1
DEPARTMENT OF HEALTH AND HUMAN SERVICES
FOOD AND DRUG ADMINISTRATION
CENTER FOR DRUG EVALUATION AND RESEARCH
ARTHRITIS ADVISORY COMMITTEE
Tuesday, July 30, 2002
8:00 a.m.
Holiday Inn Bethesda
Versailles I and II
8120 Wisconsin Avenue
Bethesda, Maryland
2
PARTICIPANTS
Gary S. Firestein, M.D., Chairman
Kathleen Reedy, R.D.H., M.S., Executive Secretary
MEMBERS
Jennifer Anderson, Ph.D.
Kenneth D. Brandt, M.D.
Leigh F. Callahan, Ph.D.
John J. Cush, M.D.
Ildy M. Katona, M.D., CAPT, MC, USN
Susan M. Manzi, M.D.
Wendy W. McBrair, R.N., M.S., C.H.E.S.
Yvonne S. Sherrer, M.D.
GUESTS
ARTHRITIS ADVISORY COMMITTEE
Steven B. Abramson, M.D.
Raymond A. Dionne, D.D.S., Ph.D.
Janet D. Elashoff, Ph.D.
Clifford J. Woolf, M.D.
ANESTHETIC AND LIFE SUPPORT ADVISORY COMMITTEE
Michael Ashburn, M.D., M.P.H.
Nathaniel P. Katz, M.D.
Mitchell B. Max, M.D.
NONPRESCRIPTION DRUGS ADVISORY COMMITTEE
Alastair Wood, M.D.
INDUSTRY REPRESENTATIVE
Charles H. McLeskey, M.D.
GUESTS
David Borenstein, M.D.
John T. Farrar, M.D. MSCE
Vibeke Strand, M.D.
3
C O N T E N T S
PAGE
Call to Order and Introductions,
Gary S. Firestein, M.D. 4
Meeting Statement,
Kathleen Reedy 6
Welcome,
Lee S. Simon, M.D. 9
Comments, Charge,
Lawrence Goldkind, M.D. 10
ABC Metrics of Acute Pain,
James Witter, M.D., Ph.D. 11
Estimates of Dosing Intervals,
Lawrence Goldkind, M.D. 28
Edward D. Bashaw, Ph.D. 30
Lawrence Goldkind, M.D. 41
Safety Databases for Acute Analgesics,
Lourdes Villalba, M.D. 59
Discussion Points #1, 2, 3 77
Open Public Hearing:
Eugene Laska, Ph.D. 113
Nijab Babul, Pharm. D. 125
Further Discussion of Proposal for Criteria to
Obtain a Chronic Global Pain Indication 132
Responder Index, a Model,
Vibeke Strand, M.D. 174
Discussion Point #4 208
Summary,
Lee S. Simon, M.D. 230
4
1 P R O C E E D I N G S
2 Call to Order and Introductions
3 DR. FIRESTEIN: Good morning, and welcome
4 to the second day of the Arthritis Advisory
5 Committee meeting. I am Gary Firestein still. I
6 think because there may be some people here today
7 that were not here before we can just go around the
8 room again quickly with introductions since this
9 represents a separate meeting. Then, we can have
10 the meeting statement from Kathleen Reedy. Again,
11 I am Gary Firestein.
12 DR. SHERRER: I am Yvonne Sherrer,
13 rheumatologist.
14 DR. CUSH: Jack Cush, rheumatologist,
15 Presbyterian Hospital, Dallas.
16 DR. CALLAHAN: Leigh Callahan,
17 epidemiologist, University of North Carolina,
18 Chapel Hill.
19 DR. WOOD: Alastair Wood, Vanderbilt.
20 MS. MCBRAIR: Wendy McBrair, nurse and
21 health educator, consumer representative, with
22 Virtua Health in New Jersey.
23 DR. WOOLF: Clifford Woolf, Harvard
24 Medical School and Massachusetts General Hospital.
25 DR. DIONNE: I must have said something
5
1 offensive yesterday because they took my mike
2 away--
3 [Laughter]
4 --but I am Ray Dionne. I am a clinical
5 pharmacologist, from NIDCR.
6 DR. WITTER: Jim Witter, from FDA.
7 DR. GOLDKIND: Larry Goldkind, FDA.
8 DR. SIMON: Lee Simon, Division Director
9 550, FDA.
10 DR. MCLESKEY: Charlie McLeskey, from
11 Abbott Labs, and serving as the industry
12 representative.
13 DR. STRAND: Vibeke Strand,
14 rheumatologist. I teach at Stanford and I am a
15 consultant.
16 DR. BORENSTEIN: David Borenstein,
17 rheumatologist, clinical professor, George
18 Washington University.
19 DR. FARRAR: John Farrar, neurologist,
20 Instant Pain Management at the University of
21 Pennsylvania.
22 DR. ELASHOFF: Janet Elashoff,
23 biostatistics, Cedars-Sinai and UCLA.
24 DR. ASHBURN: Michael Ashburn,
25 anesthesiologist, University of Utah, Pain
6
1 Management Center.
2 DR. ANDERSON: Jennifer Anderson,
3 statistician, Boston University Medical Center.
4 DR. KATZ: Nathaniel Katz. I am a
5 neurologist from Boston.
6 DR. MANZI: Susan Manzi, rheumatologist,
7 University of Pittsburgh.
8 DR. ABRAMSON: Steve Abramson,
9 rheumatologist, NYU Hospital for Joint Diseases.
10 DR. KATONA: Ildy Katona, pediatric
11 rheumatologist from the Uniformed Services
12 University.
13 DR. BRANDT: Ken Brandt, rheumatologist,
14 Indiana University.
15 MS. REEDY: Kathleen Reedy, Food and Drug
16 Administration.
17 Meeting Statement
18 And, this is the meeting statement for the
19 Arthritis Advisory Committee meeting of July 29th
20 and 30th, 2002. It is the same one; you can sing
21 along if you like.
22 The following announcement addresses the
23 issue of conflict of interest with respect to this
24 meeting and is made a part of the record to
25 preclude even the appearance of such at this
7
1 meeting.
2 The Food and Drug Administration has
3 approved general matters waivers for the following
4 special government employees which permits them to
5 participate in today's discussions: Gary
6 Firestein, Kenneth Brandt, Ildy Katona, Yvonne
7 Sherrer, Susan Manzi, Jennifer Anderson, John Cush,
8 Alastair Wood, Nathaniel Katz, Michael Ashburn,
9 Janet Elashoff, Mitchell Max, Raymond Dionne,
10 Steven Abramson.
11 A copy of the waiver statements may be
12 obtained by submitting a written request to the
13 agency's Freedom of Information Office, Room 12A-30
14 of the Parklawn Building.
15 In addition, Leigh Callahan, Frank
16 Davidoff and Wendy McBrair do not have any current
17 financial interests in pharmaceutical companies,
18 therefore, they do not require a waiver to
19 participate in today's discussions.
20 We would like to note for the record that
21 Ms. McBrair's employer's interests in two drug
22 companies are exempt under 2640.203(g).
23 The topics of today's meeting are issues
24 of broad applicability. Unlike issues before a
25 committee in which a particular product is
8
1 discussed, issues of broad applicability involve
2 many industrial sponsors and academic institutions.
3 The committee participants have been
4 screened for their financial interests as they may
5 apply to the general topics at hand. Because
6 general topics impact so many institutions, it is
7 not prudent to recite all potential conflicts of
8 interest as they apply to each member, consultant
9 and guest.
10 FDA acknowledges that there may be
11 potential conflicts of interest, but because of the
12 general nature of the discussion before the
13 committee these potential conflicts are mitigated.
14 We will also like to note that Dr. Charles
15 McLeskey is participating in today's meeting as a
16 non-voting industry representative. As such, he
17 has not been screened for conflicts of interest.
18 In the event that the discussions involve
19 any other products or firms not already on the
20 agenda for which FDA participants have a financial
21 interest, the participants' involvement and their
22 exclusion will be noted for the record.
23 With respect to all other participants, we
24 ask in the interest of fairness that they address
25 any current or previous financial involvement with
9
1 any firm whose product they wish to comment upon.
2 DR. FIRESTEIN: Thank you very much. Now
3 Lee Simon will welcome everybody again.
4 Welcome
5 DR. SIMON: I think that yesterday was an
6 intriguing day for the committee members and I
7 think certainly for us, over here at the agency.
8 Again, I would like to thank you all for making the
9 effort to come and participate even for the second
10 day. I am even more impressed--everybody is still
11 here and suffering through the heat wave we are
12 having, although I am told it is not so much the
13 heat wave; it is the expectation of Washington.
14 I would like to make mention of two
15 things. One is that, again, this is a combination
16 committee from 170, OTC and the Arthritis
17 Committee. So, there are members from everywhere
18 and I think it is very important for us to have a
19 mixture of people commenting on these particular
20 issues.
21 Secondly, as we had a meeting with the NIH
22 and the FDA in March, we are proposing to have
23 another meeting in some months on the issue of
24 function, healthful quality of life and outcomes in
25 pain, both acute and chronic. Ray Dionne and Jim
10
1 Witter are planning to apprise the wonderful
2 experience we had previously, and I have been
3 advised to inform everyone here in the audience of
4 that. In fact, for the companies' benefits, the
5 sponsors' benefits, this meeting will include your
6 participation so that we can truly get opinions
7 from all aspects of interest in this particular
8 field. So, look forward to receiving invitations
9 for this particular upcoming meeting sometime this
10 winter.
11 Back to you, Gary.
12 DR. FIRESTEIN: Thank you. There will be
13 some comments and discussion of our charge from Dr.
14 Goldkind.
15 Comments, Charge
16 DR. GOLDKIND: Thank you. Again, I want
17 to thank the committee members for taking time out
18 of their schedules to spend two days with us.
19 Yesterday we dealt with a lot of
20 conceptual issues primarily related to chronic
21 pain. While there wasn't unanimity and closure on
22 every point, the discussion we had was very helpful
23 and, hopefully, enlightening for you as well.
24 Today we will be shifting a little bit and talking
25 primarily about acute pain, probably a little more
11
1 detailed in terms of study design and analysis, and
2 we look forward to another fruitful and stimulating
3 day.
4 DR. FIRESTEIN: Thank you. In addition,
5 at some point during the day, probably during the
6 10:45 to 11:45 block, Lee has asked us to revisit
7 some of the issues from yesterday strictly with
8 regard to global pain indications, and we are going
9 to end up going around the table and soliciting
10 two-minute opinions. That goes for everybody,
11 two-minute opinions on the two questions of how
12 many indications might be required and how many
13 domains do you think would be important. So, we
14 will come back to that a little bit later on this
15 morning.
16 The first speaker today is Jim Witter, who
17 is going to talk about ABC metrics for acute pain.
18 ABC Metrics for Acute Pain
19 DR. WITTER: Good morning. Kathleen, I
20 was looking for the bouncing ball before so I could
21 follow you!
22 [Slide]
23 As Larry said, we are going to have a
24 little bit of a shift today and we will start off
25 talking about acute pain and, hopefully, go from
12
1 there. But we will be transitioning eventually
2 back to chronic pain by the time the day is over.
3 [Slide]
4 In terms of acute pain, the argument I
5 guess could go that what we need to do is to be as
6 informative--again, we are discussing labels so we
7 want to be as informative as possible about the
8 information that goes into the label for something
9 to treat acute pain. We had a discussion yesterday
10 about acute pain versus treatment in an acute
11 situation.
12 But what we have I think are really two
13 scenarios. We have an outpatient setting and an
14 inpatient setting where we might find ourselves in
15 need of acute analgesics. For example, for
16 outpatient settings, most of us have experienced I
17 think minor injuries, such as a sports injury.
18 Some of us have experienced dysmenorrhea.
19 Hopefully, fewer of us have had a major injury such
20 as a motor vehicle accident. Then, some of us
21 actually volunteer to have surgery. Now, the
22 analgesics that are applied in these situations are
23 for the most part oral, not exclusively but mostly.
24 On the other hand, in an inpatient setting
25 we again are looking at surgical settings and these
13
1 are of the non-elective and the elective type.
2 What I have indicated here by the stars are some of
3 the models or some of the clinical situations in
4 which drugs have been studied and ultimately have
5 been approved so this isn't that we are taking a
6 major change of course here.
7 [Slide]
8 I would like to take a second and talk
9 about the analgesic box. Some people would call it
10 the analgesic black box. What I have tried to
11 depict here is a pain relief curve. There is some
12 event over here that causes one to have pain and
13 you take a drug and, at some point in time then
14 there is this concept known as onset of relief.
15 The pain relief continues and goes to a certain
16 amount. This has been described in the old 1992
17 guidance document and in the EMA document as the
18 peak. We talked about it yesterday as the pain
19 curve, the whole thing, and today I am now calling
20 this the effect size. So, this pain relief goes up
21 and lasts for a period of time and then it ends.
22 We should be able to, particularly in a
23 single-dose experience, really define these
24 parameters of onset-- what I am calling here effect
25 size, and duration quite accurately if we do our
14
1 homework.
2 [Slide]
3 For acute pain the needs are to look at
4 these concepts of the onset of meaningful pain
5 relief, its duration, the effect size and we should
6 establish these then in circumstances of acute
7 inpatient and outpatient settings.
8 [Slide]
9 That leads us then to what we have termed
10 the ABC's of acute pain metrics, that, in fact, you
11 may not be able to accomplish all of these tasks in
12 one trial and you may need to break this up. So,
13 that is what we have done.
14 The A component is really getting at the
15 concept of onset of meaningful pain relief. What
16 we need to do is, to the best of our ability,
17 establish this time very accurately. This onset
18 should occur more frequently in drug versus placebo
19 patients. It should be established in a variety of
20 outpatient and inpatient settings, as I have
21 described. And, this is really the single-dose
22 experience.
23 [Slide]
24 I have depicted here a pain by time curve,
25 a little bit different than the other presentation,
15
1 the other slide. We have pain intensity which is
2 decreasing in a general sense. I have depicted two
3 patients here, patient 1 and patient 2 and at some
4 point along this curve these patients will let us
5 know that they have established the onset of
6 meaningful pain relief.
7 This is something that is not necessarily
8 the same for everybody. So, I think what we need
9 to do is make sure that while we are measuring pain
10 intensity we also, particularly in the beginning,
11 are measuring pain relief so we know how these two
12 correlated because this is really a patient-derived
13 outcome.
14 [Slide]
15 If we take an individual responder
16 approach to this situation and this would seem to
17 make sense--process analytical technology for an
18 analgesic and for pain because pain is such an
19 individual experience. So, the individual
20 responder approach then focuses on a single person,
21 not the group. It allows efficacy assessment to be
22 very individualized, which we will be talking about
23 later as well. It has the potential of eliminating
24 imputation. We talked yesterday about forward
25 filing of diaries. Michael Hufford talked to us
16
1 about that, and we all thought that was almost
2 comical. We heard from Dr. Lu about last
3 observation carried forward and other metrics to
4 complete data. But if we can eliminate this, I
5 think we all agree it would be better.
6 [Slide]
7 An individual response then for acute pain
8 in terms of onset and duration for a single dose--I
9 think the argument could be made that pain
10 intensity should be measured throughout the entire
11 trial. This includes not only the beginning but
12 also at the end, when a patient either rescues or
13 is censored, so we understand what is going on
14 throughout the trial. Pain relief probably should
15 be measured at least early to establish meaningful
16 pain relief. If we do this properly, we should be
17 able to really capture 100 percent of information
18 on the patient's response to the analgesic,
19 particularly during the single-dose experience.
20 [Slide]
21 What I have tried to do here is give us
22 some idea of what I guess might be meant by the
23 effect size. I have drawn some theoretical lines
24 here. This says threshold; this says complete
25 response. What I have depicted is the placebo drug
17
1 which crosses this threshold and goes to a certain
2 point. We then have a drug which crosses the same
3 threshold but goes beyond where the placebo
4 response was and ends here. This is the concept of
5 complete pain relief which is not happening,
6 obviously, in this case.
7 But can we say then that the difference
8 between the two blue lines here is really what we
9 mean by the effect size? In fact, the difference
10 between this line and this line is what we mean by
11 that concept of a minimally clinically important
12 difference. This is what we are searching for
13 really because that is the difference from placebo.
14 Can we, in fact, then really quantitate this
15 response in a meaningful way?
16 [Slide]
17 The B of the ABC really refers to
18 duration. What it is attempting to do in these
19 series of studies would be to define the dosing
20 interval, again, based on clinical data once more
21 from outpatient and inpatient settings. So, here
22 we are talking about the day 1 experience but it is
23 the multiple-dose experience on day 1 if that is
24 applicable for this particular drug.
25 We would then need to factor into these
18
1 metrics the concept of rescue in an outpatient
2 setting or the use of concomitant medication such
3 as opioids in an inpatient setting.
4 [Slide]
5 The C component is really meant to give us
6 an idea of the minimally effective dose, and that
7 is important because, you recall, yesterday one of
8 the things that we discussed was our concern about
9 carrying forward with analgesics, particularly
10 analgesics studied in an acute setting, where the
11 doses may be different than the doses that are
12 carried forward in a more chronic setting.
13 If we have compounds which are not always
14 going to be applicable and utilized, for example
15 something like NSAIDs which we know are going to be
16 used for the most part for something like OA, but,
17 if we have medicines that have a very narrow
18 therapeutic window but are really intended for an
19 acute setting, we want to be sure that if they are
20 used in what really would be off-label use that we
21 have the lowest effective dose to be used in that
22 situation. So, that is what the C portion of the
23 studies are really intended to do.
24 Again, this is not intended to really
25 inform chronic use. If there is a reason that
19
1 these compounds can be used in a chronic setting,
2 we would encourage sponsors to do those studies and
3 go for the indication. Again, establishing this in
4 two settings, outpatient and inpatient, and this is
5 a multiple dose over several days and the metrics
6 may need to change in the sense of what we are
7 interested in, as Dr. Lu had talked about
8 yesterday, the area under the curve versus the
9 onset peak duration mentality.
10 Once again, we are trying to establish the
11 safety and efficacy here and we begin on day 2.
12 So, day 1 in this particular series of studies is a
13 time frame where we wouldn't have to be looking at
14 any components of efficacy. These patients could
15 take basically anything that they wanted. The
16 randomization would then begin on day 2. So, what
17 we are most interested in is from day 2, day 3 and
18 on.
19 [Slide]
20 Acute pain has special issues with it,
21 some of which we talked about yesterday. Pain is
22 not equal in intensity or duration in various
23 settings. For example, the pain after a dental
24 extraction is not necessarily the same as after
25 having bypass surgery, although maybe Leigh might
20
1 disagree. Pain does tend to improve with time.
2 That is something we discussed yesterday. We will
3 hear more today from Dr. Bashaw, but PK estimates
4 in clinical results may really describe different
5 aspects of pain relief in that PK may be more
6 informative about early onset, for example, and may
7 also then inform us about safety later. What I
8 mean here is that if we have a compound that
9 supposedly has a short half-life but in fact hangs
10 around, for whatever reason, for days, and days,
11 and days and has a very narrow therapeutic window,
12 if the pain scores suggest that needs to be dosed
13 more than once day we may have an issue of
14 toxicity. In fact, we are faced with such issues.
15 [Slide]
16 So, the label in an acute pain setting
17 should be as informative as possible and should
18 contain information regarding onset, duration and
19 minimally effective dose from two clinical
20 settings, outpatient and inpatient.
21 [Slide]
22 If we cast in stone, so to speak, these
23 concepts of acute and chronic, and if this were a
24 river of pain I guess we are concerned about the
25 bridging that needs to be done here because it may
21
1 be a time, as has been argued, that there is a
2 transition from acute to chronic, wind up,
3 plasticity, those types of issues. So, we should
4 be paying attention to this interval between here
5 and not lose sight of it.
6 If studies are conducted properly we may
7 be able to support meaningful labeling claims for
8 safety. We may, in fact, be able to get something
9 for chronic pain if the studies would be supportive
10 to push in that area, and we would encourage that
11 if it makes sense. Or, this may also be
12 informative for mechanistic claims that we talked
13 about yesterday. So, it may be that this is the
14 perfect time to be studying for some of these
15 mechanistic claims, this time interval.
16 [Slide]
17 Why all the concern? Why don't we just
18 leave things the way they are? Things have been
19 working okay. Here is a drug that is in the PDR.
20 I have given it the designation of X just to
21 anonymize it a bit. This is the clinical study
22 section. This is the entire section. It says: "In
23 single-dose studies of post surgical pain
24 (abdominal, gynecological, orthopedic) 940 patients
25 were studied at doses of one or two tablets. Drug
22
1 X produced greater efficacy than placebo" and I
2 have left out a few words here just to try to
3 maintain the blind, "no advantage was demonstrated
4 for the two-tablet dose." So, this looks like one
5 tablet is pretty effective.
6 [Slide]
7 Elsewhere in the label, under dosing and
8 administration, it says that this is "indicated for
9 the short-term (generally less than 10 days)
10 management of acute pain."
11 [Slide]
12 "The recommended dose of drug X is one
13 tablet every 4 to 6 hours, as necessary. Dosage
14 should not exceed 5 tablets in a 24-hour period."
15 The question is how did the clinical
16 trials inform this dosage and administration
17 scheme? There seems to be a gap here. This is an
18 approved compound.
19 [Slide]
20 Could we make the case then that some of
21 the ideal characteristics for a pain metric in this
22 situation should be that it should be easy and
23 understandable to patients and clinicians in the
24 labeling and in clinical trials. It should be
25 applicable across studies to facilitate IND
23
1 development and eventual NDA approval. It should
2 define a clinically meaningful result so that it is
3 a useful addition to our pain armamentarium. It
4 should be valid in a variety of pain conditions,
5 and it should be achievable with current meds, but
6 also we need to think about having some kind of a
7 tiered structure, which we have been talking about,
8 so that we can really define and acknowledge
9 important differences in drugs as they are
10 developed.
11 [Slide]
12 Taking a responder analysis plan into a
13 pain setting, it has the potential to characterize
14 pain, as I have said, at an individual level in
15 both acute and chronic situations, and Dr. Strand
16 will be talking about the chronic situation later.
17 This may then be useful to allow a
18 comparison of relative efficacy. This is against
19 placebo or standard of care, not between drugs, in
20 clinical trials in acute pain and in chronic pain.
21 [Slide]
22 If the hypothesis is correct, if it is
23 properly constructed and validated, a responder
24 analysis could be a major advance in clinical
25 analgesia because it is currently not used. Later
24
1 we will be having more discussion about the concept
2 of outcomes and domains, but I will discuss them
3 here too. I think what we can say at this point is
4 that if we can come to an agreement on outcomes or
5 domains, we can do that even if we don't
6 necessarily have the instruments because we can
7 develop the instruments later. But if we can agree
8 on the domains, that is definitely a step forward.
9 [Slide]
10 Just to step back for a second and look at
11 the responder analysis that we do have in the
12 division, the ACR, American College of
13 Rheumatology, 20 responder analysis, and this is
14 for rheumatoid arthritis, and this is really in a
15 lot of ways a symptomatic responder analysis. What
16 you have then to be approved for the indication of
17 the signs and symptoms of rheumatoid arthritis, if
18 you are successful with this metric you can then be
19 approved, assuming you are safe. So, what you have
20 to do is have a 20 percent improvement in swollen
21 and tender joint counts. Those are required
22 endpoints for this particular analysis. Then you
23 can have three of the five following, a patient or
24 physician global, a pain score, a modified health
25 assessment questionnaire or some kind of an acute
25
1 phase reactant.
2 As Lee had mentioned and we talked about
3 yesterday, we had the NIH-FDA workshop back in
4 March. At that meeting we had a discussion of the
5 responder approaches and certain domains were
6 discussed. These included pain, rescue medication,
7 patient global, health-related quality of life,
8 physical function/disease specific measures,
9 economic organ damage concerns, the issue of
10 suffering which you heard about from Dr. Verburg
11 yesterday, and adverse events. These were
12 discussed as possible domains to be in some kind of
13 an analgesic responder approach.
14 [Slide]
15 For the discussion this morning, I have
16 whittled these down to the following that we should
17 maybe be considering if we want to take this
18 tactic, pain, concomitant medications, rescue
19 medications, patient global, health-related quality
20 of life, physical function, adverse events. Those
21 are the ones that maybe make the most sense in this
22 particular situation.
23 [Slide]
24 Were we to take this approach, could we
25 begin to think about fashioning a responder
26
1 analysis by looking at our studies, our A, B and C
2 type studies, and thinking through what needs to be
3 applied or characterized in those settings? For
4 example, for pain intensity the argument would be
5 that that should be in all these studies. Pain
6 relief, maybe more so in the onset and dosing
7 interval. It may not be as important in the
8 multiple-day use settings. Patient global might
9 apply in all the settings, and continuing along.
10 So, we may be able to already begin to get a sense
11 of what a responder analysis might look like in an
12 acute pain setting.
13 [Slide]
14 Let's just take a hypothetical example.
15 It might be a bit hard to see. It is an AR20/12.
16 So, AR then would imply that analgesic relief has
17 been established. With an NSAID type compound that
18 has generally been within an hour, but that time
19 frame isn't necessarily applicable, for example,
20 were we to develop a compound that would treat
21 neuropathic pain, something that occurs
22 sporadically like trigeminal neuralgia. That might
23 not be the right kind of a time frame but, in any
24 event, AR20 would refer to percent pain relief over
25 the standard of care/placebo, and 12 would refer to
27
1 the hours of relief.
2 [Slide]
3 So, let's take a hypothetical drug that
4 has two forms. This comes in a 100 mg and 300 mg
5 variety. This is what a future potential trial
6 session might look like and it would describe in
7 there then the A, B and C, how the onset dosing
8 interval and lowest effective dose were actually
9 established in outpatient and inpatient settings.
10 [Slide]
11 So, this drug at the 300 mg strength in
12 the indication section may look something like
13 this: Drug X is indicated for acute pain. It is
14 described as AR90/24 so it is a pretty potent
15 medicine; it lasts for 24 hours. See the details
16 in "clinical trials" and daily use should not
17 exceed five days. Again, what we are trying to
18 establish here is that in acute setting with some
19 of these medicines, they may not be able to safely
20 be used in a more chronic setting.
21 [Slide]
22 With the 100 mg strength of this
23 particular compound, it may look as follows. It is
24 also indicated for acute pain. Here, it is
25 described as an AR20/24, and it would say daily use
28
1 should end when the pain has resolved or can be
2 managed in another way, getting at this idea that
3 acute pain for the most part resolves.
4 [Slide]
5 Without further delay, I would like to
6 introduce Dr. Goldkind, who will be talking to us
7 more, along with Dr. Bashaw, about the uses of dose
8 and dosing interval. Dr. Villalba will be talking
9 about some of our experience with certain compounds
10 in the division. Dr. Strand will be giving us some
11 more thoughts about the responder analysis,
12 particularly in a chronic pain setting. Then, our
13 own Dr. Simon will wrap everything up for us later.
14 Estimates of Dosing Intervals
15 DR. GOLDKIND: Thank you, Jim. I want to
16 highlight the extent to which our discussions and
17 our talks today are really aimed at labeling
18 information. A lot of Jim's talk and, hopefully,
19 mine will really focus not only on minimum
20 requirements for approval but actually what kind of
21 data we should be collecting to inform the label.
22 [Slide]
23 I will be playing tag with Dr. Bashaw, who
24 is the team leader that is affiliated with our
25 division. He is in the Division of Pharmaceutical
29
1 Evaluation.
2 [Slide]
3 An ideal analgesic is one that would be
4 once a day, 100 percent effective in 100 percent of
5 patients without adverse effects. Unfortunately,
6 most drugs available today don't meet those
7 criteria. Most of the time we have multiple doses
8 per day that are needed in the acute setting,
9 suboptimal pain relief and dose-limiting
10 toxicities.
11 [Slide]
12 Therefore, the majority of patients and I
13 imagine everybody in this room as a patient, if not
14 as a prescribing physician, has been faced with
15 patients or oneself has had several critical
16 questions to ask when their pain recurs or doesn't
17 respond in the first place. "What do I do till the
18 next dose? Do I change medications? Do I call and
19 get a new prescription? Do I simply redose early?
20 Do I take another drug concomitantly with unknown
21 synergy or safety concerns?"
22 The reality is that there is no ideal dose
23 interval in our current world, but the goal is to
24 optimally characterize, particularly as I will be
25 speaking of duration of drug effect, and have that
30
1 in labeling and be sure that that is not associated
2 with toxicity that is unacceptable.
3 [Slide]
4 So, the question is how, in the real
5 world, do we generate dose interval instructions?
6 I will be using dose interval and dose duration
7 somewhat interchangeably. The first step in drug
8 development is pharmacokinetics and I will turn
9 this over to Dr. Bashaw.
10 DR. BASHAW: I would like to thank the
11 previous speakers, both Dr. Goldkind for the
12 introduction and Dr. Witter, for their fine
13 presentations, and also the fact that most of what
14 I am going to speak of today, the groundwork has
15 been laid yesterday in our discussions about
16 chronic pain and pain metrics.
17 For the most part, as has been talked
18 about already, PK/PD and analgesic response has
19 been primarily geared towards onset. The dental
20 pain model is certainly very good for that. As you
21 go from no pain to almost instantaneous pain very
22 quickly it is very reproducible for all those
23 factors we have talked about. But there are some
24 problems with its duration because eventually pain
25 does resolve in that model in a very short period
31
1 of time relative to most chronic pain.
2 During my presentation I am going to
3 briefly go over some data from a dental pain trial
4 as it relates to onset and dose optimization, and
5 contrast it to where we are going with chronic pain
6 and also with duration metrics. However, because
7 it is still early in the morning, or relatively
8 early in the morning, I promise I will not take you
9 through any model derivations or any model
10 simulations because that is way beyond the scope of
11 the time of the talk this morning.
