1 DEPARTMENT OF HEALTH AND HUMAN SERVICES FOOD AND DRUG ADMINISTRATION INFECTIOUS DISEASES SOCIETY OF AMERICA INTERNATIONAL SOCIETY OF ANTI-INFECTIVE PHARMACOLOGY FOOD AND DRUG ADMINISTRATION ANTIMICROBIAL DRUG DEVELOPMENT WORKSHOP Friday, April 16, 2004 9:05 a.m. Advisors and Consultants Staff Conference Room 5630 Fishers Lane Rockville, Maryland 2 3 PARTICIPANTS John E. Edwards, Jr., M.D., Moderator INDUSTRY Lisa Benincosa, Ph.D. Mike N. Dudley, Pharm.D. Barry Eisenstein, M.D. Dennis M. Grasela, Pharm.D., M.D. Timothy J. Henkel, M.D., Ph.D. John H. Rex, M.D. Frank Tally, M.D. Gregory A. Winchell, Ph.D. ACADEMIA David Andes, M.D. William A. Craig, M.D. Hartmut Derendorf, Ph.D. George L. Drusano, M.D. Jerome J. Schentag, Pharm.D. George Talbot, M.D. Paul M. Tulkens, M.D. FDA Renata Albrecht, M.D. Chuck R. Bonapace, Pharm.D. Phil Colangelo, Pharm.D., Ph.D. Edward Cox, M.D., MPH John Lazor, Pharm.D. J. Robert Powell, Pharm.D. John Powers, M.D. David Ross, M.D., MPH Janice Soreth, M.D. Donald Stanski, M.D. Jenny J. Zheng, Ph.D. MISCELLANEOUS John S. Bradley, M.D. Dennis M. Dixon, M.D. J. Todd Weber, M.D. 4 C O N T E N T S Page Dose Selection in Antimicrobial Drug Development Incorporation of Pharmacokinetics and Pharmacodynamics Opening Remarks: John E. Edwards, Jr. 4 Introduction, FDA: John Lazor, Pharm.D. 9 I. Overview of Use of PK/PD in Streamlining Drug Development Academic Perspective: William A. Craig, M.D. 17 Industry Perspective: Mike N. Dudley, Pharm.D. 29 FDA Perspective: John Powers, M.D. 42 Discussion 62 II. In Vitro/Animal Models to Support Dose Selection Academic Perspective: Dave Andes, M.D. 75 Industry Perspective: Lisa Benincosa, Ph.D. 90 FDA Perspective: Chuck Bonapace, Pharm.D. 103 Discussion 119 Current Status of Dose Selection in Antimicrobial Drug Development Programs Academic Perspective: George L. Drusano, M.D. 165 Industry : Dennis Grasela, Pharm.D.,Ph.D. 185 FDA Perspective: Frank Pelsor, Pharm.D. 202 IV. Improvement in Dose Selection through Clinical Applications of PK/PD in Antimicrobial Drug Development Programs Academic Perspective: Hartmut Derendorf, Ph.D. 214 Industry Perspective: Gregory A. Winchell, M.D. 228 FDA Perspective: Jenny J. Zheng, Ph.D. 252 Discussion 266 Concluding Remarks: John E. Edwards, Jr., M.D. 295 5 1 P R O C E E D I N G S 2 Dose Selection in Antimicrobial Drug Development 3 Incorporation of Pharmacokinetics and 4 Pharmacodynamics 5 Opening Remarks 6 DR. EDWARDS: Good morning and welcome 7 back. 8 I presume that all of you were given the 9 revised schedule when you came in this morning, and 10 we have made some changes, which I don't think will 11 compromise the quality of the meeting at all, but 12 as you can tell, we are scheduled now to end at 13 least an hour and a half earlier than the original 14 schedule you received yesterday. 15 I am anticipating there are not going to 16 be a lot of major objections to this change. If 17 there are, I will be happy to entertain them at the 18 break. 19 For those of you who were not here 20 yesterday, my name is Jack Edwards and I am from 21 Harbor-UCLA Medical Center, and I am also a member 22 of the Antimicrobial Availability Task Force of the 6 1 IDSA. 2 I am going to begin by taking the 3 chairman's prerogative and taking just a few 4 moments to give a very brief summary of yesterday. 5 We will be preparing an extensive executive summary 6 of this meeting, which will be on the IDSA web 7 site, and I think it may also be on the FDA web 8 site, but at least for certain, it will be on the 9 IDSA web site. 10 Last night, the members of the 11 Antimicrobial Availability Task Force had a meeting 12 that lasted quite long. It was a very stimulating 13 and animated meeting, and I must say we enjoyed 14 very much the discussions yesterday, and I just 15 want to take an opportunity to once more thank the 16 individuals involved with organizing this meeting 17 and especially the FDA for hosting this meeting 18 here as we have found the conversations and the 19 discussions extremely valuable and extremely 20 stimulating. So, let me thank you again. 21 By way of brief summary, I am going to 22 make just basically four points. The progress that 7 1 has occurred since the meeting of November 2002, as 2 described as both the IDSA and the FDA 3 presentations, is very substantive and very 4 encouraging. To summarize in just a concept for 5 each area, the IDSA is in final stages of 6 preparation of the white paper and actively 7 entertaining, beginning to engage the resources 8 necessary to create legislative recommendations 9 that will begin going to the Hill. 10 FDA gave us a very beautiful summary of a 11 large number of meetings and forums which they have 12 created, this being one of them in a sense, and 13 used to continue to develop many of the concepts 14 that came from the November 2002 meeting and 15 meetings both prior and after that. 16 We also heard a very strong commitment to 17 focus on the guidance document finalizations. 18 The three specific points that I wanted to 19 make that came from the discussions from the IDSA 20 Task Force last night are as follows: 21 The Task Force felt that it would be 22 highly useful to continue to have the guidance 8 1 documents developed with the most desirable 2 timeline being their getting to a form where they 3 could be posted on the web site for external review 4 and then finalized by the end of the year, end of 5 2004. 6 We listened to the pros and cons of the 7 guidances and also the response of industry to 8 them, but we have felt that from the IDSA 9 perspective as we have gone to industry repeatedly, 10 we consistently, over and over again, from all 11 types of industry hear the importance of the 12 guidance documents surface as a number one 13 priority, if you will. 14 We fully understand the complexities of 15 producing those guidances, and we also understand 16 the desirability for additional resources for the 17 guidances to be done and are very sensitive to 18 those issues, but nevertheless, the guidance 19 documents conceptually seem to be of extremely high 20 priority on the part of industry. 21 As we moved through the day yesterday, and 22 addressed the issues of the surrogate markers in 9 1 the primary indication for staphylococcal 2 bacteremia, again, the group felt that it would be 3 very useful to try to focus at this point on the 4 prosthetic joint infection issue in form of 5 development of a plan to study that would 6 incorporate the usefulness of a surrogate marker 7 which would be culture negativity. 8 So, focus on that particular entity was 9 thought to be very desirable, and focus to the 10 point of perhaps future discussions in the 11 Anti-Infective Advisory Committee meeting on the 12 issue of the staphylococcal bacteremia as a primary 13 indication was highlighted as a second area for 14 overall focus. 15 So, I have tried to be just be very 16 concrete, and in the interest of time, I am not 17 going to go through each of the individual 18 discussions although there will be basically a 19 summary of everything in our final summary 20 document, and I will leave the summary for the 21 moment at that now. 22 Thank you for your attention to that 10 1 issue, and now we are going to get directly into 2 our program, and I will ask John Lazor from the FDA 3 to begin with some introductory comments. 4 John. 5 Introduction 6 DR. LAZOR: Good morning. I am John Lazor 7 with the Office of Clinical Pharmacology and 8 Biopharmaceutics. 9 First of all, I would like to welcome the 10 members of the panel and the speakers for today. I 11 would also like to thank IDSA and ISAP for 12 cosponsorship of this workshop. 13 [Slide.] 14 You may ask why we decided to select dose 15 as a topic for the second day of the workshop. 16 There is many articles published in the literature 17 that have described PK/PD relationships for 18 antimicrobial drug products. 19 In addition, there seems to be a 20 systematic approach for dose selection based on 21 PK/PD that has surfaced. Because of this abundance 22 of literature, we put together an internal working 11 1 group to try to assess how this information is 2 being utilized in the antimicrobial drug 3 development programs. 4 We did this by trying to look at how these 5 approaches were being used for dose selection. 6 During this process, we did discover that there was 7 a wide spectrum of approaches, and I should say a 8 wide spectrum of rationale for the selection of 9 dose.l 10 It ranged from what I would perceive as 11 empiric or at least not very transparent to us, to 12 what I would refer to as a science-based approach. 13 So, because of these observed 14 inconsistencies and because of questions that were 15 raised by the working group with respect to the 16 methodologies, the assumptions, and the 17 extrapolations, it was decided that a workshop 18 would be beneficial. 19 In addition, getting the dose right is 20 embedded in many FDA and CDER initiatives. 21 [Slide.] 22 Three out of the five elements of the FDA 12 1 strategic plan stresses the importance of dose. 2 For instance, the plan has identified that one of 3 the reasons for the decline in the number of new 4 applications is the number of multi-cycle reviews. 5 Many of these multi-cycle reviews have been related 6 to safety and efficacy issues. 7 One can only question what the role of 8 dose was in these efficacy and safety issues. 9 Getting the dose right is important for improving 10 patient safety and for drug development for 11 counterterrorism measures. 12 [Slide.] 13 Yesterday, you heard about the FDA's 14 Critical Path Initiative which was launched last 15 month. I won't go into detail, but I just want to 16 reiterate that an objective of the program is to 17 use science and technology to create new tools to 18 improve product development process. 19 Some examples could be the development of 20 animal or computer models to predict outcomes, 21 development of biomarkers, and the development of 22 new clinical evaluation techniques. 13 1 [Slide.] 2 The critical path has been defined as 3 starting at the time when a compound has been 4 determined to move forward into product 5 development, and it is to end at the product 6 launch. This slide represents again that research 7 is a major component of the critical path. 8 [Slide.] 9 The FDA has published guidance documents 10 that promote science-based dose selection. The 11 exposure-response guidance, which was published 12 last year, is an example. This guidance promotes 13 the use of exposure-response relationships to guide 14 the design of Phase III trials and to support dose 15 and dose range selection. 16 [Slide.] 17 The end of Phase II meetings is a new 18 program designed to create opportunities for 19 sponsors to have informative discussions with the 20 FDA. One objective of the meetings is to discuss 21 the use of quantitative methodologies in the drug 22 development program. 14 1 These meetings could address the use of 2 models to forecast clinical outcomes and the use of 3 exposure-response for better informed 4 decisionmaking. Another goal of these meetings is 5 to discuss dosing strategies for Phase III. Other 6 issues, as appropriate, could be topics for these 7 meetings, as well. 8 [Slide.] 9 At the local level, the Office of Clinical 10 Pharmacology and Biopharmaceutics, through good 11 review practices, emphasizes the importance of 12 knowing exposure-response and whether dose regimen 13 and dose adjustments for subpopulations are 14 rational based on these relationships. 15 [Slide.] 16 So, there is a general theme across many 17 programs, and that is getting the dose right. The 18 goal is to optimize efficacy and minimize risk. 19 [Slide.] 20 A paradigm that has surfaced for 21 antimicrobial dose selection is to determine the 22 PK/PD in in vitro and animal models. Human PK is 15 1 added, so that a dose is identified to give a high 2 probability of success. This dose is evaluated in 3 Phase II and then it moves on to Phase III. 4 In Phase III, we have clinical outcomes, 5 we may have microbiological outcomes, and sometimes 6 we may have measures of drug exposure--I should say 7 measures of drug plasma or serum concentrations as 8 a measure of exposure. 9 What seems to be absent is an integration 10 of the outcome, the micro, and the exposure. It is 11 not known how the results from the Phase III study 12 relate to the initial predictive PK/PD promise. It 13 is important to have this understanding, so that 14 knowledge gained can be applied to the specific 15 product, so that it can be used for the advancement 16 of PK/PD in antimicrobial drug development. 17 [Slide.] 18 Today's workshop will have four sessions. 19 We will begin with an overview of PK/PD in 20 antimicrobial drug development, and then we will 21 move into a discussion on the use of in vitro and 22 animal models. 16 1 After lunch, we will talk about the 2 current status of dose regimen selection, and then 3 we will end with a session on what can we do 4 better. 5 [Slide.] 6 In today's discussion, factors important 7 for the selection of dose and dose interval will be 8 discussed, however, we cannot forget that duration 9 of therapy is an important part of an optimal dose 10 regimen. 11 [Slide.] 12 Based on the say the agenda has been 13 constructed, much of today's discussion will be 14 focused on dose regimen with an emphasis on 15 efficacy. We all recognize that it is extremely 16 important to minimize risk, so efficacy needs to be 17 balanced with safety. 18 Resistance is a third dimension that needs 19 to be considered in antimicrobial dose 20 optimization. 21 [Slide.] 22 There are some PK/PD terms used in the 17 1 antimicrobial area that are not used in other 2 therapeutic areas. For example, PK/PD index or, 3 which is the Pk-PD parameter, is a measure of drug 4 exposure. It is not only a measure of the 5 exposure, but it is linked to a measure of potency 6 relative to the pathogen. 7 You may hear reference to the terminology 8 PK/PD target. This is the magnitude of value of 9 the PK/PD index associated with either a 10 microbiological effect or an endpoint. 11 [Slide.] 12 What do we expect the outcome of today's 13 meeting to be? Well, through the presentations and 14 discussions, we hope to learn what works, we would 15 like to know what doesn't work, and we would like 16 to know the assumptions and limitations in the 17 approaches used to getting the right dose. 18 It is expected that we will learn ways to 19 better utilize the tools that we have and hear 20 proposals for improving the current approaches. 21 In fear of being I guess pinned up against 22 the wall, I also will say that one of our goals is 18 1 to evaluate the need to update the guidance 2 document developing antimicrobial drugs general 3 considerations for clinical trials, and this is 4 with respect to the Clinical Pharmacology Section. 5 [Slide.] 6 The outcome that we want to avoid is being 7 in a situation of not knowing what the dose should 8 be at the end of the product development program. 9 I look forward to the presentations, the 10 discussions, and a productive workshop. 11 Thank you. 12 DR. EDWARDS: Thank you very much for that 13 very nice introduction, John. That last slide is 14 priceless. 15 I am now going to call on Bill Craig from 16 the University of Wisconsin to begin the 17 discussion. 18 I. Overview of Use of PK/PD in Streamlining 19 Drug Development 20 Academic Perspective 21 DR. CRAIG: Thank you, Jack. 22 [Slide.] 19 1 Why all the interest in pharmacodynamics? 2 Well, if you look over the years, it has always 3 come up of interest whenever there is a narrow 4 difference between the drug exposure and the MIC of 5 the organism, and this even goes back to the early 6 days of penicillin. You can find a lot of PK/PD 7 studies in the old literature. That was because 8 the penicillin doses that were used back then were 9 very low, but as we started to find that the drug 10 was non-toxic and we could give much higher doses, 11 PK/PD sort of disappeared and there was not much 12 interest until Pseudomonas started to become a 13 significant pathogen in the 1960s and '70s, and 14 again we started to see more and more papers 15 occurring looking at PK/PD, and then the latest 16 explosion has really been with the emergence of 17 resistance to Strep pneumo, MRSA, a whole variety 18 of different organisms, and I think, though, that 19 PK/PD now is going to stay because it has other 20 applications as was mentioned for deciding on dose 21 development and dose selection for clinical trials. 22 [Slide.] 20 1 Now, for clinicians, PK/PD has had a 2 variety of different applications. It has been 3 used to help establish more optimal dosage 4 regimens, for example, once daily aminoglycoside 5 use is very common throughout the United States 6 even though none of the package inserts from the 7 FDA talk about this dosage regimen. 8 Prolonged or continuous infusion of 9 beta-lactams is also used at various institutions, 10 and again this is not always information that one 11 can find in the FDA package. 12 It has also been used to help establish 13 more reliable susceptibility breakpoints. For 14 example, the NCCLS has used PK/PD to establish new 15 breakpoints for the oral cephalosporins and oral 16 penicillins, and then also it has been used for the 17 parenteral cefotaxime and ceftriaxone for newer 18 breakpoints for pneumococci. 19 [Slide.] 20 It has also been used for preventing the 21 emergence of resistance, and I show here one of 22 Jerry Schentag's group's study in which they look 21 1 at fluoroquinolones and found that if a value had 2 an AUC to MIC ratio less than 100, resistance was 3 very common when you look at the gram-negative 4 organisms. If that value was 100, it was 5 significantly less, but I should point out on this 6 slide that is you used a combination of drugs, it 7 was even less, and I think that is what happens to 8 most of us in clinical practice now with 9 Pseudomonas is we actually drug combinations. Very 10 rarely do we use a single drug. 11 [Slide.] 12 It has also been useful for guideline 13 development. For example, the CDC guidelines for 14 pneumonia and otitis media clearly used PK/PD in 15 coming up with those guidelines, and the Sinus and 16 Allergy Health Partnership Guidelines for sinusitis 17 also have a heavy input from PK/PD. 18 With all the Pharm D.'s at hospitals now, 19 and many of them with infectious disease training, 20 what we are also finding, that PK/PD is being used 21 for formulary decisions as to which drugs actually 22 get on the formulary based a lot on their PK/PD. 22 1 [Slide.] 2 But what we are here today to talk about 3 is the application of PK/PD for new drug 4 development, and clearly, where it has been 5 applied, as was mentioned, is for dose selection 6 for Phase II and III studies. 7 What was also sort of given is the usual 8 way to do that is from in vitro or animal studies 9 to identify the PK/PD target for efficacy, and then 10 to use your Phase I pharmacokinetic studies to 11 determine which doses reach the target with a high 12 probability. 13 Now, this has been applied mostly to 14 antibacterials, but we are starting to also see it 15 now with antifungals. 16 Now, in terms of the susceptibility 17 breakpoint selection, clearly, this is required by 18 NCCLS now by their M23 document. It is clearly one 19 of the four issues that the committee looks at for 20 breakpoint selection, the others being the 21 population distributions, the mechanism of 22 resistance in the organism, and then, of course, 23 1 clinical results. 2 As far as the FDA is concerned, sometimes 3 it is used and sometimes it is not, so it seems 4 that it is more variable at least from what I have 5 been able to see as far as the FDA. 6 [Slide.] 7 What companies would like to do is to 8 start doing more studies actually in the Phase II 9 and the Phase III clinical trials. The techniques 10 are clearly there. There is optimal sampling 11 techniques, so we can reduce the number of blood 12 samples that have to be done. 13 There is interest in getting more frequent 14 data, so time to events. There is also statistical 15 strategies to model both clinical and microbiologic 16 outcomes. Just in talking about those two types of 17 outcomes, I said there the bacteriologic cure is 18 harder to obtain. What I meant to say is it is 19 more conservative, it usually requires more drug to 20 get a good microbiologic cure than it does to get a 21 clinical cure. 22 Again, even when you look at some of the 24 1 old data, as you see here, from Dr. Schentag's 2 group and Alan Forest, you always find that 3 clinical cure is higher than what one finds for 4 microbiologic cure. 5 In the studies that Dr. Drusano did with 6 levofloxacin, the area under the curve to the MIC 7 was about half of what it was for microbiologic 8 cure to develop a clinical cure. So, looking at 9 microbiology is actually a more conservative 10 approach. 11 [Slide.] 12 The problem with this, though, is doing 13 these kind of trials increases the complexity of 14 the trial, it also increases the cost, and at least 15 as right now, there is no established benefit with 16 regulatory agencies for doing it, so as people 17 talked about yesterday, doing a better job upfront, 18 with a small number of patients you would like to 19 help, but that would enable you to reduce the 20 number that you would like to have to use later on, 21 and maybe something like that can happen in the 22 future. 25 1 Clearly, PK/PD was used with the 2 fluoroquinolone to reduce the number of cases for 3 inclusion of penicillin-resistant pneumococci in 4 the label. 5 [Slide.] 6 Now, what do we need and what kind of 7 questions should we be asking when we are looking 8 at PK/PD? I think, first of all, you want to know 9 what indices best determines the efficacy of the 10 drug, and it really does require animal or in vitro 11 studies, because you really do need to use a whole 12 variety of different dosage regimens to reduce the 13 interdependence between the parameters, because if 14 you just use one dosage regimen, you will come up 15 with, for example, Dr. Schentag's first paper on 16 fluoroquinolones talked about time above MIC. 17 His second paper on fluoroquinolones 18 talked about area under the curve MIC. George 19 Drusano's paper on the same topic talked about peak 20 to MIC, so you can pick any one of the parameters, 21 they are all going to be correlated and you don't 22 really know which is the correct one, so it really 26 1 does require in vitro or animal studies to select 2 that out. 3 What is the magnitude of the indices 4 required for efficacy? Again, as Dr. Andes will 5 show to you later, free drug is really the 6 important thing, protein binding does have its 7 importance and does need to be considered, and the 8 other thing I think we try to do, at least at our 9 institution, in our work, is to try and link it 10 also with survival. 11 While we may be talk about a certain 12 number of organisms that we want to kill or reduce, 13 we try and also link that to some clinical outcome, 14 which is going to be survival. 15 You also want to know, since most of the 16 time neutropenic animals have to be used in order 17 to get the organism to grow, what effect does white 18 cells have on the parameters because most patients 19 that we treat are not going to be neutropenic? 20 How does the magnitude vary with different 21 organisms and especially this is the time of 22 bringing in resistant strains to see if the 27 1 magnitude varies there. 2 Does the magnitude vary with different 3 sites of infection? I think clearly, there is an 4 area where we clearly can see some differences. 5 [Slide.] 6 For example, if we just measuring serum, 7 there is probably a very good correlation with 8 interstitial fluids and with fluid collection, such 9 as sinusitis, an acute otitis media, but as we keep 10 moving down the line, it will start getting to 11 poorer and poorer correlation. 12 Clearly with ELF, I think we have good 13 data now even from human trials that there are 14 higher values for macrolides and epithelial lining 15 fluid, and decreased values for vancomycin and 16 daptomycin, and I think just recently it has been 17 found that daptomycin also binds to surfactant, 18 which is another factor that would reduce the 19 amount of drug and cause some problems for treating 20 pneumonia. 21 So, one needs to know this and just to 22 point out one potential problem, I do not know 28 1 which animal model best has ELF levels that are 2 similar to what we see in humans, and without that 3 knowledge, we are sort of advising most of the 4 companies that do those studies in humans to get 5 those values until we can eventually find an animal 6 model. 7 I will tell you right now I do not think 8 that it's the mouse because we find for macrolides, 9 the same amount of drug works in the lung as work 10 in the thigh model, so we don't see this markedly 11 elevated level that are 10 to 15 times higher in 12 human ELF fluids. 13 [Slide.] 14 Lastly, the last two questions again is 15 does the magnitude of the PK/PD is required to 16 prevent the emergence of resistance. This is 17 obviously becoming a more important question all 18 the time. Unfortunately, for some of the drugs, 19 the parameters or the magnitudes that are required 20 to prevent the emergence of resistance are so high 21 that they are never going to be reached with the 22 current doses that are being used, so combination 29 1 therapy is really what is going to be required. 2 [Slide.] 3 Lastly, one wants to know about the 4 kinetics of the drug, can, with non-toxic doses of 5 the drug in humans, reach the magnitude of the 6 PK/PD index that is required for efficacy, and also 7 for prevention of resistance with a high 8 probability. 9 It is becoming a challenge all the time 10 because marketing has gotten into making some of 11 the decisions on drugs as to how frequently they 12 can be administered, and because of that, that is 13 starting to put the challenge on PK/PD in being 14 able to come up with an adequate dose. 15 I will just give the old example of the 16 old penicillins. If you gave them four times a 17 day, you wouldn't need as much drug. On the other 18 hand, nowadays, where we are trying to give the 19 drugs once a day, at most twice a day, we have to 20 increase the doses significantly in order to be 21 able to reach the parameter that is important for 22 efficacy. 30 1 So, with that, I will stop and turn it 2 over to the next presentation. 3 DR. EDWARDS: Thank you very much, Bill. 4 Next, we will call upon Mike Dudley from 5 Diversa. 6 Mike. 7 Industry Perspective 8 DR. DUDLEY: Thank you. Good morning. I 9 would like to also thank John and John for the 10 invitation to speak to you this morning about an 11 industry perspective on this very important area. 12 [Slide.] 13 In the beginning, I want to pick up on 14 some ideas that were raised yesterday before I 15 really focus on PK/PD, and that is, that it was 16 discussed yesterday about where are the new drugs 17 going to come from and, in fact, a question was 18 asked about what is the rate of submissions of 19 INDs. 20 What I wanted to show here, although this 21 slide summarizes what happens in the discovery 22 phases for a variety of targets, I think the 31 1 experience of certainly myself and other colleagues 2 in the area would say that this is also the case 3 for anti-infectives, as well. 4 This is a slide drawn from a recent 5 survey, so it is very unscientific, but I think it 6 depicts for you what the problem is with respect to 7 drug discovery overall and particularly I think it 8 also describes a lot of experience of small 9 companies and large companies alike in trying to 10 find novel agents, as well. 11 What you see here is that when one look 12 then at the very early stages of discovery where 13 one is trying to find novel targets, and then 14 progress, though, then hits from high throughput 15 screens or other methods, then, of those hits into 16 a preclinical candidate stage, and then finally, 17 from preclinical to IND, you can see that the 18 highest dropout rate here, which range in some 19 companies between 10 to 80 percent, and again that 20 being target dependent or therapeutic area 21 dependent, is around 57 percent of that attrition 22 from taking the hit to preclinical candidate. 32 1 Overall, of course, it is 75 percent then 2 taking things from target to IND. The point is, is 3 where the problem exists is trying to find good 4 leads that are going to be drug-like, that can be 5 brought forward. 6 This is what I think is the real 7 difficulty, then, for small companies and large 8 companies alike to try to find good leads. It is 9 important and it is also risky business, and I 10 think was mentioned yesterday, is that small pharma 11 can't take this on by itself because of all the 12 risks being up here in terms of getting a 13 preclinical candidate, we need a partner to share 14 the risk for that. 15 So, because it is so difficult, then, to 16 try to find drugs in the early setting, good leads 17 to take forward into that, we really rely very 18 heavily on the notion of being able to find good 19 drugs and using PK/PD very early on as a means of 20 trying to find dose selection. 21 One can think about the idea as that one 22 is trying to find the zip line that is going to get 33 1 you, then, from the early stages of drug discovery 2 and into the clinic, and we believe that PK/PD is 3 one of those tools that can allow this to occur. 4 [Slide.] 5 So, it really starts in the beginning of 6 drug discovery. It is an integral part now in many 7 companies and the part of candidate selection in 8 the drug discovery process, and certainly 9 progresses through to the idea of selecting 10 compounds for preclinical development. 11 It enables programs to move forward and it 12 rightly oftentimes kills the drug leads, as well. 13 A critical step oftentimes in the 14 discovery process is the in vivo proof of concept. 15 As was being mentioned yesterday, if you are trying 16 to raise money or you are trying to interest then a 17 large pharma partner, everyone talks about a proof 18 of concept study, but no one every really defines 19 what the proof of concept study is. 20 If you have a PK/PD as a tool, it is that 21 tool that links the effects the one sees in vitro 22 at a given concentration of the drug, on the bug, 34 1 on the large organism itself, to the in vivo 2 exposure and the effect that happens in vivo at the 3 same concentration of drug. 4 So, a PK/PD proof of concept says that an 5 effect that is associated with a drug in vitro can 6 be translated if one gets the same concentration in 7 vivo to an effect on the microorganism in vivo. 8 Thus, as I think it was mentioned in 9 John's opening comments, it is a translational 10 science. It takes us from very early stages, then, 11 of drug discovery in the preclinical setting, and 12 then, of course, through all the phases of drug 13 research, but I think also particularly the 14 opportunity in Phase IIA where one can identify 15 these relationships for effects and validate and 16 refine it all the way through the clinical 17 development process. 18 [Slide.] 19 Now, one example then of how you can do 20 this, and as Bill mentioned to you, is the use of 21 in vitro models of infection where one can, in 22 fact, start to study these issues before we even 35 1 know what the pharmacokinetic properties may 2 actually be in an in vivo system, where one can 3 then expose growing cultures of an organism, either 4 a bacteria or a virus, to changing concentrations 5 of drugs. 6 7 [Slide.] 8 This is an example for a novel agent, in 9 fact, before it has actually gone into man, where 10 one can look at the effects here against MRSA, 11 using a predictive pharmacokinetic profile based 12 upon preclinical animal species and then one can 13 begin to get insights in terms of both dose and 14 dose frequency that is required then to get an 15 antibacterial effect against target organisms. 16 [Slide.] 17 So what does drug industry then view as an 18 important use of PK/PD in streamlining drug 19 development? Well, it goes without saying for 20 dosage regimens for clinical development, and the 21 subsequent speakers are going to focus upon that, 22 as well, Dr. Craig has spoken about in vitro 36 1 susceptibility in resistance breakpoints. 2 This is, in fact, a very transparent 3 process where one can then use scientific data and 4 common criteria for then an effect that helps us 5 then to determine how clinicians can use these 6 drugs. Breakpoints are used in the clinic to help 7 define, then, what drugs are going to be used in an 8 individual patient. 9 One thing that I think that is needed is 10 what about what I will call the care and feeding of 11 these breakpoints, what happens then as resistance 12 changes or as new data become available for old 13 drugs about the existing breakpoints that are in 14 the labeling? 15 Presently, right now that task is taken up 16 only by the NCCLS, which strives to harmonize what 17 is happening in the regulatory environment, as well 18 as within the clinic. 19 Labeling for resistant organisms, which is 20 what Dr. Craig talked about, as well, and we will 21 hear more about that today, about PK/PD exposures 22 that may be relevant from animal models, and we can 37 1 link that to human pharmacokinetics, and then, of 2 course, I think, which was brought up in one of the 3 other workshops before, is the idea of being able 4 to provide PK/PD parameters or indices for 5 organisms, such as the AUC to MIC, even though a 6 full indication or a full, well-controlled clinical 7 trial has not been made available with the proviso, 8 of course, that these observations may not have 9 been validated in clinical studies in patients. 10 [Slide.] 11 What I think is clear, though, is that 12 these can be used in very, very useful 13 relationships, and I have drawn from one example 14 from Dr. Ambrose and colleagues where he went and 15 pulled out information for pneumococci across 16 several clinical trials for fluoroquinolones, and 17 what you see is a very, very consistent picture in 18 terms of free drug AUC-to-MIC ratio and the 19 probability of eradication here in patients with 20 lower respiratory tract infections involving 21 Streptococcus pneumoniae. 22 So, these relationships do work within the 38 1 clinic, and they can be used then to guide the 2 development process, as well. 3 [Slide.] 4 What about pharmacokinetics? I think as 5 Dr. Lazor and Dr. Craig mentioned already, as well, 6 is that we now have techniques for getting 7 pharmacokinetics in clinical trials. This should 8 not be an excuse for not doing the proper 9 experiments, so as one Bush's once said, "Read my 10 lips," that we should be able to be able to do this 11 through the techniques of sparse sampling, 12 population pharmacokinetics and Monte Carlo 13 simulation, and that can be done, not only taking 14 into account, then, the concentrations in the 15 dosing or serum compartments, but it also can be 16 taken into account in specialized tissues as has 17 been recently shown by the Albany Group, such as 18 modeling the prostate. 19 [Slide.] 20 What about PK/PD in the response or the 21 endpoint? First, I think it is very important 22 especially in light of the discussion yesterday to 39 1 remember that these are definitely not surrogate 2 markers. Although this has oftentimes been used 3 erroneously in the literature, PK/PD parameters or 4 indices are not surrogate markers. They are at 5 least maybe two steps removed from a surrogate 6 marker based on the discussion yesterday. 7 But the analyses that are generated from 8 this can help us to understand how to get to those 9 endpoints, as well. We need consensus on those 10 relevant clinical endpoints and those markers to be 11 able to really move the science forward. 12 One issue may be, in fact, using validated 13 composite endpoints in the clinic. We know, for 14 example, now that there are these endpoints that 15 are used for making treatment and hospitalization 16 decisions within patients, so what about using, 17 then, composite endpoints for these patients? 18 [Slide.] 19 Finally, what about, then, getting more 20 information from smaller trials or more focused 21 studies, which I think were some of the themes that 22 have been brought up already here, and this is just 40 1 one example, which is really the same as the 2 oseltamivir studies that were described yesterday 3 where it may be, in fact, the speed of response 4 that one can see by taking serial measurements that 5 may distinguish, then, both the dose, as well as 6 the type of therapy that are used within individual 7 patients. 8 So, by getting information earlier and 9 sequentially within individual patients, we may be 10 able to define differences that are real and 11 important between both dosage regimens, as well as 12 agents in there, as well. 13 That can have an enormous impact, then, 14 upon the number of patients that may be required 15 for us to be able to detect differences in a 16 clinical trial. From a paper that will be 17 published short in Clinical Infectious Disease, one 18 can see that, in fact, that for the sample sizes 19 that may be required for looking for a 20 time-to-event analysis, here being negative sinus 21 cultures in patients who are then having serial 22 measurement for recovery of bacteria in sinus 41 1 aspirates, one can certainly see that one can, 2 using hours or time-to-event, one can have 3 meaningful data in as few as 26 to 50 patients. 4 [Slide.] 