FOOD AND DRUG ADMINISTRATION














                       Wednesday, April 14, 2004


                               8:30 a.m.




             Advisors and Consultants Staff Conference Room

                           5630 Fishers Lane

                          Rockville, Maryland






         Arthur H. Kibbe, Ph.D., Chair

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




         Gerald P. Migliaccio, Industry Representative

         Marvin C. Meyer, Ph.D.

         Patrick P. DeLuca, Ph.D.

         Charles Cooney, Ph.D.

         Melvin V. Koch, Ph.D.

         Cynthia R.D. Selassie, Ph.D.

         Nozer Singpurwalla, Ph.D.

         Jurgen Venitz, M.D., Ph.D.

         Marc Swadener, Ed.D., Consumer Representative




         Paul H. Fackler, Ph.D.

         Gordon Amidon, Ph.D., M.A.

         Judy Boehlert, Ph.D.

         Leslie Benet

         Charles DiLiberti

         Laszlo Endrenyi




         Gary Buehler, R.Ph.

         Ajaz Hussain, Ph.D.

         Helen Winkle

         Lawrence Yu, Ph.D.




                            C O N T E N T S


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


      Conflict of Interest Statement,

         Hilda Scharen, M.S.                                     4


      Bioequivalence of Highly Variable Drugs,

         Lawrence Yu, Ph.D.                                      8


      Why Bioequivalence of Highly Variable Drugs is an


         Charles DiLiberti, M.S.                                11


      Highly Variable Drugs:  Sources of Variability,

         Gordon L. Amidon, Ph.D.                                35


      Clinical Implications of Highly Variable

         Dr. leslie Benet                                       60


      Bioequivalence Methods for Highly Variable Drugs,

         Laszlo Endrenyi, Ph.D.                                 81


      Bioequivalence of Highly Variable Drugs Case


         Barbara Davit, Ph.D.                                   99


      FDA Perspectives, Sam Haidar, Ph.D.                      125


      Bioequivalence of Highly Variable Drugs Q&A,

         Dale Conner, Pharm.D.                                 130


      Bioinequivalence: Concept and Definition,

          Lawrence Yu, Ph.D.                                   176


      Statistical Demonstrations of Bioinequivalence,

         Donald Schuirmann, M.S.                               182


      Update--Topical Bioequivalence, Lawrence Yu, Ph.D.       225


      Establishing Bioequivalence of Topical


         Products, Robert Lionberger, Ph.D.                    225


      Future Topics--Nanotechnology, Nakissa Sadrieh,

      Ph.D.                                                    257


      Conclusions and Summary Remarks, Ajaz Hussain,

      Ph.D.                                                    270




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


  2                          Call to Order


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


  4   think we are at 8:30.  It looks like our


  5   electronics are working so we will be in good


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


  7   the conflict of interest statement.


  8                  Conflict of Interest Statement


  9             MS. SCHAREN:  Good morning.  I am Hilda


 10   Scharen.  I am the executive secretary for the


 11   Advisory Committee for Pharmaceutical Science and I


 12   am going to be going through the conflict of


 13   interest statement for the committee.


 14             The following announcement addresses the


 15   issue of conflict of interest with respect to this


 16   meeting and is made a part of the record to


 17   preclude even the appearance of such at this


 18   meeting.


 19             Based on the agenda, it has been


 20   determined that the topics of today's meetings are


 21   issues of broad applicability and there are no


 22   products being approved at this meeting.  Unlike


 23   issues before a committee in which a particular


 24   product is discussed, issues of broader


 25   applicability involve many industrial sponsors and




  1   academic institutions.  All special government


  2   employees have been screened for their financial


  3   interests as they may apply to the general topics


  4   at hand.


  5             To determine if any conflict of interest


  6   existed, the agency has reviewed the agenda and all


  7   relevant financial interests reported by the


  8   meeting participants.  The Food and Drug


  9   Administration has granted general matters waivers


 10   to the special government employees participating


 11   in this meeting who require a waiver under Title


 12   XVIII, United States Code Section 208.


 13             A copy of the waiver statements may be


 14   obtained by submitting a written request to the


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


 16   of the Parklawn Building.


 17             Because general topics impact so many


 18   entities, it is not prudent to recite all potential


 19   conflicts of interest as they may apply to each


 20   member and consultant and guest speaker.  FDA


 21   acknowledges that there may be potential conflicts


 22   of interest but, because of the general nature of


 23   the discussion before the committee, these


 24   potential conflicts are mitigated.


 25             With respect to FDA's invited industry




  1   representative, we would like to disclose that


  2   Gerald Migliaccio is participating in this meeting


  3   as an industry representative, acting on behalf of


  4   regulated industry.  Mr. Migliaccio is employed by


  5   Pfizer.  Dr. Paul Fackler is participating in this


  6   meeting as an acting industry representative.  Dr.


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


  8             In the event that the discussions involve


  9   any other products or firms, not already on the


 10   agenda, for which FDA participants have a financial


 11   interest, the participants' involvement and their


 12   exclusion will be noted for the record.  With


 13   respect to all other participants, we ask in the


 14   interest of fairness that they address any current


 15   or previous financial involvement with any firm


 16   whose product they may wish to comment upon.  Thank


 17   you.


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


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


 20   like to ask everybody to introduce themselves and


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


 22   Lawrence?


 23             DR. YU:  Lawrence Yu, Director for


 24   Science, Office of Generic Drugs, Office of


 25   Pharmaceutical Science, CDER, FDA.




  1             DR. BUEHLER:  Gary Buehler, Director,


  2   Office of Generic Drugs, Office of Pharmaceutical


  3   Science, CDER.


  4             DR. HUSSAIN:  Ajaz Hussain, Deputy


  5   Director, Office of Pharmaceutical Science, CDER.


  6             MS. WINKLE:  Helen Winkle, Director,


  7   Office of Pharmaceutical Science, CDER.


  8             DR. AMIDON:  Gordon Amidon, University of


  9   Michigan.


 10             DR. VENITZ:  Jurgen Venitz, Virginia


 11   Commonwealth University.


 12             DR. SELASSIE:  Cynthia Selassie, Pomona


 13   College.


 14             DR. BOEHLERT:  Judy Boehlert, and I have


 15   my own pharmaceutical consulting business.


 16             DR. SWADENER:  Marc Swadener, consumer


 17   representative, retired from University of


 18   Colorado, Boulder.


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


 20   Professor of Pharmaceutical Sciences at Wilkes


 21   University.


 22             DR. MEYER:  Marvin Meyer, formerly


 23   University of Tennessee professor, now living in


 24   Boca Raton, Florida.


 25             DR. SINGPURWALLA:  Nozer Singpurwalla,




  1   George Washington University.


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


  3   Center for Process Analytical Chemistry at the


  4   University of Washington.


  5             DR. COONEY:  Charles Cooney, Professor of


  6   Chemical and Biochemical Engineering at MIT.


  7             DR. DELUCA:  Pat DeLuca, University of


  8   Kentucky.


  9             MR. MIGLIACCIO:  Gerry Migliaccio, Pfizer.


 10             DR. FACKLER:  Paul Fackler, industry


 11   representative, Teva Pharmaceuticals.


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


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


 14   our discussion.  Lawrence?