12 [Slide]
13 As I said, we basically have very good
14 single-dose metrics looking at blood level onset
15 and pain relief. One can pretty much look at a
16 successful development of many OTC analgesics and
17 even prescription analgesics and see that we do
18 have a very good handle on onset, and the next step
19 is where do we go from there when we need a second
20 dose, and how we get from it.
21 [Slide]
22 This is what one typically sees. In this
23 particular case we have a dental pain trial where
24 we are comparing three different doses of a
25 nonsteroidal. Here we have what is calculated to
32
1 be a no effect dose; what was assessed to be a
2 mid-range dose; and what was expected to be an
3 antirheumatic dose but was put in the trial just to
4 see what the performance would be for a new
5 analgesic.
6 You can see this is where we would start
7 off with pharmacokinetic data, concentration versus
8 time. From this type of material one can get the
9 standard pharmacokinetic analysis of varying the
10 curve, Cmax, Tmax and those parameters which we
11 normally work with.
12 In terms of making the next step, linking
13 this to some kind of effect, analgesia being
14 duration or whatever we are looking for, one has to
15 make the next step as to how one combines this
16 information with the dynamic response.
17 [Slide]
18 This is one representation I have. I
19 tried to make it as simple as possible. Basically,
20 what our theory is, is that we pretty much have
21 optimized input rate. Input rate gets into the
22 blood, gets into the plasma and then we have drug
23 migrating into some effect site concentration that
24 then exercises the effect.
25 The dynamic compartment is a theoretical
33
1 compartment. We tend to draw it as a separate box
2 but in reality the effect site is subsumed within
3 the central compartment within the blood and within
4 the plasma. But for modeling purposes it is much
5 easier to have this over here because it explains
6 some of the things we see with the drug onset in
7 terms of lag time, in terms of dose response
8 issues.
9 Primarily what one needs to just remember
10 from this slide is that effect site concentrations
11 is what we are really trying to look at. However,
12 we can't measure them directly. We can measure
13 plasma blood levels, but we cannot measure the
14 concentrations at the effect site. These are all
15 theoretical and based on our simulations. However,
16 we do know that the rate constant, if you model it
17 this way, the Keo value, is equilibration between
18 these two compartments. It is what is going to
19 drive duration. It is what is going to drive the
20 redosing issue because it is going to control time
21 to accumulation at the effect site; time to onset;
22 and also time for levels to go back in the plasma.
23 So, that is really what we are trying to look at in
24 terms of driving this situation.
25 [Slide]
34
1 Here is what we normally see. Again, we
2 are taking our dental pain example. We have taken
3 our concentrations and now plotted them against a
4 dynamic effect. In this particular situation this
5 PID score and placebo are corrected. Here is our
6 no effect dose, some effect but not very much.
7 Here is our mid-range dose which is getting a PID
8 at maximum of about 1. Here is our antirheumatic
9 dose which is getting up there but there is some
10 lag time here.
11 This pretty much shows one of the problems
12 you have when you try to direct correlations
13 between concentrations and effect. You can see,
14 for example right here with the mid-range dose,
15 that we have concentrations of approximately 5
16 ng/ml and you get a PID change of only 0.2. Yet,
17 up here at 6 hours you have the same drug, same
18 dose and the same concentration but it has a PID
19 change of 1.
20 What is going on there? How can you have
21 the same concentration giving two different
22 responses? Part of that is due to the fact, again,
23 of the model. It is 6 hours in the dental pain
24 model and pain is starting to resolve. So, even
25 though your concentrations have dropped you are
35
1 seeing resolution of their pain relief because of
2 other factors, which again shows the limitations of
3 this model.
4 [Slide]
5 One of the things we do with this data in
6 trying to develop a relationship is we try to
7 collapse these responses. We call these hysteresis
8 loops because or their curvolinear nature. This
9 particular nonsteroidal is very typical of what you
10 see, counter-clockwise hysteresis, as one sees
11 here. This is basically due to one of three
12 reasons: There is a significant time lag between
13 drug entering the central compartment and going out
14 to the theoretical effect site. Possibly also it
15 would act on the metabolite if you were just
16 following the parent and the activity is due to the
17 metabolite. That is also going to give you a
18 disconnect which is going to result in
19 counter-clockwise hysteresis.
20 And, important for a situation with
21 nonsteroidals, it is due to the fact that we are
22 not having a direct effect here but a secondary
23 effect due to the effects of arachidonic acid.
24 Nonsteroidals, unlike opiates which work on mu
25 opioid receptors, kappa receptors, etc. and have a
36
1 direct pain activity, nonsteroidals, of course,
2 have to work through the arachidonic acid cascade
3 and that is going to cause a lag time because it
4 takes time first to use up those factors that have
5 already been formed, and then when the drug wears
6 off it takes time for the cascade to reestablish
7 itself. This also results in that disconnect
8 between concentration and effect, which is one of
9 the problems we have in modeling this data.
10 [Slide]
11 But if one continues on with the same
12 dental pain trial and you collapse the loops, this
13 is what you can derive. You can derive a
14 relationship, shown in this particular case using
15 an Emax model, and you can make a response between
16 dose and effect. You do see noise out here and
17 this, again, is due to the duration issues. But
18 one can see in this particular case that we do have
19 effect of concentration. There is an Emax of about
20 1.2 PID units, which is about what you are going to
21 see for maximum effect.
22 From a response like this, one could then
23 go back and look at your doses, look at your dosage
24 form and pick a dosage that would give you the
25 efficacy you want, depending on how you define it.
37
1 Once you have a PK/PD relationship, you can look
2 back and say you want to have a certain duration, a
3 time above a certain EC50 or EC75. If you want
4 what Dr. Witter was talking about, a 90 percent
5 change or 75 percent change depending on what
6 metric you are using, if you are using a quality of
7 life metric or if you are using PID scores, or
8 whatever, it is very analogous to how you go back
9 and do this and look at time above for duration.
10 These are analogous to what is done in the
11 surgical area where you use neuromuscular blockade
12 and you have a train of 4 measurements, where you
13 are looking at a pharmacological response in terms
14 of muscle blockade and you must calculate your
15 duration based on how long you want to have
16 neuromuscular blockade, and a train of 4 is a way
17 of doing it. It is very analogous to trying to
18 look at duration of action issues with analgesia,
19 except that we don't have as well defined a metric
20 or observation.
21 [Slide]
22 As I said before, one of the primary
23 reasons you have counter-clockwise hysteresis is,
24 of course, the fact that one has this cascade of
25 pro-inflammatory precursors and pro-pain precursors
38
1 that have to be used up in the modeling. The time
2 it takes up for these precursors, both to ramp up
3 in the case of the drug wearing off and to be
4 consumed and onset, is what affects our hysteresis
5 loops. It really is the modeling problem for
6 duration.
7 For onset we have very good metrics. We
8 have shown that and pretty much we have optimized
9 drug delivery to deal with the onset. But what
10 about duration? How can we deal with that in the
11 drugs that don't have direct response?
12 [Slide]
13 We can model duration of action using
14 indirect PK/PD models that allow for downstream
15 activities. However, it requires, as I think has
16 been reiterated before, an understanding of the
17 underlying physiology; an understanding of the
18 dynamics of the response; patient factors; and does
19 require a large number of PK and PD observations
20 across a number of doses.
21 With this kind of information together,
22 understanding exactly whether or not it is, as Jim
23 pointed out this morning, moderate or severe pain
24 from a dental pain trial or from coronary-artery
25 bypass graft pain, you have to understand the
39
1 underlying physiology of the pain. You have to
2 understand the dynamics of response of the patient
3 factors and how the patients are going to perceive
4 their pain; how they are going to relate it back to
5 you in terms of its intensity or their degree of
6 pain relief. Then, from a calculational
7 standpoint, you do have to have a large number of
8 observations, both PK and PD, so that you can make
9 predictions across a number of doses.
10 [Slide]
11 What one can get from an analysis such as
12 this--this is some simulated data we worked on for
13 an intravenous analgesic and what basically one can
14 do when one has enough data. This is the
15 probability of obtaining a certain PID score over
16 time for a certain dose of the drug. You can do
17 this for many different doses. What we see here is
18 that if you are looking for a PID change of 1, we
19 have a very good onset and we have maintenance of
20 that PID score for at least an hour and a half.
21 Right there is the last observation in this trial.
22 For this trial, here, the probability of a PID
23 score of 2 is about 0.5 and then it starts dropping
24 off when you start getting out to 40, 45 minutes.
25 PID score 3 is really not going to happen here.
40
1 But using simulations, using PK/PD and
2 understanding the models one can, using indirect
3 modeling, develop probabilities using a Monte Carlo
4 simulation that can then be related back to
5 duration of effect and the maintenance of effect
6 over time. If one has enough data-- this is
7 obviously for one particular dose level--one can
8 take multiple doses, plot together and actually do
9 three-dimensional response surface mapping and look
10 at the effect of various factors, concentration,
11 effect, time, duration, etc. and decide what is an
12 optimal dose that can then be tested in clinical
13 trials.
14 [Slide]
15 Before I hand it back to Dr. Goldkind,
16 from a pharmacokinetic standpoint looking at
17 exposure response analysis, you know, with opiates,
18 because of their mechanism of action where they
19 have direct binding to receptors, we have good
20 assessments of onset and we can do pretty good work
21 with duration because it is a direct receptor
22 interaction situation. With nonsteroidals, the
23 mechanism of action being indirect and they don't
24 actually have pain relief themselves but work
25 through other mediators, through a cascade effect,
41
1 we certainly can do onset. We have actually done a
2 lot of work over the last couple of years
3 optimizing drug delivery for onset.
4 Duration is more problematical, as we have
5 said this morning. It is model dependent. It
6 requires an understanding of the physiology. It
7 requires an understanding and identification of
8 relevant patient factors. Also, it requires
9 certainly a good amount of data to work with
10 because if you don't have the data your simulations
11 and your work just won't have the power you want to
12 have to make proper dosing selections.
13 With that, I will turn it back over to Dr.
14 Goldkind.
15 DR. GOLDKIND: Thank you.
16 [Slide]
17 We now know that PK can take us so far in
18 assessing dose duration, but only so far and the
19 question is how do we add to that with clinical
20 data? I will be talking about the endpoints that
21 are used in adding value to PK data in assessing a
22 dosing interval.
23 First I would like to go through the
24 guidance that we have, both from the FDA as well as
25 from EMEA. The 1992 guidance, in the section that
42
1 does deal with metrics for assessing the duration
2 of analgesia, and I quote directly: Similar onset
3 of analgesia, there are various approaches to
4 defining the duration of analgesia. Examples
5 include from the onset of study drug or the onset
6 of analgesia until either intensity of pain returns
7 to baseline; the patient indicates that analgesic
8 effect is vanishing, which are similar; patient
9 requests rescue, and the time to rescue is
10 sometimes designated as TTR, can either be measured
11 in the mean or the median; and the percent of
12 patients who do not rescue during the specific
13 interval. You can look at the converse, the number
14 that do and the specific interval can be over a
15 longer period than you anticipate a dose interval,
16 or the dose interval you anticipate and end the
17 study at that point.
18 [Slide]
19 The European Medicines Evaluation Agency's
20 draft guidelines from 2001 state that a real effort
21 should be made to obtain data on the best dose and
22 interval regimen, time to onset of peak effect and
23 duration of effect. The endpoints that are
24 referenced a little bit further on in that document
25 refer to duration of analgesia, which isn't a
43
1 metric per se but just reiterates that that issue
2 needs to be dealt with, and time to rescue is
3 mentioned as a metric.
4 [Slide]
5 I would like to go through the different
6 metrics now and discuss them. The return to
7 baseline pain metric, I believe, is a flawed one.
8 [Slide]
9 This graph, which is taken from real data
10 but the specific drugs are not relevant, is a good
11 example and reflective of what we see in I would
12 say most curves for analgesics. The top two lines
13 are both active drugs and the lower curve is
14 placebo. As we all know, there is a substantial
15 placebo effect. There is an onset for placebo as
16 well as the active drugs. But what you see as you
17 go out is that pain relief is pretty much steady
18 going all the way out to 12 hours. Interestingly,
19 the placebo response drops a little bit but nothing
20 comes down to baseline. That is not uncommon in
21 the studies that we see.
22 [Slide]
23 As Dr. Bashaw mentioned, acute pain
24 resolves and that is just part of the model. So,
25 you really rarely get a true return to baseline in
44
1 these studies. Therefore, this metric would give
2 you a bias, extending the apparent dosing interval,
3 if we were to use a return to baseline. In
4 addition, during acute pain studies you typically
5 have repeated questioning every 15 minutes, half
6 hour, for the first short interval, and then every
7 hour after going out variable periods of time. So,
8 it is actually quite an artificial setting to
9 collect data to begin with. I would imagine that
10 as you ask patients what pain relief they have now
11 compared to one hour ago, compared to two, three
12 and four hours ago you really introduce a lot of
13 bias and there is a lot of suggestibility. So, a
14 return to baseline pain inherently is problematic.
15 In fact, I think most pharmaceutical companies
16 realize this, and this metric is rarely used in
17 drug development, although it is mentioned in the
18 guidance.
19 [Slide]
20 So, how do we generate dose interval
21 instructions in clinical trials? Well, the first
22 thing I will say is that true dose interval ranging
23 studies, meaning to test out what you would get at
24 fixed intervals, fixed doses rather than waiting
25 for a sense of rescue or "I can't wait any longer"
45
1 are actually not done. Metrics primarily come from
2 single-dose studies. There is some qualitative
3 data that I will mention briefly later that does
4 come from multiple dose studies but this is limited
5 in amount and applicability.
6 [Slide]
7 Getting back to the other possible metrics
8 from single-dose studies and, again, I want to
9 reiterate that what these metrics describe are
10 rescue, not optimal. Percent of patients who
11 rescue during a study period is largely affected by
12 the study design and the study execution.
13 What I mean by that in study design is
14 quite fundamental. If you have a study that is
15 explained to an investigator and a patient as a
16 12-hour study, let's say, and you tell them that if
17 they need rescue to let you know, as they approach
18 that 12-hour period they may well see the 12-hour
19 mark as a threshold, as a success point, and simply
20 hold out to ask for remedication. If it is a
21 24-hour period, that will affect how it is
22 perceived. Likewise, a short study interval--if
23 somebody knows that the study is going to be over
24 in four hours, they may wait to that point.
25 Actually, the last hourly acute pain
46
1 measurement is kind of a flip side of the study
2 duration. In most studies you will have hourly
3 pain measurements up to a period of, let's say, 12
4 hours and then there will be a final pain
5 measurement session at 24 hours if the study is
6 designed that way, if the thought is that possibly
7 it is a 24-hour drug. If it is a much shorter
8 acting drug the last measurement may be at 12
9 hours, with a gap of these hourly measurements.
10 There are expectations that are
11 transmitted to the patients through the very trial
12 design that affects their behavior. We have
13 actually seen this in studies, particularly the
14 shorter intervals. A study that has hourly
15 metrics, going out to four hours, with a follow-up
16 later on, has a tremendous rescue rate right after
17 that fourth hourly measurement. It is very
18 profound when you see how the study design affects
19 the patient responses.
20 In terms of the execution, simply the
21 monitor behavior and how encouraging or
22 discouraging the monitors are of rescue, whether it
23 is called remedication or rescue, the very presence
24 of a monitor--does the monitor walk around if there
25 is more than one patient in the center? Do they
47
1 leave the room? Is the medication left on the
2 table to take truly ad lib or do you have to come
3 up and ask the monitor that may look like Nurse
4 Ratchet or may look like an inviting personality?
5 [Slide]
6 The time to rescue varies also depending
7 on the setting. Major surgery is different than
8 minor surgery; is different than dysmenorrhea. I
9 will actually show some case examples of this in a
10 little bit. Whether you are measuring the time
11 from the dose or the time from the onset of relief
12 obviously changes the metric.
13 The statistic you use, whether you use the
14 median or the mean--the median is obviously less
15 susceptible to outliers and the mean will shift
16 responses towards the shorter interval based on
17 patients who simply don't respond to the analgesic
18 to begin with.
19 [Slide]
20 I will be talking about this population
21 for analysis a little bit more. Let me define
22 things better so I don't confuse what I mean by
23 responder and responder analysis that will be
24 discussed later.
25 If you use the all-treated population to
48
1 analyze a dosing interval, then you are including
2 patients who either rescued within an hour and who
3 didn't rescue at all. This usually shifts the
4 dosing interval towards the shorter time period,
5 particularly in models of severe pain where there
6 is a high rescue rate. So, we could call that
7 either the all-treated group which does, as I say,
8 include people who had no response; we could call
9 it the ITT population.
10 The responders that I am referring to are
11 those subjects who register some form of pain
12 relief early on in the study, and there is
13 variability, in fact, at that point as well. You
14 can be defining a responder as somebody who had
15 analgesia and, therefore, they are a valid subject
16 to capture information on how long that analgesia
17 that they obtained lasted. You could do it by time
18 to onset of relief, and that can be broken down
19 into either perceptible, meaningful, adequate or
20 some prespecified either VAS or categorical
21 improvement. So, you may want to say a patient
22 doesn't really enter the analysis of duration of
23 their drug effect if that drug effect didn't at
24 least meet some minimal level. It could either be
25 subjective or you can try and objectify it with,
49
1 let's say, a pain relief score of at least 1 or 1.5
2 on a scale of 4.
3 [Slide]
4 As I mentioned earlier, there is
5 variability based on the clinical setting. What we
6 have seen is not surprising. The percent of
7 patients who rescue is highest in general surgery
8 settings, whether it is orthopedic or gynecologic.
9 Dental rescue rates tend to fall below surgery.
10 Dysmenorrhea rates are very frequently very low,
11 regardless of whether you are looking at 12 or 24
12 hours and almost regardless of the drug or placebo,
13 and we will see that. The median time to rescue
14 medication which in a sense is derived from the
15 same database as the percent who rescue, obviously,
16 then has the converse. Dysmenorrhea studies have
17 the longest dosing interval based on time to
18 remedication; dental, a little shorter; and
19 surgery, shorter yet.
20 [Slide]
21 In summary, there is a lot of variability
22 in the metrics that we use. At this point in time
23 they are not well standardized. So, we see
24 different analyses presented by different sponsors.
25 The study design, the study conduct, which
50
1 statistic is used, what population is analyzed, the
2 definition of relief, the setting and, actually I
3 didn't discuss this earlier but I put it in the
4 summary, even from trial to trial in the same
5 model, roughly same study design has variability,
6 as you would expect in nature.
7 [Slide]
8 Now I am going to go through some case
9 studies. The first ones will deal with this issue
10 of the population that is included for analysis.
11 The stopwatch technique is very frequently used.
12 What that means is that it can be either a single
13 or a double stopwatch. The patient is given a
14 stopwatch and when they feel that they have gotten
15 perceptible, meaningful, adequate relief, they
16 click that stopwatch. A two stopwatch technique
17 attempts to differentiate perceptible from
18 meaningful. So, the first stopwatch click is "I
19 feel something is happening" but it may not be very
20 meaningful for them. The second one is when "gee,
21 this is significant for me."
22 [Slide]
23 In this dental pain study, median time to
24 remedication and, again, the drug isn't really
25 relevant but the half-life is worth noting because
51
1 we will talk later about how much there is or there
2 is not correlation between PK and clinical results.
3 Placebo I will call zero half-life. We could
4 debate that. This is the all-comers or the ITT
5 analysis. You can see that placebo has almost a
6 2.5- hour median time to remedication. A 2-hour
7 drug has a 6-hour median time to remedication; and
8 a 17-18-hour drug has a 9.5-hour median time.
9 When you only look at those who responded,
10 based on the perceptible definition of response,
11 you see that this stretches out. If you were to
12 base a dosing interval instruction for a label on
13 these data, you would have to ask yourself do I go
14 with just onset, those who had onset? Just the
15 ITT? Some kind of a gestalt approach between the
16 two?
17 [Slide]
18 I am just going to show a slide
19 demonstrating variability from study to study even
20 in the same model. There is a second dental pain
21 study added to this slide. Within study 1 and
22 study 2, which really were conducted identically,
23 there is some difference that you see in the two
24 studies. Is that tremendous? Is it surprising?
25 No, that is variability that you see, but if you
52
1 were interested in drug Y, you wouldn't really know
2 whether to push this to Q8 hours. Should this go
3 to Q8 hours? Should it go to Q12 hours? Then, if
4 you are guided by the analysis of only those with
5 onset, do you go to 12 to come up with some kind of
6 a combo here, or do you go to the Q24-hour
7 interval? I think that we would all agree that it
8 is kind of difficult to know from these data what
9 is the ideal dosing interval. For drug X, it is a
10 2-hour half-life. Is it a Q4-, 6- or 8-hour? For
11 drug Y, is it Q8, Q12, Q24?
12 [Slide]
13 In summary, for dental pain studies we see
14 that there is an effect of the population you are
15 using for analysis. There is a limited
16 relationship between PK and clinical data. The
17 time to rescue and the percent who rescue within an
18 interval are informative but not definitive. Then
19 the question that, in a sense, we are asking
20 ourselves, asking the committee for input, is would
21 there be benefit in studying a multi-dose in the
22 sense of at least a minimum of a second dose where
23 you actually look at a fixed dosing interval to get
24 an idea of whether, beyond the placebo effect,
25 there actually is a pharmacodynamic effect of an
53
1 earlier dose compared to a longer dose that may be
2 chosen based on convenience and perception of
3 safety?
4 [Slide]
5 We will look briefly at dysmenorrhea. As
6 I mentioned earlier, these are two studies. This
7 is a 12-hour drug Z and a 17 to 18-hour drug Y. As
8 you can see, the median time to remedication is
9 very long even in placebo. The percent who rescue,
10 and this was within 12 hours, you can see is quite
11 low. Obviously, the greater than 24-hour median
12 tells you that at 24 hours it remains very low.
13 What this slide tells us is that
14 dysmenorrhea is not generalizable to other
15 settings. I don't think we would want to apply
16 these data to the label in a generic way. And, it
17 tells us that dosing interval for dysmenorrhea is
18 not going to be well guided by this.
19 [Slide]
20 Just looking briefly at postoperative
21 models, and this is an orthopedic study begun day
22 after surgery or when the patients came off patient
23 controlled analgesia and when they reached a
24 certain VAS of pain, I believe it was the
25 threshold when patients where entered into the
54
1 study.
2 We have placebo, drug Z 12-hour half-life,
3 drug Y 17 to 18-hour half-life. I only have the
4 ITT population analysis for this study but you can
5 see it is very short. It doesn't even resemble the
6 other two models. The percent who rescue in 12
7 hours is extremely high in all groups. Again, if
8 you were going to use this model to generalize to
9 dysmenorrhea and dental, it would be problematic.
10 We do see this across studies and across other
11 major surgery models. Do we need a totally
12 separate dosing structure for postop pain? Is drug
13 Z a Q4 hour drug? Is it a Q6 hour drug? Is Y a Q4
14 or Q6?
15 [Slide]
16 As I mentioned, the surgical setting is
17 quite different than the dental and dysmenorrhea.
18 The question is how do we establish dose interval
19 for postoperative pain and, again, if drugs Y or Z
20 can't be safely given during that shorter interval,
21 what do we do? Do we contraindicate it? Do we
22 indicate it for postop pain but in conjunction with
23 a rescue medication that should be available
24 because we know that the interval will be short?
25 [Slide]
55
1 Now I will just briefly talk about the
2 qualitative data we get for multi-dose studies to
3 add to the single-dose study metrics I have
4 discussed. Use of supplemental or rescue
5 medication over a period of time is frequently
6 collected. Patient global evaluation over
7 subsequent days is frequently collected, as is
8 average pain intensity scores over a period. These
9 endpoints generally are not really sensitive and
10 informative enough to give us information on a
11 dosing interval.
12 [Slide]
13 Let's not forget risk/benefit. We could
14 say take the drug every hour but that will have its
15 problems. We are reminded of this in this "B.C."
16 cartoon, "What's the strongest over-the-counter
17 pain killer you got?" And, the answer is a mallet
18 over the head. Is it effective? Yes. Is there
19 going to be remedication at all? Probably no. But
20 is this the ideal analgesic? Obviously not.
21 [Slide]
22 We need to balance safety and efficacy,
23 and that is an issue that we need to directly
24 address in labeling. Obviously, you want
25 convenience. You want adequate pain relief,
56
1 optimal pain relief, but you have to balance safety
2 and metrics, whiich particularly in the acute pain
3 setting, for safety are usually not very
4 informative. If you have a drug that has a very
5 high toxicity during a short-term period, you don't
6 have a drug. So, it is hard before marketing to
7 really know how that will play out. If you make a
8 drug a BID instead of once a day, you are not going
9 to see in that safety database, even if you collect
10 it for a week, substantial differences that you may
11 see in safety after it is marketed. Increasing the
12 dose may well increase efficacy but it also
13 increases adverse effects.
14 [Slide]
15 I am just going to discuss a case study of
16 attempts in labeling to optimize that information
17 on risk and benefit. It is the tramadol label. In
18 the clinical trial section it states that Ultram
19 has been given in single doses of 50 mg, 100 mg,
20 150 mg and 200 mg in patients with pain. In the
21 dosage and administration section it states that
22 for patients with moderate to moderately severe
23 pain, not requiring rapid onset of analgesic
24 effect, the tolerability of Ultram can be improved
25 with the following titration schedule, and it goes
57
1 on describing a titration schedule that has been
2 studied, and describing in some detail the extent
3 to which it spared some toxicities.
4 [Slide]
5 A little bit later in the dosage and
6 administration section it states that for the
7 subset of patients for whom rapid onset of
8 analgesic effect is required and for whom the
9 benefits outweigh the risks of discontinuation due
10 to adverse events associated with the higher
11 initial doses, Ultram 50-100 mg can be administered
12 as needed for pain relief every 4-6 hours. There
13 is a statement that clearly says not to exceed 400
14 mg per day.
15 [Slide]
16 So, we have a label that really attempts
17 to put in all the different metrics and information
18 available, and it really is a juggling act for the
19 prescribing physician. This is an example,
20 frankly, of what you would need to try to cull from
21 any label. You need to ask yourself what is the
22 best starting dose for my patient? Shall I give
23 them a loading dose that is high, or are they going
24 to tolerated it better if I start lower? What
25 timing interval should I give them? That, to an
58
1 extent, is left to patients. There is nothing
2 wrong in saying take it every 4-6 hours depending
3 on how you need it. But then you have to deal with
4 the maximum dose over a 24-hour period. You have
5 kind of taken from Peter to pay Paul. If you want
6 a high dose in the beginning you are going to have
7 to lower it later. Of course, there is titration
8 of dose which is frequently an issue with opioids
9 particularly.
10 [Slide]
11 In conclusion, the duration of analgesia
12 is guided by PK data. The return to baseline pain
13 metric is not an adequate endpoint to assess dose
14 interval. The clinical setting affects the
15 apparent duration of analgesia and remedication
16 use.
17 [Slide]
18 The analysis of time to remedication is
19 dependent on what population you are analyzing,
20 those who have some onset versus those who are
21 enrolled in the study and may well not have onset.
22 The percent who rescue is informative, but it
23 doesn't distinctly and clearly define any optimal
24 dosing interval. The current metrics, as I have
25 described them with the limitations, are not
59
1 standardized.
2 [Slide]
3 Additional information on dosing interval
4 is needed. More formal study of dosing schedules
5 may further characterize optimal dosing intervals,
6 and different acute pain settings may need to be
7 addressed in labeling.
8 I do want to say at this point that, with
9 the second point on this slide, we are kind of
10 venturing into a new area here. We don't really
11 know what those studies will tell us if we ask for
12 them, and that is one of the questions for the
13 group this morning, to discuss how valuable such
14 studies might be. Thank you.
15 DR. FIRESTEIN: Thank you. The next
16 speaker is Lourdes Villalba, on safety databases
17 for acute analgesics.
18 Safety Databases for Acute Analgesics
19 DR. VILLALBA: I am a medical officer in
20 the Division of Anti-inflammatory Analgesics Drug
21 Products.
22 [Slide]
23 Throughout our presentations at this
24 meeting, we have tried to emphasize how important
25 it is to collect adequate data to write a label
60
1 that is informative to patients and physicians.
2 [Slide]
3 I am going to talk about the kind of
4 safety databases that we would like to see. I
5 think my talk actually was titled safety in acute
6 analgesia trials, but I need to spend some time
7 talking about chronic requirements. Actually,
8 instead of chronic, this should be long-term use.
9 [Slide]
10 We do have some guidelines. We have the
11 ICH, International Conference Harmonization
12 guidelines that were published in 1995 and refer to
13 the use of products intended for long-term in known
14 life-threatening conditions. Long-term is defined
15 as continuous or intermittent use for six months or
16 more.
17 The minimum requirements are 300-600
18 patients for 6 months, and 100 patients for a year,
19 and a total exposure of 1500 patients including
20 single-dose and short-term multiple dose studies.
21 These numbers are given as a minimum guidance, and
22 exposure should be available at clinically relevant
23 doses or doses intended for clinical use.
24 However, the same guidance has said that a
25 larger N or longer-term safety databases may be
61
1 needed. That is in the case when there are
2 specific safety concerns. For example, if during
3 drug development in preclinical studies or early
4 Phase I for some reason we may identify some
5 specific event, or we may think that some adverse
6 event may be more frequent with time and that the
7 hazard rate will increase with time, in that case
8 we may need larger and longer safety databases.
9 Or, when there is need to make risk/benefit
10 decisions such as in the case when a new drug has a
11 tiny effect size and, therefore, even if an adverse
12 event is not very frequent we need to quantitate
13 how often that happens in order to make those
14 decisions.
15 [Slide]
16 As I mentioned, the guidance says that
17 exposure should be in doses intended for clinical
18 use. However, one of the safety concerns that we
19 do have, which applies to all analgesics, is the
20 dose creep phenomenon. Dose creep is the use of
21 medications at doses above the recommended dose.
22 That means doses above the demonstrated doses that
23 are effective and safe in clinical trials.