5 Finally, what are, then, some of the 6 provisions then that we can use for early market 7 entry of new drugs or for infections due to 8 priority-resistant organisms, which I think was 9 certainly the context of the discussion yesterday, 10 and I think that one thing that we would like to 11 see is what can we build on these CFR fast-track 12 provisions, can we use that now against the target 13 pathogens on the priority lists that were discussed 14 last year where one has full delineation of these 15 relationships in animal and in vitro models of 16 infection, then, using accepted endpoints or 17 surrogate markers from well-designed and executed 18 Phase II studies, then, to demonstrate then that we 19 have efficacy at these target exposures, we may 20 need to include comparators in there to ensure that 21 we have got sensitivity, as well as the comparator 22 regimens are optimized, and then to make agents 42 1 then available on a limited basis, much as what was 2 done in the nineties with the HIV agents, and then 3 build in the post-marketing phase, then, trials 4 that really continue to build on this PK/PD zip 5 line, if you will, and then also expanded then to 6 include both susceptible organisms and resistance. 7 I think it is important, of course, that 8 safety does need to be demonstrated, and it does 9 need to be demonstrated in comparative trials that 10 are going to need to be taking place, but it all 11 comes down to risk management. It all comes down to 12 whether or not the risk you are willing to take 13 with respect to the resistance that's at hand. 14 [Slide.] 15 So, to summarize, much is known about 16 PK/PD of drugs very, very early on and prior to 17 entry of man. It isn't that we go into man and 18 then try to figure this out, but it is oftentimes 19 bred into the drugs that are moving forward. 20 Streamlined evaluation, I think, of 21 efficacy, as you will see, can be obtained from 22 data-rich PK/PD Phase II trials, and then safety, 43 1 of course, is important and will ultimately, 2 though, need to be determined in the comparative 3 trials. 4 Thank you. 5 DR. EDWARDS: Thank you very much, Mike. 6 I will now call on John Powers for the FDA 7 Perspective on PK/PD issues. 8 FDA Perspective 9 DR. POWERS: Thanks, Jack. 10 [Slide.] 11 What I would like to do is to give a 12 little background on this information today, and 13 one of the main messages that I want to get across 14 is that we do feel that this information is useful. 15 I know in talking with Dr. Craig a couple 16 of times before, he has told me about uncertainty 17 about does the Agency find this information useful, 18 and we definitely want to get across that we do, 19 but then to discuss some of the potential strengths 20 and limitations of PK/PD and the overall drug 21 development program, which I think the previous two 22 speakers have already touched upon, and then talk 44 1 about some of these applications in clinical 2 trials. 3 What is PK/PD actually going to do for us 4 in shrinking the overall size of the clinical 5 development program or in being able to shrink the 6 overall size of an individual trial, and then talk 7 about some of the applications for prescription 8 drug labeling or potential applications. 9 [Slide.] 10 So, we had previous discussions at this 11 meeting in November of 2002, and also at various 12 advisory committees about what is the role of PK/PD 13 in clinical development programs, and then the IDSA 14 sent a letter to Commissioner McClellan in November 15 of 2003, and one of the suggestions on that letter 16 was for FDA to find ways to incorporate PK/PD to 17 shrink the size of clinical trials. 18 So, again, we do find that this can be 19 useful and, in fact, this is an integral part of 20 the FDA's Critical Path Initiative is using PK/PD 21 as one of the development tools. 22 So, what are some of the things that PK/PD 45 1 can do for you? Well, let's look at it from the 2 other point of view. If you don't use it, and you 3 select the wrong dose, you can end up actually 4 having a bigger clinical trials' database because 5 the cure rate comes out lower, or even worse, your 6 drug comes out ineffective in the clinical trials. 7 That gets to the issue that John Lazor 8 brought up, if your drug doesn't work, then, you 9 have got to go back to square one, and that results 10 in a multi-cycle review and that takes you a longer 11 time to get your drug approved. 12 I think one of the things, when we talk 13 about clinical approval times that gets lost in 14 that discussion, is two things can happen that can 15 result in a multi-cycle review--well, three things. 16 One, we have seen things that are just 17 sort of technical problems in that the submission 18 that comes in, we can't evaluate because it doesn't 19 work in the computer or something. 20 The second one is that the drug has 21 efficacy issues, and we need to go back and study 22 it more thoroughly, or the third one is that a 46 1 safety signal pops up that requires further 2 exploration. 3 The other thing is selection or the 4 inappropriate dose, and I think we can't forget 5 about this one. It may impact your development 6 program, but it is going to impact patients, too. 7 We don't want to be selecting the wrong dose and 8 have more people be failing from those diseases. 9 So, again, failure to show efficacy may 10 require further trials, but even if your drug does 11 look better than placebo, which is the regulatory 12 hurdle for approval, coming out with a lower 13 success rate than your competitor doesn't help you 14 in the marketplace either, so picking the proper 15 dose to get the highest success rate will actually 16 help your drug overall. 17 Then, of course, picking the proper dose 18 may limit dose-related adverse effects, as well, 19 and it may give some clues--and we talked about 20 this yesterday, I kind of tacked this on at the end 21 this morning when we were talking about the 22 endocarditis discussion--it may give you some clues 47 1 as to which indications you should study and which 2 indications you should avoid. 3 If you do some preclinical work and it 4 looks like your drug isn't so good for Pseudomonas, 5 hospital-acquired pneumonia is probably not 6 something you should go after. On the other hand, 7 if it looks like you are good against something 8 like E. coli, urinary tract infections, et cetera, 9 it may be where you want to go. 10 [Slide.] 11 So, can PK/PD shrink the size of 12 individual trials? Well, if PK/PD, optimizing the 13 dose results in a higher success rate for your drug 14 in the trial, the answer to this question is yes. 15 [Slide.] 16 And I showed this slide yesterday again. 17 So, if you can just increase the success rate by 10 18 percent in your clinical trials, you can shrink the 19 clinical trials' database from 252 patients per arm 20 with a 10 percent non-inferiority margin to 142 21 patients per arm. 22 So, yes, it can help as long as you can 48 1 have some reliable impact on the success rate, 2 clinical success rate in that trial. 3 [Slide.] 4 But the question then comes up is PK/PD 5 sufficiently accurate to predict relatively small 6 differences in success rates between drugs, is 7 PK/PD best at predicting which drugs will be 8 effective and which drugs will be ineffective 9 rather than selecting differences between effective 10 drugs? 11 I remember somebody from a pharmaceutical 12 company was sitting next to me once and said if 13 drug X is so bad, how come we can't beat it. So, 14 their drug being better on the PK/PD parameters, 15 yet, when you do the clinical trials, the drugs 16 come out equivalent to each other. 17 Now, is this just because you need to do a 18 10,000 patient trial to show those small 19 differences? Then, you have got to ask yourself 20 the question, if you have got to do a trial that 21 big, are those differences clinically relevant at 22 that point. 49 1 So, the reasons why may be do hosts and 2 other effects predominate in affecting the clinical 3 outcomes and those differences in the microbiologic 4 effects get lost in the wash there. 5 [Slide.] 6 So, what are some of the other potential 7 uses? The previous speakers have touched upon 8 this. They can come up with preliminary 9 information to come up with hypotheses for 10 potential susceptibility breakpoints. So, I used 11 the word "preliminary" and the word "hypothesis" in 12 the same sentence. I have to correct Dr. Craig. 13 We have used this. Al Sheldon, who is sitting out 14 here, worked tremendously on this--well, we do look 15 at them. 16 Mike Dudley and I talked about this at 17 ICAAC last year. It gives us a hint that the drug 18 ought to work, might work, and should work, but the 19 level that we need to actually put this in a drug 20 label is proven safe and effective. 21 So, Mike and talked about ceftriaxone 22 should work at an MIC of 16 for Strep pneumo, but 50 1 when we go and look at the databases, we don't see 2 any organisms with an MIC that high, and then the 3 question is--and John Bradley has brought this 4 up--what do we use these breakpoints for? I think 5 there is a big distinction between what the 6 Europeans use them for and what the Americans have 7 used them for in the past. 8 Are we just trying to separate out two 9 populations, or are we trying to describe for 10 clinicians which drug may be effective in what 11 situation? I would argue as a clinician it is the 12 second one, and we know that clinicians use these 13 drugs to say, well, if I got the big R on that lab 14 sheet coming back, I am not going to use it. So, 15 it really has clinical implications for people. 16 Now, after having said all that, we don't 17 want to talk about that today, because that is 18 going to require a big discussion. We do think 19 this is really important, but there is a lot of 20 stakeholders in this--and I never realized this 21 before either--not only is there the NCCLS, there 22 is the device manufacturers who put together the 51 1 plates that actually get tested, and we need to get 2 all those stakeholders together and talk about that 3 at some time in the future. 4 The next issue is the potential to prevent 5 the development of resistance, but again we have to 6 ask the question of what is the clinical effect of 7 preventing development of resistance. 8 There was a study back in the late 9 eighties, I think, of ciprofloxacin versus imipenem 10 in hospital-acquired pneumonia, where they showed 11 that ciprofloxacin selective for fewer resistant 12 pseudomonas in the sputum, but only one person got 13 sick. But that does mean that that is not an 14 impact? 15 Well, there is the question does it impact 16 that person? Does it impact other patients? What 17 we would need as an agency is that clinical data to 18 show what does the prevention of resistance 19 actually translate to in the clinical setting. 20 Again, this is something we touched upon 21 yesterday, and Dr. Ross brought this up, drug 22 labeling for antimicrobials is either you are 52 1 treating a disease or preventing a disease, and we 2 need that clinical information to show that. 3 The other issue when you are talking about 4 resistance is what do we care about here, so you 5 can prevent resistance in pseudomonas when the 6 person has a pseudomonal infection, but what 7 happens to what I refer to as collateral organisms. 8 I think about the governor of California's movie 9 "Collateral Damage." 10 So, what happens if I prevent the 11 resistance to pseudomonas, do I then get resistance 12 in some other organism as well because my drug 13 isn't as good against the gram-positives, and now I 14 select out resistance to Streptococcus pneumoniae 15 or other commensal flora. 16 So, when we talk about developing 17 resistance, we need to look at that information. 18 The idea of combination therapy certainly has a lot 19 of play in the antifungal world at this point in 20 time, and some of the discussions that have come up 21 there is, well, if you use combination therapy at 22 that point to try to increase efficacy to prevent 53 1 resistance, what happens on the toxicity side, are 2 you going to cause more toxicity, and that balance, 3 we actually need the clinical data to actually 4 show. 5 [Slide.] 6 What are some of other limitations here? 7 Well, typically, PK/PD and anti-infective drug 8 development has focused on the effects on 9 microbiological outcomes, and Mike Dudley already 10 mentioned this, that in terms of a surrogate 11 marker, we are a couple of steps removed from that 12 even. 13 So, in many situations, as we discussed 14 yesterday, the validity of that microbiologic 15 outcome as a surrogate for clinical outcomes 16 remains unclear. I like Mike's example of Dr. 17 Ambrose's study in sinusitis. 18 We think that is really important and 19 using that in a Phase II proof of principle is 20 great, but when we talked about this, is that going 21 to garner you an approval all by itself, and the 22 answer is no, because at this point, we don't what 54 1 eradicating the organism means in terms of the 2 clinical outcomes in acute bacterial sinusitis. 3 So, the other issue I think that is really 4 important, that folks who don't work at the Agency 5 don't realize, is a lot of times in these diseases, 6 the microbiological outcomes are imputed from the 7 clinical outcomes, so they look the same because 8 they are the same. 9 So, you have a person who has 10 Streptococcus pneumoniae in their sputum at 11 baseline in a pneumonia trial. They come back 12 after 10 days of treatment, on day 14, and they are 13 clinically well, they are not coughing up any 14 sputum, and they feel fine. 15 The person checks off the box they are 16 cured, and that goes down as "Presumed 17 microbiological eradication" when we don't have 18 that information. Now, we can't get it obviously, 19 the person is not making any sputum, but when 20 people then come to us and say, well, there is this 21 great correlation between micro and clinical 22 outcomes, it is because you didn't have any micro 55 1 data at the end anyway. 2 So, where we would really like to see that 3 is places where we can get that information. Some 4 of the information I showed yesterday says you 5 can't do this in some diseases like otitis media, 6 where over 60 percent of the kids with these 7 presumed eradicated, when actually the double tap 8 studies are done, the bug is still there, so it 9 turns out that "presumed" probably is incorrect in 10 some situations. 11 Obviously, we need to know if the effect 12 on the organism translates into clinical outcomes. 13 [Slide.] 14 So, again, we have this question of can 15 PK/PD differentiate. It looks like it can 16 differentiate ineffective drugs or doses, but can 17 it differentiate between effective drugs or doses, 18 and Dr. Craig brought this up, this issue of 19 getting on formulary. 20 So, how does a person on a P & T Committee 21 look at this information and say, well, drug X has 22 an 85 percent cure rate in community-acquired 56 1 pneumonia and drug Y has 84 percent cure rate? 2 They look pretty much the same to me, but this 3 guy's PK/PD looks better than that guy's. What 4 does that mean to me when I am going to decide 5 about putting this drug on formulary? Again, are 6 there other factors that are more important? 7 For instance, the mortality in severe 8 community-acquired pneumonia remains at 30 percent 9 despite the introduction of more active drugs in 10 vitro. 11 [Slide.] 12 So, again, some of the other issues that 13 may come up here that may dissociate the 14 microbiological and clinical outcomes are things 15 like pH at the site of infection, and one of the 16 issues I hope we really touch upon today is what 17 endpoint do we want to use in some of these PK/PD 18 studies, is it static growth, is it 1 log decrease, 19 is it 2 log decrease, what should we be using as 20 that target. 21 We talked a lot yesterday about direct 22 immunologic effects of the drug on the host, and 57 1 immunologic effects on the host by the organisms. 2 The other issue is does PK/PD give us 3 enough information on non-dose related adverse 4 effects, and all of this is just a prelude to 5 saying that we still need clinical trials to 6 determine the effects of the drug on clinical 7 outcomes in a given disease entity. 8 [Slide.] 9 And why am I bringing this up? Because 10 several pharmaceutical companies have come to us 11 with the suggestion that they should receive what 12 they have termed "follow-on indications" based on 13 PK/PD data alone. So, what we presume they mean by 14 this is we go out and we do a community-acquired 15 pneumonia trial and then the FDA should grant us 16 indications for sinusitis, otitis, and AECB based 17 on our PK/PD information. 18 Again, as Mike has pointed out, we can't 19 even use these as surrogates yet at this point, so 20 the point we are trying to make is we do find PK/PD 21 useful, but we actually still need that clinical 22 information in those trials, and we still need 58 1 clinical information from patients infected with 2 resistant bacteria to be able to make that 3 decision. 4 The other thing is we have clearly seen 5 differences across the safety and efficacy of drugs 6 in various diseases. Now, is this related to the 7 population? Maybe. For instance, it is 8 fascinating to look at the indication of acute 9 bacterial sinusitis. 10 Acute bacterial sinusitis, when you just 11 look at the spread, is usually younger, healthier 12 women, and we have seen a number of adverse events 13 with various drugs pop up in that patient 14 population more commonly. Is that related to the 15 disease or is it related to the host? 16 Be that as it may, we definitely see 17 different side effect profiles across different 18 drug indications. 19 [Slide.] 20 So, the issue here is we still need 21 clinical data from each indication, and at the 22 March 2003 Anti-Infective Drugs Advisory Committee, 59 1 we did talk about this issue that Dr. Cox brought 2 up yesterday, of clinical data from one indication 3 supporting another, and Dr. Talbot brought up this 4 idea of what does supportive actually mean. 5 Well, supportive presumes there is still 6 at least one trial in each indication since you 7 have to have something there to support, and this 8 also goes for the idea of resistant pathogens. We 9 have had several cases recently where folks came to 10 us and said, well, here is my MRSAs in 11 hospital-acquired pneumonia, and you should just 12 give us an indication for MRSA community-acquired 13 pneumonia, and then the question we ask is we still 14 need to see that, first of all, that organism is 15 relevant in that disease, so we want to see some 16 cases, and we have asked for as little as 10 cases 17 there. 18 Then, the other issue is we do need to see 19 there are differences across those diseases and 20 across those hosts, how the drugs actually work 21 there. 22 [Slide.] 60 1 So, what about the issue of prescription 2 drug labeling? Dr. David Gilbert, who couldn't be 3 with us this time, brought this up at the last 4 November 2002 meeting, that the FDA should put 5 PK/PD information in labeling. 6 So, we went back and we thought about that 7 some more, and the first question that came up was 8 what information should we put in labeling, and 9 then the second came up, why should we put it in 10 labeling. 11 So, first of all, clinicians don't have 12 the information needed to make these PK/PD 13 assessments in a lot of cases. Cultures aren't 14 commonly done in some diseases like uncomplicated 15 UTI or even when clinicians make their best efforts 16 like community-acquired pneumonia, we can't find 17 the organism in about 50 percent of cases. 18 Also, drug concentrations are rarely 19 available for the clinician to make these kinds of 20 decisions although we can make some guesses based 21 on modeling, but which concentration is relevant, 22 is it the concentration in the blood or the 61 1 concentration at the site of infection, which the 2 clinician will almost never have in making that 3 decision. 4 Then, one of the other issues we have 5 really hoped to get at today is, is there this "one 6 size fits all" PK/PD parameter. So, if I hit an 7 AUC MIC over 100, does that just fit every 8 gram-negative organism for every disease 9 indication, or are there some differences across 10 there? 11 I can tell you that is what clinicians 12 think. George brought up when I was the University 13 of Maryland, and I remember standing there with a 14 fellow attending who went to prescribe a quinolone 15 at 2,000 milligrams for a person with Strep pneumo 16 bacteremia, and said, oh, but it works better 17 because it is concentration-dependent, so we will 18 just jack up the dose. 19 [Slide.] 20 So, that gets us to a really important 21 piece for us, as the FDA, is would PK/PD 22 information in labeling imply a superiority claim 62 1 for one drug over another that has not been 2 demonstrated in the clinical trial, and would PK/PD 3 information spur clinicians to use a higher 4 unstudied dose that may not be as safe in hopes of 5 improved efficacy based on looking at that 6 information in a label. 7 The final thing for us, that question that 8 came to our minds is how does this information 9 actually help practicing clinicians to prescribe 10 the drug appropriately in their patients, which is 11 what our goal is when we put something into the 12 label in the first place. 13 [Slide.] 14 So, our discussions today would focus on 15 dose selection in clinical trials because we all 16 agree that that is the place where we can really 17 use this to streamline the development process, but 18 this requires a discussion amongst all the parties 19 here today, is what constitutes an adequate PK/PD 20 database for a drug development program. 21 We still do need to have discussions in 22 the future about what are some of these issues 63 1 related to selecting breakpoints, and we do want to 2 do that in the future, but that is going to require 3 getting all the parties together and we will 4 discuss that at a point in the future. 5 So, I will stop there. Thanks very much. 6 DR. EDWARDS: Thank you very much, John. 7 I am just going to make one request to the 8 future speakers, and that is, if we could just 9 please not refer to the governor of California. 10 This is a very sensitive issue, and I will allow 11 you, John, but that's it for today. 12 We actually have very few minutes for 13 discussion of this topic at this particular 14 interval. 15 Would someone like to begin? George. 16 Discussion 17 DR. DRUSANO: The single most important 18 thing is the idea needs to be understood that PK/PD 19 targets are fully stochastic, so that when somebody 20 says it's an AUC to MIC of, or time above MIC of, 21 fill in the blank, that that is a point estimate 22 with a 95 percent confidence interval about it, and 64 1 one size does not fit all. 2 One need only look at pneumococcus 3 relative to gram-negative organisms, and not all 4 gram-negative organisms are the same. Now, we can 5 make judgments, we can make conservative judgments 6 because if you pick the one that is the highest, 7 then, you will pick up all the ones that are the 8 lower, that are lower than that value. 9 But I think if we have a further 10 discussion on this area today or on some other day, 11 one of the real critical pieces about the use of 12 PK/PD is to prevent its misuse, which is prevalent 13 even now in terms of how this is getting used. 14 The other issue that you really have to 15 get to is what do the physicians do with it. Most 16 physicians really could care less about PK or PD 17 except--I think all physicians are the same in one 18 respect--they want their patients to get the best 19 available therapy. 20 So, one of the ways to do that is to, in 21 the labeling practice, push it back a notch. We 22 have a tool, the Monte Carlo simulation and then 65 1 expectation over MIC distributions. You can use 2 that tool to back the PK/PD target information back 3 into the labeled doses. At this labeled dose, you 4 can expect hitting this particular target a certain 5 fraction of the time. 6 You can have warnings that this does not 7 necessarily impute that you are going to have a 8 good clinical outcome. It does impute that you are 9 going to have this kind of microbiological effect 10 and that there is uncertainty about it. 11 Those are the kinds of things I think that 12 you can do to roll it back a notch into a language 13 that a good clinician is familiar with and can 14 apply to his or her patient without the danger of 15 saying I'm going to give 2 grams. 16 DR. CRAIG: I feel very strongly the same 17 thing. We use PK/PD at our hospital as a guideline 18 for when combination therapy should be used. 19 Again, we have already done the Monte Carlo and 20 have that data, so we tell our clinicians if they 21 are using a fluoroquinolone and the MIC is below a 22 certain level, monotherapy would be okay, but if 66 1 the MIC gets high enough, you are not going to get 2 an adequate PK/PD and drug combination should be 3 used. 4 So, there is information that you can 5 glean from PK/PD that can be used clinically by 6 clinicians. 7 DR. EDWARDS: John. 8 DR. LAZOR: George, I really appreciate 9 your comment with respect to the PK/PD target, and 10 it not being a single number. Unfortunately, that 11 is all we hear. We always hear the single point 12 determination, and we never know what the 13 variability is around that target. 14 Also, on your point with respect to 15 labeling, even though we have put that information 16 in the label, for instance, you know, you have X 17 probability of achieving a target, I still don't 18 understand what a physician would do with that. 19 I can imagine that company A would have X 20 probability, company B would have Y probability, 21 and then somebody is going to interpret that as Y 22 being better than X, of superiority. That is one 67 1 of the things that needs discussion, because that 2 is one of the things that you don't want to happen. 3 DR. DRUSANO: I have got 84, and he only 4 has 82. I mean that shouldn't happen. That is not 5 what it was supposed to do, and the problem comes 6 is that, you know, we need to do the math a little 7 better, so that we can actually get, not only a 8 point estimate of what that target should be for a 9 specific endpoint, but also its 95 percent 10 confidence. 11 Oftentimes we do this by classification 12 and regression tree analysis, and because it is a 13 recursive partitioning algorithm, it just chops 14 things up, and you don't know whether it's here to 15 here, or right in the middle, or anywhere in 16 between. All you know it is around here, and that 17 is a message that we have been terribly remiss at 18 getting out. 19 DR. EDWARDS: Dr. Rex. 20 DR. REX: I have a very small comment for 21 Bill Craig, a different direction. You put on 22 several slides, the blanket statement that it is 68 1 always free drug. That may not always be true. 2 DR. CRAIG: I would agree there are some 3 situations where a drug may still have activity 4 when bound to albumin, and those are going to be 5 primarily sites where the drug does not have to get 6 inside the cell, but I would say if the drug has to 7 get inside the cell in order to reach its target, 8 that free drug would be the important determinant 9 of efficacy. 10 DR. REX: Well, I think that you need to 11 entertain the possibility that it is not always 12 albumin that things are bound to. Things do pop 13 off and on of proteins, and it may be that the 14 available drug at the site of action is differently 15 measured. 16 The point I want to make is that you need 17 to do a test in which you decide, you know, most 18 crudely, is the MIC affected by proteins. 19 DR. CRAIG: That is what we do and, for 20 example, I can tell you for a membrane drug like 21 daptomycin, for a membrane drug like amphotericin 22 B, protein binding is not as effective in reducing 69 1 activity, but, as I said, those drugs do not have 2 to get inside the cell for their activity. 3 For drugs that do have to get inside the 4 cell for their activity, protein binding is 5 important. 6 DR. DRUSANO: If I could just amplify on 7 that just a little bit, the other issue where it 8 isn't quite mathematical in that way, is when you 9 have receptor that the drug has to bind to, that 10 has about the same KD for the drug as it does for 11 its binding site. 12 I don't care whether it's alpha-1 13 glycoprotein or albumin, or whatever it is bound 14 to, the closer the KDs get, the more it is a crap 15 shoot as to whether the drug is going to go that 16 way or that way. 17 Having said so, I think it is not always 18 free drug, but if you look at 95 to 99 percent of 19 the instances, it is free drug, and we define the 20 likely effect by virtue of free drug. There are 21 exceptions, but you just have to be cognizant of 22 where those exceptions are, but the general 70 1 principle remains true. 2 DR. EDWARDS: Jerry Schentag. 3 DR. SCHENTAG: Thank you. I have to, of 4 course, say something about those target things 5 since everybody looks at me every time every says 6 100, and I, first of all, want to say that I 7 believe also that we should avoid abuse of this. 8 I don't mean by that what some of you may 9 have thought that that means George shouldn't use 10 this technology. What I mean is we need to--and 11 John's challenge is appropriate to all of us--we 12 need to link the PK/PD marker to both clinical and 13 micro outcome, and not issue a value unless we have 14 done that, and we need to do that in patients. 15 I think it is really where we get into the 16 most trouble in those of us that do it in the lab 17 or in animals, we don't always have that clinical 18 outcome linked to it. I come from a field where 19 either you kill the bug or the bug kills you, so I 20 always thought that there was a 1 to 1 link between 21 killing the organism and your target. 22 I think that just because naively, I go 71 1 along trying to deal with nosocomial pneumonia on a 2 regular basis, or bacteremia, and finding out that 3 it works, and it is pretty much always the same 4 number when I do that. 5 That is probably wrong of me to have made 6 that statement without more effective validation, 7 and I stand in front of all of you and say that I 8 think a commitment that we all ought to try do, and 9 I am going to try to do that very soon, is to try 10 to create a validation model for the PK/PD target 11 that we use, and bring you that data. 12 I would like to argue that that is 13 something ISAP and IDSA could very well do together 14 and involve the FDA in that process, too. There is 15 data that we could work together on to do that, and 16 I would be pleased to help with that. 17 DR. EDWARDS: Thank you for those comments 18 and it leads me to do something I am going to have 19 to do here. We are really at the end of the time 20 we have available for discussing the details of 21 this, and there are many people who have comments 22 they would like to make, and I apologize right now 72 1 for having to move on. 2 But I just wanted to call on George Talbot 3 to ask a question he and I both have regarding the 4 big picture within this area. 5 George. 6 DR. TALBOT: Thanks, Jack. I think that 7 my comment is a natural follow-on from what Jerry 8 was mentioning. I was sitting here listening, with 9 great interest actually, to the discussions about 10 protein binding, and so forth, but my eye fell on 11 the title of the session, which was Overview of Use 12 of PK/PD in Streamlining Drug Development, and I 13 guess the question or challenge I would pose is 14 given that we are looking for ways to move forward 15 to address this question, could we come up maybe 16 with written comments to the Agency with two or 17 three consensus action items, such as perhaps Jerry 18 mentioned, for moving forward to determine how we 19 could, in fact, streamline drug development using 20 this other than just some of the points that John 21 mentioned. 22 What I am asking for is a focus on 73 1 answering the question and coming up with 2 constructive ideas. Perhaps FDA could suggest a 3 format that would be useful to them. 4 DR. POWERS: I think one of the reasons we 5 are doing the rest of this today is to try to get 6 that information to come up with that. I think, 7 George, we keep that in mind while we are doing the 8 rest of today's discussions, that that would be 9 very useful. 10 DR. TALBOT: I guess what I see is, 11 though, dose selection, dose selection, dose 12 selection, and yes, I had hoped we could think a 13 little bit outside the box, understanding very well 14 your point that you need clinical efficacy and 15 safety data to approve and label a drug, but the 16 questions might come up, for example, as to 17 whether--and I think you alluded to this--the 18 number of clinical trials could be reduced. You 19 still would need clinical trials. 20 So, again, rather than starting the day 21 with a conclusion that things can't happen, I would 22 just urge the group to focus on how could we make 74 1 something happen here. That is the philosophical 2 point I had. 3 DR. DRUSANO: Could I ask John a quick 4 question? This will be very quick. 5 John, if you did a Phase II PK/PD bridging 6 study in which you had two or three indications, 7 and then you picked your right dose and you did 8 your large traditional Phase III in each of those 9 areas, could you use the PK/PD Phase II bridging 10 trial as supportive information to hang up the 11 tent, if you will, for one adequate and 12 well-controlled trial per indication? 13 DR. POWERS: We have done that before and 14 that is the suggestion we are trying to get to. 15 Not only that, if you are studying various 16 indications, those Phase III trials support each 17 other, as well, so the whole thing hangs together. 18 But I think one of the issues that we would like to 19 get to today was the end of yesterday's discussion. 20 If you are company that wants to go study 21 endocarditis, what kind of Phase II trial can you 22 do that would make people feel comfortable to go 75 1 forward there. One of the things, that maybe Dr. 2 Stanski can comment on that, he has been 3 instrumental in helping develop this critical path, 4 is the idea that what we see is a lot of companies 5 skipping over Phase II, going from Phase I, and 6 just going right to the Phase III. 7 Then, when it doesn't work, what do you 8 do? You are stuck, and you come back and say 9 doggone it, the FDA is taking so long to approve 10 our drug, because you had to go back and do the 11 trials all over again. 12 So, I think that is what we are trying to 13 do, we look at this--I think Dr. Craig sort of 14 mentioned it--it is almost like an investment 15 upfront, to do the right thing, so that when you 16 get to Phase III, that you don't have to go back to 17 square one and start over again. 18 DR. DRUSANO: And if you choose the right 19 dose, and you have the highest possible clinical 20 response rate, as you showed, the numbers of 21 patients get smaller. 22 DR. EDWARDS: We will move on then and we 76 1 are now going to enter the area of the discussions 2 on animal models to support dose selection. 3 I will start with David Andes from the 4 University of Wisconsin, who will begin the 5 discussion from the academic perspective. 6 David. 7 II. In Vitro/Animal Models to Support Dose 8 Selection - Academic Perspective 9 DR. ANDES: I would like to first start by 10 thanking the organizers for the invitation to speak 11 today. 12 [Slide.] 13 What I will discuss this morning is how 14 one can begin to use, looking at the relationship 15 between a measure of drug exposure in animals, 16 pharmacokinetics in animals, looking at the 17 relationship between that, a measure of potency in 18 vivo, in vitro, the MIC, and a variety of outcomes 19 to aid in dose selection for clinical trials. 20 [Slide.] 21 As Bill Craig mentioned, there really are 22 four primary questions that animal models can help 77 1 to address, that can help in dose selection. 2 First, what is the pharmacokinetic 3 parameter that drives efficacy, what PK 4 characteristic do I need to optimize? 5 From the standpoint, then, of dose 6 selection, what PK/PD target or what magnitude of 7 this parameter drives efficacy, how much drug do I 8 need? 9 What I will spend most of this morning on 10 is discussing the variables that one might consider 11 that may impact the amount of drug or the magnitude 12 of the pharmacodynamic parameter that drives 13 efficacy or leads one to some outcome. 14 Then, most importantly, does any of this 15 matter, are the predictions from animal 16 pharmacodynamic studies predictive of what one can 17 see in clinical trials? 18 [Slide.] 19 Although I will not spend any time at all 20 talking about how one addresses the first question, 21 certainly, animal models have been critical because 22 of their ability to look at a wide variety of dose 78 1 levels and dose fractionation schedules in 2 determining which pharmacodynamic parameter best 3 drives efficacy. 4 Here is one example looking at therapy 5 with the beta-lactam ceftazidime in a Pseudomonas 6 pharmacodynamic model, here the thigh-infection 7 model, and one can clearly see that when looking a 8 wide variety of dose levels and dose 9 fractionations, that here, as we all know, time 10 above MIC is the pharmacodynamic parameter that 11 drives efficacy with the beta-lactams. 12 [Slide.] 13 But what I will spend most of this morning 14 talking about is again what magnitude of that 15 parameter or what pharmacodynamic target is one 16 looking to achieve. 17 [Slide.] 18 Most importantly, what variable should be 19 considered in looking or defining the magnitude of 20 pharmacodynamic parameter that leads to efficacy, 21 and these can include, although this list is not 22 all-inclusive, do drugs within the same class 79 1 require the same pharmacodynamic target? 2 Does the dosing regimen that one uses 3 impact the amount of drug you need? As we have 4 already begun to discuss, does protein binding 5 matter, should we consider free drug levels or 6 total drug levels, does that impact the amount of 7 drug you need? 8 Does the site of infection or the animal 9 model you use give you a different answer when you 10 are looking at pharmacodynamic targets? 