 15             Bioequivalence of Highly Variable Drugs


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


 17   are in a different file so I will give my


 18   introduction without the slides.


 19             Dr. Kibbe, Chair of the FDA Advisory


 20   Committee for Pharmaceutical Science, members of


 21   the FDA Advisory Committee for Pharmaceutical


 22   Science, distinguished speakers, distinguished


 23   guests and distinguished audience, I am Lawrence


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


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




  1             This morning it gives me great pleasure


  2   and privilege to introduce to you the first topic


  3   of bioequivalence, bioequivalence of highly


  4   variable drugs.  The objectives of this discussion


  5   are to explore and define bioequivalence issues of


  6   highly variable drugs, to discuss and to debate


  7   potential approaches in resolving them,


  8   specifically the pros and cons of the solutions and


  9   the benefits and limitations of these potential


 10   approaches.


 11             The bioequivalence issues of highly


 12   variable drugs have been discussed in many


 13   conferences and meetings nationally and


 14   internationally.  The issue is obvious because of


 15   the high variability of the drugs or drug products


 16   that require a large number of subjects or


 17   volunteers in order to pass the confidence interval


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


 19   despite many, many publications in scientific


 20   literature, to date there is no consensus and no


 21   solutions have ever been reached.  In fact, there


 22   is no regulatory definition with respect to the


 23   high variability drugs or drug products.  So, there


 24   are various approaches in resolving this in the


 25   scientific literature, for example, expansion of




  1   the bioequivalence limits; for example, using


  2   scaling approaches.


  3             We have invited a panel of distinguished


  4   speakers this morning to discuss this issue related


  5   to the bioequivalence of highly variable drugs from


  6   various perspectives, from practical difficulties


  7   of bioequivalence of highly variable issues, from


  8   mechanistic understanding of what causes the high


  9   variability of drug or drug products, from


 10   understanding of different approaches to resolve


 11   understanding of clinical implications why high


 12   variability drugs are safer, from case studies and,


 13   finally, from regulatory options.


 14             At the end of these presentations you will


 15   be asked to discuss or address the following


 16   questions.  First, what is actually the definition


 17   for highly variable drugs or drug products?


 18             Second, with respect to expansion of


 19   bioequivalence limits, what additional information


 20   should we gather in order to answer this question?


 21   We also ask you to comment on scaling approaches.


 22             With this introduction, I want to turn the


 23   podium over to our first speaker, Charlie


 24   DiLiberti.  Charlie?


 25           Why Bioequivalence of Highly Variable Drugs




  1                           is an Issue


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


  3   I start I need to disclose the potential conflict


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


  5   also a shareholder and option holder in the firm.


  6             Also, before I get into the actual


  7   discussion I would like to say that in the context


  8   of preparing this presentation I had numerous


  9   discussions with many of my colleagues in the


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


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


 12   to portray in my presentation are quite widely held


 13   in the industry.


 14             [Slide]


 15             With that, let's start off with the


 16   definition of highly variable drugs.  Oftentimes,


 17   highly variable drugs are defined in the context of


 18   within-subject variability in terms of a


 19   bioequivalence study.  I would like to take it one


 20   step further and look at variability within the


 21   patient and what does this high level of


 22   variability mean to an individual patient taking


 23   the drugs.


 24             Commonly, the often used definition of


 25   highly variable drugs is those drugs whose




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


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


  3   approximately 30 percent or more.  I will use that


  4   as my guideline for the rest of this presentation.


  5             [Slide]


  6             What are the current criteria?  Just very


  7   briefly, for bioequivalence they involve a


  8   comparison between test and reference product,


  9   involving the natural log transformation of the


 10   data.  The current criteria are that the 90 percent


 11   confidence intervals around the geometric mean


 12   test/reference ratios have to fall entirely within


 13   the range of 80-125 percent.


 14             These criteria really apply to all drugs


 15   here, in the U.S., regardless of the inherent


 16   variability of the drugs.  These criteria do have


 17   other implications.  For example, they can be used


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


 19   justify a substantial formulation change so it is


 20   not just in the context of approving a generic.


 21             [Slide]


 22             This really speaks to the crux of the


 23   issue with highly variable drugs in that it


 24   portrays the number of subjects that you would have


 25   to plan on using in a two-way crossover




  1   bioequivalence study given a particular


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


  3   drugs the number of subjects required is fairly


  4   small and quite manageable from a practical


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


  6   that the number of subjects required can increase


  7   to quite large numbers, possibly in the hundreds.


  8             [Slide]


  9             Why do we possibly need alternative


 10   criteria for highly variable drugs?  Well, first of


 11   all, we have an ethical mandate to minimize human


 12   experimentation.  Second of all, the prohibitive


 13   size of some bioequivalence studies for some highly


 14   variable drugs impacts on the availability of a


 15   generic version of that drug, which may mean that


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


 17   afford the reference product so they may go either


 18   untreated or they may be subdividing their doses


 19   contrary to the prescription.


 20             Also, changing criteria will reduce the


 21   number of participants in the BE studies and I


 22   think it can't be done without compromising the


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


 24   experience elsewhere in the world with criteria


 25   other than 80-125 percent.




  1             [Slide]


  2             This slide shows some of the


  3   bioequivalence criteria in other countries and


  4   regions in the world.  These are not specific to


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


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


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


  8   certainly, there is experience with certain drugs


  9   in these different regions with confidence


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


 11   the case of Canada for many drugs there is no


 12   confidence interval criterion, just a point


 13   estimate criterion.


 14             [Slide]


 15             What types of drugs are highly variable?


 16   Well, the types of drugs really cut across all


 17   therapeutic classes and include both new and older


 18   products.  The potential savings to American


 19   consumers could possibly be in the billions of


 20   dollars if generics are approved.  In saying this,


 21   I want to be clear that the bioequivalence issues


 22   for many of these drugs are not the only barriers


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


 24   patent issues or formulation issues as well, but


 25   still the bioequivalence issues do represent some




  1   sort of a barrier.


  2             What are some examples?  This is a very


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


  4   to give you some kind of representative examples of


  5   drugs that cut across many therapeutic areas, some


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


  7   give a flavor.


  8             [Slide]


  9             Another issue is that as of last year we


 10   now have to meet confidence interval criteria for


 11   fed bioequivalence studies.  So now the variability


 12   under the fed state is of concern.  There is


 13   generally very little information available on the


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


 15   found that some drugs do show more variability


 16   under fed conditions than under fasting conditions,


 17   leading to the potential for bioequivalence


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


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


 20   lack of information on many drugs under fed


 21   conditions, there may in fact be many more highly


 22   variable drugs than we are led to believe.


 23             [Slide]


 24             Why aren't the current criteria


 25   appropriate for some highly variable drugs?  Well,




  1   I will start off by saying that the current


  2   criteria are, I believe, appropriate for drugs with


  3   low to moderate variability because the


  4   dose-to-dose variability that a patient would


  5   experience is comparable and consistent with the


  6   width of the criteria.


  7             However, in the case of highly variable


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


  9   variability experienced by a patient may often be


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


 11   illustrate this point later on with some graphs.