24 We do have an example of the dose creep
25 phenomenon from the Celebrex NDA. In the
62
1 randomized controlled trials part of the NDA,
2 Celebrex showed efficacy in osteoarthritis at the
3 100 mg BID dose and efficacy in rheumatoid
4 arthritis at the 200 BID dose. There was no
5 obvious efficacy advantage of higher doses of 200
6 mg and 400 mg respectively. Of course, they were
7 also efficacious but there was no major advantage
8 of higher doses.
9 In the open-label part of the development
10 program patients were allowed to increase the dose
11 up to 200 mg BID in the osteoarthritis study and
12 400 mg BID for the rheumatoid arthritis patients.
13 Actually, it was shown that most patients, 70
14 percent of the patients increased the dose and most
15 of them moved to a dose twice as high as the
16 initial dose even though there was no evidence of
17 worsening efficacy right before they increased the
18 dose and there was no evidence of improvement in
19 efficacy after they increased the dose. So, this
20 is just an example and the good news is that there
21 were no major safety concerns observed with these
22 increases in dose.
23 [Slide]
24 Therefore, out of a summary regarding
25 exposure requirements for long-term use, more than
63
1 fulfilling a minimum number, what we want to see is
2 an adequate safety database that will address
3 specific issues that may arise during drug
4 development. We want to see minimum ICH guidelines
5 at the highest labeled dose. We also want to see
6 special populations addressed, particularly the
7 elderly and the pediatric populations.
8 [Slide]
9 Now I am going to talk about exposure
10 requirements in acute or short-term use. The
11 approach that we have had in the division for the
12 last several years is to require as much as if it
13 were intended for chronic use. The reason for this
14 approach is that we know, I think everybody is
15 aware, that drugs are used for longer than
16 approved. There is no analgesic that is going to
17 be used only once. Even if the label states that
18 the recommendation is for short-term, they are used
19 for longer term.
20 I have two examples here. One is from the
21 Vioxx database and the other is Duract, bromfenac
22 sodium.
23 [Slide]
24 This slide was presented at the advisory
25 committee meeting in February of last year so it is
64
1 a little outdated but it makes the point. Vioxx 50
2 mg was approved for the treatment of acute pain.
3 It was recommended in the label to be used for five
4 days. This dose is twice the dose approved for
5 chronic use, the highest dose approved for chronic
6 use in osteoarthritis and twice the dose approved
7 for rheumatoid arthritis.
8 At that time, the total number of drug
9 appearances was approximately 13 million. Of
10 those, 5 percent were for the 50 mg strength. Of
11 those, one-fifth were for more than 30 days. So,
12 this is just to show you some numbers because with
13 the next example, which is actually much more
14 dramatic because of the public health issues that
15 came with it, we do not have numbers or
16 denominators.
17 We have also seen with Vioxx that there
18 are some patients who used the 50 mg dose twice a
19 day, that is, 100 mg a day. That actually is very
20 unwise, I would say, because there are very limited
21 data on the 100 mg dose in long-term exposure.
22 [Slide]
23 This is the next example. This is an
24 unfortunate example but very enlightening for us,
25 for the division and for the agency. Bromfenac was
65
1 a nonsteroidal anti-inflammatory drug approved in
2 July, 1997. There was a voluntary withdrawal in
3 June, 1998 due to reports of hepatic failure.
4 This is a very interesting example because
5 the original development program was towards acute
6 pain, dysmenorrhea and osteoarthritis and there
7 were also some rheumatoid arthritis studies. The
8 proposed dose in the original NDA was 25-50 mg
9 every 6-8 hours up to 200 mg a day.
10 At filing, it was noted that there was
11 insufficient exposure for the osteoarthritis
12 indication. Therefore, the osteoarthritis
13 indication was withdrawn but chronic safety data
14 from the chronic studies was submitted.
15 [Slide]
16 I want to show you the size of the
17 database which is actually a very good size if you
18 look at total numbers. The total exposure was
19 close to 2200, with 1000 patients exposed in the
20 acute pain studies, close to 400 patients in the
21 multiple dose, up to one week studies. There were
22 also some dysmenorrhea studies of 250 patients and
23 the chronic exposure was about 900 patients in
24 osteoarthritis and rheumatoid arthritis. So, if
25 you look at the total numbers these look very good.
66
1 [Slide]
2 However, if you break it out by dose and
3 duration of exposure--this is the dose in
4 milligrams a day and this is the duration in days
5 of exposure, you see that the number of patients
6 exposed to the 200 mg dose for a year or more were
7 only 24. The bulk of the exposure was at doses
8 below 150 mg.
9 At the safety update there were more
10 patients, and when we get to the 900 patients
11 exposed for more than three months--I do not have
12 the breakout of these numbers but it was mentioned
13 in the medical officer's review that there was
14 sufficient exposure to support the 150 mg dose a
15 day and, again, the dose was 25-50 mg up to three
16 times a day.
17 [Slide]
18 I don't want to go into details but just
19 to show you that this was a very good database in
20 the sense that there were placebo control studies,
21 active control studies up to one year with several
22 comparators. They used fixed dose, as I said,
23 25-50 mg BID, TID and four times a day but in fixed
24 dose, not in flexible dose. There was a good
25 number of patients with OA and RA, and there also
67
1 was an open-label experience up to four years and
2 that involved flexible dose, some of them up to 225
3 mg a day.
4 [Slide]
5 Therefore, the safety review showed that
6 acute pain studies that were conducted at the 50 mg
7 and 50 mg single doses, and also in short-term
8 multiple dose studies conducted with the 25 mg and
9 50 mg a day dose, showed absolutely no safety
10 concerns. There was some nausea, some vomiting, a
11 little allergic reaction but there was not even
12 mention of any liver effects.
13 [Slide]
14 However, the chronic studies showed a flag
15 for hepatotoxicity. This is what the NDA review
16 showed regarding liver function test elevations, 15
17 percent of patients had mild elevations, that is
18 less than 3 times the upper limit of normal, and
19 2.8 percent had clinically significant elevations
20 of LFTs, 3 times the upper limit of normal or
21 higher. Of note, the NSAID template mentions that
22 LFT elevations in clinical trials of NSAIDs are
23 usually seen in 15 percent of patients. Therefore,
24 the number of patients with mild elevations of LFTs
25 was nothing outstanding. The clinically
68
1 significant elevation was higher than what the
2 template says, which is 1 percent but, again, it
3 was not something terribly dramatic here. This
4 number is actually similar to what was observed in
5 the diclofenac NDA.
6 The elevation of LFT particularly
7 clinically significant events were dose related.
8 They were observed at the 100 mg dose but most of
9 the cases were at higher doses. Most of them were
10 reversible after drug discontinuation. Some of
11 them were reversible even without drug
12 discontinuation. The majority occurred within the
13 first 90 cays, but the important observation was
14 that the earliest occurred around day 30.
15 [Slide]
16 Based on those observations, the drug was
17 approved with warnings for risk of hepatic effects.
18 Short-term use for pain should be less than 10 days
19 and, because of the risk of hepatotoxicity, if
20 longer therapy is needed, LFTs should be monitored
21 after 4 weeks. So, we think it was pretty clear
22 that there was some concern with liver toxicity
23 here. In addition, the maximum daily dose would be
24 limited to 150 mg a day, and there was removal of
25 any reference to treatment of osteoarthritis,
69
1 chronic pain and dysmenorrhea.
2 [Slide]
3 Within months the agency started to
4 receive postmarketing reports of liver toxicity,
5 including hepatic failure, need for liver
6 transplantation and death. Most of the reports
7 were at doses within the labeled dose, but most of
8 them were exposed for longer than 10 days. The
9 majority was for 2-8 months, and some of them were
10 exposed for a couple of years.
11 We have this unfortunate example, but I
12 think that reflects something that everybody knows,
13 which is that drugs are used for longer than
14 initially intended. As was discussed yesterday, if
15 a drug is approved for acute use but somebody
16 thinks that it may work for chronic pain physicians
17 are going to use it.
18 [Slide]
19 In summary, short-term safety studies are
20 certainly insufficient to address safety concerns
21 that may come up with some patients who will be
22 using the drug for longer than intended.
23 Drug development for acute pain drugs
24 should address the potential safety concerns of
25 dose creep, use for longer than the intended, and
70
1 potential for abuse which is another whole issue.
2 We propose that for a short-term
3 indication, unless there is a contraindication
4 based on safety, formal efficacy studies should be
5 done in a chronic setting. I think this is the new
6 concept that we would like your opinion on. We are
7 not saying that off-label use needs to be addressed
8 for every indication because that is impossible,
9 but for a drug that belongs to a class that is used
10 for a chronic indication it is very reasonable to
11 ask for some efficacy studies. If it doesn't work,
12 if it is not efficacious in the chronic indication,
13 then we can put that in the label, that this
14 doesn't work for chronic pain; do not use it. So,
15 we think that this would be a way to address the
16 possibility of off-label use and also allow us to
17 do a better risk/benefit assessment. That is the
18 end.
19 DR. FIRESTEIN: Could I ask a quick
20 question? Do you think that that final
21 recommendation would essentially nullify
22 yesterday's discussion about having separate acute
23 and chronic indications? I mean, if for an acute
24 indication you are going to require formal chronic
25 safety and efficacy what is the value then of
71
1 having separate tracks?
2 DR. VILLALBA: Well, we are not going to
3 require replication in three different models for
4 the chronic indication. What we want is at least
5 to have some efficacy studies. For example, for a
6 new NSAID or a COX-2 inhibitor, if someone would
7 come with only the acute pain indication, then we
8 would ask for osteoarthritis studies to see if that
9 worked in the chronic setting. That would provide
10 also better safety data because safety data
11 collected in an open-label way is not the same as
12 safety data collected in a controlled way, with
13 placebo control and active control studies. But we
14 actually would like to hear your opinion. Thank
15 you.
16 DR. FIRESTEIN: Are there any other
17 comments or questions from the group?
18 DR. MAX: I have some questions regarding
19 the dosing interval. I think a lot depends upon
20 what you want to tell people about. My question is
21 has the FDA studied what percentage of patients
22 whom you are trying to inform who are taking acute
23 analgesics take two doses total versus three doses
24 or four doses? Because if you mostly want to tell
25 people about the second dose, single-dose duration
72
1 is enough. If there is a large number of people
2 who take three doses, the second dose is important,
3 and so on.
4 DR. GOLDKIND: That question will really
5 depend on what studies show the dosing interval
6 should be. There may well be off-label usage TID
7 for a BID drug, but if the best studies have
8 identified a twice a day regimen, actually PK and
9 some Phase II clinical studies should give you an
10 idea of the ball park. I mean, we don't have
11 examples of every two-day drugs or drugs that are
12 taken very infrequently. I think, as you pointed
13 out, you need to at least get data on doses going
14 out beyond the first interval that you would be
15 prescribing in terms of usage data on how many
16 patients go beyond the frequency advised. We don't
17 have that.
18 DR. MAX: Yes, my question is have you
19 studied general use of analgesics for acute pain
20 and how many people just have one day treatments or
21 one dose treatments, or two day, three day
22 treatments?
23 DR. GOLDKIND: We don't have that, no. In
24 clinical studies it is hard to get a model that
25 will get you out multiple days. So, I think that
73
1 answers the question to some extent. Most people
2 only take acute analgesics in the postoperative
3 setting or acute injury setting for several days on
4 a regular basis.
5 DR. MAX: But do you understand what I am
6 referring to?
7 DR. GOLDKIND: If I do understand, we
8 don't have usage data to tell us how many days
9 patients take acute analgesics for most
10 indications. I don't know if that is available. I
11 don't know if IMS data could give us that.
12 DR. FARRAR: As somebody who has focused
13 primarily on chronic pain as an area of study, I
14 would admit to this being the first time that I
15 have sort of seen the full scope of the approval
16 process for acute pain. I commend the FDA for
17 reexamining the entirety of the approval process
18 because I think there are a clearly a number of
19 issues that can be addressed that aren't currently
20 being addressed, some of which were being hinted at
21 by Dr. Max.
22 One of the things that strikes me is that
23 I have never, ever seen a drug that is used as a
24 single dose, ever. It may be tested that way; it
25 may be used that way perhaps in a hospital setting,
74
1 but if it is over-the-counter it just doesn't
2 exist. Therefore, I think it is probably necessary
3 to study certainly the effect of several doses over
4 a period of time. I think that that would clearly
5 generate a completely different set of data
6 perhaps.
7 The second issue that I will just raise,
8 and I am just raising all of these and I think they
9 would need discussion at length in a different
10 setting, but the second issue relates to the safety
11 data. Dr. Goldkind showed very nicely sort of the
12 need to look at risk/benefit ratios. It seems to
13 me that it doesn't make obvious sense to look for
14 safety day in use over six months and not look at
15 least in some way at efficacy data over the same
16 period in terms of just thinking about how a
17 medicine is going to be used in terms of the
18 general public.
19 What that raises is really the last point
20 that I want to make, which is that we know that
21 these drugs are going to be used in a variety of
22 different ways by different patients and different
23 physicians. And, I think it is imperative that we
24 look at the way in which the drug is going to be
25 used and use that information to guide us in terms
75
1 of both the safety and the efficacy data that we
2 would want prior to or following approval.
3 There are two points that were made in the
4 last presentation which I think really speak to
5 this. With the Celebrex example, the fact that 70
6 percent of people increased their dose when allowed
7 to do so tells you two things. It tells you, one,
8 that that is the way it is going to be used. It
9 also tells you that even though the study was not
10 large enough to show that a larger dose provided
11 better efficacy, or that there was some development
12 of--I don't want to call it tolerance but getting
13 used to the medicine, whatever you want to call
14 that, that over time an increased dose was more
15 beneficial. The patients were telling you that.
16 The patients said when given the option I will take
17 this medicine at a higher dose because it works,
18 number one and, number two, doesn't cause acute
19 side effects. That really is telling and indicates
20 that there needs to be at least some approach to
21 the concept of if given free access to the
22 medication, if it was placed at the bedside so the
23 patient can take it without asking the monitor, be
24 that person nice or not nice, then they will use it
25 in the way in which they would probably use it at
76
1 home and that would perhaps dictate the way in
2 which a study could be conducted.
3 The very last thing that I would like to
4 point out is that we need to keep in mind with all
5 of the PK data, all of the time to effect data, all
6 of the time to return to baseline although I think
7 I agree that that is a lousy measure, time to
8 remedication, those are all mean values. What a
9 mean value indicates is that there are 50 percent
10 of the people who did either better or worse. I
11 don't think that 50 percent is the number we are
12 actually targeting in terms of what a reasonable
13 dosing schedule would be. I certainly would never
14 treat my patients and allow 50 percent of them to
15 suffer for an hour or two before I gave them a
16 second dose.
17 I think that needs to be dictated very carefully by
18 the risk/benefit or the minimum amount that they
19 can take to be effective and the maximum amount
20 they can take and still be safe.
21 DR. FIRESTEIN: I think actually you are
22 referring to median, not mean. Actually, the
23 points that you raise bring us to the first point
24 of discussion. I think based on what we have heard
25 and our own clinical experience, it is reasonable
77
1 to expect not single-dose studies but at least
2 multi-dose studies involving a variety of metrics.
3 I would like to open this for discussion with
4 regard to what sorts of metrics people might feel
5 would be appropriate. Susan?
6 Discussion Points # 1, 2 and 3
7 DR. MANZI: I just wanted to make one
8 other comment first. I agree that I think the
9 purpose of clinical trials is to accurately
10 simulate clinical practice. As I was listening to
11 these talks, I said I can't even imagine where you
12 would use single-dose analgesic even in the most
13 acute situations. So, I would agree with multiple
14 dosing.
15 The only other point, and I guess this is
16 the epidemiologist's hat that I wear, is that when
17 you are looking at how to figure out dosing, you
18 really learn a lot from the outliers. It is the
19 people who extend beyond the bell curve where you
20 get the most information. My point would be that
21 if you look at time to rescue, you shouldn't
22 exclude the non-responders in that because in
23 clinical practice we can't predict who those
24 non-responders are going to be and when they are
25 going to need some additional dosing. I think most
78
1 people don't take a drug and say "it didn't work;
2 I'm not going to try it for another dose."
3 So, my point is that I would assume the
4 most narrow time based on the outliers for time to
5 redosing and test safety of that in that setting.
6 DR. FIRESTEIN: Clifford?
7 DR. WOOLF: To come back to the issue of
8 onset and duration, Dr. Witter's presentation, the
9 context of when even a single drug is given,
10 whether it is given pre- or postoperatively may
11 profoundly change both of those metrics.
12 DR. FIRESTEIN: Coming back to the
13 question of what the appropriate metrics might be,
14 a series of possibilities were raised, and I can't
15 remember in which presentation it was but is the
16 gold standard for an acute pain medication going to
17 be quality of life, or is it simply pain?
18 Somebody?
19 MS. MCBRAIR: I would go for pain relief.
20 I don't think we are worried as much in the short
21 term about the quality of life, especially for a
22 post surgical patient. They are going to be,
23 hopefully, in a hospital setting and well
24 monitored, and they need pain relief and we would
25 not hold it back from them.
79
1 DR. CUSH: I would also say that when
2 looking at the metrics you should rely upon,
3 obviously, pain is where we are going to go.
4 Unlike other diseases where our metrics are maybe
5 multivariate where we are going for a disease
6 response, here we are looking for a symptom
7 response across many different diseases, and having
8 a multivariate definition of response might be very
9 difficult to arrive at, as we discussed yesterday.
10 But if we had an acceptable measure of pain relief
11 that was universally agreed upon, we could go for
12 the variables that Jim was looking for. For
13 instance, if you defined an acceptable response of
14 50 percent, pain relief of 50 percent, you could
15 then define the time of response and the percentage
16 of patients actually receiving that response in a
17 placebo population and in an active treatment
18 population and then also maybe even define the
19 duration of response with a PR 50, or something
20 along those lines.
21 DR. FARRAR: I think the point about
22 quality of life as a measure in an acute pain
23 process brings up an important point, which is that
24 the quality of life is defined differently in
25 different circumstances. I would argue that
80
1 adequate pain relief postoperatively is, in fact, a
2 very good measure of a postoperative six-hour
3 period of quality of life.
4 But I think ultimately that pain is the
5 primary outcome. What I would like to point out
6 though is that it is not a single measure of pain
7 that is paramount. Certainly, in treating
8 postoperative patients, clinicians are aware that
9 the onset of action is vital to the control of pain
10 and you certainly would not give a medication to a
11 postop patient who is writhing in pain a drug that
12 would take two hours.
13 So, the onset of action is of extreme
14 importance, as well as the duration of action only
15 inasmuch as it dictates dosing. The duration by
16 itself--you know, a long-acting medication may well
17 be of benefit but if you have a short-acting
18 medication, as we know, in terms of intravenous use
19 of various short-acting opioids, they can be very
20 effective and the short-actedness can be overcome
21 with either an infusion or multiple dosing.
22 So, I would argue that there needs to be
23 pain measurement as a primary outcome with at least
24 two issues. One is the onset of action and then
25 the duration of action as it dictates the use of
81
1 the drug.
2 DR. KATZ: Just to continue the discussion
3 of appropriate metrics for onset, first of all, I
4 wonder if somebody could explain to me what the
5 relevance is of placebo response to measuring
6 onset? That doesn't seem to make any sense to me
7 at all. If you are lying there in bed, looking up
8 at the nurse giving you the medication, you want to
9 know how long it is going to take this thing to
10 work. You don't want to know when is the
11 pharmacodynamic of the response of this medication
12 going to separate from placebo. That is a
13 completely noon-intuitive and clinically irrelevant
14 measure. I would propose that for onset we look at
15 actually onset, when the medication starts to work
16 as opposed to when it separates from placebo.
17 The second issue I have with onset is that
18 it is not at all clear to me why we are only
19 interested in drugs that have onset within one
20 hour. There are other characteristics of onset,
21 aside from time to onset, that are also relevant.
22 For example, in an NSAID I don't know what the
23 typical rate is of responders that you see, but if
24 you see that, for example, 60 percent of your
25 patients will respond within an hour, I also might
82
1 be interested in a drug where 95 percent of
2 patients respond but it takes an hour and a half
3 and there are other ways to bridge the gap. So, I
4 am not sure why we have this rigid notion that you
5 have to meet your onset criteria, whatever that is,
6 within an hour.
7 DR. FIRESTEIN: Can you clarify your point
8 about differentiating from placebo? You don't
9 think it is important to differentiate from placebo
10 during that first hour?
11 DR. KATZ: Let's say, for example, that
12 you give your drug to a group of patients and the
13 median time to onset of the drug itself is one
14 hour. In other words, you have a clinical sense
15 that it is going to take on average an hour for
16 that medication to work. If it doesn't separate
17 from placebo for an hour and a half, what is the
18 difference?
19 DR. FIRESTEIN: Because then you could
20 just treat with placebo.
21 DR. KATZ: No, no, no, that is not true at
22 all. The confusion I think is between looking at
23 measures of efficacy of the drug compared to
24 placebo versus looking at onset compared to
25 placebo. Obviously, you have to show that your
83
1 drug is better than placebo in some way--a SPID or
2 one of your measures that has been shown to be
3 effective for that. But in terms of giving
4 clinically important information about when the
5 drug works, the clinician wants to know when the
6 drug works; he doesn't want to know when the
7 placebo works. So, whether the drug separates from
8 placebo within that hour or it takes an hour and a
9 half or two hours, or what-have-you, is a
10 completely separate question, and I don't think the
11 separation from placebo is a clinically useful
12 metric of onset. The drug works when it works.
13 The effectiveness of a drug is a combination of its
14 pharmacological effectiveness and whatever placebo
15 or non-specific effect it brings to bear, but in
16 the real world both of those issues are operative.
17 DR. FIRESTEIN: One would wonder if you
18 can't distinguish it from placebo whether or not it
19 is truly a pharmacologic effect.
20 DR. KATZ: No, no, no, that is not my
21 point at all.
22 DR. FIRESTEIN: I understand. Dr. Ashburn
23 and then Janet.
24 DR. ASHBURN: I hesitate to speak before
25 the biostatistician speaks, but I just have a
84
1 couple of issues that I wanted to point out or
2 bring to the table. First of all, I want to remind
3 folks that pain measurement in the acute pain
4 setting needs to be both at rest and with movement,
5 particularly in patients who are undergoing major
6 operations, because that has been predictive of
7 good quality of outcome.
8 The other one is onset, and in an acute
9 pain setting I would reinforce Dr. Katz's remark.
10 There is not necessarily a limit of one hour with
11 regard to meaningful analgesia in the acute pain
12 setting. There are medications that can be given
13 preoperatively that do have a longer duration of
14 effect, which is no longer relevant if you are
15 trying to use a long-lasting medication and
16 prophylax, if you will, for analgesia at the end of
17 the operation. So, a one-hour onset may not
18 necessarily be important when looking at a
19 medication still intended for acute pain use.
20 Duration of effect, depending on the route
21 of administration, may be very important. A
22 24-hour duration of effect in a patient who is
23 going to be NPO for the first hour after surgery
24 may actually be a very meaningful, important aspect
25 of a different medication.
85
1 The other one is that adverse side effects
2 tend to be overlooked with regard to blending that
3 in with safety. Adverse side effects can be very
4 important in a postoperative period. If a
5 medication has a very low incidence of nausea and
6 vomiting, for instance, that will be perceived as a
7 marked advantage over parenteral opioids which do
8 have a fairly high incidence of nausea and
9 vomiting.
10 Of course, safety is paramount in these
11 areas because one would tend to not tolerate a
12 medication that even has a fairly low incidence of
13 a catastrophic event. A medication that is
14 relatively safe, that doesn't have opioid-induced
15 risk of respiratory depression may actually have
16 marked advantage even if it is equally as good as
17 an opioid analgesic.
18 DR. FIRESTEIN: Excellent points. Dr.
19 Elashoff?
20 DR. ELASHOFF: I wanted to comment on the
21 issue of what was being called separation from
22 placebo, which I assume means statistically
23 significant separation from placebo, which is a
24 combination of whatever the true separation is and
25 the sample size that you used to look at the issue.
86
1 So, the whole issue of when they get far enough
2 apart is both the issue of a clinically meaningful
3 separation and the issue of whether the study is
4 actually big enough to address that question.
5 DR. FIRESTEIN: Thank you. I always enjoy
6 being chastened by the biostatisticians! Yes?
7 DR. KATONA: Just looking at the world
8 from the pediatric point of view, even in other
9 situations we do not like to do placebo-controlled
10 trials. I am just wondering, in the acute pain
11 situations, especially the postop pain, in special
12 circumstances like with the children and the
13 elderly, is that something that we need to compare,
14 the active drugs with placebo, or could we do some
15 other designs? I personally even wonder about the
16 general population, if we could design these
17 studies as comparison studies or some other ways.
18 DR. WOOD: Gary, I wanted to return to the
19 point that you were raising right at the beginning
20 of this discussion, and that is how long do we need
21 safety data for, and how will that duration of
22 safety data affect the potential for indications.
23 It seems to me that we have excellent
24 data, going back to the question Mitch was asking,
25 to say that labeling changes are not very effective
87
1 and are generally not followed. I mean, if we
2 think of the example of fen-fen, the example of
3 truplidazone, or the example of even Accutane,
4 which has extraordinarily rigid labeling,
5 physicians and/or patients are still not following
6 these. Certainly with truplidazone the liver
7 function tests were ratcheted down week by week and
8 with relatively little effect.
9 So, the lesson from all of these, it seems
10 to me, is that even a drug that was approved
11 exclusively for acute use, such as one that was
12 limited eventually to ten days' use in the example
13 that was shown, was used for much longer than that.
14 So, common sense would dictate that we should have
15 safety data that extends for a much longer period
16 than just a single dose.
17 If that is the case, you have to then say,
18 well, how are you going to get that safety data?
19 You could give patients or volunteers an analgesic
20 for a long time for no indication which would seem
21 to me to be dubious ethics and you are probably
22 unlikely to get lots of volunteers. So, it seems
23 almost inevitable, therefore, that if you are going
24 to look for safety data that goes longer than the
25 acute setting, you are going to insist de facto
88
1 that you look at chronic pain relief even for a
2 drug that you might initially be looking at for
3 only the acute setting.
4 I don't see a way around that, and you
5 sort of touched on that in your question but I
6 think we need to return to that because that
7 actually is pivotal to how we think about this
8 whole issue of development, perhaps not labeling
9 but certainly how you develop it. If you are
10 unable to go forward without chronic studies, then
11 that is important to think about in terms of how
12 you pitch your development program.
13 DR. FIRESTEIN: Would you require
14 efficacy?
15 DR. WOOD: I would.
16 DR. FIRESTEIN: For the acute indication?
17 If you propose that you would look for efficacy
18 endpoints simply as a safety study, would you
19 require efficacy in the chronic study in order to
20 have approval for an acute indication?
21 DR. WOOD: Well, let me rephrase the
22 question, if I may. I don't think the question is
23 would I require efficacy data in the chronic safety
24 study necessarily. I think it is improbable that a
25 company or that you would advise a company to not
89
1 do an efficacy study if they were collecting
2 chronic data because, otherwise, you would be doing
3 a study in which you are giving an analgesic to
4 somebody chronically for no very obvious reason,
5 and I think it would be tough to get volunteers for
6 that, frankly. Therefore, for relatively little
7 additional cost you could get the efficacy data. I
8 think most people would do that.
9 If someone came to you and said we don't
10 want to do that, you would almost wonder why. I
11 mean, is the reason that they don't want to do that
12 because they have data that suggests it doesn't
13 work chronically or it is toxic chronically? As a
14 regulator, it would make me very uncomfortable if
15 someone was adamant that they didn't want to do an
16 efficacy study chronically when you were telling
17 them they had to collect safety data chronically.
18 DR. SHERRER: I think that goes back to
19 one of the original questions for why we came, and
20 that is should we really then be dividing into
21 acute and chronic pain? Because if we say that we
22 are going to give these drugs for acute and chronic
23 pain, in a sense we are saying that they work for
24 both. Maybe the dosing is different but, in fact,
25 the drugs work for both acute and chronic pain. In
90
1 practice that is really what is happening. So,
2 does that go back into the mechanistic differences
3 again, and are we really back to saying well, pain
4 is pain? You know, we treat one way for acute and
5 a different way for chronic.
6 DR. WOOD: Well, I think my point is a
7 little more than that. I think that even if we
8 could divide it into acute and chronic pain, and
9 even if we really thought that that would be a good
10 division to make--and I am not arguing for or
11 against that--de facto, we have come to recognize
12 that physicians and their patients are relatively
13 poor at following that advice. And, it is not just
14 true of pain; it is true of lots of other drugs.
15 You know, fen-fen was taken for much longer than it
16 was supposed to be. Truplidazone was taken without
17 the appropriate liver function tests being done.
18 Dosage creeps occurred with other drugs.
19 That is not a criticism; that is the
20 reality of the marketplace. That being the case,
21 it seems to me foolhardy to say that we are going
22 to ignore all that data and say if a drug comes in
23 only for acute pain we are not going to require a
24 safety database that goes beyond that, even if we
25 could make recommendations about how it should be
91
1 used and hope that it would be used in that way.
2 DR. FIRESTEIN: Dr. Max and then Dr.
3 Farrar.
4 DR. MAX: I would like to comment on the
5 metrics in the multi-dose studies. I think now the
6 standard metric in looking at doses past the first
7 dose is the choice of the patients when to rescue.
8 I see nothing wrong with that because you are
9 really using that just to tell patients when to
10 expect to do that. The problem is this, I have
11 spent many horrible afternoons sitting with drug
12 companies, trying to massage a bunch of repeated
13 dose data into some meaningful information and you
14 can't get anything out of it generally because
15 there are PRN doses with one regimen. The beauty
16 of dose response studies is that you make the dose
17 regimen the independent variable, and when you have
18 the dose also be the dependent variable you muck it
19 up completely.