11 Is the pharmacodynamic the target for all 12 organisms, or does it vary from species to species, 13 and within a species, does it vary when you are 14 looking at different resistance mechanisms? 15 Is the immune system important? As Dr. 16 Craig mentioned, we commonly look at neutropenic 17 animals, but also look at normal animals, and what 18 impact does that have on the amount of drug that 19 one requires or is necessary for efficacy? 20 Lastly, as John Powers mentioned, what 21 treatment endpoint is important? We often look at 22 a variety of microbiologic outcomes, does this have 80 1 any relationship when we look clinically? 2 [Slide.] 3 Here is a dataset that begins to show how 4 one can look at a variety of these factors. 5 Firstoff, if you look on the lefthand side here, of 6 this graph, we are looking here a microbiologic 7 outcome in two animal infection animals. 8 First, the traditional pharmacodynamic 9 model, the thigh infection model, and the lung 10 infection model here, a therapeutic model, and you 11 can see here that despite the fact that we are 12 looking at, in this case, two infection sites, 13 outcomes seem to be the same. 14 Here, we are also looking at a wide 15 variety of drugs, all within the beta-lactam class, 16 with carbapenems in red, penicillins in aqua, and 17 cephalosporins in yellow, and you can see that 18 despite the fact that we are looking at a variety 19 of drugs, within these drug classes, the 20 relationship is very strong. The amount of drug or 21 the PK/PD target in this case, the time above MIC, 22 if one were to look for maximal efficacy, if one 81 1 looks at the red circles here, the carbapenems, you 2 see maximal efficacy with times above MIC of 20 to 3 40 percent among all of the drugs within the 4 carbapenem class. 5 [Slide.] 6 Here is an example of a dataset that looks 7 at the impact of protein binding, and as George 8 Drusano mentioned, certainly, what I will 9 demonstrate here is what we found for certainly, I 10 would argue more than 95 percent of the case, here, 11 we are looking at the impact of protein binding 12 among 7 fluoroquinolones, and what we are looking 13 at here is that microbiologic outcome in a 14 pharmacodynamic model, the thigh infection model, 15 and the endpoint we are looking at here is 16 microbiologic endpoint, the amount of drug, or in 17 this case, the 24-hour, AUC to MIC ratio, that was 18 needed in this case to produce a static effect or a 19 static dose. 20 You can see here, looking across, the 21 amount of drug necessary for each of these 22 fluoroquinolones, looking at total drug levels, 82 1 they all look to require about the same amount of 2 drug until you run into two of the drugs that have 3 higher degrees of protein binding, in which case it 4 would look as if you would need much more 5 gemifloxacin or garenoxicin when considering just 6 total drug levels. 7 However, when you correct for protein 8 binding, the same amount of drug is needed to 9 achieve efficacy, in this case, the static dose. 10 Again, I could also show you examples of 11 where this doesn't fit, but those examples are few 12 and far between. 13 [Slide.] 14 Here is an example of a dataset that 15 addresses two additional variables. First, it 16 addresses the impact of the infecting species on 17 the pharmacodynamic target. You can see here data 18 with the cephalosporins, penicillins, and 19 carbapenems, in treatment again in a thigh 20 infection model, the pharmacodynamic model, looking 21 at the impact of treatment of gram-negative 22 bacilli, pneumococci, and staphylococci. 83 1 You can see here that there are slight 2 differences, so the infected species does matter 3 although sometimes only very slightly. 4 Here, also, you can see the impact of 5 looking at different treatment endpoints. On the 6 left here, you can see the amount of drug or, in 7 this case, the time above MIC needed to achieve a 8 static effect, a net static effect versus the 9 amount of drug or the time above MIC needed for 10 maximal microbiologic efficacy, and you can see 11 certainly there is a step-up when you look at these 12 two different endpoints. 13 [Slide.] 14 Here is a set of data that looks at the 15 impact or the variable of resistance within the 16 organism. Certainly, the target within these 17 organisms is changing over time. We are seeing a 18 creep in MICs. 19 Here is a dataset looking at therapy with 20 two beta-lactams in the thigh infection model 21 against pneumococcus, and organisms in this case 22 had MICs varying roughly 100-fold, and the endpoint 84 1 here we are looking at, here again is microbiologic 2 endpoint, in this case, the net static effect. 3 You can see here that the amount of drug 4 or the time above MIC that was necessary was not 5 impacted by resistance in the organism in this 6 case, with amoxicillin and cefpodoxime. 7 [Slide.] 8 Here is an example of looking more closely 9 at the impact of infection site on the magnitude of 10 the parameter needed for efficacy, here again with 11 the beta-lactam amoxicillin looking at two 12 infection models. 13 Here again, our primary pharmacodynamic 14 model, the thigh infection model, as well as the 15 therapeutic model, the pneumonia model. You can 16 see here again, looking at microbiologic efficacy 17 again, that the relationships are very similar at 18 these two infection sites. 19 Bill Craig mentioned that there are 20 exceptions to this, and macrolides are one good 21 exception where again if I were to show you this 22 data for macrolides, they would also look very 85 1 similar, but we know very well that ELF levels in 2 the mouse are not the same as ELF levels in 3 patients, so one certainly needs to be careful and 4 look at pharmacokinetics also at sites of infection 5 in situations where the infecting pathogen is not 6 just in the interstitial space. 7 [Slide.] 8 Now, what I have shown you primarily so 9 far is data in the animal models using 10 microbiologic endpoint. We also look at a 11 therapeutic endpoint looking at mortality. Here is 12 data both from Bill Craig's laboratory, as well as 13 data from the literature, looking at the impact of 14 the target and mortality with the beta-lactams 15 penicillins and cephalosporins, and this is data 16 from three animal species and four sites of 17 infection. 18 One can see that the pharmacodynamic 19 target, in this case, the time above MIC, that is 20 needed for this therapeutic endpoint, in this case, 21 survival in the animals is very similar to what one 22 sees when one looks at microbiologic efficacy in 86 1 these animal models with times above MIC of 30, 40, 2 50 percent needed for maximal survival with the 3 beta lactams. 4 [Slide.] 5 Here is a closer look at the relationship 6 between microbiologic outcomes in these animal 7 infection models and a therapeutic outcome or 8 mortality in these infection models, and I will 9 direct your attention, first, to the graph on the 10 right, where you can see here, looking at the 11 relationship between the microbiologic endpoint, 12 the static dose, and the therapeutic dose, in this 13 case, 50 percent survival in the animals, and one 14 can see a very strong relationship, and one can 15 sort of dissect this relationship across a variety 16 of microbiologic endpoints to the point where one 17 can say I would expect maximal survival or 100 18 percent survival with this microbiologic endpoint 19 if you look on the left. 20 We have looked at this with a variety of 21 drug classes, a variety of infection sites, and a 22 variety or organisms, and this has really sort of 87 1 held true. What this data also suggests and looks 2 at is the impact of treatment duration. With the 3 majority of our microbiologic studies, we look at 4 24 to 48 hour endpoints. With the survival data, 5 we are looking at anywhere from 5 to 7 days. 6 What we find is that for the majority of 7 the situations that we have looked at, the 8 treatment duration from 24 hours to 7 days, we find 9 the same pharmacodynamic target. 10 [Slide.] 11 One thing you will find in the literature 12 when you look at therapeutic endpoints, however, is 13 variability in when one looks at mortality or 14 survival in relationship to when therapy ends. 15 We have traditionally looked at survival 16 in these animals at the end of therapy, as 17 represented by the yellow circles in this case. 18 You will find in the literature again quite a bit 19 of variation. 20 Some people will look, at antibacterial 21 therapy, 7 days or even 2 weeks after therapy has 22 ended, and when you add that into the equation, 88 1 what you get then is a few organisms left behind 2 regrowing in these animals and subsequently killing 3 the animals, really limiting one's ability to 4 derive a strong relationship. 5 [Slide.] 6 The last variable that I will discuss this 7 morning is the impact of the immune system in the 8 animals and the impact of the immune system on that 9 pharmacodynamic target. The example I will use in 10 this case is with the fluoroquinolones. 11 Looking at the left, as one might expect, 12 in healthier animals, in this case, animals that 13 aren't neutropenic, one requires less drug than in 14 animals that are neutropenic. 15 I will tell you that the degree of this 16 impact varies for drug classes, and it varies for 17 organisms, as one can see in the righthand graph. 18 Here again is looking at treatment with a variety 19 of fluoroquinolones in both normal and neutropenic 20 animals. 21 On the lefthand side, you see treatment of 22 pneumococcal infections. Here, you see quite a 89 1 marked effect in the amount of drug necessary for 2 the endpoint, in this case, the static dose in 3 normal animals is roughly 10-fold less than the 4 target needed in the treatment of neutropenic 5 animals. 6 However, if one moves go gram-negative 7 infections, the impact of the immune system, in 8 this case, neutropenia, is much less marked, 1- to 9 2-fold. 10 So, certainly, I think the immune system 11 is important to consider because you can see 12 differences that are organism and drug class 13 related. 14 [Slide.] 15 Lastly, and most importantly, does any of 16 this matter? Bill Craig has shown the slide on the 17 right, and many of you have seen this before, and 18 this is one of about 10 datasets that are now out 19 there, clinical datasets where one has either 20 clinical or clinical and microbiologic endpoints 21 that can be related to pharmacokinetics in the 22 patients. 90 1 For each of these 10 clinical datasets 2 that have been available, what we have seen in the 3 animals has ended up being supportive or predictive 4 of what one has seen in the clinical datasets, 5 here, looking at the left, mortality in the 6 animals, which we found really best correlates with 7 what one sees in these clinical datasets, in this 8 case with fluoroquinolones, looking at treatment of 9 gram-negative bacilli and fluoroquinolones, and 10 survival, where one sees, as Bill Craig pointed 11 out, maximal survival in these animals when the 12 fluoroquinolone 24-hour AUC to MIC reaches a 13 magnitude of roughly 100. 14 Again, as pointed out in the clinical 15 dataset, one sees a similar relationship here, both 16 clinically and microbiologically, and this is just 17 one example, and one example of hopefully, many to 18 come in clinical trials. 19 I will end there and thank you for your 20 attention. 21 DR. EDWARDS: Thank you very much, David, 22 very nice summary. 91 1 I am now going to call on Lisa Benincosa 2 from Pfizer to continue with the discussion. 3 Industry Perspective 4 DR. BENINCOSA: Good morning. 5 [Slide.] 6 I would like to begin with just a brief 7 summary of the objectives of early drug 8 development. During our early drug development, we 9 have two primary objectives. We are interested to 10 identify critical risk factors prior to investment 11 in full clinical development, and in answering this 12 question, we hope to select the most promising 13 compounds to move forward. 14 Some of the critical risk factors that we 15 might identify at this point are the therapeutic 16 index or anticipated therapeutic index, the dosing 17 regiment is inadequate or the early identification 18 of the need for controlled release development. 19 In addition, we hope to provide critical 20 data to identify safe and effective dose and dose 21 regimens, and this, as we have discussed this 22 morning, can lead to more efficient development. 92 1 We hope to create optimal study designs, as well as 2 efficient development strategies. 3 If we think about the tool kit, as we have 4 been discussing, we have animal models of human 5 disease, we have development of biomarkers, we have 6 pharmacodynamic modeling and clinical trial 7 simulation. So, all of this has advanced our 8 sophistication in being able to deliver on these 9 objectives. 10 [Slide.] 11 So, when we think about pharmacokinetic/ 12 pharmacodynamic modeling in drug development, it 13 really is a continuum. In 1993, Dr. Gerhard Levy 14 published an article, The Case for Preclinical 15 Pharmacodynamics, and really the premise is that 16 the effective concentration or the concentration on 17 the biophase will translate from animals to human, 18 and that is a 1 to 1 translation. 19 If you look at dose, for example, in this 20 publication, the effective dose range varied 21 10,000-fold, but in this survey, the effective 22 concentration across animals and humans was at most 93 1 about a 2-fold variation. So, that is the premise 2 for preclinical pharmacodynamic research. 3 [Slide.] 4 So, we begin in experimental models of 5 human disease, and we understand the 6 exposure-response relationship. We then move into 7 healthy subjects, and we look to refine that model, 8 that understanding, later moving into patient 9 studies, looking at proof of concept. 10 Now, the endpoint may change from 11 biomarker to a clinical endpoint, and we validate 12 the translation of that relationship between 13 biomarker and our clinical endpoint, and later 14 confirming the safety and efficacy in our pivotal 15 studies. 16 The red arrows to my left show the 17 validation of this relationship, so as those data 18 emerge, and we refine our understanding, the 19 usefulness of these biomarkers in these models, the 20 types of decisions we can make and how we can 21 leverage that information will increase for 22 compounds, may be the second in the class once that 94 1 validation with clinical data exists. So, we are 2 going to make decisions with much more confidence 3 and certainty following the validation. 4 Another point that I would like to make 5 here is that when we think--Dr. Lazor mentioned 6 this morning about safety, and truly another place 7 for us to impact is on the dose selection and 8 making sure that the dose is not too high. 9 So, particularly in antibacterial 10 research, our contribution in pharmacokinetic 11 dynamic modeling will also be in understanding 12 exposure response for safety and optimizing that 13 dose and dose regimen, and thinking about the 14 therapeutic index. 15 [Slide.] 16 Preclinical models. Listed on the top 17 here are the animal models that are used at Pfizer. 18 You see various murine animal models, as well as 19 gerbil otitis media model. 20 The advantages of the preclinical 21 pharmacodynamic research using these models is that 22 we can explore in vivo exposure-reponse 95 1 relationship. We can explore hypotheses, which may 2 be very difficult to do in the clinical situation. 3 We can assess pharmacodynamics a suboptimum doses 4 and supra-therapeutic, allowing us to explore the 5 full dose range. Also, the assessment of tissue 6 distribution. 7 The challenges to use are to validate 8 these animal models for the extrapolation of the 9 results into the clinical setting. 10 [Slide.] 11 As Dr. Craig discussed this morning, 12 around the global indices, these 13 pharmacokinetic/pharmacodynamic indices, there are 14 some limitations, so we have been exploring new 15 approaches in addition to the global indices where 16 we incorporate the time course using 17 mechanism-based pharmacokinetic/pharmacodynamic 18 models. 19 When I say "mechanism-based," the models 20 are consistent with the pharmacology of the drug 21 and also incorporate both what would be a placebo 22 response or disease progression, which really 96 1 characterizes the system, so you have your 2 drug-specific parameters and the parameters of the 3 modeling that characterize the system. 4 [Slide.] 5 The example that I would like to use to 6 illustrate here, thinking about the time course, 7 really the true pharmacodynamics, is to illustrate 8 this through a study of azithromycin. 9 [Slide.] 10 This was published by Dr. Girard showing 11 the results in a gerbil otitis media model, 12 following azithromycin administered, the same total 13 daily dose, a single dose versus the dose split in 14 a 2-day or 3-day regimen. 15 You can see the curve in blue, that the 16 most rapid eradication occurred following one 17 single dose rather than splitting this dose into 18 different regimens. So, the total daily dose 19 stayed the same, area under the curve would be the 20 same, but clearly, a difference in the performance 21 of the single-dose regimen. 22 [Slide.] 97 1 So, based on these results, the hypothesis 2 was that front-loading, or giving one dose, appears 3 to be more effective although the 1- and 3-day 4 regimens were also effective. 5 Now, the original study design, the data 6 were not captured to allow for 7 pharmacokinetic/pharmacodynamic modeling, so an 8 additional series of experiments were designed to 9 address this. The goal was to develop a 10 pharmacokinetic dynamic model for azithromycin to 11 quantitate the effect of front-loading and to 12 differentiate from 3-day and 5-day regimens. 13 [Slide.] 14 We used the gerbil otitis media model and 15 studies were done to identify--and this is an 16 important point--threshold doses, and I will 17 discuss this again in just a minute, that would be 18 around the ED50, which would allow us to make these 19 comparisons and being in the most sensitive dose 20 range. 21 We had to what we are going to call 22 humanize the pharmacokinetic profiles, basically to 98 1 take account with the differences in 2 pharmacokinetics to manipulate the dose, so that 3 the shape of the exposure over time would mimic 4 what we would see in humans. 5 Two strains were used and plasma 6 concentrations were determined over 72 hours, so 7 you see that we did capture the full time course, 8 and a group of controls were included, as well. 9 [Slide.] 10 Just to stop a minute on the rationale for 11 the dose selection. The most informative region 12 for us to work in is going to be between the ED20 13 and ED50 in the steep portion of the dose-response 14 curve. 15 If the experiments are done, which 16 typically we see, where all doses result in very 17 good response, you can't identify the true 18 differences if all doses were administered and 19 maximal response was achieved, and the same for the 20 lower end of the curve. 21 [Slide.] 22 This illustrates if we were to look at a 99 1 global pharmacodynamic measure, and shown here, if 2 you just follow the blue circles, those are all 3 from administering azithromycin as one single dose. 4 You see that there appears to be a relationship for 5 this regimen, but we don't have a universal 6 description of the data because you see the 3- and 7 5-day at a particular dose. If you just select one 8 dose, there certainly was suboptimal performance of 9 the 3- and 5-day regimen. 10 Now, in pharmacodynamics, we hope to 11 understand, our drug should be characterized with 12 an EC50 and an Emax. That is a drug-specific 13 parameter and it is independent of dose regimen. 14 So, you can see in looking at the data this way, 15 you really could not explain all of the data 16 simultaneous and really to understand the 17 characteristics of these results. 18 [Slide.] 19 Now, I am just going to move briefly into 20 the pharmacokinetic and pharmacodynamic results. 21 This is just to illustrate to you, and you can see 22 the little pulses on the dashed lines in gerbils, 100 1 that the way we emulated the shape of the 2 concentration time course was to give--you know, 3 azithromycin has a very, very long half-life, so 4 you can see the pulse doses at 24, 48 hours to 5 mimic that disposition. 6 [Slide.] 7 This is the pharmacodynamic model. It is 8 a very generalizable model, and it is shown here in 9 its most generalizable form, but basically, it is 10 to describe the time course of bacterial growth. 11 This can be done using a mixture model of 12 different subpopulations, if that is necessary. 13 The model incorporates bacterial replication as a 14 capacity-limited function. The important point 15 here, the first order constant for death. For 16 azithromycin in this case, the drug acted to 17 stimulate that first order rate constant. 18 Shown below is the equation, but I would 19 just like to say that in thinking about this, what 20 you can test and evaluate is whether your drug has 21 effects on replication, enhancing the cell death. 22 So, in this case, we were able to 101 1 characterize that drug-specific effect through what 2 is shown here is the equivalent of an EC50 and an 3 Emax. This becomes important in the interpretation 4 and guiding your dose regimen. 5 [Slide.] 6 This shows the pharmacokinetic results. 7 You can see the green symbols of the observed data 8 and the simulated profiles, which I showed 9 previously. 10 [Slide.] 11 This shows the eradication for both the 12 1-day and 3-day, and you see again, consistent with 13 the results that Dr. Girard had presented, very 14 rapid eradication following one large single dose. 15 When we administered that over 3 days, you 16 see there was eradication for a period of time, 17 some regrowth, again eradication and regrowth. So, 18 very striking differences between the 1- and 3-day 19 regimens. 20 [Slide.] 21 The real interest was in looking at the 22 time course to understand the appropriate dose 102 1 regimen. Shown on my left is just a repeat of the 2 prior slide, and what I would like to focus on is 3 on my righthand panel. 4 Remember that I told you that we 5 characterized the effects of azithromycin on this 6 rate constant, so what you are looking at here, the 7 red shows the eradication, but you are looking at 8 the percent of baseline for this rate constant, so 9 what we are able to understand here and delineate, 10 these two different dose regimens, that 11 azithromycin concentrations rose, and they were 12 sufficient to shift this rate constant to its 13 maximal value. 14 It was sustained for a period of time. If 15 we just look right here, this is the very rapid 16 eradication for 1 day, the 3-day, there is some 17 eradication, and eradication again. This line is 18 that rate constant, so you see there was 19 stimulation of the rate constant for a long enough 20 duration of time to allow for complete eradication. 21 Then, as the azithromycin concentrations 22 declined in plasma, that effect on that rate 103 1 constant disappeared. 2 [Slide.] 3 So, in summary, the preclinical modeling 4 using this animal model provided us a means to 5 quantitate this effect of front-loading the dose of 6 azithromycin. There was more rapid and complete 7 kill, and there was a concentration-related 8 amplification of the bacterial death. 9 This research allowed us to understand 10 that having the highest AUC at the time of greatest 11 bacterial counts would result in greatest kill 12 possible, and optimizes the likelihood of positive 13 clinical outcome. 14 So, overall, the utility of preclinical 15 pharmacokinetic dynamic modeling enables us to 16 support the dose selection, and what is most 17 important here is dose regimen really enabling us 18 to optimize our dose regimen. I don't want to 19 underestimate our interest in understanding the 20 time course of pharmacodynamics. 21 We would propose that the best surrogate 22 of efficacy should be identified in utilizing the 104 1 mechanism-based pharmacodynamic models, and that in 2 some cases, in the question that you are asking, 3 the global indices may not be optimal. 4 [Slide.] 5 In conclusion, preclinical PK/PD models 6 can be used to support the overall clinical benefit 7 of the proposed clinical dosing regimen. 8 DR. EDWARDS: Thank you very much, Lisa, 9 for bringing out those very interesting and 10 important points in this area. 11 I am now going to call on the last speaker 12 in this section, Chuck Bonapace from the FDA. 13 Chuck. 14 FDA Perspective 15 DR. BONAPACE: Good morning. 16 [Slide.] 17 My objectives for this morning are to 18 discuss the role of in vitro and animal models in 19 drug development, to discuss the characteristics in 20 animal models that should be stated in study 21 reports submitted to the Agency, to discuss the 22 endpoints of in vitro and animal models in relation 105 1 to Phase II and III studies, and finally, to 2 discuss the limitations of therapeutic animal 3 models. I am going to discuss these terms a little 4 bit later. 5 [Slide.] 6 I just want to begin with the overall use 7 of in vitro and animal models in drug development. 8 As we heard this morning, the primary purpose of 9 this is to determine the principal PK/PD index. By 10 the "principal index," this is the index which is 11 the best predictor of the outcome. 12 The reason why this is common to do in in 13 vitro and animal models is that we can easily 14 assess a wide range of doses and dosing intervals. 15 This is not always possible in clinical studies due 16 to the limitations of dosing, we usually see a 17 single dose, and the ethical considerations, the 18 ramifications of a subtherapeutic dose. 19 We have heard already this morning about 20 the common PK/PD indices, and these consist of the 21 24-hour AUC to MIC ratio, Cmax to the MIC ratio, 22 and the time above MIC. 106 1 The definitions of these indices are not 2 standardized, and we would encourage other indices 3 to be considered with future submissions. 4 Consideration of the principal PK/PD index 5 and the magnitude of that index should be used when 6 determining the appropriate regimen to be evaluated 7 in Phase II and III studies. 8 We haven't heard this morning of any data 9 from in vitro models. We have been told that 10 sponsors perform these models early in drug 11 development, but the data is not submitted. We 12 would encourage submission of in vitro model data 13 in addition to animal model data with submissions. 14 This would be used to support the regimens 15 for Phase II and III studies. The information 16 obtained from in vitro and animal models has the 17 potential to increase the probability of a 18 successful outcome in clinical studies. 19 One of the things we want to avoid is 20 unsuccessful outcomes. It is a probability, it 21 does not guarantee that using the dose that is 22 likely to increase the probability of a successful 107 1 outcome will have a successful outcome. 2 The next few slides discuss the 3 considerations of using in vitro and animal models. 4 We have plenty of this, this morning, with protein 5 binding. I just want to state that it is not 6 always taken into consideration. 7 It is common reported as total 8 concentrations, sometimes it is not even stated 9 whether they are total of unbound. Our advice is 10 to report total and unbound and clearly state which 11 one is which. 12 The considerations of initial inocula, 13 depending on the initial inocula used in in vitro 14 and animal models can determine the outcome with 15 similar exposures from drugs. We would ask that 16 sponsors clearly state the initial inocula used and 17 justification for why they chose that initial 18 inocula. 19 The pre-treatment interval, this is the 20 interval between the inoculation and the initiation 21 of drug therapy. It ranged from an immediate, which 22 is sort of a prophylaxis, to several hours. Again, 108 1 it should be clearly stated and any justification 2 of why one was chosen and may be relation to an 3 established infection in human be stated in study 4 reports. 5 [Slide.] 6 Duration of experiment. It is a variable. 7 It is usually about 24 hours. Longer experiments 8 have been shown to reveal different outcomes. 9 Again, we would ask that this is clearly stated in 10 submissions and justification provided for the 11 rationale of the determination of the duration of 12 the experiment. It may be limited to the 13 practicalities of the experiment. 14 Surface area-volume ratio is something 15 that is only used with in vitro models, 16 hollow-fiber models are very common. There are 17 other in vitro models that are used. We would ask 18 that the sponsors state the surface area-volume 19 ratio and any justification because the surface 20 area-volume ratio can impact the concentration of 21 the time profile in the peripheral compartment if 22 it is a two-compartment model. 109 1 One benefit of in vitro models over animal 2 models is that the half-life is easily altered, so 3 it is relatively easy to mimic human or animal 4 pharmacokinetics or half-life, and again, this is 5 something that we would ask to see in study 6 reports. 7 [Slide.] 8 The next few slides go over the 9 determination of the principal index. This usually 10 involves a large number of doses and dosing 11 intervals known as dose fractionation. The degree 12 of fractionation is not consistent between studies. 13 We have seen studies that only study long intervals 14 in a sense. 15 Other designs that have the same rationale 16 are acceptable. We just saw one where we are 17 looking at the most relevant portion of a curve 18 from the ED20 to an 80, and then usually, those 19 doses are fractionated, so you are not studying 20 unnecessary doses in a sense. 21 Consideration should be given to the 22 dosing frequency with concentration-dependent 110 1 drugs. We know that overextending the interval can 2 lead to different results especially with 3 concentration-dependent drugs. 4 We have seen submissions where there is a 5 possibility that this is what happened, but again, 6 we don't know early on. 7 [Slide.] 8 Many of these models will normally assess 9 an ATCC strain for select gram-positive and 10 gram-negative organisms. It is not to say that this 11 is not representative of a clinical isolate, but it 12 may not represent the organisms likely to be 13 encountered in clinical infections. 14 What I mean by that is we have seen 15 submissions in which an animal model was performed 16 with an organism, such as Streptococcus pneumoniae, 17 and this is used to predict outcome in complicated 18 skin and skin structure. 19 So, the organisms that are studied should 20 be relevant to the indication that it is likely to 21 be supporting. 22 Because a single isolate is used to 111 1 represent a genus and species, we commonly do not 2 see a range in MIC values, and because of this, the 3 range and the effect from an in vitro and animal 4 model is based on altering the dose and the dosing 5 interval, and not on the MIC, and thus, we don't 6 get information for the impact of the MIC on the 7 principal PK/PD index, so this is usually not 8 assessed. 9 Sometimes we do see clinical isolates 10 studied in addition to an ATCC strain, many times 11 we don't. 12 [Slide.] 13 The determination of the PK/PD index is 14 usually based on the coefficient of determination 15 of a sigmoid Emax analysis. It sounds simple. 16 This is not always presented, and many times the 17 method that was used to determine the principal 18 index is not stated, so we can't confirm what was 19 the basis of that. 20 When results from in vitro and animal 21 models are submitted, we would like to see the 22 results in agreement, so that they are supporting 112 1 that in vitro and animal models are given the same 2 conclusion, the same principal index. 3 Ideally, and I emphasize the word 4 "ideally," results should be confirmed for more 5 than one preclinical study. An example of this may 6 be submission of results from an in vitro and an 7 animal model, animal model studies performed by the 8 same investigator, so we confirm one study with 9 another, or animal model studies from more than one 10 investigator, so again we have some confidence 11 that, in fact, the principal index has been 12 confirmed in more than one study. 13 [Slide.] 14 My next several slides talk about the 15 magnitude of the principal index. The magnitude of 16 the principal PK/PD index is dependent upon the 17 endpoint chosen. It will depend whether it is a 18 static of 1 log, a 2 log, or Emax effect. 19 The endpoint essentially should depend on 20 the endpoint that is associated with the clinical 21 outcome. Now, this is not always known early on, 22 especially if this is the first drug in a new 113 1 class, but for drugs and which it is a new class, 2 and there are other drugs on the market, there 3 should be some information to suggest what outcome, 4 what endpoint is associated with clinical outcome. 5 Also, the impact of immune function is not 6 always addressed in animal models. We don't always 7 see neutropenic and immunocompetent animals. We 8 would like to see neutropenic and immunocompetent 9 animals, also, the effect on the immune system in 10 animals, and how does the neutropenic and 11 immunocompetent animals relate to clinical 12 infections. 13 For instance, hospital-acquired infection 14 versus community-acquired infections, where 15 patients with hospital-acquired infections may be 16 more immunocompromised than those who are 17 relatively healthy patients in the general 18 population in the community. 19 [Slide.] 20 Again, the degree of protein binding. I 21 know I have said this already, but this has a big 22 impact on the magnitude of the principal index. 114 1 So, again, they should clearly state whether it is 2 based on total and/or unbound, and we would 3 certainly like to see submissions using both, if 4 not unbound only. 5 [Slide.] 6 How is this information used to determine 7 dosage regimens in Phase II and III studies? Well, 8 consideration of the target population should be 9 used, and whether this is again going to be a 10 community-acquired infection drug or nosocomial 11 infection drug. 12 Again, it brings into the immune status of 13 the patients. We are encouraging dose-ranging 14 studies at least in Phase II. We have heard some 15 other speakers mention this, this morning. These 16 are examples of ways doses could be determined in 17 Phase II studies, but it depends on the 18 indications, the patient population, and so forth, 19 so isn't something that I am saying to be used in 20 every one. 21 One example would be to use an endpoint of 22 like a 2-log kill for the highest dose studied, and 115 1 then an endpoint, such as 1-log or a static 2 endpoint for a lower dose. So, essentially, you 3 are going to have a dose-ranging based upon what 4 might work. 5 Another way to do this is to base it on 6 the percentage of the population that is achieving 7 an endpoint. So, if an endpoint has been shown that 8 it correlates or is associated with clinical 9 outcome, the high dose would be a dose that 100 10 percent of the patients who will receive that drug 11 will achieve that target, and than a lower 12 percentage. 13 In this case, I have 80 percent, but it 14 could be any number would achieve that target would 15 be a lower dose. 16 [Slide.] 17 I just want to talk a few minutes about 18 animal therapeutic models. These are a little bit 19 different than a PK/PD model in the sense that they 20 may, if it's certainly not a valid model, but a 21 model that may describe a little better what is 22 going in clinical infections, may give some insight 116 1 into infections in humans, and there are various 2 models. 3 Three of them, which I have up here, is 4 like an endocarditis, pneumonia, and a meningitis 5 model. A well-designed animal therapeutic model 6 can provide information for clinical trials. It 7 may give a little insight on the potential efficacy 8 based on drug concentrations at the site of 9 infection assuming the concentrations in the animal 10 and human are similar, and provide insight for the 11 dosage regimens to be evaluated in clinical trials. 12 Dose-ranging PK/PD studies performed in an 13 animal therapeutic model may provide additional 14 information to support clinical efficacy. We 15 commonly do not see this. Many of the animal 16 therapeutic models that are submitted contain one 17 or two doses. We do not see dose-ranging studies 18 and many of them do not even have the 19 pharmacokinetics in the model, so we can't do a 20 PK/PD analysis. 21 The utility of this will depend on the 22 applicability to a clinical setting. The 117 1 difference between treatment of an established 2 infection in humans versus prophylaxis, so if the 3 drug is administered almost immediately after the 4 infection is initiated, difference in 5 pharmacokinetics and tissue penetration between 6 animals and humans, and also the outcome may be 7 dependent upon the animal species. We may see 8 outcomes in different animals. 9 [Slide.] 10 Some limitations with animal therapeutic 11 models, the efficacy may be dependent upon various 12 factors, the virulence of the bacteria, the growth 13 phase that the bacteria is in, the initial 14 inoculum, the time to initiation of treatment, host 15 immune reponse, animal pharmacokinetics, protein 16 binding difference between humans and animals. 17 Drug concentrations at the site of 18 infection again may differ between animals and 19 humans. So, there are a lot of differences between 20 human infections and animal therapeutic models. 21 The predictability of outcome in animals, 22 which is usually based on a microbiologic endpoint 118 1 or survival endpoint, to humans, which is generally 2 a clinical endpoint, is not always known. 3 One thing I haven't talked on is the 4 toxicity of the drug, and this is important because 5 the toxicity of the drug may differ between humans 6 and animals, or between different species of 7 animals. 