Food and Drug Administration s3
Total Page：16
File Type：pdf, Size：1020Kb
1 1 2 1 2 3 FOOD AND DRUG ADMINISTRATION 4 5 CENTER FOR DRUG EVALUATION AND RESEARCH 6 7 8 9 10 11 CLINICAL PHARMACOLOGY SUBCOMMITTEE 12 13 OF THE 14 15 ADVISORY COMMITTEE FOR PHARMACEUTICAL SCIENCE 16 17 18 19 20 21 22 23 24 25 26 27 8:34 a.m. 28 29 Tuesday, April 22, 2003 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 1 2 2 1 Conference Room 2 5630 Fishers Lane 3 Food and Drug Administration 4 Rockville, Maryland 20857 1 3 2 1 ATTENDEES 2 3 SUBCOMMITTEE MEMBERS: 4 5 JURGEN VENITZ, M.D., PH.D., Acting Chair 6 Associate Professor 7 Department of Pharmaceutics 8 Virginia Commonwealth University 9 School of Pharmacy 10 Box 980533 11 410 North 12th Street 12 Richmond, Virginia 23298-0533 13 14 KATHLEEN REEDY, R.D.H., M.S., Executive Secretary 15 Advisors and Consultants Staff (HFD-21) 16 Center for Drug Evaluation and Research 17 Food and Drug Administration 18 5600 Fishers Lane 19 Rockville, Maryland 20857 20 21 EDMUND V. CAPPARELLI, PHARM.D. 22 Associate Clinical Professor 23 University of California, San Diego 24 9500 Gillman Drive 25 La Jolla, California 92093 26 27 HARTMUT DERENDORF, PH.D. 28 Professor, Department of Pharmaceutics 29 University of Florida College of Pharmacy 30 P.O. Box 100494, Health Science Center 31 Gainesville, Florida 32610-0494 32 33 DAVID FLOCKHART, M.D., PH.D. 34 Professor, Departments of Pharmacology 35 and Medicine 36 Indiana University School of Medicine 37 Division of Clinical Pharmacology 38 Wishard Memorial Hospital WP OPW 320 39 1001 West 10th Street 40 Indianapolis, Indiana 46202 41 42 WILLIAM J. JUSKO, PH.D. 43 Professor, Department of Pharmaceutics 44 State University of New York at Buffalo 1 4 2 1 School of Pharmacy 2 Buffalo, New York 14260 1 5 2 1 ATTENDEES (Continued) 2 3 SUBCOMMITTEE MEMBERS: (Continued) 4 5 GREGORY L. KEARNS, PHARM.D. 6 Professor and Division Chief 7 Pharmacology and Toxicology 8 Children's Mercy Hospital 9 2401 Gillham Road 10 Kansas City, Missouri 64143 11 12 MARY V. RELLING, PHARM.D. 13 Member, Pharmaceutical Sciences 14 St. Jude Children's Research Hospital 15 332 North Lauderdale, Room D-1052 16 Memphis, Tennessee 38105 17 18 WOLFGANG SADEE, DR.RER.NAT. 19 Chair, Department of Pharmacology 20 College of Medicine and Public Health 21 Ohio State University 22 5072 Graves Hall, 333 West 10th Avenue 23 Columbus, Ohio 43210 24 25 LEWIS B. SHEINER, M.D. 26 Professor, Laboratory Medicine 27 University of California, San Francisco 28 Box 0626, C-255 29 San Francisco, California 94143 30 31 MARC SWADENER, ED.D., Consumer Representative 32 2235 Dartmouth Avenue 33 Boulder, Colorado 80305-5207 34 35 36 GUEST SPEAKER: 37 38 MATS KARLSSON, PH.D. 39 Uppsala University 40 Sweden 1 6 2 1 ATTENDEES (Continued) 2 3 FOOD AND DRUG ADMINISTRATION STAFF: 4 5 PETER LEE, PH.D. 6 LARRY LESKO, PH.D. 7 NHI NGUYEN, PH.D. 8 HE SUN, PH.D. 9 GENE WILLIAMS, PH.D. 10 JENNY J. ZHENG, PH.D. 1 7 2 1 C O N T E N T S 2 3 AGENDA ITEM PAGE 4 5 MEETING STATEMENT 6 by Ms. Kathleen Reedy 7 7 8 INTRODUCTION TO THE MEETING 9 by Dr. Larry Lesko 10 10 11 TOPIC 1: QUANTITATIVE RISK ANALYSIS USING 12 EXPOSURE-RESPONSE FOR DETERMINING DOSE 13 ADJUSTMENT FOR SPECIAL POPULATIONS 14 15 Introduction - by Dr. Peter Lee 19 16 17 Example 1 - by Dr. Nhi Nguyen 26 18 19 Example 2 - by Dr. Jenny Zheng 42 20 21 Committee Discussion 61 22 23 Example 3 - by Dr. He Sun 82 24 25 Committee Discussion 93 26 27 OPEN PUBLIC HEARING 116 28 29 Example 4 - by Dr. Mats Karlsson 118 30 31 Committee Discussion 141 32 33 TOPIC 2: PEDIATRIC POPULATION PHARMACOKINETICS 34 STUDY DESIGN TEMPLATE AND ANALYSES OF THE FDA 35 PEDIATRIC DATABASE 36 37 Introduction - by Dr. Larry Lesko 150 38 39 Pediatric PPK Template - by Dr. Peter Lee 158 40 41 Committee Discussion 165 42 43 Pediatric Database Analysis - Dr. Gene Williams 188 44 1 8 2 1 Committee Discussion 203 1 9 2 1 P R O C E E D I N G S
2 (8:34 a.m.)
3 DR. VENITZ: I'd like to call the meeting to
4 order, please.
5 Welcome, everybody. This is the Clinical
6 Pharmacology Subcommittee meeting. We have a full agenda,
7 as you can tell.
8 I'd like to open the meeting by introducing
9 the individuals around the table, maybe starting with Dr.
10 Derendorf, please.
11 DR. DERENDORF: Hartmut Derendorf, University
12 of Florida.
13 DR. CAPPARELLI: Edmund Capparelli, University
14 of California, San Diego.
15 DR. FLOCKHART: Dave Flockhart from Indiana
16 University.
17 DR. SHEINER: Lewis Sheiner, University of
18 California, San Francisco.
19 DR. SWADENER: Marc Swadener, Boulder,
20 Colorado.
21 MS. REEDY: Kathleen Reedy, Food and Drug
22 Administration. 1 10 2 1 DR. VENITZ: Jurgen Venitz, Virginia
2 Commonwealth University.
3 DR. JUSKO: William Jusko, University at
4 Buffalo.
5 DR. KEARNS: Greg Kearns, University of
6 Missouri, Kansas City.
7 DR. RELLING: Mary Relling, St. Jude
8 Children's Research Hospital in Memphis.
9 DR. SADEE: Wolfgang Sadee, Ohio State
10 University.
11 DR. LESKO: Larry Lesko, Office of Clinical
12 Pharmacology and Biopharmaceutics at FDA.
13 DR. LEE: Peter Lee, FDA.
14 DR. VENITZ: Thank you.
15 Our next order of business is the conflict of
16 interest statement. Kathleen Reedy will read the conflict
17 of interest statement.
18 MS. REEDY: Acknowledgement related to general
19 matters waivers, Clinical Pharmacology Subcommittee of the
20 Advisory Committee for Pharmaceutical Science, April 22,
21 2003, an open session.
22 The following announcement addresses the issue 1 11 2 1 of conflict of interest with respect to this meeting and
2 is made a part of the record to preclude even the
3 appearance of such at this meeting.
4 The topics of this meeting are issues of broad
5 applicability. Unlike issues before a committee in which
6 a particular product is discussed, issues of broad
7 applicability involve many industrial sponsors and
8 academic institutions.
9 All special government employees have been
10 screened for their financial interests as they may apply
11 to the general topics at hand. Because they have reported
12 interests in pharmaceutical companies, the Food and Drug
13 Administration has granted general matters waivers to the
14 following SGEs which permits them to participate in these
15 discussions: Dr. Edmund Capparelli, Dr. William Jusko,
16 Dr. Gregory Kearns, Dr. Howard McLeod, Dr. Wolfgang Sadee,
17 Dr. Lewis Sheiner.
18 A copy of the waiver statements may be
19 obtained by submitting a written request to the agency's
20 Freedom of Information Office, room 12A-30 of the Parklawn
21 Building.
22 In addition, Dr. Hartmut Derendorf, Dr. David 1 12 2 1 Flockhart, Dr. Mary Relling, and Dr. Marc Swadener do not
2 require special matters waivers because they do not have
3 any personal or imputed financial interests in any
4 pharmaceutical firms.
5 Because general topics impact so many
6 institutions, it is not prudent to recite all potential
7 conflicts of interest as they apply to each member and
8 consultant.
9 FDA acknowledges that there may be potential
10 conflicts of interest, but because of the general nature
11 of the discussion before the committee, these potential
12 conflicts are mitigated.
13 With respect to FDA's invited guest speaker,
14 Dr. Mats Karlsson reports that he has contracts and/or
15 grants with AstraZeneca, Oasmia, Pfizer, and Servier. He
16 also receives consulting fees from AstraZeneca, Ferring,
17 Lilly, Pfizer, and Roche; and speaker fees from Johnson &
18 Johnson and NovaNordisk.
19 In the event that the discussions involve any
20 other products or firms not already on the agenda for
21 which FDA participants have a financial interest, the
22 participants' involvement and their exclusion will be 1 13 2 1 noted for the record.
2 With respect to all other participants, we ask
3 in the interest of fairness that they address any current
4 or previous financial involvement with any firm whose
5 product they may wish to comment upon.
6 DR. VENITZ: Thank you, Kathleen.
7 Before we proceed to the official business of
8 today, I'd like to welcome a few new committee members.
9 Dr. Swadener to my right is the consumer representative
10 who also serves on the Advisory Committee for
11 Pharmaceutical Science. We've got Dr. Shek who couldn't
12 make it today who is the industry representative, also
13 serving on the Advisory Committee for Pharmaceutical
14 Science. And Drs. D'Argenio and Davidian who couldn't
15 make it today.
16 I would also like to thank the two outgoing
17 members of the committee, Dr. Lalonde and Dr. Hale, as
18 well as Dr. Jusko to my left, for chairing the previous
19 committee meeting.
20 With that said, I'd like to turn over the
21 meeting to Dr. Lesko who is going to introduce the topics
22 for the next day and a half. Larry. 1 14 2 1 DR. LESKO: Thank you, Jurgen.
2 Well, good morning, everybody and again
3 welcome back to Rockville. This is the second Clin Pharm
4 Subcommittee meeting. We had our first back six months
5 ago, on October 22-23, and as reflect back on that
6 meeting, it was an extremely productive meeting for us.
7 The advice we received at that time was excellent, and
8 we've been thinking about it since then as we move forward
9 with the initiatives we introduced at that first meeting.
10 I do want to say thanks again. As I look
11 around the room, I recognize everyone here as very busy in
12 their own work, and taking time to come to Washington and
13 participate with us is extremely important. I think
14 you'll find that the mission that we have for this
15 committee is a noble one relating to some exciting times
16 in the area of drug development generally and clinical
17 pharmacology specifically.
18 Let me talk a little bit about today's
19 meeting. Let me start with what's new since we last met
20 in October, and there are really three exciting things
21 that are new that really impact the topics that we'll talk
22 about today. 1 15 2 1 The first is an FDA-wide announcement that our
2 Commissioner made back in January called Improving
3 Innovation in Medical Technology Beyond 2002. What this
4 initiative entails is quite lengthy. There are several
5 goals, but basically it revolves around improving the
6 process, including the drug development process, for
7 bringing medical innovations, treatments, and devices to
8 the marketplace as quickly as possible that would benefit
9 public health.
10 A second part of that, however, is improving
11 the review process at FDA through a quality systems
12 approach. This is the goal that we are going to sort of
13 use to couch today's topics because a quality systems
14 approach, if I think of it in terms of goals, has the
15 goals I have on the slide basically.
16 Application of advances in science. Well,
17 what are those advances in science? They're clinical
18 trial design. They're clinical pharmacology study
19 designs, statistical approaches, modeling and simulation,
20 use of dose response in PK/PD in interesting ways.
21 Use of new technology. Use of new technology
22 embraces things like the quantitative methods we're going 1 16 2 1 to talk about today. It embraces the integration of
2 pharmacogenetics into drug development and review.
3 Rigorous analytic reviewer tools. This is the
4 how to do it. What are the tools that we can make
5 available to our reviewers to achieve this quality systems
6 approach? We'll be talking about one of those in the
7 first topic.
8 And finally, the overall goal of this
9 initiative is to provide for high quality reviews. That's
10 translated into effective reviews, efficient reviews, and
11 consistent reviews.
12 Initiative number two that's occurred since
13 the last time we met is an initiative under our
14 Prescription Drug User Fee Act, PDUFA. It's the
15 premarketing risk assessment initiative. We only recently
16 had our first public meeting having to do with the risk
17 assessment initiative, and Bob Meyer, our ODE II director,
18 defined risk assessment as the process of identifying,
19 estimating, and evaluating the nature and severity of risk
20 of a drug product. You'll see some links between this and
21 the first topic in today's presentations.
22 In that meeting on April 9th, Bob Temple 1 17 2 1 described the ideal safety database that we ought to be
2 striving for, and that included a complete
3 characterization of the clinical exposure-response
4 relationship certainly for efficacy, but also for drug
5 safety. And he made the point that this is important for
6 making decisions about dosing adjustments, particularly in
7 many of the clinical pharmacology studies when exposure
8 goes up for a variety of reasons.
9 He recommended a good search for
10 individualization factors. This obviously involves
11 studying individual plasma drug levels, and he made the
12 point that it's always necessary to assess polymorphic
13 drug metabolism, and you'll see that this relates to one
14 of our topics in this meeting.
15 Finally, he touched upon pediatrics, an
16 important topic, and made the point that these pose
17 special issues related to dose, PK, and PD, and we'll be
18 touching upon that as well today.
19 The third initiative that's been launched
20 since October is the one that relates to our FDA Science
21 Board. We had this meeting last week on the same day as
22 our risk assessment meeting. The theme of the Science 1 18 2 1 Board meeting was integrating scientific advances into
2 regulation with the emphasis on pharmacogenetics. Dr.
3 Woodcock made a presentation at the Science Board and
4 stated that genetic contributions to variability in
5 toxicity include differences in drug metabolism, for
6 example, thiopurine methyltransferase, which was a topic
7 of our last meeting, but more broadly recommended that we
8 look at the use of genetic tests for metabolizer status to
9 predict dosing.
10 So think of those three broad initiatives, and
11 I hope it gives you a context for the discussion that
12 we're going to have today and tomorrow.
13 We have four proposals, four topics on the
14 agenda. The first is a proposal that we initiated our
15 discussion in October and we've refined the proposal and
16 we'll present with examples today the idea of a
17 standardized approach to quantitate the impact that
18 changes in exposure, related to efficacy or safety, result
19 from changes in PK caused by a variety of intrinsic and
20 extrinsic factors.
21 What we're trying to accomplish with this
22 proposal and standardized approach relates to what I 1 19 2 1 mentioned about the Commissioner's initiative, quality
2 systems and review. We want to achieve a rational
3 scientific basis for dosing adjustment that's quantitative
4 and that links to the assessment of risk.
5 We have a goal of identifying
6 individualization factors, and through our examples today
7 you'll see some of those factors.
8 And finally at the end of the day, we hope to
9 develop a standardized method that would rely on many
10 different tools, but a standardized thinking process that
11 we hope would bring consistency to the label
12 recommendations that we make in terms of dosing
13 adjustments.
14 Our topic number two is pediatrics, and again
15 going back to October, we opened the discussion of
16 pediatrics very briefly the last time. Today we'll be
17 making a proposal. The proposal will relate to a
18 pediatric population PK design template. We'd like to
19 recommend the use of this template for getting information
20 about pediatrics during drug development. We feel that
21 this approach is efficient in many cases. We feel it's
22 under-utilized for a variety of reasons relating to 1 20 2 1 perhaps a lack of understanding, perhaps related to a
2 concern that FDA will not accept this type of study
3 approach. But we'd like your input on the template that
4 we'll be presenting.
5 Related to that, at the last meeting I had
6 mentioned that we have this database at FDA related to
7 pediatric studies. These are studies that were done under
8 our pediatric rule. We felt this database is loaded with
9 information that we could capitalize on by studying it,
10 looking for trends, and learning something about the
11 pediatric clinical pharmacology situation. We'll update
12 you on our progress. It's been slow. It's been difficult
13 because we don't have access to an electronic database,
14 and much of our time is simply gathering and assembling
15 data. But nevertheless, we'd like to share with you some
16 of the things that we've learned so far, but more
17 importantly what we'd like to do going forward and look
18 for input on designing the studies of this database.
19 The third topic, which we'll talk about
20 tomorrow morning, is what I'll call a work in progress.
21 We'll all familiar with the human genome. We've been
22 bombarded with information about it, particularly this 1 21 2 1 month on the 50th anniversary of identifying the double
2 helix structure for DNA. Certainly the dream of genomics
3 is to develop new and better treatments for disease
4 states. But we feel there's a lot to be gained. There's
5 a lot of substantial improvement that could be made by
6 integrating pharmacogenetics into the treatments that we
7 now have and using this science as it matures for
8 identifying more optimal doses for subsets of the
9 population.
10 We'll continue to talk about this tomorrow.
11 What we're going to emphasize is moving forward with the
12 knowledge that we have on polymorphic drug-metabolizing
13 enzymes that influence variability in drug response,
14 especially toxicity. It's a challenging area. Many
15 questions come up in the context of this, things like how
16 much variability will genetics explain, how much of an
17 effect on drug dose will genetics explain, how important
18 is it.
19 But I think more importantly where we're
20 heading is to create a general construct for looking at
21 improvement in existing therapies, existing therapies
22 being approved drugs, to determine what criteria we ought 1 22 2 1 to be thinking about that would warrant updating labels
2 for products to optimize drug dosing using genetic
3 information.
4 Last time we talked about specifically
5 thiopurine, TPMT. Tomorrow we'll touch upon that, but
6 we'll also be looking for a broad way to best program in
7 this area and what we need to be thinking about in
8 assessing data, assessing evidence to update labels. So a
9 rational scientific basis.
10 Now, our fourth topic today is going to be a
11 new topic. We'll actually talk about it tomorrow. We've
12 been working pretty much over the last year in the area of
13 drug-drug interactions. It continues to be a major
14 problem if you read the current literature in JAMA and the
15 New England Journal of Medicine about adverse drug
16 reactions and the high fraction of those that are related
17 to drug interactions. We have some ideas on revising our
18 guidance on drug-drug interactions. We have some
19 questions on transporter based drug-drug interactions. As
20 was stated in our risk assessment workshop, some matters
21 always need assessment in regulatory review, including new
22 interactions that we may not have paid as much attention 1 23 2 1 to in the past, such as interaction involving
2 glucuronidation and transporter interactions like P-gp.
3 So we're going to bring this topic forward
4 tomorrow with some issues and questions. We'll talk about
5 the use and extension of a classification system for 3A4
6 inhibition for single and multiple drug interactions.
7 This is a classification system that we can see moving
8 towards label language for bringing some consistency to
9 how we report drug interactions in the label.
10 And a big question that we frequently get from
11 sponsors during the drug development process and we ask
12 ourselves is when and how should the role of P-
13 glycoprotein in drug interactions be investigated. This
14 is an emerging area and we're beginning to see clinical
15 evidence that this important and we'd like to develop a
16 path forward that's reasonable and rational.
17 Each of the presenters is going to have some
18 specific questions on the topic, but I'd like you to think
19 about some of the broader questions that we have for the
20 session. For example, you'll hear many proposals during
21 the next day and a half. Aside from the specific
22 questions about the subtopics of today's meeting, think 1 24 2 1 about the rationality of these proposals. Are they
2 reasonable? Are they feasible? Overall, do you think
3 these will enhance the quality of drug development and
4 regulatory review? Are these the priorities that we
5 should be looking at in our clinical pharmacology program?
6 We have works in progress, topics number 3 and
7 number 4. That means we need input as we move forward
8 with these programs and some advice on whether these
9 objectives are worthwhile. And in particular, what is the
10 best way to integrate new science and technology, whether
11 it's genetics, whether it's P-gp transporter information,
12 into therapeutics and regulatory review?
13 Well, that's the introduction to the lineup
14 for today. I look forward to the discussion. It was a
15 terrific discussion in October. Again, as we look around
16 the table, the expertise of this committee is really
17 substantial, and we're looking forward to defining and
18 expanding upon the proposals that we're going to make
19 during this meeting. Thanks.
20 DR. VENITZ: Thank you, Larry.
21 Now we're moving to our first topic. As Larry
22 indicated, we're going to talk about exposure response as 1 25 2 1 a way of justifying dose adjustments. The introduction
2 will be given by Peter Lee. He's Associate Director of
3 the Office of Clinical Pharmacology and Biopharmaceutics.
4 DR. LEE: Good morning. The first topic we're
5 going to discuss today is quantitative risk analysis using
6 exposure response for determining dose adjustment for
7 special populations.
8 This topic has been discussed in our previous
9 meeting back in October 2002. In the last meeting we
10 talked about three main topics. We had proposed a
11 standardized approach to estimate the probability of
12 adverse events in special populations using exposure-
13 response information. We also discussed a regulatory
14 decision tree for recommending dose adjustment in special
15 populations. And lastly, we also discussed the potential
16 application of utility functions for risk and benefit
17 assessment.
18 So today we're going to present several
19 examples to illustrate what we talked about in the last
20 meeting. We're going to present examples of a
21 standardized approach for using exposure-response
22 information to adjust dose in special populations. We 1 26 2 1 also are going to present one example of using population
2 analysis to obtain PK/PD information from the large
3 clinical trials. And in the last example, we're going to
4 present a methodology to applying utility function for an
5 optimal dosing strategy.
6 Just to give a little bit of background
7 information, as you know, many of the NDAs may contain up
8 to 20 or more clinical pharmacology studies, and in these
9 studies different intrinsic and extrinsic factors may be
10 studied and these factors may influence the
11 pharmacokinetics of the drug in these special populations.
12 Therefore, we need a consistent approach to determine the
13 dosing adjustment requirement in these populations.
14 Here's one example. In this particular
15 example, we have about 11 factors that have been studied
16 in the NDA. As you can see, the area under the curve of
17 the drug may change depending on what factors from 0
18 percent, which is no change, to a 60 percent increase in
19 the special populations.
20 So the question is, how do we make the dose
21 adjustment according to the pharmacokinetic results?
22 Where is the cutoff? Do we adjust the dose at 30 percent 1 27 2 1 increase of AUC, or do we adjust the dose at 60 percent
2 increase of AUC?
3 So the answer is that we had to look at PK/PD
4 information and determine what is the clinical
5 significance of this AUC change.
6 So there are some issues related to the dosing
7 adjustment in drug labels of NDA submissions. Quite often
8 we have seen inconsistency in dosing adjustment
9 recommendations in the initial label of NDA submissions.
10 Exposure-response information, as is required to interpret
11 the pharmacokinetic change, is now always available in the
12 NDA submission. The FDA reviewer had to conduct
13 additional exposure-response analyses in order to
14 interpret the AUC change. Therefore, we feel that a
15 standard for analyzing and interpreting the exposure-
16 response information will be critical and beneficial to
17 regulatory decision making in terms of dose adjustment in
18 special populations.
19 So to improve the current status, in the last
20 meeting we had proposed to develop and evaluate a
21 standardized approach for the reviewer and possibly for
22 industry to quantitatively assess the impact of exposure 1 28 2 1 change on either safety or efficacy that results from a
2 change in pharmacokinetics due to intrinsic and extrinsic
3 factors.
4 This is the standardized approach that we had
5 proposed in the last meeting. In this example, basically
6 we have seen an increase of exposure of the test
7 population compared to the reference. Using exposure-
8 response information, we can estimate the distribution of
9 response in both reference and test populations. If we
10 could determine what is the critical value of response,
11 which is considered clinical significance, which is the
12 vertical line here, then we can calculate the probability
13 of a clinically significant response based on the PK
14 change, as well as the PK/PD relationship.
15 So in order to interpret the clinical
16 significance of pharmacokinetic change, we need to have
17 observed data in pharmacokinetics in special populations.
18 We also need data of the PK/PD relationship. With this
19 information, then we can estimate the probability of
20 adverse events in the special populations with the
21 response that is greater than the clinically significant
22 critical value. 1 29 2 1 However, in order to determine the clinically
2 significant critical values, we need to base it on a risk
3 and benefit assessment of the drug therapy. Currently
4 we're doing that on a case-by-case basis through a
5 discussion between clinical pharmacology and the medical
6 reviewer. But in the last meeting with the committee, we
7 also proposed that we can use a utility function to assess
8 the risk and benefit of pharmacokinetic change in the
9 special populations.
10 In the last meeting, we also discussed a
11 decision tree for dosing adjustment recommendations.
12 Since the last meeting, based on the recommendation from
13 this committee, we have made some modifications of the
14 decision tree, and this is the current decision tree.
15 Basically we ask a number of questions.
16 First, according to our current guidance, we
17 ask whether the 90 percent confidence interval of test
18 over reference is within the default no-effect boundary.
19 A no-effect boundary could be, for example, 80 to 125.
20 Now, if the answer is yes -- we think it is the no-effect
21 boundary -- then there's no dose adjustment required for
22 the special populations. But if the answer is no, which 1 30 2 1 means the 90 confidence interval is outside the boundary,
2 then we ask the next question, whether we have a PK/PD
3 relationship.
4 If we do have a PK/PD relationship, then we
5 will take the standardized approach to estimate the
6 probability of adverse events and probability of
7 effectiveness in the special populations and ask the
8 question whether that's a clinically significant change
9 from the typical population. If it is considered
10 clinically significant, then we will recommend a dose
11 adjustment or precaution or warning in the drug label.
12 As I mentioned, there will be several examples
13 discussed in today's meeting. The first two examples will
14 be used to illustrate the use of the proposed standardized
15 approach for estimating the probability of toxicity in
16 special populations using exposure-response information.
17 And the next example will be used to
18 illustrate the potential utility of population analyses to
19 obtain exposure-response information from large clinical
20 trials. We think this is a very important topic because
21 the large clinical trials represent a unique opportunity
22 to obtain exposure-response information from the studies. 1 31 2 1 The last example will be used to demonstrate a
2 method of applying utility functions to optimize a dosing
3 strategy.
4 Today we have four speakers to present the
5 examples. The first speaker is Nhi Nguyen from DPE I, and
6 she will present the first example. The second speaker is
7 Dr. Jenny Zheng from DPE III. She'll present a second
8 example of the standardized approach. And the third
9 presenter will be Dr. He Sun from DPE II, and he will
10 present the population PK/PD approach. The last speaker,
11 our guest speaker, is Dr. Mats Karlsson from Uppsala
12 University, and he will present an example illustrating
13 the utility functions.
14 After each presentation, we're going to ask
15 one or two questions to the committee. I'd like to
16 present the questions now so that hopefully the committee
17 can keep those questions in mind when listening to the
18 presentations.
19 The first two questions relate to the
20 standardized approach. We would like to ask, under what
21 treatment circumstances, for example, intrinsic or
22 extrinsic factors or therapeutic areas, would this 1 32 2 1 standardized approach not be applicable? We also ask a
2 second question. Does the exposure response differ
3 between special populations and typical populations? If
4 so, how can the differences be detected?
5 The next questions will be related to the
6 population PK/PD analysis, and there are two questions
7 related to that topic. The first question is, what are
8 the utility and general limitations of linking
9 pharmacokinetics obtained from the population analysis to
10 the response endpoints? And the second question is what
11 are the general considerations in using exposure response
12 for dose adjustment in special populations, especially
13 using the population approach to obtain the exposure
14 response?
15 The last question is related to the utility
16 functions, and the question will be, can the presented
17 approach of utility function be generalized to other
18 scenarios?
19 So with that, I want to introduce our first
20 speaker of the examples, Dr. Nhi Nguyen. She will present
21 the first example of the standardized approach.
22 DR. NGUYEN: Good morning. This morning you 1 33 2 1 will hear a presentation on a method of analysis and how
2 it was applied in regulatory decision making.
3 This slide will summarize how we did the
4 analysis for this NDA.
5 The first step is to develop your exposure-
6 response models. The exposure-response models should be
7 based on large, randomized clinical trials, trials that
8 explore a wide exposure range and include a large number
9 of people and are of the longest duration.
10 The second step is to have an expectation for
11 your target window of exposure. How much benefit does one
12 need and how much risk is one willing to assume?
13 The third step is to example what happens to
14 exposure response when various intrinsic and extrinsic
15 factors are introduced. Typically studies in special
16 populations include pharmacokinetic information and are
17 underpowered to provide good response data. So with the
18 appropriate assumptions about exposure response in these
19 special populations, we took the data from the special
20 population studies, the individual data, not just the mean
21 data, and integrated it into the exposure-response models
22 and then determined the probability of effectiveness and 1 34 2 1 safety.
2 So probability is on the y axis here and the
3 sum of these bars equals 100 percent. So you can see that
4 we not only have a feel for the maximum likelihood of
5 benefit or risk, but we also have a feel for the tails of
6 the distribution.
7 This slide summarizes how we did the analysis
8 for this review.
9 I will take one more slide to further explain
10 the last step, determining the probabilities. And for
11 this example, I've chosen the QTc interval which is a
12 surrogate for torsade, a fatal ventricular arrhythmia. A
13 clinical trial that examines QTc prolongation may have a
14 distribution of baseline QTc intervals that look like
15 this. Modeling the data may result in concentration QTc
16 slope distributions that look like this. So if we want to
17 see what happens when an intrinsic or extrinsic factor is
18 introduced, we took the results from the PK study, and for
19 this example I'm illustrating Cmax and overlaid it into
20 the known concentration-QTc relationship.
21 So, in essence, we sampled from each of these
22 distributions and created a virtual patient, and we did 1 35 2 1 this 1,000 times to determine empirically what happens
2 with a concentration and achieving a specific QTc. So by
3 doing these simulations, we were able to test combinations
4 of the tails of the distributions that were untested.
5 So that leads me to our objective which was
6 only to quantitate the risk-benefit of a drug. A decision
7 about what to do about the risk-benefit should be made
8 with the whole review team or the domain experts.
9 In developing the exposure-effectiveness
10 model, for the primary endpoint, we chose the largest
11 clinical trial which was also of the longest duration.
12 The primary effectiveness endpoint and pharmacokinetics
13 were collected at baseline, week 2, and week 4. This
14 study also explored the largest exposure, and these doses
15 have been changed for the purpose of this presentation,
16 but let's just say that 1 milligram a day is the sponsor's
17 recommended starting dose, with titration to 2 milligrams
18 a day. So this study explored a dose greater than and a
19 dose less than the sponsor's recommended dose.
20 Next we developed the exposure-safety or
21 exposure-risk models. For these models, we used the
22 adverse event data from all the pivotal clinical trials. 1 36 2 1 Now, you can imagine that large clinical trials may have
2 hundreds of adverse events. So after discussions with
3 other members of the review team, we prioritized these
4 adverse events and focused on these six: dizziness,
5 edema, liver toxicity, palpitations, tachycardia, and
6 vertigo.
7 We also analyzed QTc prolongation because of
8 drug properties suggestive of QTc prolongation. For this
9 analysis, we chose the study that had the most information
10 on the time course of QTc prolongation. You will note
11 that this was a drug-drug interaction study, and the
12 sponsor used half the recommended starting dose. ECGs
13 were only measured up to 4 hours post dose, so there were
14 some study design limitations. And the study contained 24
15 hours of drug concentration data.