20 So, I heartily endorse what I hear in your
21 talks. Should we use the dose response type
22 regimen and take multiple different regimens,
23 either doses or times, and try to stick to it and
24 use some other drug for rescue and find out what is
25 too high, what is too low, and what is just right
92
1 for Goldilocks? That is the way to go about it.
2 There is one other finer point, and I
3 think you have to define whether your main
4 orientation is towards exploring the clinical
5 pharmacology or usage study. That gets to the
6 issue of whether you include placebos. Say you
7 want to compare a six-time a day regimen of the
8 same drug with three-time a day, there are some
9 studies I have seen where they give placebos
10 intermittently and then people say, well, the
11 placebos gave analgesia and you really can't count
12 them. It may be that if you really want to mimic
13 usage, you want to do it unblinded so you get the
14 full impact of the placebo effect of taking extra
15 pills. But I think you need to spell this out so
16 sponsors won't go ahead and use placebos or not use
17 placebos and have the study be voided.
18 DR. FARRAR: I would like to pick up on
19 something that Mitchell just finished with and get
20 back to something that was said before. There are
21 designs that are possible and completely valid to
22 look at the way in which patients use medications.
23 Two of them that are specific, one of which our
24 group has suggested to some drug companies in terms
25 of ways to look at long-term use but have not been
93
1 adopted.
2 The first one is in terms of the onset of
3 effect and the efficacy, and that has to do with
4 whether a patient at the end of the pharmacologic
5 time period where they should have their maximal
6 effect, whether or not they decide they need
7 something else to treat that pain. That is very
8 clinically oriented and it is a valid measure of
9 whether the drug is ever effective.
10 The second thing has to do with long-term
11 use. I think it was suggested before that giving
12 patients drugs for a long period of time with no
13 indication is a problem. What I would like to
14 suggest is that one possible mechanism for dealing
15 with that is, in fact, to do a very tight and
16 carefully controlled study for a period of 4, 6, 8,
17 10, 12 weeks, whatever seems to be appropriate for
18 the drug. In the long-term study it is possible
19 simply to continue to give patients the medication
20 as long as they want to take it. That sounds a bit
21 odd perhaps, but ultimately what we are asking is
22 how are patients going to use that, and is the drug
23 safe for the period of time that they use it? If
24 you want to study it long term, as in a safety
25 study, you would give them the medication; follow
94
1 them as long as they are willing to take it,
2 meaning if it still helps them, they claim it helps
3 them for whatever reason; and look at the safety
4 data over that period of time.
5 There is actually a more elegant way to do
6 that which would in fact, be to continue to give
7 the patients the medication in a blinded fashion
8 long term. One of the arguments against that has
9 been how can you possibly give somebody a placebo
10 over the long term? My argument is to reverse that
11 and to say if the placebo is providing real relief
12 for the patient, then why not give it long term?
13 One of the ways of knowing whether a drug,
14 in fact, works better than the placebo long term
15 would be simply to give it blinded for a long time
16 and follow, as was suggested yesterday, the number
17 of dropouts.
18 DR. WOOD: But how would that differ from
19 a placebo-controlled, long-term study? I mean,
20 giving a placebo and an active drug for long term
21 in a blinded fashion sounds to me like a
22 placebo-controlled, randomized, controlled trial,
23 which is what I am saying we need to do.
24 DR. FARRAR: Right, it is. The difference
25 is the following, which is that in most of our
95
1 placebo-controlled trials there is a monitor that
2 calls you every day and says, "have you used the
3 drug? Did you write in your diary? Did you use
4 your electronic diary?" What I am suggesting is
5 that over a brief period of time, 4, 6, 8, 12
6 weeks, whatever is decided, that is reasonable.
7 But what you want to then study is the
8 actual use of the medicines. So, what you want to
9 do is to give them the medicine for, let's say, two
10 weeks or a month, a month's supply and have them
11 come back to visit you, and nobody calling them in
12 between and finding out whether they took it or
13 not; whether they filled out their diary. The
14 issue is you use simply the continued use of that
15 medicine and metrics that you measure once a month
16 to determine whether or not they actually used it.
17 There is very clear evidence, as I think
18 was suggested earlier, that if the monitor is
19 somebody who makes you feel like you want to do
20 what is right, or scares you into doing "what's
21 right" you may use the medicine in a way that is
22 very different than the chronic, normal use of that
23 medicine.
24 DR. FIRESTEIN: Dr. Strand?
25 DR. STRAND: I just want to comment that
96
1 that is a rather standard design in, say,
2 rheumatoid arthritis trials, and that is that
3 patients are allowed to continue if they have had a
4 response, open-label treatment for continued safety
5 analysis.
6 But another thing that we have also done
7 with placebo-controlled trials is that the
8 responders, not unblinded, are allowed to continue
9 treatment and that treatment is maintained blinded.
10 We have actually had patients take placebo for as
11 long as three years who respond clinically.
12 DR. CUSH: The limitations of that are as
13 far as recruitment. I mean, I tell patients up
14 front that you may be on placebo for three years
15 and that is somewhat of a deterrent.
16 DR. STRAND: I think we say not that but
17 that on or after a certain period of time, if you
18 are not responding, you are allowed to go to active
19 treatment. Then, all responders can go on to
20 continued treatment and that way we don't imply
21 that they will be on placebo for a long period of
22 time.
23 DR. FIRESTEIN: Dr. Woolf?
24 DR. WOOLF: I would like to come to the
25 issue of dose creep and the relevance of that for
97
1 the primary outcome measure, which I think we have
2 all agreed should be pain. But I think the fact
3 that patients tend to take higher doses than have
4 been demonstrated to be effective might be a
5 reflection of the fact that our measurements of
6 what is effective are insensitive, and that
7 patients may be getting a greater benefit than we
8 can actually detect.
9 So, while primary pain outcome measures
10 clearly are appropriate, there may be other aspects
11 of the treatment that are making the patient feel
12 better in a way that we are not detecting.
13 DR. FIRESTEIN: Yes, Dr. Brandt?
14 DR. BRANDT: Fundamentally I agree with
15 what is being said about long-term placebo studies.
16 But, as Vibeke said, there are practical problems
17 with IRBs that are very significant in being able
18 to do this.
19 DR. STRAND: It is not that they were told
20 that they had to be on placebo; it is that everyone
21 was offered to drop out for documented lack of
22 efficacy, and only those people who responded
23 stayed in and, therefore, we selected for a small
24 group of patients who were placebo responders.
25 I would say part of any of these designs
98
1 would be the same thing, and that is people could
2 not continue treatment beyond, say, the blinded
3 time of the trial unless they were responders. But
4 you can maintain a blind and find out some
5 interesting information.
6 DR. FIRESTEIN: And even open-label
7 extensions with safety rather than efficacy as a
8 primary endpoint would not raise the bar that much
9 higher for an acute indication.
10 There were a couple of other issues that
11 were raised that the agency has requested that we
12 discuss. One has to do with the parameter used for
13 assessing dose intervals for acute analgesic drugs.
14 The other, item three, is the issue of how one
15 measures clinically important differences.
16 Actually, I think Dr. Katz yesterday used a quote
17 that I think I am probably going to put on my
18 slide, which is if a difference doesn't make a
19 difference, then what is the difference? Or some
20 variation of that.
21 What I would like to do is try to steer us
22 towards addressing those two issues right now. One
23 is if anybody has specific thoughts on what sort of
24 dosing interval studies would be required, or
25 whether that is appropriate. Dr. Elashoff?
99
1 DR. ELASHOFF: Specifically with respect
2 to 2(b), which is median time to rescue, and to (a)
3 as well, which is the T 1/2, part of what was
4 remarked earlier is that just looking at the mean
5 or just looking at the median is not bringing in
6 variability from patient to patient. One kind of
7 thing which could be helpful in that is looking at
8 the 25th percentile or the 75th percentile, that
9 sort of information as well to help characterize
10 how typical, in some sense, the median is of people
11 and to try and get into the variability from one
12 patient to another issues.
13 DR. KATZ: I am happy to say I was
14 actually going to say the same exact thing. We
15 have been talking a lot about how to get a precise
16 estimate of duration by whichever metric, whatever
17 that will wind up being, 8 or 11 hours, but to have
18 some sense of how variable that is I think is very
19 important. If two-thirds of your patients are
20 within an hour of that, that is different than if
21 two-thirds of your patients are within 4 hours of
22 that and informs clinical practice better I think.
23 DR. FARRAR: I agree with what has been
24 said, and I think what was just being suggested is
25 actually best described as a box plot. It is a
100
1 very simple mechanism for actually displaying in an
2 understandable format the 25th, 50th and 75th
3 percentiles.
4 I think what it brings to mind is a second
5 issue which is that patients are really quite
6 different. In trying to help physicians understand
7 how to use the medication what we really need to
8 tell them is what is the minimum time that a
9 patient should wait before they take an additional
10 dose. That really is dictated by safety data. The
11 question really is if a patient only waits an hour
12 to take a second dose, an hour to take a third, and
13 an hour to take a fourth they are clearly going to
14 take much more medicine than if somebody waits
15 three or four hours.
16 The example that comes to mind is when we
17 prescribe medications for a patient 2-4 mg every 3
18 hours. What our patients will do sometimes is to
19 take 2 mg but then, because they have taken the 2
20 mg they decide they have to wait the full 3 hours
21 before they take an additional 2 mg, even though
22 the intention was for them to be able to take up to
23 4 mg in that period of time.
24 What I am suggesting really is that in the
25 label what it probably ought to say is something
101
1 along the lines of the minimum time a patient
2 should wait before taking a second dose is two
3 hours, and that would be dictated more by the onset
4 of action rather than the time at which the
5 medication would run out, and that the maximum
6 number of pills allowed in the first 24 hours is
7 such-and-such, and allow physicians essentially to
8 give patients the right to take enough medicine to
9 achieve the relief that they are entitled to get in
10 a safe circumstance.
11 DR. FIRESTEIN: Larry?
12 DR. GOLDKIND: Particularly for an opioid
13 that may be a good model. The problem is if you
14 have a non-opioid, there is a whole different
15 mechanism where the dose response curve is not
16 quite as clean. If you tell somebody, based on
17 safety, you can take another dose in a couple of
18 hours, we don't really know that that second dose
19 will benefit other than the placebo effect.
20 DR. FARRAR: Could I respond to that? I
21 agree with that, in which case I think the issues
22 that were brought up before about the 25 percent
23 non-response, or the time point at which 75 percent
24 of the patients still have an effect would be a
25 reasonable dose interval where 25 percent had
102
1 started to take an additional dose, as long as that
2 is a safe dosing regime.
3 DR. GOLDKIND: We do get data submitted
4 that has it in quartiles and the median is simply
5 the one that is highlighted. It doesn't really
6 help in decision-making. It may help in terms of
7 approvability. It may help in labeling to have
8 that data displayed so people know when the median
9 will rescue. We would have to deal with the
10 variability of whether, again, it is responders or
11 whether it is all patients. Frankly, in the model
12 are we going to apply the dental pain or the
13 surgical setting to that description? We could end
14 up with a ten-page label if we were as informative
15 as we may discuss here.
16 DR. FIRESTEIN: Dr. Borenstein?
17 DR. BORENSTEIN: To follow-up on that
18 point, I think part of the responder aspect may be
19 the half-life of the drug. While in the label it
20 may be a certain half-life, human biology, when it
21 comes to the clinic, seems to have a much wider
22 range. So, there are some people who say, yes, I
23 can take this drug and it truly is once a day, and
24 other people really say it is twice a day and I
25 need to take it because I really experience the
103
1 lack of efficacy. So, it will have an effect
2 partly on your response, but also if you can get
3 the data which shows the range of what it may be in
4 a variety of patients so you actually can tell
5 that. Tthat actually may make for a better label,
6 that it is a range and that when you have that you
7 have individuals maybe on the short side and the
8 long side. So, you may find with your dosing that,
9 in fact, what may be once a day in some patients
10 may actually end up being twice a day and to get
11 efficacy for those individuals you will need to
12 dose it that way and the drugs will have a wider
13 range of effect.
14 DR. DIONNE: I was going to endorse the
15 proposal that Jim Witter made about acute pain
16 responders as an alternative to doing mean or
17 median responses. We are probably at the point now
18 where we are going to have a better potential for
19 understanding the basis for individual variation
20 due to genetic factors. If we have the data that
21 we are using to analyze the range of responses, we
22 could possibly better interpret what is going on
23 not only on an individual basis due to the genetic
24 variation, but also we would eventually be able to
25 form, I think, more reasonable judgments about the
104
1 safety or efficacy of a drug.
2 If there was a drug that had a very
3 effective median dose, nice duration but one out of
4 a thousand patients had a very serious adverse
5 response, we might be much less willing to see that
6 as a drug for acute pain use or eventually consider
7 it for over-the-counter use versus having the
8 perception that this drug has significant
9 liabilities or significant variabilities that
10 affect its clinical use. So, if we had a formal
11 way of doing responder analysis we could get at
12 that variability.
13 The only problem is I would hope that we
14 would derive that due to some data-driven process
15 rather than just some sort of an opinion-driven
16 process. It might take a couple of years for that
17 to evolve.
18 DR. FIRESTEIN: You mean actually use
19 evidence-based medicine?
20 DR. DIONNE: Something like that.
21 [Laughter]
22 DR. FIRESTEIN: Dr. Wood?
23 DR. WOOD: It is important to recognize
24 that the duration of effect is not a simple
25 relationship to the pharmacokinetic half-life. The
105
1 duration of effect would depend on the time for
2 which the plasma concentration is above the minimum
3 effect of concentration. At a high dose that might
4 be very long and at a low dose that might be very
5 short, both of which might not be obviously related
6 to the half-life. So, the pharmacokinetic
7 half-life is not a good measure of the effect and
8 duration, and probably should be ignored, except in
9 the sense that, obviously, a drug with a very short
10 half-life will likely last less time than a drug
11 with a very long half-life unless the drug with the
12 very short half-life can be given at doses that are
13 way above the minimum effect of concentration.
14 DR. FIRESTEIN: Let's spend the last
15 couple of minutes talking about point three, which
16 is how does one determine if a difference makes a
17 difference. Would you like to get us going since
18 you are the one who generated that pithy quote?
19 DR. KATZ: Sure. I think it is actually
20 Yogi Berra or somebody like that. But I think it
21 is an empiric question and just needs to be
22 explored empirically in the context of whatever
23 model one is looking at. John Farrar has done some
24 very nice work in looking at clinically important
25 difference in neuropathic pain and I think, John,
106
1 you found that it was about 30 percent reduction in
2 pain.
3 We have done some work in a chronic back
4 pain study that Dr. Borenstein participated in. In
5 the analyses that we have been doing it looked more
6 like 50 percent pain relief was associated with
7 global measures and other signs that were the
8 marker for meaningful pain relief. So, I think it
9 depends on the individual model and it is an
10 empiric question.
11 DR. FIRESTEIN: Vibeke, in the arthritis
12 studies with visual analog scales, what have you
13 found to be something that is significant?
14 DR. STRAND: I will show you this during
15 my talk, but basically we found that it is about 30
16 percent, 30-36 percent, looking at correlations
17 with patient global assessments for various other
18 parameters, such as HAQ, disability index and so
19 on. It is about 18 percent above placebo. As we
20 just talked about, Dr. Farrar's work across ten
21 trials, randomized, controlled trials in multiple
22 different kinds of pain was very consistent. It
23 was approximately 30 percent. By VAS, we think
24 that the test/retest variability, if you are using
25 100 mm scale, is about 20. So, when you get to
107
1 about 30 you have a minimum clinically important
2 difference. That seems to work no matter what kind
3 of a VAS scale you are using. Again, I will show
4 you some of that data later.
5 DR. FIRESTEIN: Dr. Sherrer?
6 DR. SHERRER: I might be assessing a
7 rescue medication use because I think that is the
8 patient's indirect way of telling us what is
9 adequate if the pain medicine is adequate by itself
10 and they don't have to be rescued. If they have to
11 be rescued, no matter what the pain relief was, to
12 me, it was not adequate. It doesn't mean that that
13 drug is not useful. It may be useful in
14 combination but, to me, if the patient has to be
15 rescued they are telling us whatever it did, it
16 didn't do enough.
17 DR. DIONNE: I was just going to add to
18 the discussion of what is the minimally effective
19 increment of pain improvement. We did a study in
20 the oral surgery model with about 125 patients
21 starting with either moderate or severe pain. We
22 slowly titrated a nonsteroidal anti-inflammatory
23 drug IV until they reached a point where they
24 pressed the stopwatch, and then we had them fill
25 out their category in VAS scales. It was startling
108
1 that it came out to be about 50 percent pain relief
2 across the different types of pain intensity in
3 different scales.
4 DR. MAX: I have two concerns about
5 setting a minimally significant clinical
6 difference. One is that I am afraid of approval
7 creep. Now it is enough, given a reasonable safety
8 record and a sense of clinical usefulness, if you
9 just beat placebo within an acceptable alpha level.
10 I am afraid if you establish that you need to have
11 really 15 percent pain relief, the requirement may
12 creep into being that the studies need to be
13 statistically significant above that level.
14 Alternatively, I want to point out that it
15 really depends upon the context and the side
16 effects. If you had an analgesic that looked safe
17 and had no, say, cognitive side effects, you could
18 add it to most of the analgesics that are sedative,
19 and even if you only got five percent or ten
20 percent additional relief, it is cheap enough and
21 it would be a very welcome addition. So, I would
22 want to leave this to the case by case judgment of
23 the agency.
24 DR. STRAND: Could I just clarify for a
25 minute? I don't think we are talking about MCID
109
1 based on one outcome measurement as defining
2 clinical response. That is why I would like to put
3 this off until this afternoon when I present.
4 But I think what we are really trying to
5 talk about is where do we see minimum clinically
6 important differences in various parameters. The
7 way they become useful is if you now combine those
8 parameters that are not closely related into some
9 type of an analysis for responder. All of this has
10 to be done as evidence based.
11 DR. MAX: Yes, and it just depends
12 comparing to the safety profile of the clinical
13 context.
14 DR. FIRESTEIN: Dr. Cush and then Dr.
15 Elashoff, and then we will take our break so that
16 we don't have break creep as well.
17 [Laughter]
18 DR. CUSH: I just want to go back to
19 Yvonne's suggestion, and I agree that the use of
20 rescue medication is certainly an important measure
21 and I think one that is useful for analysis, but I
22 am also bothered in doing clinical trials where we
23 use rescue medicine, especially in osteoarthritis,
24 by the number of patients who refuse to use rescue
25 medication despite their pain. I can't quite
110
1 explain that. I know they have pain but they
2 continue to not want to use the analgesic medicine
3 we give them. So, I somehow fear that we may be
4 missing an important outcome if we rely too heavily
5 on that one measure. That needs to be included but
6 I don't know that it can be a primary outcome
7 measure.
8 DR. ELASHOFF: Any time one feels one
9 needs multiple measurements in order to understand
10 what is going on, you are either left with trying
11 to sort of put them together after the fact, after
12 they have all been measured, or defining some
13 arbitrary combination of them. There is always an
14 arbitrary character to that, and if you define
15 things ahead of time then you are liable to lose
16 information later on. But there is always a
17 tradeoff. There is no way to totally win this
18 situation.
19 Dr. Cush's remarks about the rescue
20 medication issue are certainly important ones. The
21 advantage of that particular type of outcome--or at
22 least if we don't think of it so much as rescue but
23 amount that they would actually take if left on
24 their own, the advantage of that kind of outcome
25 measure is that it is directly related to the
111
1 safety issue in a much clearer way than some of the
2 other outcome measures one might be talking about.
3 DR. FIRESTEIN: Dr. Simon?
4 DR. SIMON: Just before the break, if you
5 will give me a minute, there are a couple of
6 questions that arose in the previous discussion
7 that weren't really answered by us. One was Dr.
8 Katona's question about were there other
9 alternative designs besides a placebo-controlled
10 trial. That would be appropriate and, yes,
11 obviously an active comparator would be an
12 acceptable way to go for an acute pain trial in
13 children, elderly, in any number of different ways
14 to do that, background therapy, withdrawal therapy
15 as has been done in children before, though I am
16 not that enthusiastic about withdrawal therapy in
17 adults despite what came up yesterday and I am sure
18 we will discuss that part again.
19 Number two, there was an interesting
20 discussion about acute pain, time to onset of acute
21 pain, differentiation from placebo and preemptive
22 anesthesia. I would like to point out that we are
23 willing to consider that as an entirely
24 disassociated issue, meaning, we have to create a
25 label that patients understand how to use drugs.
112
1 We believe the time to onset of an hour
2 may be important to patients as opposed to two
3 hours, although I do not want to get into a
4 discussion, as we did in '98 on fast, faster or
5 fastest because, in fact, that is not really
6 informing us anything. The reality is that there
7 may be the need for an entirely different
8 indication of preemptive anesthesia rather than
9 acute pain because, in fact, that is a different
10 issue and it would affect different patients.
11 There are not a lot of patients walking around with
12 a toothache who need preemptive anesthesia as
13 opposed to acute pain relief.
14 The third issue is the issue of effect
15 size that Dr. Elashoff referred to before. It
16 refers back to what Dr. Max was talking about,
17 which is that we have to be familiar with MCID
18 because if we don't consider that the sponsors, not
19 because they are bad people but because they have
20 accrued a lot of patients in a trial, can then have
21 enough patients to show a statistically significant
22 difference from placebo yet, in fact, the effect
23 size is entirely unimportant.
24 Part of that is bias and a take on how big
25 is the effect size. It might be nice to know that
113
1 an effect size is evidence-based and defined by
2 what is minimally clinically important, and that
3 may be very important because of the number of
4 patients you could recruit. You can't just make
5 your study be positive.
6 DR. FIRESTEIN: Thank you very much for
7 clarifying those issues, and we will take a break
8 now. We will start again in exactly 15 minutes,
9 10:45.
10 [Brief recess]
11 DR. FIRESTEIN: Can the members of the
12 committee please rejoin us? In this session we
13 have an open public hearing. Then, we are also
14 going to try to clarify or revisit some of the
15 questions that were raised yesterday with regard to
16 chronic pain indications. We have two speakers,
17 Dr. Eugene Laska who has been allocated ten
18 minutes, and then Dr. Nijab Babul who has been
19 allocated five minutes, and I would like to welcome
20 them. Dr. Laska?
21 Open Public Hearing
22 DR. LASKA: Thank you.
23 [Slide]
24 This little presentation is sponsored by
25 Merck, whose folks I would like to thank for their
114
1 stimulating comments and stimulating discussions
2 which led to the clarification of several issues
3 among the contributors, their ideas, particularly
4 Al Sunshine whose name I want to mention. The
5 ideas here are ones I have talked about before. I
6 apologize for repeating some of them. Lee Simon
7 and Jim Witter and Ray Dionne also deserve special
8 recognition because they are clearly attempting to
9 open up the box and make the business of
10 registration more transparent. Some day a drug
11 company will know whether they are going to get
12 approved before they make a submission rather than
13 wait for the surprise of the letter.
14 As I mentioned yesterday, the goals of a
15 randomized, controlled trial are to allow causal
16 inference; to allow the conclusion that the drug is
17 the reason for the effect we observe.
18 I want to add to that that another major
19 reason for doing clinical trials is to get point
20 estimates of very important parameters which
21 characterize what the drug is all about. It is
22 instructive in trying to design clinical trials to
23 contemplate how one would use the information that
24 comes out of them; what kind of information one
25 really wants.
115
1 If one thinks about onset, duration and
2 dosing intervals as if you knew the entire story,
3 you know, the probability distribution of onset and
4 duration and response rates, you would see that it
5 is a complicated, multidimensional space that would
6 be very hard to characterize. And, what we are
7 looking to do in these clinical trials is to find
8 very, very minimally informative point estimates
9 which describe to some degree the amount of the
10 effect that we are talking about, median time to
11 onset and the like.
12 Too many measures, as Janet says, are not
13 necessarily useful, and for these trials for the
14 longest period of time we have collected data on
15 both relief, which refers to original time, and
16 current intensity. I am pleased to see the agency
17 moving to the notion of dropping redundancy at
18 least in the notion that it may be redundant in the
19 beginning but certainly long term. Good thinking.
20 The same thing is true about all of these
21 parameters. They are functions of pain intensity
22 levels. So, again, the hyper space in which these
23 characteristics are described is very, very high
24 dimensional.
25 [Slide]
116
1 Let me start by talking about stopwatch
2 and measure onset. I believe that it is important
3 to eliminate the two stopwatch theme that has been
4 used by many companies in the recent past and
5 return to the one stopwatch approach that measures
6 meaningful relief because I believe that that is
7 the most useful concept that can be measured, and
8 that the redundancy in having a second watch to try
9 to capture perceptible relief merely adds
10 complexity and does not really bring in enough new
11 information to warrant or justify its use. And, I
12 think that second stopwatch is a very useful tool,
13 which I will mention in a second, that cay be used
14 to look at duration.
15 [Slide]
16 Once one collects the data, I think it is
17 important to conceptualize the ideas associated
18 with onset as representing two subpopulations, one,
19 people who will not respond or who have not
20 responded; and the second, the group that has
21 responded. That is characterized statistically by
22 the top equation. It is called the cure model. We
23 won't talk about it today but it has been described
24 in the reference in the bottom of the slides. That
25 particular model conceptualizes the outcomes as
117
1 falling into two groups, the responders group and
2 the non-responders group.
3 I believe that the regulatory indications
4 of collecting data the way I have described and
5 breaking up the population into these two subgroups
6 flows very naturally. The clinical trial's
7 objective will be to estimate the proportion of
8 patients who respond, who get this meaningful pain
9 relief, and look at the survival distribution
10 including the median time to obtaining meaningful
11 relief.
12 [Slide]
13 The regulatory implications that flow from
14 that I believe fall in two camps. One is a
15 comparative camp and the other is a numerical
16 estimate camp which has to do with characterizing
17 the drug independent of another drug or placebo.
18 So, the first requirement would be that Pd
19 is bigger than Pp for the placebo group. The
20 proportion or response must be demonstrated to be
21 statistically superior on the drug than the
22 proportion who respond on placebo. Perhaps a
23 minimal difference in the proportions is called for
24 so that sample size doesn't dominate the decision
25 as to whether there is a proportion.
118
1 [Slide]
2 But then the issue of whether or not a
3 drug works within an hour or more generally within
4 T units is characterized by the second requirement
5 which only talks about absolutes, not comparators.
6 That is, the median time to onset among the
7 responders on this drug ought to be within some
8 period of time, perhaps an hour, perhaps an hour
9 and a half but more generically T. T, of course,
10 may depend on the pain intensity, the model setting
11 and a variety of other things relating to the
12 individual and the biological response that that
13 individual represents.
14 [Slide]
15 Perhaps more difficult to contemplate is
16 the question of duration.
17 [Slide]
18 Let me suggest to you that the FDA's
19 concerns about using the various interferences that
20 are introduced by the nurse or whoever is
21 collecting the data or deciding whether or not to
22 give that second dose is mitigated by putting that
23 second stopwatch that used to be used for something
24 else, so they are around and there is no extra
25 expense--that second stopwatch can be used to
119
1 answer the question when is the patient no longer
2 getting pain relief.
3 The agency used to worry about what they
4 called back then the minute wars of the first
5 interview for onset at 15 minutes, demonstrating
6 efficacy, would provoke another drug company to
7 collect its first interview data at 14 minutes so
8 that they could claim faster onset. Well, the
9 stopwatch eliminates that problem and it does so
10 here as well. It removes the bias, the
11 interpersonal possible interference that the nurse
12 observer or the person who could give the next
13 medication introduces.
14 The estimating functions that would derive
15 from collecting data of that sort are exactly
16 analogous to what we would obtain in the onset
17 story. We would estimate the survival distribution
18 of time to rescue and the proportion who respond.
19 Very importantly, they do not impute a value for
20 those people who never got onset.
21 The question of how long a drug works
22 after it has worked is not informed by the
23 percentage of people or the time at which those
24 people rescue if they never got onset. it is a
25 different question. The answer to the question of
120
1 when shall I remedicate when a person is not doing
2 well on the drug I gave him is a very different
3 question from the one that asks when do I
4 remedicate after there has been a long period of
5 time where the patient has responded.
6 A number of the things that can be
7 reported along the way are the proportion who
8 respond at the various times that are convenient,
9 like 6, 12 and 24 hours; median time to rescue
10 among responders who do rescue.
11 Let me focus on that for a minute. It is
12 useful to say ten percent of the patients respond,
13 and among the ones who do--sorry, median time to
14 rescue. Among the people who rescue, how long does
15 it take before they need rescue? That is going to
16 depend on severity and the like, but that informs
17 the notion of the time to rescue and is a
18 complement to the proportion who don't rescue.
19 Those different arms are the reason I described in
20 the beginning the hyper dimensionality of the
21 outcome space when you do a clinical trial of this
22 kind. To mix them up is to blur and lose
23 information about what is actually transpiring.
24 [Slide]
25 The regulatory implications of choosing a
121
1 dosing interval on this basis has to do with, in my
2 view, a compromise between the wide range of dosing
3 intervals that are absolutely necessary, that all
4 of the clinicians on this panel discussed in the
5 last hour but, nonetheless, if the agency chooses
6 to characterize with one number, I think that
7 number is the median despite the comment that I
8 don't want the other half of my patients to do
9 poorly because the dosing interval is honored in
10 the breach. So, if this is the one number you want
11 to produce, I think you are stuck with the median
12 and, therefore, the dosing interval is some number
13 less than or equal to the median time to rescue.
14 I believe the limitation that you place on
15 providing information in the label is a very
16 artificial one, and the notion of posting
17 information on the web doesn't need to be defended.
18 You don't need to hide behind the label to describe
19 what happened in the trials; put them out some
20 other way. Once they are out, clinicians will find
21 a way to use them if they care to find out the
22 information.