8 So, it may be important when you are doing 9 a dose-ranging study, and the dose is toxic, and 10 the efficacy you see may be--for instance, if 11 outcome was survival--it may be related to the 12 drug, and not the lack of efficacy. 13 [Slide.] 14 In conclusion, in vitro, animal models can 15 serve as the foundation upon which anti-infective 16 drug development should be based, and we emphasize 17 this. This is where we really see this information 18 being used, as the foundation of drug development. 19 In vitro, animal models represent an 20 important tool for determining the principal index. 21 They may be used to identify dosage regimens for 22 evaluation in Phase II and III studies, and 119 1 ideally, the results from clinical studies should 2 be used to confirm the in vitro and animal model 3 endpoint associated with efficacy. 4 It sounds pretty simple, but this is 5 something that we usually do not see. 6 [Slide.] 7 Closing the loop is a phrase that we have 8 coined for this. Essentially, what this means is 9 using the information determined by in vitro and 10 animal models, combine with pharmacokinetics from 11 Phase I to predict regimens, which are studied in 12 Phase II, and then the clinical efficacy, which is 13 determined in Phase III, that information should be 14 fed back to sort of confirm the original 15 hypothesis. 16 Although this sounds simple, we rarely see 17 this. So, one of the things we need to advance the 18 whole field or area is sort of a confirmation of 19 the results from Phase II and Phase III and how 20 this relates in PK/PD to the original hypothesis. 21 One of the problems with doing this is 22 that PK/PD is usually not done in Phase II and III, 120 1 and actually is fairly common to see plasma 2 concentrations even obtained in Phase II and Phase 3 III. So, many times we do not have the tools or 4 answers we need to confirm the original hypothesis. 5 Thank you for your time. 6 DR. EDWARDS: Thank you very much, Chuck. 7 At this point, we are going to take a 8 15-minute break and then we are going to have a 9 relatively large block of time for discussion 10 before lunch. We will come back in 15 minutes. 11 Thank you. 12 [Break.] 13 Discussion 14 DR. EDWARDS: We have 30 minutes now 15 approximately for the discussion of this very 16 complicated, complex area. I might just make a 17 couple of comments before we start. 18 I think what we would like to do with this 19 discussion is to make it in some ways more general 20 rather than highly specific in terms of small 21 points that have to do with PK/PD indices and 22 interpretation. 121 1 I think what we have heard is one thing, 2 one concept, is that the PK/PD issues in some ways 3 have been laid down over many years. The 4 fundamental principles are there, but the execution 5 is an issue that Chuck clearly and explicitly laid 6 out for us. 7 So, in many ways, a lot of the tools are 8 available, but the utilization is not at the 9 optimum level at the present time. I think that is 10 a fairly fair summary of what Chuck was telling us. 11 I think one of the issues that would be 12 highly desirable to perhaps focus much of this 13 discussion on is getting back to the central 14 question of how can the PK/PD be used to shorten 15 trial time, and I would like to start this part of 16 the discussion refocusing on the issue that we 17 spoke about for quite a long time yesterday and 18 have touched on very briefly today, and that is the 19 use of the PK/PD for approval for endocarditis. 20 John, let me ask you to do two things. 21 One is to tell us what you would like to get from 22 this discussion, where you would like to see the 122 1 precious few minutes we have for this huge area, 2 which, you know, has been the whole focus point of 3 a task force in the past, and tell us that and let 4 us get focused in the direction you would like us 5 to go, understanding that we are interested in the 6 primary issue of how can we devise perhaps, not 7 only specific issues, but also a process to get 8 closer to the goal of being able to use PK/PD to 9 shorten trial time. 10 DR. POWERS: I think what we are going to 11 do is this afternoon's sessions, the two of them 12 are what are we doing now and then the second 13 session this afternoon will be what can we do 14 better, so I think we will have more opportunity to 15 discuss that this afternoon. 16 I guess the issues that come up here, if 17 we can use sort of the endocarditis template as an 18 example, and that is where we ended yesterday and 19 George Talbot brought up I would feel uncomfortable 20 in giving this drug to somebody without it having 21 been used clinically, and George Drusano brought up 22 wait a minute, is there a way we can look at this 123 1 in Phase II that would give us some comfort. 2 I would ask the people in industry that 3 are here of what their view on all of this is, 4 because as I said earlier, what we see is some 5 people skipping from Phase I directly to Phase III. 6 We implied that what that means, and we don't know 7 this for sure, is that industry is seeing Phase II 8 as slowing them down and that since it is important 9 to get their drug to market as quickly as possible, 10 that perhaps they are not seeing any benefits of 11 this Phase II. 12 So, I would ask both people around the 13 table and in the audience and industry to say do 14 you find this useful, is this something your 15 management is going to say, look, forget it, this 16 Phase II stuff is taking too long, and is it help 17 for, as an agency, to try to do this and say, look, 18 this is how we can find it to be helpful in those 19 terms. 20 Again, let's go back to the endocarditis 21 example, is there some in vitro, hollow-fiber, and 22 animal model that would give people some comfort 124 1 that they could roll in an endocarditis study into 2 their Phase III development program rather than 3 have to do all your other, you know, complicated 4 skin, hospital-acquired pneumonia first and then go 5 back and do an endocarditis study, which may not be 6 as palatable to some people in terms of prolonging 7 their development process. 8 DR. EDWARDS: Bill. 9 DR. CRAIG: We have gone back and looked 10 at all the data in the literature on animal models 11 for endocarditis to see if we could model anything 12 of such a paper is published. 13 The problem that you have is these 14 one-dosage regimens, so whatever parameter you want 15 to pick up will show correlation because of their 16 interrelationship, but at least for the 17 fluoroquinolones, it appears, in terms of 18 magnitude, to be somewhat similar to what we see 19 for gram-negative organisms in the thigh. 20 The problem again is that there is 21 different sampling times. Some people look after 22 day one, some look at day three, some look at day 125 1 seven. There is a whole variety of different 2 sampling times, so it is not consistent and it 3 makes it much harder to look at, but it is an area 4 in which if there was better design, one could look 5 at, because oftentimes what is done is one dose is 6 looked at, and that is all you get, not a 7 dose-response or anything. 8 These models are expensive and ethically, 9 you know, you can only use so many animals, so you 10 can't oftentimes get as much as you want from 11 endocarditis models as you can get from a thigh 12 model. 13 DR. EDWARDS: George. 14 DR. TALBOT: Thank you, John. I think it 15 would be helpful to go in that direction, to focus 16 some questions. I view this really as related to 17 the FDA's wonderful document on the critical path, 18 which makes a point that innovation is stagnating, 19 the costs and difficulties are going up, and what 20 is needed is really translational research or 21 applied research to move forward from the 22 technology base that exists. 126 1 So, just to expand a little bit, the ideas 2 you mentioned, it seemed to me that some of the 3 specific concrete questions that could be asked 4 could be the following. Should, in fact, Phase II 5 studies be more robust as opposed to less, would 6 that, in the end, save a lot of time? 7 If you collected more dose-response and 8 exposure-response data in Phase II, and I would 9 add, not just efficacy, but also safety, realizing 10 that sample size and numbers of adverse events in 11 small trials may limit the power of your analyses, 12 but shouldn't safety be more a part of this whole 13 PK/PD question both in Phase II and Phase III. 14 Another concrete question we could ask is 15 what degree and precision of characterization of 16 both efficacy and safety PK/PD relationships in 17 Phase II, based on the animal data that Bill and 18 others create, what degree of characterization 19 would allow, for example, a single trial for 20 indication, and how would that apply, if was a 21 within-class, that is, another fluoroquinolone 22 versus a new class. Chuck sort of mentioned that 127 1 point. 2 And if you could do a single trial based 3 on a robust Phase II program and in a well-defined 4 animal in vitro database, what type of trial would 5 that have to be? You know, would it be a trial of a 6 standard size, such as you would do when you are 7 doing two, well-controlled, statistically adequate 8 trials per indication, or would it have to be more 9 robust, allowing for more sensitivity analyses? 10 I would hope that some points like that 11 could be discussed, because that would take us back 12 to where we started yesterday, which is how to make 13 this easier, how to reinvigorate the process, how 14 to keep those people who are in, in, and move 15 forward. 16 DR. EDWARDS: Yes. 17 DR. EISENSTEIN: John, you asked from an 18 industrial point of view, and given that Cubist is 19 in the midst of an endocarditis study, I thought I 20 could make a few comments. 21 Maybe as an overarching point, it seems 22 that what we are all talking about here is a 128 1 balance between human health, which is obviously 2 the primary point, that could actually then be 3 subdivided into several issues. One of them is the 4 ethics for the individual patient. The second is 5 the larger public health issues, and clearly, the 6 Agency should take a lead role in thinking in that 7 regard. 8 There is the science that we are thinking 9 about that allows us to come up with predictive 10 models. Then, as a point that the companies have 11 to deal with, as well as all of those others, 12 particularly as you have well pointed out in 13 dealing with upper management or the business 14 aspects. 15 It seems to me that what we are really 16 trying to describe today is a way to manage risk, 17 and risk management can be looked at in the various 18 concepts of the individual patient. We want to be 19 as ethical as possible. We also want to manage 20 risk at the public health level, so that we are 21 capturing as much value as we possibly can. 22 From the business standpoint, one is 129 1 dealing with a variety of various issues. It all in 2 a way boils down to return on investment, which 3 means cost, time, likelihood of success, and 4 likelihood of success is another way of saying 5 managing risk, so risk can be viewed in that 6 regard, as well. 7 Now, insofar as our own indication and 8 experience with that indication with endocarditis 9 and daptomycin development, we have significant 10 background information that we think has enabled us 11 to move appropriately into the study of this very 12 difficult-to-treat disease. 13 In terms of Phase III studies, clearly, we 14 already have an indication with two pivotal complex 15 skin, complicated skin and skin structure infection 16 studies, as well as significant safety data 17 including pulmonary data, as well as a lot of other 18 data at a higher dose. 19 From the preclinical animal model data, 20 there is significant PK/PD data specifically 21 looking at various organisms and various models of 22 endocarditis at various dosing intervals, and then 130 1 from the Phase II analysis, we not only have the 2 old Lilly bacteremia and endocarditis data, but 3 Cubist's own bacteremia data. 4 It is with that combination that we feel 5 that we can move appropriately into this very tough 6 indication. Let's remind ourselves that there 7 hasn't been a registration trial for endocarditis 8 in 20 years. We feel that if one is going to work, 9 that this is the more likely one to do so, but only 10 in the background of the significant amount of 11 scientific and medical and ethical and business 12 considerations. 13 DR. POWERS: I hadn't thought about it 14 that way, Barry, but maybe one way to look at this 15 process is the traditional and perhaps the best way 16 to manage risk about determining whether a drug is 17 safe and effective is in Phase III randomized, 18 controlled trials. 19 What we are hearing is that is also a 20 difficult way to manage risk, so what we are trying 21 to do is say if we only have one Phase III 22 randomized, controlled trial, is there some of this 131 1 other information that we can then use to give us a 2 better level of certainty about whether that drug 3 is safe and effective or not. 4 The concept of using other information 5 isn't new, and the Code of Federal Regulations 6 actually specifies that in certain situations, you 7 can accept one clinical trial plus supportive 8 information, and part of what we are trying to get 9 at today is okay, if we don't want to do two big 10 Phase III clinical trials, what is the level of 11 that information that is done earlier in the 12 clinical development program that would give us 13 that level of certainty, maybe not equal to what we 14 would get in a Phase III randomized trial. 15 George, one of the points you bring up is 16 making Phase II trials more robust, and one of the 17 things I can think of right off the bat is putting 18 controls in those, that uncontrolled Phase II 19 trials, would you rather see a 100-patient 20 uncontrolled trial or a controlled trial with 50 21 and 50, which one of those would actually be able 22 to give you more information. 132 1 Jack, Mike Dudley has had his hand up for 2 about a half an hour. I don't think you can see 3 him beyond the podium. 4 DR. EDWARDS: I am sorry, Mike. 5 DR. DUDLEY: Let me go back to the 6 endocarditis question, as well, because I think 7 there is two points that are relevant I think 8 towards bridging with the animal model and the 9 observations in humans, and then I think also to 10 address, I think, what George exhorted us to do 11 earlier about an action item. 12 One is I completely agree with the idea of 13 controls. The trial that George Drusano mentioned 14 yesterday, that involves cefonicid in right-sided 15 endocarditis where there was failures, was done at 16 an institution with a lot of experience on treating 17 right-sided endocarditis in San Francisco General 18 Hospital. 19 They know what the response is in patients 20 who are getting adequate antibiotic therapy with 21 clearance of cultures and defervescence of fever. 22 In that study, those patients did not respond 133 1 within 4 days. When they were immediately switched 2 on to nafsone [ph], immediately responded in all of 3 the cases. 4 I do know that in industry, it is 5 oftentimes, particularly in thinking about 6 staphylococcal disease, that in a clinical trial, 7 in a Phase II trial where you might have a 8 comparative group like that, where you can look for 9 those, is that one can get a very quick read across 10 dose and across patients with respect to response. 11 These patients, as you know, have 12 staphylococcal bacteremia for several weeks before 13 they finally present to be treated, so one is able 14 to assure adequate safety in those individuals that 15 if you are on a failing regimen, you can switch on 16 to something else. 17 So, I do think that there are examples 18 where you can go into those diseases and 19 particularly with good animal model data, which in 20 that case, it actually did work in the bunnies 21 because the drug was only 92 percent bound in 22 rabbits, so therefore, they saw efficacy. 134 1 With respect to the other point of 2 preclinical models and looking at, we have had 3 experience with an anti-MRSA cephalosporin whereas 4 if you go into the rabbit model of endocarditis and 5 you dose it, so that you are 25 percent time above 6 MIC, which according to Bill's slide is still 7 within the--for anti-MRSA cefs or beta-lactams for 8 staphylococci, is an effective regimen within the 9 thigh, you do get efficacy in the rabbit model. 10 You get about a 3-log drop in the MRSA in 11 the rabbit over the 4-day treatment period starting 12 from a CFU per gram of vegetation of 10 8. However, 13 if you go to 33 percent time above MIC, you can get 14 a 6-log drop in those organisms with 6 out of 8 of 15 the rabbits, if I remember, having sterile 16 vegetations by 4 days. 17 So, what I am saying is it depends on what 18 the endpoint is, and what we have to do in the 19 animal models now is decide what is the endpoint in 20 the animal model that corresponds to the outcome 21 within patients. 22 So, whether you want to take the clinical 135 1 endpoint or whether you want to take the 2 microbiologic endpoint, we do have to close Chuck's 3 loop. 4 We have to go back and look and say for an 5 acceptable clinical outcome or microbiologic 6 outcome within patients, what does that correspond 7 to in terms of exposure, and then what does that 8 exposure then go back into the animal model to get, 9 is it the static dose, is it a 2-log drop, or is it 10 a 3-log drop. 11 We have the information now out there. 12 You can go back as best you can from the 13 literature, but there are better studies now, 14 particularly in the thigh model, where we could go 15 back now for fluoroquinolones and say is it a 2-log 16 drop, is it a 3-log drop that we need to get in the 17 models to do that. 18 Azoles are an example where the clinically 19 effective dose of fluconazole corresponds to just a 20 1-log drop in the CFU in the kidneys of a mouse in 21 that model, but yet we can really force the dose up 22 and get a 5-log drop if we really want to. 136 1 So, calibrating those animal models to the 2 outcomes that are seen in patients would be a 3 helpful way to make them at least one step closer 4 as being a surrogate marker. 5 DR. POWERS: But what we see, on the other 6 hand, is the target has become everything. So, in 7 other words, a company starts off and says, well, I 8 need to get the AUC over MIC of 40, so I will go 9 back and I will pick the endpoint that gets me 10 there rather than doing it the other way around of 11 saying I am going to pick an endpoint of 2-log 12 kill, what target do I need to get to 2-log kill. 13 The literature says 40, I am going to go 14 back, and I didn't make it for 2-log kill for 40, 15 so I am going to change it to static. 16 I think that is one of the issues that we 17 are getting at today, is what are the ways in which 18 we can tell what the endpoint ought to be. 19 DR. EDWARDS: George. 20 DR. DRUSANO: I would like to kind of get 21 back to George Talbot's issue about robust Phase 22 II's, and I am going to, after lunch, show you some 137 1 bridging studies where the same drug was used 2 preclinically and also in a clinical circumstance 3 with significant correlation. However, let me put 4 this out there. 5 Let's do some war gaming. Suppose a 6 sponsor goes and does a robust Phase II trial. 7 Sponsors are different, so some sponsor might want 8 to see an 80 percent target attainment right with 9 the dose that they take, some 85, some 90, some 95. 10 Maybe it is dependent on whether it is your mom or 11 your mother-in-law that is going to get treated. 12 But at the end of the day, suppose that 13 they actually do pick a dose that is going to give 14 you a 90 or 95 percent target attainment rate for 15 whatever you choose. 16 You then take that into your robust Phase 17 II trial, and everybody dose well. Now, what is 18 the inference from both the regulatory perspective 19 and from the company perspective that should be 20 drawn from such an outcome? 21 Now, I have my own ideas, but I would 22 certainly like to hear some of the sponsors around 138 1 this table say something, and I would certainly 2 like to hear the FDA response and to how they would 3 respond to such a clinical trial, because if you 4 are getting 95 percent responses, it is not that 5 you won't get a relationship, but the likelihood 6 decreases as your obvious response rate goes up. 7 So, what would you guys do with that? 8 DR. POWERS: I am not sure I understand 9 the question, George. 10 DR. DRUSANO: Okay. Let's back 15 yards. 11 DR. POWERS: So, you have got a cure rate 12 of 95 percent, and we are supposed to be unhappy 13 about that? 14 DR. DRUSANO: The point is, you know, we 15 say we want to validate. If you say, you know, you 16 go to the animal data, right, and you want to see a 17 pharmacodynamic relationship in Phase II, but you 18 have chosen the dose correctly, is that okay? I 19 mean if you get that 95 percent response rate, but 20 you don't get a pharmacodynamic relationship, does 21 that count as a validation of the preclinical data? 22 DR. POWERS: But one of the other things 139 1 you might get out of that is what was the safety 2 between those different doses, as well. So, again, 3 we can't forget there is more than just the 4 efficacy piece. 5 DR. DRUSANO: Oh, absolutely, but as I 6 think somebody else pointed out, if you are going 7 to do 50 to 100 patients, and you have 50 of them 8 on controls, your probability of getting a safety 9 relationship approaches zero from the left. 10 I mean we did 272 patients in the first 11 prospective pharmacodynamic clinical trial 12 published in the literature with the prospective 13 analysis plan registered with the FDA, for levo, 14 published in JAMA, and we saw zero relationship 15 amongst 272 patients with good PK, and their 16 toxicity outcomes. 17 DR. POWERS: Part of the issue about the 18 critical path is I mean there are opposite 19 examples. Five patients that got the drug in a 20 Phase I trial, 2 of them become hypotensive. So, 21 the question is part of the critical path is 22 selecting out the drugs you don't want to move 140 1 forward with, as well, so even though you don't 2 answer every safety question, you can pick out the 3 drugs that are highly likely to have significant 4 safety problems, and do you want to go forward with 5 those, or do you want to cull them out at that 6 point. 7 If you guys got the response, would you 8 say that validated the preclinical data? 9 DR. ZHENG: I think it is difficult. You 10 can qualitate to say the drug works or not, but I 11 think what we want is to verify that quantitative 12 relationship. 13 DR. DRUSANO: Then, you need to have some 14 failures, and therefore, then, you are not going to 15 pick necessarily the right dose, you are going to 16 have a wrong dose in order to get some failures. 17 DR. ZHENG: That is why I think Phase II 18 study is important. If you designed a Phase II 19 trial very well, I think if the approach is so 20 robust, you should be able to get some relationship 21 from Phase II study. 22 That is why I think we recommend 141 1 dose-ranging study in Phase II, because usually 2 Phase III is a fixed dose and very often, as George 3 said, because the response rate is so high, we 4 couldn't identify that relationship. We do have 5 many, many examples. 6 DR. EDWARDS: John. 7 DR. REX: I want to inject two notes of 8 caution. The first one is something that I was 9 taught many years ago, which is that any dummy can 10 measure an MIC. The trick is to measure the right 11 MIC, and everything we are talking about here today 12 has MICs as part of the equation, but also has 13 hidden in it this other choice, which is what is it 14 that you are measuring in your PK/PD model, and we 15 have had several discussions on that. 16 I just want to make that point again, and 17 I will make it really clearly. I can only make it 18 about MICs in particular, because you give me any 19 bug and any drug, and tell me what MIC you want, I 20 can measure that number for you, and I can make the 21 MCI vary by 3 orders of magnitude by changing the 22 way that I do the assay, 3 orders of magnitude. No 142 1 sweat. 2 I am sure you can also do that with PK/PD 3 parameter measurements, because MIC is built into 4 that, ergo, I know I can vary the denominator by 3 5 orders of magnitude, so surely I can fiddle around 6 with the numerator, as well. 7 So, keep that in mind when you are 8 thinking about using this to drive what you are 9 doing, and the whole thing has to come back to your 10 clinical response. 11 Let me reflect on what happened with 12 fluconazole, a drug where we actually really have 13 real human PK/PD observations. We have them in 14 esophageal candidiasis where we had the misfortune 15 for those individuals of having high MIC isolates. 16 We had the fortune of a variety of doses and blood 17 levels. We were able to establish a pretty strong 18 Cardoso response curve for fluconazole in human 19 beings at a time when we did not have animal data 20 to support that idea. 21 Then, David Andes came along and showed in 22 an animal model that actually, what you might have 143 1 guessed was the breakpoint, turned out to actually 2 be the breakpoint in his animals, and that actually 3 then happened to validate the human scenario for 4 invasive disease, not just esophageal disease, but 5 invasive disease. 6 So, all of it came together, but the 7 closing of the loop was really a sloppy mess, and 8 we never really closed the loop for the disease of 9 greatest interest, which is invasive candidiasis. 10 To this day, I can't tell you what the 11 minimum effective dose of fluconazole is, nor do I 12 care, I am not interested in playing a game of 13 limbo with fluconazole in treating these patients. 14 The other thing is that everybody in this 15 room would handle the drug differently, so to find 16 the minimum effective dose, I can find the minimum 17 effective dose for George, okay, or the minimum 18 effective dose of George, I am not sure which it 19 is, but that doesn't mean that that is the correct 20 dose for the next person with all of their various 21 modifications. 22 So, when we get into this issue of target 144 1 attainment rates and variabilities, you know, I 2 think that we have to inject a great note of 3 caution because we run the risk of making a whole 4 series of very conservative assumptions, and 5 getting to the end of the day saying, well, what 6 you really need is 14 grams a day of the compound 7 in order to guarantee a 90 percent target rate for 8 all of the lefthanded redheads who have a variable 9 modification of their handling of the drug. 10 You have to be very careful about that 11 because then that just drops out all the other 12 possibilities for sub-MIC effects, there are just a 13 whole host of other things that get buried in that, 14 so that is my note of caution here. A lot of these 15 numbers are very slippery, and do not get too hung 16 up on 80 percent or 85 percent or 89 and 17 three-quarter percent attainment rates. I think 18 that is a dangerous slippery slope. 19 DR. EDWARDS: Yes, please. 20 DR. GRASELA: I would like to make two 21 points. The first is I think as we go forward with 22 the evolution of PK/PD information, it is going to 145 1 be really hard to validate things. I think the 2 comment that we are making is that if you were to 3 take a quinolone into development right now, you 4 could work a quinolone up really well, and there is 5 such concordance between the in vitro animal and 6 clinical data that you could take all the factors 7 into consideration, protein binding, tissue 8 distribution, and you could zero in with Monte 9 Carlo simulation, zero in on what the dose is or 10 the dose that is most likely to be the most 11 effective, and if you don't have any toxicity 12 limiting you on the other side, you could push the 13 dose to the point where everybody who is going to 14 get better, is going to get better. 15 I turn that back around to the utility of 16 a robust Phase II study, because I have an example 17 in my presentation, but it took us pooling two, 18 Phase III clinical trials, pooling the comparator 19 quinolone, we had to throw it in there, as well, to 20 get 68 well-studied cases of Strep pneumo. 21 If you are looking for failures, you have 22 to have large numbers, so again the more that we 146 1 learn, the harder this is going to become unless 2 you are really going out into totally new targets 3 where we don't know anything and we are starting 4 all over again. 5 Then, I would say yes, use the tools as 6 much as we can, design a Phase II clinical trial 7 that gives you the answer, but for some of the 8 well-trod paths, this is going to be incredibly 9 difficult, hence, George's question to you, if we 10 developed a Phase II study and we had 95 percent 11 response rate, would you take it and let us move 12 forward to Phase III. 13 DR. POWERS: I think, Dennis, and one of 14 the things maybe Chuck can talk about this, when we 15 have looked at this internally, you point out 16 quinolones, give us another example, that's about 17 the only place where we have seen this 18 characterization done in a way that would be 19 useful. 20 Part of this issue is we do want to try to 21 see new targets, right, that is one of the goals 22 here is to try to come up with new antimicrobials, 147 1 so we want to be able to utilize that information. 2 The validation part, going backwards, is 3 helpful for further indications for that drug if 4 you want to study that, but let me give the example 5 of what we have seen, where you see a Phase III 6 trial, and the efficacy doesn't look so hot 7 relative to the comparator. 8 I walk over to the Biopharm people and say 9 how did they pick this dose, because it looks to me 10 like they didn't shoot for it, and everybody just 11 shrugs their shoulders. You know, you see Phil 12 scratching his head going, you know, how did they 13 come up with this. 14 That is the position we don't want to be 15 in. There has got to be some rationale for going 16 forward. 17 The other issue here is, we keep 18 acting--and John is pointing out some of the holes 19 in the science here--as if we know everything there 20 is to know about this and let's just call it a day. 21 There are still some things we have to work out 22 here about how useful this can be to us. 148 1 DR. GRASELA: Let me just put one more 2 point on the table. The first is which of the 3 furry individuals was John? I couldn't get those 4 on those two slides--I mean Phil--which one of 5 those furry slides was John? 6 The second question is how much, if 7 discovery is going to come from the smaller 8 companies, and that is something that was talked 9 about before, and we talk about the body of 10 knowledge and risk management, how much data from 11 the compounds that fail will you allow for the 12 support of future compounds? 13 For instance, we have three or four 14 possible leads. You pick the best lead, the best 15 biopharmaceutics' properties, et cetera, and work 16 it up fully in animal models, in vitro, et cetera, 17 and in the IND, toxicology studies, dose compounds 18 fail. 19 The question I would ask is how much can 20 you bring of that hard work forward when you pick 21 the backup compound that is very similar 22 structurally, and has very similar properties? 149 1 DR. POWERS: That is a really good 2 question. One of the parts of the critical path, 3 though, when Dr. Woodcock talked about this a 4 couple of weeks ago, is we see things other people 5 do not at the Agency. We can't publish that, we 6 can't make it public because it is proprietary 7 information, but one of the ideas of this end of 8 Phase IIA meetings is the idea of companies coming 9 in and talk to us, so we can tell them, look, we 10 don't think this is such a good idea. And do they 11 always listen to us? That is a different story, 12 but that we can actually try to give some of that 13 advice based on what we have seen maybe with other 14 compounds within the class. 15 DR. EDWARDS: John. 16 DR. BRADLEY: I just want to bring up a 17 couple issues, one, I deal with pediatrics and they 18 are a special population, and it's tougher to go 19 through IRB, and virtually, every drug we get has 20 gone through adult trials, so at least we have the 21 opportunity to view some of these Phase II 22 dose-ranging studies and see what doses fail in 150 1 adults, and don't even come near that. 2 But, secondly, within pediatrics, there 3 are different disease entities that we treat, 4 perhaps the same organisms and the same drugs, but 5 meningitis and cellulitis are two completely 6 different concepts for me in terms of target 7 attainment rates. 8 A non-bacteremic cellulitis, I am willing 9 to achieve 85, 90 percent. That is probably not 10 unrealistic, but for meningitis, I can't accept 11 anything less than 100 percent. I can't sacrifice 12 any children, and if you look at, you know, George 13 was mentioning everything is distribution, so 14 distributions of MICs, of organisms that cause 15 meningitis, distribution of PK serum values, 16 distribution of what you get in the CSF, all of 17 this stuff is very important in modeling how we 18 come up with a dose to treat children with 19 meningitis. 20 I am not saying it isn't complicated, but 21 these are all factors that we share with you in 22 dialogue in coming up with the dose that we use, 151 1 but in Phase II, to use a dose in children which 2 would fail is almost unethical. 3 DR. POWERS: That gets back to this issue 4 of validating that PK/PD piece, but let's take a 5 step back at what we are really trying to do here, 6 and it is question I asked in the beginning. 7 Are we trying to make these fine cuts 8 between a drug that is 96 percent effective versus 9 94 percent effective, or are we just trying to 10 select this drug will make it, and this drug will 11 not? 12 Look, if it comes out 95 percent versus 90 13 percent target attainment versus 20 percent, don't 14 those tell you something? I think that is getting 15 to John Rex's point about you can monkey with the 16 numbers to make it look like you want, but it might 17 be a little tougher to make a 20 percent look like 18 90 percent than it would be to make a 95 percent 19 look like 90. 20 DR. EDWARDS: Yes. 21 DR. TALBOT: It seems to me that those are 22 two very important questions, but they are 152 1 separate, (a), you know, which drugs can you kill 2 clearly and without question in Phase II given the 3 scientific and commercial constraints, and so 4 forth, and then of those, based on efficacy, look 5 as though they would almost certainly succeed, can 6 you further differentiate those based on any safety 7 information or exposure-response relationships, and 8 also, then, how can you leverage what you have 9 learned in a good Phase II and preclinical programs 10 to optimally design an efficient Phase III program. 11 But I think those are two different goals, 12 both of which have major potential impacts. They 13 would potentially at least reduce cost. They 14 certainly might save time, of not time in clinical 15 trials, time in review, multiple cycles, et cetera. 16 I think to the extent they can do that, 17 they reduce uncertainty, which as we discussed 18 yesterday, is the bane of existence of senior 19 management. 20 DR. POWERS: We have seen examples where 21 these principles were not applied optimally, and 22 companies went out and did 5 and 6 Phase III 153 1 trials, and the drug was not effective. That would 2 seem to be the biggest waste, first of all, it is 3 bad for patients. I tried to come back to that in 4 the beginning, we don't want patients to be failing 5 in serious and life-threatening diseases, but then 6 get beyond that, there are the issues of that is 7 not cost effective clearly to have done a big Phase 8 III program and have a drug that does not look like 9 it is effective in that setting. 10 I thought Dr. Eisenstein's way of calling 11 this "managing risk" was a really good way to look 12 at it in terms of if you are a company, is it 13 better to manage that risk in a smaller Phase II 14 trial and find out what is going on than to get to 15 a large Phase III program and get an unwanted 16 surprise. 17 DR. DRUSANO: John, let me make one small 18 proposal, and that is, if I am a sponsor and I go 19 and do what everybody would say would be a robust 20 Phase II trial, and I have taken my preclinical 21 data and done the best job I can with it. 22 I have chosen what everybody says is a 154 1 reasonable microbiological endpoint, I do a Monte 2 Carlo simulation, I do an expectation over the MIC 3 distribution. I get a population target attainment 4 rate, and it is something that everybody is happy 5 with. 6 I now go forward and do that robust Phase 7 II. You have, just from straight clinical and 8 microbiological outcome, you have a point estimate 9 of the response rate, and it is 95 percent 10 confidence bound. 11 In my view, and I would like to hear your 12 response, is if the Monte Carlo simulation is 13 somewhere in that 95 percent confidence bound, that 14 would serve to me as a validation of the 15 preclinical PK/PD even if you couldn't get a 16 traditional, you know, exposure-response 17 relationship because you had too few failures. 18 So, how would you respond to that? 19 DR. POWERS: I don't think your question 20 for us is the validation of the PK/PD target, is 21 that coming back to the loop part, but what we are 22 using it for in that particular drug product is to 155 1 decide am I going to go forward to a Phase III 2 program. 3 Let me turn it back and ask you, would you 4 be happy with that information if you had that in a 5 endocarditis trial-- 6 DR. DRUSANO: I would be delighted. 