16 So now that you've seen the exposure
17 effectiveness and the exposure risks that were assessed,
18 let's take a look at the models.
19 This is the exposure-primary effectiveness
20 model, and the asterisk in the following slides will
21 indicate the mean Cmax of the sponsor's recommended
22 starting dose of 1 milligram. These lines indicate the 1 37 2 1 mean Cmax for the 1, 2, and 4 milligram dose, and you will
2 note that the increase in concentration is more than dose
3 proportional. The maximum effect was 7.6 and the
4 concentration that produced half the maximal effect was
5 about .2 units.
6 In the following slides, I show a mean Cmax
7 line to keep the slide clean, but really we are
8 considering the entire distribution of individual Cmax's.
9 So it's something that may look like this with some
10 overlap between the 1 and 2 milligram dose. So that's the
11 exposure-effectiveness model.
12 When you look at the risks, each blue line on
13 this slide indicates one individual's concentration/QTc
14 prolongation relationship. The QTc corrections shown here
15 are individual corrections obtained by nonlinear mixed
16 effects modeling.
17 There was a statistically significant
18 relationship between concentration and QTc prolongation.
19 However, you can see that there is a lot of variability.
20 The starting dose in some patients results in no QTc
21 prolongation. However, in other patients, it results in
22 substantial QTc prolongation. You will also note that we 1 38 2 1 have very little data around the concentration of the mean
2 Cmax for the 2 milligram dose.
3 For our analysis of other adverse events, we
4 found three adverse events to be statistically significant
5 and that was tachycardia, palpitations, and edema.
6 However, since the analysis of all these adverse events
7 was similar, I will only present one for the sake of time.
8 This slide shows the probability of
9 tachycardia by the effective dose, and the effective dose
10 is an adjustment of the actual dose to account for the
11 saturable first pass of the drug. So, 1, 2, and 6 really
12 correspond to 1, 2, and 4 milligrams of drug.
13 The probability of tachycardia was dependent
14 on weight and dose. So in a 70-kilogram patient, you can
15 see that there is a very small probability of tachycardia,
16 and this probability does not increase with a six-fold
17 effective dose increase. Whereas, in a 50-kilogram
18 patient, there is about a 5 percent probability of
19 tachycardia, and then this increases about 1 percent with
20 a six-fold effective dose increase.
21 So now you've seen the exposure-effectiveness
22 model, the exposure and QTc model, and then the 1 39 2 1 probability of tachycardia by effective dose. So now
2 we're equipped with the models necessary to interpret
3 results of the special population studies.
4 This slide shows results of two special
5 population studies presented in terms of changes in AUC
6 and Cmax. Ketoconazole with a half a milligram of drug
7 resulted in a 13-fold increase in AUC and a 7-fold
8 increase in Cmax, and grapefruit juice with 1 milligram of
9 drug resulted in a 7-fold increase in AUC and a 6-fold
10 increase in Cmax.
11 So the next step is to see how these data
12 integrate into the known exposure-response relationship.
13 This is the same figure you saw earlier, only it's
14 smaller, of the concentration effectiveness. When a half
15 milligram of drug is given, you would expect to see an
16 effectiveness around 3. Taking ketoconazole with a half
17 milligram of drug increases concentrations about 7-fold,
18 reaching the Emax of about 7.6. Taking 1 milligram of
19 drug with grapefruit juice results in about a 6-fold
20 increase in concentration, and no additional
21 effectiveness. Now, hypothetically if ketoconazole were
22 given with 1 milligram of drug, we might expect to see a 1 40 2 1 similar increase in concentration.
2 If we look at the concentration/QTc
3 relationship, a half milligram of drug would result in
4 about this amount of QTc prolongation. Taking
5 ketoconazole with a half milligram of drug pushes patients
6 from this amount of QTc prolongation over to this amount.
7 Taking 1 milligram of drug with grapefruit juice results
8 in about a 6-fold increase in concentration, and the
9 concentrations are then off the figure and we do not have
10 QTc data there. And then hypothetically again, if
11 ketoconazole were given with 1 milligram of drug, we might
12 expect to see a similar response. So in this situation,
13 we would not be able to make any conclusions about what
14 happens at these higher concentrations on QTc prolongation
15 because we do not have data there.
16 Now, I also want to remind you again that
17 there is a distribution of Cmax's and slopes. So really
18 we are looking at data that looks like this. So if we
19 want to consider the worst case scenario, we have to
20 consider both of these distributions, and the results of
21 some of those simulations will be presented in the next
22 slide. 1 41 2 1 Now, if we look at the probability of
2 tachycardia, taking a half milligram of drug with
3 ketoconazole in a 50-kilogram patient barely increases the
4 probability of tachycardia, whereas taking grapefruit
5 juice with 1 milligram of drug increases the probability
6 of tachycardia about 1 percent in the 50-kilogram person.
7 In a 70-kilogram person, you can see that this slope is
8 pretty much a straight line and there is not much of an
9 effect on the probability of tachycardia. Then again if
10 ketoconazole were given with 1 milligram of drug, we might
11 expect to see a similar response as that with grapefruit
12 juice.
13 So to summarize the data integration,
14 ketoconazole with a half milligram of drug results in a
15 13-fold increase in AUC and a 7-fold increase in Cmax.
16 And this translated into a 4-unit effect, and we did
17 simulations to determine that 5 percent of the population
18 may have a prolonged QTc greater than 32 milliseconds.
19 Realize that these simulations are determined from the
20 data in the ketoconazole study. So we could present this
21 data in other terms, such as change from baseline or
22 percent or the time above a certain threshold QTc. 1 42 2 1 And then the probability of tachycardia with a
2 half milligram of drug and ketoconazole was barely
3 increased or affected. Grapefruit juice with 1 milligram
4 of drug resulted in a 7-fold increase in AUC and a 6-fold
5 increase in Cmax, and this translated into no additional
6 effectiveness, and we were unable to conclude anything
7 about the effect on QTc because we did not have data at
8 those higher concentrations. And then the probability of
9 tachycardia increased about 1 percent in the 50-kilogram
10 patient, and there was negligible effect in the 70-
11 kilogram patient. And then again if ketoconazole were
12 given with 1 milligram of drug, we may expect to see a
13 similar response as that seen with grapefruit juice.
14 So at this point we would present the table to
15 the review team and weigh the effectiveness and risks and
16 realize the assumptions of our models. One is that it is
17 the higher drug concentrations, not the intrinsic or
18 extrinsic factor itself, that alters response. We
19 recommended to conduct an appropriate QT study, one that
20 explores a wider concentration range and one that collects
21 QT data over 24 hours.
22 DR. VENITZ: Thank you, Nhi. 1 43 2 1 We have time for questions. Go ahead.
2 DR. SHEINER: I gather that the various parts
3 of the various models were gathered from different data
4 sets sometimes.
5 DR. NGUYEN: For the effectiveness, we used
6 the largest clinical trial, and then for the safety and
7 risks, we used the largest clinical trial and the other
8 pivotal trials.
9 DR. SHEINER: I guess the question is this.
10 You've got several models that are translating from A to B
11 and then B to C and so on.
12 DR. NGUYEN: That's correct.
13 DR. SHEINER: And the question is, were enough
14 of them gotten from the same set of people so that you
15 could look at things like correlations? Does it turn out,
16 for example -- not that it should -- that the people with
17 the high concentrations essentially -- in other words,
18 your model is kind of assuming that this association that
19 you see, there are no correlations. So it's not
20 necessarily true that somebody who has, let's say, a raise
21 in level and doesn't have toxicity will also not have
22 efficacy or something like that. 1 44 2 1 You've got a set of relationships that
2 translates from concentrations before you add
3 ketoconazole, let's say, to afterwards, and then you map
4 from concentration to effect, but there is no part of this
5 thing that says, well, when you raise the concentration to
6 the ketoconazole, maybe the relationship of concentration
7 to effect is not the same. I'm not saying that it is, but
8 you don't have any evidence to say one way or another. Is
9 that right?
10 DR. NGUYEN: Yes.
11 DR. VENITZ: Any other questions?
12 DR. FLOCKHART: One thing directly to that
13 point. There is some data -- but this might be
14 addressable -- to suggest that ketoconazole itself can
15 affect cardiac repolarization.
16 DR. NGUYEN: That's right.
17 DR. FLOCKHART: You might have data on that
18 from the control arms of the smaller trials. If they show
19 no effect, that would be reassuring. It doesn't
20 completely get to Lew's point because it's still possible
21 that the concentration-effect relationship is different in
22 the presence of ketoconazole. 1 45 2 1 DR. CAPPARELLI: Just a clarification so that
2 I understand the terminology. When you say probability of
3 tachycardia, you're speaking of sinus tachycardia in this
4 case, not torsade de pointes?
5 DR. NGUYEN: That's correct, sinus tachycardia.
6 DR. CAPPARELLI: I just wanted to be clear
7 because I think one question that I had is, did you take a
8 similar approach to heart rate that you did to QTc?
9 Because your sinus tachycardia is going to be relative to
10 where you start from, at least the risk.
11 The one thing that I think was brought up as a
12 question is, are there extrinsic factors that we need to
13 think about in these models? Clearly, strictly from a PK
14 standpoint, I wouldn't expect a 50-kilogram person to have
15 a different response based on weight. So I think there
16 clearly is an extrinsic factor that's linked to the 50-
17 kilo patient rather than the 70-kilo patient. But I'm not
18 certain that it's gotten here. So I have some questions
19 about the classification scheme based strictly on weight,
20 and maybe linking to where their baseline heart rates
21 would be of help from that standpoint.
22 DR. NGUYEN: Actually for the tachycardia 1 46 2 1 analysis, we would have preferred to analyze it by heart
2 rate, but the data were collected like that. So it was
3 sinus tachycardia, yes or no. So we did a logistic
4 regression.
5 DR. CAPPARELLI: Was there an age effect or
6 other disease effects that you looked at in terms of heart
7 failure? It's kind of difficult to look at this and see
8 where you expect a large change in concentration such that
9 the 70-kilo person at the highest dose is going to have
10 much higher concentrations than the 50-kilo person at the
11 lowest dose. And yet, you're seeing this differential PD
12 response. Without understanding what's causing that, I
13 think it becomes very difficult to extrapolate from this
14 aspect of the analysis.
15 DR. NGUYEN: Probably the 50-kilogram person
16 did receive more of a dose, milligram per kilo, than the
17 70-kilogram person. But the analysis -- they were given
18 the same dose. So we didn't have concentration data to
19 analyze data by concentration and probability of
20 tachycardia. We only had dose data.
21 DR. SHEINER: Just a comment. I think we're
22 getting at the fundamental problem that what you want to 1 47 2 1 do is extrapolate to new circumstances, people having
2 these other co-factors. You want to get some reasonable
3 guess as to what's going to happen, what's dangerous and
4 what isn't. Yet, you're extrapolating from observational
5 data based models, which is the hardest thing to do
6 because you don't know where causality resides in those
7 models.
8 One of the solutions in the past is to just
9 not do it, and I don't think that's the right solution.
10 But I do think there is no easy solution, and we have to
11 be quite careful about things and recognize that we're
12 talking about outer boundaries and recognize that we're
13 talking about sort of worst case scenarios or maybe even
14 best case scenarios. We can't be sure. We have to
15 somehow get comfortable with the increased degree of
16 uncertainty that arises in this activity. But as I say, I
17 think we should do it because the alternative is even
18 greater uncertainty.
19 DR. DERENDORF: There are a lot of straight
20 lines in your concentration-effect and dose-effect
21 relationships. Is there enough evidence that you have
22 linear relationships between those parameters, 1 48 2 1 particularly when you use them to extrapolate?
2 DR. NGUYEN: Which one are you referring to?
3 DR. DERENDORF: The concentration-QTc
4 prolongation plot and then also the one below the dose-
5 tachycardia plot. You just have straight lines there that
6 suggest that concentration and effect are linked that way.
7 Do you know that?
8 DR. NGUYEN: No. I mean, that was the data
9 that we had. So like for the QTc, they measured ECGs at
10 0, 1, 2, and 4 hours post dose, and they had 24 hours of
11 concentration data. So that straight line is the
12 relationship between the concentration and QTc.
13 DR. DERENDORF: You think it is or you know it
14 is?
15 DR. NGUYEN: Well, that's what it was in that
16 population. They could have gone with a higher dose
17 range, and then we could have seen what type of model the
18 relationship is. So I don't know.
19 DR. DERENDORF: Then the other question that's
20 related to the previous question that I really am puzzled
21 with is this 50- versus 70-kilo situation where you have a
22 6-fold dose and a 70-kilogram person doesn't have 1 49 2 1 probability versus a one-sixth of a dose in a 50-kilogram
2 person. So there would be quite different exposures and
3 very different risks.
4 DR. FLOCKHART: It just means to me that the
5 relationship between dose and weight isn't a simple linear
6 relationship. You're right. There may be something else
7 involved, but that's not uncommon.
8 DR. SADEE: I have a comment also on the
9 variability from one patient to the next. You're looking
10 at two interactions. It's maybe one of the classic
11 examples in pharmacogenetics where you have a variety of
12 different genetic variance from one patient to the next,
13 and actually that could affect the interaction between the
14 two drugs so that you may have specific cases in the
15 single patients that are totally different in their
16 exposure than others. So there would be no way of
17 extrapolating from that because you're disturbing the very
18 relationship with the dose response that you're looking at.
19 DR. FLOCKHART: My difficulty here is the FDA
20 is faced with the problem of trying to make a rational
21 prediction. We can sit as academics and ding them all
22 over the place for it. You know, you can't do this and 1 50 2 1 you can't do that. But the reality is you have to try.
2 And I think Lew's point is salient. We have to try and
3 include the error.
4 DR. KEARNS: And that's the point that I think
5 I want to make. Back to your ECG slide. It's not to be
6 critical. It's quite exciting to see 20 percent of people
7 have a response that way and then one outlier at the top
8 who really had one.
9 But I was struck by the recommendation that
10 you showed on your slide which was, okay, we did this.
11 Now we go back and recommend a trial with more
12 concentrations, wider range. And as Dr. Sheiner was kind
13 of getting at, there's a lot that's riding on this
14 extrapolation. I think from a public side, there's always
15 the question of how much additional time is that going to
16 take. From a medical side, there's always the question
17 about that one or two outliers. Are those the people who
18 die in that trial because they have some hERG channel
19 defect that's not recognized at the outset?
20 So I think trying to go back and do the
21 diligent thing is to wire up the model as best you can.
22 For instance, if that was a pediatric population and you 1 51 2 1 looked at baseline QTc's, you'd see quite a different
2 dispersion based on age for no treatment than you would in
3 an older free-living population.
4 So is the applicability of the model approach
5 -- can we go across populations? It depends I think. But
6 to jump to the study and to add the patients, to add the
7 concentrations, to maybe add the risk until you've taken
8 all the flies that you can out of the ointment could be
9 premature.
10 DR. VENITZ: Thank you, Nhi.
11 Let's move to the next presentation. Dr.
12 Jenny Zheng. She is a pharmacometrics reviewer in the
13 Division of Pharmaceutical Evaluation III. She's going to
14 give us a second example.
15 DR. ZHENG: Good morning. Today I'm going to
16 present another example to illustrate how the dose-
17 response relationship was used for recommending dose
18 adjustment.
19 In our review process, it's very often to see
20 the pharmacokinetics of a drug is influenced by intrinsic
21 factors such as age, gender, impaired renal and hepatic
22 function and extrinsic factors such as drug-drug 1 52 2 1 interactions. In this situation, we have to ask the
2 question what is the clinical significant of the changes
3 in concentrations.
4 Currently the decision will be made based on
5 the clinical assessment based on the clinical experience
6 and totality of the evidence. But the disadvantage of
7 that approach is it's not a quantitative and standardized
8 approach. The assessment could be pretty subjective. In
9 other words, the decision may not be the same based on who
10 makes the assessment and from what perspective.
11 Therefore, we propose from a clinical pharmacology
12 perspective to use the exposure-response relationship to
13 bridge the response and exposure data to quantitate the
14 influence after changes in the exposure.
15 The example I'm going to present will focus on
16 the drug concentration increase and the safety assessment.
17 This is drug Z. It's a noncardiac drug. From both
18 preclinical and phase I studies, it shows that the drug
19 caused QT prolongation and this QT prolongation is
20 concentration dependent.
21 The phase I PK studies showed three factors
22 increased the drug concentration. In an age study, it 1 53 2 1 shows that the mean steady state maximum concentration was
2 100 percent higher in elderly subjects as compared with
3 Cmax in young subjects. The renal study demonstrates
4 steady state Cmax in severely renally impaired subjects
5 was 50 percent higher as compared with healthy subjects.
6 And drug interaction studies showed ketoconazole increased
7 the steady state Cmax by 60 percent.
8 Knowing the concentration increase in this
9 situation, the question raised is, should dose be adjusted
10 in elderly, renally impaired subjects or when co-
11 administered with ketoconazole? To answer that question,
12 we need to understand the effect of increase in drug Z
13 concentration on the QT prolongation which will rely on
14 the exposure-response relationship.
15 The exposure-response relationship was
16 obtained from several phase I studies. They were all
17 placebo-controlled crossover studies. The doses included
18 in the study were a clinical dose and two times of the
19 clinical dose and three times of the clinical dose. The
20 higher dose is important here to provide the wide range of
21 the concentration which is the key for obtaining exposure-
22 response relationship. From all these phase I studies, 1 54 2 1 blood samples were collected for drug measurement. Also
2 the QTs were measured.
3 The results of the analysis are shown in this
4 slide. The QT prolongation is represented as delta QTc,
5 which is the QT change from the baseline. So the
6 association between the delta QTc with the concentration
7 was described by a simple linear regression model. The
8 linear mixed effect model was used for analyzing these
9 data. The dashed lines represent individual regression
10 lines. The solid line represent the population regression
11 line. The wider band of the lines indicates that the
12 inter-subject variability is quite high. The estimated
13 slope ranged from 1.5 to 7.6, indicating that for some of
14 the subjects, the delta QTc change is sensitive to the
15 concentration change. In some of the subjects, the change
16 is not quite as sensitive.
17 An outlier analysis is a very important part
18 of QT assessment. We want to know how many subjects would
19 experience the delta QTc longer, for example, 10
20 milliseconds, 20 milliseconds, 30 milliseconds, or 40
21 milliseconds. Unfortunately, the phase I study usually
22 included a limited number of subjects which limits its 1 55 2 1 ability for that type of analysis.
2 For the phase III study, even though hundreds
3 of subjects may be included in that analysis, the QT
4 measurement is not as intensive as the phase I study. So
5 it's difficult sometimes to capture the outlier from the
6 phase III study. On the other hand, if you're interested
7 in the special population, even a phase III study may not
8 provide sufficient number, for example, severely renally
9 impaired subjects.
10 So in order to make an outlier comparison
11 between the population, a simulation exercise was
12 conducted here. Most specifically, the phase I data, the
13 concentration data was modeled assuming the logarithmic
14 distribution in the PK parameter. Using that model, 2000
15 maximum concentration was simulated for young subjects,
16 elderly subjects, for renally impaired subjects. The same
17 approach is used to simulate 2000 Cmax for when
18 ketoconazole is co-administered with the drug Z. So we
19 have 2000 concentration in each special population,
20 special situation. Then we used the exposure-response
21 relationship as described in the previous slide to predict
22 delta QTc. 1 56 2 1 The results of the age effect are presented in
2 this slide. The data is presented as the percent of the
3 subjects who would have the delta QTc longer than 10
4 milliseconds, 20 milliseconds, 30 milliseconds, 40
5 milliseconds. These results indicated that about 2
6 percent of young subjects would have delta QTc longer than
7 40 milliseconds and for the elderly, the percentage will
8 increase to 7.3 percent. So for delta QTc longer than 30
9 milliseconds, in young subjects it's about 8 percent; in
10 the elderly, it's about 19 percent. So a similar trend
11 could be seen for a delta QTc longer than 20 milliseconds
12 and 10 milliseconds.
13 This slide presents the results for renal
14 function. As you can see, most subjects with severe renal
15 impairment would have longer QT prolongation than the
16 normal subjects. For example, for the normal renal
17 function subjects, 2 percent would have delta QTc longer
18 than 40 milliseconds, but if you have severely renally
19 impaired function due to the concentration increase, there
20 will be 5 percent of the subjects who would have a delta
21 QTc longer than 40 milliseconds.
22 The results in this slide show ketoconazole 1 57 2 1 increased the percent of subjects who experienced a
2 certain extent of delta QTc. Like if you're taking the
3 drug alone, 2 percent of subjects would experience delta
4 QTc longer than 40 milliseconds. But if you take drug Z
5 with the ketoconazole, the percentage will increase to 6.2
6 percent.
7 The percent of subjects with delta QTc longer
8 than 40 milliseconds is summarized in this slide. You can
9 see that the risk of having a delta QTc longer than 40
10 milliseconds is higher in elderly subjects, in severely
11 renally impaired subjects, and when the drug is co-
12 administered with ketoconazole.
13 Examination of the creatinine clearance
14 indicated that the age effect might be partially
15 attributed by reduced renal function. So in the age
16 study, creatinine clearance was 50 percent lower in
17 elderly subjects as compared with the young subjects. So
18 it's believed that age effect would be reduced if the
19 renal function effect was corrected by dose reduction.
20 Since the consequence of the worst event could
21 be very severe, the question was asked in the review team,
22 what would be the effect of ketoconazole in subjects with 1 58 2 1 severe renal impairment? Not many subjects would belong
2 to this group, even from a phase III study. So in order
3 to make that assessment, a simulation was conducted.
4 First, steady state Cmax in severely renally
5 impaired subjects was simulated as I described earlier.
6 In the second step, the Cmax ratio of drug Z at presence
7 and absence of ketoconazole was obtained from the
8 crossover study so that the ratio actually characterized
9 the ketoconazole effect. From that study the ratio ranged
10 from 1 to 4. So the combined effect for both factors was
11 simulated by just randomly multiplying the maximum
12 concentration from step 1, which is the maximum
13 concentration for severely renally impaired subjects, and
14 the ratio from step 2 which characterized the ketoconazole
15 effect.
16 The results are shown in this slide. As you
17 can see, 19 percent of subjects who are severely renally
18 impaired would experience a delta QTc longer than 40
19 milliseconds when co-administered with ketoconazole.
20 This slide just simply summarizes all of the
21 factors, the effects. It summarizes young subjects. It's
22 the percent of subjects with a delta QTc longer than 40 1 59 2 1 milliseconds. For young subjects, it's about 2 percent.
2 In the elderly, it's almost triple the percentage, up to
3 7.3 percent, and more than double that percentage in
4 severely renally impaired subjects. And when drug Z is
5 co-administered with ketoconazole, the effect could be
6 very dramatic if the two factors are combined.
7 That analysis leads to the conclusion that the
8 increase of concentration by age, severe renal function,
9 and co-administration with ketoconazole resulted in
10 increased number of subjects with a delta QTc longer than
11 40 milliseconds. The effect is more significant when two
12 factors are combined.
13 Based on that analysis and the consideration
14 of the nature of an adverse event, a dose reduction was
15 recommended in severely renally impaired subjects. A dose
16 reduction was also recommended when drug Z is co-
17 administered with ketoconazole.
18 DR. VENITZ: Thank you, Jenny. We have about
19 5 minutes for questions.
20 DR. DERENDORF: I think it's the same issue as
21 in the last case. You're assuming that the exposure-
22 response relationship that you got from your phase I study 1 60 2 1 is a constant and it doesn't change in elderly or in
2 severe renal impairment. So the calculations that you're
3 making are all focused on exposure, and then at the very
4 end, you convert that into expected --
5 DR. ZHENG: I don't quite understand your
6 point. Actually the delta QTc versus concentration, that
7 is the relationship between the effect versus
8 concentration. I don't think we make any assumption with
9 a constant concentration.
10 DR. DERENDORF: No, not constant
11 concentration. But the relationship between the exposure
12 that you have in your different cases and the outcome --
13 you take that linear relationship that you have from your
14 phase I study where you have concentration versus change
15 of QTc and apply that to all of these cases assuming that
16 this relationship holds true for all of these.
17 DR. ZHENG: Okay. So you have the problem
18 with the extrapolation from the young healthy subjects to
19 the elderly population.
20 DR. DERENDORF: I don't see any evidence that
21 it holds.
22 DR. ZHENG: In one of the phase I studies, 1 61 2 1 they included not only the young subjects but the elderly.
2 So we do look at the relationship there. We don't see
3 much difference in terms of the slope, the relationship.
4 So that's one of the information we could have.
5 In terms of the drug-drug interaction and the
6 severely renally impaired subjects, yes, we don't have
7 data to show that they are going to have the same
8 relationship. But you are right. That's the assumption
9 we have to make for these type analyses.
10 DR. CAPPARELLI: Just as a follow-up on that,
11 because this is a recurrent issue -- I mean, this
12 particular QT prolongation looking at drug concentration
13 effects -- has there been a systematic look at several of
14 these drugs looking at especially, say, with renal
15 impairment where you're going to have changes in
16 electrolyte abnormalities and looking really at sort of a
17 population dynamic model to identify the covariates?
18 So I think as Hartmut was saying, we're going
19 forward with the assumption that the exposure-response
20 relationship is totally uncorrelated with the changes in --
21 the disease states that are causing the changes in PK. I
22 think this is actually a great example where maybe in some 1 62 2 1 across-study evaluations, one could actually look at some
2 of these populations not only at the variability in
3 response in subpopulations, but maybe the electrolyte
4 differences in your renal failure patients may change that
5 slope entirely. So adding these effects as we go along in
6 the chain, it's nice along the way to test some of these
7 assumptions.
8 DR. ZHENG: I think if we could have enough
9 information, definitely that's a good thing to do. But I
10 think here, unfortunately we just don't have that much
11 information. So the focus here is simply the effect of
12 concentration on the delta QT.
13 DR. SHEINER: Of course, the answer is get
14 more information. But the answer to the problem when you
15 don't have enough information is not that you have to make
16 some assumption and go with it, but that you have very
17 carefully display. It is sort of what I was indicating
18 before. You have to carefully display the limits of your
19 knowledge.
20 So if I look at, for example, the last page
21 and the last several slides you showed, you've got these
22 bars that are just heights, the amount of change with the 1 63 2 1 elderly or renal failure, and there are no uncertainty
2 intervals on them. Yet, this is exactly what you need to
3 pay a lot of attention to, it seems to me, in this kind of
4 a situation so that you can have a rational dialogue with
5 other people.
6 And where do the uncertainties come from? And
7 we have techniques whereby you say, well, look, I have
8 this linear relationship between QTc change and the
9 concentration, but I could put in some uncertainties
10 about, let's say, whether it applies to other populations.
11 Then I can actually build that into my projections, and I
12 can see that instead of having 30 percent of people above
13 QTc of 40, it will be anywhere from 10 to 50 percent, or
14 whatever the numbers are.
15 That's the point. You've got the computers to
16 do it. It doesn't cost money. And that's the way I think
17 to deal with the problem that there are so many
18 assumptions that have to be made, sensitivity analyses and
19 honest uncertainty intervals which involve model
20 uncertainty as well as data uncertainty. And then
21 everybody is talking about the same thing. It may well be
22 that the conclusion stays the same. 1 64 2 1 DR. LEE: Dr. Sheiner?
2 DR. SHEINER: Yes.
3 DR. LEE: To follow up, if we don't know the
4 true relationship in different populations, how do we
5 build into the model the uncertainty due to population
6 difference?
7 DR. SHEINER: Well, let's say we'll talk about
8 the average slope. Let's say you're willing to assume
9 it's linear, but it's the average slope that's different
10 in different populations. So then you just talk to a
11 bunch of people and you say, how big do you think it could
12 be, and you just build that uncertainty in. Now it
13 spreads out all of your predictions, and it means there's
14 a larger fraction of people who have low values, but
15 there's a larger fraction of people who have high values.
16 So it's a matter of assessing the risk. It's
17 what's the probability based on everything we know,
18 including all the uncertainty, that the value will be
19 greater than this. And that will be your most educated
20 guess.
21 The point is it will be everybody's most
22 educated guess, and anybody who says I don't think that 1 65 2 1 will happen, you'll say, well, you're pointing to the 40
2 percent that's still below the line because we have
3 uncertainty. And I understand you're betting on that 40
4 percent, but we're worried about the 30 percent. So
5 that's the way we're going to go.
6 The point is you're never going to get the
7 answer from doing the numerical calculations. All you get
8 is an honest statement of what you know and that everybody
9 can agree on. I think that's the big thing, is that
10 everybody can agree this is the state of our knowledge.
11 Therefore, if you're going this way and I'm going that
12 way, it's because we're valuing different outcomes
13 differently, and so the expected value comes out
14 differently.
15 Another example here of a place for an
16 opportunity for this is in the discussion -- well,
17 actually, I'll let it go. But I think you get the idea.
18 DR. VENITZ: Let me just follow up to that,
19 Jenny. Whenever you do an outlier analysis, which is
20 really what you're trying to do, worst case scenario, what
21 are the few that have a large change in QTc, distribution
22 assumptions are very important in terms of what your final 1 66 2 1 outcomes are. I look at your simulation slide. You're
2 talking about a logarithmic distribution. I'm assuming
3 you mean a log normal distribution.
4 DR. ZHENG: Right.
5 DR. VENITZ: How did you then actually
6 simulate the changes due to the disease states or the
7 drug-drug interaction? Did you just change the mean or
8 did you change variances as well?
9 DR. ZHENG: Actually I just changed the mean
10 because the model is fitted to the raw data. For example,
11 the young subjects -- we modeled that. So we know the
12 mean for that group.
13 DR. VENITZ: But what about the variance? I
14 guess what I'm worried about, whenever you look at
15 outliers and you have a change in variance, in other words
16 your elderly or your renal population are probably more
17 variable than your young reference population even in
18 terms of kinetics.
19 DR. ZHENG: We used the same model to model
20 the data for young subjects and the data for elderly. I
21 mean, the same compartment model.
22 DR. VENITZ: But in terms of your parameter 1 67 2 1 variability, did you use the same variance in your --
2 DR. ZHENG: No. The data --
3 DR. VENITZ: So you used the actual data
4 variance.
5 DR. ZHENG: Yes. I used the real data
6 variance.
7 DR. VENITZ: Just to follow up on that, I'm
8 assuming when you looked at your slopes, you assumed that
9 the slopes followed normal distribution or log normal
10 distribution?
11 DR. ZHENG: I did that analysis using NONMEM.
12 So it's an additive model. So it's normal distribution.
13 DR. VENITZ: Well, based on what I've seen or
14 based on the previous example, that may not be a good
15 assumption. It could be that you just have a few outliers
16 and have very steep slopes, but the rest of them have a
17 fairly shallow slope. Whenever you do an outlier
18 analysis, just as a general rule, the distribution
19 assumption of the variances that you make really determine
20 what your final outcome is.
21 In addition to that, I would reinforce what
22 Dr. Sheiner said, and that is, I was missing the fact that 1 68 2 1 you didn't really give us an idea of the uncertainty --
2 DR. ZHENG: Right. I think that's something I
3 should have included. Probably I don't have enough
4 information to speak to severely renally impaired
5 subjects, what the relationship will be. But I do have
6 the information about uncertainty of the parameter
7 estimate. So I agree 100 percent.
8 DR. VENITZ: Just look at your three slides
9 where you tell us what happens for the young individuals.
10 Let's say the QTc of less than 10 is 41, 44, and 4 and
11 42.7.
12 DR. ZHENG: Right.
13 DR. VENITZ: Those are three different
14 simulations. So you just do that a couple times and you
15 know how much --
16 DR. ZHENG: Right, yes. The uncertainty of
17 that estimate should have taken into that exercise. I
18 agree.