23 So, the regulatory implications are that
24 the percentage of patients, the second point, who
25 need rescue is significantly less than the
122
1 proportion of patients who need rescue on placebo
2 among the people who responded to placebo. That
3 would need to be demonstrated statistically.
4 The first point, the comparative one, the
5 absolute is that the proportion of responders is
6 less than some fixed time point, and that is less
7 than a half.
8 [Slide]
9 Just one comment quickly on Larry's
10 feeling that return to baseline is a flawed metric.
11 I think one can conceptualize this whole idea as
12 the complement, the counterpoint to the responders
13 analysis. If you like, this is the failures
14 analysis and patients will return to baseline
15 individually. The argument that the mean does not
16 return to baseline doesn't mitigate against the use
17 of return to baseline or no longer getting
18 meaningful relief on an individual basis, and it is
19 the counts of how many of those people there are as
20 well as the time to the event that makes the game
21 playable.
22 [Slide]
23 So, clearly informed by PK and informed by
24 the experience of the clinical trials in the acute
25 phase, one has to look at multiple days and the
123
1 question is what to do in that context and I had to
2 think about it. My view is that this is not the
3 place to be exploring dose response. In the very
4 mild pain circumstances where pain is almost gone
5 the next day, it makes no sense to me statistically
6 as a statistician to impute data from day one to
7 day two to show artificial differences which are
8 not real.
9 I believe that you can only sustain the
10 notion of what the effective dosing interval that
11 has been proposed and see if it makes patients
12 "happy." So, at the end of day in these mild cases
13 there should be no need to demonstrate superiority
14 to placebo, but the proportion of patients who
15 require rescue ought to be smaller than some
16 absolute number that is credibly determined on a
17 judgment basis.
18 [Slide]
19 For more serious pain or perhaps severe
20 pain models were PRN narcotic is required, I see no
21 alternative to the idea of using the dose sparing
22 property of the drug.
23 [Slide]
24 There is an old rule that every animal
25 pharmacologist will ascribe to, I am sure, that
124
1 says if you fix dose, study outcome. If you fix
2 outcome, study doses. In the dosing sparing
3 setting where you use PRN narcotics you are fixing
4 an outcome. Patients titrate to adequate relief.
5 The only thing to study is the amount of narcotic
6 that is spared. It is sensible and there are
7 caveats raised by others in the group here about
8 interaction, about promoting side effects.
9 Remember, this drug has been studied in the acute
10 setting. It is known to be an analgesic. Now the
11 question is what does it do on day one, two or
12 three and that kind of sparing relationship, in
13 face of the knowledge from the earlier trials, is
14 pretty clearly evidence if you believe in the
15 hidden assumption--as Jim pointed out, there is a
16 hidden assumption and in this case it is that there
17 is a dose response to the narcotic being used. So,
18 dose sparing makes sense to me as the way to
19 sustain that data.
20 [Slide]
21 One last situation then, we are in
22 long-term use, and I am anxious to hear the
23 objection. If chronic pain situations where
24 patients on placebo drop out at very high rates,
25 once again we are into the game of projecting
125
1 forward; we are making up data--statisticians call
2 that imputation, to justify whether the drug still
3 works at week W where W is a big number like 12.
4 I think that makes no sense. It is a
5 circumstance, again, where we are only trying to
6 sustain the notion that this drug continues to work
7 after 12 weeks. We are not trying to prove
8 effective here; it is does the drug still work?
9 The best way to answer that question is not with
10 respect to placebo patients who drop out earlier;
11 it is with respect to patients in whom the drug is
12 working, it is withdrawn and superiority to placebo
13 in a randomized, controlled trial is demonstrated.
14 I believe that this kind of an approach is
15 a rational way of looking at onset and duration and
16 choosing dosing interval. And, I thank you for
17 listening.
18 DR. FIRESTEIN: Thank you. The next talk
19 will be from Dr. Babul, from TheraQuest.
20 DR. BABUL: Good morning.
21 [Slide]
22 I would like to address the committee and
23 the division on the issue of multi-dose analgesic
24 development. This is one of the questions that the
25 division has asked the committee to consider in
126
1 terms of evaluating analgesics in acute pain.
2 [Slide]
3 I have previously provided a conflict of
4 interest statement and that stays on record so I
5 won't repeat it here.
6 [Slide]
7 This slide shows the essential approach
8 that we have been taking for the last two decades
9 to evaluation and approval of analgesics in acute
10 pain. Certainly from an efficacy perspective, we
11 do some of those studies by screening a patient,
12 initiating some sort of an acute insult, having
13 some sort of a period of recovery when the pain
14 stimulus reaches a particular intensity, moderate
15 or severe usually. We will then dose the patient.
16 We evaluate the response over a single dose and
17 then we terminate assessments either after the
18 dosing interval is over, which is generally 8, 12
19 or 24 hours, or at the time that the patient
20 requests their first rescue analgesic.
21 [Slide]
22 There are compelling reasons why
23 pharmaceutical sponsors have not gone down the path
24 of efficacy evaluations in the multi-dose arena,
25 and I would like to address these and propose some
127
1 potential solutions.
2 [Slide]
3 There is no doubt that there is no growing
4 request for data. I recall that even at the Vioxx
5 advisory committee meeting there was discussion of
6 the availability or relative lack of multi-dose
7 data in the dossier. There have been increasing
8 requests from both Division 550 and 170 for such
9 data.
10 I think the challenge here is, if I can
11 just be frank and I guess this is for the record,
12 that our collective rhetoric perhaps outpaces the
13 actual science of drug development. In other
14 words, our methodologic ability, to echo what Dr.
15 Laska was saying, to actually tease out some of
16 those differences is not always there.
17 In order to address this issue of
18 multi-dose analgesic evaluation from an efficacy
19 perspective, we need to ask ourselves precisely
20 what our objectives are. Are they to establish
21 efficacy? Are they to demonstrate effectiveness?
22 Are we trying to establish dosing frequency? Are
23 we trying to prospectively test a draft package
24 insert? Or, are we merely trying to provide some
25 sort of supportive safety data in a perioperative
128
1 setting where perhaps patients might be critically
2 ill and otherwise compromised?
3 [Slide]
4 Here are some of the challenges to
5 evaluating these drugs in acute pain. The first
6 issue, and this has been alluded to earlier, is
7 that the natural trajectory of acute pain is such
8 that, whether treated or untreated, for the most
9 part it diminishes. To be sure, and Dr. Katz
10 referred earlier to thoracotomy patients or lumbar
11 laminectomy patients who may have somewhat
12 long-term pain. To be sure, some patients may have
13 a longer trajectory, but a majority of these
14 patients have a relatively short trajectory. So,
15 this introduces an issue that most analgesiologists
16 have called assay sensitivity.
17 We are also faced with a reduced duration
18 of hospitalization. A significant number of
19 patients after major surgery are home within four
20 days to a week's time.
21 There is also a growing trend towards
22 surgical techniques that reduce surgical pain. For
23 instance, hip arthroplasty, as is currently being
24 conducted, requires substantially less
25 postoperative opioids than perhaps 10 or 15 years
129
1 ago and this presents a bit of a challenge.
2 Furthermore, patients will sometimes
3 refuse to consent to multi-dose placebo controlled
4 studies. It is one thing to convince patients to
5 do a single-dose placebo controlled study, but to
6 tell them you are going to repeatedly be give
7 placebo over the next five or seven days presents a
8 bit of a challenge.
9 We also have this issue of data
10 contamination when you give rescue analgesia, and
11 we have a problem in terms of availability of
12 trained analgesic observers or nurse raters. This
13 is a very specific discipline requiring an
14 exceptionally well-trained individual who truly
15 understands analgesic methodology, and there is a
16 real shortage of such folks. Your most senior
17 study coordinator usually wants to work the day
18 shift so you have 72 hours more to go beyond that
19 to evaluate the patient.
20 [Slide]
21 I would like to suggest some proposed
22 approaches without getting too prescriptive. Some
23 of these have really been spurred through
24 discussions with Division 550 with Dr. Witter and
25 Dr. Simon and others. One option clearly is to use
130
1 active controls, with the Division's prior consent.
2 That is certainly one possibility to consider.
3 The other option is to use what I call
4 pseudo placebos. So, these would not be placebos
5 but would be perhaps ultra low dose of an approved
6 agent, to allow us to get some assay sensitivity.
7 Yet another option, and this was discussed
8 previously by Dr. Laska, is to use rescue analgesia
9 as an endpoint. This has been used successfully
10 but only with a modest degree of success in the
11 past.
12 We can also integrate rescue and pain
13 assessment data, and there are some techniques
14 available for that. Of course, because of the
15 shortage of trained study coordinators, we can
16 perhaps consider doing serial assessments long
17 term. We can use recall instruments to assess
18 pain.
19 [Slide]
20 The rationale for integrating rescue and
21 pain scores to come up with some composite scores
22 is given on this slide, and I am going to be brief
23 here. Traditional studies have tended to discard
24 rescue after the first dose. The issue is that
25 rescue tends to confound our analgesic evaluation.
131
1 Furthermore, rescue differentially confounds the
2 analgesic response. David Silverman, for instance,
3 has suggested a rather elegant but simple approach
4 to integrating rescue and analgesia scores.
5 [Slide]
6 Alternative approaches that are available
7 involve the use of recall instruments. We know
8 that recall, at least among analgesiologists, is
9 viewed as somewhat suspect but we, and others, have
10 shown and have published data demonstrating that
11 recall is actually quite sensitive. We have done
12 studies where we have looked at recall in
13 orthopedic pain and other models, and we think that
14 this allows you perhaps to conserve on the
15 resources that are a problem in multi-dose studies.
16 [Slide]
17 The last potential option that one ought
18 to consider is rescue analgesia as an endpoint. I
19 believe it is a potential endpoint. It does have
20 some risks because the variability is not
21 insignificant.
22 [Slide]
23 These are data that were presented in 1998
24 at the Arthritis Advisory Committee in the review
25 of rofecoxib submission. As you can see in this
132
1 particular study, over day two to five there was a
2 difference between placebo and rofecoxib in terms
3 or rescue consumption. It was a one tablet per day
4 difference. Now, whether this is clinically
5 meaningful is a separate issue but it certainly
6 provided some assay sensitivity in an attempt to
7 look for differences.
8 In summary, the methodology for multi-dose
9 efficacy evaluation is not quite cooked; it is not
10 established. I think there are some possible
11 options that are available, but we need to
12 understand that there are some compelling reasons
13 why single-dose evidences have formed the primary
14 basis for efficacy evaluation. None of these
15 techniques can meaningfully, in my opinion, answer
16 questions related to the time course of effect and
17 dose response. Those questions, and they are
18 critical questions, need to be addressed in
19 single-dose efficacy evaluations. Thank you.
20 Further Discussion of Criteria for
21 Chronic Global Pain
22 DR. FIRESTEIN: Thank you very much. At
23 this point Lee has asked us to revisit our
24 discussion of the proposal for the criteria to
25 obtain a chronic global pain indication. Just to
133
1 remind people, there are two essential issues. One
2 is that for such an indication the proposal was
3 that three separate models would need to be
4 explored, and in each of them there would be three
5 separate domains that would have to be all
6 positive.
7 So, what we are going to do now is
8 actually go around the table and get people's
9 opinions on those issues. I would ask that people
10 restrict their comments to two minutes or less.
11 Please don't feel obligated to use the entire time
12 because there are about twenty of us and it will
13 take quite some time if we wax poetic.
14 I will go ahead and start and then people
15 can take various and sundry pot shots at my
16 comments, either amplify or deny them.
17 DR. ELASHOFF: I am still unclear on the
18 question.
19 DR. FIRESTEIN: The question is what do
20 the individual members feel about, number one, what
21 the criteria should be for a chronic pain
22 indication, with the initial proposal that there be
23 three separate indications explored in order to get
24 labeling for chronic pain.
25 DR. SIMON: Global chronic pain indication
134
1 with three areas of etiopathogenesis that would
2 have to be studied with three domains as
3 co-primaries in replicate trials.
4 DR. FIRESTEIN: So, those are the two
5 separate issues that we should comment on. Does
6 that clarify that?
7 DR. ANDERSON: But what are domains?
8 DR. SIMON: To remind you, they were
9 patient global, function and a pain score. It is
10 just in chronic pain. I know we have just talked
11 about acute pain but we didn't get enough clarity
12 yesterday for us to know exactly what you all felt
13 about our proposal.
14 DR. FIRESTEIN: We were appropriately
15 obtuse. So, I will start and then we will just go
16 around the table. For introductions we went to my
17 left and this time we will go to my right.
18 There were a number of other proposals
19 that were also made with regard to the number of
20 indications. First of all, I think that the bar
21 should necessarily be high for a global chronic
22 pain indication. The question whether it should be
23 two, three, four or five indications is really not
24 well defined by evidence-based medicine but, based
25 on opinion, three doesn't sound like a lot and four
135
1 sounds okay and five sounds like a lot. So, by
2 process of elimination, four sounded reasonable to
3 me.
4 The other issue is whether or not you need
5 replicate trials for a global pain indication. It
6 seems to me that the indication is global pain, not
7 the individual models. So, for instance, a
8 confirmatory trial would not be a second OA trial
9 but a second trial in another indication,
10 preferably different mechanism, and I think there
11 needs to be considerable care with regard to
12 choosing how one selects the different models,
13 making sure that there is adequate representation
14 from multiple mechanisms--neuropathic pain,
15 musculoskeletal pain, cancer pain, etc. So, from
16 my perspective, it seems to me that a single trial
17 with more indications makes sense.
18 With regard to the domains, the main issue
19 is that function may not necessarily be a
20 reasonable endpoint for some of these indications,
21 as was pointed out yesterday, and I think there
22 needs to be some flexibility in endpoint selection.
23 Pain is obviously going to be the more important
24 one and function may be less important in certain
25 patients where strictly comfort is all that
136
1 matters.
2 So, why don't we move off to the right?
3 Dr. Brandt? Was that clear enough?
4 DR. BRANDT: Fundamentally, I think I
5 agree with Gary. The complexities in the science
6 that drives chronic pain, as we heard yesterday, I
7 think are very significant and it makes it hard to
8 reduce this in terms of a limited number of models
9 of disease states in which a drug shows efficacy to
10 be comfortable that that truly gives enough
11 information for a global pain indication. So, I am
12 more comfortable considering pragmatics. I think
13 it would be reasonable.
14 I think we regard to the outcome measures,
15 certainly pain, certainly patient global, and I
16 think that you have to look at function in terms of
17 the specific disease state that is more relative to
18 certain diseases than it is to others, as we heard.
19 But I think the greater breadth that would be
20 provided by demonstrated efficacy in four disease
21 states for chronic pain has appeal to me, and
22 perhaps more than looking at three times with the
23 six-pack.
24 [Laughter]
25 DR. KATONA: Looking at the issue from the
137
1 pediatric point of view, for the chronic model it
2 will be very difficult to recruit enough patients
3 since out of the four proposed models really the
4 only one which could be found in children in great
5 numbers is the cancer pain. Children have no OA,
6 very rarely low back pain, a low incidence of
7 neuropathic pain. So, I think the study is going
8 to be limited. The acute model I think is very
9 important in children. So, those two will have to
10 be concentrated on.
11 As far as efficacy, I think we always rely
12 a lot on the adult trials and I think we definitely
13 will do the same. However, I think the PK studies,
14 the dosing schedule and especially the safety are
15 going to be extremely, extremely important in
16 children. So, I think those are going to have to
17 be conducted and these have to be long term. Thank
18 you.
19 DR. ABRAMSON: I would maybe take a
20 slightly different position at least from Ken and
21 Gary on this. I mean, chronic pain is a very broad
22 term. Although it is clinically a very important
23 issue, the name of the term itself is like the 1899
24 Merck Manual of Hepatology or lumbago and I think
25 we have to be careful in setting a bar for a
138
1 broader indication that the elements within that
2 indication are robust in the way that they are
3 looked at from the term etiopathogenesis that Lee
4 used.
5 Therefore, whether a global pain
6 indication requires three, four or five individual
7 etiopathogenic syndromes, I think the bar for each
8 of those syndromes has to be as high as it would be
9 for anything else that a drug is getting approved
10 for, namely, two replicate pivotal studies for
11 example.
12 When you talk about domains in these
13 studies, the domains may vary within the syndrome
14 you are looking at, whether it is neuropathic pain,
15 low back pain, osteoarthritis pain, etc. So,
16 clinical outcomes, meaningful clinical responses,
17 things that you might tag on to look for mechanisms
18 of pain will vary within each of those.
19 So, I would make the argument for keeping
20 the bar very high for any individual entity of the
21 individual syndromes that need to be looked at,
22 recognizing that fibromyalgia is different from low
23 back pain and the musculoskeletal indication for
24 example.
25 Then, whether one gets for marketing
139
1 purposes a more global indication will depend on
2 three, four or five very highly rigorous standard
3 replicate studies that would have been required for
4 independent registration.
5 DR. FIRESTEIN: Lee, would you just
6 comment on whether or not this would change the bar
7 for individual indications? In other words, that
8 is a separate issue I think.
9 DR. SIMON: No, in fact, the bar, as we
10 have described it in my earlier discussion, for any
11 one indication with two replicate trials with three
12 domains is obviously open to discussion based on
13 which domains, but we would like patient global
14 pain and a functional domain. It is particularly
15 applicable to osteoarthritis but it may not be
16 applicable to all of them. So, that would not
17 change an individual indication issue.
18 What we are really discussing here is, is
19 that high bar too high for the global chronic pain
20 indication? And, we each have our opinion and that
21 is what we are waiting to hear.
22 DR. WITTER: I just want to add a thought,
23 and I think Dr. Katz brought it up yesterday. As
24 you think about this, I mean, we are interested in
25 labeling that makes sense to you as clinicians and
140
1 also to your patients. So, were we to construct
2 chronic pain, the big claim, you know, I think you
3 need to think through your current repertoire of
4 medicines and ask if they should be able to reach
5 that hurdle. If they do, then what implications
6 does that have for whatever claim structure we
7 might set up because would we be creating something
8 and everybody would get it and may not have what we
9 had hoped down the road. So, I think maybe you
10 want to think about that as well.
11 DR. MANZI: I think when I was thinking
12 about this the one assumption here that is probably
13 true is that the number one biggest problem
14 probably in the U.S. is that we under-treat chronic
15 pain, more than abuse of medications or
16 over-treatment. So, with that in mind, I said what
17 would the advantage be of having a global
18 indication more than industry incentive in some
19 way? What advantage to the patient?
20 I guess from that perspective, I actually
21 would presume that a global indication may open the
22 door for a broader application of some of the
23 potential medications in patients with chronic
24 pain.
25 With that in mind, I would say what are
141
1 the downsides? The downsides may be that it is not
2 as effective in certain disease states or that
3 perhaps in certain subpopulations it may not be
4 safe. I think those are clear concerns.
5 With that in mind, I guess my perspective
6 is that I might actually consider lowering the bar
7 a bit and say is it really safety issues and
8 efficacy that we are worried about, or do we really
9 want to open up to our patients the availability of
10 a broad range of potentially helpful agents for
11 treating chronic pain?
12 With that said, this is arbitrary but I
13 would say I would go a little lower with perhaps
14 the three entities not having to capture every
15 pathophysiologic mechanism for pain because I am
16 not sure that is even possible, obviously, keeping
17 the individual rigor that the FDA does already with
18 each of those entities. So, I think I would favor
19 more a slightly lower overall bar to get a global
20 label for the reasons that I mentioned.
21 As far as the domains, I agree with the
22 previous speakers that I think you have to a priori
23 determine which domains are relevant to the disease
24 state that you are looking at and decide what the
25 success is in each of those and not make a standard
142
1 requirement across the board for each population.
2 DR. KATZ: I feel more comfortable
3 articulating some general principles relevant to
4 this discussion, rather than just throwing out a
5 number of five, three or something like that. So,
6 I don't know if my comments will help you in any
7 way but I will go ahead and take my two minutes or
8 less anyway.
9 First of all, there has been a great
10 debate as to whether giving an overall
11 categorization for acute pain, chronic pain, or
12 what-have-you, is appropriate. My feeling is that
13 the opioids have taught us that it is possible to
14 have a class of drugs that are broad spectrum
15 analgesics for just about all kinds of pain. So, I
16 think that the notion of a broad spectrum analgesic
17 does have construct validity.
18 Number two, I think the opioids have also
19 taught us that just because a drug has broad
20 spectrum applicability in acute pain, chronic pain,
21 it doesn't mean that it is going to work for all
22 subcategories or all populations or all people. I
23 think that is fine and it should not dissuade us
24 from giving a broad sort of labeling, although it
25 would be nice if we had some way, through the label
143
1 or otherwise, to educate physicians that just
2 because a drug has a broad label doesn't mean it
3 will work for everybody and it doesn't relieve them
4 of their responsibility to manage their individual
5 patient or different disorders.
6 I think acute pain as a category does have
7 construct validity and I think chronic pain as a
8 category does have construct validity too. It
9 seems to me that in order for something to be
10 called a medication for chronic pain, it needs to
11 work for neuropathic pain as a broad construct and
12 also for musculoskeletal pain because drugs that
13 work for musculoskeletal pain may not work for
14 neuropathic pain, and vice versa. So, it is
15 inconceivable to me that something could be called
16 a medication for chronic pain without working
17 robustly in both of those different categories.
18 So, I wouldn't see it possible to label a
19 drug for chronic pain unless one could also label
20 it for neuropathic pain broadly and one could also
21 label it for musculoskeletal pain broadly, with
22 whatever robustness of evidence one would need in
23 each of those individual subcategories.
24 We have just had a meeting for a whole day
25 and talked about neuropathic pain and what sort of
144
1 trials would be necessary for that. People have
2 thought that you would need a six-pack or more just
3 for peripheral neuropathic pain, let alone chronic
4 pains. That is a big discussion and I am not going
5 to try to summarize it all here, but I think it is
6 important to just say that you have to be confident
7 of neuropathic pain before you get to the point of
8 chronic pain.
9 In terms of the issue of replicate trials,
10 personally I find it much more useful to see
11 different trials in different disease entities than
12 in the same entity. For example, two identical
13 replicate trials in osteoarthritis don't help me
14 nearly as much as one good trial in osteoarthritis
15 and one good trial in some other kind of
16 musculoskeletal pain like low back pain or
17 rheumatoid arthritis, or something like that. I
18 think that is where the information comes in. So,
19 personally I would discourage replicate trials and,
20 if you are looking for a broad categorization, then
21 try to get as broad an experience as possible of
22 disease entities within that category.
23 Lastly, in terms of the issue of the
24 requirement for the three co-primaries, my
25 experience suggests to me that that is an
145
1 absolutely wrong approach. I think it is obvious
2 that if a drug reduces pain but does not
3 necessarily improve function, quality of life or
4 whatever, it is still an analgesic.
5 On the other hand, I think that those are
6 very, very fundamentally important secondary
7 outcome variables that will differ from disease to
8 disease and can also help us understand the meaning
9 of the primary and borderline cases or unusual
10 cases. I think the data should definitely be
11 collected. It should be required but not as
12 co-primaries for developing analgesics.
13 DR. ANDERSON: I actually agree with quite
14 a lot of what Dr. Katz said, although I disagree
15 about the domains. First, I didn't like the idea
16 of this global indication at all because I just
17 don't think a single drug can do it all and also
18 retain function. Also, it seems to me that it
19 would be abused in the sense of, you know, you had
20 all your three areas or even six areas where you
21 showed it worked it would be used in many more
22 where it might not work at all or might be unsafe.
23 So, I think that you should just stick
24 with what you have at the moment, which is for any
25 particular indication, pathogenesis area or
146
1 whatever, you have to have two trials, perhaps with
2 a different disease.
3 I think that the three domains are all
4 important. Okay, this is an analgesic but it is
5 more than an analgesic. You know, for an analgesic
6 which is just for acute pain, then, okay, pain is
7 the only outcome that matters. But for an
8 analgesic that is for chronic pain or long-lasting
9 pain, then it is not much use unless the person can
10 have function unless you are talking about terminal
11 illness where there is no hope for that. But I
12 think that we would want to use these drugs in
13 cases where people want to retain and improve
14 function. So, function, patient global and pain
15 score I think are equally important and should all
16 be kept and be required.
17 DR. ASHBURN: I am an anesthesiologist who
18 has left the OR to take care of patients who have
19 chronic disease over long periods of time. So, as
20 a result, I am used to having conflict within
21 myself.
22 [Laughter]
23 I think that this is one of the areas
24 where I have mixed feelings. In a global area I
25 think it is really important to recognize that
147
1 individuals who have complex chronic pain disorders
2 require more than one medication. They frequently
3 benefit from polypharmacy with medications targeted
4 towards specific issues and specific individual
5 patients. They frequently have depression; they
6 frequently have sleep disorders; frequently have
7 anxiety. They also have social issues that need to
8 be addressed by cognitive behavioral therapy. They
9 also have physical dysfunction and require
10 activating physical therapy. To a certain degree,
11 it is almost disingenuous to think that one
12 medication could be useful as a global indication
13 for chronic pain.
14 The other thing that even makes it more
15 difficult in that area is that pain management
16 physicians and physicians in general tend to be
17 enamored with the use of unproven techniques in
18 this patient population. I think that that poses
19 some concern with regard to safety.
20 On the other hand, six well-controlled
21 trials for the indication seems to be an extremely
22 high bar. Drilling down to the specifics, I am a
23 little bit worried about the specific definitions
24 of group as far as how you define, how you group
25 patients. One concern that was already brought up
148
1 is how would you study children, and for
2 essentially orphan children who have chronic pain
3 in these areas. Clearly, designing six
4 well-controlled clinical trials that include
5 adequate numbers in children would be extremely
6 difficult. Do you do it by mechanism? Do you do
7 it by cancer? We have already heard discussions
8 that patients who have metastatic cancer don't
9 necessarily have one etiology of their pain but
10 frequently have multiple ones that are working
11 simultaneously, and is that a meaningful patient
12 population to study? Or, do you do it by body
13 location, which also is fraught with all sorts of
14 problems?
15 My concern is that if you set the bar too
16 high companies will go for a narrow indication,
17 which may be appropriate but, on the other hand, a
18 narrow indication will lead towards less data on
19 safety in different patient populations, which I
20 think would be very helpful in guiding use.
21 With regard to a patient global
22 indication, I think that this is something that
23 probably ought to be required but I have a concern
24 about it being used as a primary endpoint to
25 determine approval. I think having six positives
149
1 is very, very difficult. Also, I don't know that
2 the patient global assessment is well defined in
3 the literature, and whether or not that assessment
4 tool, which has become very common, has been
5 validated in a meaningful and appropriate way and
6 is used in a uniform and consistent manner.
7 Lastly, most of the function scales have
8 multiple different measurement tools and they have
9 to be well defined with regard to how you would
10 affect function. The usefulness of a tool will
11 vary by patient populations. So, it is possible
12 that you will be offered different function
13 assessment tools for different patient populations
14 and you will not be able to combine that in a
15 meaningful way. Again, with pediatrics there is
16 very little data on validated disease-specific
17 measures of health in children with pain, and even
18 less data on children at the end of life. As a
19 result, children are again going to be orphaned.
20 An alternative is to require the use of
21 validated, as best one can, disease-specific
22 measures of health specific for the population to
23 be studied in each individual trial and use that
24 data, not necessarily solely for determination of
25 approvability, but use that to inform the label.
150
1 Thank you.
2 DR. ELASHOFF: I don't feel well enough
3 informed to comment on the issue of how many
4 separate indications one might make or what they
5 would be. However, I do feel that each one going
6 into that should have sufficient information. So,
7 I feel very strongly that you should have replicate
8 studies.
9 In terms of the outcome domains, probably
10 each indication is going to need somewhat different
11 ones, but the whole issue that I am concerned about
12 is that all this needs to be extremely carefully
13 defined before the study is started or, perhaps
14 even before you talk about an indication for a
15 specific area which things ought to be measured.
16 the whole issue of exactly how one is going to deal
17 with multiple co-primaries on a statistical basis,
18 what you are going to do about alpha levels what
19 the implications of this are for power, you will
20 probably need to look very closely for each
21 indication at how correlated these things are
22 because that is going to have a great deal of
23 influence on the powering of the study. If they
24 are very highly correlated you are in essence only
25 asking for one of them. If they have very low
151
1 correlation, then you may well need bigger sample
2 sizes.
3 The other thing that wasn't put into the
4 question, although some people have mentioned it,
5 is that I think the safety requirements, the safety
6 information that you would need if you are going to
7 have a global indication should be far greater than
8 for any single indication.
9 DR. FIRESTEIN: Dr. Farrar, you are up.
10 DR. FARRAR: I guess from my perspective,
11 understanding that no drug is going to be perfect
12 and that every drug is going to fail at something
13 and that FDA approval is being used more and more
14 to limit payment for therapies by insurance
15 companies, I am in favor of a global indication to
16 allow me to use medications in patients for which
17 there is good clinical trial evidence that they
18 work but which may not have been submitted to the
19 FDA for formal approval, which is really very often
20 driven by costs and marketing considerations.
21 As such, I think it is reasonable to think
22 of a global indication. In fact, I would favor two
23 trials in syndromes which are clearly neuropathic
24 and would also request that those be in separate
25 entities but clearly neuropathic, and two trials in
152
1 what are clearly somatic pain, also two separate
2 entities as being the bar for efficacy.
3 In addition, since patients really are the
4 defining factor in terms of whether a medication
5 works or not, I think that the global outcome is
6 exactly the right measure provided it is done
7 correctly, and I think it can clearly be done
8 incorrectly. By correctly, what I mean is that it
9 is supported by several other outcomes that are all
10 going in the same direction. To have a global
11 outcome that is by itself I think would be
12 incorrect.