 12             Highly variable drugs are oftentimes wide


 13   therapeutic index drugs.  In other words, they have


 14   shallow response curves and wide safety margins.  I


 15   want to qualify this statement by saying when I say


 16   highly variable drugs, highly variable in a patient


 17   with respect to the parameter that is variable.  If


 18   a patient experiences high variability, that means


 19   that the drug is safe and effective despite this


 20   wide variability in the patient.  Therefore, I


 21   believe that modifying bioequivalence criteria on


 22   highly variable drugs to reduce the number of


 23   participants in bioequivalence studies could be


 24   accomplished while still maintaining safety and


 25   efficacy assurance.




  1             [Slide]


  2             Different highly variable drugs may


  3   require different approaches.  One size may not fit


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


  5   that I had plotted, obviously the number of


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


  7   coefficient of variation is very different from the


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


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


 10   considerations that we have to take into account.


 11             [Slide]


 12             Probably one of the more important


 13   considerations is whether the drug accumulates in a


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


 15   of a drug that does not experience significant


 16   accumulation to steady state in a patient.  These


 17   are typically short half-life drugs, in other


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


 19   interval.  Here are some examples.  We could


 20   possibly consider some sort of modification to the


 21   criteria for both AUC and Cmax because an actual


 22   patient would experience significant dose-to-dose


 23   variability for both Cmax and AUC because neither


 24   is smoothed out at steady state.  Therefore, the


 25   drug could be considered to exhibit a wide




  1   dose-to-dose variation in blood levels irrespective


  2   of chronic dosing.


  3             The same sort of logic could potentially


  4   apply to a highly variable drug that is not dosed


  5   chronically.  One particular application of the


  6   scenario of a relatively short half-life drug that


  7   does not undergo accumulation might be the case of


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


  9   variability where there is also a metabolite that


 10   needs to be measured which has a much longer


 11   half-life and low variability.  I could easily


 12   envision the case where the confidence interval


 13   criteria are somehow modified to accommodate the


 14   higher variability of the parent drug but, in the


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


 16   applied to the metabolite.


 17             [Slide]


 18             Now let's look at the case of accumulation


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


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


 21   relative to the dosing interval so there is some


 22   accumulation going on.  Here are a few examples.


 23             In this case, because the accumulation


 24   process will tend to reduce the fluctuation in AUC


 25   and Cmax, both at steady state, actually in




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


  2   highly variable because the variability may be


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


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


  5   At steady state the test/reference ratio for two


  6   drugs, assuming linear accumulation, will be about


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


  8   a single dose study because the accumulation


  9   process preserves that test/reference ratio.


 10             However, for Cmax, generally speaking, the


 11   test/reference ratio that we see at single dose


 12   conditions will be the most extreme and the


 13   test/reference ratio observed upon accumulation to


 14   steady state will go closer and closer to unity,


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


 16   to consider these two cases differently in the case


 17   of a drug that accumulates.


 18             [Slide]


 19             The other possibility with drugs subject


 20   to accumulation is to actually conduct the steady


 21   state study but this has all sorts of practical


 22   limitations for some drugs, including toxicity.


 23             [Slide]


 24             What I have tried to do in this graph is


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




  1   fluctuations in a pharmacokinetic parameter--I have


  2   plotted this as if it were Cmax but it could


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


  4   does not undergo accumulation.


  5             What is plotted here, in orange, is


  6   simulated data representing the sequential


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


  8   patient taking a single drug over the course of 30


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


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


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


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


 13   see that the drug is fairly well controlled within


 14   a fairly narrow range.  Just as a yardstick for


 15   variability, I have plotted the bioequivalence


 16   limits, the 80 percent limit and the 125 percent


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


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


 19   just plotting them here to give some sense of


 20   scaling.


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


 22   a different formulation, formulation B of the same


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


 24   percent higher mean than this.  CV is still 10


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




  1   magnitude of change that one would expect upon


  2   switching a patient from one formulation to a


  3   second formulation with a higher mean.  You can see


  4   that there is some degree of overlap between the


  5   second formulation and the first but, just


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


  7   there is visually some discernible shift in the


  8   overall levels.


  9             [Slide]


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


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


 12   here that there are many more excursions on a


 13   single formulation outside the range of 80-125


 14   percent.  Overall, there is much more overlap


 15   between formulation B and formulation A despite the


 16   fact that these two formulations differ by 25


 17   percent.


 18             [Slide]


 19             Let's increase the variability one notch


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


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


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


 23   You can see now that the overlap between


 24   formulation B and formulation A, again a 25 percent


 25   difference here, is almost not discernible at all




  1   to the eye.


  2             [Slide]


  3             Finally, let's turn it up one notch


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


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


  6   single formulation with no switch involved, with a


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


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


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


 10             In light of this, suppose that this is a


 11   reference drug that is already approved by the


 12   agency and known to be safe and effective, that


 13   safety and efficacy is true in spite of the wide


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


 15   cannot have a narrow therapeutic index and must


 16   necessarily have a relatively wide therapeutic


 17   index if it is safe and effective despite such wide


 18   variation.


 19             Also, you can see that the switch-over


 20   product, formulation B, again a 25 percent higher


 21   mean, is virtually indistinguishable now from the


 22   range of blood levels that you see with formulation


 23   A.


 24             I think that the criteria, which are still


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




  1   commensurate with the degree of overlap that we are


  2   trying to achieve between formulations.  Even


  3   though these are the criteria, I would like to


  4   point out that in order to pass the criteria the


  5   actual observed mean in a bioequivalence study


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


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


  8   your chances of passing a bioequivalence study on a


  9   very variable drug are very, very poor.


 10             [Slide]


 11             There are certain special considerations


 12   that we need to take into account in the discussion


 13   of highly variable drugs, one of which is where


 14   parallel studies are conducted for long half-life


 15   drugs.


 16             Oftentimes you can't do a crossover study


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


 18   Powering parallel studies depends on between


 19   subject variability rather than within subject


 20   variability.  Between subject variability is often


 21   large, necessitating large bioequivalence studies


 22   just as with highly variable drugs.  However, the


 23   high between subject variability does not


 24   necessarily imply high within subject variability.


 25   Instead, it could be due to inter-individual




  1   differences in absorption, metabolism, etc.  So,


  2   these drugs, from a clinical perspective, may not


  3   really be highly variable but we are still faced


  4   with the powering problems in terms of conducting


  5   bio studies.  In these cases, generally speaking,


  6   multiple dose studies are not feasible, and we


  7   might consider some sort of alternative criteria


  8   for such studies.


  9             [Slide]


 10             A second issue that arises and is directly


 11   related to the issue of highly variable drugs is


 12   the issue of pooling data from multiple dosing


 13   groups.  Because of the large number of subjects


 14   often required for highly variable drugs,


 15   oftentimes you have to split up dosing into


 16   multiple dosing groups.


 17             Currently, the FDA requires a statistical


 18   test for the poolability of the data from these


 19   multiple dosing groups and the test is a measure of


 20   the significance of the group by treatment


 21   interaction terms in the analysis of variance.  If


 22   this interaction term is statistically significant,


 23   then you are not permitted to pool the data from


 24   the multiple dosing groups.  The consequence of


 25   this is that each group is then evaluated on its




  1   own merit and, because each group is generally


  2   considerably smaller than the total pool of


  3   subjects, each group will be grossly under-powered


  4   to achieve bioequivalence and, therefore, if you do


  5   have a statistically significant interaction term,


  6   overall you are likely to have failed the criteria.