19 DR. VENITZ: I think we have one more question.
20 DR. JUSKO: I imagine you're using the best
21 available metrics on evaluating the exposure-response
22 relationships. But you might consider using the 1 69 2 1 availability of these data to examine additional
2 possibilities. For example, if you look at absolute
3 changes in QTc, there might be the possibility that a
4 change of 10 or 20 in the elderly is a bigger problem than
5 a change of 10 or 20 in the young subjects. And perhaps
6 something in relation to baseline values should be
7 considered.
8 Secondly, you're using Cmax as the exposure
9 index, and it would seem to me that in addition to that,
10 the duration of time that a person has an abnormal QTc
11 interval could be an additional hazard that could be
12 factored in in exploring bigger sets of data as you may be
13 doing.
14 DR. VENITZ: Okay. Thank you. Thank you,
15 Jenny.
16 Before we go on a break, just an announcement.
17 For those of you on the committee who haven't handed in
18 your lunch orders, now is the time to do it or you're
19 going to starve.
20 With that, we're going to reconvene at 10:10.
21 (Recess.)
22 DR. VENITZ: I'd like to reconvene the meeting 1 70 2 1 please.
2 All right. While Peter is posting the
3 questions that the FDA is asking the committee, are there
4 any additional specific questions to the two presenters,
5 Dr. Zheng and Dr. Nguyen?
6 DR. KARLSSON: Yes. Just a question regarding
7 this last presentation. The elderly showed quite a change
8 when you looked at the distribution; 7.3 percent would be
9 above 40 milliseconds. Maybe that would be mitigated by
10 the renal impairment dose adjustment, although I guess
11 even in the elderly population, it wouldn't be that many
12 that's below 30 mls per minute in the elderly population.
13 But I guess you could look at that through simulations as
14 well.
15 But another question is, when looking at the
16 percentage of a particular subpopulation that's outside,
17 is it only the percentage within the population you're
18 looking at, not the size of the population at all?
19 Because I guess the elderly population is very large in
20 absolute numbers compared to maybe severe renal impairment
21 or ketoconazole.
22 Did I make myself clear? 1 71 2 1 DR. ZHENG: Actually could you repeat your
2 second question?
3 DR. KARLSSON: Well, if we're looking at dose
4 recommendations, is it only the percentage within the
5 population that's interesting? Is it not also the size of
6 the population as such?
7 DR. ZHENG: The simulation I did is 2,000
8 subjects. So it may change if you change the sample size
9 to --
10 DR. KARLSSON: No. In essence, what I mean is
11 that the elderly population is maybe like 80 million
12 people in the U.S. and the severe renal impairment
13 population is maybe 1 million. I don't know. 2 million.
14 So that would come into play as well when making dosing
15 recommendations.
16 DR. ZHENG: Okay. Yes, I think at the time we
17 make a decision, we should consider the population who use
18 the drug, the impact.
19 DR. LESKO: Mats, I'm not clear how you would
20 consider it, though. Would you consider it in the context
21 of saying that equal changes in a population that's larger
22 number would get more weight in a dose adjustment scheme? 1 72 2 1 Or how would you think about it as far as that issue goes?
2 It's like saying because the elderly
3 population is so large, there's a greater overall risk to
4 public health than there would be with patients with
5 severe renal function. But it would seem in labeling a
6 product, I'm not sure that would be taken into account for
7 dose adjustment. Or if it is, I'm not sure how it would
8 be.
9 DR. SHEINER: It would be if you were thinking
10 about what dose sizes to make, for example.
11 DR. LESKO: Okay, from a manufacturer's
12 standpoint.
13 DR. SHEINER: It's more convenience. If 90
14 percent of people are going to use this to make them safe
15 and 10 percent -- you know. They're the ones who are
16 going to get out their pocket knives and hack the thing in
17 half.
18 DR. VENITZ: Peter, do you want to review the
19 questions for the committee one more time?
20 So those are the questions that we are asked
21 to discuss with regard to the approach that we just two
22 examples of using exposure-response information as a way 1 73 2 1 of predicting probabilities of, in this case, adverse
2 events as a way of deciding about dosing adjustments or
3 not.
4 Lew?
5 DR. SHEINER: Did you want discussion on that?
6 DR. VENITZ: Yes.
7 DR. SHEINER: Well, I think we're back to
8 where we were sort of in the very beginning. It's a good
9 thing to do, but if it's not done with a little extra
10 care, then maybe it's not a good thing to do. So I think
11 it really comes down to that.
12 As I was saying to Jenny at the break, if you
13 do a well-designed, even clinical experiment in which you
14 know exactly the question you want to ask, you've got
15 adequate data by good design, and you analyze it, in a
16 funny way the statistics are relatively unimportant. The
17 signal to noise is usually pretty high, and it's usually
18 pretty clear what the result is. Yet, that's where most
19 of the statistics that most of us have seen have been
20 applied, in making sure that type I error is controlled.
21 And there's nothing the matter with that. It's a good
22 idea. But it's not really where you need it. And it's 1 74 2 1 not that you need sure inference here because you can't
2 get sure inference when you're this uncertain.
3 But what we need is we need to have a good way
4 of displaying what we know so that everybody is looking at
5 the same thing and understands the uncertainties. It
6 seems to me what we didn't see were two ways in which I
7 feel that that needs to be done.
8 One, as soon as you generate a simulation
9 model, you have to show me that that simulation model can
10 simulate the data it was derived from. There are lots of
11 different ways of going about convincing me of that. Some
12 of them treat the data it was derived from as though they
13 were new by leaving it out and then making the thing and
14 remaking. There are many, many different techniques. But
15 the fundamental idea is show me that the sorts of
16 conclusions that you want me to draw about extrapolations
17 are at least verified on the data that you built the thing
18 from when you apply them to those data. So that's number
19 one. I want to see a lot of that.
20 And then I want to see a real honest
21 uncertainty in my simulation. We all understand we're not
22 talking about anything that's sure here. But I want to 1 75 2 1 know how big the uncertainty is and I want to have some
2 way of knowing where it came from. I personally really
3 want to see model uncertainty as well as data uncertainty.
4 In fact, I'm more concerned about model uncertainty than
5 data uncertainty.
6 What do I mean by that? I mean if you have
7 100 patients from whom you've generated the data set, I
8 understand the next 100 patients are going to have
9 somewhat different numbers, and so you're going to get
10 somewhat different conclusions. And we all estimate that
11 uncertainty, and it's not that tough to estimate. And
12 sometimes it isn't that large because we have a fair
13 number of patients.
14 What's really uncertain is whether or not the
15 relationship we discovered on this population is going to
16 apply to that population. There we have no data if we
17 haven't studied that population, if we're extrapolating to
18 it. So there we need just some reasonable guesses. How
19 different have populations been with respect to this kind
20 of thing in the past with similar sorts of things? This
21 is where the science comes in. This is where the judgment
22 comes in. But you can build those model uncertainties in, 1 76 2 1 and I can get to see how big they are. That's kind of
2 like a robustness test. It's kind of like a way of saying
3 how much will conclusions vary if I vary my assumptions.
4 Assumptions there will always be. I'm not
5 against assumptions. What I'm against is making
6 assumptions look like facts. It turns out that the things
7 we don't know anything about we put the least uncertainty
8 on, and that's very peculiar. We choose a form of a
9 model. So it's a bi-exponential. And then we say, boom,
10 that's it. No questions about that. And that's the thing
11 we know the least well. What we do know well is the data
12 we observed, and that we say, aha, that's got noise. So
13 it's kind of like backwards. I want to see the model
14 uncertainty.
15 This is not to be critical. I believe in
16 modeling and I believe in trying to be quantitative about
17 conclusions. But without that kind of thing, you won't
18 ever get people around a table to agree on what you know,
19 and if you can't do that, they won't agree on where you
20 ought to go.
21 DR. VENITZ: Any other comments to the first
22 question? Can we think of any specific examples or 1 77 2 1 circumstances, therapeutic areas where this approach may
2 not be applicable?
3 DR. KEARNS: Yes. I think one glaring one --
4 and Dr. Sheiner again speaks of model uncertainty. As I
5 might understand it, it would be in the context of the
6 facts of an experiment that we saw examples of. But as
7 Dr. Flockhart mentioned, for QTc studies where a patient
8 may ingest a medicine that can have effects on its own,
9 that's not necessarily part of model uncertainty. I don't
10 know that you could predict the rate of co-ingestion of
11 those drugs. And I would argue that with some
12 combinations that are available, the relationships that
13 you so nicely shared with us could look quite different.
14 So are there treatment circumstances that the
15 approach might not be applicable as it was presented?
16 Yes, and I think that's one example.
17 DR. VENITZ: What about the second question?
18 I think that's something that we talked about last time.
19 Differences in the exposure-response relationship between
20 the typical and special populations.
21 DR. DERENDORF: Yes. I think that there are
22 some examples in the literature where there are clearly 1 78 2 1 differences in exposure-response relationship. If you
2 think of benzodiazepines, for example, with the
3 sensitivity and all the patient changes, so for the same
4 concentration, you'll get a different response. But
5 there's actually very little hard data available in the
6 literature because it's hard to study. If you want to do
7 it right, you have to do a complete PK/PD study, and just
8 focusing on exposure is simply easier and therefore it's
9 done more frequently. But I think that's what we need:
10 more clean PK/PD studies in different populations to see
11 how much variability and how much systematic change we
12 have in the exposure-response relationship.
13 DR. VENITZ: Larry.
14 DR. LESKO: There are actually two levels of
15 uncertainty that we're dealing with. The first -- and I
16 think it was the first example. We were talking about
17 exposure-response relationship across various special
18 populations. The second example illustrated a different
19 problem and that was the assumption of exposure-response
20 relationships between healthy volunteers and then
21 patients. That's just the fact of the way, at least
22 currently, drugs are developed. So we have to find ways 1 79 2 1 to think about that, and it would seem there are two
2 things I thought about.
3 One was in the pediatric decision tree or in
4 the pediatric rule, we make the assumptions, or at least
5 we ask the questions, about disease progression being the
6 same in adults and kids and whether or not the mechanism
7 of action in the exposure response is the same in adults
8 and kids, and then we proceed down a path of logic that
9 requires perhaps a dose being changed based on simply
10 pharmacokinetic differences to achieve the same type of
11 exposure.
12 It gets to the question, though, is my base
13 assumption that exposure response is similar in the
14 absence of hard data, and then I look for reasons, perhaps
15 mechanistic reasons, why it wouldn't be, or do I look and
16 say, well, let me assume it's different and find
17 mechanistic reasons that it should be the same? For
18 example, in the pediatric adult area, you might ask the
19 question, does a beta receptor's either density or
20 sensitivity change and is it safe to assume that with a
21 beta blocker I'm going to have similar exposure-response
22 relationships? 1 80 2 1 It just seems to me that there's a way to
2 think about it mechanistically if one understands the way
3 the drug is working and the changes that are occurring in
4 the special population. Like in the QTc, for example, if
5 renal patients have altered potassium levels, then we know
6 that affects sensitivity in terms of drug effects on QTc,
7 and that could be kind of a rationale for including some
8 assumption about heightened sensitivity or something like
9 that or a change in the exposure-response curve. But in
10 the absence of that kind of mechanistic information, it
11 would seem we have to go with the assumption that these
12 exposure-response relationships are the same.
13 I mean, does that line of thinking make sense?
14 DR. VENITZ: That would be the way I think.
15 My default position is there is no difference between my
16 typical population and the special population unless I
17 have either hard data to show that it is, which is rare,
18 or I have mechanistic reasons based on the pathophysiology
19 of the disease of that special population and/or the
20 mechanism of action of the drug to suspect that it is.
21 Then I either have to question the need for additional
22 studies to show whether it exists or not or build it in as 1 81 2 1 an uncertainty in my model.
2 DR. KEARNS: And Larry, I think another answer
3 to your question that you posed is it depends on the
4 surrogate chosen to assess effect. For example, if we
5 look at studying a proton pump inhibitor in a child,
6 there's convincing physiologic evidence that the
7 maturation of the proton pump occurs very early and that
8 the children respond to those medicines in ways that are
9 very similar to adults.
10 But if you go pick a surrogate far from the
11 tree of effect or drug action and you apply it and say, is
12 gastroesophageal reflux in a 6-month-old the same as in a
13 46-year-old, and then try to make arguments about
14 bridging, you'll find that the pillars that you've
15 constructed the bridge out of are not worth traversing.
16 So it depends on how close your surrogate is to where the
17 medicine works.
18 DR. VENITZ: Something else I think we're
19 going to talk about in a minute that I would also consider
20 -- and I'm pretty sure you do that in your briefings with
21 the medical reviewers -- is what are the consequences of
22 being wrong. In other words, what's the utility of 1 82 2 1 whatever assumptions you may not be very certain about?
2 Sometimes that severity or that consequence may be
3 relatively inconsequential, and then it really doesn't
4 make a difference. Forget the fact that you have
5 statistical uncertainty associated with it.
6 DR. LESKO: There are many ways these kind of
7 data are handled for purposes of dosing adjustment, and
8 maybe that's one of the reasons we're trying to arrive at
9 a standardized approach to doing it.
10 It would seem the safest way of doing it is to
11 simply adjust the dose based on an area under curve
12 change. The question then becomes what is the threshold
13 level for that area under curve change to trigger that.
14 And that's where the difference of opinion occurs because
15 you don't have a method on the table that allows one, as
16 people have said, to discuss this in a quantitative way.
17 So there may be, in essence, a lot of
18 likelihood of not optimal dosing by doing it that way,
19 either making dose adjustments when you don't need them or
20 not making them when you should based on people's
21 interpretation of the data without a methodology to
22 discuss. 1 83 2 1 DR. VENITZ: But in the examples that you've
2 shown, the endpoints, as far as I can understand them --
3 QTc. That's a surrogate of fatal arrhythmia. So you're
4 worried about a potential fatal consequence. There's a
5 high, in my terminology, negative utility associated with
6 it. On the other hand, things tachycardia or palpitations
7 would rank much lower on the totem pole of my concerns.
8 But I'm not sure how you quantitatively incorporate that
9 short of using utility functions.
10 DR. FLOCKHART: This is an editing, small
11 point. If you're going to talk about changes in AUC of a
12 compound, I think particularly when you're talking about
13 the QT -- but this may be representative of other things --
14 the area under the exposure curve is not necessarily the
15 main thing. The time of exposure to a drug is not the
16 trick. Parameters like the rate of rise to Cmax can be
17 very important and the QTc max at a given dose can be very
18 important. There are dis-relationships, blocks between
19 the time-effect curve so the time of the concentration
20 Cmax is absolutely not necessarily the time of the effect
21 Cmax. It can be later. So it's possible there would be
22 situations where a parameter other than a change in the PK 1 84 2 1 AUC would be the appropriate parameter. It could still be
2 a PK parameter, but it might be Cmax itself or it might be
3 the rate of rise to Cmax. And that would be a drug-
4 specific question.
5 If, for example, you looked at quinidine,
6 quinidine has a very poor relationship to the QT interval.
7 If you were able to talk about torsade, what really
8 matters there is the rate of rise, how quickly you get to
9 Cmax. And if you get to a very nasty Cmax very slowly,
10 it's not a terribly dangerous thing it looks like, but if
11 you get to the same Cmax very quickly, then it can be a
12 very dangerous thing.
13 That's a drug-specific question, and I would
14 caution about always using AUC. I mean, you can think of
15 examples related to Greg's example too. Above a certain
16 point, changes in the AUC of a proton pump inhibitor do
17 nothing. They're meaningless. I guess that just
18 emphasizes the point that you need to know the
19 pharmacodynamic relationship first.
20 DR. SHEINER: True as what you say is, I quake
21 at the notion that things as uncertain as area under the
22 curves, which are essentially integrals and consequently 1 85 2 1 smooth out error, and how you're telling us we're going to
2 have to take derivatives, which augment error --
3 DR. FLOCKHART: Well, it's taking a smaller
4 part of the data.
5 DR. SHEINER: We may never be able in sort of
6 a naturalistic setting to estimate a derivative with any
7 kind of accuracy.
8 It would work at the level of a preparation.
9 If you had a preparation that was rapidly absorbed and one
10 that wasn't, and that was pretty consistent, then you'd
11 know that you'd have more danger from one than another in
12 that sort of circumstance. But that we'll ever discover
13 who are the people who absorb more rapidly by sort of
14 surveying the world and then trying to put it together
15 across several models -- and I'm the mad modeler.
16 DR. FLOCKHART: But, Dr. Sheiner, shouldn't
17 that come out of some models? In other words, you would
18 be able to see in a large population study whether people
19 who get fast absorption get a QT longer.
20 DR. SHEINER: Well, I don't know the rate of
21 rise because I don't know when their level was drawn. I
22 put it on the graph at a certain point because that's what 1 86 2 1 they told me, but I can have two levels that are 10
2 minutes apart and they're really 2 hours apart.
3 I guess what I'm trying to say is that -- and
4 I was just being facetious, but I think we do need to
5 temper the kinds of conclusions we hope people to draw
6 from sort of messy clinical data. I'm just mentioning
7 that derivatives are really hard.
8 DR. VENITZ: But that's the empiricist
9 talking. As a pharmacologist, I say maybe I can understand
10 something about what's responsible for orthostatic
11 tachycardia and it may well be that my rate of change in
12 concentration or my Cmax is much more physiologically
13 important, and I know that without having to do an
14 empirical study.
15 DR. SHEINER: Right. What I'm saying is the
16 implications of what I'm saying, to be serious rather than
17 just making trouble, would be if that's the kind of thing
18 you want to know, if you're not sure about it, then you
19 need to do a very well-controlled experiment. You're not
20 going to be able to learn that from the same kind of data
21 you might be able to learn that area under the curve was
22 the determinant. 1 87 2 1 DR. VENITZ: Either that or you have some
2 mechanistic understanding how the drug concentration leads
3 to a response. That's the point that I'm making.
4 Hartmut.
5 DR. DERENDORF: Well, I think this discussion
6 shows that it's really impossible to even try to have a
7 standardized approach in terms of a parameter like a
8 bioequivalence approach, that you have a single criterion
9 that would summarize it all up. I think each drug, each
10 class of drug is different, and each situation is
11 different. I'm not sure if we can find a standardized
12 approach as we're asked to.
13 DR. LEE: I guess by standardized approach, we
14 mean a standardized conceptual approach, which means we
15 always like to calculate the probability of an adverse
16 event, rather than saying that we're going to standardized
17 an Emax model as the method to be used or the magical AUC
18 or Cmax. So, again, we're trying to standardize the
19 conceptual approach.
20 DR. SHEINER: I think it's really worthwhile
21 focusing on the positive here, which is this is a
22 difficult problem and the very fact that people involved 1 88 2 1 in regulation are acknowledging that it's worthwhile to
2 try to be quantitative about things that are extremely
3 uncertain and to try to come up with a better way to be
4 more quantitative about a problem where there will never
5 cease to be disagreement about any particular case because
6 you'll never nail anything down close enough -- you'll be
7 saying this is what we think we ought to do in terms of
8 dosage recommendations. And it will be based upon an
9 information base which would allow a rational person to
10 say, no, you don't need to do that. That's where we're
11 going to wind up, and to wind up anywhere else would be so
12 prohibitively expensive that it would not justify the
13 effort.
14 So I'm extremely encouraged. It's not as it
15 has been in the past, hands being thrown up and we can't
16 do this well enough, so we won't do it at all. That's not
17 the right attitude. And that you're seeking advice on how
18 to do this difficult thing I think is a very good thing.
19 But I do think that some kind of
20 standardization, for example, about turning things into
21 probabilities and utilities in an honest way so that
22 everybody can be on the same page -- they can all 1 89 2 1 understand what you know and what values you're applying.
2 DR. LESKO: A lot of the context for the
3 discussion in the case studies so far have been what we've
4 seen, what has come in in an NDA, but is there a way to
5 translate the methodology we're talking about, let's say,
6 to a drug development program in order to get studies
7 designed that would provide for information that would
8 reduce some of the uncertainty that we work with in the
9 absence of some formal recommendation to do studies a
10 certain way?
11 In other words, let's say a standardized
12 method evolves over time, and let's say that that could
13 perhaps evolve into a guidance on study design that would
14 provide for information that would be better apt to
15 provide the information that we're asking here in terms of
16 dose adjustment and quantitation of risk. Is that a
17 logical follow-through on the path we're on in the minds
18 of people?
19 DR. VENITZ: But is the uncertainty and the
20 consequence of the uncertainty that we currently have so
21 large that we really need to do a whole lot more
22 experimental work, short of what you're doing right now, 1 90 2 1 which is on a case-by-case basis, evaluate whether the
2 information is sufficient for you to assess the risk-
3 benefit, and then as in Jenny's case, recommend to the
4 sponsor that they would have to do a larger study to look
5 at high exposures?
6 DR. LESKO: I guess I was sort of asking the
7 question -- in Jenny's case, for example, QTc was obtained
8 for the first 4 hours. Would a study design that looked
9 at a longer period of time -- wait a minute. Was that
10 your case? Well, it was one case. Sorry, Nhi. I should
11 know these data.
12 There was one case where the QTc was obtained
13 for only 4 hours and blood levels were obtained for a
14 longer period of time. And would a different study design
15 have provided a better basis to make the recommendations
16 that people were trying to make with the analysis of the
17 data? That's sort of where I'm heading with that.
18 DR. FLOCKHART: I think the answer to that is
19 obviously it depends. 4 hours might have been long enough
20 for that drug, but there are other drugs -- haloperidol
21 comes to mind -- where that would not have been enough.
22 To go back to my point, I think actually it 1 91 2 1 is, Dr. Lee, very generalizable. I think there is a
2 generalizable conceptual approach. My point about
3 bringing up just sticking to the AUC was just to be
4 educated about that. I suspect that the AUC would very
5 often be a valuable parameter, but you have to be open to
6 using other things when that's biologically and
7 pharmacologically appropriate.
8 DR. VENITZ: The only thing I would add is, as
9 you've heard the committee talk about last time, as well
10 as this time, I think there's a lot of favorable
11 sentiment. As Dr. Sheiner likes to point out, it's better
12 than what we currently have. It beats the competition.
13 The one thing that I would reinforce, though,
14 is that it's very important to communicate it
15 appropriately, and that has to do with all the assumptions
16 that are being made. Are they verifiable to some extent
17 or not? Do you want to err on the conservative side or on
18 the more liberal side? So that the people that deal with
19 the clinical pharmacology reviews interact with the
20 medical reviewers. They may not understand the technical
21 side of it, but they're the domain experts and they can
22 follow those kind of thoughts. So it's really a matter of 1 92 2 1 risk communication in my mind more than it is the actual
2 process.
3 DR. KEARNS: And to pick up on a point that
4 Dr. Flockhart mentioned too, it has to be driven by
5 biological or pharmacological plausibility. To use an
6 approach, a guidance across the board can create
7 information that is not factual.
8 For example, I had the occasion to look at a
9 new molecule just last week with a sponsor to talk about a
10 study design. Of course, they had received some input
11 about that study design, which included multiple ECGs over
12 time that was coincident with the sampling time for the
13 pharmacokinetics. When I inquired as to the preclinical
14 data about the ability of the molecule to prolong QT,
15 about the only way that I could be convinced it could
16 happen is if the structure could be inserted in the chart
17 of the alphabet and somehow got between the letters Q and
18 T.
19 (Laughter.)
20 DR. KEARNS: So what will we see when we do
21 the trial? We do multiple ECGs, in this case, on
22 children. What happens if we see a relationship come out 1 93 2 1 of that that can be described by a host of models with all
2 the appropriate variability nested in? Will we have
3 proven something that wasn't shown by prudent preclinical
4 testing, or will we be finding ourselves in the midst of
5 yet another epi phenomenon that has implications about how
6 the drug might be used?
7 So I think one has to use caution in making
8 sure that when we do these things, we have good reason to
9 do it based upon what we know. I'm not saying that we
10 will always know everything up front. We clearly, clearly
11 don't. It's an imperfect science in an imperfect world.
12 But to just cast it out there as indiscriminate use of an
13 approach carries with it some liability that might not
14 serve the public at the end of the day.
15 DR. DERENDORF: Just a follow-up comment to
16 what Dr. Sheiner said. I fully agree that it is
17 worthwhile doing it and it is a good thing to do it. But
18 the standardization -- really my point was that it stops
19 at the point where we say each drug is different and the
20 more you know about the exposure-response relationship for
21 that particular drug, the more we can use it to make
22 predictions and the better they will be. That's a trivial 1 94 2 1 conclusion, but I think that's where the standardization
2 ends. Then from there on, really each case is different
3 and needs to be dealt with individually.
4 DR. VENITZ: Would it be helpful, as far as
5 the internal workings are concerned, to come up with a
6 list of questions that you typically consider when you go
7 through this process and for the committee to have a look
8 at them? I'm not sure whether the approach is something
9 that can be unified, but maybe the kind of questions that
10 should be asked every time you do this can be found
11 consensus on. Does that make sense to the committee? So
12 maybe at a future meeting, the questions that you would
13 ask, what surrogate markers do you have, what
14 relationships do you have, do you use areas or Cmax, those
15 kinds of things that you go through every time that you
16 have to review an NDA based on your experience.
17 DR. SHEINER: I think you can go a little
18 further. I think there are sort of best practices. Maybe
19 that's the way to think about it in doing this kind of
20 thing. Since I don't know anything about anything in
21 particular, I've been dwelling on the generals of showing
22 clearly what you know and what you don't know and somehow 1 95 2 1 checking your models against your current data and so on.
2 So I think there are best practices in this and I think
3 there are some things you can say in general, although I
4 agree with you, when you get down to putting the labels on
5 the x axis and the y axis, then suddenly you're in the
6 domain area and you've got to talk to the right folks.
7 DR. VENITZ: Any further comments by the
8 committee or any further questions from the FDA staff?
9 Larry.
10 DR. LESKO: Yes. Maybe this is a deeper
11 question and there isn't time to discuss it, but it does
12 lead us down the path if we develop a standardized
13 approach, the question that I have, in terms of labeling,
14 comes into my mind. Right now we put in the label
15 descriptive information, for example, in the clinical
16 pharmacology section about a change in an area that
17 describes the, let's say, drug interaction or a special
18 population change, and then if it warrants, a change in
19 the dosage and administration section as to what to do
20 about it. But if you have more data in hand, i.e., the
21 likelihood of a risk or the probability of a risk or the
22 probability of an increase in risk or other things that 1 96 2 1 might come out of a standardized approach, the question
2 would be to what extent would this information be helpful
3 to prescribers or would it be a distraction to the
4 prescribers and how can we enhance labels. Because we now
5 know there are certain pieces of information that go into
6 labels that at least the consumers, public and physicians
7 tell us are not helpful to them and they can't interpret,
8 and drug interaction seems to be one of those areas we
9 frequently hear about.
10 So we're thinking of ways of improving labels
11 in terms of consistent language, the scope of information
12 that goes into it, and with this kind of standardized
13 approach, it could lead to some interesting ways of
14 revising labels to convey different information to
15 prescribers and patients.
16 DR. VENITZ: Any comments?
17 (No response.)
18 DR. VENITZ: Okay. Then let's move to the
19 third example for today, which is going to be presented by
20 Dr. He Sun. He is a pharmacometrics reviewer in the
21 Division of Pharmaceutical Evaluation II.
22 DR. SUN: Good morning. I will try to discuss 1 97 2 1 some general questions in my discussion, and we may switch
2 the specific detail in numbers to general issues.
3 These will be the questions I'm going to ask
4 after the presentation, but I just put it up front to get
5 some initial feelings.
6 The first question is, if we get adverse
7 reaction data from clinical studies, these data can be
8 treated as either a continuous variable or a categorical
9 variable. Then, what should we do? Do you have any
10 preference, and why? I will show some examples to
11 illustrate it further for this.
12 The second question is, in phase III clinical
13 trials, lots of subjects, but we may not have observations
14 in special subjects. Therefore, the population PK
15 approach may give us the opportunity to either simulate or
16 predict the exposure parameter for the population who
17 don't have exposure observation but do have response
18 observation. So what's the limitation and utility of this
19 approach?
20 Now, if we have a PK model based on the above
21 approach, we get some kind of conclusion on side effect
22 versus drug concentration relationship. How do we make a 1 98 2 1 dose adjustment recommendation for subpopulations?
2 There's some limited information here to
3 present what data distribution pretty much looks like. On
4 the slide here, in this corner it shows what the data
5 distribution may look like. It can be dense data from
6 phase II trials or it can be sparse data from phase III
7 clinical trials, or some kind of a combination with an
8 unbalanced situation.
9 Now, the safety information can be either a
10 single critical key adverse reaction parameter which is a
11 continuous variable, like QTc variable or high blood
12 pressure and so on. But it can also be a categorical
13 variable like pain or "yes or no" for liver toxicity and
14 so on.
15 But these two actually are switchable. Let's
16 say blood pressure. You can set up a cutoff marker that
17 says if above such and such, it is abnormal, below such
18 and such, it is normal. So the continuous variable
19 actually can be changed to a categorical variable.
20 Now, a categorical variable, although it can
21 be a "yes or no" situation, but for the group with "yes,"
22 you can also give a score of 1, 2, 3, 4, 5 or from 0 to 1 99 2 1 10. So it becomes some kind of a continuous variable.
2 So these two actually have no clear cut.
3 That's why I ask this question in this presentation. If
4 you have this situation, which one do you prefer? Of
5 course, this will also change your data analysis process.
6 So you can also have a combination of both
7 with multiple ADR observations. But for phase III trials,
8 pretty much we have this kind of situation: mixed types,
9 multiple ADRs, and unbalanced. Then that is why
10 population approach can play here.
11 Let me first show some data sets. Then we go
12 back to see what analysis process we can apply. This data
13 set is used just to illustrate the question or the process
14 I mentioned before. Forget the exact numbers and the true
15 terms. Sometimes I have to modify this.
16 Let's say we have two clinical trials, very
17 big size, 1,500 evaluable treatment patients. And we also
18 have multiple dose levels from X to 4 times higher. And
19 the patient plasma drug concentration was measured,
20 although for some are dense and for some are sparse.
21 Therefore, the total data set is kind of unbalanced.
22 Then we have endpoints for safety measurement. 1 100 2 1 This can be some blood chemistry variables which usually
2 is a continuous variable at the beginning, but a clinician
3 can define some value as a cutoff point shows this
4 variable as normal/abnormal to claim at such situation
5 there's no ADR and others will be ADR.
6 The ADR can be also present as a "yes or no"
7 situation for some, like headache, liver toxicity,
8 phototoxicity values. And this variable again can be
9 changed to different scores for the extent of headache,
10 like mild, moderate, or severe.
11 Now, PK results. Let's say drug-drug
12 interaction causes AUC to increase by almost 300 percent.
13 The Cmax changes by 150 percent. AUC and Cmax may also be
14 changed by age, gender, or so on and so forth, even
15 between ethnic groups.
16 The safety results. We will not focus on
17 efficacy in the presentation. We will only focus on
18 safety parameters. Safety parameters usually are very,
19 very small in percentage and very sparse. So there are
20 some cases where you never have any so-called "maximal
21 effect" for side effect terms.
22 The efficacy results. Let's make this 1 101 2 1 discussion a little simple. We see efficacy has no such
2 exposure-efficacy relationship detected although we see
3 there's a demonstration of clinical efficacy in total.
4 Now, with these data sets, let's come back and
5 see what process we usually can do. First of all, of
6 course, there are managing/editing data processes we can
7 use. For this part, we start to have a question: shall
8 we treat the data as a continuous variable or should we
9 treat the data as a categorical variable?