13 In this setting, however, the most
14 important issue and the thing for which the bar
15 needs to be set the highest is safety. If the drug
16 is going to be used or potentially used in a wide
17 variety of patients, it needs to be shown to be
18 safe in those populations, in specific, the elderly
19 and children. It may be hard to find enough
20 children to demonstrate efficacy in all of these
21 areas, but if I know that it is going to be safe I
22 would be willing to try it, and maybe clinical
23 trials that are done outside of FDA approval will
24 help to guide my therapy.
25 Lastly, I would like to suggest that
153
1 perhaps there needs to be a different study that is
2 called perhaps a labeling study. We look at dose
3 in a Phase II trial, but maybe we need to look at
4 dose in Phase III(c) or perhaps even in Phase IV to
5 help us answer some of these questions that have
6 been raised in terms of whether a 50 percent
7 response time is the appropriate dosing schedule if
8 it, in fact, limits our use of the medication. In
9 actual fact what we are talking about is limiting
10 the use as opposed to providing real benefit in
11 terms of the guidance for use. So, those would be
12 my suggestions.
13 DR. BORENSTEIN: My thoughts on the
14 subject have to do with trying to follow the
15 clinical situation with the clinical setting. If
16 we are going to have a chronic pain indication on a
17 general basis, those situations for an individual
18 neuropathic pain versus low back pain versus even
19 osteoarthritis may not be quite the same. My hope
20 would be that the FDA would allow studies to be
21 done that could show potential efficacy that would
22 mirror the clinical situation. Now, it may make it
23 a little bit more difficult because the trials may
24 have a different look to the patients that would be
25 admitted and things of that sort. But it would
154
1 have greater applicability to what the clinical
2 situation is.
3 So, whether that would be three or four
4 settings where it would follow what would be
5 happening in the clinical situation, that would
6 make it much more applicable. So, this idea of
7 either having multiple drugs and adding or
8 withdrawing would then be allowed so that a trial
9 for osteoarthritis might look different than one
10 with neuropathic pain versus one with low back
11 pain, but would still be accepted and how many
12 would be needed, whether that would be two of each
13 in neuropathic and somatic versus three, I think
14 would still need to be decided.
15 I think also very important is the idea of
16 safety and that the studies be done at least long
17 enough for us to get a handle on how these agents
18 would be used in these clinical situations. I
19 think that is very important because it is all well
20 and good to have a single drug and see whether it
21 is safe but in the real world many patients are on
22 three, four or five different drugs. They are
23 hypertension drugs; diabetes drugs. And, it is the
24 interaction of the new agent with the other ones
25 which makes it, once again, clinically applicable.
155
1 So, I think the closer we can get to the real world
2 and still do good science would certainly be quite
3 useful.
4 The last point I would make regards the
5 domains. I think a global assessment is clearly
6 very important, but I think as an analgesic, we
7 want to be sure that patients are achieving pain
8 relief and that should be the primary outcome of
9 studies. But every study should look at patient
10 satisfaction and global outlook. So, I think those
11 two at least. Then, in the appropriate setting how
12 that is affecting their daily function and using
13 the appropriate outcome measure to measure that
14 would once again be important. But, once again, I
15 think it is the clinical situation, as close as we
16 can get to it, the greater will the impact will be
17 of the information which is actually observed from
18 these studies.
19 DR. STRAND: Well, I would like to perhaps
20 give a little bit of a preview to what I was going
21 to say this afternoon, after lunch. The group that
22 I led at the NIH breakout meeting finally decided
23 on five domains that they felt were essential as a
24 minimum number of domains to be assessed in
25 clinical trials of chronic pain. They were pain;
156
1 patient global; some type of measure of physical
2 function or health-related quality of life, a
3 generic measure of health-related quality of life
4 and adverse events.
5 So, what we are really talking about here
6 I think is that these need not necessarily be
7 co-primaries. As has been done in other diseases,
8 and I am not trying to shove this into the
9 rheumatoid arthritis model, one could ask for any
10 number of these five domains assessed by different
11 instruments to show improvement without the others
12 showing deterioration.
13 We could perhaps elevate patient global to
14 something like a health utilities measure, which is
15 more like the way the patient would weigh all risks
16 and benefits from the intervention in terms of
17 their pain and assess what they think of it.
18 Certainly, we talked about physical
19 function and belabored the point that it doesn't
20 work in metastatic cancer pain. I would simply
21 argue that what we need to be doing is looking at
22 the instrument. There are plenty of different
23 instruments that would assess domain of some type
24 of function--the ability to perform activities of
25 daily living, the ability to even get out of bed,
157
1 whatever. They can be disease specific even down
2 to the type of cancer that there is. So, I think
3 there always is some instrument that would help in
4 the clinical setting that we are looking at the
5 pain.
6 Clearly, we have to ask about pain. A
7 reason to look at a generic measure of
8 health-related quality of life, besides economic
9 assessments which might be important in
10 noon-malignant types of pain, would also allow us
11 to compare interventions across different kinds of
12 pain. If we are talking about doing, say, three
13 different models or four different models of
14 chronic pain, somatic, musculoskeletal, or
15 inflammatory as I would like to think of it, versus
16 neuropathic.
17 Adverse events are obviously quite
18 important and that was, of course, the fifth
19 domain. In terms of the fact that these domains
20 would not be closely related, if they are combined
21 in some type of a responder analysis that should
22 decrease the sample size quite significantly. It
23 certainly is true with rheumatoid arthritis. In
24 terms of saying that perhaps both the global and
25 the pain measures, whatever they might be, have to
158
1 be required as improved and then the others must
2 not show deterioration, or whatever, that is
3 another way to make sure that the domains that
4 everyone thinks are most important are specified.
5 But it also makes it a lot easier than requiring
6 that any three domains be co-primaries which is
7 very difficult.
8 Finally, not to do any of this that isn't
9 evidence based. I have been a part of predefining
10 responder analyses on the basis of consensus with
11 there being no data, and those are fraught with
12 very much of a likelihood of failure, as Jane
13 Elashoff has mentioned. But it could be done based
14 on looking at data in Phase II with the product and
15 then defining a responder analysis based on the
16 data dredging from the Phase II studies.
17 DR. MCLESKEY: I would like to reiterate
18 what I said basically yesterday, that I think we
19 are all in this together. Our purpose, as I
20 believe I mentioned yesterday, is to advance the
21 practice of medicine and how might we best go about
22 doing that
23 The concern that I expressed yesterday, I
24 will reiterate today, and that is to study a new
25 agent in three different models of disease, each
159
1 studied in a replicate fashion; each having three
2 co-primary requirements that all have to hit in
3 order to obtain a claim is, in fact, a high hurdle,
4 perhaps too high a hurdle, perhaps a hurdle that
5 you simple cannot get over. I am just concerned
6 that if industry feels that it is such a high
7 hurdle that it can't be achieved then that might,
8 in fact, stifle innovation, which is the antithesis
9 of what we are all about.
10 So, I just restate that again. I hope
11 that I am reflecting adequately what industry in
12 general feels, but it seems to me that the hurdle
13 that has been proposed as a possibility seems a bit
14 high and potentially challenging to a degree we
15 can't meet.
16 Another issue, and it has been raised by
17 previous panelists around the room, is that some of
18 those co-primaries may actually be inappropriate in
19 certain models of disease and, therefore, maybe
20 those co-primaries need to be reexamined and
21 reduced a little bit in their importance in certain
22 circumstances. Also as was previously mentioned,
23 the question of validation of some of the tools
24 also potentially deserves a closer look.
25 The discussion yesterday regarding
160
1 multiple alternatives that has been reiterated
2 today reminded me of a an advisory meeting that was
3 held a couple of months ago, which Gary had
4 mentioned earlier. It was a discussion of
5 neuropathic pain and there were multiple
6 possibilities mentioned at the time, one of which I
7 will just reiterate for this group today, those who
8 were not in attendance, because I haven't heard
9 this particular possibility alluded to yet. As a
10 suggestion, it was that one method or one model
11 disease could be studied in replicate and then
12 other models of disease studied not in replicate
13 but in single form, sort of a combination or merge
14 of the two different proposals. At that meeting, I
15 heard mentioned that we might do a replicate
16 analysis of one model and then look at maybe two
17 other models of disease in a single study format to
18 justify a broader claim.
19 Just as an aside, Lee, I would like to
20 compliment you for mentioning yesterday and then
21 highlighting again today the fact that you are
22 proposing a subsequent meeting to examine these
23 kinds of issues more closely, more carefully,
24 perhaps in a more focused way in the presence of
25 the academic community, the presence of the
161
1 regulatory community and perhaps a more meaningful
2 presence from the industrial community as well,
3 with representatives with a more substantial
4 presence at that occasion. That is reassuring
5 certainly to the industry members in the audience
6 today.
7 As an aside also, I think some of the
8 industry people would also like to be reassured, if
9 that were possible, that the arrangements that are
10 already under way and the commitments that have
11 already been made will, in fact, be honored as
12 these new guidance proposals are development and in
13 process, some reassurance there would be
14 appreciated, I know, by some in the room.
15 Also, just as an aside or perhaps as a
16 commentary, some of the industry people have come
17 up to me during the breaks and they are reflecting
18 on the following, and that is the issue of idealism
19 versus realism. There are many physicians and
20 healthcare providers at this table in practice;
21 there are many in the regulatory agency; there are
22 many in the industrial organizations and sponsors
23 that are in the room today and all of us know, as
24 has been mentioned by many of the clinicians at the
25 table, the variability in patients and the
162
1 variability in their circumstances. It is that
2 variability that makes some of these trials so
3 difficult to accomplish and complete in a fashion
4 that would satisfy the proposal that is before us
5 today.
6 That is why I am concerned that the hurdle
7 might be set too high. We just must not lose
8 perspective of the variability in patients and in
9 their situations and in their circumstances which
10 would make it very difficult to hit on all of the
11 targets that have been proposed.
12 DR. FIRESTEIN: Before we move on, I would
13 just like to remind people to please keep their
14 comments to about two minutes, and let's try to
15 answer the specific questions that have been
16 raised. Dr. Max?
17 DR. MAX: Regarding the models, I agree
18 with Dr. McLeskey that people are going to want to
19 do replicate trials in one condition anyway to get
20 the drug on the market. It would make sense to me.
21 I would rather have a broader representation of
22 diseases and I don't need any more replication.
23 So, whether the number would be two and one, plus
24 two additional conditions or three additional
25 conditions, I would recommend that the FDA do a
163
1 careful economic analysis, and if you could get
2 more conditions without killing the wonderful
3 engine of industry, I would make it five trials, if
4 not four trials, and you can figure that out.
5 I think in each condition you should try
6 to either make it relatively homogeneous
7 mechanistically for clinical criteria, or at least
8 allow the information to be there. For instance,
9 if you study cancer pain, mixed cancer pain means
10 very little mechanistically. We should be able to
11 look at bone pain separately and, similarly in back
12 pain, the people with root injury are different
13 from those with central back pain. So, try to use
14 the clinical criteria to allow some mechanistic
15 inferences.
16 Regarding the issue of the three proposed
17 co-primaries, I again disagree with that. I think
18 that pain should be the primary outcome. I agree
19 that a global outcome and function are important
20 things to measure but they should be secondary
21 outcomes and, obviously, if over the pattern of
22 studies globals deteriorate and function
23 deteriorates there is something wrong with the drug
24 and it won't be approved. But I would make pain
25 the only primary. And, I think general chronic
164
1 pain is a great idea as it will drive the science
2 forward.
3 DR. DIONNE: Well, I have very little
4 experience with chronic pain so, presumably, I
5 don't have the basis for an intelligent opinion but
6 that hasn't stopped me before.
7 I just wanted to reiterate the concept of
8 some sort of a data-driven regulatory practice for
9 analgesic drug development in this particular
10 question that might take the form of a
11 meta-analysis of the existing drug classes that are
12 generally accepted for chronic pain, be it
13 tricyclics and NSAIDs, and look back and see if
14 there is enough evidence to support the application
15 of these criteria that are being considered
16 prospectively when we look at the evidence that
17 exists for drugs that have been studied for 50 to
18 100 years. Then, on the basis of that we might
19 determine that the standard is too high, too low,
20 if it doesn't actually apply to drugs that have
21 already been approved, and then make the subjective
22 evaluations that have to be made about the
23 prospective criteria at least on the basis of the
24 data for the drugs that are already out there.
25 DR. WOOLF: I must admit, I am concerned
165
1 about this notion of there being a global chronic
2 pain analgesic in the absence of evidence that such
3 a drug exists. I think that is the key issue.
4 This needs to be evidence based. I am worried that
5 we don't know which trials, whether they be three
6 or five, in which conditions are going to be
7 predictive of whether any drug is going to be
8 effective across a wide range of different chronic
9 pains.
10 So, the issue to me is how happy are we
11 going to be living with an analgesic that has a
12 global pain indication and, yet, is not effective
13 in subcategories or different diseases? If we
14 don't have a basis yet for predicting which of the
15 suitable trials, whether it be low back pain or
16 fibromyalgia or age-related neuropathy, it is pure
17 guess work as to which of these we can select and
18 how many to try to come to an assessment of whether
19 any individual treatment is going to be effective
20 across a wide range of conditions.
21 The other issue that hasn't been discussed
22 yet is in these trials are we looking for
23 placebo-controlled trials or active comparators?
24 If so, since they are going to be so different what
25 would the active comparator be if you are going to
166
1 compare fibromyalgia versus neuropathic pain in the
2 conduct of these trials?
3 MS. MCBRAIR: I too am concerned about a
4 global assessment. It seems early on and what I
5 would really like to see us do is a really good job
6 with each one of the indications or diseases or
7 health problems and be able to give the very best
8 guidance to the practitioners that are using these
9 medications and to the patients. I think we need
10 to focus on that first before we go towards a
11 global assessment.
12 As far as the domains, I think they are
13 all important based on the individual health
14 problem. I do think patients need to be able to
15 function if they are supposed to, and that is the
16 goal of the medication in part. Certainly in
17 rheumatoid arthritis, if we are just covering the
18 pain we may not be addressing the inflammatory
19 process and that needs to be paid attention to as
20 we are looking at these individual situations. But
21 I think the domains are very important to the
22 people that we are trying to serve.
23 DR. WOOD: It is getting late. I agree
24 with much of what has been said before,
25 particularly by Dr. Abramson. I also agree with
167
1 what Dr. McLeskey said, that there are worries
2 about having multiple primary endpoints and merging
3 these into a composite endpoint rather than just
4 having your primary endpoint being the reduction in
5 pain which is, after all, the indication we are
6 looking for.
7 On the other hand, a global indication
8 seems to me to go beyond the science. If you think
9 of other areas, we don't give global indications to
10 improvement in cardiovascular health. We say
11 cholesterol agents do one thing; beta blockers to
12 something else; ACE inhibitors do something else.
13 All of these drugs, in fact, produce mortality but
14 we have a recognition about the specific
15 indications for their use to reduce that mortality
16 and that seems appropriate here; it is just that
17 the science isn't as far advanced.
18 The one thing that has not been discussed
19 that I would want to put on the table is that it
20 seems to me there is an underlying assumption being
21 made up till now that all our studies are going to
22 come out positive in a global indication. What are
23 we going to do with studies that come out
24 negatively? Never mind how many positive studies
25 you need, how many negative studies do you need?
168
1 Does one negative study immediately knock you out
2 of the park? I mean is that it? That you can no
3 longer get a global indication?
4 I would be particularly concerned that
5 that is going to give rise to gaming of the system.
6 You know, I think we can reliably expect that we
7 will hear about all the positive studies. The
8 negative studies may not be presented in this room.
9 So, I think the idea that somehow all the studies
10 will come out positive and really all we are
11 arguing about, as Bernard Shaw said, is the number
12 is unrealistic. Some are going to come out
13 negative. And, I think there is a big danger for
14 industry in going for a global indication because,
15 clearly, if you go for a global indication and one
16 of your studies comes out negative you are dead in
17 terms of a global indication. There is a
18 possibility that one of your competitors may come
19 out with a study that is negative and that is then
20 used to undercut your global indication.
21 So, I think there is a risk in that and I
22 think we should be cautious about extending to
23 indications for which we don't have obvious data to
24 support them.
25 DR. CALLAHAN: Well, I think Dr. Woods
169
1 made a very good point about if there is a negative
2 indication. So, based on that, I would like to say
3 I would like to see two replications of whatever
4 indications, and the numbers I think would depend
5 on sort of the feasibility within the company in
6 terms of how many indications they could look at.
7 Clearly, you need to look at different types of
8 mechanisms within that.
9 In terms of the domains, I think pain
10 should be a primary outcome, not have the
11 co-primary, but I would like to see some sort of
12 disease specific function included, as well as
13 patient global. Then, I very much like the idea of
14 a general health-related quality of life so that
15 they can be compared across conditions.
16 DR. CUSH: There is a benefit to going
17 late; you get to listen to everybody else's ideas
18 and be swayed by them. I will back off. I was
19 very much in favor of this when it was first
20 presented and I would say I am against it.
21 [Laughter]
22 DR. FIRESTEIN: I am going to have to go
23 around the table again now, so be careful!
24 {Laughter]
25 DR. CUSH: I think that there is an issue
170
1 regarding under-treatment of pain, but I think that
2 doesn't rest with the lack of available options or
3 drugs that could be labeled as globally effective
4 therapies. I think that rests more with poor
5 education and poor understanding of pain and pain
6 control. I think if you look at drugs that we
7 might call sort of global drugs, widely used drugs,
8 broad-spectrum antibiotics, while they may have
9 been helpful there has also been a certain degree
10 of misuse, and the problems that that may have
11 arisen from that I don't think were anticipated.
12 When we look at our arthritis drugs, we
13 have drugs like methotrexate and disease-modifying
14 drugs. They tend to be used globally, sometimes
15 outside of indications because we don't have
16 options. Sometimes that is done because we
17 understand the mechanism of disease. Sometimes it
18 is done quite blindly and quite stupidly, and with
19 no apparent effect and maybe with great expense or
20 maybe toxicity. I think that there are drugs that
21 are out there that are being used in this manner
22 currently, drugs such as the COX-2's and narcotics,
23 are basic globally used pain medicines. Currently
24 they are used in a way that basically forces the
25 physician to be intelligent and understand the
171
1 mechanisms of disease and what is going on with the
2 patient, and also act as an advocate on behalf of
3 the patient to go for those indications and write
4 letters to explain why this is indicated.
5 So, you know, would a global indication
6 actually help a payer, an approver of drugs that
7 they may not be indicated for? So, would they
8 actually approve the use of a new, novel pain
9 medicine for phantom limb pain, acute gout or
10 visceral pain associated with losing to the
11 Yankees? I don't know.
12 [Laughter]
13 I still think it forces me to have to
14 still write those letters to get these drugs
15 approved, and for this reason I would say that we
16 should not have this indication.
17 I will close by just saying I think we
18 have an issue of nomenclature here that was raised
19 yesterday by Dr. Ashburn. The whole use of words
20 "acute" and "chronic" are a little bit
21 disconcerting and I think we should try to maybe
22 redefine the terms we use and maybe go for things
23 such as short-term therapy or long-term therapy.
24 In this instance, general global pain indication is
25 a bit too obtuse clinically and unrestrictive to be
172
1 useful. Thank you.
2 DR. SHERRER: I am last but I didn't
3 change my mind. So, some of us can stay steady.
4 While it is true that we do, in fact, use
5 medications that are on the market with restrictive
6 indications broadly, nevertheless, as a clinician,
7 I think it would be very useful to me in
8 prescribing to know that a drug has utility across
9 different types of pain. If the studies were
10 useful and really are showing me that, for
11 instance, if you do osteoarthritis and low back as
12 two of your models I am not so sure that you are
13 looking at different pain. On the other hand, if
14 you look at cancer bone pain and you look at
15 diabetic neuropathy and you look at OA, you
16 probably are looking at different pains and it
17 would be very useful for me to know that that has
18 been demonstrated.
19 In terms of looking at the domains, I am
20 one of those who believes that we need to look at
21 the total impact of the drug as an outcome. So, I
22 would favor looking at least at three, if not four
23 of them. I think pain is useful but the total
24 impact of a drug is even more useful to my
25 patients. In fact, that is why some won't take
173
1 certain pain medications, because of the side
2 effects, because of their effect on quality of
3 life. So, I would use several of those, and most
4 important to me would be pain, would be patient
5 global and some appropriate assessment for the
6 particular disease of function or quality of life.
7 One thing I haven't heard that I would
8 like to bring up, and maybe it would be a
9 secondary, is steroid sparing because I think that
10 in certain chronic pain disorders where steroids
11 are an important part--I said steroid sparing,
12 opioid sparing--many patients are very concerned
13 about opioids and so are we, and if a drug spares
14 opioids, that would be very important to me.
15 DR. FIRESTEIN: We are done. We have gone
16 all the way around the table. So, we will break
17 for lunch and we will reconvene at 12:55, which
18 means we will start at 1:00.
19 [Whereupon, at 12:05 p.m., the proceedings
20 were recessed for lunch, to reconvene at 1:00 p.m.]
174
1 A F T E R N O O N P R O C E E D I N G S
2 DR. FIRESTEIN: I am happy to introduce
3 Dr. Vibeke Strand, who is going to talk about
4 responder index, a model.
5 Responder Index, a Model
6 DR. STRAND: Thank you, Gary. We have
7 been more or less talking around this topic for the
8 last day and a half, and perhaps we should have
9 started sooner with this discussion.
10 [Slide]
11 What I would like to do is basically
12 present to you a discussion that was started at the
13 last NIH-FDA meeting on pain. Just to point out
14 something that we have talked about before,
15 responder analyses have face and content validity.
16 They do allow the assessment of multiple domains.
17 They probably could better help us categorize
18 analgesics.
19 They should also help facilitate
20 comparison of efficacy across products and disease
21 populations and indications. I think in analgesia,
22 as in rheumatology, most of our patient populations
23 are quite heterogeneous and this would help
24 considerably.
25 This might or might not lead to a tiered
175
1 approach in label indications as has been done in
2 rheumatoid arthritis but really has not yet been
3 done otherwise. The precedent, as we have talked
4 about previously, is the ACR responder criteria in
5 rheumatoid arthritis.
6 [Slide]
7 Jim Witter pointed this out to you this
8 morning. it is a model for other responder
9 analyses. One could say that the two criteria
10 here, which are tender and swollen joint count,
11 could be required in a responder analysis in pain,
12 for instance, whatever assessment of pain could be
13 required and perhaps also the patient global
14 assessment could be required. The others could be
15 included.
16 One of the things we do know is that it is
17 probably too stringent to require all components of
18 a responder analysis to be improved. It is
19 possible to choose the majority of them to be
20 improved. It is also possible to indicate that the
21 remaining ones should not be deteriorated.
22 If we want to talk about a definition of
23 no deterioration, however, we have to allow that
24 statistical definition to account for test/retest
25 variability, which we have alluded to before in our
176
1 discussions around changes in visual analog scales.
2 [Slide]
3 The strength of the rheumatoid arthritis
4 guidance document is that it has had a proven track
5 record and since its inception we now have six
6 products approved for the treatment of rheumatoid
7 arthritis, some of them just for the signs and
8 symptoms, as in the COX-2 products, but many of
9 them now for improvement in signs and symptoms in
10 either 6 or 12 months and then inhibition of
11 radiographic progression at 12 months, and
12 subsequently improvement in physical function
13 without deterioration in health-related quality of
14 life over 2 to 5 years. In this case it has been
15 over 24 months.
16 These outcomes have all been achieved in
17 single protocols using prespecified outcome
18 criteria, whereby the first outcome criterion must
19 be satisfied statistically significantly, p less
20 than 0.05. Then one may look at the subsequent, in
21 sequence, criteria, provided each one remains
22 statistically significant without taking a p value
23 correction. That is a very valuable way to look at
24 multiple different aspects of a disease and how it
25 affects the disease population.
177
1 [Slide]
2 When we had this breakout session at the
3 workshop, in May, the definition for the
4 workshop--and I am not saying that is a definition
5 we have been working on today, but the definition
6 for chronic pain was randomized, controlled trials
7 of at least three months duration in pain of at
8 least three months duration, regardless of the
9 underlying cause. That was simply taken as a
10 definition so we could have the discussion we were
11 going to have.
12 We agreed in that discussion that we would
13 not specify specifically different diseases. We
14 agreed that maybe there might be some differences
15 specifically for chronic cancer pain, but for the
16 purposes of the discussion we would not
17 distinguish.
18 [Slide]
19 We were considering musculoskeletal
20 indications such as rheumatoid arthritis,
21 osteoarthritis and low back pain, as we have talked
22 about in the last two days, also fibromyalgia,
23 neuropathic pain, the examples being diabetic
24 neuropathy, post herpetic neuralgia, trigeminal
25 neuropathy. For cancer pain, we agreed that it
178
1 wouldn't necessarily be for a three-month duration
2 in terms of trial and that we would be thinking
3 about rapidly progressive disease and adjust
4 intervention as the disease progresses which is, of
5 course, a very important thing around cancer pain.
6 [Slide]
7 We agreed to select the domains regardless
8 of the clinical indication; that we would consider
9 the available instruments and whether or not they
10 were validated and whether or not they had been
11 validated in pain trials; just that they had been
12 used in previous randomized, controlled trials but
13 not necessarily in pain; and whether they were
14 disease specific or generic was sufficient.
15 The point really was that the outcome
16 measures in rheumatology clinical trials, the
17 OMERACT international consensus process has
18 actually helped to define the ACR responder
19 criteria, and is helping to define responder
20 criteria in osteoarthritis, but the first decision
21 is around the domains to be used, not the specific
22 instruments, and that there is some flexibility
23 around which instruments might be utilized to
24 satisfy each of the domains.
25 [Slide]
179
1 We did believe, however, that the strength
2 of choices in terms of domains was based on
3 multiple available instruments and our own prior
4 clinical experience. So, the choices, as they were
5 thrown out and written up, were pain and we talked
6 a lot about the multiple different measures of pain
7 that probably should be important to be included in
8 a given trial under a single domain, including the
9 patient global assessment; including the assessment
10 of rescue medications; and time to treatment
11 failure--all of these which we talked about this
12 morning.
13 Suffering was suggested as a domain, as
14 was pain relief, a disease specific measure of
15 improvement and/or physical function and/or
16 health-related quality of life was proposed. So
17 was health-related quality of life, and we have
18 been throwing around the term quality of life. I
19 think it is important that we specifically mention
20 that it should be health-related quality of life in
21 all the way health affects you. Because,
22 certainly, political circumstances, economic
23 circumstances and the presence or absence of food
24 and money are not part of health-related quality of
25 life but certainly are part of quality of life.
180
1 Patient global assessment, adverse events and
2 specifically how they are perceived by the patient
3 which is something we are not very good at in
4 clinical trials; we usually trust the physician to
5 report those adverse events and often not with very
6 much input from the patient, other than that the
7 complaint has been offered. Damage, whether it is
8 due to the disease or its treatment, and
9 specifically indicating that it is irreversible,
10 and economics.
11 [Slide]
12 After a relatively brief series of
13 discussions, we came up with the final vote the
14 first time when everyone was allowed to vote on
15 basically three parameters: Unanimous decision for
16 pain; an almost unanimous decision for a disease
17 specific or a disease relevant measure. We have
18 been talking a lot about physical function but, as
19 I said to you before, I think it can be basically
20 perceived as a disease relevant or specific measure
21 of either function or health-related question.
22 Health-related quality of life as a generic measure
23 was an almost unanimous decision. Patient global
24 and adverse events followed.
25 So, this was felt to recommend a minimum
181
1 core set of required domains, and that other ones
2 could certainly be added but if we were to speak
3 about trying to do a responder analysis, these
4 should be the components to be considered at a
5 minimum.
6 [Slide]
7 We have talked a lot about defining
8 improvement in pain, but I think the point we are
9 all trying to get at is defining improvement
10 multidimensionally. We know that patients
11 experience pain and they report pain, but they
12 report it specifically as they feel on the day they
13 are reporting it. So, if they are forward filling
14 their diaries, it is based on how they are feeling
15 that day. If they are back filling, it is also
16 based on how they are feeling that day.
17 One of the important things too is that
18 their expectations of what they can do and what
19 they should be able to day change according to how
20 their pain is. So, if they have already had
21 significant pain relief their expectations have
22 changed and become even greater than they were when
23 they, for instance, first entered the study and
24 were suffering considerable pain.
25 What we are trying to do, obviously, is
182
1 separate the experience of pain from functional
2 impairment and disability which may or may not
3 occur because of the pain or follow the pain. We
4 want to separate physical impairment from
5 disability. It is important, I think, to use
6 individual responder analyses because it allows us
7 to define responder, non-responder. We don't have
8 to impute data. All cause dropouts before the
9 endpoint are then considered non-responders.
10 Therefore, from a statistical analysis it can be a
11 more robust analysis. I think it is important that
12 we use both disease specific or disease relevant
13 measures as well as generic measures.
14 [Slide]
15 Something to quickly point out is that
16 disability is really in the eyes of the beholder.
17 It is, of course, age and gender appropriate. It
18 is important and pertinent to the work, the family
19 and the social setting. But, in fact, someone who
20 has had cerebral palsy since birth and is
21 wheelchair-bound may not perceive themselves as
22 being disabled even though we would certainly
23 consider them to be far more than just physically
24 impaired.
25 The other part of it is that impairment
183
1 may be due to pain or it may be due to structural
2 alterations, and functional limitations are
3 certainly something that we can measure. There are
4 arguments about disease specific or disease
5 relevant measures of physical function and how
6 accurate they are in that those of us who are
7 rheumatologists often note that our fibromyalgia
8 patients are far more severely impaired than our
9 rheumatoid arthritis patients. But, by and large,
10 if we can choose the right types of instruments we
11 can usually find some type of a valid report that
12 is consistent with the other self-reports that the
13 patient may offer.