7 DR. POWERS: That is the point for us. 8 So, to then say, well, can we use it, go backwards 9 and validate the PK/PD target, that may be useful 10 for that company in terms of studying another 11 disease, but in terms of going forward for that 12 particular development program, it may not be as 13 relevant. 14 DR. DRUSANO: Would you allow that, then, 15 as a study that would serve as a supporting study 16 for the single, well-controlled Phase III? 17 DR. POWERS: I keep saying, George, we 18 have already done that. We have already gotten to 19 that point, and that is already specified in the 20 CFR that one study plus supportive information, and 21 we have already set the precedent of accepting 22 that. 156 1 But one of the things that I want to get 2 back to was I wrote down while Dr. Eisenstein was 3 talking about this target attainment, isn't this 4 dependent on the disease, and John Bradley brought 5 that up, so I think some of this is going to 6 be--so, in other words, if somebody brings us in 7 something for complicated skin, and then they want 8 to study their drug in endocarditis or meningitis, 9 those may be different situations. 10 DR. EDWARDS: Mike. 11 DR. DUDLEY: I guess the usual situation 12 might be, to take George's example further, would 13 be that we may have a signal from preclinical 14 studies, and even Phase I, that says maybe we are 15 not quite sure about the safety at that point. 16 So, in fact, you may want to go back and 17 say, well, then, what is my, not AUC to MIC 18 parameter, but what is the absolute AUC, and you 19 still may want to study lower doses because that 20 may save you from having that oops in Phase III 21 that you have got an unacceptable safety signal in 22 that, so you still have to go down, not that we are 157 1 going to purposely design trials that are going to 2 have patients fail, but we are going to study them 3 over a suitable dose-response curve even in Phase 4 II, so that if we have signals from the preclinical 5 and even Phase I setting, that we can start to look 6 for that and do that risk management in terms of 7 the low dose versus the high dose. 8 DR. EDWARDS: John. 9 DR. LAZOR: That was the point I was going 10 to make, because you don't want to be faced with 11 the oops in Phase III, and the question will come 12 up, would a lower dose have been just as effective 13 without the toxicity. 14 DR. EDWARDS: I am going to conclude this 15 discussion now and by way of summary, let me say I 16 think we have heard two things really clearly. One 17 is that some of the tools are available, and we are 18 not using them in a way that is user-friendly for 19 evaluation by the FDA. 20 We have clearly discussed in depth the 21 advantages of a robust Phase II clinical trial, and 22 I think that point has come through very clearly. 158 1 What isn't so clear to me is a mechanism that we 2 have devised to answer the question of what do we 3 need to get an endocarditis indication, not the 4 details, but the process, and I don't think we are 5 going to be able to discuss that further now, but 6 maybe we will have a chance to come back to it 7 later this afternoon. 8 If I could end there, we are going to have 9 an hour for lunch, and we will resume at 1:15. 10 I have some announcements. The panel 11 members should leave with the FDA staff, who want 12 to use that lunch facility. Anyone returning this 13 afternoon, please leave their badge at the table, 14 and someone will be here to watch all the luggage. 15 Thank you very much. 16 [Whereupon, at 12:15 p.m., the proceedings 17 were recessed, to be resumed at 1:15 p.m.] 159 1 A F T E R N O O N P R O C E E D I N G S 2 [1:15 p.m.] 3 DR. EDWARDS: Before we start, we are 4 running about 15 or so minutes behind, and we are 5 going to run just a little bit later than the 6 schedule shows, but not much, because some of us, 7 myself included, have got to get to the airport at 8 the right time, so don't worry about this dragging 9 on a significant amount of time beyond what you see 10 on your schedule there, but it will be a few 11 minutes. 12 Before we start the second part this 13 afternoon, I want to ask Bob Powell to make a few 14 comments from the podium. I have not called on him 15 in two sessions here, and I do not let him make his 16 comments, which are very important, I am going to 17 be persona non grata magna. I am trying my best to 18 stay out of that position. 19 Bob has some very important points to 20 make, so, Bob, let me let you do that now. 21 DR. POWELL: Thanks very much. 22 Of course, when someone says that someone 160 1 has very important points to make, the skeptics 2 will try and prove that that is not true, so I hope 3 not to disappoint. 4 Let me tell you I have spent some time in 5 academics. I spent about the last 16 or so, 17 6 years in the industry, and then I came to the FDA 7 in January. My sense is that with regard to 8 antibiotics, which I hadn't, aside from working on 9 developing antibiotics and antifungals and 10 antivirals some, my feeling is the conversation 11 really hasn't changed much from about 10 or 15 12 years ago, that I was hearing in terms of what is 13 the number in terms of whether it should be AUMC or 14 AUC ratios or peak or whatever, and I think what 15 John Lazor said in the very beginning was 16 absolutely key. 17 He was talking about this feedback loop 18 and that that doesn't exist, so that my curiosity 19 about that, I started thinking about, well, why 20 wasn't does that exist, because that is fixable, 21 and it has been fixable for other therapeutic 22 areas. 161 1 So, if you guys are stuck, then, you ought 2 to try and do something different, and think about 3 why that might be, because it is not the science, 4 the science exists to do what you want to do, and 5 yet why doesn't it get done. 6 I would suggest to you that one of the 7 main reasons is inside the drug company, it is not 8 at the FDA, but the FDA can potentially help solve 9 it. 10 The people that develop the science, 11 coming from preclinical through Phase I, all the 12 stuff that you are talking about and that you are 13 interested in, and that you want to get in here, by 14 and large, the people that develop that work tend 15 to have very little influence on the design and the 16 execution of the Phase III trials, so that the 17 piece that really needs to be fixed is inside drug 18 companies, the people in experimental medicine or 19 clinical pharmacology, or however you are 20 organized, really have to have an impact on Phase 21 IIB or Phase III trial designs, how they are put 22 together. 162 1 For example, it is fairly uncommon in 2 Phase III trials to have concentration data that 3 could be collected in such a way to develop the 4 correlations that you are looking for, or when 5 there is an adverse drug effect, to have the case 6 report forms. 7 I want to give you some specifics. The 8 case report forms that can be designed in such a 9 way that you know when the adverse effect occurs, 10 when that is in relation to the dose, and to have 11 it in the case report form, it is going to trigger 12 someone to go out and draw a plasma level, if that 13 is warranted, and also to describe the time that 14 the adverse effect is there. 15 I can tell you that is generally not 16 present. People may be asking for it inside the 17 company, but it is blocked. Now, why is that? The 18 people that are running--and they may be in this 19 room--the people that are running the Phase III 20 trials may not be listening to the requests for 21 dose ranging, and the other thing is that it is 22 clearly the closer a drug gets to the marketplace, 163 1 so that for an antibiotic, that frequently occurs 2 after Phase I, certainly in Phase II, marketing has 3 a tremendous impact on the way those trials are 4 going to be designed. 5 If someone only wants one dose, that is 6 hard to fight against. Now, how can you fight 7 against that? There is actually a couple of 8 leverage points. 9 By the way, INDs, the target in an IND is 10 the first dose in man is set 6 months before you 11 start dosing. That is set up with the CRO, when we 12 are going to take this sucker into humans. 13 Now, when the company puts in the IND, 14 they are trying to present as small a target to the 15 FDA as they can. I can tell you that, because I did 16 it, because they don't want to be stopped, so there 17 is nothing on what the proof of principle strategy 18 is, and there is nothing on what the dose finding 19 strategy is. It ain't there. 20 If you want to have that dialogue, there 21 is a provision to have a pre-IND meeting and 22 discuss what the drug development strategy is, 164 1 where you can get into how we are going to do proof 2 of principle and how we are going to do dose 3 finding. That is not done very often. 4 The second thing is that at the end of 5 Phase I, there will be a guidance that will be out 6 shortly. It will be probably called an End of Phase 7 IIA Guidance, but it is actually an end of Phase 8 I/IIA guidance. 9 That is an opportunity for the company to 10 summarize the data quickly, you would have to be 11 doing things on the fly, and come in and say what 12 you have learned. 13 Modeling and simulation can be used, if 14 you are quickly putting the information together 15 and you are doing PK/PD type relationships in Phase 16 I or Phase IIA, that could be put together to then 17 use along with--certainly PK/PD does not predict 18 outcome--but compliance does. 19 You talked about the neutropenic state of 20 the patients or not. All those sorts of things can 21 be accounted for in trial design. You were talking 22 about why aren't these Phase III trials being more 165 1 efficient and using fewer subjects, why can't you 2 just do one trial instead of two or three or four. 3 Well, if you want to do that, then, my 4 experience with the FDA over the years is that the 5 people generally respond very well to good science, 6 so that if you present a good scientific argument 7 at the end of Phase I or the end of Phase II, IIA, 8 or end of Phase II about what you are planning to 9 do in Phase III, then, that is a great dialogue to 10 have. 11 Why doesn't that happen more often? I 12 don't know, but the reality is people that are 13 interested in this topic are generally not as 14 influential as they need to be inside companies. 15 Now, I think you can use the FDA in the 16 ways that I have talked about, to try and gain that 17 leverage. That is what I primarily wanted to speak 18 about. 19 Thanks. 20 DR. EDWARDS: Thank you very much. That 21 is a very important point, and in side discussions, 22 it has come up on several occasions, especially 166 1 today, as well. 2 I think we are going to have to go on. 3 This afternoon's session is the current status of 4 dose selection for the first part, and then the 5 second part is improvement in dose selection 6 concepts. We will start now with the current 7 status, and I am going to ask George Drusano to 8 begin these presentations. Most of the discussion 9 is going to be at the end of all of the 10 presentations actually, so, George, let me ask you 11 to start off. Thank you. 12 III. Current Status of Dose Selection in 13 Antimicrobial Drug Development Programs 14 Academic Perspective 15 DR. DRUSANO: Thank you, Jack. I have 16 already been told that if I go over, I will be 17 shot, so that is clear, and I would first like to 18 start by thanking the Agency for providing the 19 invitation to give this presentation on getting the 20 dose right and the view from academia. 21 [Slide.] 22 Well, what are the determinants of 167 1 "getting the dose right?" Well, first, you have to 2 answer a question, and the first question that 3 needs to be addressed is what outcome is it that 4 you desire. 5 Multiple outcomes are reasonable. You 6 want a good clinical outcome, microbiological 7 outcome. You want to suppress resistance, you want 8 to have minimization of concentration-related 9 toxicities. 10 Well, one or more of those is what you 11 want you want to accomplish. Now, what are the 12 determinants of getting the dose right? 13 [Slide.] 14 Well, we can do all of those things and 15 then dose choice becomes an issue basically in the 16 Phase I time frame. Well, it is actually earlier, 17 but the proper data becomes available at this time, 18 because I, for one, I think that dose scaling, 19 allometric scaling, or PD/PK, is useful, but it can 20 be a little dangerous. 21 You really want to have human PK. Now, 22 because of this, the choice of outcome is limited. 168 1 Clinical outcome cannot at this time be an outcome 2 measure because you haven't at this time put the 3 drug into a patient that is infected. 4 So, the most common measure, then, is a 5 microbiological outcome determined either from in 6 vitro or animal model data. 7 [Slide.] 8 Now, I point this out to you. This comes 9 from our laboratory. By request, the name of the 10 drug has been changed to drug X, and all the data I 11 am going to show you today is from one single drug 12 and different manifestations of studies. 13 So, what one can see here is if you do 14 multiple doses of this particular drug--and this is 15 Pseudomonas aeruginosa--this a mouse thigh 16 infection model. You start out with around 7.9 17 logs of organisms, in the presence of granulocytes, 18 by the way, you can see that you need to have about 19 45 to 1 AUC to MIC ratio to achieve stasis, about 20 80 for a log kill, and up to 130 for 2 log kills. 21 Bacteremia usually gets shut off somewhere 22 between a 1 log and 1 log kill in this 169 1 circumstance. So, remember that number of about 2 80, so it is somewhere between 80 and 130. 3 [Slide.] 4 So, once a target exposure is chosen, what 5 other information do you like to have? Well, the 6 first thing that you would like to have is to 7 identify the sources of variability in the 8 circumstance. 9 So, as every is want to point out, I am a 10 little on the porky side, so that if you take a 11 look at somebody like Dr. Powers, and you gave him 12 a dose of drug, and you gave me that same dose of 13 drug, we would have two very different profiles of 14 drug, because he is thin and I am not. 15 So, how do we deal with this? We actually 16 take populations of individuals and give them doses 17 of drug, and we do population pharmacokinetic 18 analysis. This quantifies the variability in how 19 the drug dose is handled by a population. 20 We also have different target pathogens, 21 and for each target pathogen, there is a 22 distribution of MICs for the drug in question. So, 170 1 we have to have relatively large databases of these 2 MICs for the target pathogen that we care about, 3 and you also have to know your target product 4 profile, because it is going to make a difference 5 whether you are treating skin and skin structure, 6 meningitis, endocarditis, pulmonary infections. 7 The other thing that we need to know is 8 that, in general, only free drug is 9 microbiologically active. I don't want to go back 10 into that, we all know that there are times when it 11 is not quite mathematical, but, in general, to a 12 high degree of certitude, you want to pay attention 13 to free drug. 14 [Slide.] 15 Now, we can evaluate how well a specific 16 drug dose will attain the desired pharmacodynamic 17 target, and we do this through Monte Carlo 18 simulation, and all the Monte Carlo simulation is 19 doing for you is to give you some measure of the 20 variability of the drug exposure in the population. 21 You have to remember the limitations. 22 At this point, we are almost always 171 1 looking at volunteer data, so that is going to be 2 skewed relative to the data that one sees in 3 honest-to-God target populations. Usually, but not 4 always, that skew is conservative, so that the 5 clearances are, in general, higher, making the 6 inferences drawn somewhat conservative. 7 [Slide.] 8 So, just to take a 1 log drop, so the AUC 9 to MIC we saw for that was 81. On the x axis, you 10 see the MICs. We have a distribution of MICs for 11 Enterobacter cloacae and for Pseudomonas 12 aeruginosa. 13 We then do a 10,000 subject Monte Carlo 14 simulation, and what one sees is that at this MIC, 15 it's 100 percent target attainment, 100 percent, 16 100 percent, and then it falls off the end of the 17 table. When you get down to 1, you have got about 18 a 50 percent target attainment, and goes down to 19 zero. 20 Now, is that any good? The answer is if 21 all the MICs were out here, that would be 22 spectacular, if all the MIC distribution was out 172 1 here, it would really stink, but we can actually 2 get an idea of how good it is by doing a target 3 expectation. 4 So, you just take the fraction of 5 organisms at that MIC, times the target attainment 6 rate at that MIC, this one times that one, this one 7 times that one, and so on, and you add it all up. 8 It's a fancy name for taking a weighted average. 9 So, when we do that for Pseudomonas, this 10 particular dose of drug gives you an expectation of 11 hitting 81 at about 66 percent. For Enterobacter, 12 that number is about 88 percent. Remember, that is 13 a little on the conservative side because this is 14 volunteer data. 15 [Slide.] 16 Okay. What are the questions now? First, 17 was the target correct, because as we talked about 18 earlier, this thing has a 95 percent confidence 19 bound around it. There is no one right target, and 20 it even may change by organism and by site. 21 The second question, is the target 22 attainment rate adequate? Well, we will talk about 173 1 that in a second. 2 So, the target, a 1 log decline is 3 reasonably conservative, but you have to be careful 4 about the infection site. 5 Now, the answer to the second question 6 depends on the patient and the consequences of 7 being wrong. For instance, if you have a 8 granulocytopenic patient, the consequences of being 9 wrong are severe. The same thing for a meningitis 10 patient. If you have an uncomplicated skin and 11 skin structure patient, it is not so bad that you 12 could be wrong. 13 Now, is it adequate, the target attainment 14 rate? Well, that depends on whether you are 15 treating your mother or your mother-in-law, it's in 16 the eye of the beholder. 17 [Slide.] 18 So, now, what now? Now, it is important 19 to recapitulate the analysis in real patients in 20 the Phase II environment. Why? 21 Well, the PK that you got from that is 22 determined in the target population. Two, the 174 1 correlation, if you can make it, with preclinical 2 animal or in vitro data, provides near certainty of 3 "no surprises" in Phase III except for toxicity 4 issues, and you are seeing real world organisms, 5 and these can be gauged against the original MIC 6 distributions. 7 Again, when you do this, it actually, 8 usually comes out conservative. So, Monte Carlo 9 simulations can then be recalculated with real data 10 before you move into the Phase III environment. 11 [Slide.] 12 Here is an example of a drug that was 13 taken from a nosocomial pneumonia study, and 750 14 milligrams of this drug was given as a 1.5 hour 15 constant rate I.V. infusion. There were 58 16 patients that were studied. 17 The sampling design was a stochastic 18 design that we put together at our group, and we 19 only really had 6 samples, but we were able to 20 capture a lot of the system information with those 21 6 samples. 22 I only point out the clearance here for 175 1 this drug, that the clearance in a normal volunteer 2 population would be about 11 to 12 liters per hour. 3 The mean and median clearance, I apologize for 4 this, is between 6 and 7 liters per hour. So, this 5 is telling you of the difference between volunteer 6 populations and true target populations 7 particularly in terms of important factors like the 8 drug clearance. 9 [Slide.] 10 We were then able to have 47 of those 58 11 patients where we had a recovered pathogen from 12 that nosocomial infection study, the MIC to the 13 drug in question, an outcome that was a 14 microbiological outcome in this particular 15 instance, and the PK, so that we could do the 16 population modeling, get MAP Bayesian estimates for 17 each of the patients, and calculate a specific AUC 18 to MIC ratio. 19 The first thing we did was we took 20 classification regression tree analysis to try to 21 find a breakpoint, and the breakpoint that we came 22 up with was a breakpoint of 87, which was really 176 1 quite close to the AUC to MIC of 81, that we had 2 gotten in the mouse thigh model for a 1-log drop. 3 We then did some model building. There 4 was a preregistered analysis plan, and with that 5 preregistered analysis plan, there were only two 6 things that had an impact upon the microbiological 7 outcome. 8 One was whether or not you attained the 9 AUC to MIC of 87. The second was the age of the 10 patient, and the model was statistically 11 significant. We did look at the positive and 12 negative predictive values, the sensitivity and the 13 specificity, and it was quite reasonable, and these 14 data will be published in the May 1st issue of the 15 Journal of Infectious Diseases. 16 [Slide.] 17 For those who like pictures, here, on the 18 x axis is the age of the patient; on the y axis is 19 the probability of organism eradication, and this 20 curve is for those that attained the breakpoint 21 value of 87, and this curve is for those who did 22 not. 177 1 As you can see, there is clear separation 2 between these curves, and actually, when you look 3 at it, what you can show is that if you are above 4 65 years of age, this is statistically 5 significantly worse. So, there is actually a 6 breakpoint in age, as well, as it is really the 7 older patients that benefit the most from attaining 8 the breakpoint AUC to MIC value of 87. 9 [Slide.] 10 We can then close the loop. We have 11 identified an AUC to MIC breakpoint in patients 12 that is quite concordant with the number that we 13 got out of the mouse thigh infection model, and now 14 we can redo our AUC to MIC target of 87 and do the 15 Monte Carlo simulation and the expectation. 16 Now, this is a little different when you 17 actually start to take a look at this distribution. 18 [Slide.] 19 When you do the expectations for this AUC 20 to MIC breakpoint for Pseudomonas, now it is 72 21 percent, and for Enterobacter, the target 22 attainment rate is around 92 percent. 178 1 [Slide.] 2 What about resistance suppression, can we 3 do this? 4 [Slide.] 5 These are data that were published from 6 our laboratory in the Journal of Clinical 7 Investigation last year, and what you see is 8 oftentimes when we beat on a large population of 9 organisms, we see an initial fall followed by 10 regrowth. 11 This is the effect of having one drug 12 exposure on two disparate populations of organisms, 13 one that is susceptible and one that is less 14 susceptible, and when you deconvolute the effect of 15 the single exposure on the two populations, a funny 16 thing happens on the way to the forum. 17 When you look at the sensitive population, 18 you really, really have an impact. When you look 19 at the exposure and its impact with the resistant 20 population, you see unrestricted amplification of 21 that resistant subpopulation, so that when you put 22 the results of that single drug exposure on both 179 1 populations together, it's not at issue where this 2 comes from. 3 [Slide.] 4 We population modeled a number of 5 different exposures simultaneously using the Blue 6 Horizon machine out at UC/SD Supercomputer Center. 7 These are the point estimates of the parameters. I 8 have been enjoined by the Agency from not showing 9 the five differential equations we used to do this, 10 because nobody wanted to make everybody nauseous 11 after lunch. 12 So, we really did do it, that's what we 13 got. 14 [Slide.] 15 Just to show you, you can then show that 16 this is the MAP Bayesian posteriors predicted 17 observed plot. This is for the total bacterial 18 count. What you see is the r 2 is about 0.93, the 19 slope near 1, and we have a relatively small Y 20 intersect. 21 So, we did a good job, the model did a 22 good job of describing the effect of the drug on 180 1 the total population. 2 [Slide.] 3 Here is the effect of the drug on the 4 resistant population, again, slope near 1, small y 5 intersect, and an r 2 of 0.94. So, we were able to 6 do this for this drug and Pseudomonas aeruginosa. 7 [Slide.] 8 You can then use those numbers to 9 calculate an AUC to MIC that will shut off the 10 growth of the resistant mutants, and that number is 11 about 157 to 1 of total drug. It's about 110 to 1 12 of free drug. 13 [Slide.] 14 We decided to do a prospective validation. 15 Now, I am showing you the AUC to MIC of 52, which 16 we calculated would optimally amplify the resistant 17 mutants, and the solid lines are not best-fit 18 lines, they are prediction lines from what we had 19 done previously, and we scattered around those 20 prediction lines what happened to the total 21 population and the resistant population. 22 If you actually look, you would say that, 181 1 hey, that's not so bad if you did not look at the 2 resistant subpopulation. You would show that you 3 were able to drop the total population about a half 4 a log with that AUC to MIC ratio. 5 When you, however, look at the resistant 6 population, you are amplifying the resistant 7 mutants. You are killing off a few of the 8 susceptible organisms, but you are replacing them 9 with resistant organisms. 10 So, again, we were able to predict quite 11 nicely, what was going to happen to the resistant 12 mutant population as a function of a suboptimal 13 exposure. 14 [Slide.] 15 When we used 157 to 1 to shut off the 16 amplification of the resistant mutant population, 17 it did not amplify. I will also tell you we have 18 recapitulated this study in an in vitro 19 hollow-fiber infection model with another drug of 20 the same class, and with the same Pseudomonas 21 aeruginosa, and you find exactly the same thing. 22 So, in that circumstance, you can do this. 182 1 [Slide.] 2 So, we were able to determine how the 3 overall population responds to pressure from this 4 drug. More importantly, we were able to model the 5 resistant population, choose a dose based on 6 simulation to suppress the resistant mutants, and 7 the prospective validation, the only one in the 8 literature of which I am aware, demonstrated that 9 the doses chosen to encourage and suppress the 10 resistant mutants did, indeed, work. 11 [Slide.] 12 We are almost done. What about having a 13 high probability of attaining a good outcome while 14 not encumbering the patient with a high probability 15 of a concentration-related toxicity? 16 Now, we will use aminoglycosides for this 17 where there has been developed concentration-effect 18 and concentration-toxicity probability 19 relationships. 20 [Slide.] 21 The efficacy probability function was 22 generated by Angela Kashuba and Joe Bertino. They 183 1 were kind enough to ask me to do the math on this, 2 published in AAC in '99, and the toxicity 3 probability function was developed out of Mike 4 Rybak's lab as part of a randomized, double-blind 5 trial, and again they were kind enough to ask me to 6 do the math. 7 [Slide.] 8 So, for the effect, you see on the x axis 9 AUC to MIC ratio, and the probability of resolution 10 by different days of therapy. For this, we will 11 use 7 days of therapy. 12 [Slide.] 13 For the toxicity, this again was published 14 in AAC, and you see on the x axis the daily area 15 under the curve of aminoglycoside with and without 16 concurrent vancomycin. For this, we will use the 17 without vancomycin curve, but you can clearly 18 relate the probability of nephrotoxicity to the 19 daily exposure to the aminoglycoside. 20 [Slide.] 21 But here we have both of them together, 22 and on the x axis, one is AUC to MIC, one is AUC, 184 1 and so in order to get around that, we put AUC on 2 the x axis in all instances, and then probabilities 3 of both effect and toxicity, and then we sorted out 4 by MIC, so that if you have a nice, sensitive E. 5 coli or Klebsiella, where you have an MIC of 0.25, 6 you can get a 90 percent or so probability of 7 resolution, and not encumber much more than 1 or 2 8 percent probability of toxicity. 9 As you go to 0.5, now you can see it is 10 around there, it is around 3 or 4 percent 11 intersecting the toxicity probability curve, so it 12 is a little narrower of a therapeutic index. 13 When you go 1, now the wheels start to 14 come up, and what you see is in order to have an 90 15 percent probability, you have to encumber the 16 patient with about a 33 percent probability that 17 they are going to be toxic. 18 At 2, you may as well not even bother 19 because to get 90, is out here, and at that point, 20 you have a 95 percent probability of getting toxic. 21 So, what this is telling you is we can do 22 this. This is the simple way to do it. There is a 185 1 fully stochastic way that has been worked out with 2 multiple model stochastic control with linear 3 dynamics and quadratic cost functional, but this is 4 just the easy way to do it. It can be done. 5 [Slide.] 6 So, why go through all this mathematical 7 nonsense? Because at the end of the therapy is a 8 patient who would appreciate getting better without 9 toxicity. Also, because choosing the dose 10 correctly, this allows trials with anti-infectives 11 to have the lowest Phase III failure rate of any 12 therapeutic area. 13 In so doing, we can push the development 14 of these drugs faster, cheaper, and better, so that 15 we can keep those in the industry that are still 16 in, and hopefully, attract back some of the people 17 in industry that have abandoned this therapeutic 18 area. 19 In the words of my favorite Hollywood 20 movie star, "That's all, folks." 21 DR. EDWARDS: Thank you very much, George. 22 That was a very nice discussion. 186 1 We are going to move right along now to 2 Dennis Grasela from BMS. 3 Dennis, please go right ahead. 4 Industry Perspective 5 DR. GRASELA: I would like to begin by 6 thanking the conveners of this meeting for the 7 opportunity to speak here, and to also say that 8 this is very difficult presentation, for two 9 reasons. 10 One, it is an hour after lunch, and, two, 11 it follows Dr. Drusano. 12 [Slide.] 13 The selection of dose for the industry is 14 a critical milestone in the drug development 15 process, for lots of reasons, most of which have 16 been discussed today, but of all those reasons that 17 deal with the patient, with outcome, with 18 regulators, with everything else, there is whole 19 other large component of the industry that is 20 mobilized when you say what the dose is. 21 It determines how much is manufactured, 22 what dosage forms are produced, for some insoluble 187 1 drugs it defines whether or not a dose formulation 2 can be produced, and in some cases, a valuable drug 3 may not be taken forward if we think the dose is 4 600, the solubility is brick dust, and they say 5 they can't put any more than 10 milligrams into a 6 pill. This is an incredible challenge for all of 7 us. 8 In summarizing yesterday, Dr. Edwards 9 punted a little bit, but I will summarize it in one 10 way - we are all in this together, and this is 11 really hard to do, and I hope that more of these 12 discussions happen, so that we can get it. I don't 13 think we will ever get it completely right, but we 14 will get as good at it as we possibly can. 15 [Slide.] 16 There are three parts to my presentation. 17 First, I would like to talk about the way that we 18 have embraced exposure-response approach to dose 19 selection at Bristol-Myers Squibb. 20 Second, the optimist's view of the factors 21 that are driving the use of PK/PD based drug 22 development. 188 1 Third, my view of potential cost savings. 2 Again, this is my perspective. 3 [Slide.] 4 As I begin the exposure-response 5 discussion, I just want to preface it by saying it 6 is assumed, I want to assume two things. One, that 7 we know a lot about the product, we factored in 8 protein binding, we factored in tissue 9 distribution, we factored in all of the things that 10 we have talked about are important to know about 11 the drug biopharmaceutics, et cetera. 12 The second piece is to ask your indulgence 13 in that some of the examples I will show, the data 14 became available well after the decisions were 15 made, but we utilized all that information for the 16 second compound that came along with it, which was 17 the driver for my question about how much can you 18 bring forward from what you know before to get it 19 right the second time around, particularly when 20 development or discovery compounds come in 21 clusters, you are usually trying to pick between 22 one, two, or three compounds. 189 1 I will digress by saying the FDA has 2 helped a lot by the concept of the screening IND, 3 which enables us to take some uncertainty and some 4 risk out of it by taking more than one compound in 5 at a time. 6 [Slide.] 7 The road to dose selection, in my view, is 8 really four basic principles. 9 First, is you need to know the 10 microbiological hurdle, and that can be figured out 11 in some ways by the in vitro MIC values. Given Dr. 12 Rex's comment earlier on, I am not so sure how much 13 we know about the hurdle after the smoke clears. 14 The second is to define the 15 pharmacodynamically-linked parameter and the target 16 value as best we can from in vitro hollow-fiber 17 models and in vivo animal models of infection. 18 The third is to use PK/PD modeling in 19 proof of concept studies to examine the 20 exposure-response relationship. This is really 21 hard to do. It is often about how you can not only 22 garner support from your clinical research 190 1 colleagues, as indicated by Dr. Powell, but it is 2 also how responsive the clinical sites can be to 3 collect the minutia, and that is the way they think 4 about the information we are trying to collect is 5 minutia. Heck, they are trying to just make sure 6 the patient shows up that day and they can remember 7 what day of the week the person actually came in. 8 We are asking them to write down what time they 9 took the dose. 10 Writing the case report forms is 11 incredibly difficult. For one study, we needed to 12 know what the time of their last meal was. They 13 told me about the bowl of ice cream they had the 14 night before. It was a once-a-day drug. I wanted 15 to know what time breakfast and the meal was the 16 day before that, and unless you are doing real-time 17 data clean-up, this is incredibly difficult because 18 what you get at the end of the day is a lot of 19 unusable data, and you wind up, when you consider 20 that, as Dr. Drusano had talked about, you need an 21 organism, you need a patient that has some outcome. 22 You actually need to know what his exposure is, and 191 1 it all has to happen in the same person, and you 2 need to know the date and time of the dose and when 3 those samples were collected for that individual. 4 When you take all of that together, the 5 recovery rate of subjects is incredibly low, 6 despite how many patients you include in your Phase 7 II and III clinical trials. 8 When all that is said and done, we try to 9 combine all that information, use Monte Carlo 10 simulation to define the dose and schedule for 11 Phase III clinical trials. I might add that the 12 guys that manufacture the stuff want to know this 13 actually before you actually got it into your first 14 patient. 15 [Slide.] 16 The hurdle. The MIC distribution. I am 17 sure if I have asked people in the room when they 18 got their first discovery compound, they had an MIC 19 90. What they forgot to tell you it was an on an N 20 of 10 isolates. 21 I think if we looked across the history of 22 compounds, you realize that the MIC 90 floats. 192 1 Sometimes they go up, sometimes they go down. The 2 only message is here, and we have the luxury of our 3 SENTRY program, so that we know a lot about the MIC 4 distribution of our compounds as we bring them in, 5 but you need to know that you need to continually 6 monitor that to make sure you are not way off on 7 that assessment, and you also need to know what the 8 shape of that curve looks like. 9 This distribution is very different than 10 when it is skewed to the left or skewed to the 11 right. 12 [Slide.] 13 The second part that I talked about is the 14 in vitro models, and we have heard a number of 15 eloquent presentations about the in vitro 16 hollow-fiber. This happens to be an example out of 17 Phil Lister's lab for one of our quinolones, 18 essentially doing dose fractionation studies, et 19 cetera, from the history of quinolones. 20 I hope we can agree that the AUC to MIC 21 ratio is the target. In vitro hollow-fiber gives 22 you a window of free drug AUC to MIC ratios, and I 193 1 don't want to debate free drug. From an industry 2 perspective, I am always going to err on the 3 conservative side, and if I think free drug is 4 going to work, I am going to calculate the dose 5 based on free drug. I can always be a little wrong 6 the other way, but I am going to calculate it based 7 on free drug. 8 This data suggested that AUC to MIC ratio 9 of about 30 is what we needed for the pneumococcus. 10 [Slide.] 11 We next went to an animal model, and this 12 is with Dave Nikalel's [ph] lab, and this is in a 13 neutropenic mouse model. Essentially, an AUC to 14 MIC ratio of somewhere between 30 and 50 gives you 15 100 percent survival with a free AUC to MIC ratio 16 of somewhere between 30 and 36, 37, so we are 17 getting information. 18 [Slide.] 19 We know what the MIC distribution is, we 20 know what the in vitro hollow-fiber is telling us, 21 and we know what the animal models are showing. 22 Now, here is the hard part. For this 194 1 particular quinolone, we introduced population 2 pharmacokinetics into Phase II and Phase III. This 3 picture is actually combining two, Phase III trials 4 in two different respiratory infections and 5 throwing in the comparator quinolone to get enough 6 information to get an exposure-response 7 relationship. 