  7             This procedure results in discarding and


  8   having to repeat about 5 percent of studies based


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


 10   underlying effect.  The concern here I think is


 11   that even if there were some sort of underlying


 12   explanation for the statistical significance of the


 13   interaction term, for example differences in


 14   demographics among the dosing groups, I believe


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


 16   all the dosing groups because had they been dosed


 17   together in a single group it would be perfectly


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


 19             [Slide]


 20             Conclusions--while the current


 21   bioequivalence acceptance criteria I believe are


 22   appropriate for drugs with ordinary variability, I


 23   believe that they may not be appropriate for some


 24   highly variable drugs.


 25             Current bioequivalence acceptance criteria




  1   make it difficult or impossible to develop generics


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


  3   effectively denying treatment to many patients


  4   because of affordability issues.


  5             I believe that practical, scientifically


  6   sound alternative bioequivalence acceptance


  7   criteria could be implemented for highly variable


  8   drugs to reduce the bioequivalence study size while


  9   still maintaining assurance of safety and efficacy.


 10             Different approaches may be needed for


 11   different types of drugs depending on accumulation


 12   following multiple dosing, and also depending on


 13   the variability of the drug.  And, other related


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


 15   multiple dosing groups should also be considered in


 16   conjunction with any changes to acceptance criteria


 17   for highly variable drugs.  Thank you.


 18             DR. KIBBE:  Does anybody on the panel have


 19   questions for our presenter to clarify information?


 20   Nozer?


 21             DR. SINGPURWALLA:  Certainly, I do.  I


 22   have four questions and five comments.  Do I have


 23   time?


 24             DR. KIBBE:  You have until everybody


 25   leaves to go to the airport!




  1             DR. SINGPURWALLA:  The first question is a


  2   question of clarification.  What is Cmax?  when


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


  4             MR. DILIBERTI:  That represents the


  5   maximum because concentration achieved within a


  6   given patient or subject over the course--


  7             DR. SINGPURWALLA:  So, it is maximum blood


  8   concentration?


  9             MR. DILIBERTI:  Yes, it is maximum blood


 10   concentration.


 11             DR. SINGPURWALLA:  Thank you.  What is


 12   AUC?


 13             MR. DILIBERTI:  Area under the curve,


 14   which is generally taken to be a measure of the


 15   extent of absorption.


 16             DR. SINGPURWALLA:  The third question is


 17   why did you take natural logs?


 18             MR. DILIBERTI:  It is conventional in the


 19   analysis of bioequivalence data to do a log


 20   transformation.  This is already established as


 21   standard--


 22             DR. SINGPURWALLA:  Log transformation of


 23   the whole data or just the maximum?


 24             MR. DILIBERTI:  You would log transform


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




  1   by appropriate analysis of variance.  The same log


  2   transformation also applies to the individual AUCs


  3   prior to analysis of variance.


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


  5   log transformation of all the data to get


  6   approximate normality if the distribution is log


  7   normal.


  8             MR. DILIBERTI:  Yes, that is true.


  9             DR. SINGPURWALLA:  Just taking the log of


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


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


 12             MR. DILIBERTI:  The geometric mean is what


 13   results from the log transformation.  You do the


 14   log transformation and conduct analysis of


 15   variance.  From the analysis of variance you get a


 16   least-squares mean on a log transformed variable.


 17   When you back-transform that by exponentiating it


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


 19             DR. SINGPURWALLA:  Okay.  Now we will go


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


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


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


 23   should really be looked at as a bivariate problem.


 24   You have two variables.  One variable is the extent


 25   of absorption and the other variable is the rate of




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


  2   because the following is possible, suppose you have


  3   a drug which has a low variability with respect to


  4   absorption but high variability with respect to


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


  6   what we need is a better measure of classifying a


  7   highly variable drug which is a bivariate measure.


  8   That is the first comment.


  9             You proposed, I think, abolishing the


 10   confidence limit notion.


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


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


 13   here to really identify what the concerns and


 14   problems are.


 15             DR. SINGPURWALLA:  Okay, but do you have


 16   any sense of what is an alternative?


 17             MR. DILIBERTI:  Various alternatives have


 18   been proposed, including reference scaling or some


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


 20             DR. SINGPURWALLA:  But you are not putting


 21   those forward?


 22             MR. DILIBERTI:  I am not really here to


 23   discuss that.


 24             DR. SINGPURWALLA:  So, your basic focus is


 25   criticizing what is there but without an




  1   alternative in mind?


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


  3   later speakers will address the issue of potential


  4   solutions.


  5             DR. SINGPURWALLA:  Now, in these charts


  6   that you showed, how did you choose the particular


  7   patient whose charts you were showing?


  8             MR. DILIBERTI:  It is simulated data.  It


  9   is log normally distributed random independent


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


 11   thought that that was clear.  It is entirely a


 12   computer simulation just to give some sense of the


 13   relative magnitude of the variability.


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


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


 16             MR. DILIBERTI:  No, no, no.


 17             DR. SINGPURWALLA:  --those data you were


 18   showing.


 19             MR. DILIBERTI:  No.


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


 21   show it because if it is simulated we can


 22   appreciate it.  The last point is when you talked


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


 24   group defined?  What constitutes a group?


 25             MR. DILIBERTI:  By the day on which dosing




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


  2   100 patients or subjects in a clinic all on the


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


  4   today and maybe the other half several weeks from


  5   today.


  6             DR. SINGPURWALLA:  So the groups are


  7   random depending on who shows up.


  8             MR. DILIBERTI:  Essentially, yes.


  9             DR. SINGPURWALLA:  Suppose one were to


 10   think about forming these groups based on some


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


 12   a certain way, conceivably you could justify


 13   pooling.  This is completely random.


 14             MR. DILIBERTI:  Right, and I believe that


 15   the way that the groups are conventionally arranged


 16   in a typical bioequivalence study pooling may be


 17   justified even if you do have a statistically


 18   significant interaction term.


 19             DR. SINGPURWALLA:  See, what I am afraid


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


 21   you had the same policy of pooling at random you


 22   may see a completely different result in the sense


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


 24   Well, thank you.


 25             MR. DILIBERTI:  Thank you.




  1             DR. KIBBE:  Anybody else?  Go ahead.


  2             DR. SELASSIE:  You mentioned that


  3   potential savings to patients are in the billions


  4   of dollars if generics are approved.  Can you tell


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


  6   actually be the lack of savings due to the fact


  7   that there are no generics for each of these as


  8   opposed to other patent issues?


  9             MR. DILIBERTI:  That is very difficult to


 10   assess because, for example, in looking at patents


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


 12   Some of these formulations have patents that are


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


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


 15   I do know from personal experience that the


 16   difficulties in meeting bioequivalence criteria do,


 17   in fact, pose a very real barrier to the


 18   development of some generics.


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


 20   your wife is on premarine you know you insurance


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


 22   currently available because of bioequivalence


 23   issues, instead of $5.00.


 24             MR. DILIBERTI:  Right.


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




  1   just add that I do agree with you about pooling


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


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


  4   measurement.  A week later another patient comes in


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


  6   those separately.  So, unless there is really some


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


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


  9   silly not to put them together.