10 Then we can conduct a population PK analysis
11 based on exposure data such as building a base model, add
12 variability, add covariate, and so on and so forth. Now,
13 there's one problem: if we cannot find a significant
14 covariate in the PK model, then the next step for
15 predicting E for the new population or new individuals
16 will havew a little problem. But let's say we have the
17 model built and the model validated. Then we can go to
18 the next step and determine individual exposure or
19 subpopulation exposure parameters. There are two parts
20 here. We can do post hoc for the subjects who are already
21 included in the study, or we can do a simulation trial to
22 determine exposure parameter for the population who was 1 102 2 1 not really in the trial or the observation was not
2 available in the particular patient.
3 The next step will be to derive secondary
4 exposure parameters, such as AUC, Tmic, effect compartment
5 concentration and so on. And another important factor
6 here I want to emphasize is that the accumulative exposure
7 parameter can be estimated and determined, like say what
8 if after multiple dose or long exposure situation.
9 With these exposure parameters available
10 through the above processes, we can determine exposure-
11 response relationship for individuals or for special
12 populations. We have lots of methods here. We can use
13 classification method. I will give you some examples
14 later on. We can do classification or regression tree
15 analysis, logistic regression for binary data, and so on.
16 Then we consider the accumulative exposure
17 time, then perform statistical analysis on response data.
18 Now, this again correlates with what do we do with the
19 data? If our data is a continuous variable and we have
20 odds ratios with uncertainty built in, we can do
21 statistics. But sometimes if the variable purely is
22 categorical and is divided to either above the mean or 1 103 2 1 below the mean, the statistics will be hard.
2 Now, with all this situation, the next step,
3 finally we will make a dose adjustment. What are we going
4 to do especially if we have multiple variables? Let's say
5 the exposure-response depends on age, gender, body weight,
6 and blah, blah, so on. If we take all of this condition
7 together for making a dose adjustment, it may be too
8 complicated in drug labeling. Shall we only consider the
9 one which is critical, or shall we consider the one which
10 has most frequently occurred, or some other method? I
11 want to hear some discussion on this.
12 Let's see some results. If we're dealing with
13 this process, what result can we get?
14 Classification. The first part we can see is
15 to divide the whole population into some equal populated
16 segments. Let's say every 25 percent subjects from low
17 exposure to high exposure. Or we can divide this whole
18 population into equally distanced segments, the
19 percentile; that is, the first 25 percent in
20 concentration, the second 25 percent, and so on and so
21 forth. Then we count what's the frequency of ADR. For
22 example, the results can be presented as total ADR is 18 1 104 2 1 percent if AUC is greater than the mean and only 5 percent
2 if AUC is less or equal to the mean value. That makes the
3 whole discussion for this kind of classification results.
4 Of course, there are lots of pros and cons. I really want
5 to hear a discussion later on.
6 Now, the second one is that we can do a
7 classification analysis based on severity. Let's say we
8 reclassified R values, the response values, as different
9 class or different scores, as severe, moderate, or mild,
10 and so on. Then we see the example. Severe ADR occurs if
11 Cmax is greater than 10 but only mild ADR is apparent if
12 Cmax is less than 2, although the frequency probably
13 between these two has no significant difference. But in
14 this situation, we see it looks like 10 is some kind of
15 cutoff value to avoid severe ADRs.
16 Then we can throw this data into a computer to
17 search for the best maximum split, maximum split
18 distinctions between R values as by a classification tree
19 or regression tree. One result I present here, for
20 example, the first split on ADR frequency was at AUC
21 equals to 871. So when the split occurs on ADR, it was 23
22 percent if AUC is greater than 871 and would be 2.5 1 105 2 1 percent if AUC is less than 871.
2 Next, we will do modeling work. We can do
3 modeling work for the same data set for different ADR or
4 the same ADR parameters. When we do modeling work, there
5 are several ways. I do not want to discuss further on
6 this part. I only want to show the possible ways. We can
7 base on purely statistical models to do the modeling work,
8 or base on some kind of physiological-based, meaningful
9 models. There's a lot of discussion on the pros and cons
10 for each. But let's go to the next one. Our model can be
11 a linear model or a nonlinear model. Of course, there are
12 uncertainty parts when building nonlinear models, adding
13 fixed effects and random effects, again, with this model.
14 So two examples. We can do a simple
15 regression analysis based on so-called logistic
16 regressions. So, for example, we can get a result with
17 even a 95 percent confidence interval for the logistic
18 regression results for odds ratios. For example, we can
19 see the odds ratio for acute tissue rejection increases 23
20 percent if AUC 0-24 decreases by 10 percent. That's one
21 way to present this data.
22 Then we can treat all the data as a continuous 1 106 2 1 variable, do as the next few examples for modeling work.
2 We can get an equation that says the HDL drops below
3 normal on day 95 if the concentration, average
4 concentration, is greater than 10 micrograms per ml on day
5 95 for patients with high body weight and low initial HDL
6 at baseline. So there's a covariate effect built in and
7 it also has this kind of a drug concentration curve
8 profile.
9 The back pain we treat as 0 to 10 scores, some
10 kind of semi-categorical variable. It's nonlinearly
11 correlated with plasma drug daily AUC and the number of
12 treatment days and the dose regimen. So three factors.
13 The result here found was b.i.d. actually has less
14 incidence of back pain. T.i.d. will have more, although
15 the total daily doses are equivalent. And the number of
16 treatment days significantly correlate with or predict
17 back pain scores.
18 QTc prolongation. Now, we can find some
19 models. The relationship between QTc and the drug
20 concentration can be described by an Emax model. Then you
21 can find some E-R parameter like E0, Emax, ED50, and so
22 on. But these parameters can be, as we discussed before, 1 107 2 1 correlated with either gender or age or some other
2 subpopulation variables.
3 Phototoxicity. Now, we only have two
4 variables, yes or no. We get a "yes" value on day 10 if C
5 average on day 10 is greater than 8. This will occur only
6 in female subjects. So this can be due to either data
7 limitation or this really is a true result that female
8 subjects are more sensitive to the drug in terms of having
9 phototoxicity occurring.
10 Liver toxicity. We can get some kind of
11 frequency linearly correlated with C in the initial few
12 hours of time of exposure. It may not correlate exactly
13 with the Cmax, but it correlated with the initial range of
14 concentration average.
15 Blood chemistry. Now, blood chemistry
16 actually is a continuous variable, but we can treat the
17 blood chemistry variable as a categorical "yes or no,"
18 normal or abnormal, or give them a score from 0 to 10.
19 Now, if we have a score from 0 to 10, we can get some kind
20 of correlation with plasma concentration, either AUC or
21 number of treatment days.
22 So these are examples I want to show. So one 1 108 2 1 single variable can be treated by different ways, but in
2 one study we have lots of different variables, and
3 different variables may be treated by different ways, and
4 use different data sets as different base for information.
5 And there are others, probably we never find
6 any relationship, never can find a cutoff point.
7 Descriptive. Here I give some examples. Some we just can
8 have a definition of normal/abnormal values, but may not
9 see any difference between different subgroups like these
10 four examples I show here.
11 So now we come to the end. We have all the
12 information, by all different methods, with different
13 bases, from different data sets, so on. We definitely
14 already used population PK, did it two times. One is
15 predicting exposure parameter for subjects who do not have
16 observation in exposure but do have a response
17 measurement, and in the second part, use a population
18 approach, nonlinear mixed effect modeling, to show whether
19 the E-R relationship depends on some other co-variables.
20 So with all this population of subjects and
21 information, now we will make a dose adjustment. See, for
22 example, we can do this: the average upper therapeutic 1 109 2 1 limit is probably around 10 for Cmax and 871 for AUC.
2 Remember, these two variables are gathered from previous
3 toxicity analysis. And the female subjects seemed the
4 most sensitive to phototoxicity, and the concentration
5 average should be less than 8. Then we see a b.i.d. dose
6 regimen is preferred because it reduced one of the
7 particular toxicity results.
8 Now let's go back to my questions with all the
9 data we have seen in the examples. First is what are the
10 utility and general limitations of linking PK obtained
11 from population analysis to response endpoints? And what
12 are the general considerations in E-R based dose
13 adjustment for special populations? And should we treat
14 this data as a continuous or categorical variable? What's
15 the preference?
16 Again, I really want to hear a discussion more
17 focused on the general concept and ideas based on the
18 experience you have and let us know the pros and cons for
19 each situation rather than focusing on the numbers because
20 I have modified the values somewhat to make the
21 presentation smooth. Thanks.
22 DR. VENITZ: Thank you, He. 1 110 2 1 Any questions about Dr. Sun's presentation
2 before we delve into his proposed questions?
3 On one of your slides, you mentioned Cint.
4 Can you tell me what that meant?
5 DR. SUN: This is the initial concentration
6 exposure.
7 DR. VENITZ: Oh, initial concentration. Okay.
8 Do you want to pose the questions and then let
9 the committee bat it around?
10 DR. SUN: Okay.
11 DR. VENITZ: The first question regarding the
12 limitations of linking PK from a population analysis to
13 response endpoints. Do you want to elaborate on that
14 question what specifically you had in mind?
15 DR. SUN: Okay. This question was the utility
16 and general limitations of using population PK for
17 population for this analysis. As I mentioned, we have two
18 places we can use nonlinear mixed effect modeling work for
19 doing data analysis for this data. The first part is in
20 clinical phase III trials, we may not have a concentration
21 exposure measure for every subject, but we do have a
22 response measure for every subject. In this situation, if 1 111 2 1 we have sufficient information, use the population PK, get
2 the model, then we can predict or estimate concentration
3 or other exposure parameter for the population we see in
4 the phase III trial. Or in some situation, patients only
5 have one or two trough measures and we want to determine
6 the total exposure and the time of exposure. So this is
7 one place population PK can be used.
8 The second part is after we have E data and R
9 data, either categorical or continuous variable, now we
10 can use mixed effect modeling to see whether these two
11 variables are related to each other based on some
12 covariate factors.
13 So that's my question. What are the utility
14 and general limitations on this if we do it this way?
15 DR. SHEINER: I can't decide if what you're
16 asking is, is there a manual for how you treat any given
17 set of data to come up with the conclusions that you're
18 going to find most believable. We can't address that. So
19 when I look at that first question, your description of
20 what you might do sounded sort of like something I might
21 do.
22 The only thing I can say, the only serious 1 112 2 1 general limitation about which even good data analysis
2 cannot help you is the problem of confounding. Both the
3 PK and the responses are endpoints, are outcomes, and
4 whenever you try to relate outcomes to outcomes, you have
5 the problem that you can't tell which way causality runs.
6 And you base that conclusion, if you do, on external
7 information in the way of science or other things. You
8 can't tell it from your data. So that's the limitation.
9 Now, that doesn't mean we don't proceed every
10 day, based on observational data, to make the most
11 important decisions in our lives. We do. But we have to
12 understand that in a regulatory context, there are other
13 forces operating. You want to be cautious in certain
14 ways. And that's the serious problem.
15 Most of the other stuff, it seems to me, that
16 you brought up were technical issues. And I'm not sure
17 that we really want to spend -- even though I'm a real
18 techno-wonk when it comes to model-based analyses, I don't
19 think you want to hear me dilate on that.
20 I think the basic thing there is if you get
21 different conclusions when you treat your data as
22 categorical versus continuous or when you use one kind of 1 113 2 1 a model or another, then there's something wrong. So you
2 ought to get all the same conclusions. What happens as
3 you turn data from continuous, if it's got a lot of
4 information, to categorical, you're losing information.
5 So some things will drop out. Some things will appear no
6 longer to have relationships that did appear before to
7 have relationships. That's got to happen as you limit the
8 information in your data.
9 But other than that, I think you want to
10 basically use the data representation that's most relevant
11 to the people who are going to use it and that keeps the
12 information, et cetera, all the good rules of modeling.
13 But I don't think we can get into too many details.
14 So maybe if you have particular instances
15 where you think, looking at data in different ways, the
16 same data in different ways led you to very different
17 conclusions, I think that's something that I might be
18 interested in hearing about. Otherwise, I don't know what
19 we can say in general.
20 DR. LEE: Can I rephrase the question a little
21 bit? The reason we're making this presentation is because
22 phase III studies actually present a unique opportunity 1 114 2 1 for us to look at exposure and response, especially the
2 safety endpoint that we don't frequently observe in a
3 phase II study which is too small to capture a rare
4 adverse event. That's why we wanted to ask the committee
5 whether a population approach would be a good approach to
6 look at exposure response in the phase III studies.
7 However, there may be some limitation in terms
8 of study design. For example, we may not have enough
9 plasma samples or maybe the sampling time between the PK
10 and the pharmacodynamic endpoints is different.
11 So this is the type of question we're trying
12 to ask the committee, whether internal study design,
13 whether there's any limitation to conduct such type of a
14 population PK analysis. And if there are certain
15 limitations, can we recommend to the sponsor to design the
16 study differently in the future so that we can get a
17 better quality PK/PD relationship out of the phase III
18 studies?
19 DR. SHEINER: Let me just say one more thing
20 about that. So you're talking about wanting to use this
21 confirmatory study for learning purposes. And there are
22 certain kinds of learning data elements that don't 1 115 2 1 interfere with your design that would make your life a lot
2 easier. Whether they're worth it or not, you can only say
3 afterwards. But measuring things serially, whether it be
4 toxicity or efficacy or both, rather than just the 6-month
5 and 1-year endpoint, or whatever it is that was the
6 primary thing; measuring compliance, that is to say, what
7 drug did they actually take; measuring PK.
8 I think, by the way, my guess is that
9 adherence is more important an influence on outcome than
10 PK is for most cases. But when they say measuring PK, you
11 want to measure that in the case where adherence is
12 assured so that you have those as two separate variables.
13 And so on.
14 Basically the idea is measuring biomarkers,
15 whether they be adherence or chemicals, you know, along
16 the causal path from the prescription to the effect, and
17 measuring them serially over time. That's the best you
18 can hope for without changing the design radically. And
19 if you want to be able to do these kinds of analyses,
20 that's the kind of data that you need.
21 But techniques for dealing with missing data,
22 techniques for dealing with other problems that arise, 1 116 2 1 with mixed kinds of data, both continuous and categorical,
2 and so on, those I don't think are essential. They exist.
3 They make your life a little tougher or a more
4 interesting, depending on where you come from. But
5 they're there and you should be able to get the
6 information out of the data.
7 DR. SUN: He was asking whether I have an
8 example regarding how the data can be treated either
9 continuous and categorical.
10 DR. SHEINER: And reach different conclusions.
11 DR. SUN: Yes, reach a different conclusion.
12 Let's say this situation. If you treat data
13 as a "yes or no" situation and you divide the
14 concentration distribution to be above the mean and below
15 the mean, do you will find the only conclusion you can get
16 is the frequency of "yes" when concentration or AUC above
17 the mean is 23 percent. If lower than the mean, it will
18 be 5 percent. That's all you can get.
19 If you treat it as a continuous variable, you
20 can get some kind of sigmoid models, correlation between
21 scores of adverse reaction versus the concentration. Then
22 from the curve, you can pick up -- say you want to limit 1 117 2 1 less than 10 percent of subjects has a score less than 2 --
2 a concentration. So this becomes a different decision.
3 And the curve becomes a smooth curve. You pick up a point
4 at which you limit two factors. Percent of subjects reach
5 a score of XYZ. Compared with the first one, you only can
6 get a result if above the mean will be such and such, if
7 below the mean will be such and such.
8 Then in labeling, it will be different. In
9 labeling, when make a dose adjustment, say due to drug-
10 drug interaction, if the population Cmax change, still
11 somewhat below the mean values, for the overall
12 population, or you can get a feeling what the frequency of
13 side effects due to drug-drug interaction will be. But
14 in the second situation, if you have a continuous curve,
15 you can estimate when concentrations switch a kind of 10
16 percent, what's the percentage of patients will have a
17 score of 2 or 3 increase by such and such.
18 So when we recommended these two suggestions
19 to clinical or to the labeling committee for NDA review,
20 these two really makes different. That's why the question
21 comes. Any parameter really we can treat as one of, or we
22 can switch between the two. And what really we want to do? 1 118 2 1 DR. DERENDORF: I think as was said earlier,
2 whenever you move from continuous to categorical values,
3 you throw away information. I think it comes down to a
4 compromise that you have to make with the information that
5 you have and how you want to communicate it. If you make
6 it too complicated and include everything you know, nobody
7 is going to use it. So you have to find a way to focus on
8 the important things, but still come up with an accurate
9 conclusion.
10 You have a great example that you can overdo
11 it and make it too simple. One of the conclusions you
12 have here, the total ADR is 18 percent when the AUC is
13 above 1,200 and 5 percent when it's below. That may be
14 true for the data set that you have, but it's totally
15 useless for someone who wants to extrapolate it for a
16 certain situation because obviously you can have very,
17 very low AUCs that wouldn't have any ADRs or you could
18 have very, very high where the probability would be much
19 higher. So that would be a misrepresentation of the
20 actual information that you have by making it too simple.
21 So I think there's the trick that you have to find the
22 right balance. 1 119 2 1 DR. LESKO: We're in the world of safety here,
2 and it would seem like in the clinical trial design,
3 there's going to be prespecified endpoints of safety that
4 the sponsor provides. It also seems to me that there's
5 going to be a convention for a safety biomarker, let's
6 call it, that will be continuous or categorical.
7 For example, if my concern with the drug is
8 heart rate, I'm going to look at that maybe as a
9 continuous variable because that might be the way it's
10 measured and the way it's analyzed. On the other hand, if
11 I was looking at a hematological toxicity like
12 neutropenia, I might be concerned about grading the
13 severity of that.
14 So I guess the question becomes, given there
15 are certain prespecified endpoints in terms of severity
16 and frequency, and given that there's a conventional way
17 of presenting data as continuous or categorical, what
18 would be the motivating factor to change continuous to
19 categorical? What would the benefit of that be in terms of
20 the pragmatic aspect of it, in terms of dose adjustment?
21 Because you think about a drug that is going to be used
22 therapeutically and being monitored by these same 1 120 2 1 endpoints and doses being adjusted by those same
2 endpoints. So it's sort changing something from the usual
3 to the unusual, and you'd have to say there's a reason to
4 do that. And I guess it wasn't clear to me what the
5 reasons for me to do that would be.
6 DR. SUN: I will give one example. Liver
7 toxicity, one situation we saw in one NDA. If you see the
8 frequency of liver toxicity based on concentration above a
9 cutoff point, below a cutoff point, you don't see a
10 difference. So it looks like all concentrations show --
11 if you only define liver toxicity as a yes or no
12 parameter, you don't see a correlation between these two.
13 But if you use actually the measurement of blood chemistry
14 values, which is used as an indicator for liver toxicity,
15 you're able to correlate concentration with the blood
16 chemistry variable. So clinically it may be easy to see,
17 well, 9 subjects or 10 subjects have liver toxicity, but
18 we don't know correlation with concentration.
19 But on the other side, you can see actually
20 when concentration or AUC increases, the chemistry value
21 have some trend increase. Then you define a cutoff at
22 that point that says on the curve where is the cutoff, 1 121 2 1 then where is the concentration's cutoff? So there are
2 two ways to present this data. The same data set we can
3 do two ways.
4 My own experience in the drug labeling is that
5 it looks like the first way is easier, the second way
6 somehow, like you mentioned, has to be communicated to
7 others in different professions as well.
8 DR. FLOCKHART: I would just emphasize that
9 point. I think obviously you lose power with a
10 categorical variable, but if you have a change that is
11 picked up by the categorical variable, as well as by the
12 continuous one, in general it communicates better. And we
13 have a humongous problem with communication. That's an
14 understatement. So I think when it's possible to state
15 something in stark, clear yes or no terms to somebody
16 who's practicing medicine in the area of drug interactions
17 or recommendations within a label -- we're all aware of
18 the vast majority of the label is just dust anyway. So
19 when you can simplify it, that helps.
20 DR. VENITZ: Can I just make a general
21 comment? Looking at some of the, I guess, labeling
22 language or some of the statements that you reviewed with 1 122 2 1 us, He, the only time they're going to be useful for a
2 practitioner is if they actually draw blood levels because
3 otherwise the fact if I'm above some certain level or some
4 certain area, something happens or doesn't happen, it
5 won't help me as a practitioner. The only thing that I
6 want to know is can I change the dose and how do I change
7 the dose.
8 So when you translate the information that you
9 get from what you just reviewed for us and translate that
10 into labeling language, you really have to target doses
11 not concentrations, unless part of the therapeutic
12 management with this particular agent requires dose
13 titration based on plasma levels. Otherwise, I don't see
14 how that is useful information for the practitioner to
15 translate into practice.
16 DR. SUN: You prefer dose-response
17 relationship rather than concentration-response.
18 DR. VENITZ: The only thing that the clinician
19 can change is the dose unless part of the management
20 requires taking blood levels, and then depending on the
21 blood level, I can make certain adjustments.
22 DR. LEE: Can I say something here? I think 1 123 2 1 what we propose here is we're not going to use a different
2 approach either as population analysis or regular PK/PD
3 analysis. So what we're presenting here is can we use
4 population analysis to get a PK/PD relationship. But once
5 we get a PK/PD relationship, we're going to follow the
6 standardized approach to estimate the probability of an
7 adverse event for special populations. So that's what
8 we're trying to propose here.
9 DR. VENITZ: The statements that he reviewed
10 for us, at least most of them, have some statement about
11 levels or areas that are too high, too low. And I'm
12 saying from a practitioner's point of view, unless part of
13 the management with this particular agent requires drawing
14 blood levels so I can actually measure levels, it's
15 useless. I need to know what to do with my dose because
16 that's the only thing I can change short of changing the
17 drug itself.
18 DR. SUN: But on the other side, let's say,
19 same dose level due to drug-drug interaction change the
20 concentration. This information will give us an idea if
21 the concentration changed by such a degree, what's the
22 probability in the ADR will be. So in terms of decision 1 124 2 1 on this side, we still have to rely on this rather than
2 only rely on dose. Although at the end, we can see -- the
3 end and the level and we see in terms of drug-drug
4 interaction, you have to deduct the dose by 50 percent,
5 but a 50 percent deduction was got from concentration
6 dose-response relationship.
7 DR. VENITZ: I don't have a problem with your
8 conclusions. I'm questioning the usefulness of
9 incorporating the conclusions as they are in a label to
10 convince a practitioner to change a dose. That's all.
11 DR. SHEINER: Getting back to this issue of
12 dichotomous versus continuous. There's a big difference
13 as to whether you translate -- and I think this is what
14 David was getting at -- from continuous to dichotomous,
15 let's say, on the way in -- that is to say, you change the
16 data and then analyze the data that are now dichotomous --
17 versus on the way out where you take this complex model
18 that deals with all the variables in their full complexity
19 and then draw a very simple conclusion based upon
20 simulations of that that says half the time it's going to
21 be bigger than this if you do that. There's no problem
22 with the latter, and that's very important for 1 125 2 1 communication, to make things simple, to talk about the
2 key issues.
3 The problem with the former is that you're
4 losing information. The places you see it worst are in
5 these clinical scores where you have 15 or 20 questions
6 that you ask and then you add up the number of yeses and
7 that's a number. They're clearly not combined optimally.
8 You don't know whether one or another question would have
9 more information than other things.
10 What they do is they allow you to do your
11 analysis simpler. The only time that in my mind it's
12 justified to simplify the data before you analyze them is
13 when the loss of noise exceeds the loss of signal and you
14 can afford some loss of signal and you're losing more
15 noise by dichotomizing, let's say.
16 And I think most of us actually are probably --
17 I don't know. It would be interesting to take a survey of
18 how many bits of biological information people believe
19 there is in something, let's say, that's three significant
20 figures like a serum sodium. Is there really the 12 or so
21 bits that it takes to represent a three significant digit
22 number or is it really only three or four bits? The 1 126 2 1 sodium really 140 to 143 is all the same, et cetera. I
2 think there's probably a lot less information in these
3 continuous variables than we think there is because of the
4 high feedback systems that we're dealing with.
5 But the main point, it seems to me, is the
6 three questions that I always say you've got to ask before
7 you do anything. So this one is, what's the question? If
8 you want to write a label that says that as you increase
9 the dose by these increments, the probability of this
10 toxicity will go up by those increments, then you have to
11 have a model that somehow represents continuous
12 probability versus continuous dose. If you don't want to
13 write a label that says that but only says don't give the
14 drug to people who have values beyond these, then you
15 don't need that. So first figure out the question.
16 There I think maybe is where one could
17 sometimes schedule a meeting to talk about that. What
18 should recommendations for dosage changes in special
19 populations look like? What kind of statements ought we
20 be trying to make? And that will determine the data we
21 want to gather and how we want to analyze it. I'm not
22 sure that's all settled, but I'd like to hear at some 1 127 2 1 point what the agency does think about what constitutes a
2 complete set of statements and what degree of precision
3 and what kinds of words you use and what kinds of things
4 you want to be able to say.
5 DR. VENITZ: Any comments, Larry or Peter?
6 DR. LESKO: No. We've sort of thought about
7 that last statement that Dr. Sheiner brought up. I think
8 it's a very valuable statement, but I think we need to
9 think about it and put the story together and bring it
10 before the committee. But I think it would be a very
11 interesting discussion to have, what elements of a good
12 label should there be in terms of a probability of risk
13 and an intervention of some sort and how do you present
14 that in a consistent way across special populations or
15 something like that.
16 DR. VENITZ: A couple of more comments, I
17 guess more general comments, not specific to your
18 question. But how to translate information that you would
19 gather from this kind of analysis and translate them into
20 labeling language. How is the drug being administered?
21 Is it titrated on some kind of effect? That would make a
22 big difference in terms of how some of those things would 1 128 2 1 translate into a dosing recommendation because you may not
2 have to make a dosing recommendation because you're going
3 to pick it up as part of the normal way of therapeutic
4 drug monitoring. And it might not be a drug level. It
5 might be some surrogate marker, some biomarker.
6 Obviously, what are the available ways of adjusting the
7 dose? Do they have the dosage forms to accommodate that
8 or can you not?
9 And then along the titration route, do we know
10 anything about intra-individual variability that would
11 allow us to assess for a given patient how likely they're
12 going up and down, and is our dosing algorithm going to
13 pick that up?
14 Those are questions that aren't really
15 addressed in what you're talking about, but I think
16 they're very relevant for translating this information
17 into recommendations in the labeling.
18 DR. LESKO: Just to add to that in thinking
19 about maybe a future discussion with the proposed labeling
20 rule that the agency has out, there's going to be some
21 revamping of the label such that certain information gets
22 to the top of the label out of the individual sections. 1 129 2 1 And a question could be raised as to what criteria in the
2 context of what we're talking about would warrant raising
3 information to a more prominent part in the label. If
4 that is done for catching the attention of a prescriber,
5 it gets to that language and how you present the
6 information in translating it into therapeutics. So I
7 think the whole thing sort of flows as a future issue.
8 DR. SUN: Yes. We do see situations where a
9 label can be -- let's say it's an IV formulation or all
10 doses are available. You can put a table. It shows about
11 every year with a different dose. But when the
12 formulations only have 15 and 30 milligrams, you only can
13 do is categorize them to two classes. That is really true.
14 Let me summarize. As will be -- if we
15 translate from continuous to categorical, we lose
16 information. But on the other hand, maybe categorical is
17 easier to communicate. And it also depends on the
18 outcome, what are you going to do before you handle the
19 data. Before you even start an analysis, how do you use
20 this information.
21 Thank you very much.
22 DR. VENITZ: Any other comments? 1 130 2 1 DR. DERENDORF: Maybe it's way out there, but
2 it seems that one problem is not so much the information
3 and the data analysis, but it is really the ability of the
4 user to do something with it. I think maybe one of the
5 reasons is that we're limited in the label to just a
6 written document, and with modern technology, a lot of
7 that information can be presented to someone in a
8 palatable way like in a little computer simulation where
9 all the information is entered, and then there's a
10 recommendation or an assessment that comes back. That may
11 be something to consider in the future.
12 DR. RELLING: I'm new to this, but you talk
13 about special populations and you've presented different
14 examples of covariates. is there some sort of list that
15 you have of what you consider to define special
16 populations and what you consider covariates that should
17 always at least be asked about? We heard about
18 ketoconazole and something else, age or renal dysfunction.
19 Are you letting individual studies drive these things?
20 Are you going through some algorithms to define what you
21 should look for a priori? How are you deciding what
22 you'll even include or think about looking at? Because 1 131 2 1 there are all kinds of examples where we're making dumb
2 mistakes like looking at ketoconazole but not
3 itraconazole. How are you going through this stuff?
4 DR. SUN: The first part you're asking how to
5 define the special populations. In the legal terms, they
6 define special population as disabled patient/subject,
7 blind subject or some others. In our clinical
8 pharmacology term here, we refer to subpopulation like a
9 pregnant patients, pediatric subjects, or other ethnic
10 groups rather than the legal term defined those special
11 populations. That's the first part.
12 The second part. When we do clinical trials
13 in phase III trials, we do have some inclusion/exclusion
14 criteria. A majority of the time it's how the drug in the
15 future will be given to patients in the clinical setting.
16 So most likely we will include other subjects who
17 potentially will take this drug.
18 Did I address your question?
19 DR. RELLING: Not really.
20 DR. LESKO: Let me add to it. First of all,
21 the answer to the question is somewhat drug dependent, but
22 there is a standard range of assessments that are expected 1 132 2 1 within the drug development program. And most of these
2 are revolving around the changes in pharmacokinetics. So,
3 for example, there are demographic factors. We would want
4 to know age, gender, and race and the effects on
5 pharmacokinetics and whether that's pertinent to dosing
6 adjustments that might be necessary in those
7 subpopulations or special populations.
8 We then move next to intrinsic factors, as we
9 call them, and they're predominantly disease states that
10 handle drug disposition. So renal disease and hepatic
11 disease is a standard study that's in most NDAs.
12 And then finally the issue of extrinsic
13 factors and predominant amongst those are co-administered
14 drugs. So there's a heavy emphasis, ideally
15 mechanistically driven, to look at clinical drug-drug
16 interactions that are likely to be important in the
17 clinical setting.
18 So the categories of special populations are
19 defined by these demographics, the intrinsic, and the
20 extrinsic factors.
21 Now, depending on the drug used, there may be
22 additional special populations that would be looked at, 1 133 2 1 but I would say the ones I just mentioned are the standard
2 covariates that you would be interested in.
3 DR. RELLING: Do you require those to be
4 addressed a priori in the trial? And what makes the
5 difference whether you decide -- when are you going to
6 start looking at genetics? Where do you start drawing the
7 line of saying you've got to look at something besides a
8 creatinine and a bilirubin or whatever it is you are
9 asking for?
10 DR. LESKO: "Require" is kind of a harsh word.
11 I'd like to think of it as recommendations. By and large,
12 everything I just said is contained in guidances to the
13 industry that say what the expectations of the agency are,
14 and if those types of analyses aren't available, the
15 burden of proof then is on the company to say why it's not
16 important in the case of that particular drug.
17 When you get into what I would call evolving
18 areas or evolving covariates -- and you mentioned genetics
19 I think as one of them -- we have now appearing on sort of
20 the drug development scene the ability to look at changes
21 in DNA sequence that would describe what we typically have
22 called phenotypes, poor metabolizers and extensive 1 134 2 1 metabolizers. It's a logical extension, to me at least,
2 to begin to ask the question, given the availability of a
3 test to measure a genotype, at least answer the question
4 somewhere in drug development, is that an important
5 covariate. It's no different in my mind. The fact that
6 it's genetic doesn't make much difference to me in a sense
7 because it's a covariate that could be identified that may
8 have a significant effect on exposure and then
9 subsequently response. So it would seem to me you either
10 need to have some information that would say I need to
11 worry about it or I have information that would say I
12 don't need to worry about it.