8 Essentially, this is 58 organisms and 58 9 patients that we had all of the information come 10 together on. Basically, what it shows is that an 11 AUC to MIC ratio above 33 gives you a high 12 probability of clinical response, hence, the 13 feedback loop gets closed to around a free AUC to 14 MIC ratio of about 30 for this quinolone and the 15 pneumococcus. 16 [Slide.] 17 When we use that target and look at the 18 target hit rates using Monte Carlo simulation, it 19 suggested if the target is 30, that the dose we had 20 selected for this drug provides a 94 percent 21 response rate--I am sorry, not a response rate--but 22 a target hit rate for this compound. 195 1 When we looked at the same patients that 2 we pooled from those two, Phase III clinical 3 trials, we had microbiologic eradication in 92 4 percent of those subjects. 5 I would just like to digress for a minute. 6 This is talking a lot about the efficacy piece of 7 the equation, but dose selection continues on 8 through Phase III and the post-marketing 9 surveillance. 10 For low-frequency safety issues that may 11 be related to pharmacogenomic piece, that needs to 12 be picked up in Phase III and post-marketing 13 surveillance, and needs to continue to be looked 14 at. 15 In addition, with this information, as 16 MICs begin to drift as the compounds are on the 17 market, it provides the opportunity, in my view, 18 for a course adjustment, perhaps a dose adjustment 19 unless there is major shifts in the MIC values. 20 [Slide.] 21 Now, from my optimistic opinion, what are 22 the factors driving the use of PK/PD based drug 196 1 development? John Lazor called it "science-based 2 drug development," I will call it "knowledge-based 3 drug development." I think we are heading in the 4 same direction. 5 These principles allow dose selection, 6 more importantly, dose conformation. They allow 7 you the ability to examine the effect of 8 administering a dose not studied during the 9 development process. 10 What is the probability of if you lower 11 the dose or raise the dose, what is the probability 12 of your target hit rates? 13 I think we can use this as select target 14 indications. I think Dr. Powers was alluding to 15 this a little bit. If your drug isn't going to get 16 the job done, you probably ought to not go there, 17 or you need to combine it with another drug to fill 18 your spectrum gap, and just an overall enhanced 19 understanding of the drug. 20 [Slide.] 21 Regarding that enhanced understanding of 22 the drug, I had indicated we included population PK 197 1 into our Phase II and III trials for this 2 particular drug. I am a fan of population PK for a 3 number of reasons. 4 One, I think it can help explain 5 differences in response among individuals receiving 6 the same dose. Is it a covariate, is it a genomic? 7 It can help identify at-risk subpopulations and 8 define risk-benefit ratios and/or risk management 9 strategies. 10 We talked a little bit about the 11 clarithromycin study yesterday, and I am intrigued, 12 I will read up on that, but I wonder whether the 13 folks that had toxicity from clarithromycin were 14 predisposed to that by some reason, and knowing 15 that might enable us to do a risk assessment in 16 terms of using that dose should it be warranted in 17 selective patients. 18 Also, to help guide the use of 19 pharmacogenomics, if you don't have an exposure 20 reason for an adverse event, and you don't have an 21 individual demographic reason for that adverse 22 event, can it lead you down the path of a genomic 198 1 reason for that event. 2 [Slide.] 3 What are the external factors? We talked 4 about some of those today. There is FDA 5 expectation and there are opportunities. The 6 expectations are set through the guidances and the 7 opportunities, in part, are provided by the 8 opportunities to discuss the plan, the pre-IND 9 meeting, the End of Phase IIA meeting, and other 10 venues. 11 The FDA Modernization Act of 1997, which I 12 will talk about in next couple of slides, also 13 provides opportunities to streamline the drug 14 development process, but there are also global 15 expectations, data-driven responses to regulatory 16 questions. Regulatory agencies contain really 17 smart folks that come up with really good 18 questions. Those questions are really hard to 19 answer when you don't have data. 20 The other piece that is becoming more 21 important as you do risk management assessments or 22 "what if" scenarios. The stronger your base, the 199 1 stronger your foundation, the more able you are to 2 answer those questions in a collaborative way. 3 Also, the notion of benefit-risk 4 assessment. Should you run into a low-frequency 5 toxicity, what are you going to do? You need to 6 come up with risk-benefit assessment for that, for 7 yourself, for the company, and for the public at 8 large. 9 [Slide.] 10 Just coming back to the FDA Modernization 11 Act. Most of you probably know much about it. 12 There are two elements - Section 111, which talks 13 about bridging PK studies, bridging PK studies in 14 pediatrics. 15 [Slide.] 16 And Section 115, which we have talked 17 about a lot, and that is the single adequate and 18 well-controlled trial when there is supporting 19 PK/PD information. 20 [Slide.] 21 Potential cost benefits. These are the 22 arguments that I try to use when I am running into 200 1 resistance to incorporate population PK into some 2 of the trials. One is that population PK can 3 sometimes obviate the need for selected clinical 4 trials, the age-gender study, renal impairment 5 studies, or modified renal impairment studies, 6 these are easy wins if you collect the data 7 appropriately. 8 To position the company to be able to use 9 ICH E5 guidelines, to use PK bridging studies for 10 submissions in Japan and other regions, and to also 11 position the sponsor to use FDAMA, rewrite issues, 12 Sections 111 and 115, as potential benefits. 13 [Slide.] 14 Other cost benefits are selection of 15 indications based on PK/PD evaluation of the 16 antimicrobial spectrum. I alluded to this a few 17 seconds ago, in that if your drug is not going to 18 be able to get the job done, it is better to know 19 now, and not pursue that indication. 20 Smaller sample sizes associated with 21 exposure-response versus dose-response. 22 Selection of optimal dose may lower sample 201 1 size requirements for non-inferiority trials. 2 [Slide.] 3 I am not a statistician, but this is my 4 sort of fundamental approach at looking at this. 5 If you have an anticipated response rate 6 for comparators of 85 percent, and you have a 7 projected response rate or target hit rate for your 8 compound of 80, 85, 90, or 95 percent, your sample 9 size are listed to your right. 10 If you remember the Monte Carlo simulation 11 for our compound, it suggested the hit rate was 94 12 percent. The clinical trials data for Strep pneumo 13 suggests that the efficacy was at 92 percent. If 14 you assume the 90 percent difference, your sample 15 size would be 83 per arm versus 219 per arm. That 16 is a substantial savings in my view if you have the 17 gumption to go there. 18 [Slide.] 19 Cost benefits. Higher quality 20 submissions. I consider a submission like a 21 manuscript. You either pay now or pay later. If 22 you put together a really good submission or really 202 1 good manuscript, your reviews are minimal and you 2 are published quickly. 3 Similarly, if you are a reviewer and you 4 get a really good manuscript, you are likely to 5 review it very quickly and get it in the 6 literature; if it's not, you are likely to have a 7 real struggle time with it. 8 I think a good submission will always 9 facilitate a regulatory review. It will enhance 10 the relationship with our regulatory colleagues and 11 minimize post-submission questions. Once you have 12 made your submission, it is sometimes difficult to 13 go back and pick through all that data to answer 14 very specific questions. 15 I think it can facilitate transition with 16 novel dosage forms based on PK studies if you know 17 the PK/PD relationships and provide a basis for 18 market-driven differentiation. 19 [Slide.] 20 In summary, selecting the optimum dose for 21 the treatment of infection is important in order to 22 maximize efficacy, minimize toxicity, and minimize 203 1 resistance development. 2 I think you have seen that we have the 3 tools to at least get an idea of how to do this 4 best, and using an exposure-response approach to 5 dose selection can facilitate knowledge-based 6 decisionmaking, optimize trial designs, and 7 streamline development and post-development costs. 8 Thank you. 9 DR. EDWARDS: Thank you very much, Dennis, 10 that was a very nice discussion. 11 We will now move to Frank Pelsor from the 12 FDA. 13 Frank. 14 FDA Perspective 15 DR. PELSOR: Good afternoon. I will be 16 presenting an FDA perspective on the current status 17 of dose selection and antimicrobial drug 18 development programs. 19 [Slide.] 20 I have two objectives for my presentation 21 this afternoon. The first is to discuss why a 22 well-developed rationale for dose selection is 204 1 important to the antimicrobial drug development 2 process, and secondly, to discuss some of the 3 Office of Clinical Pharmacology and 4 Biopharmaceutics experience with dose selection and 5 antimicrobial drug product applications. 6 [Slide.] 7 Why is dose selection important? We have 8 had this question come up several times during the 9 course of the couple of days here, and I feel like 10 I am preaching to the choir, but nevertheless, stay 11 with me on this, and then I will take up some 12 examples that I think will be informative. 13 Antimicrobial drug therapy is a complex 14 endeavor that includes more components than the 15 target of the drug, the bacteria. The patient 16 status, for example, immunocompetence, and the 17 adaptability of bacterial must be taken into 18 consideration when selecting an antimicrobial drug 19 regimen for therapy. 20 In addition, the drug itself may have 21 complementary effects, such as an inflammatory 22 activity which will be beneficial to the patient. 205 1 However, for our discussion today, we must limit 2 our focus. At some risk of oversimplification, our 3 attention must be focused by distilling 4 antimicrobial drug therapy into a couple of 5 concepts. 6 The first concept. For an effective 7 drug-bacterial interaction, the requirement is to 8 deliver free drug to the site where the infectious 9 organism is located. Drug delivery can be 10 accomplished by varying the magnitude and timing of 11 input. 12 The second concept is that we must deliver 13 effective therapy while remaining within the domain 14 of acceptable toxicity. Adjusting the inputs can 15 help manage the toxicity. 16 [Slide.] 17 The next slide, which is already up, 18 illustrates the relationship between these two 19 concepts using dose-response curves for efficacy 20 and toxicity. 21 Traditionally, the dose for an 22 antimicrobial drug is selected from the upper 206 1 plateau of the efficacy dose-response curve, and 2 that is the top arrow, and the toxicity has been 3 relatively low, the second arrow. 4 In contrast, newer antimicrobial drugs 5 appear to be accompanied by dose-response curves 6 for toxicity that are closer to the dose-response 7 curves for efficacy. This shift in curves is 8 illustrated by the center curve. 9 [Slide.] 10 For cases where management of these 11 unwanted toxic effects may be possible, perhaps 12 through dose reduction, it is not clear what effect 13 the dose reduction will have on efficacy of the 14 antimicrobial drug, because we usually have 15 information for a single dose only. 16 In addition, for special populations where 17 dose adjustment may be necessary, it is not always 18 clear what effect the dose adjustment will have on 19 efficacy. Ultimately, failure to have the 20 necessary understanding of the relationship between 21 effective dose response and toxic dose response can 22 result in delaying drug product approval or even in 207 1 denying approval. 2 [Slide.] 3 Drug toxicity management attempts to 4 maintain optimal drug therapy within constraints 5 imposed by the drug's toxicity. We know that we 6 can use dose adjustment for higher risk patients, 7 such as pediatric patients and patients with renal 8 and/or hepatic impairment to achieve desired 9 exposures. 10 In order to deliver optimal therapy in 11 these situations, we must have a thorough 12 understanding of the location of the dose efficacy 13 response curve. Appropriate interpretation and use 14 of pharmacokinetic and pharmacodynamic information 15 can help us to locate the dose. 16 As part of our review of information that 17 sponsors submit to the Agency, we asked some 18 critical questions about the rationale for dose 19 selection with an emphasis on exposure-response 20 relationship, we asked the following: Are the dose 21 and dosing regimen consistent with the known 22 relationship between dose or concentration and 208 1 response? And, secondly, are there significant 2 risks related to the clinical pharmacology issues, 3 for example, any changes in exposure related to 4 intrinsic or extrinsic factors, and we try to 5 determine how these risks should be managed, 6 perhaps through dose and/or dosage adjustment. 7 [Slide.] 8 To this point in the presentation, I have 9 been describing the importance of dose selection, 10 however, we recognize that dose selection is an 11 intermediate output of a process. Therefore, I 12 want to shift gears now and focus on the process. 13 As Agency reviewers, we expect to see dose 14 selection rationale based on a multi-stage process 15 shown in this diagram. Following the diagram in 16 clockwise fashion, each stage serves as a building 17 block for the next. 18 Proceeding from preclinical in vitro and 19 animal model studies through Phase I human 20 pharmacokinetic studies, and Phase II human dose 21 and activity studies, we should have explored the 22 PK and PD indices and the targets. By the time we 209 1 establish proof of concept in Phase II, we should 2 understand pertinent PK and PD relationships and 3 potential doses that may be selected for Phase III 4 clinical trials. 5 Phase III clinical trials should be 6 designed in such a way that the data coming out of 7 the trials not only can be used to test hypotheses, 8 but that the data provides feedback about the 9 relationship between PK/PD indices and the dose. 10 The process also allows for modification 11 of future studies by permitting feedback at various 12 stages, but I want to emphasize we do not view the 13 process as bidirectional. I am separating 14 bidirectional now from feedback. That is, the 15 selection of dose followed by construction of a 16 database to support the dose is not appropriate in 17 our view. 18 Ultimately, we would like to appreciate 19 how the various pieces of the puzzle fit together, 20 the pharmacokinetics, the pharmacodynamics, the 21 relationships, the dose, the efficacy, and safety, 22 and how the information can be further transformed 210 1 into knowledge for us in evaluating other 2 antimicrobial drug products. 3 [Slide.] 4 The rationale for dose selection is highly 5 variable in the regulatory packages that we see. 6 It really is not always clear to us how the sponsor 7 selected the dose to pursue an indication. 8 We have seen some well-described 9 rationales here today, and we have see 10 well-described rationales in submissions. I will 11 discuss an example later. However, we have seen 12 rationales that have been based on information that 13 really is not informative. I will discuss as 14 example, as well. 15 The third example that I will present is 16 an example where it appeared to us that there were 17 other issues driving the dose selection, such as 18 marketplace and patient compliance or convenience. 19 Marketing and patient management issues certainly 20 merit consideration, however, the focus of our 21 discipline and our review is on a PK/PD based 22 rationale. 211 1 [Slide.] 2 This slide describes the framework in 3 which we review a sponsor's package with respect to 4 dose selection. I have taken it from knowledge 5 engineering and data mining literature. You may be 6 more familiar with it in terms of the knowledge 7 pyramid. I call it the Data Information Knowledge 8 Continuum, however, I have added a dimension 9 described how the continuum relates to levels of 10 understanding. 11 (Slide.) 12 This is the first example of the 13 rationale. This slide illustrates a 14 not-too-uncommon description of the dose-selection 15 rationale. The PK/PD target for time above MIC is 16 derived from the mean concentration time profile 17 and mean MIC values. 18 The sponsor tells us that the 19 100-milligram dose was selected because it provides 20 concentrates greater than the MIC for the 21 appropriate pathogen for 40 percent or about 10 22 hours in a 24-hour dosing interval. 212 1 It turns out that the Phase III clinical 2 trials provided the necessary evidence to support 3 approval of the product, but in terms of being able 4 to understand the rationale for the dose selection, 5 this really is what I would call a black-box 6 approach. 7 The dose may be effective, but the 8 description of the rationale really is not 9 sufficient to be of any help to us. 10 (Slide.) 11 In this slide, I am giving an example 12 where we have a package. In this case, the Agency 13 reviewers analyzed the information, which contained 14 MICs for pertinent organisms, drug concentration 15 time profiles from Phase I studies, and protein 16 binding data to determine if the dosing regimen 17 tested by the sponsor could be supported. 18 Our finding was that dose would not be 19 appropriate to test in a clinical trial for the 20 indication being sought. The results of the 21 clinical trial revealed that the drug was inferior 22 to the control. 213 1 The example points out the detailed 2 consideration of PK and PD information for dose 3 selection can be a powerful tool in antimicrobial 4 drug development. The ability to identify drugs 5 early that will fail should be important knowledge 6 to have. 7 [Slide.] 8 Lastly, this is an example where the 9 package contained detailed information for 10 microbiological data in vitro and animal model 11 data, and Phase I pharmacokinetics data. The Phase 12 II study design included several doses based on an 13 analysis and information we were able to understand 14 the relationships that results in the doses 15 selected for Phase II. 16 [Slide.] 17 In summary, I have shared with you the 18 view that we feel strongly that the rationale for 19 dose selection of antimicrobial drug products 20 should be linked to the supporting studies and 21 data. 22 The basis for dose selection is critical 214 1 to understanding how to manage changes in patient 2 drug exposure while maintaining acceptable safety. 3 However, it usually is not clear to us where we are 4 on the dose-response curve. Should the aim of dose 5 selection be the search for a single point to find 6 the dose with the highest probability of efficacy? 7 Further attempts to relate PK and PD 8 targets to clinical outcome are rare. 9 [Slide.] 10 I have also shared with you that we see 11 large differences in the level of detail included 12 in rationale to support dose selection. The 13 rationale is seldom presented in terms of a data 14 information knowledge continuum. Often the dose 15 selection rationale is based on PK/PD targets 16 derived from mean data, so that we have no 17 appreciation for the variation. 18 The rationale for dose selection is not 19 always apparent and dose adjustments based on PK/PD 20 targets is often difficult to assess. 21 Thank you. 22 DR. EDWARDS: Thank you very much, Frank. 215 1 I really apologize for that technical problem. 2 That is next to the worst nightmare, the worst 3 being that the slides don't appear at all. 4 We are going to move right along and now 5 enter the second phase of the discussion, which is 6 an exploration of methodology that we might utilize 7 to enhance the current tools which are available. 8 Let me begin right away with Hartmut from 9 the University of Florida. Please go right ahead. 10 IV. Improvement in Dose Selections Through Clinical 11 Applications of PK/PD in Antimicrobial 12 Drug Development Programs 13 Academic Perspective 14 DR. DERENDORF: Thank you. 15 First of all, let me also on behalf of 16 ISAP thank the FDA for hosting this meeting and 17 making it possible. I think this is so important 18 to bring everybody together who has interest in 19 this field and have this exchange of ideas, and the 20 discussion so far has been great. 21 [Slide.] 22 What I would like to do in the next 10, 15 216 1 minutes is make some proposals of how we can do 2 things better, and let me start by this very basic 3 slide that shows you again what we are doing with 4 PK/PD. 5 We are trying to break down the final 6 outcome that we have, and that, of course, is the 7 relationship between effect as a function of time, 8 and PK/PD separates the contributions that we get 9 from exposure, which is the time-concentration 10 relationship, and the response, which is the 11 concentration-effect relationship. 12 I would like to point out that we are in a 13 very unique situation with anti-infectives, because 14 that is one of the few cases where we can look at 15 the pharmacodynamics separately, we can move it 16 outside the body and look at it and study it in an 17 in vitro system or the animal model. That is truly 18 unique and that is something that we should take 19 full advantage of. 20 The pharmacodynamic portion here, this is 21 what microbiology contributes, and we have to find 22 intelligent ways to link that to the exposure. 217 1 [Slide.] 2 If you look historically at the field in 3 the last 10, 15 years, three indices, as they are 4 called, have emerged and are used today that link 5 the PK to the PD, and they are, as you know, the 6 time above MIC, the peak MIC ratio or the area 7 under the curve MIC ratio, and depending on the 8 group of drugs, we have certain parameters, certain 9 indices that we prefer and use. 10 However, if you look at them closely, you 11 see that for the PK information, in all cases, we 12 use the serum concentration, and for the PD 13 information, in all cases, we use the MIC, so this 14 is our raw data that goes into these indices. 15 As was pointed out by John Rex very 16 eloquently earlier, the MIC has its problems, and I 17 will come to that. 18 So, what can we do better? This is the 19 gold standard, this is what we are doing right now. 20 It works to a certain extent as we have seen. 21 Sometimes it works very well, sometimes it doesn't 22 work, and what can we do differently. 218 1 [Slide.] 2 Let's start with the PK. We use the serum 3 level. Is that the right way to go? Well, there 4 are certainly two areas where we can improve. One 5 has been addressed here several times, and that is 6 protein binding, and I am glad that there is 7 consensus now that we really should use unbound 8 concentration. 9 If you go to the literature, the data is 10 overwhelming, not only in the anti-infectives 11 field, it is a concept that holds true in all of 12 pharmacology, that the unbound concentrations are 13 responsible for the activity. There may be some 14 exceptions, but this is truly rare. 15 [Slide.] 16 This is not a new finding. This is an old 17 study from 1973, where a bunch of penicillins were 18 compared with increasing degrees of protein 19 binding. We have ampicillin here with 22 percent, 20 all the way up to cloxacillin with 95 percent, and 21 you see for each of these compounds three bars. 22 The one on the left is the MIC in broth in 219 1 the absence of proteins. The one in the middle is 2 the MIC in serum in the presence of proteins, and 3 you will see as you come to the drugs with the high 4 protein binding, you need much more drug to kill 5 the bug. 6 The third bar on the right is the unbound 7 concentration of the sample in the middle, and that 8 agrees very nicely with the MIC in broth, so 9 showing that it is, indeed, the unbound 10 concentration. 11 Again, there are many, many studies that 12 have shown the same thing, so I really think we can 13 put this issue to rest now. 14 [Slide.] 15 The second issue that is tougher to deal 16 with is the question of local exposure. In all of 17 these indices, we use serum concentrations, and the 18 simple reason is it is easy to measure. It is easy 19 to get a blood sample, however, that is not usually 20 where the infection is. 21 If you look at drug distribution in this 22 cartoon here, we can separate the body in the 220 1 vascular and extravascular spaces, and as you know, 2 the infection is usually in the extracellular space 3 here. 4 So, if we look at this fluid, that also 5 contains protein, and we apply the concept that we 6 should use unbound concentrations. The target 7 concentration that usually is applicable to 8 anti-infectives is the unbound concentration in the 9 extracellular fluid or the free tissue level, as it 10 is called. 11 There have been numerous ways to measure 12 this experimentally, mainly with blister fluid 13 data. Fortunately, there is a better method that 14 we have used in our group extensively, and that is 15 microdialysis. 16 [Slide.] 17 Microdialysis works with a probe that you 18 stick at any site where you want to know the local 19 concentration, and it measures directly the unbound 20 concentrations, so it measures what you want to 21 know, and it has tremendous potential. 22 [Slide.] 221 1 To show you one example here, this is a 2 comparison of two oral cephalosporins, cefpodoxime 3 and cefixime. In white, you see, after oral 4 administration, the serum concentrations. 5 If you do the PK, you find that the AUC is 6 exactly identical of the two, however, if you 7 measure the local free concentrations, and the 8 tissue in this case is the muscle, these are the 9 purple lines, and you clearly see that this drug 10 cefpodoxime has about twice as much local exposure 11 at the tissue level than the cefixime has. 12 So, the total serum concentration can fool 13 you, and this allows you to measure locally. 14 [Slide.] 15 Let's move on to the pharmacodynamics. 16 MIC has its problems, we heard about that already. 17 It is imprecise, it has an effect of 2-fold when 18 you determine it, so that right there makes any 19 calculation questionable. 20 It is monodimensional, and I will come 21 back to that, what that means. 22 It is used as a threshold. I think this 222 1 is conceptually the biggest problem. We draw a 2 line in serum concentration, and this is where we 3 want to be, or we want to be above that line. That 4 is not really what happens in pharmacology. We 5 rarely do have black and white events, as we imply, 6 with this threshold. 7 We have learned to get around this issue. 8 When the MIC doesn't work, then, we have some 9 patches, such as post-antibiotic effect or sub-MIC 10 effect that we use to explain what it should be. 11 [Slide.] 12 So, how can we do it better? Well, the 13 better way is to use kill curves, I believe, and we 14 have seen several models. This is the one that we 15 are using. We can reproduce any kind of exposure 16 profile that you are interested in. 17 What we are doing, we are doing the 18 microdialysis and then reproduce the unbound 19 concentration profiles in this in vitro model. 20 Then, we can compare different dosing regimens and 21 see what they do. 22 [Slide.] 223 1 This is an example of a penicillin, 2 piperacillin. We have in the top row here 2 grams 3 and 4 grams. We can look at multiple doses, once a 4 day, q8, and 6 times a day treatments here, and you 5 will always see two curves. The black curve, the 6 top curve is the control, and the red is treated 7 curve under the same conditions. 8 So, we can then compare the effects and, 9 for example, if you compare these two pictures 10 here, the 2 grams 6 times a day, you get an 11 excellent response, whereas, with the 4 grams 3 12 times a day you get a marginal response. 13 Now, the total daily dose is the same, 14 it's 12 grams a day, so you can deduce from this 15 experiment that giving the drug more frequently 16 will give you more effect, and that, is course, 17 consistent with the approach of constant rate 18 infusions with beta-lactams. 19 [Slide.] 20 Now, what other problems does the MIC 21 have? The MIC sometimes can give you misleading 22 information, that is what I mean by 224 1 monodimensional. This is a simple comparison of 2 two bugs here. On the left you see Strep 3 pneumoniae with an MIC of 20; on the right, 4 Hemophilus, and these are a bunch of kill curves 5 with increasing concentrations of ceftriaxone. 6 If I pick up the concentration of 10 7 nanograms product milliliter, we see here, this is 8 this blue curve, that for this concentration, 9 clearly, the effect on Hemophilus is stronger. 10 [Slide.] 11 However, if you to high concentration, you 12 get an exact opposite response. The effect on 13 Strep pneumoniae is better than on Hemophilus. 14 Same bug, same drug, same MIC, you get different 15 outcomes depending on concentration. The reason is 16 they have different kill rates. 17 [Slide.] 18 Another example of that here. This is a 19 simulation of a situation where you have exactly 20 the same PK in these two cases here. You have also 21 exactly the same growth rate of the bacteria. You 22 have the same MIC. So, all these three indices 225 1 that we just looked at would be identical. 2 However, these two bugs have different 3 EC50, different sensitivity, and different maximum 4 kill rates, therefore, the outcome, which is the 5 pink curve here, is different in these two 6 situations. You would not pick them up by just 7 looking at the three conventional indices. 8 [Slide.] 9 Another example of a penem antibiotic. 10 This is a PK/PD study to look at the effect of food 11 intake. In the top curve, you have, for the same 12 dose, the pharmacokinetics after a meal. On the 13 bottom, the fasted situation, you clearly see the 14 impact that the food intake has. 15 On first glance, you would think, well, 16 this is clearly a lower exposure, therefore, the 17 response probably is less. However, if you do the 18 PK/PD modeling, then, you will be surprised to see 19 that actually this exposure profile will result in 20 a better effect than the one that you see here. 21 The reason is that you get an extended release by 22 the food. You cut off the peak levels, but yet, 226 1 over time, the release covers better. 2 [Slide.] 3 Another example again where you can get 4 fooled. This is a comparison of piperacillin in 5 healthy subjects versus intensive care patients and 6 again using the method of microdialysis. You see 7 on the left here, the serum concentrations is on 8 the open symbols, and the concentration in muscle, 9 and in comparison here, the exposure in patients. 10 Clearly, if you compare these two down the 11 same scale, the exposure in patients is lower, but 12 you also see that the half-life is longer, and then 13 again results in a better effect in the patients 14 than compared to the healthy subjects. 15 [Slide.] 16 These are the kill curves when you use 17 these exposure profiles and reproduce them in vitro 18 and incubate the material, you clearly get a better 19 effect here in patients than you do in the healthy 20 subjects. 21 [Slide.] 22 So, to sum up, I believe that a simple 227 1 comparison of serum concentration and MIC is 2 helpful, but frequently not sufficient to evaluate 3 the PK/PD relationships of anti-infective agents. 4 We need to consider protein binding, that 5 we need to consider tissue distribution when we 6 talk about the exposure, and that microdialysis may 7 be a method that can help us. There are other 8 methods also are emerging, such as imaging, and I 9 think a lot of new developments are going on in 10 this area. 11 PK/PD analysis based on MIC alone can be 12 misleading, and kill curve analysis provides more 13 detailed information about the PK/PD relationships 14 than simple MIC determination. 15 [Slide.] 16 In general, I believe that intelligent 17 PK/PD modeling can help to streamline rational 18 clinical dose selection. It has clearly its 19 limitations and the final dose that comes out of 20 these predictions needs to be confirmed in a 21 clinical trial. That is always necessary. 22 [Slide.] 228 1 I would like for those of you interested 2 in this topic to mention this meeting. This will 3 happen in September this year in Nurnberg, Germany. 4 "Dosing the Magic Bullets" is in honor of Paul 5 Erhlich's 150th birthday. Outside the room, there 6 are flyers of the meeting, and it is going to be a 7 very, very exciting program. 8 [Slide.] 9 I would also like to thank all of the 10 people in my group who have contributed to the data 11 that I have shown you. 12 Thank you. 13 DR. EDWARDS: Thank you very much. 14 I am going to deviate from the schedule at 15 the moment. I think we would probably benefit by 16 taking a 10-minute break. We have no break 17 scheduled for this afternoon. Is that acceptable? 18 I think so. So, let us come back in 10 minutes. 19 [Break.] 20 DR. EDWARDS: We are on a tight schedule 21 at the end of the day. My estimate is that we are 22 going to be finishing sometime between 4:00 and 229 1 4:15. So, if that helps some of you, if possible, 2 we will be on the earlier side of that, but I think 3 we are going to be right in that zone. 4 I know for absolute certain we are going 5 to be finished by 4:30, at least I will be, so that 6 we can count on for sure, but I am hoping more for 7 between 4:00 and 4:15. 8 We will just go right ahead. I told Greg 9 that, as he was walking up right ready to present, 10 and I call a break, it is a little bit like calling 11 time out just before the last field goal to win the 12 football game or lose the football game, so he has 13 assured me he is ready to go now, and I am sorry 14 for that. 15 Industry Perspective 16 DR. WINCHELL: First of all, I would like 17 to start with the standard disclaimer that it is 18 not at all clear to me that my perspective is that 19 of the industry. In fact, it is not at all clear 20 to me that my perspective even matches that of most 21 of the rest of Merck, hence, the sub-title of my 22 talk. 230 1 Also, one of the problems with speaking 2 this late in the program, of course, is everybody 3 has already covered everything I am going to say. 4 Therefore, and given the time, I will try to be 5 brief. 6 [Slide.] 7 Moving into it, first of all, the dose 8 selection in drug development, a point that has 9 been made numerous times, but can't be emphasized 10 enough, of course, is that what we are trying to 11 identify is not just the dose, of course, but the 12 dosing regimen. 13 Another point, though, is that it is not 14 just a single decision that we are making through 15 the course of drug development, but it is actually 16 a series of decisions we are making before each of 17 the stages of trial - Phase IIA, Phase IIB, if you 18 do both of them, Phase III, and then what you put 19 in the label for marking the product. Hopefully, 20 each of these decisions is made with increasing 21 certainty. 22 Then, beyond that, in the label, you want 231 1 to consider, unless your drug has an unusually wide 2 therapeutic index, you want to know whether you 3 need dosing adjustments in the subpopulations. 4 Even beyond that, if you really have a narrow 5 therapeutic index drug, you have to take your 6 dosing adjustments down to the level of 7 individualization. 8 In fact, the value of PK/PD actually 9 matches that, as well, with a very wide therapeutic 10 index drug, you really don't care about the PK/PD, 11 you just give them a large enough dose, so the vast 12 majority of your population gets a response, and 13 you don't worry about it. 14 The other angle is on the 15 individualization where you are actually using 16 PK/PD and therapeutic drug monitoring to pick the 17 dose for an individual patient. Of course, the 18 vast majority of drugs and the vast majority of 19 antibiotics falls somewhere in between those two. 20 [Slide.] 21 This is sort of my generic PK/PD slide, 22 and I actually like Hartmut's more, so I may have 232 1 to steal it from him. I would like to make two 2 points with this. 3 First of all, in dividing pharmacokinetics 4 and pharmacodynamics, both kinetics and dynamics, 5 coming from an engineering background, are related 6 to time. They are time-dependent events. So, what 7 we are interested in, in pharmacokinetics, is drug 8 concentration with time, and what we are interested 9 in, in pharmacodynamics, is drug effect with time. 10 The other point, too, is, with an 11 engineering background, is thinking of this really 12 as a black box. When you are doing just 13 dose-response relationships, you really have a 14 large black box where you are going from the two 15 ends, from dosage to effect. 16 One of the objectives of PK/PD really, as 17 I see it, is to divide that black box into two, 18 pharmacokinetic angle, where you have the 19 intermediate measure of drug concentration with 20 time, and then the second black box. 21 [Slide.] 22 To amplify on that a little bit, to start 233 1 with pharmacokinetics, obviously, we all know that 2 pharmacokinetics depends on many different factors, 3 however, as I think George demonstrated very amply, 4 I think we are pretty good at this point at 5 characterizing the pharmacokinetics of drugs, and 6 with population pharmacokinetics, including the 7 variability of that drug, so it is critically 8 important. 9 But that part of it, I think we really 10 know how to do, and, of course, there is the added 11 complication of the concentration site of action, 12 but that is already I think, I knew would be 13 covered amply by other speakers, and I won't touch 14 on that, but I would add the point that if you 15 aren't able to measure drug at the site of action, 16 you can incorporate distribution to tissue into 17 your pharmacodynamic side of the model. 