 10             DR. KIBBE:  Paul?


 11             DR. FACKLER:  If I could just make a


 12   couple of comments, one addressing the issue of AUC


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


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


 15   say that from our experience it is the other way


 16   around.


 17             DR. SINGPURWALLA:  I am sorry, I missed


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


 19             DR. FACKLER:  I am saying that there are


 20   very few examples of drugs that are highly variable


 21   on AUC but not highly variable at Cmax.  Generally


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


 23   as Cmax.


 24             DR. SINGPURWALLA:  So, it makes my point


 25   that you may have a bivariate situation.




  1             DR. FACKLER:  Yes, absolutely.


  2             DR. SINGPURWALLA:  Thanks.


  3             DR. FACKLER:  One of the things I wanted


  4   to ask Charlie was on the simulated data you


  5   represented 80 percent and 125 percent.  I am


  6   wondering did you happen to calculate the


  7   confidence intervals for the simulated data sets to


  8   show where the 90 percent confidence intervals


  9   would have resulted?  Because I am certain they are


 10   far beyond 80-125.


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


 12   not go through that calculation.


 13             DR. FACKLER:  The last point I wanted to


 14   make was that on the graph of the number of


 15   subjects needed to get to 80 percent power versus


 16   the variability, it is important to recognize that


 17   80 percent power means that one out of five studies


 18   under those conditions will fail to show


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


 20   even if a product is tested against itself with,


 21   for instance, 30 percent variability, using the


 22   number of subjects in that particular graph one out


 23   of five studies will fail to show that the product


 24   against itself is bioequivalent.


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




  1   Gordon, you are up.


  2          Highly Variable Drugs: Sources of Variability


  3             DR. AMIDON:  I am going to talk about


  4   sources of variability and emphasize mechanisms of


  5   absorption and focus on bioequivalence from an


  6   absorption point of view.  It is the approach I


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


  8             [Slide]


  9             If you think about bioequivalence where we


 10   are comparing drug products, then the question of


 11   bioequivalence is really a dissolution question.


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


 13   mechanism and dissolution and processes that are


 14   controlling absorption and develop our tests around


 15   that mechanism, what is controlling the process.


 16             Of course, plasma levels are the gold


 17   standard.  Our business is to ensure that plasma


 18   levels match the innovator product used in the


 19   clinical testing.  That is the criterion, no


 20   question about that; no argument about that.  The


 21   question is what test.


 22             [Slide]


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


 24   We tend to focus on bioequivalence from a plasma


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




  1   plasma which is the gold standard.  But if


  2   absorption is controlled by the dissolution


  3   process, dissolution controls the presentation of


  4   drug along the gastrointestinal tract and,


  5   therefore, controls the rate and extent of


  6   absorption.  If the rate and extent of absorption


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


  8   same.  So, in the question of bioequivalence then


  9   the real scientific issue is how do we set a


 10   dissolution standard?  My position may be a little


 11   extreme because no one seems to want to think about


 12   that very much but that is the reality of the


 13   science.


 14             [Slide]


 15             So, I think if you have two drug products


 16   that present the same concentration profile along


 17   the gastrointestinal tract, they will have the same


 18   rate and extent of absorption and systemic


 19   availability.  You may want to think about that,


 20   the same rate and extent of absorption implies the


 21   same systemic availability.  So, we need to focus


 22   on product.


 23             [Slide]


 24             Some of the processes in the


 25   gastrointestinal tract that can lead to the




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


  2   processes here--would be the gastric emptying,


  3   intestinal transit, luminal concentration both of


  4   pH and surfactants, phospholipids, presence or


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


  6   are a lot of sources of variability just in the


  7   gastrointestinal tract.


  8             [Slide]


  9             Systemic availability--what should our


 10   testing ensure?  It is the gold standard, no


 11   question about it.  But the question then is what


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


 13   plasma levels?  And, when plasma levels are


 14   difficult to measure or, in the case of highly


 15   variable drugs where it requires a lot of subjects,


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


 17   the source of that variability and then what type


 18   of test might we set.


 19             I would argue that if two highly variable


 20   drug products dissolve the same way in the


 21   gastrointestinal tract they will be bioequivalent.


 22   It might require 100 subjects to show that.  I


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


 24   with a dissolution test and the answer will be far


 25   simpler.




  1             [Slide]


  2             So, what are some of the physicochemical


  3   factors?  Clearly, particle size and distribution;


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


  5   in some cases chemical instability such as prodrugs


  6   and esterases and peptidases in the


  7   gastrointestinal tract can lead to highly variable


  8   absorption and, hence, systemic availability.


  9             [Slide]


 10             I just put one graph in here showing the


 11   dependence here of dissolution time, ranging up to


 12   30 hours, and gastrointestinal transit time as a


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


 14   in this presentation but the dissolution time


 15   increases dramatically as the drug solubility


 16   decreases.  Particle size becomes a critical factor


 17   for low solubility drugs.  Of course, everyone


 18   realizes that but it is not particle size that we


 19   put into the formulation, it is the particle size


 20   that comes out of the formulation in the


 21   gastrointestinal tract.  So, those process


 22   variables are important.


 23             [Slide]


 24             Some of the factors in the


 25   gastrointestinal tract then are gastric emptying,




  1   intestinal transit, position dependent permeability


  2   along the gastrointestinal tract--duodenum,


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


  4   mucosal cell metabolism, and in particular CYP3A4


  5   which is highly expressed and differentially


  6   expressed along the gastrointestinal tract, and


  7   potentially PGP expression along the


  8   gastrointestinal tract.


  9             [Slide]


 10             To give you an example of variability in


 11   gastric emptying rates, we can just look at the


 12   light blue because that is administered with 200


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


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


 15   involved in drug regulatory standards and realized


 16   that a glass of water was the U.S. standard; not


 17   the standard in Japan.  We are trying to figure


 18   that out, what is a glass of water in Japan.  So,


 19   with 200 ml you can see that the variability in


 20   gastric emptying.  Depending on when you dose in


 21   the fasting state, it ranges from 5 minutes to


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


 23   variation in gastric emptying rate depending on


 24   when you administer to a particular subject.  This


 25   is because of the different contractual activities




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


  2   and 4.


  3             [Slide]


  4             Clearly, intestinal transit--again, this


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


  6   presentation--transit through the gastrointestinal


  7   tract where the drug is released in the duodenum.


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


  9   minutes through the duodenum, jejunum, ileum and


 10   colon.  The dissolution rate, particularly of a low


 11   permeability drug where the permeability appears to


 12   be the rate-determining step to absorption, the


 13   permeability profile along the gastrointestinal


 14   tract is very important.


 15             [Slide]


 16             There are about 10 L of fluid processed in


 17   the gastrointestinal tract per day, actually


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


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


 20   actually ingested as external.  The other 8 L are


 21   ourselves.  We are continually secreting and


 22   reabsorbing not only fluids but cells and proteins


 23   and other ions that are secreted into the intestine


 24   so there is a tremendous amount of variability and,


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




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


  2   the variability and dissolution and absorption in


  3   the gastrointestinal tract.