13 If one thinks about the size of special
14 populations, certainly a genotyped group, i.e., a poor
15 metabolizer for a 2D6 substrate, is much larger in size
16 than would be, say, a patient population defined by their
17 renal function. So I think we're moving in that direction
18 as the science evolves and as our ability to identify
19 these covariates becomes more commonplace and available.
20 DR. SADEE: It appears to me that these models
21 all still assume exposure in terms of how much drug is
22 there. So the alternative is to look at special 1 135 2 1 populations that have a clearly different dose-effect
2 relationship. So what I'm struggling with, in terms of
3 trying to understand how to simplify this and how to still
4 get reality in there, is if you have two different
5 populations with the outlier, the toxicity occurrence, is
6 because somebody is genetically or environmentally
7 predisposed in a way that cannot be predicted by the
8 model, that it's just looking at exposure, how much drug
9 is in the body. And if you try to merge the two, you're
10 obviously making an error. On the other hand, it's clear
11 that even for those patients, the more drug you have, the
12 more likely it is that there would be an adverse effect.
13 So I think we're talking about two different
14 models here. One assumes we have this relationship
15 between how much drug is there and the effect, and the
16 other one would be there's a completely different
17 relationship. Those two have to be merged, I suppose,
18 without then incurring too large of an error.
19 DR. LESKO: Yes. I think that sort of brings
20 home a very interesting point, and we kind of talk about
21 it but don't know exactly what to do with it. And that is
22 the paradox of drug development, as I call it, that drug 1 136 2 1 development revolves around population signals and
2 population data, whether it's the efficacy signal, the
3 safety issue, or the pharmacokinetics. You're looking at
4 a population average.
5 Yet, as Dr. Sadee has mentioned, when you're
6 treating patients, you're worried about the individual
7 situation and how you can best optimize a dose in the
8 individual. When we talk about individualizing or
9 optimizing dose, I'm sure we're talking about it in the
10 context of a subpopulation or special population defined
11 by something I could measure or observe as opposed to the
12 individual that genetically may be predisposed.
13 I think it's easy to focus on pharmacokinetics
14 and much harder to focus on the factors that influence
15 receptor sensitivity or the things such as long QT and
16 things of that sort because they're a little bit more
17 complex, especially in the polygenic nature of being more
18 complex.
19 DR. VENITZ: Any further comments?
20 (No response.)
21 DR. VENITZ: Then it looks like we're going to
22 get an early break. We're going to break from, I guess, 1 137 2 1 right now until 12:45.
2 We have nobody signed up for the public
3 hearing. So we are starting at 12:45 with Dr. Karlsson's
4 presentation, the example number 4, in using exposure
5 response to recommend dosing adjustments.
6 So enjoy your lunch and we'll be back at 12:45.
7 (Whereupon, at 11:40 a.m., the subcommittee
8 was recessed, to reconvene at 12:45 p.m., this same day.)
9
10
11
12
13
14
15
16
17
18
19
20
21
22 1 138 2 1
2
3
4
5
6
7
8 AFTERNOON SESSION
9 (12:50 p.m.)
10 DR. VENITZ: Can we reconvene the meeting,
11 please?
12 Our next presentation is Dr. Mats Karlsson,
13 the faculty of Uppsala University of Sweden. He is going
14 to talk about optimizing dosing strategies for defined
15 therapeutic targets. Dr. Karlsson.
16 DR. KARLSSON: Good afternoon. I'm really
17 grateful for this opportunity to get some insight into the
18 American regulatory process.
19 So I'm going to talk about optimizing dosing
20 strategies for defined therapeutic targets. I'm going to
21 talk about target definition in relation to dose finding.
22 We have done some work, mainly simulation 1 139 2 1 work, but then also applications to a few real drugs under
2 development, and I'm going to focus more on those as
3 examples of what we've been doing.
4 The dosing strategy alternatives that there
5 are are, first of all, the single, one dose fits all
6 always, which of course is very convenient for patients,
7 clinicians, and producers, if it's appropriate, but often
8 variability and pharmacokinetics and pharmacodynamics will
9 lead us down the individualization route.
10 There are two types of individualization.
11 First, we can individualize based on patient
12 characteristics, observable, or feedback individualization
13 based on some measurement or observation of the patient,
14 or we can have a combination of these two.
15 So the next thing is the defined therapeutic
16 target. Anybody who is involved in decision making on
17 dosing strategies have to have some implicit target
18 concept. This might not always be quantitative. It might
19 not always be stated, but there has to be some target.
20 And if we were to quantitate it and actually spell it out,
21 we could base it on, most reasonably, the weighted balance
22 between beneficial effects and side effects. You could 1 140 2 1 also consider other endpoints like only one side of the
2 coin or drug concentrations or biomarkers, especially when
3 it comes to individualization in subpopulations, as we
4 have been discussing today. And, of course, the target
5 might differ between different patient subpopulations. We
6 want to treat more severe disease maybe more aggressively.
7 We need not only to know the target, but also
8 seriousness of deviation from the target, and we can use
9 an all or none criteria which we recognize from
10 pharmacokinetics as the simplest concept of therapeutic
11 window where all concentrations within the therapeutic
12 window are equally desirable and all concentrations
13 outside equally undesirable. And the pharmacodynamic
14 correspondent is the responder/nonresponder concept here.
15 But we usually think that biology works in a
16 more graded manner and we might want to actually value the
17 deviation from target in a more graded manner. Of course,
18 what's important there is the clinical picture, but we
19 might approximate that with various statistical
20 distributions. And also the seriousness of target
21 deviation may vary between patients, although I haven't
22 seen any examples of such applications. 1 141 2 1 When I'm talking about the target concept and
2 the penalty function, of course together these form the
3 utility function, and I understand that that was the topic
4 of talks at the last committee meeting.
5 So one scenario for selecting a dosing
6 strategy would be that it's based on some implicit
7 criteria on the target and penalty function, and then that
8 more or less stops there.
9 Another scenario would be the same thing, but
10 then take it a step further. If one has a population
11 model for the dose, two-target variable, one could
12 actually estimate based on the decided dosing strategy and
13 the model, what target and penalty function does that
14 correspond to and then assess whether it seems to be a
15 reasonable target, and if it is, stop there. Otherwise,
16 revise your dosing strategy.
17 A simple example that we came across when
18 collaborating with a drug company was a drug under
19 development for a disease which had two harmful events.
20 The frequency of one was decreased by the drug and the
21 frequency of the other was increased by the drug. Of
22 course, what exposure, what dose you would choose would 1 142 2 1 depend on how you evaluate these two against each other.
2 If this effect is deemed more harmful than this, you would
3 choose a low dose. Otherwise, the opposite, the high dose.
4 So we did some calculations. So the black
5 line here corresponds to what I was just saying. If we
6 weight the adverse event high compared to the event that
7 is beneficially affected by the drug, then we would choose
8 a very low dose, and if they're equally weighted, we would
9 choose a much higher dose.
10 In this case, the project team had already
11 selected a dose of 1, which corresponded to a weight of 3
12 to 1 for the two events. And when we presented the
13 project team with this, they said, well, this seems to be
14 a reasonable weight.
15 However, there were two sub-diagnoses that had
16 actually different PK/PD relationships, but they had
17 decided to go with the same dose in both subpopulations.
18 So, of course, that meant that the weighting was different
19 for the two subpopulations. So in one it was 4 to 1 and
20 in the other 2.5 to 1, and again the project team said,
21 well, that's reasonable because in the sub-diagnosis, when
22 the harmful event occurs, it's more serious than in the 1 143 2 1 other.
2 Then finally for renally impaired patients, a
3 dose of a quarter of that for the main population was
4 selected and it corresponds to the same weight between
5 adverse event and beneficial event.
6 So this is a way to rationalize a selected
7 dosing strategy after the effect.
8 Of course, one might have more use of defining
9 the target and penalty function beforehand, and I'm
10 certainly no expert in this area, but what seems to be
11 wise is to ask a clinician because they're really the ones
12 who are sitting on the information here although maybe not
13 used to formulating these type of functions in
14 quantitative terms.
15 If there is a drug first in class, consult
16 preclinical phase I data to assess what tolerability
17 issues there might be.
18 Consult literature and marketing which might
19 have done surveys in patients and clinicians of what are
20 deemed important features of a drug therapy for a certain
21 disease.
22 And then develop a few alternative targets and 1 144 2 1 penalty functions, and apply it to historical data, maybe
2 make up a bank of hypothetical patients to be ranked.
3 And then again ask a few clinicians about the
4 developed utility functions. Most likely they won't
5 agree, which might be a source for revising the utility
6 function, but even so, you might not always get full
7 agreement between different clinicians, and just as Lewis
8 said earlier today that we need to incorporate uncertainty
9 in all our aspects, maybe we need to include uncertainty
10 or variability in the utility function as well.
11 So if we actually have a defined utility
12 function, then we can proceed in a more rational manner.
13 If we have the utility function, we have a population
14 model for the dose-to-target variable, we can estimate the
15 best dosing strategies given different constraints such as
16 we're going to give everybody the same dose. We're going
17 to individualize, but only with two doses and based on a
18 covariate, or we might individualize based on feedback, et
19 cetera, and then select the dosing strategy based on
20 target fulfillment and practical considerations.
21 So what we would do in more detail for the
22 first step, if we want to optimize the one-size-fits-all 1 145 2 1 dose, would be to, based on the utility function and
2 population PK/PD model, we would actually maybe not need
3 the full PK/PD models that we usually use if we're only
4 considering steady state concentrations' relation to
5 effect. We might only need the model for clearance.
6 Covariate models are essentially superfluous because
7 they're not going to affect our dosing, and we would,
8 based on these models, simulate a large number of
9 hypothetical patients. We would obtain a prediction of
10 each individual's deviation from the target for a certain
11 dose, and then obtain the optimal dose by minimizing the
12 overall loss. In this case, we can do this simply by just
13 repeated simulations, trying different doses, but we could
14 recast the problem as an estimation problem and estimate
15 the dose instead.
16 This actually is just a pictorial slide
17 showing the same thing.
18 So if we want to actually do
19 individualization, the questions become more and it's more
20 problematic of how to do it best. I'm only going to focus
21 on this first question for the case when we want to do
22 feedback individualization. What dose strength should be 1 146 2 1 made available?
2 We came into such a problem when collaborating
3 with a company that had developed a drug. It was planned
4 to go into phase III as a fixed dose size, everybody
5 getting the same dose, but in light of high variability in
6 PK and PD, partly because of polymorphic 2D6, they were
7 contemplating maybe doing individualization. And the
8 question was, what would we gain by that? Would it be
9 sufficient? And they wanted the gain to be measured on a
10 responder scale and the overall responder rate.
11 We went ahead and did the estimations based on
12 one-dose-fits-all or a feedback individualization with two
13 dose strategies where we estimated the lower dose size,
14 the higher dose size, and the fraction of the patients
15 that would be preferentially treated with the lower and
16 the higher dose. I won't go into technical details here,
17 but we used the $MIXTURE function in NONMEM.
18 DR. SHEINER: (Inaudible.)
19 DR. KARLSSON: No. This is feedback
20 individualization without --
21 DR. SHEINER: This was feedback on the
22 response? 1 147 2 1 DR. KARLSSON: Yes.
2 DR. SHEINER: So actually somebody was
3 observed to be a responder or not a responder, and then
4 the dose was changed.
5 DR. KARLSSON: Yes.
6 We had built population PK and PD models for
7 both the satisfactory effect and for the side effect
8 previously. These were more elaborate models with
9 continuous and ordered categorical type data, but they
10 could be easily reduced to the dichotomous question of
11 were they responders or not responders.
12 And when we estimated the single dose size to
13 be given, it was very close to what the project team
14 actually had come up with, and it resulted in 47 percent
15 overall responders.
16 The best two-dose strategy was two doses, one
17 lower and one higher, a 4 and 18, with about 60 percent
18 gravitating towards the higher dose. This was predicted
19 to increase the overall responder rate to 63 percent.
20 This maybe was one piece of the puzzle that made the
21 company actually in the phase III to go with both fixed
22 dose and individualization in parallel. 1 148 2 1 We did simulation studies on similar problems,
2 and just to relate two observations there, one was that
3 although a particular dosing strategy may not be the most
4 optimal for one utility criteria, it may be near optimal
5 across all relevant utility criteria, and therefore may be
6 superior to other dosing strategies.
7 Also, another observation that all-or-none
8 type responder definitions, like in this example, seems to
9 favor individualization to a higher degree than gradual
10 utility functions. Although this was obtained from a
11 single example, it seems reasonable that if you have these
12 very harsh, steep benefits of changing maybe somebody just
13 a little bit on a continuous scale into having a utility
14 of 0 to have one of 1, that that is more sensitive to
15 individualization.
16 Moving on from feedback individualization to
17 individualization based on covariates identifiable up
18 front when the patient is to be started on a therapy, we
19 will face questions such like what is the best covariate
20 to base dosing on? What should the number of dose sizes
21 be, and what covariate intervals should each dose size be
22 applied to? Just illustrating here, the covariate here 1 149 2 1 might be organ function, body size, age, et cetera.
2 If we want to go with two dose groups, the
3 parameters we need to identify are what is the optimal
4 cutoff value, what is the dose in the higher group and the
5 dose in the lower group? So that's three parameters to
6 actually estimate.
7 If we wanted to have three dose groups, that's
8 five parameters, et cetera. The problem becomes more
9 complex.
10 We would proceed to estimate those parameters
11 in a very similar fashion as before, but with some
12 differences. First of all, in this case it's, of course,
13 very important to have covariate models for the covariates
14 that we are intending to be using for dosing decisions,
15 and we also need to have distributions of these covariates
16 in the target patient population. We can obtain those
17 from simulations, but more relevant maybe from empirical
18 databases from previous studies. What we're estimating
19 are then the dose sizes and the cutoff values for them,
20 where to change dose size.
21 We also had an example of this in relation to
22 drug development. I can actually name this. This is NXY- 1 150 2 1 059, a drug under development by AstraZeneca. We had a
2 publication coming out in Clinical Pharmacology and
3 Therapeutics in January this year.
4 This is a drug to be used for stroke. It's
5 acute dosing, a 72-hour infusion with a 1-hour loading
6 infusion.
7 The project team was worried about too high
8 variability if one were to give everybody the same dose in
9 particular since this is entirely -- not entirely, but
10 mainly renally cleared compound, and they were worried
11 about the end of loading infusion and the maintenance
12 infusion concentrations.
13 So we tried to see what individualization
14 could do in terms of bringing down variability to a
15 reasonable level. The target that was set was a free
16 concentration of 100 micromolar. The penalty function
17 used was quadratic loss in log domain which means that
18 half the target concentration is as bad as twice the
19 target concentration.
20 We had a pop PK model developed from the first
21 patient study which showed clearance being highly
22 dependent on creatinine clearance and volume on body 1 151 2 1 weight.
2 We used empirical covariate distributions from
3 previous phase III studies of stroke patients.
4 And for the loading infusion, we considered
5 one dose, the same to all, or two-dose groups either based
6 on creatinine clearance or on weight.
7 For the maintenance infusion, it was clear
8 that we could not give everybody the same dose, but two to
9 four dose groups were explored and dosing were to be based
10 on creatinine clearance.
11 An additional constraint was made, which said
12 that the therapy has to be fulfilling the following
13 criteria, namely that 90 percent of the patients have to
14 be above 70 micromolars and less than 5 percent above 150
15 micromolars.
16 As it turned out, actually these criteria
17 could be met by just giving one loading dose of 2,400
18 units to everyone, but for the maintenance infusion, it
19 was necessary to give three different infusion levels,
20 depending on the creatinine clearance, with the cutoffs at
21 80 and 50 mls per minute. And you can see the dose units
22 there. 1 152 2 1 This dosing design was implemented in a phase
2 II study, and the target fulfillment was acceptable, with
3 more than 92 percent above the lower limit but more than 7
4 percent above the upper, which is slightly more than what
5 was desired.
6 So actually before that, we had done some
7 simulations to look at a priori dosing based on
8 covariates, and we took dosing on creatinine clearance as
9 an example. The standard approach often used when
10 individualizing doses based on renal function is to use
11 predetermined cutoff values for the renal function and
12 quite large dose decrements, often a factor of 2 or
13 higher, when going down to lower renal functions.
14 We wanted to explore what would be optimal
15 approaches to renal based dosing and we wanted to see what
16 drug characteristics and what other factors influenced
17 what would be the optimal approach. So we did simulations
18 where we changed the drug characteristics of hypothetical
19 drugs, where we changed the creatinine clearance
20 distribution in the target population, and also the
21 utility function shape.
22 And two factors came out as by far the most 1 153 2 1 important. For the selection of what would be the optimal
2 cutoff value in the patient population, the median
3 creatinine clearance in the patient population was most
4 important. And if only two dose groups are to be used,
5 the cutoff should be ideally positioned close to that
6 median value regardless of other drug characteristics.
7 For the dose ratio, the ratio of the high to
8 the low dose, the one factor that again by far was the
9 most important was the strength of the covariate
10 relationship, and for renally cleared compounds that can
11 be expressed as the fraction excreted unchanged. So when
12 the fraction excreted unchanged is 1, the higher to the
13 lower dose ratio would be around 1.7. And for all other
14 situations, the ratio would be lower than that ideally.
15 These two pictures are quite in contrast to
16 what the practice is today which is using cutoff values
17 below the median value and often dose ratios higher than
18 1.7. This might have practical and other reasons, but
19 it's also to be maybe recognized that this has an impact
20 on the target fulfillment.
21 This is just a picture showing some of the
22 gains that could be made from doing individualization 1 154 2 1 based on fraction excreted unchanged and the number of
2 dose levels.
3 This is actually a picture very similar to the
4 one that got us involved in this area. It was when again
5 we collaborated with a drug company who had already
6 decided upon a dose individualization scheme where they
7 had actually selected a low cutoff value and a large dose
8 decrement. So what in effect they were doing was they
9 changed around the doses, but they didn't manage to lower
10 the variability in exposure. It was different types of
11 patients that were in the tails of the distribution, but
12 the overall variability in exposure was not reduced by the
13 individualization. So doing this might actually not be
14 the most simple task.
15 So to summarize with respect to target
16 definition or utility function definition, this is
17 something that certainly can aid data collection. If one
18 knows what parameters are the most important and therefore
19 go into the utility function, that will aid both data
20 collection and modeling efforts, being able to maybe take
21 both quantitatively and qualitatively better data for
22 those variables and also do modeling more focused on those. 1 155 2 1 To improve communication within project teams
2 or maybe between project teams and those outside. My
3 experience is that many of the important factors for the
4 utility is something that resides with only a few persons
5 within the project team and many of the others that are to
6 contribute to the dosing decisions are not particularly
7 well informed about the weighted balance between effects
8 and side effects or between different types of effects.
9 And of course, it's important to appropriately
10 value the drug compared to other drugs.
11 And the last point. This was a slide prepared
12 for a meeting in Europe last December that had the title
13 "Getting the Dose Right." If you ever want to say that
14 you got the dose right, obviously you need to know what
15 you're aiming for.
16 Separate from defining the utility function,
17 which in its own right has a lot of benefits, dosing
18 strategy estimation might have additional benefits to
19 motivate the choice of dose or dosing strategies and to
20 obtain conditions for optimal individualization and
21 thereby assess the maximal potential value of
22 individualization to justify doing it or, oppositely, 1 156 2 1 maybe to justify not doing it before because the benefit
2 of doing it isn't large enough. And if you do know the
3 optimal individualization strategy, then you can directly
4 offset any practical consideration that simplifies dosing
5 against a decrease in target fulfillment.
6 I know that some people don't believe me when
7 I say it's easy, but compared to doing population PK/PD
8 modeling, compared to defining utility functions, which
9 are the two more difficult tasks, I think this is very
10 easy.
11 Thank you.
12 DR. VENITZ: Thank you very much, Dr. Karlsson.
13 Any questions or comments by the committee?
14 Let me ask you, in your simulations, you used a symmetric
15 loss function. Right?
16 DR. KARLSSON: We used both symmetric and non-
17 symmetric loss functions. Obviously, if you use non-
18 symmetric so you penalize the high concentrations or
19 adverse events more, then you're going to gravitate
20 towards lower doses. If you are non-symmetric in the
21 other direction, you're going to penalize the low doses,
22 so you're going to get a higher overall dose. 1 157 2 1 DR. VENITZ: So your conclusions with regard
2 to the effect of mean creatinine clearance, that was based
3 on a symmetric loss function. Right?
4 DR. KARLSSON: We actually did explore various
5 loss functions also there, and across a range of
6 reasonable loss functions that we thought are reasonable,
7 it was pretty stable towards that. But obviously if you
8 have a very asymmetric loss function, you will tend to get
9 other values.
10 DR. VENITZ: And my guess would be that's the
11 reason why your recommendation is different than what's
12 currently done because what people build in is a loss
13 function where they're very worried about overdosing, less
14 worried about under-dosing. So the easiest way to account
15 for that is by adjusting the dose in people that have
16 renal failure.
17 DR. KARLSSON: Yes. I think that's true, that
18 they are usually dosed to an average AUC that's actually
19 below what's seen in the main patient population. So if
20 that is what is decided, then certainly that's the case.
21 For the example we had here, where actually we're
22 operating at one point with different loss functions for 1 158 2 1 people with lower renal functions because if there was a
2 concern about renal function, then we wanted them to have
3 a lower target as well. So there is certainly the
4 possibility of incorporating all these types of
5 considerations.
6 I think the reason why we see more lower
7 cutoffs and large dose decrements is probably the way drug
8 development pursues with more healthy patients to start
9 with and then inclusions of larger and larger. So the
10 initial dose levels are based on those with higher renal
11 function and then the other ones are added as a tail more
12 towards the end.
13 DR. SHEINER: I just wanted to sort of raise a
14 point in the questions. You were romping through your
15 slides. So you looked at individualization based on that
16 drug that you told us the name of, that slide you just
17 showed a moment ago where it had the 90 percent and the 5
18 percent?
19 DR. KARLSSON: This one?
20 DR. SHEINER: No. At the bottom of the slide,
21 it had selection of dosing strategy.
22 DR. KARLSSON: Okay, yes. 1 159 2 1 DR. SHEINER: Yes, there.
2 Correct me if I'm wrong. I think one
3 important thing to realize is that those numbers down at
4 the bottom, of course, might not be attainable by any
5 strategy. But to discover whether they are or are not
6 attainable by some strategy, you fix on a model for the
7 process. Then those percentages there have got to do with
8 variability among patients. So this doesn't contain that
9 uncertainty that I was talking about.
10 The model that says how frequently you'll get
11 toxic at a given level, that very model is itself
12 uncertain because it's based on assumptions and it's based
13 on data. It would be very likely with not a great deal of
14 data -- it would be impossible to attain a 90 percent
15 probability of being in the right range or whatever it is
16 if you incorporate that kind of uncertainty because that
17 uncertainty says, I don't know how the world works. So
18 how can you possibly get 90 percent certainty on anything?
19 And there's nothing wrong with this. This is
20 the right way to proceed. Then you want to look --
21 presumably you did -- at the sensitivity of those two
22 various assumptions that went into your model. 1 160 2 1 But this whole business of being clear with
2 people about what uncertainty you're bringing in when,
3 when it's appropriate and when it's inappropriate -- it
4 would be here inappropriate to bring in model uncertainty
5 when you're trying to find the strategy because it's got
6 to condition on some state of nature.
7 So I just wondered what your experience was
8 with dealing with all of those, I think, somewhat subtle
9 issues with people who are not used to speaking this
10 language.
11 DR. KARLSSON: It is true that we actually
12 didn't -- at first, when we saw these, we thought they
13 were too stringent criteria. With normal variability,
14 this would be very hard to achieve, and as you say, it
15 might not actually be possible.
16 When doing these calculations, we did not take
17 the structural model uncertainty into consideration, but
18 we did do simulations looking at the uncertainty in the
19 population PK parameters, what the impact of that would
20 be. And it wasn't actually as large because they were
21 relatively well defined. But this is, as you might guess,
22 done with the point estimates. 1 161 2 1 In general, I do find it often difficult to
2 discuss these matters with the project team I think maybe
3 because it's not usually done, talking about quantitative
4 models in this sense. The type of utility function that
5 does seem to be used is the responder/nonresponder
6 criteria. So it's easier if it's simpler, but again, the
7 discussions around the responder criteria was where to put
8 the cutoffs and people differed in their opinion of where
9 to put the cutoff, which is the same thing.
10 DR. LESKO: Mats, I want to bring you back to
11 a little more maybe pragmatic question. But clearly this
12 can be done in the context of what we talked about this
13 morning in having a data set in front of us and then
14 looking at an approach like this to adjust -- or make a
15 decision based on adjust dose.
16 However, as I listened to you speak about the
17 project team where there's a need to define a target and
18 the penalty function, this method requires a weighted
19 balance between the effectiveness and the safety. I guess
20 in some cases that's very clear, depending on the nature
21 of what the effectiveness and safety is.
22 If you think of risk as sort of an overriding 1 162 2 1 issue in drug development as opposed to efficacy -- in
2 other words, the risk of an adverse reaction, limiting
3 approval, limiting dosing, limiting the label -- does the
4 notion that a safety consideration would drive the
5 relative agreement that you would have in a project team
6 on the utility function -- in other words, I'd give more
7 weight to a safety side of the drug's effect as opposed to
8 the efficacy side. As an example, if I had a QTc issue,
9 that would seem to weigh heavily in terms of my utility
10 function consideration even against the most promising
11 efficacy that I might be speaking about. So it would be
12 the driver, if you will, in trying to get an agreement on
13 what weights to assign to the utility functions.
14 Can you speak to that a little bit? How does
15 that work? And we sort of talked about this last time.
16 You said ask a clinician, but we have actually asked
17 clinicians after our October meeting and we do get quite
18 different views of how a utility function would serve the
19 purpose of what we're talking about. So I wondered if you
20 can sort of pursue that a little bit.
21 DR. KARLSSON: I wouldn't have any hopes or
22 belief that utility functions within the very near future 1 163 2 1 would drive the decisions so that you would just define a
2 utility function and then forget what went in there. I
3 think a more reasonable way is to actually come up with
4 some initial utility function and then use that in
5 parallel maybe to illustrate different consequences of
6 decisions. Then that would maybe show up where utility
7 functions would actually fail because it wouldn't take
8 something into consideration that is of importance, which
9 would point to how they need to be refined, what needs to
10 be considered in them.
11 When it comes to the safety, I think in some
12 situations where there are tolerability issues of maybe
13 not so severe nature as QTc prolongation, it's easier to
14 incorporate them. Obviously, severe side effects are --
15 by its nature, you're not going to see very many of them,
16 and you're going to not want to expose patients or
17 volunteers up to the range where you get a very good
18 handle on what the function is at the dose levels. So
19 you're going to have a large uncertainty in the models at
20 that range, and you need to take that into account I
21 think. So you're going to have a much larger uncertainty
22 on the upper end than on the lower end probably. 1 164 2 1 DR. SHEINER: I know you know my response to
2 this, Larry, but I just wanted to say it keeps on
3 reminding me of the old data analysis argument about the
4 Bayesians and frequentists and the issue of I'm trying to
5 fit a model that's too big for my data, and so I'm going
6 to fix some parameters. The Bayesian, of course, hearing
7 that, says well, okay, you may not think you'll be acting
8 like a Bayesian but fixing parameters is the same as
9 saying they have a prior distribution that's got point
10 mass at the value that you said. That can't ever be as
11 good as giving it a little bit of wiggle room.
12 Well, it's the same kind of thing here with
13 utility. Of course, if you have trouble convincing people
14 that they ought to sit down and do a utility function,
15 then there's some implicit utility function that's
16 dominating usually by the most aggressive person or the
17 person highest up in the organization who happens to be at
18 the table. And you don't know what it is. If you believe
19 in decision theory, it's got to be there somewhere, and
20 it's what's overriding everything else.
21 Just listening to the talk and saying,
22 wouldn't I like to be in a room where the discussion was 1 165 2 1 about how much do I believe that these data really do
2 support this notion of what's going to happen, how much
3 weight do I put on versus how much you put on, the various
4 side effects. It seems like such a rational and sensible
5 discussion to have rather than, well, I think we ought to
6 go with 25 milligrams. What do you think, Joe?
7 DR. VENITZ: Any further questions for Dr.
8 Karlsson?
9 (No response.)
10 DR. VENITZ: Thank you again for your
11 presentation.
12 Peter, do you want to pose the questions for
13 the committee?
14 DR. LEE: I think the question is very simple.
15 Can this methodology be generalized to other scenarios?
16 That's pretty much the question that we have.
17 DR. VENITZ: The question for the committee
18 is, can this utility approach be generalized to other
19 therapeutic areas?
20 DR. SHEINER: Well, why isn't it just about as
21 general as it gets? This is one of those things that
22 really is -- you know, it applies to drugs and automobiles 1 166 2 1 and airplanes. How can you make any decision if you don't
2 know what you're deciding about and what your values are
3 and what the likely state of nature is? That's all we're
4 really talking about here. Then after that, it's
5 computation.
6 DR. VENITZ: I obviously second that, but I'd
7 also like to maybe point out an approach of how we can
8 convince other stakeholders that this is something useful.
9 And it goes towards identifying utility values generically
10 for certain kinds of adverse events, just like you would
11 assign generically speaking for certain efficacy, life-
12 threatening disease versus quality of life changes, and
13 get agreement on that regardless of what drugs you're
14 looking at rather than what you presented are specific to
15 that drug where a decision was made this is how I define
16 my utility function relative to my target, which makes it
17 vary case by case.
18 But I think if you want to get consensus,
19 let's start on the safety side. We agree certain adverse
20 events are going to get certain utility values associated
21 with it regardless of what causes it. By the same token,
22 on the efficacy side, certain disease interventions get 1 167 2 1 certain utility values assigned regardless of whether it's
2 a drug treatment, device treatment, or whatever it is.
3 Then at least there's some common acceptance I guess on
4 what the ranges are.
5 Otherwise, you're really getting into this
6 swamp of, well, everything is a case to case. Show me the
7 drug and that's the only way I can come up with a utility
8 function, which defeats the purpose in my mind at least of
9 using utility functions.
10 DR. LESKO: It just strikes me what you said
11 would lead to agreement particularly on the safety side.
12 I think you can identify certain safety signals that
13 people would agree are bad and are most serious, and then
14 weigh those, in turn, against a range of efficacy or
15 benefits that one might get from a drug and against the
16 loss of efficacy if you were to somehow misappropriately
17 adjust the dose.
18 I think you can define boundary conditions in
19 a sense with regard to certain safety signals, with regard
20 to certain categories of drugs for efficacy. For example,
21 hepatic toxicity, QTc. I think you'd get general
22 agreement these are bad. We focus on those extensively 1 168 2 1 irrespective of what the efficacy side is offering. By
2 the same token, a disease state where efficacy is of
3 extreme importance, there may be a different view of the
4 safety signals. But you need some anchor points, it would
5 seem, and they might be those boundary conditions, and
6 then the gray areas fall in between somehow.
7 DR. VENITZ: But if you look on the adverse
8 event side, in oncology there is already a consensus of
9 how to categorize and how to rank order certain adverse
10 events according to organ function. It's unanimously
11 accepted by the whole oncology community. Now, they don't
12 necessarily use it in the context of utility. They use it
13 to define dose limit of toxicity.
