product that is the same as the reference product.

18 So,
in this case I would say the variability was

19
comparable, test versus reference.

20
One study used 33 subjects. The
other,

21
this would obviously fall into a category where it

22
necessitated a lot of subjects because this was not

23
only 77 subjects, it was also a replicate design so

24 it
meant that each of these 77 subjects received

25 the
drug product four times, on four occasions.

108

1 So,
this was quite an extensive study.

2
[Slide]

3 Another question we wanted to ask
was are

4
there ever cases in which the pharmacokinetic

5
variability is a function of the drug product as

6
opposed to the drug substance. We
found two

7
instances last year, two different drug products

8 and
I will call them drug C and drug D. This
was

9 the
same RLD for both studies for drug C and the

10
same RLD for both studies with drug D.
Drug C was

11 an
extended release tablet. Drug D was an

12
immediate release tablet.

13
We will look at drug C first. In
one

14
study, conducted by one applicant, using I guess 33

15
subjects in the fasted and 35 subjects in the fed,

16
this product would not qualify as a highly variable

17 drug. Notice root mean square errors of 0.18,

18
0.11, 0.21, 0.24. However, for
the same reference

19
product, in other words it is the same product,

20
different formulation, another company, 0.31, 0.38,

21
0.25, 0.34.

22
This could be due to a number of reasons.

23 I
looked at the data and, obviously, the extended

24
release dosage forms are more complex than the

25
immediate release dosage forms and the two

109

1
formulations were quite different.
So, there could

2
have been, you know, differences in variability due

3 to
the formulation. Also, the
bioequivalence

4
studies were done at different sites.
I looked at

5 the
assays. They were both LCMS assays. I didn't

6 get
the specifics of the extraction methods but I

7
noticed that the two studies had different limits

8 of
quantitation and there were different doses in

9 the
two studies. I am not sure how much of a

10
factor this was. This was an
extended release

11
product for which I believe there were three

12
different strengths. One company
submitted a study

13 on
the highest strength and I think used two times

14 15
mg, which was 30. The other company did
studies

15 on
5 mg and used 4 times 5 mg, which was 20.
So,

16
different doses in the two studies.
So, there are

17 all
these factors that could be contributing to the

18
variability. At least, those are
the factors I

19
could think of.

20
Drug D--this was an interesting issue.

21
Once again, in the hands of one sponsor, one

22
applicant, we saw root mean square errors of 0.16,

23
0.25, 0.13 and 0.2; the other applicant, 0.38,

24
0.55, 0.22 and 0.24. This was an
immediate release

25
product and I noticed that the formulations of

110

1
these two were qualitatively identical;

2
quantitatively there were some differences.

3
These were done at two different sites and

4 in
this particular application the bioanaytical

5
methods were done at a CRO that we have had some

6
issues with in the past. They
seemed to be having

7
problems with some of their data.
So, it could

8
have been a contributing factor here.

9
I would like to stress that of all the

10
applications that we saw last year, these were the

11
only four in which we saw that there was a

12
difference which was possibly due to drug

13
formulation or possibly due to where the studies

14
were done that was contributing to the high

15
variability.

16
[Slide]

17
Then we thought we would look at how many

18
study subjects are usually enrolled in these

19
studies. Once again, I emphasize
that this is

20
really a biased sample because we only see the

21
studies that have passed. We
don't know how many

22
tries this represents. We don't
know how many

23
studies were done where the company just couldn't

24 get
the study to pass the confidence interval

25
criteria so these are just the passed studies.