  4             [Slide]


  5             I show here just ranitidine, a low


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


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


  8   get human data although there is some data


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


 10   a significant difference in permeability.  So, you


 11   can envision a slowly dissolving ranitidine


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


 13   releasing in the ileum would have very poor


 14   absorption.  So, dissolution for a low permeability


 15   drug is probably more important because, in


 16   general, the permeability in the upper part of the


 17   gastrointestinal tract is more important or higher,


 18   I should say.


 19             You know, we used to use language like


 20   "rapidly but incompletely absorbed."  You would see


 21   that in the literature after analysis of


 22   pharmacokinetic data and I would say how can that


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


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


 25   has to be position-dependent permeability and the




  1   absorption rate must decrease dramatically at some


  2   point very quickly after the drug is administered.


  3   Presumably, that is the result of drug getting into


  4   the ileum or distal in the small intestine where


  5   there is lower absorption.


  6             [Slide]


  7             PGP--this is some immunoquantitation


  8   results on CYP3A4 showing the variation in the


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


 10   so that there is less metabolism, particularly if


 11   there is a controlled release formulation releasing


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


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


 14   more about the metabolism source of variability,


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


 16             [Slide]


 17             I am going to propose that we classify the


 18   drugs, highly variable drugs using BCS.  Here is


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


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


 21   see a list of drugs perhaps based on the


 22   variability of reference products, whatever we


 23   could find today, develop a list of highly variable


 24   drugs or that we think might be highly variable,


 25   and then look at their properties and decide what




  1   are the likely sources of variability.


  2             Anyway, I know there are certain so-called


  3   highly variable drugs that are Class I drugs.  They


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


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


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


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


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


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


 10   variability; nothing to do with the product


 11   variability.  Again, that is a hypothesis.


 12             Probably the majority of the drugs that


 13   are highly variable are in Class II where there is


 14   low solubility, potentially Class IV for some


 15   higher molecular weight compounds.  There, the


 16   solubility-dissolution metabolism interaction can


 17   be difficult to separate and that is where we would


 18   need to look more carefully at the drug products to


 19   determine whether it is the solubility and


 20   dissolution variability or whether it is a


 21   metabolism variability that is leading to the high


 22   variability in plasma levels.


 23             [Slide]


 24             So, I think that the BCS classification


 25   can help focus on the source of the high




  1   variability.  Then, in the case of rapid


  2   dissolution of Class I and Class III drugs a


  3   dissolution standard may be enough.  There may not


  4   be too many highly variable drugs because I think


  5   the majority would be the low solubility Class II


  6   or Class IV drugs and there I think metabolism


  7   and/or dissolution can be the source of


  8   variability.  In the case of metabolism, the


  9   metabolism variation would be due to the


 10   variability and dissolution and presentation along


 11   the gastrointestinal tract.  So, again, it comes


 12   back to a dissolution issues.


 13             In fact, I would propose that we look more


 14   carefully at the highly variable drugs, the sources


 15   of variability, again asking the critical question


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


 17   will go back to the original implementation of BCS


 18   in the case of high solubility, high permeability,


 19   rapidly dissolving drugs.  Plasma levels are


 20   telling us nothing about the product differences.


 21   It is only telling us about gastric emptying


 22   differences at the time of administration of


 23   patients or subjects.  So, again, focusing on


 24   dissolution and classification I think can help us


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




  1   not all of the highly variable drugs can be


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


  3   simplified this way.  For those drugs that are


  4   complicated, we just say they are complicated.


  5   Take a drug like premarine.  You have already


  6   mentioned that, Marvin.  I think that premarine is


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


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


  9   today because of the way we regulate drugs and


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


 11   questions by the committee.


 12             DR. KIBBE:  Questions, folks?  Jurgen?


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


 14   much in favor of identifying sources of variability


 15   and what you are presenting are obvious sources of


 16   variability, and it always bothers me when we talk


 17   about highly variable drugs and they are defined


 18   phenologically.  All we are doing is a clinical


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


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


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


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


 23   that.


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


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




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


  2   effects that excipients may have that could be very


  3   different between formulations so that may not have


  4   an impact on dissolution but may have an impact on


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


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


  7   whether that is a significant problem or not but I


  8   think it is more than dissolution that you are


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


 10   recommending, which is basically do dissolution


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


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


 13   comment.


 14             DR. AMIDON:  If we extend the dissolution


 15   to dissolution of the excipient, that is, the


 16   dissolution of the excipient and the drug, then I


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


 18   be okay.


 19             DR. VENITZ:  But if you have products that


 20   have different excipients, that is my point.


 21             DR. AMIDON:  Yes, okay.


 22             DR. VENITZ:  As you said, life is


 23   complicated.  Sometimes it works; sometimes it


 24   doesn't.


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




  1   function of what is the source of the variability.


  2             DR. VENITZ:  Yes.


  3             DR. KIBBE:  Ajaz?


  4             DR. HUSSAIN:  I worked with Gordon for


  5   many years on developing the BCS guideline, and so


  6   forth, and we actually did examine that very


  7   question of excipients and their impact not only on


  8   the dissolution process but on permeability and


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


 10   learn more about transport every day.  Therefore,


 11   clearly, I think when Gordon mentioned dissolution,


 12   we have discussed that so many times and we always


 13   include that as a source of variability and that


 14   has to be considered.


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


 16   different direction.  One of the hesitations as we


 17   developed the BCS guidance was the reliability of


 18   the in vitro dissolution test.  We were not


 19   confident that the current test really was good


 20   enough to extend it to the slower releasing


 21   products.  So, that was the reason we crafted


 22   rapidly dissolving and said dissolution is not rate


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


 24   dissolution test to do that.


 25             I think as we move forward here, I think




  1   what we have done with the PAT initiative is to


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


  3   question what are the criteria variables, what are


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


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


  6   really understand those relationships then you have


  7   a better link between your formulation and your


  8   excipients; you have your process directed to the


  9   clinical relevance.  So, that is the opportunity


 10   that technology is offering us to do that without


 11   having to do an artificial in vitro test where


 12   questions keep continuing and increasing with


 13   respect to the relevance of that in vitro test.


 14             DR. AMIDON:  I certainly obviously agree,


 15   Ajaz.  We have talked about these issues for many


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


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


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


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


 20   determine for any particular drug what might be a


 21   good representative dissolution test, and I might


 22   call that a bioequivalence dissolution test rather


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


 24   you are absolutely right.  The issue is really in


 25   vivo dissolution and how do we capture that in some




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


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


  3   why.  We use the term dissolution very generically


  4   when it should be much more specific.


  5             DR. KIBBE:  Les wants to comment and then


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


  7   are not part of the committee?


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


  9   I wanted to comment on BCS and what Jurgen brought


 10   up in terms of the excipients.  When we initiated


 11   BCS I was very strong concerning the potential for


 12   excipients on Class I drugs and we have written the


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


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


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


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


 17   at least the transporters.  But they will be a


 18   problem with Class III drugs.


 19             So, so far I have been very opposed to


 20   moving the Class III drugs because I can make a


 21   Class III formulation that will pass dissolution,


 22   any dissolution, and fail.  The reason is that


 23   Class III drugs need uptake transporters to get


 24   absorbed and, therefore, I can block an uptake


 25   transporter in the gut with a substance that has no




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


  2   little early in translating this dissolution


  3   criteria beyond Class I, but I think we were


  4   correct in Class I and the extra safeguards we put


  5   in actually really turn out not to be necessary.