14 But why can't we do that as a general approach
15 to adverse events? And you would do that regardless of
16 the underlying cause of that. Whether it's a specific
17 drug or other drugs, it's irrelevant. To get out of this
18 discussion where everything is case by case, the moment
19 you do that I don't think the utility approach is going to
20 work. It might work on specific drugs, but it wouldn't
21 work across the board because you've got nothing to
22 compare to. 1 169 2 1 DR. KARLSSON: Although I agree and I like the
2 idea. Maybe one complicating factor is those types of
3 grading 1 to 4, which I assume you're talking about, are
4 based on outcomes, aren't they? Whereas, if you have like
5 a biomarker for a safety event, that's something a bit
6 different and maybe more difficult to value.
7 DR. VENITZ: Maybe.
8 DR. FLOCKHART: I'm just thinking about
9 something I know more about which is the QT interval. So
10 the question would be, is a given QT prolongation, a
11 prolongation of 10 milliseconds, by one drug the same as a
12 QT prolongation of 10 milliseconds by another drug?
13 Unfortunately, the answer is no because drugs do more than
14 one thing. They have more than one mechanism of effect on
15 the QT interval. For example, many antipsychotic drugs
16 that affect the QT interval are anticholinergic as well,
17 so they have effects on the heart rate, and that might be
18 a little bit protective, make them a little bit
19 tachycardiac. In situations where they got more drug,
20 they would not only get more IKr blockade, but they would
21 get more mu 1, M1, receptor blockage.
22 So because of that, I think hepatotoxicity 1 170 2 1 gets even more complicated. If one drug causes X amount
2 of change in the LFT's, is that the same as another drug
3 causing the same? It's a very complicated thing.
4 Nevertheless, I think the effort is probably
5 noble. It's worth venturing along that path at least to
6 find out how different things are. I'm intuiting that QT
7 interval drugs would be different. I don't really know.
8 DR. VENITZ: But I think the current --
9 DR. FLOCKHART: You need an outcome, though.
10 You need an outcome, and that's a very important point.
11 DR. VENITZ: As far as QTc is concerned, the
12 current assumption is that QTc prolongation is bad
13 regardless of what causes it. The relationship then
14 really is what you're talking about, how predictive is a
15 biomarker of a clinical outcome? Because the biomarker
16 itself is not bad. It's the clinical outcome. You're
17 worried about the fatal arrhythmia. So in addition to
18 your utility, now you have to look at what are your
19 uncertainties involved in linking that biomarker that
20 you're measuring to the final outcome. In other words,
21 for a given change in, I don't know, 40 milliseconds in
22 QTc, how many people are likely to develop fatal 1 171 2 1 arrhythmias.
2 DR. FLOCKHART: Yes. You can put a model on
3 it, and you can also model in things like there are a
4 significant number of drugs that prolong the QT interval
5 for which torsade has never been reported, and we have
6 them on the QT.org website. You could model that in as
7 well. That's a chance that a drug in that class would not
8 do it.
9 DR. VENITZ: But from my perspective, that's
10 not a utility. That has something to do with how your
11 biomarker relates to the clinical outcome.
12 DR. FLOCKHART: Right.
13 DR. VENITZ: Because really what you're
14 assigning utility to is the bad outcome and fatal
15 arrhythmia in --
16 DR. FLOCKHART: Well, it is affecting the
17 utility, though, because it alters that number.
18 DR. SHEINER: (Inaudible.)
19 DR. FLOCKHART: Dr. Sheiner was pointing out
20 that it affects the expected utility, and he's right.
21 DR. VENITZ: Any other comments or more
22 specific questions of either Dr. Karlsson or the FDA? 1 172 2 1 DR. KARLSSON: I was just going to add to that
2 that maybe if you do start with a utility function that
3 incorporates many different aspects of the drug therapy,
4 you will find that actually in the end it's only really a
5 few of them that are going to be important and you could
6 reinspect those and the assumptions that go along with
7 those particular events.
8 DR. SHEINER: Let me just ask. Some of these
9 examples were actual examples from your interaction with
10 the industry. I guess we can't draw too much of a
11 conclusion from anecdotal experience, and also these folks
12 asked as opposed to having it pushed at them by a
13 regulatory agency.
14 Is it your feeling that this exercise was
15 appreciated, informed people, and had some consequence in
16 terms of the way in which the development plan went
17 forward?
18 And finally, do you have any information on
19 whether or not the regulatory agencies with which these
20 companies dealt -- how they responded to the
21 justifications offered through this means?
22 DR. KARLSSON: I don't think any of these 1 173 2 1 drugs have gone to the regulatory authorities yet,
2 although I might be wrong.
3 For the example that I presented, the
4 AstraZeneca one, they were very helpful and really
5 appreciated all our efforts.
6 In the other cases, I think it was more add-on
7 and maybe it was not the core of the project team that was
8 really wanting to have this information. It was more a
9 side effect of having the possibility of doing it, and
10 they thought maybe it was nice to know, but to what extent
11 it influenced their decision I'm not sure.
12 DR. DERENDORF: A follow-up question. How
13 many times do you think this is done a posteriori so that
14 the decision is made first and then you need a
15 justification for it?
16 DR. KARLSSON: I don't think that's done very
17 often. I don't know, but in my experience outside the
18 examples I've been involved in when I've actually done
19 this, I haven't heard these kind of discussions going on
20 with people in industry.
21 DR. LESKO: Just following up on the same
22 question. Maybe this isn't a fair question, but I sense, 1 174 2 1 at least in our regulatory agency, a stronger desire to
2 understand dosing strategies than there has been in the
3 past. It's partly related to a lot of things.
4 The question I sort of have is a follow-on to
5 one that was asked, and that is, is there anything in your
6 experience in working with the method, within the context
7 of drug development, that would cause any concern on the
8 part of a sponsor with regard to what a regulatory agency
9 might ask of this kind of methodology?
10 It would seem like it would be received rather
11 positively because it provides a fair amount more than we
12 normally would see in terms of a dose justification or a
13 rationale for dose selection. Therefore, my sense would
14 be it would received positively.
15 But on the other side of that coin is anything
16 different is different, and what are the issues that
17 somebody might think about with regard to if I presented
18 this, the regulatory agency might do X, Y, or Z?
19 DR. KARLSSON: Well, I've heard both
20 arguments, both that it's good to do your homework and to
21 be able to know your drug so you can argue well. I've
22 also had the response that let's not do it because we're 1 175 2 1 waking the sleeping bear, or whatever your expression is,
2 that maybe they wouldn't think about this if we haven't
3 done it.
4 DR. DERENDORF: Well, but for internal
5 decision making, I think it's always helpful to do it.
6 DR. KARLSSON: I think to some extent, though,
7 it goes hand in hand. If you don't want others to know
8 it, you don't want to know it yourself.
9 (Laughter.)
10 DR. VENITZ: Any final comments?
11 (No response.)
12 DR. VENITZ: Thank you again for your
13 presentation.
14 Then we are going to start with our second
15 topic and the last topic for today which is the pediatric
16 database. I think Dr. Lesko is going to give us an
17 introduction to that topic.
18 DR. LESKO: Thank you, Jurgen.
19 This is going to be a rather brief
20 introduction. It's to lay the groundwork really for the
21 next two presentations. The focus of this now, switching
22 gears, is pediatrics, and we're going to be talking about 1 176 2 1 two topics. The first is going to deal with a template or
2 standardized approach, if you will, for pediatric studies
3 that utilize sparse samples. The second is to discuss
4 with the committee some ideas that we have for mining the
5 pediatric database that we have as a result of the
6 pediatric rule.
7 Let me start with that. The pediatric rule is
8 something many of you are familiar with who were actually
9 here at our last meeting because we discussed it in some
10 detail. We shared with you the types of studies that
11 we've received under the pediatric rule. Basically the
12 rule encompasses the idea that we want to bridge the adult
13 data to the pediatric situation, usually not directly but
14 through the initiation of some studies in the pediatric
15 population, and depending on the nature of the adult data
16 and the nature of the assumptions that we make in doing
17 that, the types of studies that are conducted for
18 pediatrics would be either efficacy, safety, or
19 pharmacokinetics. We presented last time a pediatric
20 decision tree that sort of guided the thought process on
21 what studies to conduct.
22 Basically the idea with the pediatric rule is 1 177 2 1 to avoid large-scale studies. It's considered inefficient
2 to redesign the efficacy and safety trials when there's a
3 preexisting database in adults. Although the challenge
4 from a clinical pharmacology perspective and a clinical
5 perspective is to decide what studies are, in fact,
6 appropriate to conduct and will provide the most
7 information.
8 The pediatric rule, as you know, is intended
9 to speed up access of drugs for children, to do that in a
10 cost effective way without reinventing the entire
11 efficacy-safety spectrum. I think overall I think most
12 people feel this has been generally successful in meeting
13 the goals that were intended for it.
14 But some questions always need assessment
15 within the context of the pediatric decision tree and the
16 pediatric rule. Is it reasonable to assume a similar
17 PK/PD relationship as exists in adults? This sounds very
18 familiar to the questions that we were discussing this
19 morning in terms of different populations, and probably
20 the underlying principles are very much the same.
21 We feel there's a need -- and we've begun to
22 look at this -- a need to develop standard methods for 1 178 2 1 answering this question for specific drugs and drug
2 classes particularly where we've seen submissions in
3 pediatric patients in drug classes over a number of drugs.
4 The other question is what is the appropriate
5 dose. And to answer this question, we generally rely on
6 pharmacokinetic studies. They can be of two types: the
7 full exposure, sample-rich type study design, or the
8 sparse-sample, population PK approach.
9 The overall goal of these studies is
10 straightforward, to achieve a dosing that's intended to
11 achieve exposure similar to adults. Generally PK studies
12 aren't the only studies conducted in pediatrics.
13 Frequently a safety study or in some cases a small
14 efficacy study is also required.
15 But it's probably safe to say that a sparse-
16 sample strategy has been under-utilized in pediatric
17 studies. We have some ideas why that might be. It might
18 be that there's an insufficient understanding of this
19 approach. It might be that there's some concern that the
20 regulatory agencies won't accept such an approach. But it
21 seems like there's room for opportunity here to do these
22 types of studies that call for pharmacokinetic information 1 179 2 1 in a sparse-sample strategy that is based on good
2 principles.
3 So the first thing we wanted to talk about
4 today was a discussion of a standardized -- and I use
5 "standardized" very generally, but a standardized study
6 design template that would be useful, for example, in
7 communication between investigators or between companies
8 and the agency to agree on a sort of starting point for
9 designing both acceptable and informative PPK type
10 studies. So that would be the first thing we're going to
11 talk about, and Dr. Peter Lee will talk about that
12 primarily.
13 The next topic is a follow-on to what we
14 talked about in October. We had made the point and we
15 shared with you the specific drugs in October for which we
16 have a database on pediatric studies. We raised the
17 question about if this were your database, what would you
18 want to learn from the database and what would you look
19 for. What would be the questions you would have? We
20 didn't spend a lot of time with that. We have more time
21 today. But we've begun to look at this database and
22 you'll hear from another individual from our office, Dr. 1 180 2 1 Gene Williams, who is on detail to the immediate office of
2 OCPB to specifically look at what we can learn from this
3 database.
4 This is a work in progress. We began to
5 assemble the database, in an organizational way of age
6 groups, the PK data that we have. We've begun to look at
7 individual drugs in the database, elimination pathways,
8 the clinical endpoints that were studied.
9 I'd have to say this has not been easy.
10 Unfortunately, this database is not in a form where you
11 can just press a few buttons and pull it out. So there's
12 a fair amount of up-front work that goes into assembling
13 the database. I think in recognition of that, it's
14 important that we understand where we want to go with the
15 questions that will derive from this effort to organize
16 the database once we start moving in that direction.
17 So Gene is going to do an overview of some of
18 the research objectives that we have, with the goal of
19 generating knowledge from this database. What we think
20 we'd like to do is look at underlying mechanisms where
21 there are exposure differences between kids and adults,
22 look at breakpoints perhaps on age, look at specific 1 181 2 1 elimination pathways.
2 The reason we want to do all this is it's sort
3 of common sense. We don't feel we should be asking the
4 same questions now of pediatric studies that we asked
5 three years ago before we had 50 or 60 studies in house.
6 So the idea is to learn from the database and then use
7 that information and knowledge to revise our pediatric
8 context or pediatric decision tree for the future.
9 So the goal of this effort: to improve or
10 revise our pediatric decision tree based on identifying
11 better studies that we need to conduct in the future --
12 and by "better," I mean more informative studies -- or
13 perhaps reduce the number of studies in the areas that the
14 data would allow. So for that, Gene Williams will present
15 an update.
16 And Peter is first.
17 DR. VENITZ: Lew.
18 DR. SHEINER: Can I just ask you a question,
19 Larry? On your third slide where you have "some questions
20 always need assessment," and you say, "is it reasonable to
21 assume a similar PK/PD relationship as adults," I
22 understand that that's not particularly the focus of 1 182 2 1 today. I'm sensitized to this because of some work
2 recently that I've been doing with a topic that's related
3 to this with Novartis.
4 It's an interesting question. What would
5 constitute evidence that children are like adults with
6 respect to PD now, not the PK part? I just wanted to put
7 it on the list of many things that you have to think
8 about, that maybe we ought to address that at some later
9 point, or maybe you might want to have us address that at
10 some later point. It's a vital question. Obviously, if
11 what it takes to establish the similarity relationship is
12 more work than it takes to clear a drug to start with,
13 you're done.
14 DR. LESKO: Yes. I think it's a vital
15 question. We have a few instances, a few drugs in hand
16 where we actually have information both on the adult where
17 that happened to be done in the NDA and it was also done
18 in the kids. We're going to focus on a few of those
19 examples to try to answer that question.
20 We've often turned the question around and
21 asked how much of a difference would be important in the
22 PK/PD relationship and would that warrant necessarily a 1 183 2 1 different dosing strategy in kids. These are open
2 questions, but we do not have a lot of opportunity to look
3 at this issue based on, at least, what I've seen of the
4 database to date. Maybe Gene will comment a little more
5 on that.
6 DR. SHEINER: Right. It does relate to that.
7 But it's also a question of style in the following sense.
8 If the data are sort of anecdotal, are they really data?
9 So the issue becomes -- when you say I think that the
10 relationship is the same, and then I say to you, okay,
11 here's some evidence, and you say, well, that's not really
12 evidence because that's physicians' opinions, let's say.
13 So the question is what constitutes evidence that it's the
14 same, given that you're inclined to believe it. That's
15 the issue I was getting at.
16 DR. LESKO: Right. That's a good question
17 whether it's in the statistical domain or whether it's in
18 the modeling domain. I think we need to talk about that,
19 but I think that's a good open question for another
20 discussion.
21 The point of bringing that to the committee
22 might be when we've done our analysis of the database as 1 184 2 1 we have it and share with the committee what we've learned
2 from the data that we have and maybe what we might want to
3 know from data we don't have at the moment but might
4 recommend somebody begin to look at, not necessarily in
5 addition but maybe as an alternative to the studies that
6 are being conducted today.
7 DR. VENITZ: Peter?
8 DR. LEE: What I'd like to present to the
9 committee today is a proposal to develop a pediatric
10 population PK study design template. I believe we have
11 sent a copy of the full proposal to each committee member
12 a few weeks ago, so hopefully you would have had a chance
13 to look at the proposal already. But what I'd like to do
14 in the next few minutes is to just go over this on the key
15 points in the proposal.
16 The objective of the pediatric PK study design
17 template is the following. First, to provide a consistent
18 approach, like Larry mentioned earlier, to design and
19 evaluate a pediatric PK study. We'd like to develop a
20 computer-aided pediatric study design template which will
21 take the user-supply study design and automatically
22 estimate the study performance and study power. So it 1 185 2 1 will be making it easier for the user to determine which
2 type of study design is appropriate for their drug and for
3 their design.
4 We also want to select case studies from the
5 FDA database to test the template and refine the template.
6 Finally, through this template, we hope to
7 promote a wider use of population design in pediatric
8 studies.
9 I think Larry has mentioned this pediatric
10 study decision tree. He also presented this decision tree
11 in the last meeting. Basically we use this decision tree
12 or propose to use this decision tree to determine what
13 type of study will be necessary to bridge the adult
14 efficacy and safety data to pediatrics. Depending on the
15 answer to many of these questions, we may end up doing or
16 recommend doing a efficacy study or clinical study, a PK
17 study or a PK/PD study, or all of the above, or safety
18 studies.
19 But today I just wanted to focus on this
20 particular box here which is we will recommend doing a PK
21 study as a bridging study between adults and pediatrics.
22 So once that decision has been made, then we 1 186 2 1 will have to use the PK study for dose selection in
2 pediatric populations. So this goes back to the dose
3 adjustment in special populations, the same decision tree
4 we talked about in the morning today.
5 Based on this decision tree for dose
6 selection, we have to answer several key questions.
7 First, we have to ask whether there is a clinically
8 significant difference in pharmacokinetics between adults
9 and pediatrics. Secondly, we have to ask what is the
10 pharmacokinetic parameter in pediatrics, and we had to use
11 that pharmacokinetic parameter to adjust the dose in that
12 population.
13 So based on this decision tree, we had to
14 conduct the pediatric population PK study to get two
15 information. First, we had to identify whether there's a
16 clinically significant difference -- not just any
17 difference, but clinically significant difference --
18 between adults and pediatrics. Secondly, we had to
19 accurately estimate the parameters in pediatric
20 populations without any bias.
21 There are, of course, many factors that may
22 influence study performance, population PK study 1 187 2 1 performance. For example, the study design which may
2 include a number of subjects, demographic information
3 maybe, the number and timing of samples. Also, the study
4 conduct, such as the compliance of the patients, the
5 variability of dosing time and sampling time and in
6 missing dose and in drop-off. Of course, the
7 pharmacokinetics itself and the variability of the
8 pharmacokinetic parameter also influence the study
9 performance or how we design the studies.
10 In the proposal, we also bring up several
11 important points to be considered during designing a
12 population PK study. We believe that when we design a
13 study, we have to take into consideration the objective of
14 the study which I just mentioned to identify the
15 clinically significant difference and to estimate the PK
16 parameter in pediatrics.
17 We also believe that because the
18 pharmacokinetic parameters are very different between
19 drugs, also there are many different varieties of
20 population PK designs that will provide sufficient study
21 performance to achieve the objective that we just
22 mentioned. There is no one-size-fits-all design for 1 188 2 1 population PK. Each design should be looked at on a case-
2 by-case basis, but using a consistent approach which we
3 believe a clinical trial simulation will be a good
4 practice to estimate the study performance and study power.
5 In our proposal, we also bring up several
6 other points to consider. For example, we had to look at
7 a number of factors that may influence study performance,
8 such as dosing time, sampling time. Compliance is an
9 important factor. Of course, the number of subjects, the
10 number of patients, and how the sampling time and dosing
11 time is distributed with time. We also need to consider
12 if there's an unbalanced design in terms of number of
13 subjects as well as the number of samples between
14 different populations.
15 This is a flow chart of the proposed study
16 design template. It consists of a module where the user
17 can input their study design protocol. So this is where
18 the user can enter the number of subjects, the number of
19 patients, when do you take the samples, and what is the
20 variability of pharmacokinetics, and so on and so forth.
21 So this will be your input parameter where the
22 user can put into the module. And the template will take 1 189 2 1 this information and automatically generate or translate
2 the study design template into a simulation code. With
3 that simulation code, the software will generate a study
4 performance indicator. In this case, it will be the power
5 to determine whether there's a clinically significant
6 difference in pharmacokinetics between populations, also
7 what is the accuracy and precision and bias of the
8 parameter estimations.
9 So the input and output of the proposed study
10 design template includes the following. The input will be
11 a study design, the pharmacokinetics in adults, but also
12 the variability of the demographic or patient population
13 you will include in the studies, and also the study design
14 variables. The other input parameter will be the criteria
15 for evaluating the study performance.
16 But the output from the template will be the
17 estimated study performance which is related to the two
18 objectives of the population PK study. One is the power
19 to identify -- I want to emphasize -- the clinically
20 significant difference, and secondly, the precision and
21 bias of parameter estimations.
22 The clinical trial simulation that we propose 1 190 2 1 to be used in the study design template is pretty
2 standard. It includes the following steps. First, it
3 will generate demographic variables and pharmacokinetic
4 parameters based on the user input information. It will
5 simulate a study design as well as the study conduct, and
6 it will generate population PK data. Once the population
7 PK data is generated or simulated, then we will conduct a
8 population analysis based on the simulated data. And
9 finally, we will repeat the process perhaps a few hundred
10 times to a few thousand times to estimate the power of the
11 study as well as the precision and accuracy of the
12 parameter estimations.
13 Like I mentioned earlier, there are two main
14 objectives to measure the study performance. The first
15 one is to identify a clinically significant difference in
16 pharmacokinetics. Based on the decision tree or dose
17 selection that we had presented in the morning and early
18 in my slides, the first option is to just look at a 90
19 percent confidence interval and see whether the 90 percent
20 confidence interval of a PK parameter is within a default
21 boundary.
22 But the second option which we might prefer is 1 191 2 1 to determine whether a clinically significant difference
2 in pharmacokinetics will exist. In order to determine the
3 clinically significant difference in pharmacokinetics, of
4 course, we had to know the PK/PD relationship and we have
5 to assume that the PK/PD relationship that we have perhaps
6 from the adult population is similar to those in the
7 pediatric population.
8 For example, in the simulation we can assume
9 that there's X percentage which is considered a clinically
10 significant difference in the body weight normalized
11 clearance. With that difference, we want to ask the
12 question whether the population PK study that we tried to
13 design will be able to capture this clinically significant
14 difference. So with that we can calculate the study power
15 of the population PK studies.
16 Now, the second criteria for study performance
17 we propose is the precision and bias of PK parameter
18 estimations. Precision and bias can be presented in terms
19 of a percentage prediction average. Precision will be the
20 standard deviation of the prediction errors and bias will
21 be the average of the prediction error. The prediction
22 error is defined as the percentage difference between the 1 192 2 1 true value and the predicted value because we're doing a
2 clinical trial simulation, so we know exactly what the
3 true value is.
4 So the output again are two information to
5 measure the performance of the study. One is whether the
6 study has a power to identify a clinically significant
7 difference. The second is precision and accuracy of the
8 PK parameters.
9 So this is basically a general description of
10 the proposal. Again, the detail is elaborated in the
11 actual proposal itself.
12 I guess we'd like to ask two questions to the
13 committee. The first question is, are the proposed
14 objectives for pediatric PPK studies reasonable,
15 considering the decision tree for the dose adjustment? So
16 we have mentioned two objectives for the population PK
17 studies. We also talk about criteria for study design
18 performance.
19 The second question is, is the proposed
20 pediatric PPK study design template reasonable? This is
21 related to the clinical trial simulation approach that we
22 propose, as well as the factor to be considered, a study 1 193 2 1 design factor.
2 We also would like to ask what feature we
3 should include in the pediatric study design template so
4 that we can make it more user friendly and more useful for
5 both the clinician and the reviewer who might use it to
6 design pediatric population studies.
7 That's it.
8 DR. VENITZ: Thank you, Peter.
9 You have the questions for the committee. Any
10 comments by the committee? Hartmut.
11 DR. DERENDORF: I have a question for
12 clarification. On your decision tree, right on top you
13 said it's reasonable to assume similar response to
14 intervention. Then you have yes, and reasonable to assume
15 similar concentration response in pediatrics and adults.
16 What's the difference between the two?
17 DR. LEE: I think one example was asthma. For
18 example, it may be a different endpoint that can be
19 measured in a pediatric population than an adult
20 population. That will be the first block of questions.
21 So if we can answer that question yes, then we go to the
22 next block. But if the answer is no, then we have to go 1 194 2 1 to the clinical studies. So if the answer is no to the
2 first two questions in the top block, then basically
3 either the clinical endpoints are different between the
4 two populations or the disease is totally different or
5 disease progression is totally different in the two
6 populations.
7 Now, the second question on the right-hand
8 side of the block is that once we decide that through our
9 clinical opinion the disease and disease progression and
10 endpoints are similar, then we ask in our database do we
11 know there is a PK/PD relationship and can we assume that
12 a PK/PD relationship is similar between pediatrics and
13 adults. For example, for a proton pump inhibitor, we know
14 the PK/PD relationship of gastric acid versus drug
15 concentration.
16 Now, can we assume that the relationship we
17 have seen in the adult population is the same as the
18 relationship we will see in the pediatric? If the answer
19 is yes, then we can just rely on PK information to
20 determine the dose selection in the pediatric population.
21 DR. JUSKO: A comment and a couple of
22 questions. This, in general, seems like an excellent 1 195 2 1 idea. Of course, we always want to utilize information we
2 know ahead of time to anticipate the study design and
3 changes that will occur in a new group for study.
4 One small point that seems to be totally
5 missing is dosage form and bioavailability. Young
6 children are typically getting liquids and chewable
7 tablets, and in order to anticipate what will happen in
8 the kinetics and dynamics in young children, you may need
9 a comparable database in adults with a similar dosage form
10 or at least make an adjustment for perhaps faster
11 dissolution or absorption. So at some point some reminder
12 of that question may need to be added.
13 Then there's a certain vagueness when you talk
14 about dose adjustment or dose selection because in
15 pediatrics there's always dosage adjustments. Children
16 are seldom given the same 500 milligram tablet that adults
17 are. So something needs to be said about what do you mean
18 by dose adjustment. Are they getting certain milligrams
19 per kilo already or certain dosage sizes depending on
20 weight ranges? There's a great deal of flexibility
21 already inherent in selecting dosages in children.
22 DR. LEE: I think the answer to the first 1 196 2 1 question is I think we are talking about dose selection
2 because we know that we're going to adjust the dose
3 anyway. We're going to give a different dose. Normally
4 the dose is given on a per body weight basis or sometimes
5 we will look at the body surface area.
6 DR. LESKO: It depends on the age group.
7 Sometimes the dose is adjusted based on average exposure
8 to the drug, and depending on the age, it may be down to a
9 milligram per kilogram basis, something like that.
10 But I was thinking of the first question on
11 the response that Hartmut asked and I don't know if Peter
12 clarified. But I was thinking basically you're asking the
13 question is the response the same that you'd be interested
14 in in kids as you would be in adults. That's sort of like
15 a two-part question.
16 So the first question is, is the response,
17 let's say, FEV-1 in an asthmatic patient the appropriate
18 response? And that sort of gets to the heart of the
19 mechanism of action of the drug and the progression of the
20 disease similarity. Often data to support those
21 assumptions or answers is not available, but if you do
22 agree that, yes, I believe that's true, then the question 1 197 2 1 is, is there a concentration related to that response that
2 you previously agreed is similar to the adult? Then that
3 puts you further down in the decision tree.
4 DR. CAPPARELLI: I just had a question and a
5 comment as well. In regards to the scope of this, is this
6 looked at as part of a safety study as well, or is this
7 really structuring a population PK study with the only
8 endpoints being population PK? Because I think one of
9 things that's missing in the objectives is getting at the
10 question that we've been asking how do we assess these
11 potential age-related exposure-response relationships.
12 And the design that may be very robust for estimating the
13 PK parameters of the population may not give you all of
14 the estimations in the individuals that you may want to do
15 some of that exploratory analysis. So I think that it
16 needs to be very clear along those lines.
17 It was brought up before, and I spoke with Dr.
18 Lee before a little bit about the concept that, really,
19 kids are different. The question is whether or not we can
20 predict those differences a priori. So saying are there
21 differences, there are always differences. The question
22 really is that based on our knowledge of modeling, our 1 198 2 1 knowledge of pathophysiologic changes, developmental
2 changes, can we predict those well and then go from there.
3 I also just wanted to amplify Dr. Jusko's
4 comments about the dosage formulations. A lot of the
5 exposure issues are going to be based on what size doses
6 you have. So when you say clinically significant
7 differences in PK, it's really what sort of exposures
8 we're going to get out of the available dosage forms.
9 There may be a modest change in PK, but because of where
10 you're left with your cut-points, you may end up having
11 big changes in dosage exposures which again it needs to
12 get back to, I think, what we're interested in and at
13 least getting the exposures as comparable when we don't
14 have the information in terms of differences in exposure-
15 response relationships.
16 DR. RELLING: Is it implicit that it's only
17 worthwhile to do these pharmacokinetic studies in children
18 if there aren't good a posteriori methods of dose
19 adjustment based on more readily available clinical
20 measures like blood pressure, like immediate response to
21 anesthesia, like pain relief? Is it any part of your
22 interest in the pediatric rule that pharmacokinetic 1 199 2 1 studies be performed for drugs for which there's a narrow
2 therapeutic range or small therapeutic index and there's
3 no other good way to adjust doses?
4 DR. LESKO: No. I think the pharmacokinetic
5 studies are routine in these types of situations. I don't
6 think dose adjustments -- I'm trying to think if I have
7 any experience with dose adjustments being made without
8 the availability of PK information, for example, being
9 based on observed responses in kids relative to adults.
10 Is that sort of the question?
11 DR. RELLING: My point is that I feel that
12 there are a lot of studies being done by pharmaceutical
13 companies, which they claim are being done to fulfill FDA
14 regulations or requirements or suggestions, which are done
15 for medicines that don't need to have pharmacokinetic
16 studies done. They're done for anesthetics that could be
17 easily titrated based on the response of the patient to
18 the oxygen saturation. They're done for narcotics for
19 which the drug is going to be adjusted based on pain
20 response. And I feel like a lot of resources are being
21 expended on these studies for unclear reasons. So I'm
22 trying to figure out is this a suggestion that's being 1 200 2 1 made for all medications that would ever be used in
2 children regardless of the therapeutic index and the
3 ability to adjust doses based on other parameters besides
4 PK.
5 DR. LESKO: Yes. I'm not sure "all" is the
6 right word, but I'm thinking of the implications in
7 labeling the drug product with a starting dose in
8 pediatrics. You need to have some information, it would
9 seem to me, to begin dosing, and that generally has been
10 the pharmacokinetic studies to recommend some changes in
11 that initial dosing strategy.
12 This isn't to say this is the only objective
13 measure that one looks at in the pediatric area. There
14 are always other studies, in particular safety studies
15 and, in many cases, small efficacy studies as well.
16 DR. KEARNS: I want to thank Dr. Relling for
17 her insightful question because everybody wants to know
18 the same thing about children and that is how much do I
19 give, do I need to give a different amount as the child
20 gets older, and will it work like I want it to. That's
21 really crux of all this. I won't belabor all my soapbox
22 points about this issue, which are many, but I want to 1 201 2 1 make two points about this recommended approach.
2 First is the issue that Dr. Sheiner said we
3 should grapple with at a later date, but it's at the top
4 of the decision tree, and that is, do we believe that
5 whatever condition occurs in a child it is substantially
6 similar? And that's the language in the law,
7 "substantially similar" to what it is in an adult.
8 For the last four or five years, I have seen
9 people at all levels of the agency grapple with this as
10 though it were a very large, mean animal, and at the end
11 of the day, people rather than slay it, seem to run from
12 it and invent ways to try to avoid it. I think that's
13 tragic because what that has produced is an incredible
14 consumption of resources, not to mention the exposure of
15 children to clinical trials that is a needless exposure.
16 That doesn't get talked about enough, and I'll stop
17 talking about it now because that's probably another whole
18 day.
19 Now, with respect to "substantially similar,"
20 what comes to my mind is something very similar. I'm
21 starting to sound like one of our politicians who uses the
22 same word over and over. There's an article in CP&T, Art 1 202 2 1 Atkinson. It's near the front of the journal this last
2 month on biomarkers. It talks about the goodness or
3 badness of biomarkers on how far away they fall from the
4 trunk of the tree of drug effect. The same thing could be
5 looked at with respect to this issue in children. Drug
6 action is obviously something we can, at times, determine
7 whether it's similar. If we can't talk about action, we
8 can talk about drug effect, a physiologic response in an
9 association with a drug dose or a concentration or an
10 amount. If we can't talk about that, we can then talk
11 about disease response and lastly disease progression.