14 [Slide]
15 One of the other things about a global
16 assessment is that it is probably much more
17 important to ask the patient in all the ways that
18 your pain is affecting you, including its
19 treatment--how are you doing today? When we talk
20 about visual analog scales for patient global
21 assessments, we always talk about how are you doing
22 today, this moment? The other part of it here is
23 to make it a global assessment and to include sort
24 of the risk as well as the benefit as an important
25 thing in terms of the patient assessment of the
184
1 pain treatment.
2 Now, a transition question can probably be
3 equally sensitive, in other words, how are you
4 compared to when you first started taking this
5 medication? That may well get to the same point.
6 The other point that is quite useful is
7 that health utilities which are used for economic
8 measures are single reports sometimes, questions or
9 several questions around how patients are doing in
10 terms of what their perception of perfect health
11 would be. A health utilities index or the EQ5D can
12 be given. It is a simple questionnaire that the
13 patients can fill out. Or, one can ask the patient
14 to report, by a feeling thermometer, how they are
15 doing in terms of perfect health and death. That
16 looks very much like a visual analog scale
17 vertically.
18 [Slide]
19 We have talked a lot about minimum
20 clinically important differences. We consider them
21 to represent changes which are perceptible to
22 patients and are considered clinically important
23 and meaningful. When they were first started in
24 the OMERACT process we used patient query as well
25 as a delphi technique. Then they were demonstrated
185
1 to be consistent with patient global assessments of
2 improvement or patient global assessments of how
3 they were doing.
4 In fact, when we determined the proportion
5 of patients with clinically meaningful improvement
6 or clinically important improvement, this gives us
7 a much more interpretable result than, in fact,
8 trying to say, okay, this many patients had 50
9 percent improvement in pain or this many patients
10 had 30 percent improvement in pain.
11 [Slide]
12 If we think about this, we have now
13 noticed that changes in disease specific or
14 relevant measures of function and health-related
15 quality of life that have been statistically
16 related to much or very much improvement in patient
17 global assessments, either by visual analog scale
18 or Likert have given us very consistent values
19 across OA, RA and fibromyalgia, and I will show you
20 that briefly.
21 [Slide]
22 Briefly, measures of chronic pain include
23 a lot of different things. There is the brief pain
24 inventory, the McGill pain questionnaire, all of
25 these others. Perhaps one of the more important
186
1 new ones is the treatment outcomes and pain survey
2 which was developed as an add-on to the SF-36 and
3 has been shown to be very useful in cancer pain, as
4 well as some other non-malignant settings of pain,
5 chronic pain with multidimensional therapy.
6 [Slide]
7 The faces rating scale we have talked
8 about before. We talked about using a visual
9 analog scale that is not anchored. This one
10 actually combines a Likert scale of more or less 7
11 with a visual analog scale of 10 and is sort of the
12 example of what not to do at the same time to get
13 sensitivity and specificity, which is why I chose
14 to show this slide because I, myself, would be very
15 confused about which face to combine with which
16 number.
17 [Slide]
18 Talking about MCID, one of the nice papers
19 published by Dr. Farrar, sitting at the table, is
20 looking at the pain intensity numerical rating
21 scale and comparing that to very much improved in
22 patient global assessment.
23 These are 10 placebo, randomized control
24 trials of Pregabalin, which is not yet approved,
25 but this has been published in Pain 2000 for
187
1 diabetic neuropathy, low back pain, fibromyalgia
2 and OA. So trials across different indications of
3 chronic pain have shown that the relationship of
4 much and very much improved in PGIC and pain
5 intensity by numerical rating scale is very
6 consistent with reduction of 30 percent or two
7 points in the pain intensity scale.
8 This is really interesting given the wide
9 variety of disease states here, and this is
10 regardless of the baseline pain scores in these
11 patients. So, a robust MCID definition.
12 [Slide]
13 If we look at other measures of physical
14 function and health-related quality of life in
15 chronic pain, I just want to remind you again that
16 the top survey here is meant to look at changes in
17 health-related quality of life in individuals over
18 time, which is different from the generic measure
19 of health-related quality of life, the SF-36, which
20 I will come back to in a minute, and one other
21 measure that is an HRQOL measure in pain is the MPI
22 which specifically looks at psychosocial role
23 functioning but omits work-related activity.
24 Finally, cancer-related health-related quality of
25 life has been looked at a lot on the BPI, the brief
188
1 pain inventory, but that has not been validated in
2 non-malignant pain.
3 [Slide]
4 Generic health-related quality of life
5 measures go back as far as the sickness impact
6 profile which is, in fact, considered not to be a
7 very popular instrument because it implies to the
8 patient that they are sick.
9 The Nottingham health profile is also an
10 older measure of HRQOL and not particularly
11 popular. A very popular one is the SF-36 which is
12 expanded over the SF-12. It is designed to measure
13 health-related quality of life in large groups and
14 across different disease states. It has problems
15 if it is being used as a single measure of HRQOL in
16 pain states or in arthritis states because there is
17 a limited assessment of upper extremity function,
18 as well as upper extremity pain and facial pain,
19 and does not differentiate well between low back
20 pain and upper body pain.
21 The WHOQOL is a new instrument, but with
22 100 questions it has fallen out of favor. There
23 are some shorter version. The EQ5D is widely used
24 in Europe.
25 [Slide]
189
1 Disease specific measures of physical
2 function and/or health-related quality of life
3 include all of these. We have called them disease
4 specific. People like Jim Freis, who developed the
5 health assessment questionnaire, prefers not to
6 call it disease specific because he believes it can
7 be used across many disease states as well as
8 aging, which is not a state of disease, as he wants
9 to remind me. So, I have chosen to also call these
10 disease relevant measures.
11 Clearly, the WOMAC is something that is a
12 very good one for osteoarthritis of a knee or a
13 hip. There are others, as well as some for the
14 hand which are being developed. We talked about
15 Roland-Morris and Oswestry. There are some for
16 geriatrics and, of course, a variety of ones for
17 cancer.
18 [Slide]
19 What I would like to do very quickly is
20 just show you some examples of how these measures
21 interrelate in rheumatoid arthritis, osteoarthritis
22 and fibromyalgia.
23 [Slide]
24 So, if we look at rheumatoid arthritis, we
25 talk about the health assessment questionnaire
190
1 which has now become widely used in randomized
2 controlled trials in rheumatoid arthritis. It is a
3 measure of physical function with 20 questions. It
4 also accounts for when patients use aids or devices
5 to perform these activities.
6 [Slide]
7 The SF-36, as I mentioned to you, is
8 validated and widely used. It has been validated
9 across multiple cultures, many disease states.
10 There exist gender and age specific norms for
11 multiple populations, both in the U.S., Canada and
12 northern Europe and other countries. Then, it has
13 eight domains as well as a physical component score
14 and a mental component score. It has been shown in
15 RCTs to show change in as short a time as four
16 weeks, probably sooner than that.
17 [Slide]
18 The physical domains are physical function
19 role, physical body pain, general health. They are
20 combined positively into the physical component
21 score which then negatively also weights the mental
22 domains of vitality, social function, emotional and
23 mental health. So, positive changes here are
24 weighted positively here against the positive
25 changes in these domains, which are negatively
191
1 weighted for the mental component score. The
2 mental and physical component scores are based on
3 normative data only to a total of 50. Therefore,
4 they can show less change. And, if you are looking
5 at a disease like rheumatoid arthritis where the
6 predominant change is in the physical component
7 domains, then one is not going to be seeing much
8 improvement in mental domains because they are
9 weighed against by the improvements in these.
10 [Slide]
11 What we have learned from the various
12 trials is MCID for the HAQ disability index is a
13 score 0.22 improvement. For the SF-36 it is about
14 5 to 10 points in domains. For the physical and
15 mental component scores, 2.5 to 5 points.
16 [Slide]
17 So, if I look very quickly across some
18 clinical trials in rheumatoid arthritis you can
19 see, with the leflunomide Phase III trials across
20 all three studies, with methotrexate and
21 sulfasalazine the mean improvement over two years
22 exceeds MCID almost to twice in all treatment
23 groups.
24 [Slide]
25 If we look at the ATTRACT study, and this
192
1 is HAQ disability index over two years, again we
2 see that in the placebo group it does not quite
3 reach MCID and is about twice that in all of the
4 active treatment groups.
5 [Slide]
6 Similar types of improvements in the ERA
7 trials with Etanercept versus methotrexate.
8 [Slide]
9 If we go back to look at the U.S. study
10 with leflunomide and methotrexate, which was the
11 first to show that the SF-36 was sensitive to
12 change in rheumatoid arthritis, you can see that
13 based against age and gender matched U.S. norms the
14 patient population had significant decrements in
15 all domains of healthcare quality of life, but
16 particularly physical function, role physical,
17 bodily pain and vitality. As we know, patients
18 perceive their health-related quality of life
19 differently, and one can see the changes here in
20 the active groups actually are within MCID for
21 almost every domain, with some deterioration in
22 placebo.
23 [Slide]
24 If one then goes forward, we see that
25 these are the baselines for the treatment groups
193
1 and these are the age and gender matched norms,
2 then treatment with leflunomide and methotrexate,
3 in fact, just about bring health-related quality of
4 life up to a normative population level. That is
5 probably a very meaningful change and it certainly
6 does equal MCID in many of these eight domains.
7 [Slide]
8 There is similar improvement infliximab in
9 the ATTRACT trial. These are the two of the
10 physical domains. If we look at the PCS and the
11 MCS we see that there is very significant decrement
12 in the physical component score at baseline, almost
13 two standard deviations from the U.S. norm, and
14 treatment over one and two years brings it to
15 within one standard deviation of the U.S. norm. As
16 we might expect, the MCS was not that different
17 from expected, and it could not show a great deal
18 of improvement based on the large amount of
19 improvement in the physical domains. Nonetheless,
20 improvement is shown.
21 [Slide]
22 This is the median improvement in PCS
23 score with the ATTRACT trial showing the same type
24 of a picture, with placebo showing not much
25 improvement.
194
1 [Slide]
2 This is the early RA trial, again showing
3 baseline for the PCS, about two standard deviations
4 below the U.S. norm, and improvement to
5 approximately one standard deviation from the U.S.
6 norm with treatment.
7 [Slide]
8 So, I think you can see from this that
9 basically improvements in HAQ disability index, in
10 other words the disease relevant measure of
11 physical function and the generic measure of
12 health-related quality of life appear to be very
13 clinically meaningful, and that there are
14 consistent values for MCID across these
15 instruments. We are showing that improvement in a
16 disease relevant measure is highly correlated with
17 a generic instrument, and the generic instrument is
18 useful because we can compare it across different
19 disease states for an economic basis, but also to
20 try and understand improvement, for instance as we
21 might when we are looking at chronic pain
22 indications.
23 [Slide]
24 Quickly, lets look at osteoarthritis. The
25 WOMAC is the disease specific measure in OA of the
195
1 knee and hip. It reflects physical activities that
2 are most affected by the osteoarthritis. It is
3 composed of pain, five questions on joint
4 stiffness; two questions on physical function which
5 dominates the instrument of 17 questions out of a
6 total of 24, and is scored either by a zero to 4
7 Likert or a zero to 10 VAS scale for each question.
8 [Slide]
9 So, what we have found out looking at the
10 COX-2 trials with both celecoxib and roficoxib is
11 that basically, using a Likert scale for the
12 composite total WOMAC score, MCID was about 10
13 points and was different according to the domains
14 because they had more or less questions. If one
15 uses the VAS scale for all of the questions, then
16 we see very consistent MCID for each of the domains
17 of about approximately 10.
18 [Slide]
19 This is what this looks like in the
20 composite scores of WOMAC in clinical trials of
21 celecoxib versus placebo and the active comparator,
22 naproxen. Here is MCID.
23 [Slide]
24 If we look at it for rofecoxib using the
25 primary outcome question in the physical function
196
1 subscale we see again that improvement is evident
2 and exceeds MCID considerably.
3 [Slide]
4 If we look at the improvement in the SF-36
5 with rofecoxib and we compare it to age differences
6 in the U.S. population, we can see that there is
7 considerable improvement in the mental domains as
8 well as the physical domains, but the largest
9 improvement is in role physical.
10 [Slide]
11 Similarly, if we look at the changes with
12 celecoxib in the SF-36 in the trials that I showed
13 you previously, you can again see that MCID is
14 reached in many of the domains, particularly the
15 physical ones.
16 [Slide]
17 This actually translates again towards
18 improvement that approaches the U.S. norm. This is
19 the U.S. normative population and these are the
20 final scores with the different doses of celecoxib
21 and naproxen and placebo.
22 [Slide]
23 So, again, we see clinically meaningful
24 improvements. We see that the MCIDs are consistent
25 across agents and patient populations in this
197
1 disease, and that improvement in the WOMAC
2 correlates with the generic HRQOL SF-36 measure.
3 [Slide]
4 I don't have outcomes for fibromyalgia,
5 but I do have interesting consistent relationship
6 at baseline between pain, sleep disturbance and
7 fatigue. These are all patient reported and they
8 are highly correlated either by a pain diary or a
9 sleep quality diary or multidimensional assessment
10 of fatigue, a well-known fatigue instrument. And,
11 this is whether it is done by a numerical rating
12 scale that is ostensibly recorded daily in the
13 diary or a visual analog scale that is done at the
14 office visit weekly. It has been shown that the
15 high baseline scores indicate impaired sleep.
16 Significant fatigue, we know that our fibromyalgia
17 patients think of themselves as being very
18 physically impaired, and these correlate with low
19 scores in SF-36, particularly role physical, bodily
20 pain and vitality domains; poor sleep quality by
21 the MOSA sleep, high fatigue and also more anxiety
22 than really depression.
23 [Slide]
24 In terms of cancer, there are a lot of
25 different instruments that would be useful in
198
1 trials of cancer pain, and they can be the FACT-G
2 or FACT that is a P for prostate or any one of the
3 cancers that you want to look at. The same for
4 LASAs which can also be done for symptoms of
5 chemotherapy as well as for symptoms for cancer or
6 pain. The same kind of thing for the FLIC.
7 Basically, there are all these different
8 instruments that can be used and, again as I
9 mentioned to you before, the TOPS has been
10 developed and validated in cancer pain, among
11 others.
12 [Slide]
13 Since the TOPS was defined as an extension
14 of the SF-36 it has been a very useful instrument
15 and it really does show change in individual
16 patients over time.
17 [Slide]
18 So, the appropriate domains, based on what
19 we discussed at that particular breakout session
20 and as a recommendation to this group, would be
21 that pain would be included as a domain. There are
22 many instruments. We have talked about looking at
23 different ways of assessing pain. Perhaps we can
24 get away from some of our old visual analog scales
25 and Face scales.
199
1 A disease specific or disease relevant
2 measure of health-related quality of life and the
3 ways that the disease affects you in your day to
4 day activities could be used, or one could use the
5 TOPS which is much more generic. When it is
6 relevant to whatever the disease is, other measures
7 could be looked at. They do not necessarily have
8 to be included in the responder analysis.
9 I think you can see that the
10 health-related quality of life measure SF-36 as a
11 generic measure has turned out to be very useful
12 and sensitive to change across a large number of
13 types of diseases; and some way of asking the
14 patient how they are doing in terms of risk/benefit
15 in terms of the treatment as well as the pain; and
16 finally adverse events, which we haven't talked
17 about, might be subsumed under this global
18 assessment if it does include the treatment as well
19 as the pain.
20 [Slide]
21 Certainly for acute pain we probably don't
22 need a measure of health-related quality of life,
23 as we have discussed, and certainly we can talk
24 about all of these. We do want to remember time to
25 treatment failure and rescue medications as being
200
1 part of something that needs to be assessed in the
2 pain domain.
3 [Slide]
4 When we go to subacute pain or pain of two
5 to five days, or whatever the definition is that is
6 less than chronic pain but more than one day of
7 pain, it would appear that these different domains
8 would be equally relevant. We can show changes in
9 SF-36 over a very short period of time. Again, it
10 might be useful to use the TOPS or to use a disease
11 relevant measure.
12 [Slide]
13 In fact, again Dr. Farrar has published a
14 very nice paper on cancer-related breakthrough
15 pain, acute pain. This was in a study of oral
16 transmucosal fentanyl citrate, which ultimately was
17 not approved. But these were 130 patients who were
18 naive to the study drug, many episodes of pain, and
19 the differences in pain scores between the episodes
20 which did and did not yield adequate pain relief.
21 Again we see MCIDs for pain intensity difference
22 and maximum total pain relief of about 33 percent.
23 Again, the same kinds of changes in terms of
24 absolute pain relief and sum of pain intensity
25 differences of 205 points in a Likert scale, which
201
1 are then consistent with what we were looking at in
2 the other measures of chronic pain.
3 [Slide]
4 So, in my conclusions, a responder
5 analysis for pain randomized controlled trials
6 would make sense. I would never suggest that we do
7 it in the absence of data. I would never suggest
8 that we prospectively put it together and then set
9 out to validate it but that, instead, it be
10 developed over time using perhaps a particular
11 product and validating it from Phase II data into
12 Phase III final randomized, controlled trials. Or,
13 perhaps we would be able to work on it as a
14 concerted effort with a bit of help from
15 meta-analyses. Unfortunately, most of these
16 domains have not actually been assessed even in
17 recent clinical trials of pain relievers and that
18 will limit a lot of what we can do post hoc. I
19 think this represents minimum number of required
20 domains. We certainly want to use validated
21 instruments. As I have mentioned before, several
22 different components have to be included.
23 As with other responder analyses, it could
24 be required that the majority of them showed
25 improvement but not that all would be required to
202
1 show improvement in the domains we are talking
2 about here. As Dr. Simon had proposed, three of
3 those five would be improved. It could be added
4 that there should not be deterioration in the other
5 two, or that could be omitted. The degree of
6 improvement proposed could be based on MCID values
7 at least for those instruments that we have.
8 When we know that these different domains
9 are not closely correlated in responses, then we
10 know that we have both a very robust clinical
11 response when we get a responder analysis that is
12 positive, and that we have additive statistical
13 power which allows our sample sizes to decrease
14 considerably. That certainly has been true in
15 rheumatoid arthritis and, hopefully, it will be
16 true in some of these chronic pain studies.
17 [Slide]
18 At any rate, I would just say that there
19 is a rating scale in the "San Francisco Chronicle"
20 for movies, and so on, which has to do with the
21 little man and whether he is falling out of his
22 chair or whether he is asleep. If he likes the
23 movie he is jumping up and down, and if he hates
24 the movie he is asleep. Perhaps some day, after we
25 make all these evidence-based decisions, we can
203
1 develop a universal quality of life scale. Thank
2 you very much.
3 DR. FIRESTEIN: Thank you very much,
4 Vibeke. Does anybody have any specific questions
5 about the instruments? Steve?
6 DR. ABRAMSON: Vibeke, a question that I
7 guess that you have dealt with and the FDA has
8 begun to think about, but have you lumped together
9 diseases like RA and OA and these other pain
10 syndromes, particularly in RA where we have
11 mechanism-based therapies? So, if you treat with
12 steroids or anti-TNF blockers you get a very nice
13 response on pain. Obviously, we are going to need
14 to sort out when we look at diseases like RA what
15 it is that we are measuring.
16 I guess the related question to be
17 grappled with is that we will have pain indications
18 for OA that are separate from indications for the
19 treatment of OA. I think those are two separate
20 questions, but I guess I am mostly curious about
21 how rheumatoid arthritis would be included in these
22 kinds of studies.
23 DR. STRAND: Well, for brevity I did not
24 include the COX-2 data in rheumatoid arthritis but,
25 in fact, you can show very nice improvements by
204
1 ACR-20 responder analyses and also by SF-36 and HAQ
2 even with a medication that we would consider to be
3 largely a pain reliever.
4 Now, the magnitude of those improvements
5 is not as great as we see with our DMARDs or our
6 biologics but, in fact, most of the time patients
7 are on background therapy with those agents. So,
8 there is still some incremental improvement when
9 those patients have been taken off whatever
10 anti-inflammatory they were taking and they flared,
11 and then they would go into these trials.
12 I think the other part of that is that
13 when you see some of the improvement with the
14 COX-2s in terms of morning stiffness, which we
15 consider to be not a good component of responder
16 analysis because it wasn't sensitive to change, and
17 you see that the morning stiffness can be
18 completely abrogated in some of these clinical
19 trials you realize that we are again still looking
20 at multiple dimensions of a multidimensional
21 disease, and that the treatment of the
22 inflammation, either by an ostensibly mild agent or
23 even a much more significant agent, really impacts
24 many of these domains. So, there is a lot of
25 physical function and there is a lot of
205
1 health-related quality of life that is clearly
2 impacted by pain. Does that get at the question
3 you were asking?
4 DR. ABRAMSON: Yes, I think that is part
5 of it. I guess the other is if a drug has an
6 indication for OA, is it possible then to mine the
7 data on the pain aspects of the studies that allow
8 approval for OA and have a separate pain
9 indication? We need to cross over what we are
10 looking at in some of these clinical trials.
11 DR. STRAND: Well, I would certainly think
12 that we could try that. I mean, I think that it
13 has to do with the risk/benefit profile of the
14 product as to whether you would even argue that a
15 DMARD might be a pain reliever or might be usable
16 just in RA but, say, OA. I think we could consider
17 this the same type of thing and, clearly, when you
18 look at the data in OA that I showed and the data
19 that we just talked about in RA with the COX-2s and
20 the data with the COX-2s in various other pain
21 models, that is true.
22 The other side of it is I can't imagine
23 that if we affect structure significantly either in
24 OA or RA without a lot of other symptom
25 modification that we won't ultimately still see
206
1 improvement by patient-reported measures.
2 DR. FIRESTEIN: One of the questions that
3 comes up, and you addressed here to an extent, is
4 whether these domains must not be closely
5 correlated if they are going to be useful. This
6 has come up again and again with regard to
7 especially the arthritis clinical trials where the
8 ACR-20 or even pain measurements are very
9 closely--you are going to say no? Well, in early
10 RA the HAQ scores do correlate reasonably well with
11 pain. In late RA it is primarily with erosions and
12 joint damage.
13 So, the issue is whether or not these are
14 independent variables or whether they are dependent
15 variables, and how one takes that into account when
16 trying to set up an instrument for measuring this.
17 DR. STRAND: Our definition is different
18 around close correlations. The ACR criteria, with
19 the exception of tender and swollen joint counts,
20 correlate with each other no better than an 0.4.
21 In all of the x-ray trials physical function HAQ,
22 sed rate, CRP, ACR-20 have not correlated with
23 x-ray any better than an 0.4 and usually less.
24 Even the tender and swollen joint counts that are
25 considered to be obviously appropriately changing
207
1 together have a correlation of no better than
2 around 0.7. So, I will defer to the statisticians
3 around that, but that is one of the reasons why we
4 have been able to decrease the sample sizes.
5 In terms of x-ray, we don't actually see
6 correlations with HAQ scores until we are looking
7 at very long disease duration, and although HAQ
8 scores correlate very high in early disease
9 patients, they go down very, very quickly when they
10 get their first DMARD. So, I think we are just
11 differing about the correlation coefficients.
12 DR. FARRAR: I want to address Dr.
13 Abramson's question from the following perspective,
14 which is that I think that one of our statistician
15 colleagues indicated that looking at the outliers
16 can be very informative. From that perspective,
17 for a broken femur and intramedullary rod is a pain
18 medicine with a very slow onset but a very
19 long-acting action.
20 I think your point though is well taken in
21 that when we are treating a disease as a primary
22 disease we clearly affect all of the symptoms
23 associated with that disease and, hopefully, with
24 Clifford's help and Mitchell's and others, we will
25 be able to look at it from a mechanistic
208
1 perspective and know whether we are treating the
2 disease or the pain process primarily. However, I
3 think it would be very reasonable to say that a
4 treatment for RA that improves the disease could
5 say in its labeling that it treats pain. However,
6 it would not then end up meeting the criteria for
7 treatment of a broken bone or treatment of other
8 things where we would also want to be able to use
9 it.
10 So, I think as long as we restrict and are
11 careful about how we label what the drug is
12 treating and, to the extent that we know, how it
13 improves the overall symptomatology, then we won't
14 have that problem.
15 Discussion of Point # 4
16 DR. FIRESTEIN: One of the items that we
17 were asked to comment on is item number 4, to
18 discuss the domains and responder indices, and
19 address whether they adequately address the issues
20 of efficacy or safety. I would open that up for
21 the discussion. Obviously, Vibeke covered quite a
22 bit of this already. Are there other comments?
23 DR. KATZ: Just a question. I wonder what
24 people think the best way is to measure side
25 effects in these trials and how important that is.
209
1 DR. FIRESTEIN: Any comments? Yes,
2 Vibeke?
3 DR. STRAND: Well, we have our adverse
4 event reporting system which I do not want to
5 change, other than to improve it. But I think we
6 really do need to have some type of a patient
7 assessment, reported assessment of both the
8 positives and negatives of whatever intervention
9 they have undergone and they can weigh that.
10 Perhaps we do it best with a utility measure, but I
11 certainly see subsuming adverse events into that
12 because then it is in the eye of the beholder or
13 the experiencer how these adverse events truly
14 impact and should be weighed in their therapy.
15 DR. FARRAR: I think there are a couple of
16 things I would like to say about that. One is that
17 one person's side effect is another person's
18 effect. Just to make the point, if a drug is very
19 sedating it may be a very good sleeping medicine
20 and, you know, one can even look at nausea and
21 vomiting and say for ipecac that is the effect that
22 we are looking for.
23 So, the point is that the really isn't a
24 difference in looking at side effects and effects.
25 The measures are very often the same. I think
210
1 though that the point was just made by Dr. Strand,
2 which is that we need to allow patients to tell us
3 what is important to them, and that asking merely
4 how much of this do you have, or how frequently do
5 you have it doesn't get at the issue.
6 In a nice scale that was designed by Russ
7 Portnoid to look at systems, he asks how often, how
8 bad is it, and then how much does it bother you?
9 This is brought out by examples of patients that I
10 have treated for pain for whom the pain is a 10
11 and, yet, as soon as they develop a little bit of
12 constipation they go off the medicine because the
13 constipation is worse to them than the pain was. I
14 think it is important that we give patients the
15 opportunity to indicate whether or not they think
16 that side effect is important to them.
17 At the end of the day, I would have to
18 argue that you need to allow the patients to
19 integrate that information. I think it was said
20 before that we can come up with lots of models, but
21 none of those apply to every patient. A suggestion
22 might be the following, which is that I certainly
23 would want patients to think about all the various
24 pieces that go into how are you doing, like you
25 might ask them in SF-36, and at the end of the
211
1 SF-36, so you collect all that data and you have
2 all that for subanalysis, but at the end of the
3 SF-36 you say considering all of the above, are you
4 better, the same or worse than before I started the
5 medicine? That allows the patient to integrate all
6 of those different answers. We have assigned
7 values to each of them; we have dictated that pain
8 is a zero to 10 single measure in the SF-36 and
9 that there are three measures of being able to
10 move. So, we have said movement is three times as
11 important as pain by the way we analyze that study.
12 If we allow the patient simply to integrate that
13 for us by saying overall, in terms of your pain,
14 considering all of the above, are you better, worse
15 or the same we are certainly gaining a sense of
16 information that we don't get in any other way.
17 DR. FIRESTEIN: Isn't that essentially
18 what a visual analog scale would provide in
19 addition to these other instruments?
20 DR. FARRAR: You can ask the question any
21 way you like, and a visual analog scale would
22 certainly do it. From a global perspective, there
23 is evidence that a balanced scale is better so you
24 want to allow as many down steps as up steps to
25 really get a balanced view. People tend to look at
212
1 the middle of a scale and then go one way or the
2 other.
3 The other thing is you don't need to ask
4 globally how are you with regards to the world. I
5 think the issue was brought up before that your
6 food status, your money status and your children
7 status and all those things certainly play into it.
8 You can ask globally is your pain better, much
9 better, very much better or worse, a little worse
10 or much worse and get a global response integrating
11 the things you want.
12 DR. FIRESTEIN: Would that not be the gold
13 standard for an approvable agent? If the other
14 items were all very positive, if you were trying to
15 assess whether something is an analgesic, isn't in
16 the end whether their pain has improved the most
17 important measure?
18 DR. FARRAR: I would agree, and I think
19 you have stated the two important features, which
20 is if you got the full measure of all of these
21 subcomponents and at the end of the day you said,
22 you know, are you better and they said I am
23 spectacularly better but all of their others were
24 saying they were worse, you would have to wonder
25 about whether the questions were constructed
213
1 correctly. But as long as everything is at least
2 consistent, I think that the gold standard is then
3 overall are you better, worse or the same.
4 DR. STRAND: I would simply second that
5 because we are looking for a robust response,
6 therefore, we want to see it along a variety of
7 components. It could be made so this was the
8 primary outcome provided the others showed
9 improvement or no deterioration.
10 DR. MAX: Vibeke, there is some indirect
11 evidence from pain scores from large groups of
12 patients in pain clinics from Jenssen and MrFarlan,
13 in Seattle, that because of fluctuation in pain
14 from day to day a mean of at least seven
15 measurements over a week is more robust and may, in
16 a clinical trial, theoretically allow half the
17 sample size as a single measurement on the last
18 day. But I haven't seen any such data in clinical
19 trials. Do you want to comment on whether a single
20 pain measurement on the last day or an average is
21 more robust?
22 DR. STRAND: I will actually let Dr.
23 Farrar comment on that in one minute because my
24 experience is very limited with pain trials. But
25 in terms of looking at area under the curve
214
1 analyses, for instance, in RA trials there are a
2 lot of baseline disease activity changes over time,
3 and that is why we typically get two pretreatment
4 values to give us a baseline, both an over time
5 analysis area under the curve or a landmark
6 analysis where you are looking at responders versus
7 non-responders at the last visit, where all-cause
8 dropouts are considered non-responders, show very
9 robust findings and actually reflect what we are
10 looking at. So, I agree it could be done either
11 way provided there is a value being given to
12 keeping the patient in the trial.