  6             DR. SINGPURWALLA:  I like this concept of


  7   looking at the causes of variability.  I see this


  8   as a first step towards going to a Bayesian


  9   alternative for the existing methodology that was


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


 11   question perhaps both for you and also for the


 12   first speaker.  Has anybody looked at the


 13   reliability of the testing instrument itself?


 14   Because if the testing instrument itself shows a


 15   large variability--if the instrument itself shows a


 16   large variability then you don't know whether the


 17   variability is coming from the instrument or from


 18   the particular drug or the combination of the


 19   instrument, the drug and the patient.


 20             DR. KIBBE:  Anybody?  Who wants to handle


 21   that?


 22             DR. VENITZ:  I think by instrument what


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


 24   Are you talking about dissolution or are you


 25   talking about in vivo?




  1             DR. SINGPURWALLA:  Both.


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


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


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


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


  6   depend upon absorption and dissolution; they depend


  7   on everything that happens after the drug gets in


  8   the body, which is something we are not interested


  9   in.  If that contributes significantly to the


 10   variability, then you are looking at primarily


 11   variability and disposition which determines why we


 12   have a highly variable drug, not because there is


 13   variability in absorption.  So, your instrument


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


 15   your lingo.


 16             DR. SINGPURWALLA:  Right.  You have an


 17   instrument by which you measure these things, like


 18   a thermometer.  If your thermometer is bad--


 19             DR. VENITZ:  I am saying that for some


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


 21   instrument and the noise is not related to what you


 22   are trying to measure.


 23             DR. SINGPURWALLA:  Exactly.


 24             DR. KIBBE:  Let me just take the


 25   prerogative of the chair for half a second and then




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


  2   to understand the real noise level of the


  3   instrument.  The instrument is the bioequivalency


  4   test itself and the agency gets submissions with


  5   bioequivalency tests that are passed.  The question


  6   is how many were done that failed before the one


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


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


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


 10   interesting to look at, we would find that the


 11   instrument is very crude and the reason we live


 12   with it is that it is close to the clinical


 13   therapeutic outcomes that we really want to measure


 14   in terms of steps away from that outcome.  What


 15   Gordon is recommending is that we even eliminate


 16   the human from our decision-making process, which


 17   brings us further away from the ultimate goal which


 18   is to know that it therapeutically equivalent, and


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


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


 21   all have been struggling with for 25 years.


 22             DR. HUSSAIN:  Now I have three comments.


 23   With respect to the instrument variability, I think


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


 25   bioequivalence testing we try to minimize that and




  1   try to make it more precise and more accurate by


  2   doing a crossover study.  We test the two products


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


  4   that is our attempt to minimize that.  The other


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


  6   of more similar individuals but we wanted to move


  7   away from that in the general population because


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


  9   pointed out with respect to variability the


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


 11   we need to address that.


 12             But the point I think, going back to the


 13   key question, is what are the important questions


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


 15   bioequivalence.  For therapeutic equivalence our


 16   approach is very simple.  First you need to be


 17   pharmaceutically equivalent and then, if there is a


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


 19   pharmaceutical equivalence for solutions you don't


 20   need a bio study.  So pharmaceutical equivalence,


 21   bioequivalence and then therapeutic


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


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


 24   sense if we understand our formulations, if we


 25   understand our processes, if we understand the




  1   mechanisms, pharmaceutical equivalence essentially


  2   is defining therapeutic equivalence.


  3             DR. AMIDON:  To come back to your question


  4   about the dissolution apparatus, there is a range


  5   of dissolution apparatus in the USP that are used


  6   internationally, and you can study many of the


  7   variables that change in vivo by pH and surfactants


  8   in those apparatus.  The apparatus themselves have


  9   been proven perhaps historically to be very


 10   reliable, although you could argue maybe today that


 11   we could design a better apparatus but that is very


 12   complicated because these things are used in many


 13   companies internationally with defined procedures


 14   that are approved by the regulatory agencies and


 15   making change in an apparatus is a very complex


 16   process.


 17             But, yes, we can study the various


 18   variables in vivo and I think that a dissolution


 19   test that included changes in pH and surfactant to


 20   reflect what is happening in vivo is something we


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


 22   and follow the dissolution as a function of time.


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


 24   insightfully actually.


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




  1   use dissolution is reliable but insensitive, and we


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


  3   conversion.  Anybody else?


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


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


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


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


  8   potentially change our release specifications


  9   because our product is too variable and that is not


 10   acceptable in the manufacturing arena.  You go back


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


 12   how much data is really available on if I gave


 13   myself a rapidly absorbed drug once for the next


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


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


 16   reproducibility in a subject, unless it was the old


 17   multiple dose studies where the drug was


 18   essentially eliminated in 24 hours.


 19             So, I think we need some more information.


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


 21   the innovator firms do special populations and they


 22   find the elderly are different than the young, do


 23   they have to then go further and explain is that


 24   gastrointestinal pH, is it transit, is it


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




  1   think then we can get some background information


  2   on source of variability.


  3             Just to bounce off an idea which is


  4   undoubtedly ludicrous, do we need in a sense to


  5   prescreen some subjects so we have a calibrated man


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


  7   allowed into the study so if they have less


  8   variability they get into our study?  Could we do


  9   that?  One thing that really troubles me is the


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


 11   think I support it, of having different mechanisms


 12   of release tested against each other in a


 13   bioequivalence study, an oral study versus a


 14   particular dosage form.  Intestinal transit can


 15   have a profound difference on those two so if you


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


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


 18   vegetarian might show the oral tablet is excreted


 19   in four hours and the other person may take much


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


 21   of knowing where the problems are; can we reduce


 22   variability somehow; are subjects legitimately--is


 23   that a viable approach?


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


 25   would want to take on preselecting subjects because




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


  2   sense?


  3             DR. MEYER:  I am thinking more in terms


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


  5   do that now somewhat routinely.


  6             DR. AMIDON:  Right.


  7             DR. MEYER:  So, we might give a


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


  9   and over again anyway.  Let's characterize them


 10   first before they are allowed into subsequent


 11   studies.


 12             DR. KIBBE:  Paul, go ahead.


 13             DR. FACKLER:  If I can just comment on


 14   that, we used to do bioequivalence studies in males


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


 16   believe.  The agency has recently requested that BE


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


 18   representative of the American population so we now


 19   include females and we include the elderly, and it


 20   just makes the variability problem that much worse.


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


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


 23   with the same physical habits, generally with the


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


 25   because we have reduced the variability in the




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


  2   recently, in the opposite direction, making these


  3   products in particular less likely to pass against


  4   themselves again.


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


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


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


  8   is, are the two formulations behaving the same,


  9   should be their behavior independent of the


 10   subjects studied, and are there variabilities


 11   between product-subject interactions that might be


 12   significant in special populations.  I think it is


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


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


 15   think Gordon is suggesting is if we understood the


 16   variables we might not have to use that blunt a


 17   tool to estimate what will happen in the average


 18   patient.


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


 20   There was a wonderful report done--Les will


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


 22   the agency that looked at dissolution and tried to


 23   correlate it with bioequivalency data that they had


 24   almost twenty years ago and there was absolutely no


 25   way that dissolution predicted any of the results




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


  2   complicated than it first appears.