12 Sadly, I hear people put disease progression
13 at the top of the discussion list as they look at that box
14 because if we were to get a bunch of pediatricians in a
15 room and ask them if they agreed that the progression of
16 GERD was the same in adults and children, they would never
17 agree. They would never agree about asthma. They would
18 never agree about leukemias, other malignancies. Pick a
19 disease. No one will agree at the end of the day. Which
20 means then you punt the ball, and the ball is punted in
21 terms of time, dollars, delay, and for all the reasons
22 that people line up on the opposite side of the pediatric 1 203 2 1 argument to say it's bad things to do, it gives them fuel
2 for their fire.
3 So I think it is incumbent upon the agency and
4 those of us who you've elected to advise you to, at some
5 point in time, grapple with what is substantially similar
6 so that any well-designed pharmacokinetic approach can get
7 on the right track and do what it's intended to do. So I
8 applaud Dr. Sheiner for suggesting that and hope we can
9 talk about it.
10 With regard to population pharmacokinetics
11 which is, Peter, I think central to your presentation, my
12 question is always the same, and that is, when we use a
13 pop PK approach in a pediatric study, are we aiming to
14 explore relationships with age or, alternatively, are we
15 aiming to define them? I think we certainly can use pop
16 PK, appropriately designed, to explore them.
17 But keep in mind that for those drugs where
18 age is an important covariate with respect to metabolism
19 or perhaps pharmacodynamics or response, an exploration
20 and a definition can be very, very different with respect
21 to impact because at the end of any pediatric program
22 that's conducted, we are trying to do things on the quick, 1 204 2 1 on the cheap, and on the small. That means the
2 generalization that we want for an entire population of
3 patients that represents about 15 percent of the
4 population of the United States is predicated on a
5 fraction of the numbers of subjects and things that we
6 would do in adults.
7 So what you're proposing is very important,
8 has incredible potential if done correctly, but we've got
9 to be mindful of knowing how that tree starts and making
10 sure that at the end of the day the people at level of the
11 review divisions and the Office of Pediatric Therapeutics
12 understand the power of this tool and how it can help them
13 as opposed to what's going on now in the area of PPIs -- I
14 hate to harp on this, but it a plays a nice tune -- where
15 all the things that people around this table know, if you
16 go into the little, bitty room downstairs on the third
17 floor and you listen to the recommendations to a sponsor,
18 you would have thought you woke up in the stone age. The
19 magic is still very much there, and this approach has to
20 be used to make the process better.
21 DR. SHEINER: I'm glad we're moving off in a
22 direction away from the techno-nerd thing. 1 205 2 1 Mary, at the risk of maybe setting you up as a
2 straw man, let me just think about what you just said.
3 I'm going to take away the word "pediatrics."
4 Why wouldn't everything you said apply to a
5 drug that's easily titratable no matter who it's intended
6 for?
7 Now, I think Larry's response to that was,
8 well, we need a dose in the label somewhere. It's got to
9 be based on some kind of evidence that people can refer to.
10 So I think, if I can rephrase your question,
11 what I think you're saying, to put it in sort of a
12 Bayesian context, is there's something about having
13 studied dose-effect relationship in adults that for a drug
14 that may be isn't too toxic and is easily titratable with
15 respect to effect might mean we don't have to do any more
16 than that for children, which means, in a sense, that
17 there's a prior somehow on the doses that you ought to be
18 trying in children because, after all, the implication of
19 an easily titratable drug that's not very toxic is that
20 you don't have to get the first dose very right. So
21 you're saying I'll tolerate a much wider range. And
22 there's something about having studied it in adults that 1 206 2 1 gets me close enough to that range that if I were to work
2 it all out with a utility function and put a prior on what
3 I think I can extrapolate from adults to children, I'd
4 find I don't have to do a study at all. I'd find my net
5 benefit to society would be better served by going ahead
6 and approving it for children than doing any more studies.
7 I think that's how I'd try to put it in a quantitative
8 context.
9 All I'd like to say about that is we could do
10 that. We could put it in a quantitative context so you
11 don't have to sit there and feel a little uneasy about
12 what other folks are suggesting and Larry doesn't have to
13 feel a little uneasy about what you're suggesting. We can
14 put it all together and, just as we were talking about
15 earlier with Mats' presentation, take a look at it. What
16 values do we need to place on doing -- what negative
17 utility do we need to place on doing studies in children
18 in order to make it worthwhile to extrapolate, given our
19 sense of uncertainty from adults to children, and
20 conditional on this drug, its safety profile in adults, et
21 cetera?
22 What I'm trying to say is I think that the 1 207 2 1 same way of trying to be quantitative about this could
2 answer that question. I think it's just like the drug
3 companies ask themselves now. Am I better off going right
4 to phase III and skipping phase II? And there's a
5 decision analytic argument that might say that that's
6 really, in terms of net present value, et cetera, a good
7 idea for this drug in this circumstance given what we know
8 about it and its competitors and the fact that it's quite
9 similar chemically and so on. And all that can be worked
10 out and it doesn't have to be an opposition of people not
11 really understanding each other or thinking that they're
12 coming from different places. They're not. They may
13 value some things differently, but I doubt very much.
14 And that's a worthwhile exercise, it seems to
15 me, to do as we think about going to this in the future to
16 try to approach this whole issue from the way in which
17 we're saying people might approach the much simpler and
18 therefore easier metaphor of a dosing decision, this whole
19 issue of do we do a study at all, because it's really all
20 the same. It will get us into the habit of thinking that
21 way in a place here where we've got some time to think
22 rather than having to act right away. 1 208 2 1 Peter, I wanted to get to your question. It's
2 really I guess my question about your question. It's
3 really a question for Larry, and it's this one.
4 You called it a template and maybe words are
5 important here. But it sounds to me an awful lot like a
6 piece of software. At the sort of highest level, does the
7 FDA want to be writing software for people who are then
8 going to probably feel obliged to use that software
9 because the agency says this is what you should do to
10 design the study that you're going to do to come to us
11 with? I mean, the agency I think has generally been
12 pretty careful about that and has had best practices and
13 guidances and suggestions and all that kind of stuff. But
14 here's something we made and it's for you. That's a whole
15 new line, isn't it? And do we want to go down that line?
16 DR. LESKO: Yes. I think you're right in
17 pointing that out. I don't think we're talking about, at
18 this point, the software that we're either going to
19 develop and advocate, advance for drug development. I
20 think what we're talking about is a template that is based
21 upon software that a reviewer might use in conjunction
22 with a discussion with a sponsor that would prompt for the 1 209 2 1 critical information that would go into making a robust
2 study. It's intended to sort of be a starting point for
3 discussion or designing such a study that would be, at the
4 end of the day, generally acceptable to the people that
5 need to accept it in terms of its review and utility.
6 I think the problem now is we don't
7 necessarily see a consistency in advocating the design of
8 these studies across different opportunities to do so.
9 This represents a way of channeling the discussion into
10 the critical areas that would lead to usable results.
11 DR. SHEINER: I like it. It's a good
12 metaphor, but I think you may be getting too specific too
13 fast. It's sort of like a guidance, a statement about
14 what are good things, what you want to see, what kind of
15 principles you apply, and it's got a lot of wiggle room in
16 it. The thing about software is it hasn't got any wiggle
17 room at all. You put the inputs in and it's
18 deterministic; out comes the answer. So you have got to
19 be pretty sure that's the algorithm you like. So I'm not
20 sure I would start there.
21 But taking the metaphor of designing software
22 to say that -- and that gets us to focus on all the key 1 210 2 1 issues. That's, I think, a good idea so long as at least
2 you're contemplating maybe not going the whole route and
3 going into the software business.
4 Then I think fundamentally the questions are
5 right, the first one being what's the minimum evidence I
6 can gather that I ought to bother anymore. Is there any
7 difference? But I don't think you'll get that from the
8 same study. As I say, I think this is essentially
9 sequential. That's the difference that I'd have between
10 the way you put it. It's not going to be one study.
11 There's something you do to figure out whether I need to
12 go any further, and that may be quite different in design
13 -- although I haven't thought this all out -- than what
14 you do when you say, oh, I guess I better go further. I
15 better pin down the key PK parameters in this population
16 and how it varies with disease state and other things. My
17 suspicion is that those will be two different activities.
18 DR. LESKO: I think the problem with the
19 studies that have been done -- and it's probably why we
20 haven't seen very many of this sort -- is the
21 believability of the outcomes because these studies are
22 not as well understood as obviously a sample-rich study 1 211 2 1 design and the issues that go into analyzing a sparse-
2 sample study using NONLIN or something like that. Is the
3 information reliable and how do I know that? It's having
4 to explain that to people who have to make decisions over
5 and over or having designs that would lead to an
6 acceptable result is sort of where we're heading here.
7 That being said, if we have an optimally
8 designed study to get at the questions we're asking about
9 differences in pharmacokinetics for the purposes of dosing
10 changes, can this method be confirmatory enough to stop
11 there? I think Greg used the word "exploratory." I don't
12 know if that was a suggestion that these studies at best
13 can be exploratory for the purposes of designing another
14 study or would they be confirmatory enough to say I know
15 what the difference in clearance is between this drug as a
16 function of age and maybe more age groups if I can do a
17 sparse-sample approach, and thus I can recommend some
18 different dosing for these age groups based on the study
19 design without having to do a full study which in some
20 cases limits the number of age groups that can be looked
21 at.
22 DR. KEARNS: But, Larry, I think there are 1 212 2 1 some examples on the books where it does work. The whole
2 program on montelukast to me has been an incredible
3 success story because a very careful pop PK approach was
4 taken. We went down through all the pediatric populations
5 now down to 6 months. From my opinion, the appropriate
6 variability was considered and the parameter estimates
7 seemed to hang together, and when the data were taken the
8 next step into showing proof of concept with respect to
9 effect, the effect was there. Consequently, the labeling
10 of the drug has been changed multiple times. It likely
11 will continue to be changed based on that approach. It
12 worked. It was done right. There was a need to do it so
13 we know the dose.
14 But you're correct in that many other
15 companies have not followed suit, so to speak, and for
16 reasons that I don't completely understand.
17 On the other side of the issue too,
18 logistically -- and certainly Dr. Capparelli knows this
19 because he's kind of in the business -- for the most part,
20 if you have a pediatric study, a PK study, and you go to
21 the trouble of obtaining repeated blood samples, you're
22 obtaining them through a catheter. If you're analytical 1 213 2 1 method is such that you don't need a lot of blood, the
2 bother in getting eight samples is no more than the bother
3 in getting three or two. IRBs anymore, at least pediatric
4 ones, do not allow you to stick children several times.
5 So there are many times when a pop PK approach could,
6 indeed, be used and it would be perfectly valuable and
7 valid. But a traditional approach is very achievable,
8 more so than most people think.
9 DR. DERENDORF: Just as a follow-up, the other
10 technique that is coming on strong is microdialysis where
11 you can take as many samples as you want without taking
12 any blood out. So I think that will change the ability of
13 doing studies in children.
14 DR. SHEINER: Just a quick response, Larry, to
15 your question. The essence of having a credible
16 confirmatory analysis is controlling type I error, and
17 controlling type I error involves essentially saying what
18 you're going to do before you do it because you can't have
19 feedback from the data to the analysis. That doesn't mean
20 you can't get valid conclusions from doing that, but you
21 can't control type I error if you do that. So I think
22 that's the only issue. 1 214 2 1 If you do a well-done analysis, then you know
2 how uncertain you are when you're done. That's an issue
3 of design. That is to say, given the assumptions we're
4 willing to make and the data that we get, the sparse data
5 that we gathered, do we wind up with sufficient precision
6 on these things to make the kinds of statements we want to
7 make? That's an issue that unfortunately there isn't a
8 lot of theory for because they're complicated designs and
9 they're complicated analyses, and so you have to do it
10 through simulation.
11 But the key thing would be specifying
12 beforehand exactly what models you're going to use, what
13 procedures you're going to do, et cetera. All of us in
14 the business of doing extensive modeling have always felt
15 a little anxious about that because you can't take away
16 from me my ability to look at the data and decide what
17 model I ought to use, but you have to take that away from
18 me if you want to control type I error.
19 So I think there will be an interesting issue
20 there of how you balance that and whether, in fact,
21 controlling type I error is as important as you sort of
22 said it is by bringing it up. I don't know. I think I'm 1 215 2 1 probably with Mary and Greg on this one, that we've got a
2 lot of priors behind us and I don't think I need to pin it
3 down to a fare-thee-well.
4 DR. LESKO: One of the inputs into the model
5 is the variability within the kids or within the age
6 groups that are being studied, and I'm not sure how that's
7 been handled. I can't recall the montelukast or, Ed, some
8 of the studies you've done. But the variability
9 associated with the -- what you would expect with your
10 different age groups -- how was that generated and how
11 important is that in terms of designing these studies?
12 DR. CAPPARELLI: Well, in terms of looking at
13 simulations and sort of real data that come out, the
14 variability goes up as you go down the age group. So from
15 a pragmatic standpoint, one starts with at least an adult
16 value and goes up from there. But clearly there may be
17 thresholds, some of them drug-specific and age-specific,
18 in terms of dealing in HIV where we've got drugs that have
19 major food effects, we've got formulation effects, and as
20 soon as you cross certain thresholds, the variability is
21 going to drop down. But it adds complexity both on the
22 design standpoint and the analysis standpoint when you 1 216 2 1 have observed doses where you've got your compliance and
2 you see no drug. And this is in a CRC setting, but it's
3 just the way that it behaves in this population.
4 So there clearly are needs for evaluating
5 distributions much more intensively, especially if we're
6 interested in sort of the outlying regions, which I think
7 most of us are. But experience is that it's greater.
8 It's just variable how much more.
9 DR. SHEINER: I've got to ask about that. I
10 always thought the opposite. Once you line up kids by
11 some maturational marker, whether it's gestational age or
12 whatever it is, I thought they're all newly minted coins
13 and they all look the same. In fact, I think I remember
14 Bill Jusko saying that when you get real old, the
15 variability goes up because you've run a longer race and
16 you're sort of stretched out there at the finish line.
17 That's what I always thought the case was. There was more
18 variability in old folks than there was in little babies,
19 again lining them at the right maturational level.
20 Obviously, a 3-month premature is not a term baby.
21 DR. KEARNS: No. Actually we're finding the
22 opposite. It's quite interesting because if you look at 1 217 2 1 what Dr. Sheiner said, if you look at a 3A4 substrate in a
2 healthy adult, there's 20-30-fold variability in the
3 processing. You look at it in a 3-year-old. There it's
4 about the same. If you look at it in the 3-month-old,
5 it's about the same.
6 The problem is that as that little beast
7 travels into adolescence and adulthood, the shape of the
8 acquisition curve, if you will, changes, not to mention
9 changes in body composition which are quite evident. So
10 there are a couple of moving targets that make it a
11 particular challenge which, in designing a pop PK study,
12 trying to estimate what your real variability is when
13 you're up at the front, is not always an easy cookie to
14 get.
15 But there's got to be a way to do it. I think
16 what we hopefully will see, as we see the database, is
17 some of the information that the agency is collecting is
18 beginning to show us where these patterns might be, if you
19 will, or breakpoints might be, and that makes things a bit
20 easier to deal with.
21 DR. DERENDORF: And if enzymes change like
22 that, what makes us assume that receptors don't? 1 218 2 1 DR. KEARNS: Absolutely, right.
2 DR. FLOCKHART: The absence of a phenotypic
3 probe for the receptor.
4 (Laughter.)
5 DR. LESKO: We'll talk about genetic solutions
6 tomorrow.
7 DR. CAPPARELLI: I would also emphasize at
8 least a lot of the variability experience where there are
9 these major changes, besides the newborn, is when you get
10 into oral drugs. So I think there are, at least from my
11 experience -- again, the diet is different. Controlling
12 for when they take it relative to food, all those things
13 that may be a little bit easier to do in adults is much
14 more difficult in kids. The formulations themselves --
15 while you can do bioavailability studies in adults and
16 show similar formulations, it doesn't always extrapolate
17 to kids. So you have those sorts of things, I think,
18 contributing as well to the variabilities.
19 DR. VENITZ: Any other questions or comments?
20 (No response.)
21 DR. VENITZ: Then let's take our afternoon
22 break. It's now 2:35. So let's reconvene at 3:05, in 30 1 219 2 1 minutes. Thank you.
2 (Recess.)
3 DR. VENITZ: Let's reconvene our meeting,
4 please.
5 Our next and our last presentation for today
6 on the pediatric topic is Dr. Gene Williams. He's a
7 pharmacometrics reviewer currently on detail at the Office
8 of Clinical Pharmacology and Biopharmaceutics, and he's
9 going to give us an update on the pediatric database.
10 Gene.
11 DR. WILLIAMS: Thank you, Jurgen.
12 This is the title of my talk, kind of a long
13 title. The notion is to become better at predicting peds
14 clearance and to take advantage of what we usually know at
15 the time that we see peds submissions or proposals for
16 peds studies, that is, child age, a lot of information in
17 adults, and the knowledge of in vitro metabolism.
18 I'm going to ask four questions of the
19 committee. It makes sense to show them first so you know
20 where I'm headed.
21 The first one is sort of an overall scope and
22 method. That is, is the general approach that I'm going 1 220 2 1 to suggest in the presentation rational and logical? And
2 the approach proceeds from a very empiric method to a more
3 mechanistic method.
4 Secondly, is there anything special that you
5 think that I should be aware of, some difficulties that
6 you think I'm likely to encounter, and if you can identify
7 such, how can I avoid them?
8 The third question is of particular interest.
9 Are there data sources you could recommend? One committee
10 member has already been referred to as "in the business."
11 We'll get there I guess.
12 The fourth question I have is, do you have any
13 suggestions regarding the form of the non-physiologic-
14 based PK mechanistic models? That will become clear as I
15 proceed, I hope.
16 What brought on this project and what exactly
17 are we talking about trying to accomplish? What we'd like
18 to be able to do is construct a model that allows us to
19 predict pediatric systemic drug clearance from, as I said,
20 adult PK and in vitro microsomal metabolism data. That
21 would be a short-term goal. Obviously, we have a longer
22 vision. It seems like if you could construct such a 1 221 2 1 model, it would aid us internally. It would also be of
2 potential interest to industry scientists, and finally
3 perhaps even health professionals in the community could
4 make use of such a model.
5 It's probably appropriate to begin talking
6 about the data that we have because that largely drives
7 how we'll be able to model.
8 The most fundamental data unit we're talking
9 about using here is clearance, whether it would be from
10 sparse or dense data, and age for each individual in the
11 data set. A number of demographic data also we would use;
12 that is, the weight and height for each individual, renal
13 function for each individual, and gender and race for each
14 individual. Finally, as I've alluded to earlier, we would
15 also want to make use of what we know about in vitro
16 metabolism data for each drug.
17 I should probably add here that, as many of
18 you I believe have appreciated, FDA is well positioned to
19 do this sort of work because the data that we have often
20 is very specific. We not only see data summaries, but we
21 also see individual data, which is a limitation that if
22 you use literature data, you face, but we often don't 1 222 2 1 face. We usually get fairly raw data where we do have all
2 these values.
3 Our data set. I've taken this statistic from
4 a website that I've included here. I believe it's
5 publicly available. I don't think it's just on our
6 intranet. In mid-March, about a month ago, we had 72
7 active moieties that had received pediatric exclusivity.
8 As Larry said earlier, most if not all of those would have
9 pediatric data available.
10 As I've gotten ahead of myself a little bit,
11 the data that we see is usually raw. It's actual
12 measurements of individuals, not summaries across
13 individuals. And for the models that we want to explore,
14 that's of a lot of utility.
15 Further, our data is usually reviewable to an
16 extent that literature data is not. The analytical
17 methods, dropout, salient features of data accumulation
18 and choices made in data analysis are often presented to
19 us. So we can do a good job of assuring data quality.
20 However, there are some limitations of the
21 data we see. First, studies are often not powered to
22 compare PK across age groups. People are submitting data 1 223 2 1 to us for regulatory purposes, not always to discern
2 carefully small age effects. That's in distinction to
3 studies sometimes performed by academicians where they're
4 specifically trying to see age effects or, I should say,
5 reasonably small age effects.
6 The ages with the greatest difference from
7 adults, often the very young, are often most poorly
8 represented in the data sets that we see. The data sets
9 we see are motivated by the desire to treat, not
10 necessarily the desire to see an age effect.
11 Finally, most of the drugs that we see are not
12 probe substrates. People, again, are not asking
13 mechanistic questions. They're trying to get a drug
14 approved. So the ability to tease out effects may be more
15 difficult since we don't have good markers for each
16 individual effect that may be present. As a result of
17 this, it may be necessary to use some function of in vitro
18 metabolism such as the Km, that is, for an enzyme, as a
19 covariate when we do our analysis.
20 I'm now going to carefully consider a data set
21 that I took from the literature. I did this for a number
22 of reasons. As Larry said earlier, organizing our data 1 224 2 1 set is a considerable effort, and since this data set was
2 sitting out there, I thought I'd use it not because we
3 want to analyze it, but because it makes a good platform
4 for discussing the methods that we intend to use.
5 This is taken from Ginsberg, et al. There are
6 somewhere 21 and 27 drugs represented here. The y axis is
7 children's clearance relative to adult. You'll see at the
8 bottom of the slide I've described the units that are
9 used. These data are standardized for weight -- they
10 looked at kilograms -- and age. Age here is not a
11 continuous variable. Rather, it was grouped
12 categorically, a decision the authors made. They took
13 these data from the literature. It's not their own data,
14 and I guess the data lent themselves or, for some reason,
15 they organized for categories in this way.
16 I've shown a line at 1. That would be where
17 the child is exactly the same as adult. You can see at
18 ages 12 to 18, that's accomplished.
19 I don't want to give much attention to this
20 slide, but the question is likely going to arise as to
21 what sorts of drugs they were. This is also taken from
22 their database. This database is available on line for 1 225 2 1 anyone who wanted to explore it. As I said, I don't want
2 to discuss this, but the drugs represented a number of
3 classes.
4 Before you attempt to model these sorts of
5 data, it's necessarily to normalize clearance. The reason
6 why is you want to consider each drug on its own and not
7 have your analysis complicated when you compare drugs
8 whose adult clearances differ widely. So the method we
9 chose to normalize clearance, similar to the method that
10 Ginsberg used, is to divide each individual pediatric
11 clearance by the mean adult clearance.
12 Again, this is Ginsberg's data. The y axis is
13 clearance ratio versus age. However, unlike the plot that
14 I showed you from their paper, this data has had the
15 element of weight removed. So this is no longer adjusted
16 according to the representative body weight of the data.
17 The line shown here is a simple least squares
18 fit, no weighting has been performed. This is unlike what
19 we intend to do when we analyze our database. We'll
20 probably use NONMEN extended least squares.
21 As you can see, or perhaps as I need tell you,
22 I have fit the effect of weight on clearance in this plot. 1 226 2 1 So the equation is shown beneath the line and it's a
2 simple exponential relationship. The maximum ratio I
3 allowed to happen was 1; that is, where the ratio of child
4 to adult would be 1. So essentially this is one
5 parameter. I fixed the maximum ratio.
6 As you can see -- I was somewhat surprised --
7 it provides a reasonably good fit. This is a little at
8 early ages, and this is sort of consistent with what we
9 generally expect to happen, that during development and
10 maturation, things may be a little different.
11 DR. SHEINER: Excuse me. Just to clarify. I
12 guess I'm not sure what it is. You haven't fit the data
13 on the y axis to the data on the x axis. Your equation
14 there is in weight which is --
15 DR. WILLIAMS: Correct, yes.
16 DR. SHEINER: So tell me again what I'm
17 looking at.
18 DR. WILLIAMS: Indeed. I have not fit age
19 here. I fit weight. So what I did is, although I'm
20 representing it age because that's the thrust of our
21 interest, before I went there, I wanted to isolate the
22 effect of age as opposed to the effect of weight. So 1 227 2 1 first, I fit the effect of weight, and in the next slide I
2 will then add in the effect of age. I should have
3 clarified. Thank you.
4 DR. SHEINER: So the brown line that I'm
5 looking at there is the equation that you wrote in the
6 lower right-hand corner.
7 DR. WILLIAMS: Correct.
8 DR. SHEINER: And the way you know where to
9 plot it on the age axis is what?
10 DR. WILLIAMS: By converting each age to a
11 weight based upon standard CDC pediatric tables.
12 DR. SHEINER: Okay, but the fit was actually
13 to the blue points where you knew what those weights were.
14 DR. WILLIAMS: I did not know what those
15 weights were. I had to go by the age. So if we back up a
16 little bit, these are the ages I had, but I have summary
17 data. I don't have individual data. So what I did is for
18 each bar I took the mean age. Then I went to the CDC
19 tables to get the weight --
20 DR. SHEINER: Transformed it to a weight.
21 DR. WILLIAMS: Exactly.
22 DR. SHEINER: So if we go back to that 1 228 2 1 picture, we're really looking at a transformation -- a fit
2 of the blue points on the y direction to a to a
3 transformation, defined by these tables, of the data on
4 the x axis.
5 DR. WILLIAMS: Indeed. It would have been
6 more straightforward to plot weight on the x axis, but the
7 reason why I didn't do that is twofold. One is for
8 continuity with the next example where I'm going to fit
9 weight and age.
10 DR. SHEINER: Where you're going to have both
11 pieces of data.
12 DR. WILLIAMS: Exactly.
13 DR. SHEINER: Okay.
14 DR. WILLIAMS: Is that clear to everyone?
15 DR. JUSKO: From the previous graph, that
16 should start at .5, at the age near 0.
17 DR. WILLIAMS: We have birth, which is --
18 DR. JUSKO: The ratio at birth on that graph
19 is around .5.
20 DR. WILLIAMS: Correct. But the y axis here
21 is different because these are weight-adjusted. The y
22 axis here is child clearance divided by kilograms, 1 229 2 1 quantity divided by adult clearance divided by kilograms.
2 DR. SHEINER: You're going to regret ever
3 having shown that picture.
4 (Laughter.)
5 DR. WILLIAMS: So what I've done is I've taken
6 out the kilograms so I could fit weight.
7 DR. RELLING: What is your goal?
8 (Laughter.)
9 DR. RELLING: Why would you do that?
10 DR. WILLIAMS: The reason why I chose to do it
11 this way is because you want to independently describe
12 weight effects and age effects. You expect there to be
13 weight effects, and you also perhaps expect there to be
14 age effects. But you want to be able to independently
15 address are there age effects that are not simply a
16 consequence of weight.
17 DR. RELLING: Okay. Let's see what you have.
18 DR. WILLIAMS: I won't suffer from this
19 difficulty when I have the FDA data because it doesn't
20 initially present itself as normalized to kilograms. Is
21 this making sense a little bit more now? Okay.
22 So in spite of the fact that the x axis is not 1 230 2 1 saying so, I have fit the relationship between this ratio
2 and weight here.
3 The next model I looked at is a combination of
4 a weight effect and an age effect. The weight effect is
5 what you saw on the previous slide. Here I've added in
6 the age effect. The effect of the two summed together,
7 each of which is a simple exponential relationship, is
8 shown with the green line. The weight effect, which is
9 what I described previously, is shown with the line in the
10 middle, sort of the pink dashed line, and finally, an age
11 effect which is what's new on this slide.
12 Now, you'll notice that I fit 6 points with 4
13 parameters. My point here is not to show that I can draw
14 pretty lines. The reason why I'm presenting this to you
15 is because it shows the sort of strategy you might take
16 when we have a larger database and how we might think
17 about developing the models on our own data set.
18 Did this confuse everyone further? Can I aid
19 anyone?
20 DR. SHEINER: Yes. You don't have any
21 independent information in this particular data set.
22 DR. WILLIAMS: Correct. 1 231 2 1 DR. SHEINER: You have weight, which is a
2 deterministic function of age, and then you have age,
3 which is a deterministic function of age.
4 DR. WILLIAMS: Indeed.
5 VOICE: (Inaudible.)
6 DR. SHEINER: No. He got it from a table. So
7 it's a deterministic function of age. So if I were to
8 write your equation, it really is Rmax 1 minus E to the
9 minus Kf of age plus Rmax 1 minus E to the minus K of age.
10 So all you've done is done a shape thing. By restricting
11 it to exponentials, you get more information out of two
12 exponentials than one. It's just like when we have time.
13 And you said the right thing at the end which is this is
14 just an illustration.
15 (Laughter.)
16 DR. WILLIAMS: I would agree. What we do have
17 going here, though, is that the shapes -- I haven't looked
18 at this specifically, but the shapes -- well, actually I
19 have to an extent. The shapes are different. If you
20 plot --
21 DR. SHEINER: No. I'm saying if I used a
22 spline or some flexible function of age, I could only get 1 232 2 1 one term in age because it would be as many parameters as
2 I needed, but because you've broken it up into two
3 exponentials, you can get two terms in age because they
4 don't have the same shape because the function of age that
5 weight represents is another shape change. So it's just
6 like saying I have a polynomial in age. It's not a
7 polynomial. It's a flexible function in age.
8 DR. WILLIAMS: I would agree.
9 DR. SHEINER: But it doesn't prove that you
10 fractionated an age effect away from a weight effect.
11 DR. WILLIAMS: I would agree on that.
12 DR. SHEINER: Okay.
13 DR. WILLIAMS: Interestingly, I was somewhat
14 surprised it turned out as consistent as it did because
15 one thing I did as a check -- like I said, it wasn't
16 essential to my purpose because I'm just trying to show
17 you the kind of strategy I would employ. But one thing I
18 did do is I went back and switched the order in which I
19 added the two, and interestingly I got the same
20 relationship. I don't know if that's surprising,
21 meaningful, or what, but it did happen.
22 DR. SADEE: Can I ask you about this? Using 1 233 2 1 weight as a scaling may not be all that appropriate. So
2 rather than saying there's an age effect, is there any
3 information on body surface area which would just do away
4 with this --
5 DR. WILLIAMS: No. The answer is no. In this
6 database everything was normalized according to body
7 weight, and other than the numbers as presented, which
8 were always per kilogram, I had no raw data.
9 DR. SADEE: Well, could you translate this
10 into body surface area which would provide you with a
11 different scale and it may actually do away with the need
12 to invoke age? Because body surface area changes with
13 respect to weight and age, so it may account for both.
14 DR. WILLIAMS: Perhaps. When I actually do
15 this on our own data set, the path that I intended to
16 follow was, first, to describe the effect of weight or
17 mass or BMI, ideal body weight, BSA. I would investigate
18 a number of weight metrics. Then whichever one fit best I
19 would use and then proceed to looking at what an age
20 effect, in the absence of a weight effect, would be to see
21 if I can describe one, and if so, what it is. That's my
22 intent here. 1 234 2 1 The reason why I did this in this way is to
2 try and not have the effect of age confounded by weight
3 and the only weight metric I had, which I couldn't pull
4 out in any other way, was kilograms.
5 The path, in addition to what I've just
6 indicated for the models that we would consider -- and
7 this is speculative. It will be driven by data, but this
8 is sort of how I see it going and thought it would be
9 useful to share.
10 First, we would consider adding exponentials
11 if the data supported it. You've seen a two-exponential
12 fit, and as Dr. Sheiner said, it's effectively both
13 descriptions of weight. But the idea is if we should
14 discover that at early ages there appears to be a unique
15 phase of the curves -- it doesn't occur at later ages --
16 and is not as simple as a single exponential fit, then we
17 would consider adding additional exponentials, a
18 structural model, if you would, of more than one term.