18 On the other hand, pharmacodynamics, I 19 think, is the more difficult part, and here I have 20 just listed a bunch of the factors I think 21 influence, sort of off the top of my head, 22 pharmacodynamics. 234 1 I have seen other lists that are probably 2 more comprehensive than this, but the key point is 3 you are treating a population of bugs in a 4 population of patients, and there is a whole series 5 of steps that leads to your drug effect, any one of 6 which could be the rate-limiting step. 7 In addition, you have a whole bunch of 8 other factors that modulate that effect, and we are 9 trying to reduce all of this into a simple 10 pharmacodynamic model. The key is it really isn't 11 that simple. 12 [Slide.] 13 So, this is my attempt, and I think it is 14 pretty consistent with what was said before, to 15 summarizing what the current paradigm is for 16 identifying antimicrobial PK/PD, and I think it is 17 quite consistent with what has been presented 18 before. 19 The first part is characterizing the 20 activity in non-clinical studies, in vitro systems, 21 animal models, all of that has been covered, 22 identifying the metric of exposure. At least until 235 1 now, the standard has been to identify, you know, 2 what is the index of exposure. 3 Population PK/PD modeling in Phase I, and 4 using Monte Carlo simulation, and then I think the 5 hard part and the part that I am addressing is 6 confirming these regimens and exposure metrics in 7 clinical trials. 8 So, again, the first part I think has 9 really been covered,, where I think we still have 10 room for improvement is in the last part - once you 11 get into the clinic, what do you do with your 12 clinical PK, clinical PK/PD information to improve 13 how you select dose. 14 [Slide.] 15 So, I have listed here what I see are four 16 of the opportunities for improvement. The first 17 one has been touched on by two of the previous 18 speakers, Hartmut and Dr. Benincosa, is to use more 19 dynamic PK/PD models. This goes back to what I 20 said before, and I will amplify in a minute. You 21 really want to take time into account, if you can, 22 in what the drug effect is. 236 1 The second is really a no-brainer, which 2 is the increased use of simulation at all stages of 3 development. I really think simulation is one of 4 the really underused tools in clinical drug 5 development, and it is not just to go from Phase I 6 to Phase II. 7 As you go through drug development, there 8 are many, many opportunities to use the data you 9 have, whatever it is, preclinical data, data from 10 other compounds, data from compounds that died 11 previously, or whatever, there are many 12 opportunities to use simulation to help guide the 13 design of your next study, and I really don't think 14 overall we make full use of that. 15 Of course, the major caveat is that your 16 simulation is always only as good as the data by 17 which it is driven, so in order to do good 18 simulation, you have to develop good data 19 throughout the course. 20 If you are working off bad in vitro data, 21 you are going to get bad simulations. If you do a 22 bad job of collecting population PK in Phase II, 237 1 you are going to have bad simulations for Phase 2 III. So, you must have good data to drive your 3 simulation. 4 Another point which has also been raised, 5 but it is easy to say that we really need to do it, 6 and that is to include population PK/PD in your 7 Phase II and Phase III trials, and actually, the 8 Phase II is really the most critical, in my view, 9 because that is where, if at all possible, you have 10 the most dynamic dose range, which I will get into 11 as well in a minute, but there is just no getting 12 around it. 13 If you want to use concentration data, 14 what we have heard about a lot today is getting to 15 that Phase II and Phase III, especially the Phase 16 II, dose selection, but unless you do the 17 population PK/PD in Phase II to drive your Phase 18 III selection, and then do it in Phase III to 19 confirm all of that, you really have lost an 20 opportunity. 21 The other issue where I think we can do 22 better at that level in our pharmacokinetic 238 1 analysis is really you can characterize key 2 factors, and here, I have picked three, and it 3 reflects the fact that I work largely in 4 antivirals, what I picked. 5 But the key is to think about your drug 6 and what the key factors are, and then try to build 7 those into the model and into your analysis. 8 Again, these are just three examples, but you 9 really have to think about what it is. 10 You could do something else like the 11 natural history of disease is another aspect that 12 you are able to build in. The other one that 13 George did, which is the amplification of a 14 resistant subpopulation. All of those can be built 15 into your pharmacokinetic and pharmacodynamic 16 models. 17 [Slide.] 18 So, to amplify a little bit on the dynamic 19 aspects of PK/PD modeling, I think I start from a 20 fundamental philosophical point, for antimicrobial 21 response of actually almost any pharmacodynamic 22 endpoint, which is these responses are a function, 239 1 often complex, of drug concentrations over time at 2 the site of action, so that is why I have trouble, 3 at least initially, with focusing on exposure 4 metrics like time above MIC or AUC over MIC, 5 because I really don't think that they fully 6 account for the dynamics. 7 Again, there has been discussion of the 8 limitations of MIC, which I think are widely 9 recognized. I do think there are good relative 10 measures of what kind of exposure it takes for 11 comparing either within pathogens or even across 12 pathogens, but I think the human physiological 13 system is sufficiently different that it is not a 14 direct correlation, as has been raised many times 15 before. 16 Nevertheless, I think that they can be 17 useful, in fact, they have obviously been very 18 useful, however, starting out particularly in the 19 Phase I/Phase II part of drug development, it is 20 sort of where you want to end up, in my view, not 21 where you want to start. If you can do it, you 22 would rather start with a more complex model that 240 1 incorporates the dynamics. 2 Of course, having PK/PD models, such as 3 Dr. Derendorf has described, and Lisa, as well, 4 that you can incorporate both concentration and 5 time into the model can be more useful. 6 It allows you to account for more complex 7 data, for example, post-antibiotic effect, there is 8 really nothing about post-antibiotic effect that is 9 magic. It is really quite easy to conceive of a 10 model in PK/PD terms that will explain 11 post-antibiotic effect. 12 Again, having time in there allows you to 13 do more robust simulations. You can get much 14 better information if you have time information in 15 there. Of course, the price you have to pay to 16 have more robust data is it requires more data to 17 drive these PK/PD models. 18 [Slide.] 19 Now, again, characterizing key factors, I 20 have just listed three, but as Dr. Powell started 21 to mention, taking into account factors other than 22 just taking PK/PD are very important for 241 1 understanding. 2 Adherence, of course, is critical in the 3 HIV world, but I think it is also relevant for 4 antibiotics. Any time you have ambulatory patients 5 who are not taking observed dosing, adherence can 6 be a critical factor, and, in fact, can completely 7 obscure, in the cases of HIV, can completely 8 obscure your PK/PD relationship, because whether 9 somebody responds or not, or develops resistance or 10 not, is much more a function of whether they were 11 adherent to therapy than what their PK/PD exposure 12 was on the day they came in for PK day and had an 13 observed dose. 14 But also you can flip that around, if you 15 have a dynamic PK/PD model, you can actually assess 16 the effect, and this has been done again in HIV. 17 You can assess the effect of adherence 18 patterns if you are able to gather them, by 19 following or simulating what you think the 20 concentration over time is for the adherence 21 pattern that you have either measured or somehow 22 assessed or even simulated in that patient. You 242 1 can get some idea of what the forgiveness or what 2 the effect of that non-adherence is on your 3 outcomes. 4 Likewise, emergence of resistance is 5 another one that can be characterized. Emergence 6 can actually be a fairly difficult stochastic 7 model, but also a much simpler problem which has 8 been addressed here is just treatment of resistant 9 pathogens, which is actually fairly easy to 10 accommodate in PK/PD analyses. 11 The last one is host factors and any of 12 the factors like that, factors on my list 13 previously, which can be incorporated into your 14 population PK/PD analysis as covariates, for 15 example, neutrophilic, you can incorporate it any 16 way you want, neutropenic or not, or neutrophil 17 counts, or whatever you want to do. 18 19 Now, of course, you can do these kinds of 20 analysis just looking at your standard 21 dose-response data, but what you gain by 22 incorporating it instead into a PK/PD analysis is 243 1 you see what the effect on drug concentration is. 2 So, for a neutropenic patient, what is the 3 concentration that is required to treat a given 4 organism in that patient, and what is the 5 concentration that is required in the 6 non-neutropenic patient. 7 So, incorporating these covariates into 8 population PK/PD modeling, and this is what you 9 really gain by doing it in a large population in 10 Phase III, only in that very rich dataset or very 11 extensive dataset do you really have the numbers to 12 have any chance of identifying what the important 13 covariates are. 14 [Slide.] 15 Of course, there are challenges in trying 16 to do all of this, and the first one I think we 17 have hit on in the discussion a number of times. I 18 think the biggest single challenge scientifically 19 in trying to do this in anti-infectives is the 20 limited dynamic range that you have in clinical 21 studies. 22 In the studies that you do, you have a 244 1 limited dose range often, and even more important 2 for trying to separate out or validate effects, is 3 you have a limited number of regimens. It is very 4 hard to get people to do two times a day or three 5 times a day if people think it might be a 6 once-a-day drug, and it is very hard to get a 7 range. 8 I think something that is not listed here, 9 that also came up in antimicrobials, is you 10 actually have a limited dynamic range in your 11 pharmacodynamic endpoints, as well. When your 12 endpoints or, say, microbiology either are killed 13 or not killed, just a dichotomous answer, there is 14 really not that much information there, especially 15 if it is only measured at one time point. 16 So, I think all of those combine to say it 17 is really hard to get the necessary dynamic range 18 you need to be able to identify what these are. 19 It is even more difficult, of course, in 20 serious infections, which again is an area I have 21 worked most in, both in antifungals and HIV, when 22 you have serious infections, of course, the 245 1 benefit-risk is such that you don't want to take 2 the chance of being low, so the difficulty is you 3 immediately try to get as high as you can on the 4 dose-response curve, and it becomes very difficult, 5 then, to characterize where that curve starts to 6 fall off and where you start to lose efficacy. 7 This came up before, it is hard to do a 8 robust Phase II study in those cases because you 9 just can't generate the data that is on the steep 10 part of the dose-response curve. 11 Finally, a point that Dr. Craig, who 12 apparently has left, raised earlier is it is very 13 difficult because of this limited dynamic range, to 14 actually validate, if you choose them, those 15 metrics that you have chosen preclinically. 16 [Slide.] 17 I have a slide to illustrate that. This 18 is for antifungal caspofungin where I have plotted 19 over three different doses AUC versus C24, so 20 trough concentration, of course, that could be time 21 above MIC, you get very much the same plot. But 22 what you see, because it was the same regimen, at 246 1 the same length of infusion, what you see is a very 2 good correlation between C24 and AUC. 3 So, based on this information in this 4 study, in this Phase II study, you really can't 5 differentiate between whether it is C24 or AUC, 6 which, of course, becomes important, say, when a 7 patient has renal impairment. Now you have changed 8 the relationship between C24 and AUC, and you don't 9 know, on the basis of this data at least, which one 10 is going to be really driving efficacy. 11 [Slide.] 12 An area, which I don't think we have 13 touched on as much today, are the practical issues 14 in trying to do this, in particular, the practical 15 issues in trying to do population PK. To me, the 16 number one issue on this is timing, and that hasn't 17 come up at all. 18 If you are trying to drive dose decisions 19 using PK/PD, it is essential that you deliver the 20 data and the analysis in time to make the decision. 21 It does no good to do a PK/PD analysis and not 22 deliver, because they are going to make the 247 1 decision already, they are not going to wait for 2 the PK/PD analysis, and you are not going to hold 3 up going from Phase I to II, or from II to III 4 waiting for a PK/PD analysis to be done. 5 It is a very, very difficult problem, I 6 think it is the hardest one. You have to plan way 7 ahead, you have to take extraordinary measures to 8 get your data in-house in time and get people to do 9 the analysis in real time, so that you can deliver 10 a PK/PD analysis in time for people to use it for 11 effective decisionmaking with respect to dose, and 12 that is one of the hardest. 13 Of course, another one, especially for 14 population PK, which Dennis touched on, are the 15 logistics and resources are very difficult. Again, 16 it is the quality of the data that this really 17 drives. It is very difficult to generate quality 18 population PK data. 19 You have to have accurate sampling times, 20 you have to have accurate sample handling, and it 21 can be very difficult, particularly in a Phase III 22 setting. I actually think anti-infectives people 248 1 are used to doing this, and it is easier to do, as 2 opposed, to say, in an obesity clinic where it can 3 be very difficult to get accurate PK/PD data. 4 The last one, I think once you get into 5 combination therapy, PK/PD becomes extraordinarily 6 difficult. It is really hard to differentiate, in 7 clinical studies, it is hard to differentiate out 8 what the relative contributions of different drugs 9 are in combination therapy. 10 So, then, I would like to finish with just 11 a brief review of sort of the history of drug 12 development of anti-infectives at Merck. Actually, 13 I missed a point, which was a key point, which is 14 the organizational inertia and resistance, which 15 Dr. Powell was hitting on. 16 One of the reasons why it is difficult to 17 do population PK/PD is just it is hard to get our 18 colleagues in clinical research and senior research 19 management to agree to do this. Part of it is just 20 inertia, although in some cases, it is outright 21 resistance. 22 But nevertheless, my staff and I, we spend 249 1 an inordinate amount of time on the phone trying to 2 convince people to draw samples when we want a 3 sample, allow them to do all of that, go back even 4 to preclinical to convince our preclinical 5 colleagues to draw the samples, to do the right 6 experiments to generate the PK/PD. 7 So, again, I think that this is another 8 major stumbling block at least in the industrial 9 setting in trying to get this done. My experience 10 at Merck at least has been that you really have to 11 win this battle incrementally, as well as 12 therapeutic area by therapeutic area. 13 You have to take what they--in terms of 14 sampling, deliver on that value, then, turn around 15 and they will be much more amenable to doing it the 16 next time if you can show value from whatever you 17 can convince them to do the first time. 18 [Slide.] 19 So, with that background, I would then go 20 to my summary and how this has evolved at Merck. 21 Starting with indinavir, our protease 22 inhibitor, in this case actually, we did not 250 1 prospective PK/PD, in fact, all of the PK/PD 2 analyses that have been done for indinavir were 3 done post hoc, well after the dosing decisions were 4 made--well, five patients, was it, six. 5 One of the reasons this happened, just as 6 an instructive, is we actually had a bad experience 7 in one of the previous HIV drugs, an RTI we were 8 developing earlier, where we generated a set, in a 9 Phase II study, we generated a set of PK data. 10 It was just completely useless because it 11 was not collected properly, we didn't have good 12 sampling times. It basically was just noise, and, 13 in fact, when you tried to do a PK analysis, it 14 just added noise to the data and was a worse 15 analysis than just doing dose. 16 As a result of that, we lost credibility 17 with our Anti-infective group, so when it came to 18 developing the protease inhibitor, they were much 19 less willing to put any faith in, yes, you can 20 draw, you can actually do this in a clinical 21 setting. So, a lesson learned there. 22 The second one is our carbapenem. In this 251 1 case, the beginning part, we actually followed much 2 of the paradigm that was listed, you know, just 3 another carbapenem, you know, time above MIC was 4 largely assumed to be the driving force, and our 5 dose selection was based on time above MIC as 6 measured in the animal models done by Dr. Craig, as 7 I recall, and human PK, and that was the dose 8 selection. 9 However, we didn't do, after that, we 10 didn't do any Phase II or Phase III population 11 pharmacokinetics, after that, we relied just on 12 dose. 13 That was in, say, that was approved in 14 '91, so it was developed in 2000-2001 time frame. 15 With caspofungin, on the other hand, we 16 did manage to convince them to do prospective 17 population PK/PD in all of the clinical trials, and 18 that worked very well, and it turned out to be very 19 supportive actually of the dosing selections in 20 this case. 21 That actually was a little bit of an easy 22 case, however, because it is an I.V. drug, so it is 252 1 relatively easy to characterize, and it is given in 2 a hospital setting, so compliance isn't an issue. 3 So, it was very useful, but it was also 4 one of the easiest cases in which you are able to 5 characterize population PK/PD. 6 Just as a sign of the progress we are 7 making at least at Merck, one of the drugs we now 8 have in early clinical development, we started 9 early, we did human PK predictions to select the 10 initial dose based on preclinical data that we got 11 started. 12 We are doing prospective population PK/PD 13 and we have it already built into the plans to use 14 it for selecting dosing regimens for our subsequent 15 trials. We are using MEMS-caps or planning to use 16 MEMS-caps in order to capture compliance data, to 17 try to incorporate that into the PK/PD modeling. 18 So, I think what this shows is again you 19 can see the incremental increase in the acceptance 20 of these approaches just within the Anti-infective 21 group at Merck, and again it is a question of 22 showing value and then delivering on that value, 253 1 and avoiding the kind of mistakes we made early in 2 the HIV program. 3 Basically, that's it, and I have just 4 tried to give you some aspect at least at Merck as 5 to what the issues and opportunities are in this 6 area. 7 DR. EDWARDS: Thank you very much, Greg. 8 I will now call on our last speaker, Jenny 9 Zheng from FDA, to comment on the FDA perspective. 10 FDA Perspective 11 DR. ZHENG: Good afternoon, everybody. 12 I will talk about improvement in dose 13 selection from FDA's perspective. The views 14 expressed in this presentation actually are based 15 on FDA's current experience. 16 [Slide.] 17 I will, first, very briefly discuss the 18 studies we need to evaluate the dose selection and 19 then move on to the second topic, which is use, 20 model-based, a quantitative approach to dose 21 selection. 22 The use of PK/PD for dose selection. Dose 254 1 selection is very challenging. You have to 2 integrate both efficacy and the safety. The dose 3 should be selected to demonstrate at such a dose, a 4 drug is not only efficacious, but also safe, so 5 integration of both are important, however, the 6 safety consideration is not the main topic of this 7 presentation. I am going to focus on the dose 8 selection from efficacy perspective. 9 Resistance is a special issue for 10 antimicrobial agents, so it is important and should 11 be considered for dose selection. 12 [Slide.] 13 First, I will discuss the studies we need 14 for evaluation, the dose selection. It comes from 15 four areas, first, microbiology in vitro data, 16 which include all the susceptibility data, protein 17 binding, post-antibiotic effect data, and 18 preclinical data including the PK/PD studies in 19 various animal and in vitro model. 20 The primary objective of those studies is 21 to identify the important PK/PD indices and 22 estimate the magnitude of the PK/PD indices. 255 1 [Slide.] 2 PK study from Phase I is necessary for 3 dose selection except the PK study in healthy 4 subjects, we would recommend PK study in special 5 population will be conducted earlier, so the 6 information from that study can also be used for 7 dose selection. 8 Phase II studies are very important for 9 dose selection. For efficacy perspective, there 10 are proof or concept studies. They provide a lot 11 of information on efficacy, on the other hand, the 12 well-designed Phase II dose-ranging study can be 13 very informative to define the PK/PD relationship 14 in human, which currently have very limited data. 15 From safety perspective, Phase II studies 16 are the only studies actually we can use to 17 identify any dose-related adverse events, because 18 most of the time, in the Phase III, it is going to 19 be a fixed dose study. 20 So, integration of efficacy and the safety 21 from Phase II study is important to optimize the 22 dose for Phase III. 256 1 [Slide.] 2 Dose selection is such an important issue, 3 so they would recommend early communication with 4 us. The desirable time could be prior to Phase II 5 and Phase III study. 6 [Slide.] 7 At the current time, as Frank discussed in 8 his presentation, the rationale for dose selection 9 in the antimicrobial drug application is variable. 10 Very often the dose actually is selected based on 11 mean PK data in relation to MIC for relevant 12 pathogens. 13 Two issues are associated with this 14 approach. One is that the PK variability is not 15 considered. The underlying assumption is that 16 exposure is the same among the population at the 17 dose, obviously, it is not correct. 18 So, a second issue is it is not always 19 clear how high the concentration should be above 20 MIC, or how long the concentration should be above 21 MIC. In the situation where you have to decide a 22 dose between X or 2X, this approach barely give you 257 1 any clue of which one you are supposed to choose. 2 So, I will discuss a quantitative 3 approach. This is not new, especially as the last 4 speaker of the presentation, but I would emphasize 5 the utility of this method for dose selection. 6 [Slide.] 7 Quantitative approach is to use modeling 8 and the simulation, to quantitatively predict the 9 outcome. The advantage is quantitative, so you can 10 use the model to make prediction, to predict the 11 outcome. 12 Secondly, the decision is more 13 transparent, a decision can be made more 14 objectively than subjectively, so it may reduce the 15 chance of making human error. 16 [Slide.] 17 Quantitative approach is about predicting 18 outcome. What we are predicting here is if the drug 19 can kill the bacteria or inhibit bacterial growth 20 at a certain dose in a human who is infected by the 21 pathogen. 22 Assuming a subject is infected by the 258 1 pathogen, and dose X is given to the subject, 2 according to the PK and the protein binding, you 3 can predict unbound drug exposure for this subject, 4 but the question still remains if the pathogen can 5 be killed at such dose. 6 So, to answer that question, we have to 7 rely on PK/PD relationship obtained from 8 preclinical study. It has been found that some 9 PK/PD indices actually do correlate with bacterial 10 killing or bacterial inhibiting effect, so a PK/PD 11 index can be obtained from those studies to be used 12 to predict the bacterial killing and the inhibiting 13 effect. 14 Knowing the PK, knowing the MIC of 15 pathogen, you can calculate the PK/PD index from 16 this subject, compare that value based on PK/PD 17 indices you found from the preclinical study. If 18 PK/PD index for unbound drug in subject is above 19 the PK/PD index obtained from preclinical study, we 20 would assume the pathogen can be killed at such a 21 dose. 22 [Slide.] 259 1 The goal of this approach is to predict 2 the percent of patients who could reach a PK/PD 3 target at a range of doses. Gathering information 4 from the trial is important, however, design the 5 trial right to get the right information is more 6 important. 7 So, the results from this analysis can 8 actually guide you to design your trial. If the 9 objective of the study is to demonstrate efficacy 10 at a fixed dose, you can chose a dose such that the 11 majority of subjects actually could reach the PK/PD 12 target, however, if the objective is to design the 13 Phase II dose-ranging study to define PK/PD 14 relationship, you probably can choose the doses 15 which should be defensible [?] with regard to 16 percent of subjects reach the PK/PD target. 17 By doing that, you increase the power of 18 detecting the relationship. 19 [Slide.] 20 In the next couple of slides, I will use a 21 simulated example to illustrate this methodology. 22 Again, this is not new, but I just want to 260 1 emphasize the utility for dose selection. 2 PK data from Phase I can be used to 3 establish the PK models, which include estimation 4 on not only PK parameters, and also its viability, 5 so the model can be used to generate a PK profile 6 at a dosage even not studied. 7 So, the advantage of this approach is you 8 can explore many scenarios. The concentration time 9 profile for this simulated drug at dose X, actually 10 are generated and presented in this figure. As you 11 can see, these are actually very different among 12 the population, even the same dose is given. 13 Knowing the MIC value of a pathogen, you 14 can calculate PK/PD indices, but for this example, 15 I use MIC 90. I know it's a single value, but it is 16 a more conservative consideration. It represents 17 some distribution above this MIC, but again it is 18 more conservative approach. 19 So, based on the individual profile and 20 MIC 90, you can calculate the important PK/PD 21 index, which had been found from preclinical study. 22 [Slide.] 261 1 So, the distribution of that PK/PD index, 2 here, I just assume, I use the MIC as the key PK/PD 3 index. So, distribution of that PK/PD index at 4 such dose is presented in this histogram. 5 Assuming from the preclinical study, the 6 AUC-MIC ratio of 35 or greater is associated with 7 desirable bacteria killing effect, we can calculate 8 percent of subjects who actually can reach this 9 target level. At dose X, it is 77 percent. 10 [Slide.] 11 Using the same approach, you can calculate 12 the percent of subjects who reach that PK/PD target 13 in a range of doses for different pathogens with 14 different MIC values. I will use the term 15 "response" to represent the percent of subjects 16 with AUC-MIC greater than 35 for the purpose of 17 explaining these slides. 18 In general, the dose-response relationship 19 can be described by the Emax model, meaning that at 20 certain point, increasing dose will not result in 21 any benefit with regard to response. 22 The second point is the dose-response 262 1 relationship is going to be very different. It 2 depends on the MIC value of the pathogen. So, at a 3 certain dose, the response could be very different, 4 depends on MIC of the pathogen. 5 The right vertical line represents the 6 dose X. As you can see, at such a dose, for the 7 pathogen with MIC 0.5 micrograms, the response is 8 in the range of 80 percent, but the response rates 9 actually are very low for the other two pathogens 10 with MIC value of 1 and 2, which is not very 11 different from 0.5 actually, only 1-fold, 4-fold 12 difference. 13 So, knowing your target right is very 14 important for your dose strategy here. Knowing the 15 dose-response relationship is very informative for 16 dose selection. For example, I am using a pathogen 17 with MIC 90, 0.5 microgram per ml as an the example 18 here. 19 You wouldn't choose dose X 1 because the 20 response is going to be too low. You probably 21 don't want to choose the dose X 2 either, because 22 even though the response probably is high, but the 263 1 dose itself if not very robust. A little change in 2 the dose would result in dramatic difference in 3 response rate, so it may not be a good dose either. 4 Definitely, you don't want to pick up the 5 dose X 3 because it is unnecessarily too high. 6 Selecting this dose, you may have a safety problem. 7 The dose X 4 probably is reasonable. At first, it 8 provides pretty good response, and it is on the 9 flat curve. It is a little bit more robust. 10 So, knowing where your dose is on the 11 curve is important for successful development of 12 the drug. 13 [Slide.] 14 Successful treatment of infection in most 15 interaction of host drug and the bacteria. The 16 factors we have considered in this model include 17 the pharmacokinetics, protein binding, 18 susceptibility of the pathogen, and PK/PD 19 relationship, however, many other factors have not 20 been considered in this methodology. They could be 21 tissue penetration. We have heard the issue with 22 that. The post-antibiotic effects, some killing 264 1 rate, and more importantly, we have not considered 2 immune system [?] in this model, but we all think 3 it is important. 4 So, the factors not being considered may 5 represent the potential limitation of this 6 approach. 7 [Slide.] 8 The most significant issue is we use PK/PD 9 from animal to predict the outcome in human, so the 10 issue is the predictability of PK/PD relationship 11 in animal for treatment of infections in human is 12 not clear, because most of the time the PK/PD 13 relationship for most of the drugs was established 14 only in human. 15 So, to improve our understanding, this 16 PK/PD relationship in human, we need to gather more 17 data from Phase II studies and from Phase III 18 studies. 19 Again, designing the trial is more 20 important than gathering the information. 21 [Slide.] 22 In summary, a model based on quantitative 265 1 approach is informative for dose selection, 2 however, it raises the potential limitation, the 3 dose should be selected based on the totality of 4 the available data. 5 Dose selection is an ongoing process. At 6 first, you can gather microbiology preclinical data 7 to integrate with Phase I data, and doing the 8 analysis to help you to design the Phase II. 9 The results from the Phase II study can be 10 very useful to confirm the hypothesis you generate 11 from the early development process, and integration 12 of both safety and efficacy can help you to design 13 the dose for Phase III. 14 Selected dose for Phase III can be tested 15 actually in the trial. The results from the Phase 16 III trial can further confirm the hypotheses you 17 use at the beginning. 18 [Slide.] 19 So, in the future, we would like to see 20 more well-designed, dose-ranging Phase II studies 21 in appropriate infection. From those studies, it 22 can provide PK/PD relationship in human and also 266 1 safety information. 2 The improvement of the model used in 3 quantitative approach is needed except as a 4 prospective, we need to consider the safety and 5 resistance. I wish we could have all the 6 relationships presented by George Drusano in his 7 presentation. He present all the prospective. We 8 wish we could have the relationship before the drug 9 is approved other than after that. 10 So, evaluation of indices other than 11 AUC-MIC, Cmax-MIC, and the time of MIC is 12 encouraged because we know those parameters have 13 some limitation as the previous speaker has talked 14 about. 15 Finally, development of optimal duration 16 of therapy should be considered. 17 Thank you. 18 DR. EDWARDS: Thank you very much, Jenny. 19 Let me just update the schedule. I am now 20 predicting 4:15 as the absolute end unless there is 21 a dramatic change in that estimation, so we are 22 going to be shooting for that now. 267 1 Now I would like to open the discussion 2 for both of these sessions this afternoon. 3 Discussion 4 DR. LAZOR: We heard this afternoon, and 5 actually this morning, as well, that is important. 6 Data translates into information, and information 7 into knowledge. I also heard that sometimes the 8 data are difficult to obtain because of the 9 practicalities involved. 10 What progress has been made over the years 11 to actually collect good data? 12 13 DR. EDWARDS: Dennis. 14 DR. GRASELA: I think the progress that we 15 have made has been around the prospective design of 16 protocols and case report forms, and the use of 17 real-time data assembly to have a few back-loop 18 back to the sites as early as we possibly can. 19 One of the things that I have noticed is 20 that study coordinators, God bless their souls, are 21 incredibly efficient and incredibly consistent, and 22 if they got it right at the investigators meeting, 268 1 they will do a great job. If they got it wrong or 2 that the lag time is too long between when you 3 design your trial and when they actually enroll 4 their first patient, and they don't get it right, 5 they will do it wrong every time after that. 6 So, the thing that we have paid a lot of 7 attention to, or tried to pay a lot of attention 8 to, is to do things in a very prospective manner. 9 When I first came to BMS, the population PK 10 consisted of a boatload of data coming to us, 11 dumped on the table, and saying save us, we don't 12 think we have the right dose. 13 Now we have gone to a prospective 14 approach, which I think will ultimately keep us in 15 good stead, but it is always the balance of--one of 16 the things we didn't talk about is enriching 17 trials, and sometimes if you are going to enrich 18 your trials with a certain population or, for 19 instance, penicillin-resistant Strep pneumo, if you 20 are going to enrich your trial, sometimes it can be 21 done in the U.S., sometimes it has to be done in 22 the far reaches of the world, that can make the 269 1 infrastructures really bad, and even shipping the 2 samples, it can cost you thousands of dollars just 3 to have someone walk on site to pick up the box to 4 bring it to you. So, there are a lot of logistic 5 challenges with doing studies in developing worlds. 6 DR. SCHENTAG: The interesting thing is I 7 have been approached over the years by sponsors all 8 over the place that say what do I need to do, and 9 then when you tell them what you need to do during 10 the course of a clinical trial, either Phase II or 11 III, I have done it in both, they always say, no, 12 that is impossible, the investigators will never do 13 it. 14 I say every time, okay, if you fund this, 15 I will get the investigators to do it, and that is 16 true, you have to usually pay them a little bit 17 more, and usually you have to watch them and work 18 with them, and you have to be in real type contact 19 with them. 20 The other thing, and I agree exactly with 21 what Dennis said, too, is that if you are willing 22 to interact all the way along in the sampling 270 1 process, you can get very good samplings out of the 2 clinical investigators all over this country 3 anyway, and I haven't tried it overseas, but I 4 don't have any reason at the moment to believe it 5 couldn't be done there also. 6 It is just a matter of you need to commit 7 as much effort to it as you do to collecting the 8 patients in the first place. You can't just put it 9 in the protocol and expect it to be done right, 10 because most of the time it isn't, but I have never 11 seen a situation yet where we couldn't get the data 12 if there was real-time commitment to it. 13 DR. EDWARDS: John. 14 DR. POWERS: It seems what people are 15 saying is that if it takes that commitment, it 16 seems like basic human nature that nobody is going 17 to make that commitment unless there is something 18 in it for you. 19 So, the question would be--I was just 20 writing this down, thinking here what is the lesion 21 here--is it a lack of consensus on how to do it 22 right, is it a lack of perceived value from 271 1 industry either because it costs too much, it takes 2 too much time, or both, or is it because they can't 3 use it for marketing or labeling in a way that they 4 find useful, or is it a perception that FDA won't 5 find this useful, I can't use it to move my drug 6 forward, so I am not going to do it? 7 Greg, can I ask you that one to start off 8 and maybe some other folks can try to tackle that? 9 DR. WINCHELL: I don't think it is because 10 it doesn't drive labeling, because actually it has 11 been quite successful. A lot of population PK/PD 12 information is making its way into the labels. 13 I think it is people, at least don't 14 understand the value--we haven't demonstrated the 15 value of it is what I think is probably the major-- 16 DR. SCHENTAG: To? 17 DR. WINCHELL: To management and also-- 18 DR. SCHENTAG: To management. 