13 DR. FARRAR: I think that there are sort
14 of three ways of looking at that. Mark Jenssen has
15 done some spectacular work looking at the
16 robustness of different measures he looked at. I
17 think that, clearly, if you can reduce the sample
18 size that may be seen as being of importance.
19 Obviously, the talk we had yesterday about how
20 valid the measures are on a day to day basis would
21 be important in that evaluation.
22 But I think the question really gets back
23 to something that Dr. Simon said before, which is
24 that with a sufficient number of patients you can
25 prove anything is statistically significant. I
215
1 would raise the question of if you find that you
2 can get a smaller difference to be statistically
3 significant, which is really what we are talking
4 about--when you say cut the sample size, what you
5 mean is I can use less patients to find the
6 difference, which is what they have shown. The
7 argument has been made that the VAS is more
8 sensitive than the ten-point scale. There is no
9 question that it is; no question.
10 However, in studies that have been done,
11 as you know, the variance is something like 21 mm.
12 So, if your variance is already 21 mm, who cares if
13 you can find a difference of 5 mm on a 100 mm
14 scale? Because a 5 mm scale, at least in pain
15 management, I would argue is not clinically
16 important difference. If it was in sepsis and you
17 are providing benefit in terms of mortality,
18 improvement in mortality, I would argue five
19 percent is of tremendous importance. But in terms
20 of symptom management, I wonder whether being able
21 to detect a 5 mm change versus a 10 mm is of any
22 particular use.
23 DR. MAX: Let me respond to that. We
24 pointed out that there is essentially no data
25 looking in pain clinical trials chronically to
216
1 compare the sensitivity of what we are saying is
2 the most important value, reduction in pain. The
3 only data that I have ever seen--thank goodness for
4 the rheumatologists--a couple of years ago Nicholas
5 Belamy published two studies in rheumatoid and
6 osteoarthritis where he gave people 11 different
7 scales and he found that the most sensitive were
8 the VAS, the zero to ten point scale, and scales
9 that had only four points were cruder and had less
10 power.
11 So, I think we are crying out for
12 methodological studies to see if just averaging an
13 area under the curve or taking a single last day
14 measurement is important. John, I would agree with
15 you that to just take a few patients could be
16 misleading, but I think a more efficient, reliable
17 scale is always better because you can take the
18 same number of patients and get more subtle
19 differences, and perhaps prove that mechanistic
20 subsets exist. So, this is the question that I
21 would suggest to you needs to be answered,
22 particularly if it is our first outcome.
23 DR. FIRESTEIN: Dr. Anderson, and then Dr.
24 Goldkind and then Dr. Elashoff.
25 DR. ANDERSON: On this issue of seven
217
1 measurements allowing you to have the sample size,
2 I think that is likely in most cases to be an
3 exaggeration because area under the curve analyses
4 have been done in rheumatoid arthritis and compared
5 with change during the trial, just looking at the
6 beginning and the end. Although you get some
7 improvement in power, it is not that dramatic. You
8 know, you always want to have, of course, the most
9 precise measure of the outcome that you can, but I
10 wouldn't count on it to halve the sample size.
11 DR. GOLDKIND: I just wanted to note that
12 the term robustness and sensitivity are different
13 terms. I think that we have seen examples in the
14 agency where using end of study, just a landmark
15 analysis in chronic pain, created a p value that
16 wasn't there--I am sorry, that an area under the
17 curve did where a landmark did not.
18 The issue still remains though whether
19 something is overly sensitive, or sensitive to
20 irrelevant changes, or whether they are meaningful.
21 When you are looking at how to best identify a
22 metric that will help mechanistically, I don't
23 think that the kind of data that we are talking
24 about now will help in that regard. You need to
25 see how the model or the endpoint that you are
218
1 using to assess the mechanism is affected by time.
2 Dr. Lu's presentation yesterday I think pointed out
3 that, in a sense, the two metrics, a landmark
4 versus an area under the curve, give you different
5 ways of looking at the same picture and it really
6 depends on what you are interested in. I think one
7 of her points was that both of them add value. In
8 a chronic condition you want the landmark to show a
9 difference. On the other hand, if it asymptotes
10 out at three months and there is very little up
11 front, it is important to know that as well.
12 DR. ELASHOFF: I wanted to make comments
13 in two different areas. One is that in terms of
14 planning your studies, it is generally better to
15 have a more sensitive measure. The drug works as
16 it is going to work. If you can do more studies
17 because you can do each in a smaller sample size to
18 demonstrate that that drug works, that is a better
19 thing to have from an economic point of and for
20 more science.
21 If you are concerned about the issue of
22 finding statistical significance when you don't
23 believe it is real important, then you have to
24 address that issue in terms of clinically
25 meaningful. It isn't an argument for using a less
219
1 sensitive measure so you won't find out what is
2 going on.
3 The second point I wanted to make is that
4 it has been stated that responder analyses don't
5 require imputation. That is not true. If somebody
6 quits early you still have to impute something. It
7 is just that people are more ready to agree that
8 you should impute the answer non-responder. It is
9 not that no imputation is required.
10 DR. FIRESTEIN: Any additional comments in
11 this area? Dr. Katz?
12 DR. KATZ: Just one quick comment to just
13 again shore up what I hear as a few people's
14 recommendation of prospectively looking at symptoms
15 and the distress associated with the symptoms from
16 the patient perspective. There are few papers, one
17 written by a guy called Richard Anderson and also
18 Marcia Testa at the Harvard School of Public
19 Health, in Boston, looking at differences between
20 antihypertensive therapy and another set of papers
21 looking at differences between oral hypoglycemics.
22 Where the efficacy of the drugs was the same, the
23 side effects captured in a typical side effects
24 capture way in pharmaceutically sponsored trials
25 were equal between groups. A battery of typical
220
1 quality of life tests showed no differences between
2 groups but a prospectively administered symptom
3 distress inventory of something like 80 items
4 showed significant differences between groups that
5 then was able to predict dropouts from the trial
6 where none of the other measures predicted
7 dropouts.
8 So, there is evidence from that literature
9 anyway that sensitive methods to detect differences
10 in symptoms distress can actually more readily
11 discriminate outcomes between groups than either
12 primary efficacy AEs captured the usual way or
13 traditionally done quality of life batteries.
14 Maybe we should look at the same thing.
15 DR. STRAND: I think what we were trying
16 to say about domains and all that, and whether it
17 is a responder analysis or whether it is, in fact,
18 what you are suggesting, by indicating there is not
19 deterioration by some of these other instruments
20 would be a very fine way of looking at the
21 responder analysis. I think all we are trying to
22 argue for here is that we assess multiple different
23 aspects of the pain condition in these chronic pain
24 studies.
25 DR. FARRAR: Just a very brief comment,
221
1 which is that in every academic trial that I know
2 of we tend to prospectively collect side effect
3 data. We ask them at every visit. We give them,
4 you know, a 20-question scale to collect the data.
5 In the pharmaceutical industry the adage is to
6 basically report things that are self-reported.
7 I think that the concern was that in the
8 ask mode you are going to get a lot more side
9 effects, and that is certainly true. However, as
10 has been demonstrated in all of the last labels
11 that I have seen, if you display the side effect
12 rate within your treatment group and your placebo
13 group you can overcome that issue of having an
14 additional number of side effects and get at this
15 issue that Nat Katz was just remarking on, which is
16 that it begins to help us explain why patients
17 respond the way they do, and perhaps even get at
18 some mechanisms that Mitchell was referring to
19 before.
20 DR. FIRESTEIN: In item four it says
21 discuss how the selection of the measurement
22 instruments of metrics may impact the assessment of
23 efficacy. I don't think we can specifically answer
24 that, obviously, without knowing what the metrics
25 are. But I think that has been adequately covered.
222
1 There are a number of additional optional
2 points, some of which we have actually covered in
3 some detail, including patient global issues,
4 opioid sparing, as well as the time of onset of
5 effect.
6 One of the areas that we haven't talked
7 about, which probably we should touch on very
8 briefly, is the placebo issue and the relative
9 merits of active comparator versus placebo
10 controlled studies. This is a problem that comes
11 up frequently, and with greater frequency in
12 rheumatoid arthritis trials where the ability to do
13 prolonged placebo controlled trials has been
14 markedly attenuated by the fact that we now have
15 effective agents, and the ethics of having placebo
16 controlled studies for longer than, say, three
17 months now has become a significant issue.
18 I was wondering if we could touch on that.
19 We talked a little bit about open-label extensions
20 earlier, but are there any comments on the use of
21 active comparators versus placebo controls for
22 either acute or chronic indications?
23 MS. MCBRAIR: I, for one, would very much
24 like to see reduction in placebo, or maybe not at
25 all, especially in acute surgical pain, also with
223
1 children, and really all people. I think if we
2 didn't have good comparators, then we would have to
3 look at that differently, but we do. In that case,
4 I think we shouldn't lean toward placebo unless it
5 is absolutely necessary for some reason.
6 DR. FIRESTEIN: Yes, if there are rescue
7 methods when it is clear that placebo--excluding
8 children for obvious reasons, does that still fit--
9 MS. MCBRAIR: I think rescue methods
10 certainly help but if I have waited an hour for any
11 kind of pain medication and now I am being given
12 something that is going to take an hour, those two
13 hours following a surgical case, that is a long
14 time. Two hours is a very long time.
15 DR. FIRESTEIN: I would agree with that,
16 except in rheumatoid arthritis the issues are that
17 delay of therapy can have long-term implications.
18 Whether or not an additional hour of discomfort,
19 and when there is appropriate consent, is a
20 separate issue.
21 MS. MCBRAIR: I agree with rheumatoid
22 arthritis. I was really leaning towards the
23 postsurgical pain.
24 DR. ELASHOFF: I think the biggest issue,
25 as a statistician, to the question of whether you
224
1 use a placebo or an active comparator is whether
2 you are able, when you are using an active
3 comparator, to do a superiority trial or not
4 because as soon as you get into the non-inferiority
5 trial issues there are some very significant
6 statistical problems with interpreting the results
7 of the study and it may make it very, very
8 difficult to know what is going on, especially
9 since the definitions of what is equivalent or not
10 equivalent tend to be very problematic and you
11 could easily get a situation where, from one study
12 to another to another, you are creeping toward less
13 and less efficacy for what you are approving.
14 Although people worry a lot about not giving the
15 people placebo, it is good to remember that you
16 also are giving them something that is very likely
17 to have fewer side effects when you give them
18 placebo.
19 DR. FIRESTEIN: Go ahead, Dr. Anderson.
20 DR. ANDERSON: I agree with that, and I
21 would also like to say something about post surgery
22 trials because earlier this morning Dr. Babul, from
23 TheraQuest presented some data from a post surgical
24 trial which I scribbled down, I don't know if I got
25 it all correct but it looked as though in the
225
1 placebo group--you know, it was active versus
2 placebo, and there was a 55 percent response rate
3 in the placebo group and 75 percent in the active
4 group. There was more rescue medication needed in
5 the placebo group. But I would contend that even
6 in a post surgery trial, of course in the two to
7 five days not the first day, there is room for
8 placebo I think.
9 DR. FIRESTEIN: It is important to
10 remember that one of the main issues we have
11 discussed is safety, and for a compound that is in
12 early development we don't know whether we are
13 doing more harm than good and it may be that the
14 placebo is the preferred arm of the study under
15 certain circumstances, but who knows?
16 DR. MAX: First regarding placebo, I think
17 analgesic experts would unanimously agree with
18 Temple and Ellenburg's article defending the
19 importance of placebo in early drug development.
20 And, nowhere is it more important than in fields
21 like analgesia. In my 20 years at NIH we have had
22 thousands of people participate in trials and
23 receive placebos, and they have complained about
24 some things that have occurred during their care
25 but I don't remember anyone complaining about
226
1 having received placebo given their chance for
2 rescue and their consent process.
3 Regarding active comparators, for the
4 reasons that Temple makes very well, comparisons of
5 the new drug to an old drug without a placebo can
6 be very misleading if you don't establish assay
7 sensitivity. So, it is important in most cases to
8 include a placebo or vary doses of one drug as
9 well.
10 So far, in chronic pain studies it is
11 remarkable that there are almost no published
12 studies comparing within the same population drugs
13 of two different classes. So, when we have sat
14 down, a number of us around the table, to try and
15 write up consensus documents on how to treat
16 patients we have nothing to inform us. We have to
17 go to different trials where one drug is compared
18 to a placebo and then, in a different year and a
19 different group of patients in a different place,
20 another drug is compared to placebo, and because of
21 the conditions of the study there is such a wide
22 confidence interval that you really can't draw any
23 conclusions.
24 So, I would urge the FDA to try to
25 encourage more comparisons of a new drug to a
227
1 standard. These are hard because some people don't
2 want to be on a standard and it may reduce
3 enrollment. There are a lot of complex issues but
4 it would do an awful lot for prescribing practice
5 to have that information.
6 DR. WOOD: I agree with that. I think it
7 is very important that as far as we possibly can
8 ethically we include placebo. Bob and Susan in
9 their article very eloquently point out that
10 everything that we know about placebo-controlled
11 trials has stood on its head almost statistically
12 when we try to use active comparators. More
13 carelessness in the trial, all the kinds of things
14 that normally discipline us are overturned. So, I
15 think we use active comparators at our peril in
16 particular in an area like this. So, I think we
17 should certainly be using placebo as much as we can
18 with appropriate ethical and safety issues, like
19 using escapements and so on.
20 DR. FIRESTEIN: Yes, Dr. Borenstein?
21 DR. BORENSTEIN: I just want to point out
22 that the difficulty we have is that placebo works
23 so well, and if it didn't work so well life would
24 be much easier for us. The difficulty is placebo,
25 as pointed out, is not necessarily a bad choice,
228
1 unfortunately. When that happens we have to just
2 wonder what is happening in those individuals. So,
3 I have no trouble when asking patients to be in my
4 trials. It may not be the largest group but I do
5 think placebo is something that should be in these
6 trials, and people are willing to participate in
7 those circumstances.
8 DR. FARRAR: We aren't here to discuss the
9 pros and cons of the placebo effect, which
10 obviously could take a whole day in and of itself.
11 However, just a comment which is that every person
12 every day of their lives uses the "placebo effect"
13 to affect how they feel about what they are doing
14 and whether they go to work because they bumped
15 their leg or not. So, I think that the issue of
16 whether it exists or not and what it means is
17 important to take into consideration. As was just
18 commented, it can work really well in certain kinds
19 of syndromes, not so well in other ones. And, I
20 think that the primary issue is what Mitchell was
21 saying and what Dr. Wood was saying in terms of the
22 need to have a comparison against something that is
23 the least active, and that would be placebo with
24 the appropriate controls. It is rare that you
25 cannot come up with an ethical way to do it. Even
229
1 in a postop trial, if you are giving somebody a
2 pain medication that is supposed to work and you
3 give half of them a placebo, at the time of the
4 maximum pharmacologic dose you ask them is this
5 enough, and if it is not you give them a rescue.
6 Most patients, as I think was said, are willing to
7 participate in a study where they may have to put
8 up with some pain for a period of an hour or maybe
9 a little bit longer.
10 I think the second thing to mention is
11 that I have heard today or yesterday perhaps a
12 couple of times when people said placebo corrected
13 trials. I don't know what a placebo correction is
14 because the placebo effect is for free. You get
15 the placebo effect. When you give an active drug
16 you get the placebo effect. What we are really
17 looking at, and the advantage of a responder
18 analysis, is whether people reach a level where
19 they are satisfied with the relief in pain, or
20 whatever, and it doesn't matter what the response
21 rate is in the placebo group in terms of trying to
22 ascertain whether or not people are better. Right?
23 The question is better or not better. What then
24 matters is to decide whether the difference in the
25 response in the placebo group is sufficiently
230
1 different than the response in the active treatment
2 group. The one place where the placebo effect can
3 be problematic is if you have a population where
4 you end up at the top of a scale. If you end up
5 with the placebo effect working in 90 percent of
6 your population, then you are going to have a lot
7 of difficulty showing that last 10 percent where
8 you got a clinically important difference.
9 So, I think there are some issues but it
10 is not really related to subtracting out the
11 placebo effect. I think that doesn't get us
12 anywhere.
13 DR. FIRESTEIN: At this point, Lee, would
14 you like to summarize? Good luck!
15 Summary
16 DR. SIMON: Thank you, Gary and thanks
17 again to all the members of the committee for such
18 interesting discussions over the last day and a
19 half. I actually come up here with some humility,
20 being able to actually attempt to summarize what we
21 talked about and I hope that you will find it
22 useful.
23 There are a couple of statements that have
24 been made throughout from people on the committee
25 that I would like to be clear about. You know, we
231
1 are very open and we would like to believe that
2 this kind of meeting reflects how open the division
3 is to discuss with the sponsors and other
4 interested parties the way drugs are developed.
5 So, I think that is the first thing that needs to
6 be said, and can't be said enough.
7 [Slide]
8 We reviewed chronic and acute pain, and we
9 reviewed the concepts of the clinical approaches
10 and the concepts of the mechanistic approaches,
11 recognizing, of course, that the mechanistic
12 approaches are rather nascent in development. We
13 are not yet there and we still have to grapple with
14 those drugs that are presently in front of us and
15 to be soon in front of us, and have clear messages
16 about how these drugs can be approved for their
17 various different indications. Although we would
18 like to believe that the mechanistic approaches are
19 just around the corner, they are not yet there and
20 I don't think any of the protocols, drugs and
21 designs that we have in front of us right now are
22 actually dealing with mechanistic issues.
23 [Slide]
24 I think this sign really summarizes what I
25 mean by being clear. I don't want anybody to feel
232
1 like our division is giving you mixed messages. I
2 really would like you to believe that we are giving
3 you the real arrow to the right when it really
4 needs to be to the right.
5 [Slide]
6 So, we discussed temporal descriptions of
7 acute versus chronic for example, or intensity
8 differences such as mild, moderate to severe, and
9 we decided I think that they weren't enough to
10 really inform us about where we wanted to go. Some
11 of that is because of the issue of is chronic as
12 broad as it should be, or is it too broad, and
13 those kinds of issues.
14 So, we clearly need further clinical
15 trials to define mechanisms because we can handle
16 mechanisms better, but that is for the future, and
17 it is unknown whether there can be a global
18 analgesic right now for we know there are quite
19 different mechanisms driving the sensation of pain.
20 [Slide]
21 There is clear concern that we need, as an
22 agency, to design claims and consider proposed
23 trial designs fostering new development, new drug
24 development for pain. I actually think that is
25 very true. For the chronic pain proposal, I heard
233
1 some people thought it had merit. That was again,
2 just to remind everybody in case you have
3 forgotten, three models, three co-primary outcomes
4 of pain function and patient global, and it would
5 be replicated in nature with disparate
6 etiopathogenesis mechanisms or disease states.
7 They were replicated, necessary, when you were
8 doing studies in models with simpler mechanisms or
9 not. We weren't sure whether or not it was going
10 to need to be replicated in that particular
11 circumstance.
12 And, it seemed that in the vote we took,
13 although there was no vote but consensus building
14 that we took, although I am happy to say I
15 understand the camps, I am not entirely sure we got
16 consensus. Most people said yes to pain as a
17 measure; yes to patient global and that is a
18 measure of clinical relevance of the response; and
19 there was a qualified yes to function. We would
20 need to take that into consideration of the model
21 or mechanism or disease state that we were talking
22 about. Obviously, cancer function or a patient
23 with cancer who is functioning, that would be a
24 different issue than some other diseases.
25 There was debate of how many different
234
1 models are required to get any type of specific
2 claim for chronic pain. Are three different models
3 required? Dr. Verburg suggested four models of one
4 trial in each. Maybe Dr. Firestein resonated with
5 that a little bit. We were suggesting three models
6 with two replicate trials. Dr. Farrar suggested
7 two neuropathic models and two somatic pain models.
8 So, clearly, we will be taking back this
9 information to think more about what we should do.
10 [Slide]
11 In that context, the lumping and splitting
12 context is very important. We had thought we were
13 doing both lumping and splitting because we gave
14 the opportunity to split or lump. Dr. Abramson
15 kind of resonated with the rigor that would be
16 associated with that kind of approval, and it
17 really raised issues about whether it would be
18 iterative. You would get one indication and then
19 perhaps a much broader organ-based indication, and
20 then perhaps a whole disease indication, fully
21 recognizing, however, that the daunting nature of
22 the full, whole thing, the whole kit and caboodle
23 may be just too much and, in fact, companies would
24 opt for something easier, perhaps cheaper, and then
25 off-label use would drive that and that would not
235
1 be an ideal situation. I think it is really
2 critical for us to remember that we were providing
3 in our proposal that opportunity, for better or for
4 worse.
5 [Slide]
6 We also recognized and heard clearly that
7 acute pain is not similar to thinking about the
8 drugs that would be used to treat it. Thus,
9 actually we are thinking about short-term
10 analgesics rather than drugs for acute pain. The
11 same thing in obverse is true for chronic pain. We
12 are really thinking about drugs to be used for a
13 long period of time and that has issues regarding
14 safety and durability of response in trial design.
15 [Slide]
16 We learned something that I think we have
17 consensus on, that chronic low back pain, if
18 handled correctly, might be an indication to go for
19 independently, or actually may be part and parcel
20 of a much larger package. Although heterogeneous,
21 it consists of many different processes but they
22 can be delineated, and we could select a specific
23 patient population with some similarity in the
24 natural history, perhaps ignoring or removing those
25 patients with reticulopathy or neuropathy, and
236
1 perhaps we would have a model that we could use or
2 pain disease state that we could use for a clear
3 indication, as well as performance of a broader
4 label. It seemed that there was good consensus
5 about that if we made sure that we subtracted out
6 patients with neuropathic disease and systemic
7 disease.
8 I think we heard clearly that there are
9 two really broad patient populations that we have
10 not dealt with very well. One is the elderly and
11 one is the pediatric population, and we have to
12 recognize that the elderly are quite unique.
13 Polypharmacy is a significant issue with them.
14 Safety issues are particularly important, and some
15 of the elderly who are suffering chronic pain are
16 in unusual care-giving environments. Perhaps as
17 the baby-boomer population gets older it will be a
18 usual care-giving environment, but we have to learn
19 how to use nursing homes for actual study designs
20 and carrying out studies in those areas as the
21 patient population in them grows larger.
22 [Slide]
23 The issue of flair design was debated.
24 Some of us had problems with flair design. It
25 actually has been tried and true but, on the other
237
1 hand, it preselects those patients who both
2 tolerate the drug as well as respond. A priori
3 they have been on the drug for a period of time so
4 there are issues about that particular problem.
5 We heard about possible ways to do a
6 run-in phase and withdrawal studies, both of which
7 have problems. The run-in phase really doesn't do
8 anything differently than does the flare design.
9 It suggests that you are only taking patients who
10 are having a response and getting rid of all those
11 patients who can't tolerate the drug. So, you have
12 a true bias in the evaluation.
13 The other concept of the withdrawal phase
14 which Dr. Laska asked me to comment on was, in
15 fact, some concern about are the patients who get
16 withdrawn unblinded or not based on the symptoms
17 that emerge? So, that is an issue that I think we
18 are going to have to think about.
19 [Slide]
20 Many of us talked about the issue of
21 opioid sparing, although it is not dissimilar from
22 glucocorticoid sparing, and how important it is for
23 the assessment of outcome. It might be a good
24 response to measure. Would it be a primary
25 measure? Probably not. It might be a useful
238
1 secondary measure but we would have to debate that,
2 demonstrating that the study drug works and
3 decreases the need for opioids and, presumably, the
4 study drug in the circumstance would have less side
5 effects than the opioids so there would be a
6 warranted reason for the study. The problem, of
7 course, is that the study drug might enhance the
8 effects of the concomitant opioid therapy, thus
9 decreasing the use of opioids or, alternatively,
10 decreasing use of the opioids may be due to the
11 emergence of increasing toxic effects.
12 What I am constantly daunted by, and I am
13 not really that far off in glucocorticoid sparing
14 either, is that I don't know what it means to be
15 sparing because I don't know if 3 mg is better or
16 30 mg is really sparing, and I think we have to
17 debate what that really means. As mentioned by Dr.
18 Wood, there is the issue of the PK change and what
19 that would imply to the whole process.
20 [Slide]
21 We then moved on to the ABCs of acute
22 pain, and there seemed to be--perhaps you could
23 show me with smiles on your faces--less debate
24 about this. This seemed to be something that you
25 all bought into faster for good things.
239
1 [Slide]
2 Clearly, we want to improve the
3 information in the label by turning from inferences
4 evidence by PK modeling to data derived from
5 clinical trials. That would be the multi-dose
6 assessments. That was informed by the B of the
7 ABCs.
8 We want to improve safety analysis of
9 short-term use by analyzing long-term exposures
10 even for drugs approved only for short-term use.
11 There seemed to be some confusion as to whether or
12 not, if we were going to require some chronic
13 exposure, and maybe even efficacy trials, that that
14 actually might mean two replicate trials or three
15 co-primary outcomes, maybe even three different
16 disease states. That is not really what we were
17 suggesting. It probably would be just one trial,
18 perhaps even just very robust and perhaps just one
19 outcome measure but we would have to debate that
20 and talk about it in an open fashion to determine
21 exactly what we would want. But this was then
22 informed by proposal C of the ABCs.
23 [Slide]
24 We clearly heard that generalizing to
25 postop pain and efficacy from a dysmenorrhea trial
240
1 or dental pain trials really was a problem and we
2 have been very uncomfortable with that. So, we
3 needed to think about requiring or suggesting that
4 not only does one do an outpatient trial in such a
5 circumstance, but one might want to choose an
6 inpatient model which would give a broader aspect
7 of pain relief, thus, a bunionectomy model as well
8 as a dental pain model.
9 Additional info regarding the dosing
10 interval was needed, and that was clearly defined
11 by B of the ABCs; more optimizing of the dosing
12 schedule in responder versus non-responder
13 inclusion, which I actually found to be a
14 fascinating discussion.
15 [Slide]
16 Dose creep was brought up, and I think
17 that it is very important. and it came up several
18 times from the committee that we need to construct
19 our clinical trials in a real-world way to ensure
20 that we understand how the drugs are going to be
21 used in the real world, and that doesn't imply
22 open-label analysis; that just implies different
23 ways of thinking about trial design than we have
24 done before. Issues of longer time of use requires
25 the chronic studies, as we talked about.
241
1 [Slide]
2 The discussion went on after the
3 presentations regarding the matrix of clinical
4 trials. Again, I think everybody around the table
5 believed that they should inform us about
6 real-world use and should be labeled as such.
7 Time to rescue should include the
8 non-responders and that implies an
9 intention-to-treat analysis, not just a responder
10 analysis.
11 New designs with preemptive anesthesia
12 raises the question of whether or not we should be
13 thinking about that differently than acute pain,
14 and maybe that is a whole other world of trial
15 design, and all the consultants out there can start
16 to think about that and create new business for
17 yourselves, which is a good thing. Improved GDP
18 and all of that.
19 Short-term studies, pain relief, patient
20 global in terms of level of response for how long
21 and when is the onset; when it separates from
22 placebo; drugs not with onset within an hour but a
23 very good analgesic, do they inform about some
24 acute use? In fact, that came up several times,
25 this idea that there is the acute; there is the
242
1 chronic; but what about kind of the middle ground?
2 We need to start to think about this subacute use
3 and what that really means.
4 [Slide]
5 Also, going through dose descriptions and
6 minimum time to the next dose is informed by the
7 time to onset. It needs also to be limited by
8 total dose and dose ranges may be better described
9 by quartiles of response. I really like that idea.
10 I think that really gives us a much better handle
11 on what this all means.
12 [Slide]
13 Lastly, but not leastly, we heard about a
14 tiered responder analysis, informing patients and
15 clinicians much more so than present analyses do
16 for pain. One could see that in acute pain you
17 could define a level of pain relief, along with the
18 duration of pain effect within the same construct
19 of explanation or description. And, in chronic
20 pain it would develop an information database
21 including efficacy, kind of encompassing pain and
22 suffering relief; durability of response; time to
23 retreatment or time to treatment failure; as well
24 as function and HRQOL measures; and then also
25 safety. So, this would be a remarkably robust data
243
1 set to inform patients about what really is going
2 to go on with the therapy.
3 [Slide]
4 I want to close with this, and I don't
5 really show this entirely in jest--entirely. This
6 was actually a real traffic sign in England where
7 they actually advertised and demonstrated the
8 directions to the secret nuclear bunker. We don't
9 really hold any secrets in the agency. People have
10 come over to me and said, well, would you really
11 talk to us? Or, can we come talk to you? Or, we
12 have our stuff already in and we are talking about
13 changing, are we going to be held to a different
14 standard when we have already done all of our
15 trials?
16 Well, one, you need to talk to us. Make
17 an appointment and come in for a meeting. Call
18 your project manager and see what the status is. I
19 would prefer not to hear any complaints that we are
20 not willing to talk to you. I am being very public
21 about this. We are willing to talk to you. There
22 are no secrets here.
23 Number two, we are willing to debate with
24 you as to what might be happening in this
25 particular turbulent time of change because, in
244
1 fact, we are trying to do, and I think you all are
2 too, what is best for patients and to derive the
3 most information in the most open way. So, I
4 invite you to give us a call. Those of you that
5 have not been in for a while and have been busy
6 developing drugs, I really urge you to take
7 advantage of all the opportunities to have guidance
8 discussions because, in fact, it is much better to
9 come in and talk to us before you come in for your
10 pre-NDA meeting and be surprised.
11 So, in that context, let me suggest that
12 we show you the way to our secret nuclear bunker
13 and give you all the directions up front, and I
14 think everybody will be happy.
15 So, thank you again very much for coming.
16 Thank you to the committee for working so hard in
17 helping us and informing us about your ideas. I
18 don't know what will happen next but we will
19 certainly have another meeting about it.
20 DR. FIRESTEIN: Thank you very much. The
21 meeting is closed.
22 [Whereupon, at 2:30 p.m., the proceedings
23 were adjourned.]
24 - - -