19 One thing also I haven't shown here is an
20 offset for the age effect. That is, you might not expect
21 that the age effect would be 0 at birth. Whatever
22 processes are occurring, maturation may begin -- almost 1 235 2 1 certainly did begin -- in utero. So you might want to add
2 a term that would give an offset for the age effect.
3 Again, this is speculative, but not based upon a real data
4 set that I'm currently evaluating.
5 Finally, we would begin to look at more
6 physiologic covariates. Can you enter covariates such as
7 the percent excreted unchanged, the Km for a given enzyme,
8 the Km ratios across the enzymes for which metabolism of
9 the drug is responsible?
10 Finally, the approach I've shown is very
11 empirical. Whether or not it will be successful is a
12 question, and if it were unsuccessful, the next step would
13 be to consider models which are more mechanistic.
14 The first mechanistic model perhaps to be
15 considered would be one that is less than a full-blown
16 PD/PK model but which does incorporate what I'm calling
17 process constants, such as GFR, such as Km and the percent
18 non-renally eliminated.
19 Finally, if even such mechanistic models were
20 not successful, you might have to go into more of a full-
21 blown physiologically based pharmacokinetic model. The
22 difficulty with that is clear. Usually it's difficult to 1 236 2 1 obtain data to support such a physiology-based model.
2 Would anyone like to tell me if they think
3 this is reasonable?
4 DR. VENITZ: Thank you, Gene.
5 DR. DERENDORF: I have a question for
6 clarification. Are these all lumped together, hepatically
7 cleared drugs, renally cleared drugs, high extraction, low
8 extraction, no difference?
9 DR. WILLIAMS: Yes. The initial cut would be
10 the raw data is age versus clearance independent of the
11 physiologic mechanism by which drug is excreted or
12 eliminated.
13 DR. DERENDORF: Well, then I would not agree
14 with the first question that it's logical because I would
15 expect there to be major differences depending on the
16 mechanism of clearance. Obviously, for a renally cleared
17 drug, the enzymes don't matter, and for high extraction
18 drugs, the intrinsic clearance would matter, and so on.
19 So I think to break it down into several subgroups would
20 make a lot of sense.
21 DR. WILLIAMS: You might expect that as the
22 physiology differs, so will the relationship between 1 237 2 1 clearance and age. But I guess it's dependent upon a few
2 elements. One is to what extent is each one of those
3 different elements not a function of allometry. In other
4 words, if the enzyme maturation and, say, GFR are both
5 linear functions of weight, then you might expect that
6 this approach would work.
7 Now, the clear difficulty is at early ages,
8 there's a whole literature, which many of the members of
9 this committee have helped develop, that speaks to the
10 fact that that is sometimes or perhaps oftentimes not the
11 case. So the question is perhaps, how much data do I have
12 that can address that? Specifically, do I have a lot of
13 data at early ages or how much does it vary?
14 Would you agree that --
15 DR. DERENDORF: I'm not an expert, but if I
16 recall correctly, glomerular filtration rate in a 2-year-
17 old is almost like an adult. Right? There's no
18 difference. Whereas, clearly the number shows that for a
19 2-year-old the clearance per body weight is almost twice
20 as high. So there are differences depending on the route
21 of elimination, clearly, because you wouldn't expect a
22 difference for a renally cleared drug. 1 238 2 1 DR. LEE: Can I add to that? What we're
2 planning to do is look at different drug classes. For
3 example, we will look at a drug class which is purely
4 renally cleared and then try to see what's the
5 relationship between clearance and age. Then we will look
6 at a bunch of drugs, for example, 3A4, purely 3A4, and
7 then see how it's going to change in our clearance versus
8 age. So that will probably address your question. And we
9 know from the literature that the maturation of different
10 enzymes are very different. I mean, 2D6 may be fast and
11 3A4 may be slower. This is what we planned. We want to
12 look at different drug classes and build a model perhaps
13 about one drug class at a time, and then finally we have
14 an individual model. Then we will look at a drug that has
15 a combined pathway, maybe a drug with 20 percent 3A4, 40
16 percent 2D6, and see if the model can actually predict the
17 age effect.
18 DR. SADEE: Do you consider changes between
19 males and females and the various sexual developments and
20 so on? If you talk about maturation of enzymes, which
21 sounds a little fuzzy of a term, but males and females are
22 probably very different, but that may also depend on the 1 239 2 1 age. I don't know.
2 DR. KEARNS: Actually with respect to drug
3 metabolism, they're not very different at all. There are
4 a few examples of substrates for P450s that during
5 adolescence differ a bit and it probably has to do with
6 the things that make for differences in linear growth more
7 than sexual maturation. But for the most part, it's
8 pretty boring, boys and girls, before puberty.
9 DR. SHEINER: If you go back to that picture,
10 the one with the graph that we had the problem with, if I
11 just connected the dots, I'd have a function defined by
12 line segments of that y axis value, the relative
13 clearance, versus age. And everybody would be completely
14 clear that I had totally explained all the data with my
15 function, and consequently there's no way to partition out
16 the effect of age from the effect of something else in
17 your case that is a function of age, which is weight. So
18 the fact that you could partition out, means you made some
19 kind of very powerful assumption that allowed you to take
20 this function of the x axis and see it as a separate
21 contributor to this curve which is practically connecting
22 the dots. So that's the point I'm making. 1 240 2 1 It's nice that you started with this one
2 because here we all know there's only one variable age.
3 We don't know anything about weight, not the real weight.
4 So you've got perfect what's called multi-colinearity or
5 you've got a perfect problem that the one substitutes for
6 the other. The value from your table can be translated
7 back exactly into the ages and the other way around.
8 So this is what's called, you know, an ill-
9 posed inverse problem or an unidentifiable model. It
10 depends on what field you're in. And the only way you can
11 get something out of those -- and I'm going to get to the
12 key question here in a minute.
13 But the only way you can get something out of
14 those, if you're trying to learn about these different
15 effects, is you have to make some very powerful
16 assumptions, some assumption that allows you to separate.
17 Here, the assumption you made is the exponentiality.
18 So I could say, if I didn't know anything
19 about this, if you have a solid basis in physics for that
20 exponential equation, something at the level of theory
21 that's as powerful as physics, that says each of these has
22 to influence and the only way it can influence the spread 1 241 2 1 of these points is through a rising exponential, then I'll
2 believe that what you sort out of the two effects is right
3 because that's the key piece of information that you've
4 added that you told me essentially everybody knows this is
5 true. But if you say, well, I just -- the exponential
6 because it kind of went up and then it went over, you
7 know, then there's no reason to believe you've got it
8 sorted out right now.
9 I don't want to criticize this picture because
10 you were clear this was an example, but you're going to
11 have a similar problem with the real data. That is to
12 say, there's going to be very high correlation between age
13 and weight. So you got an almost ill-posed problem. You
14 got an almost unidentifiable model.
15 So if your goal is to sort out the independent
16 effects of things we can measure like size -- even that's
17 a surrogate for something else -- and what we will all
18 agree is an unknown age-related something called
19 maturation or something like that that we don't measure
20 when we measure renal function, that we don't measure when
21 we measure these other things, if your idea is to sort
22 that out and then look at the shape of that thing, we're 1 242 2 1 going to have a lot of trouble believing it even when
2 you're not as bad a situation as this because of that high
3 colinearity.
4 So I told you I'd get to the question. So my
5 question is, what's the question? What do you really want
6 to learn? If you want a predictive equation, you could do
7 anything. I'm not being facetious. If you want a
8 predictive equation and you're not going to interfere with
9 the system, you're not going to deliberately change
10 people's ages or weights or whatever, you can just let it
11 come as it falls, and you find some general predictive
12 equation for clearance of all drugs as a function of a few
13 easily measurable things, that would be completely valid
14 as long as you don't go in and mess up the system, even
15 though it doesn't give you an understanding necessarily of
16 what the causes are.
17 But on the other hand, if you say no, I want a
18 predictive equation for a whole new drug, then you're
19 interfering because you're not sampling from the same
20 world.
21 So I guess my question is, what's your
22 question? 1 243 2 1 DR. WILLIAMS: First of all, I would say if it
2 does turn out as you're saying and you do have this high
3 colinearity and this difficulty, it seems to me that would
4 be good news because it means that you have the ability to
5 describe the relationship between the ratio and the age as
6 only a function of weight. If the data is well described
7 by small or perhaps a single parameter, that would be
8 fantastic.
9 But to get to your later question, perhaps you
10 can educate me some more. It's not so clear to me if you
11 had the situation that you're defining where you did have
12 this high amount of colinearity and you then parsed out
13 your drugs as a function of all of the things that you can
14 look at, metabolic route, percent renally eliminated
15 unchanged, et cetera, and you found that you could not
16 identify covariates which interfered with that
17 relationship, then wouldn't it be legitimate to extend it
18 to the new drug?
19 DR. SHEINER: That would certainly be an
20 empirical basis on which you would guess that a new one
21 would look just like the other ones because you had a
22 whole bunch that all looked the same and you hadn't chosen 1 244 2 1 them for that purpose. I agree it wouldn't be a
2 mechanistic basis. That's why I said sort of a whole new
3 drug class. Somebody might say, sure, you know, you've
4 been right 85 times out of 90 so far, but you've never
5 looked at one like this before. So that's the problem.
6 That's what I'm saying.
7 I agree with you, it's good news if they all
8 look the same because it gives you more faith that the new
9 one, even though we know it's a little bit different, will
10 also look the same. But that's just counting how many
11 times have I been right out of how many times I've tried.
12 So is that your goal?
13 DR. WILLIAMS: Yes, that would be the goal.
14 I'm new to this area, but from my read of the
15 literature, I think it's unlikely that we would see that
16 at young ages. Now, the question is the quality of my
17 data. Do I have a lot of data at young ages? Because I
18 would expect that that's where they will separate from a
19 simple function of weight.
20 DR. JUSKO: Don't you at each age have data
21 from many children where there's a distribution of weights
22 at each age? 1 245 2 1 DR. WILLIAMS: Yes.
2 DR. JUSKO: So if you do have that, then you
3 have the ability to discern the separate effects of age
4 and weight. So I don't understand Lew's --
5 DR. WILLIAMS: Yes. Our data set will be an
6 improvement of that. But if the colinearity is -- if
7 they're very highly correlated --
8 DR. JUSKO: Certainly they're highly
9 correlated, but there are factors in addition to age that
10 control weight that you would be able to discern through
11 this kind of analysis.
12 What do you select as the upper limit in age
13 or weight to reach the maximum? Because for many
14 physiological functions, the graph will look like this for
15 a certain age range, but reach a maximum at about 18, and
16 then as we all know, we steadily deteriorate until we
17 reach some lower level.
18 (Laughter.)
19 DR. JUSKO: So a more insightful empirical
20 function may be more of a U-shaped type of --
21 DR. WILLIAMS: I didn't grapple with that here
22 obviously. What I did is I just fixed it to 1. But yes, 1 246 2 1 it's a question. I guess what I would do is I would try
2 and look at as great a wealth of adult data as I can to
3 see if there is an age relationship, and if so, where's
4 the maximum I suppose. But you're right. That's
5 something that will have to be worked through.
6 DR. SHEINER: Bill, I think you're right
7 except I'm not sure what you get out -- so if I've got
8 different ages and I've got a lot of weights at different
9 ages, then it's quite true that if I want to say there is
10 some sort of sum of these two effects that's operating,
11 then if it were only weight, then I should be able to go
12 across age and find it at the same weight. Everybody had
13 the same clearance. And if I didn't, then I'd have to
14 explain that by age. Is that sort of where you're -- yes.
15 DR. JUSKO: Age is likely to be the strongest
16 determinant, but then we have to bring in genetics and
17 diet and the rest as additional determinants of weight.
18 DR. SHEINER: Yes. I think Bill is completely
19 right, that having independent variability in age and
20 weight will help. It's just how much of that do you need
21 to feel comfortable about what you get out. And strange
22 things like Simpson's paradox can happen where, as you 1 247 2 1 move from age to age, the regression within each age
2 versus weight could be actually opposite the direction
3 that it was when you do the whole thing. These sorts of
4 things happen. So it's just a matter of what the data
5 contain.
6 I guess one of the things that bothers me is
7 that with kids it's going to be a very, very strong
8 relationship; whereas, with adults we expect a small age
9 effect -- the deterioration that Bill is talking about --
10 and a large weight effect.
11 But again, for predictive purposes, it doesn't
12 make any difference. If you're not going to do anything
13 different the next time, then your data has got all the
14 information. I suppose what you're saying is an
15 interesting result here would be if there was a relatively
16 simple equation that predicted most of what you see across
17 lots of drugs.
18 DR. WILLIAMS: Yes. I anticipated that I
19 would get the reaction that I believe Hartmut is
20 expressing, which is this is sort of naivete to expect
21 that it would go that way and that it's probably very
22 important to incorporate physiologic covariates. But one 1 248 2 1 of the things that drove me to think about it in this way
2 is it made sense to operate on parsimony and start simple
3 and see what the data set would support.
4 DR. DERENDORF: But you want to use anything
5 that you already know. So if you know that you have
6 gentamicin, you can use glomerular filtration rate as a
7 pretty good estimate. So if you know what the
8 physiological value is for that, you should use that and
9 not ignore it.
10 DR. KEARNS: Just two comments, one to speak
11 in support of doing this in the context of getting you in
12 the ball park because I think it clearly has the ability
13 to do that. You're right in that for very young infants
14 and perhaps even up to 6 or 8 months of age, it's not an
15 issue of weight for much of the maturation. Probably
16 post-conceptional age falls out as a way to best predict
17 these things because you do come to the field with a
18 little bit of activity depending upon when you come to the
19 field. That's clear.
20 But from a practical standpoint, I was sharing
21 at the break our experiences in doing a study with a drug
22 that was 100 percent renally cleared. This drug was 1 249 2 1 studied under that list of 72 who've been given
2 exclusivity, and they got their 6 months of extended
3 marketing exclusivity and had a big party and everyone was
4 happy.
5 But in going back in time and looking at those
6 studies, we were able to simulate the results of the trial
7 before we enrolled one patient, as you might believe you
8 could do. We made an argument that we thought was
9 reasonably passionate, but perhaps not sufficiently so,
10 that the trial that was done needed only to contain
11 children in the first 3 months of life, that everything
12 else could be predicted. We were sent away believing
13 that, indeed, some revelation would occur. We could use
14 the knowledge that we had to simplify and improve and
15 streamline the process only to find out that we were
16 wrong, that ultimately we were expected to fill in all the
17 pieces of the puzzle of the barnyard despite knowing that
18 it indeed contained animals of identity that was known.
19 At the end of it all, there was a lot of time, effort, and
20 money spent for no good reason. No good reason.
21 So for all the pimples on an approach like
22 this mathematically, this has the ability to improve the 1 250 2 1 design of pediatric studies if, within the halls of this
2 wonderful agency, it would just be used to do so.
3 DR. WILLIAMS: I guess, Larry, that helps
4 justify my detail.
5 (Laughter.)
6 DR. LESKO: I was just going to comment on
7 what is the question. Actually Greg's comments are
8 discouraging to what I was going to say.
9 (Laughter.)
10 DR. LESKO: Nevertheless, when Lewis asked the
11 question about what is the question, it would seem what we
12 want to know is not necessarily the empirical relationship
13 we're talking about, but rather the more mechanistic one
14 where you can look at a drug and it would be a whole new
15 drug, but look at it not as a whole new drug in a
16 therapeutic class per se but a whole new drug with certain
17 attributes of processes of elimination, cytochrome
18 enzymes, what we know about Bmax and Km's of those
19 enzymes, and based on that information and based upon the
20 analysis of a pediatric database, know where there are
21 breakpoints in the age groups and perhaps do a limited
22 study that might bracket age groups, and then you can fill 1 251 2 1 in the blanks in between based on some model to say I know
2 this from these relationships between routes of
3 elimination and age. I might do some limited studies, but
4 then cover age groups I haven't actually studied in terms
5 of extrapolating that information. It's kind of what you
6 were saying in terms of knowing something in the first 3
7 months and then using that to predict the rest of the
8 puzzle, but it just struck me that if there's a difficulty
9 in doing that with a drug renally excreted, the difficulty
10 becomes magnitudes more for a drug that is out the enzyme
11 system.
12 But nevertheless, that's our noble mission
13 here to try to look at the database. Maybe we just need
14 more examples of this using data we already have as
15 opposed to something new. I don't know, but I think where
16 we want to go eventually is to take attributes of a drug
17 and be able to make better predictions and maybe even
18 excuse pediatric studies from being conducted if we're
19 confident enough that we can predict clearance in those
20 age groups.
21 DR. KEARNS: And the other thing -- and what
22 I'm about to say is not mine. I really owe this to Steve 1 252 2 1 Spielberg who began preaching this some time ago -- is
2 that if you can define a breakpoint, if they really exist,
3 then it's possible to design your trial in such a way to
4 enrich it so that you can get the most information out of
5 the least numbers of subjects studied. That's okay
6 because in the process, we don't compromise the end game
7 result, and we also don't put children in trials just for
8 the sake of confirmatory purpose.
9 Parents always want to know, especially for a
10 nontherapeutic trial, they say, explain to me again why
11 this is important. As an investigator, you really have to
12 be able to tell them that. If they're convinced, you have
13 a child on your study and you have good data. If they're
14 not convinced, maybe you shouldn't be either.
15 DR. KARLSSON: Maybe in addition to what you
16 presented here, your analysis, if you looked at it, could
17 also settle the debate that was before the break regarding
18 what about variability in elimination capacity with age.
19 Does it actually decrease? Does it increase? Is that
20 dependent on the elimination pathway, et cetera?
21 DR. WILLIAMS: Yes.
22 DR. FLOCKHART: I guess I'm back to what's the 1 253 2 1 question. It seems to me, Larry, that you gave a
2 different answer to the question.
3 Greg, I think if the playing field is so big
4 and has very significant error within it, I've got to ask
5 the question, what's the point of asking where it is on
6 the playing field. I can't tell by looking at the playing
7 field whether the ball is in the goal or is at the
8 centerpoint. If it's a very vague thing from doing this
9 kind of activity before, I'm not sure knowing it's on the
10 playing field is a valuable exercise.
11 On the other hand, I guess we're all biased as
12 scientists towards believing that a more mechanistic,
13 physiologically based approach would work better. But I
14 have to say that that real hypothesis even in adults has
15 not been really hard core tested. We don't really know
16 that.
17 So this becomes, therefore, a testable
18 hypothesis. Each time you add a new drug, you're testing
19 the idea, and if it turns out even between the ages of 1
20 and 16 that things fit, that would be a tremendously
21 valuable thing to know that we kind of got gratis, we got
22 free as we went along. 1 254 2 1 The error is the key, though, because it's
2 very variable.
3 DR. KARLSSON: Are these 72 studies
4 intravenous studies, or are you going to mix IV with oral
5 studies? If so, how do you handle bioavailability issues?
6 DR. WILLIAMS: Until I look at the actual
7 data, what drugs are available and pick them, I won't know
8 that answer. In other words, how do I choose among the
9 72? Perhaps the easiest way to start would be to look at
10 IV drugs, but on the other hand, perhaps you would have a
11 willingness to accept drugs whose metabolic route is well
12 defined and is thought to be largely one process. So the
13 72 will be a mix, but which ones I choose to look at first
14 and how I develop it is something that we have to consider.
15 DR. CAPPARELLI: I would just like to echo
16 what Mats was referring to, that the oral component is
17 going to be huge, especially in the younger age groups,
18 and everything that's been said before about some of the
19 formulation issues. So I think that it's going to take
20 careful selection. I'm excited to see this direction, and
21 I lean on the mechanistic fence of things, starting off
22 with what we know and building on it. I think renally 1 255 2 1 eliminated drugs make a lot of sense from the standpoint
2 of what we know about renal function, how we can measure
3 it or how we can, at least, estimate it in different
4 pediatric populations and relate it.
5 But you start getting into drugs where there
6 are bioavailability issues in adults, it's going to go all
7 over the map, and you're going to have the additional
8 confounding issue if you've got active transport or gut
9 metabolism. And those may not parallel what's going on in
10 the liver.
11 So, again, trying to simplify it and at least
12 starting it at points that I think you'll have buy-in and
13 belief in the model I think is very important.
14 DR. SHEINER: How variable will be the way in
15 which the clearance was determined in the individual
16 children across these studies?
17 DR. WILLIAMS: I guess fundamentally you can
18 separate into sparse and dense, and we're likely to see
19 both of those. But beyond that, how studies are
20 conducted, sampling times, populations, numbers, probably
21 a wide range.
22 DR. DERENDORF: Just to clarify, because I was 1 256 2 1 under the assumption from the beginning it was only IV
2 data. Now you're saying that there was some oral data.
3 So they were the ratios of the oral clearances between
4 kids and adults that you showed in that very first table?
5 DR. WILLIAMS: My recollection of -- oh, this
6 is perhaps unfortuitous. This is slide 4, not data IV.
7 (Laughter.)
8 DR. WILLIAMS: So the answer to the question
9 is yes. Certainly some of these or perhaps all of them
10 are. I did not separate this out into oral versus IV.
11 Now, when we actually perform this on the FDA
12 database, of course, we will have the luxury of choosing
13 the order in which we consider the drugs. Obviously,
14 there's an advantage, especially given Dr. Capparelli's
15 comments, of beginning with the simplest case which would
16 be IV drugs.
17 DR. DERENDORF: The probability of getting
18 anything useful out of it then is very low in my opinion.
19 I think you have too many things that are lumped together
20 here.
21 I'm amazed at the ratios, that they come out
22 to be so close to 1 in that figure there. If oral 1 257 2 1 bioavailability is included there, it's almost hard to
2 believe.
3 DR. LESKO: But in our own data set, we can
4 control for that. We can select drugs, as we said, taking
5 care of those differences. We would combine drugs in
6 different ways that take those similarities into account,
7 whether they're IV or oral, and not mix them. He's
8 working with a published data set, but I think your
9 question and issue would be resolved if we picked from the
10 data set appropriately that we have within the FDA. Isn't
11 that right? Am I misunderstanding?
12 DR. DERENDORF: If you stick to high
13 bioavailability drugs where there is not a big difference
14 and where we don't expect a big difference, but if you
15 have high extraction drugs in the group, you would really
16 get numbers all over the place.
17 DR. LESKO: But I think one of the plans would
18 be to look at different ways of categorizing the drugs
19 that we're looking at to see if that makes a difference or
20 not.
21 DR. DERENDORF: Well, the question changes
22 completely. Initially it was a question of how do 1 258 2 1 metabolism and clearance develop, and now we have included
2 bioavailability. We have formulation issues. We have
3 transporter issues. We have intestinal metabolism, I
4 mean, a whole bunch of things that happen all lumped in
5 one number. And I think the chance of filtering out
6 anything that teaches us something is very, very slim.
7 We'll get some kind of an average curve and we can fit a
8 line through it, but what does it mean?
9 DR. KEARNS: But, Hartmut, if I told you that
10 the data set that they had had over 300 patients
11 intensively studied with midazolam, half of them on oral,
12 half on IV, from ages of 6 months to 16 years, all of a
13 sudden it becomes a little more interesting. And that's
14 in their data set. So there is some gold in there to be
15 mined. But your point is well taken in that it's not
16 something to be done in a reckless way not paying
17 attention to all the assumptions and limitations.
18 DR. DERENDORF: Again, I think where I'm
19 coming from is to utilize anything that we already know,
20 and the physiological information about blood flow, for
21 example, is something that doesn't vary that much. That
22 should be included in the data analysis. It should focus 1 259 2 1 on intrinsic clearance as the number to correlate with. I
2 think then it makes much more sense.
3 Then you have a chance that you can identify
4 maturation rates for the various enzymatic pathways that
5 you then can use to extrapolate for new compounds. Once
6 you know that for a new compound which is the breakdown of
7 different pathways, you already know how the rate of
8 maturation occurs and you can make good predictions
9 without any study.
10 DR. WILLIAMS: If you're right -- and I
11 initially tended to think that way too, but like I said,
12 this is new to me -- then we will get there because what
13 will happen is the simple models will fail.
14 DR. LESKO: Gene, do you know the size of
15 these studies, just speaking about the small to large of
16 the pediatric studies in the database? What's the typical
17 n in these studies in terms of getting an estimate of
18 precision of the pharmacokinetic measurements?
19 DR. WILLIAMS: I really don't.
20 DR. LESKO: Greg, what do you do? Or, Ed,
21 what's the typical size of a PK study in a pediatric
22 population in your experience in terms of a single age 1 260 2 1 group or all the groups?
2 DR. CAPPARELLI: When you say all the age
3 groups, that incorporates a little bit of gray as you get
4 further on down and looking at degree of maturation at
5 birth. Some of those studies get to be very large.
6 Typically most of the stuff that one sees may
7 not be optimal, but rarely do you see anything less than
8 50. Most of it is in the 100 to 200 range if it's
9 incorporated into the safety trial.
10 But the driving force isn't often the optimal
11 PK component. It's really the other aspects of the study.
12 So again, the precision issue really comes near the cut-
13 points which I think was brought up earlier. There is
14 often a lack of information where the action is,
15 unfortunately.
16 So having a large data set that has three
17 patients under the age of 2, and you get this spread
18 that's here, here, and here, and then trying to make some
19 sense that no, nothing is going on down there or there's
20 something very dramatic going on down there really is
21 based on the belief of who's looking at the graph rather
22 than any real aspect of the data. 1 261 2 1 DR. KEARNS: Larry, for the phase II things
2 done under most of the written requests that I've seen,
3 it's about 24 subjects to 36.
4 Some of it's -- I need to pick my words right
5 because we're on tape, but it's interesting some of the
6 ways the designs are done. For instance, if we had a drug
7 whose metabolism we knew or believed changed greatly in
8 the first 3 months of life, you might see a written
9 request that asks the sponsor to include 24 infants from
10 the age of birth to a year and that the infants should be
11 equally distributed across the age spectrum, so there
12 should be at least 3 infants or 4 infants in the first
13 month of life. And oh, by the way, you can study babies
14 all the way down to 800 grams. Now, the chance of coming
15 out of that at the end of the day with a revelation of
16 therapeutic utility is slim to none.
17 But unless my experience is somewhat not
18 representative, this is happening every day under the
19 context of negotiating a written request for drugs studied
20 under the Best Pharmaceuticals for Children's Act, which
21 is why I believe that until we put some of this bit of
22 science and ingenuity into the action plan, we're really 1 262 2 1 not serving the intent of the people who put that act
2 together or, even worse really, the children as we try to
3 make a fact out of fancy many times.
4 DR. VENITZ: Any other comments, questions, or
5 further discussion?
6 DR. WILLIAMS: Are there data sources anyone
7 here can recommend?
8 DR. VENITZ: Dr. Sheiner.
9 DR. SHEINER: At the risk of stupefying Mary,
10 are you going to go back to the original data, the
11 measurements of concentrations versus time, or were you
12 thinking to use those clearances?
13 DR. WILLIAMS: The notion is that we would
14 just go back to the clearances.
15 DR. SHEINER: Okay. And those clearances will
16 be some, you know, from 15 samples after a single dose and
17 a nice area under the curve.
18 DR. WILLIAMS: Right. Some of them would
19 probably come from population post hoc and some would come
20 from dense.
21 DR. SHEINER: Well, you want to think about
22 how you mix those because those posterior Bayes' estimates 1 263 2 1 are funny creatures. They're centrally biased. So they
2 do odd things to regressions when you use them either as
3 the explanatory variable or as the variable to be
4 explained. So you need to think about that.
5 DR. WILLIAMS: Yes. It seems to me it gets
6 pretty complicated if you don't do that, but would you
7 like to propose an alternative?
8 DR. SHEINER: Well, if you don't do that, it
9 doesn't get complicated. That complication goes away. If
10 you use the original data, the complication I was just
11 talking about goes away because you're not summarizing the
12 data with a strange estimate. But it does mean that you
13 have lot longer run times and a lot more modeling to set
14 up and all those nasty things.
15 I don't know. I would almost be tempted to
16 say, since you're going to use them in subsequent
17 regression and you have lots of data, that you should use
18 unbiased estimates of each individual's clearance which
19 you would get essentially by taking the prior variant
20 system infinity or fitting the trapezoidal rule with three
21 points. I mean, that's bizarre to talk about. But I don't
22 know. I have to think about it. It's not obvious. 1 264 2 1 DR. WILLIAMS: Perhaps we can discuss this
2 further and I can come to the committee as a whole or
3 perhaps even yourself showing you the actual
4 characteristics of the data set.
5 DR. VENITZ: With that, I think we're ready to
6 conclude.
7 DR. WILLIAMS: Actually if I could make a --
8 DR. VENITZ: You conclude.
9 (Laughter.)
10 DR. WILLIAMS: First, a number of the
11 committee members are very active in this area. If you
12 would like to share your data, we would certainly welcome
13 it. As I said, some limitations of our data is we often
14 don't see very young ages and we often don't see probe
15 substrates. So if you'd like to contribute, we sure would
16 welcome that.
17 Finally, should the very empirical models not
18 be successful, the form that you would use not as far as a
19 full physiologic-based model, but incorporating some of
20 these physiologic covariates into the description of age
21 versus clearance is not entirely clear to us. We've
22 worked on it a little bit, but we don't have any firm 1 265 2 1 conclusions. If anyone would like to contribute here or
2 perhaps even off line how they would see the form of those
3 equations running, we'd be grateful.
4 DR. SHEINER: Just a quick question about
5 that. Usually PBPK to me means the various compartments
6 connected in various ways and blood flow from the gut
7 going to the liver and things like that. Is there a large
8 collection of physiologic models of clearance?
9 DR. WILLIAMS: I'm not sure I understand.
10 DR. SHEINER: Well, I mean GFR and renal
11 clearance. They seem to be linked, and usually people do
12 it linearly. I haven't seen too many what I'd call
13 physiologic models of clearance. That would be a thing
14 where you'd have some model of the uptake mechanism and
15 then the transport and the metabolism. That might be
16 *McHale-Smitton or something like that.
17 DR. WILLIAMS: Well, the ones I'm most
18 familiar with sort of group all those sorts of things into
19 a global parameter, intrinsic clearance.
20 DR. SHEINER: But you're asking for a model
21 for the intrinsic clearance. Right? Or for the Q times
22 clearance over Q plus clearance, or something like that. 1 266 2 1 Because you're not asking for a model of the drug level.
2 DR. WILLIAMS: Right.
3 DR. SHEINER: That's what we usually think of
4 when we say PBPK, models of how do concentrations relate
5 to physiologic processes. But you already decided that
6 you are looking at a physiologic process called clearance,
7 and I'm not aware of an awful lot of physiologic models,
8 except I guess everybody who thinks about it could come up
9 with things they think would be more, let's say,
10 reciprocally related and things that would be more
11 directly related. But other than that, I think -- most of
12 those models are empirical, and even in the middle of a
13 population analysis, there's some little equation that
14 says that clearance is a linear function with an intercept
15 often, which doesn't make any sense, of weight, age, and
16 some measure of renal function or of hepatic function.
17 DR. VENITZ: Sounds like a topic for another
18 meeting.
19 DR. WILLIAMS: Thanks, everyone.
20 DR. VENITZ: Thank you, Gene, and thank you
21 all for hanging in for today's agenda.
22 We are adjourning the meeting. We are 1 267 2 1 reconvening tomorrow at 8:30.
2 (Whereupon, at 4:10 p.m., the subcommittee was
3 recessed, to reconvene at 8:30 a.m., Wednesday, April 23,
4 2003.)
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22 1 268 2 1
2
3