19 DR. WINCHELL: Again to clinicians within 20 therapeutic areas actually. As was pointed out, 21 clin-pharm, clinical PK, et cetera, are involved 22 only up to a certain point, and then it goes in, 272 1 and then, at least the way Merck is structured, 2 when you get to the clinicians, it is a different 3 therapeutic area each time, and you have to fight 4 the same battles and win the same victories with 5 each therapeutic area as you go along, and it is 6 just very difficult to do. 7 DR. DUDLEY: I think Dennis' comments are 8 spot on, and I have taken the approach--the 9 contract research business is very competitive, as 10 you know, and if they can't get it done, then, it's 11 go talk to the next group they get on the list. 12 A lot of centers that are now doing 13 studies particularly in the antifungal area now, a 14 lot of those initial trials are taking place 15 offshore, they are taking place in research units 16 in South America and South Africa. They are very 17 conscientious about getting those types of studies 18 done now, as well. 19 So, I think it probably depends on the 20 company and their commitment to doing it right, but 21 I think you can insist on it and get the contract 22 research organizations to respond or else you go 273 1 down the list. 2 DR. EDWARDS: Bob. 3 DR. POWELL: You were asking where the 4 critical lesion is. Really, all the pieces that 5 you discussed, but probably the most important 6 thing to understand about an organization is how 7 decisions are made. 8 So, for example, you are at the FDA. 9 Decisions are made on NDAs in terms of go/no go, 10 approval or not approval, where what's in the label 11 by the physicians that are running the therapeutic 12 areas, where everything else, is my perception, is 13 advisory to that, to those people. 14 Would you agree, more or less? 15 DR. POWERS: I like to think we work as a 16 team. 17 DR. POWELL: But that is the option of the 18 person that is running the therapeutic area, and 19 some do it more than others. It is the same in the 20 industry, so that in some therapeutic areas--and 21 people have alluded to this--that people will be 22 very team-oriented and inclusive, and some, for 274 1 sure, one of the characteristics in drug 2 development is that people want to focus. 3 You can track responsibility for making 4 decisions usually down to one person for a given 5 type of decision, and it is kind of up to that 6 person's style. Now, organizations, I would say, 7 in drug development, some organizations have 8 evolved a style because of horrendous mistakes that 9 have been made by one person where they are more 10 inclusive, but others aren't, but it comes down to 11 that is the piece, how decisions are made. 12 The other thing is that a lot of times the 13 people in clinical pharmacology, they may want to 14 be influential, but for whatever reason, they 15 haven't evolved to be as influential inside the 16 drug company as the Phase III physicians. 17 People may want to disagree with that, but 18 that is my perception. 19 DR. POWERS: So, then comes up the next 20 question, then, so outside of tying up those 21 people, throwing them in the middle of the floor 22 here, and say why don't you accept this, what would 275 1 it take to convince those people? Would anybody go 2 forward with an anti-infective program without 3 chucking their bug into a test tube and seeing what 4 the MICs were? 5 DR. POWELL: Well, I am not sure those 6 people are here. Are there Phase III people here? 7 DR. POWERS: Let me go back. How does 8 something like an MIC become so accepted, and how 9 do we get PK/PD to be accepted in a similar way 10 that people will see its value? 11 DR. POWELL: It is just as simple. The 12 clinical trialist mentality that has driven really 13 the FDA over the last 30 years has driven much of 14 clinical development. That is changing, but it is 15 changing ever so slowly, and as people have said, 16 you have to take that on, on a person-by-person, 17 project-by-project, therapeutic area-by-therapeutic 18 area. 19 Now, I would say that, in my experience, 20 infectious disease people are kind of akin to 21 oncologists, they tend to be pretty conservative in 22 terms of we will do this trial the way we did the 276 1 last trial, or the way someone else did the trial, 2 where sometimes I have seen people in other 3 therapeutic areas be much more amenable to change. 4 DR. EDWARDS: Hartmut. 5 DR. DERENDORF: I think a lot of things 6 are done because they are done that way, you know, 7 people just don't think about it, they do it like 8 the other person before them has done it, and if 9 that worked, then, they are encouraged to do it 10 that way. 11 So, from that perspective, I think the 12 role that you play, and that the FDA plays, is 13 extremely important, because that sets the tone, 14 and that new Critical Path document that just came 15 out, in my opinion, is a milestone, because it 16 really presents an opportunity that you can shape 17 drug development in this country. I mean that you 18 are not just drug police and monitoring 19 organization, but you truly have an active role in 20 bringing better products to the people quicker. 21 DR. POWELL: There is another leverage 22 point, and that is regulatory affairs inside the 277 1 company. They will usually also be very 2 conservative, but I think that they pay a lot of 3 attention to what the FDA says, so that if we can 4 start having End of Phase I/IIA meetings or End of 5 Phase II meetings, then, what you say, as the 6 regulator, at that meeting, and what you are 7 looking for, there is a lot of attention that is 8 paid to that inside the company, so regulatory is a 9 leverage point. 10 DR. BENINCOSA: I would like to echo what 11 Dr. Powell was saying. Certainly, in the last 12 year, industry has taken notice to both the 13 opportunity for the End of Phase IIA meeting, which 14 was just mentioned, his appointment, as well as Dr. 15 Stanski's appointment, so the receptivity to this 16 knowledge-based drug development, the message is 17 coming clear, and that is only going to help us. 18 I think perhaps in the past, the focus of 19 the traditional End of Phase II meeting, that 20 clinical pharmacology issues, we have not had the 21 opportunity to really discuss them as much, and so 22 the End of Phase I/IIA meeting gives us that 278 1 opportunity and puts the issue of dose selection in 2 a forum where we can have these discussions, so it 3 will only advance our abilities to do this in the 4 industry. 5 DR. EDWARDS: John. 6 DR. REX: I want to follow on to answer 7 John's question specifically about what does it 8 take to influence. Let me start with the 9 observation that the FDA doesn't have a brain, AZ 10 doesn't have a brain, Pfizer doesn't have a brain, 11 there are no corporate or federal neurons. It is 12 all about individual people and what they believe 13 and what they think. 14 This is actually one of the reasons why 15 guidances are so valuable, and it is particularly 16 valuable when guidances contain instructive 17 exemplars, instructive examples. 18 I was looking at one of the draft--I will 19 use that ugly word--draft guidances from whenever 20 the last ones were published, and buried in there I 21 found some really lovely little vignettes about 22 stories about a given case where somebody had done 279 1 or not done something and the consequences of it. 2 I really enjoyed Hartmut's examples of 3 things that worked and didn't work. We have had 4 several of those today, and that kind of thing is 5 what educates people, and that is what I can use to 6 educate senior management. 7 Before I joined AstraZeneca, my personal 8 cross to bear was that I was a hospital 9 epidemiologist, and I had to convince 10 cardiovascular surgeons that they needed to do 11 something different from what they were doing, and 12 you can't tell a cardiovascular surgeon anything, 13 but they are physicians, and they are very 14 interested in outcomes and data. 15 So, when I started to show them their 16 personal data, then, all of a sudden I had their 17 attention because it was alive and real to them. 18 The same thing is true of my bosses, of 19 Lisa's bosses. Once you show them real data, of 20 real examples, not made-up stuff, and I know that 21 you guys, you can't name names, but you can 22 summarize themes and provide examples. 280 1 The things that you can put in those 2 documents and white papers and guidances, that we 3 can use as examples, you know, Greg has got some 4 great stories about how things worked and didn't 5 work, and to have those in front of you when you 6 are trying to tell a tale is incredibly useful. 7 So, that is the thing that I would like to 8 say, is that, you know, this is why these documents 9 are so important to us on the other side. It is 10 because it gives us examples. 11 DR. EDWARDS: George. 12 DR. TALBOT: I would like to echo what 13 John was just saying. I was operating at a 14 slightly higher--well, my example was at a slightly 15 higher cerebral level, because I wasn't thinking of 16 neurons, I was thinking of memory. 17 I was going to point out two things. One 18 is that in many companies, there is a lack of 19 institutional memory, and that is also relevant 20 because again, we talk about PhRMA, but as we 21 discussed yesterday, PhRMA isn't pharma, it's 22 bigger companies maybe with a longer history and 281 1 better institutional memory, it's smaller companies 2 who are perhaps new and coming into this, and they 3 vary tremendously, in my experience, in terms of 4 their expertise and their institutional memory. 5 So, I think John's point about how useful, 6 if not official guidances, can be, but at least 7 summaries, communications that can be used in 8 educational processes, I agree with that. I think 9 all those documents are helpful in those situations 10 where institutional memory may be missing. 11 DR. EDWARDS: Phil. 12 DR. COLANGELO: During the course of 13 today, I seemed to have fixated a little bit on 14 Phase II studies. There has been a lot of 15 discussion about Phase II studies and to Phase IIA 16 meetings, et cetera, and I think intellectually, 17 scientifically, we can all understand and 18 rationalize the importance of Phase II studies 19 within a development program. 20 I guess, though, my curious question to 21 the industry colleagues and those in academia is 22 what type of obstacles, I guess, do you all face in 282 1 terms of trying to convince your management folks 2 and investigators as to doing Phase II types of 3 trials. 4 George Drusano kicked around the concept 5 of robust Phase II studies, and I think that needs 6 to be clearly defined as we try to move forward to 7 get better ideas about rational dose regimen 8 selection. 9 So, I just would like to throw it back to 10 you all and maybe here at this point in time, what 11 types of hurdles or obstacles you may face within 12 your own companies or within your investigators in 13 terms of doing Phase II trials. 14 DR. EDWARDS: Before we have a response, I 15 would just like to add something to that, Phil, and 16 ask, in addition to what you are asking, whether it 17 is possible to convince management that robust 18 Phase II trials are cost saving, and whether there 19 is a structured way to approach them with that 20 particular concept in mind. 21 Yes, Mike. 22 DR. DUDLEY: I think that if you look at 283 1 what happens in small companies, I can't comment on 2 big companies, one of the real sentinel events or 3 the milestones, of course, are filing the IND. 4 That is really sort of the Good Housekeeping Seal, 5 and not being put on clinical hold, but filing the 6 IND and being able to go into Phase I, that is a 7 real milestone. 8 Then, I think the other is the Phase IIA, 9 the first time going into patients and the 10 so-called, as we have talked about, the proof of 11 concept Phase IIA efficacy study. 12 So, that is why that first Phase IIA study 13 is so critical. I think that management, 14 especially rookies in the area, they will recognize 15 that it is going to cost some money to go to Phase 16 IIA anyway, so I think that by now, with the 17 guidances being in place, and I think the comment 18 that was made earlier is that guidances are very 19 important, the stories of the autopsies of failed 20 programs and what have happened that results in 21 problem, then, all strengthen the idea that it is 22 going to cost money, but you are enabling yourself 284 1 to, at a critical value creation point, by having a 2 successful Phase IIA study, because you are going 3 to want to go on from there with some fairly good 4 certainty about where your dose is going. 5 So, I think that the timing of where these 6 studies are placed is actually pretty good at least 7 in the smaller company, because you are positioned 8 to do those studies where there is more emphasis on 9 getting success and rather than on the cost-cutting 10 types of things that may come later. 11 DR. POWERS: Also, one of the things it 12 seems to us, that we have heard about, is that for 13 smaller companies, that getting capital is an 14 important thing. Is doing those Phase II trials 15 and showing some proof of principle helpful to them 16 in terms of doing that? 17 DR. DUDLEY: Absolutely, absolutely, and I 18 think that there is pushback oftentimes, do we 19 really need to do that, can we cut the cost of the 20 trial, and so forth, but I think with the ideas 21 that that analysis is going to really, a successful 22 Phase IIA is going to be a value creation point, so 285 1 it is worth the investment. 2 You have heard pay me now or pay me later 3 a lot of times in the meeting today, and I think it 4 does help, because it is an insurance policy that 5 you will have usable data from that Phase IIA 6 study, because if you don't have a good result, you 7 don't have a way of getting yourself out of it, and 8 understanding why. 9 DR. COLANGELO: The perception, at least 10 my perception about Big Pharma is that there is 11 really no sequential process, you know, Phase I, 12 II, III, it is really a simultaneous process and 13 that it would at least seem to us that learning 14 from Phase II first, before going to Phase III, I 15 think would be helpful in describing, as well as 16 transparency about Phase II, what you have learned 17 from Phase II with us also would be helpful. 18 But I know that there are deadlines and 19 milestones that need to be reached, so again, I 20 just welcome your comments about what types of 21 problems there are with that. 22 DR. EDWARDS: Jerry. 286 1 DR. SCHENTAG: I would just like to add 2 something about money to this, because budgets for 3 these things are, of course, always an issue. It 4 cost about the same per patient, and, you know, 5 budgets go per patient, it costs about the same per 6 patient to add on PK/PD done properly to a Phase 7 II, as it does to a Phase III, but because Phase 8 III's are typically huge trials, and Phase II's are 9 smaller, it makes it financially more attractive to 10 do this in Phase II, or the other possibility, 11 which is you do it in a subset of the Phase III 12 patients, like in some centers, not in others, so 13 it is not a money difference that prevents it from 14 being done in Phase III or forces it to be done in 15 Phase II. There are other things that have to be 16 done. 17 It is only about 15 to 20 percent more to 18 add PK/PD on to a typical trial budget anyway. 19 Most of the expense is not the cost of doing the 20 PK/PD, it is the cost of paying for that patient 21 and paying that investigator for his time. 22 Anyway, I just thought I would add that. 287 1 DR. EDWARDS: Yes, Dennis. 2 DR. GRASELA: I would like to address it 3 from actually two sides. The first is a real 4 candid question, and please don't take offense. 5 If you have two well-controlled clinical 6 trials and they are successful, they have met their 7 non-inferiority endpoints, and you have no idea how 8 they have selected the dose, will the drug get 9 approved, yes or no? 10 DR. POWERS: The answer is yes. 11 DR. GRASELA: So, we talk about some of 12 the difficulties we have with our trialist friends, 13 and that is the carrot that they will always stand 14 up. When they get pushed back from the 15 investigators about doing some of these things, 16 because some investigators will actually tell us, 17 well, I have a trial from company X and your trial, 18 you want PK and a whole bunch of other things, and 19 you are only to give me a few thousand dollars, and 20 I can do company's B and have twice the throughput 21 because I am not burdening my study coordinator 22 with that extra work. I am going to do trial B, 288 1 and the patients are going to go there, so they may 2 accept your trial, get IRB approval, they enroll 3 those subjects. 4 So, as long as the Agency says two 5 well-controlled trials and you are approved, our 6 trialist friends are always going to hold that up. 7 The second piece of the coin is I have 8 been to a couple of End of Phase II meetings, and 9 the agendas are so packed that I don't get to talk 10 to my friend Phil Colangelo about what we are doing 11 in the clinical pharmacology piece of things, there 12 is zero time. 13 So, I have been encouraged with the IIA 14 piece. Now, coming back to IIB, I have been very 15 fortunate, I have an incredible relationship with 16 my VP of Clinical Research, and I have been able to 17 introduce population PK across the board in 18 infectious diseases. 19 The problem I am having right now is 20 delivering, because when the database is cleaned, 21 our clinical colleagues push a button, and they get 22 their endpoints, their confidence intervals, and 289 1 they are done, and if you are not very prospective 2 in the way you collect the data, you are just 3 beginning to do that PK/PD analysis, and it takes 4 time to do that. 5 So, therefore, you cannot deliver by the 6 time of the end of the Phase II meeting, and they 7 say, well, we couldn't use it for decisionmaking 8 anyway, why should we fund this. That is our 9 problem, that is actually not your problem. 10 I have a good example of when we have done 11 it with quinolone, and it was very successful, and 12 we had it in time for the IIA meeting. That took a 13 lot of work, and it was very rewarding, and it sets 14 the precedent. 15 The third piece is coming back to the 16 guidances and really good examples, because I think 17 what we are getting nods from my colleagues is you 18 have to reeducate the next person. 19 When I branched into the antiviral part of 20 our pipeline, until they made all the mistakes the 21 first time, I was like salmon swimming up a 22 waterfall. Now that they have got the hard 290 1 questions, and I have been able to answer them with 2 data, now I have no problem, they are calling me up 3 and saying, okay, you need to do this for the next 4 compound, so it is those wins that help the 5 snowball run, but also having documents which have 6 examples of when it didn't work, and having the 7 folks at the Agency at large meetings like an end 8 of Phase II where senior vice presidents of 9 clinical research are present, and you folks start 10 to say, gee, how are you selecting the dose, what 11 is going on here, I mean you are going to Phase 12 III, but we are not certain, it is not transparent 13 to us, or where is your population PK, that really 14 sends a really strong message. 15 DR. POWERS: There is an old proverb that 16 says you should learn from other people's mistakes 17 because you won't have time to make them all 18 yourself. 19 It would seem we have talked a lot about 20 managing risk. Is there no trepidation on the part 21 of management that they are about to select a dose 22 and go into a Phase III trial, and spend millions 291 1 of dollars, and not know whether they are going to 2 get those two adequate and well-controlled trials 3 to work, isn't that enough of a risk to make them 4 want to know these things before they move ahead? 5 I guess not. 6 DR. TALBOT: I have a comment on that. I 7 would like my current pharmaceutical colleagues to 8 comment, but it is my recollection that senior 9 management says we have to have the NDA filed by X, 10 how are you going to do it? 11 DR. POWERS: This has got some timeline 12 deadline, and you are going to get there no matter 13 what else happens, but if you don't get there 14 because your drug didn't work, then, what have you 15 accomplished? 16 DR. POWELL: There is marked differences 17 in company culture in that regard. I have worked 18 in one company that clearly time was the piece that 19 was most valued in terms of development. You hit 20 the times and if you stood in the way of them, you 21 got into trouble. 22 When Pfizer ate [?] us a couple of years 292 1 ago, I was really amazed by their culture, and Lisa 2 may want to speak about this, but they literally 3 stopped development at the Phase II point, or they 4 used to, for on the order of a year to try and get 5 the decision right in terms of whether to go 6 forward or not, and, secondly, what the Phase III 7 trial would be designed. 8 So, I mean you see these radical 9 differences between companies in terms of exactly 10 what you are saying. It is very variable. 11 DR. TALBOT: I think another element, 12 though, just to go back to Dennis' question about, 13 you know, if you have two non-inferiority studies 14 that are successful, does it matter how you got 15 there, I think an important distinction to make in 16 the whole thought process is whether you are 17 dealing with another fluoroquinolone or whether you 18 are dealing with a new compound with a novel 19 mechanism of action, and in the latter case I 20 suspect it's--well, hopefully, it is going to be 21 easier to convince people of the necessity of 22 having an ironclad dose rationale selection. 293 1 But not to minimize all the other 2 excellent points that have been made about the 3 obstacles both at the investigative site level and 4 at the company management level and at the project 5 team level, that impede the utilization of this 6 tool. 7 DR. COLANGELO: I think that in Phase II, 8 not only do we want to obviously explore efficacy, 9 but we may be able to go to a lower dose that give 10 us the same efficacy, but with more acceptable 11 safety. You know, there is a lot of 12 advantages--disadvantages obviously--but, you know, 13 I think at the end of the day, it seems like a lot 14 of times our questions are, well, golly, couldn't 15 they have gotten by with a lower dose at the end of 16 a Phase III evaluation. 17 DR. EDWARDS: Yes. 18 DR. TULKENS: May I maybe comment on that 19 and maybe just try to rephrase the discussion? We 20 started the meeting by saying we are lack of good 21 drugs for bad bugs, so the resistance problem is 22 something that is driving us today. 294 1 Now, if I hear you, Phil, I am afraid of 2 one thing. About 10 years ago or 8 years ago, 3 companies were looking for low doses, and they were 4 introducing low doses of anything they could do 5 including antibiotics. I am pretty confident that 6 looking for low doses for antibiotics is going to 7 drive resistance, and I would submit to you the 8 following thing, is that actually taking a PK/PD 9 parameter suggests the static dose might even be 10 dangerous because the bacteria are still there, and 11 that is the best way to actually construct 12 resistance or select resistance. 13 I was surprised to see here among all the 14 various presentations that resistance was spoken 15 about, but very few data were given. The only ones 16 that I really saw was those of George Drusano. 17 I would submit again to you that we should 18 perhaps, at the level of the regulatory, try to 19 introduce a concept which is, first of all, try to 20 see whether we minimize the risk of resistance, is 21 there anything that could demonstrate that first? 22 Second, is the dose which is given to 295 1 patients going to eradicate, eradicate the bacteria 2 very effectively, because once again, if we do not 3 eradicate, we know that is the best way to drive 4 for resistance, and this is what has been done over 5 the last 20 years, because we identified dose that 6 were effective. We were curing patients, weren't 7 we? 8 But yet we have seen the resistance 9 emerging. Sometimes we do see it very quickly in 10 one patient, that is with the gram-negative 11 bacteria, sometimes in the population basis, and 12 you have to face the reality that--let's take 13 macrolides. Macrolide resistance has gone from 14 almost a few persons to 30, 35 persons in about 10, 15 15 years. 16 We believe, or I believe at least, I 17 submit to you that this is partially due to the low 18 doses of many of the antibiotics that were used. 19 So, in other words, my point here, which I tried to 20 make, is that we should perhaps in the PK/PD 21 modeling system that we use, introduce the concept 22 that resistance is an issue, and not satisfy 296 1 yourselves by just efficacy for the patient, it 2 should also be efficacy on a population basis, and 3 efficacy for the years to come. 4 DR. EDWARDS: Thank you very much. 5 I think I would like to take that comment, 6 if I could, in this formal part of the discussion. 7 It really goes back to where we started in a way. 8 So, I am going to do that. I am going to 9 actually need 10 minutes myself, if I could get 10 that, in that I feel that we really need to 11 verbally produce a summary here that hopefully will 12 engender some feedback, so that we can put a 13 summary into a written form, that will be 14 representative and not more or less personal as I 15 am going to do it now, and it really will be only 16 10 minutes, but I will need 10 minutes. 17 Concluding Remarks 18 Before I start, I need to thank everyone 19 involved with this meeting including the FDA, ISAP, 20 the NIH, and the CDC who came, the IDSA. There are 21 a number of individuals I need to single out, and 22 they include John Powers and Mark Goldberger, Ed 297 1 Cox, Renata Albrecht, Janice Soreth, John Lazor, 2 and the FDA Biopharm staff. 3 I would like to thank Bob Powell, 4 unfortunately, he just left us, for his comments. 5 Dana Schuhly and Christine Moser, and all the FDA 6 staff for their support, and Bob Guidos from the 7 IDSA, and certainly Leo Chang, who I have been 8 extremely concerned about this physical welfare 9 maintaining through this entire time. Leo has just 10 done a tremendous job of helping us keep all this 11 together. 12 And especially all of you who took the 13 effort to come down and spend your valuable time at 14 this meeting over this extremely important topic. 15 The amount of effort that I know John 16 Powers put into getting this meeting together is 17 extraordinary, and I really appreciate all of that, 18 John. I know from talking to John late at night 19 and early in the morning, and on Saturdays and 20 Sundays, what kind of effort has gone into this, 21 and we thank you very much. 22 I would suggest if any of you have any 298 1 feedback to give to John, that you probably not 2 call him after 10 o'clock tonight. 3 I am about to extemporaneously, in front 4 of an audience with collectively hundreds of years 5 of experience in this area, summarize 24 6 presentations and 7 discussions that have occurred 7 over the last two days, and if anyone would like to 8 do this instead of me, would you please raise your 9 hand. 10 I didn't have any takers there. 11 Let me go ahead and, as I say, what I am 12 going to say is at this moment, my own 13 interpretation, which hopefully will be modified as 14 we try to produce a written summary, so we will 15 have to take the comments in that context. 16 We are going to shift gears and go all the 17 way back now to yesterday morning, where we started 18 out with a much more policy orientation to the 19 discussions. During that part of the discussion, 20 the question was raised are we in a crisis here now 21 with the availability of anti--particularly 22 bacterials--but anti-infectives in general and 299 1 resistance. 2 The point was brought up by George Talbot 3 that a crisis really is in the eye of the beholder 4 and it depends on how you define the crisis, and we 5 discussed the issues that it is clear that Big 6 Pharma is leaving the anti-infective development, 7 not wholly, but a significant contingent of Big 8 Pharma is leaving that. 9 We have documentation to that effect. 10 Again, we refer to the fact that the paper that was 11 written, studying the exodus, was just went on the 12 web site yesterday, and, in fact, it did, the 13 Spillberg paper. 14 So, that is a fact, and I think we 15 understand many of the pressures that they are 16 under and many of the reasons why it is happening. 17 It is also a fact that there has been a 18 decline, no matter how you analyze it, in the new 19 molecular entities for anti-infectives. So, over 20 the past several years, there is a documented 21 decline. 22 We discussed the issue that it may be 300 1 comparable to what is going on in classes of drugs 2 outside of anti-infectives, but we also mentioned 3 the fact that that may not be very relevant to the 4 fact that there is, indeed, a decline. 5 We also know that the level of resistance 6 is increasing. That is a fact. The point was 7 brought up by Mark Goldberger that there are lots 8 of antibiotics that are effective against most of 9 the resistant drugs we have at the present time, 10 however, the impact of this emerging resistance is 11 palpably felt in the clinical setting. 12 One thing we haven't heard from during 13 this meeting is a lot of testimony to that effect, 14 but all of us who are seeing patients on a daily 15 basis would use a definition of the word "crisis" 16 that would fit perhaps best with again a term 17 George has so nicely provide for us, and that is 18 that we are in a brewing crisis, and we mentioned 19 the fact that we are all about here working on 20 trying to prevent a potential catastrophe that has 21 not yet happened in its fullest extent. 22 We reviewed the progress that has occurred 301 1 since the last meeting of November 2002, and I am 2 just going to briefly mention again that the IDSA 3 has done many things which were enumerated on a 4 long list by George, and have led to the final 5 stages of the development of a white paper, which 6 will go to Congress, and pressure will be put on 7 our legislature and legislators to provide 8 incentives where it is felt at the present time is 9 going to be the strongest lever in maintaining an 10 active anti-infective program, and the IDSA is very 11 close to actually writing legislation at the 12 present moment. 13 The FDA, Ed gave us a very beautiful 14 summary of all of the activity which has gone on in 15 the FDA, and that has been very impressive. It is 16 clear that the FDA is completely aware of the--I 17 will use the term "brewing crisis," and has been 18 very responsive, but it is all so clear that none 19 of us want to see a compromise in safety and 20 efficacy. 21 So, therefore, we have a very intricate, 22 detailed problem. As someone said earlier, we are 302 1 in a field that does not have a lot of easy 2 solutions. It is a hard field. It can be 3 characterized by the fact that there is a high 4 attrition rate of drugs developing, we are dealing 5 with human safety issues, and there are no easy 6 answers to most of the questions we are trying to 7 address. 8 So, I think all of us, and I am really 9 speaking certainly from the IDSA standpoint, 10 genuinely appreciate the openness that has occurred 11 with the FDA, and I would like to really use the 12 thought John expressed just a moment ago, is that 13 FDA is becoming a team, part of the team of 14 clinical development rather than the police, as 15 someone else used the term, and we feel that. 16 At the same time, guidance is a word that 17 has been used probably more than any other single 18 noun during these discussions, and as I mentioned 19 earlier today, the IDSA, in its summary, will 20 undoubtedly reflect on how useful getting the 21 guidances into a form that can go up for public 22 analysis and then in a finalized form by the end of 303 1 2004, how useful that would be. 2 The FDA has told us that the five 3 guidances that are they are developing, which 4 include the resistant pathogens, meningitis, acute 5 otitis, sinusitis, and AECB are moving along, and I 6 think everyone is--again, the issue was brought of 7 what does industry do with the guidances, how do 8 they use them. Whatever the answer to that 9 question is I don't think any of us have, but we 10 just do know they want them, and that they are of 11 value I think for many reasons, which we actually 12 haven't probed the depth of during these 13 discussions. 14 So, as we got into the surrogate 15 endpoints, I am going to briefly summarize what I 16 think was conceptually important about that 17 discussion, and that they are clearly a mechanism 18 of reducing costs of drug development, they are 19 desirable. 20 That doesn't need to be elaborated upon, 21 but there are very few of them that are in use at 22 the present time, and they are difficult to 304 1 validate, but they must be validated to be useful. 2 Centered around the surrogate endpoints 3 again were two points that the IDSA will again 4 reflect on as selected for use, very useful in the 5 near future, and that would be a consideration of 6 the prosthetic joint infection guidance, if you 7 will, using the surrogate endpoint of bacterial 8 eradication, and the development of a strategy to 9 study staphylococcal bacteremia. 10 The precise mechanism to complete those 11 strategies may perhaps best go to the 12 Anti-Infective Advisory Committee unless another 13 mechanism can be devised to try to bring those two 14 entities to a form of definition. 15 There is one point that I need to mention 16 that is just a little out of context right now, and 17 that is a point that was brought up by John Rex and 18 others, and that is that as we try to influence our 19 societal structure through the legislature, that we 20 need to think about that, not only as it pertains 21 to the United States, but also globally, because of 22 overlap we have with international regulatory and 305 1 areas where industry is involved in anti-infective 2 development. So, I just want to make sure I got 3 that point out. 4 The point was brought up also that the NIH 5 may be an area where development of surrogates can 6 be facilitated. Dennis Dixon has suggested some 7 examples. This may be an area that a great deal of 8 more focus could be put on in the future in terms 9 of expansion of the role of NIH in helping develop 10 validated surrogates. 11 As we moved into the PK/PD discussions, it 12 is without necessity to say that dose selection is 13 an extremely important component for clinical 14 trials and clinical drug development, and that 15 PK/PD is useful for dose selection. 16 That may be its greatest area of 17 usefulness. We reflected on many occasions on the 18 desirability of robust Phase II clinical trials, 19 and I think that bringing that out in this 20 discussion was very useful. 21 There may be other tools that are on the 22 horizon that may facilitate the development of the 306 1 PK/PD science, and area still needs, and will need, 2 clinical validation. The importance of the pre-IND 3 and Phase IIA and pre-NDA discussions was brought 4 out within the context of the fact that we do have 5 a lot of tools for PK/PD that aren't being utilized 6 as extensively as they might be within the context 7 of utilizing those discussions to their maximum 8 benefit. 9 As we moved into the dose selection area, 10 it became clear that there are highly developed and 11 scientifically sophisticated strategies for dose 12 selection at the present time, the two examples we 13 reflected on most acutely are population PK/PD and 14 modeling, population PK/PD modeling and the Monte 15 Carlo simulations. 16 The tools again it was mentioned were not 17 always brought forward by sponsors in a way that 18 provides a complete picture for analysis of data 19 for approval. Again, it was emphasized that the 20 modeling and the tools that are under development 21 are going to have to be validated by clinical data. 22 I am not going to go into some of the 307 1 details of the fact that it is obvious that the 2 PK/PD data cannot be based on MIC alone or area 3 under the curve versus MIC or the Cmax compared to 4 MIC or the time over the MIC. These parameters 5 have to be integrated and paired with other data. 6 The final comments that I will make, and 7 these really are the final comments, is that a 8 problem has been perceived as arising. There have 9 been very strong attempts to discover the details 10 and the pace and the development of the problem. 11 Certainly, industry, FDA, and certainly 12 the IDSA are aware of the brewing problem, and one 13 thing that is clear, and was an interesting 14 discussion I overheard is that those of us involved 15 with all of this are passionate about 16 anti-infectives. 17 I really don't need to mention that to 18 this audience because that is the reason all of us 19 are here and have spent a lot of time from other 20 activities to confront these issues, but we are 21 talking about a class of drugs her that in many 22 ways is different from many other classes and that 308 1 they are acutely life-saving drugs. 2 They work in all age populations of 3 patients. They can prevent things such as 4 blindness and amputations, an additionally, they 5 have been responsible for many of the advances that 6 have occurred in surgery and in cancer chemotherapy 7 development. 8 We are faced with the unique problem that 9 these drugs define their own life span and that the 10 thought leaders at the present time have a tendency 11 to be very conservative with the use of the new 12 drugs, but part of that conservatism is based on 13 the fact that the pipeline is so small at the 14 present time. 15 So, I am going to conclude now and once 16 again thank you all very much for participation in 17 this meeting, and I am sure this will not be the 18 last meeting on this topic, and we look forward to 19 developing plans for the next one. 20 Thank you. 21 [Whereupon, at 4:20 p.m., the meeting 22 concluded.]