  3             DR. AMIDON:  I got involved in this


  4   process about that time, and my position is you


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


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


  7   dissolution test to make it more relevant to the


  8   variables that we need to control to ensure


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


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


 11             DR. HUSSAIN:  The key aspect I think is


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


 13   questions and if a bioequivalence study is only


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


 15   question from the public health aspect because the


 16   product is going to be used in all populations?


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


 18   fundamentals of what is a bioequivalence study.  If


 19   it is just confidence interval criteria, then that


 20   is one aspect.


 21             DR. SINGPURWALLA:  Why not have a separate


 22   set of drugs for different categories of people?


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


 24   10 mg and you specify your milligrams based on the


 25   population.




  1             DR. HUSSAIN:  That is a major aspect of


  2   dose finding and then labeling that goes into the


  3   new drug development process itself.  The


  4   bioequivalence essentially has been a quality


  5   assurance approach to making sure that a


  6   pharmaceutically equal product has an in vivo rate


  7   and extent of absorption similar to the innovator.


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


  9   study, to make sure that your assumptions and your


 10   in vitro methods are more reliable or at least


 11   conform from that perspective.


 12             DR. KIBBE:  Thank you.  Unless someone


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


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


 15   enlighten us.


 16          Clinical Implications of Highly Variable Drugs


 17             DR. BENET:  I am older!


 18             [Laughter]


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


 20   I think the last two times I have appeared before


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


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


 23   opportunity.




 25             We have been discussing at an




  1   international level, I was reminded as I heard


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


  3   consensus conference in 1989 to try to develop


  4   standards for bioequivalence and we are still at


  5   it.


  6             [Slide]


  7             This was said by the first speaker but


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


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


 10   bioequivalence guidelines: the manufacturer of the


 11   test product must show using two one-sided tests


 12   that a 90 percent confidence interval for the ratio


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


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


 15   reference product is within the limits of 0.8 and


 16   1.25 using log transformed data.  It is


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


 18   your mythology, the Procrustes himself was a robber


 19   that took people when they came through his gate


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


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


 22   were too short he stretched them out until they fit


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


 24   Procrustean guidelines that say all drugs must fit


 25   the same criteria no matter what the issues are.




  1             Now, BCS, biopharmaceutical classification


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


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


  4   in looking at dissolution is that we didn't


  5   understand the flawed classifications.  So, the


  6   only time dissolution is going to have any


  7   relevance to bioequivalence or bioavailability is


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


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


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


 11   But I strongly believe and have suggested over a


 12   number of years that there need to be other


 13   non-Procrustean advances and that is what I will


 14   talk about today.


 15             [Slide]


 16             What are we trying to solve?  What are the


 17   bioequivalence issues and what concerns patients


 18   and clinicians so that they have confidence in the


 19   generic drugs that are approved by the regulatory


 20   agencies so that they feel there are no questions


 21   related to their therapeutic efficacy?


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


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


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


 25   FDA bioequivalence issues that ever caused any




  1   therapeutic problems in a prospective study. That


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


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


  4   out there from people who would like them to


  5   question the bioequivalence criteria.  So, this is


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


  7   that we face is not necessarily scientific but it


  8   is creating an environment where the American


  9   public has confidence in the regulations that we


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


 11             But what we have done and what our


 12   concerns are now with therapeutic index drugs, NTI,


 13   we need to have practitioners have assurance that


 14   transferring a patient from one drug product to


 15   another yields comparable safety and efficacy, and


 16   a few years ago we termed that switchability and we


 17   developed or tried to develop a number of


 18   statistical criteria to approach that.  The issues


 19   we are facing today are for a wide therapeutic


 20   index, highly variable drugs which do not have to


 21   study an excessive number of patients to prove that


 22   two equivalent products meet the preset one size


 23   fits all statistical criteria.  So, these are the


 24   issues I want to address and ask the committee to


 25   take cognizance of.




  1             [Slide]


  2             Now, it was not obvious a few years ago


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


  4   narrow therapeutic index drug it is very easy to


  5   pass the bioequivalence criteria, and that is


  6   because narrow therapeutic index drugs, by


  7   definition, must have small intra-subject


  8   variability.  If this were not true for narrow


  9   therapeutic index drugs, patients would routinely


 10   experience cycles of toxicity and lack of efficacy,


 11   and therapeutic monitoring would be useless.  So,


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


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


 14   scientific issue.  We might not have the


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


 16             [Slide]


 17             Let's look at some narrow therapeutic


 18   index drugs.  They have high inter-subject


 19   variability and they have low intra-subject


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


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


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


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


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


 25   although most of them are Class II drugs.




  1             Getting back to the reliability of the


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


  3   Look at the warfarin sodium intra-subject


  4   variability.  The clinical measure that the


  5   clinician uses to judge the status of the patient


  6   in terms of his blood thinning capability, the INH


  7   measurement, is significantly more variable.  So,


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


  9   drug is working is more variable than the patient


 10   is going to experience from dose to dose in terms


 11   of the criteria for this particular drug.  So,


 12   these are interesting questions.


 13             [Slide]


 14             Now, we tried to address this


 15   switchability issue over a long period of time with


 16   the concept called individual bioequivalence, and I


 17   chaired the expert panel for about three years and


 18   tried to address this issue.  The ideas about


 19   individual bioequivalence were that we were going


 20   to get these promises, we would address the correct


 21   question, switchability in a patient.  We would


 22   consider the potential for subject by formulation


 23   interaction.  There would be incentive for less


 24   variable test products.  Scaling would be based on


 25   variability of the reference product both for




  1   highly variable drugs and for certain


  2   agency-defined narrow therapeutic range drugs.


  3   And, we would encourage the use of subjects more


  4   representative of the general population.


  5             In fact, none of that worked and we gave


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


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


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


  9   there is no evidence that the present regulations


 10   are inadequate and that we need to be more rigorous


 11   in our definition related to switchability.


 12             [Slide]


 13             Consider that the subject by formulation


 14   interaction turned out to be an unintelligible


 15   parameter from both the agency and the exterior


 16   scientific community.


 17             Incentive for less variable test products,


 18   yes, but that could be solved by average


 19   bioequivalence scaling and that is what at least I


 20   am here to talk about today.


 21             Scaling based on variability of the


 22   reference product both for highly variable drugs


 23   and for certain agency-defined narrow therapeutic


 24   index drugs, again average bioequivalence with


 25   scaling could solve this issue.




  1             Encourage the use of subjects more


  2   representative of the general population, that was


  3   a good hope but it completely failed in terms of


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


  5   work.


  6             [Slide]


  7             I recognized in Lawrence's introduction


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


  9   variable drugs.  This is the consensus definition


 10   that came out of a number of international


 11   workshops, highly variable drugs should be those


 12   when the intra-subject variability is equal or


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


 14   therapeutic index highly variable drugs we should


 15   not have to study an excessive number of patients


 16   to prove that two equivalent products meet this


 17   preset one size fits all statistical criteria.


 18             This is because, by definition, again


 19   highly variable approved drugs must have a wide


 20   therapeutic index, otherwise there would have been


 21   significant safety issues and lack of efficacy


 22   during Phase III testing.  In fact, highly variable


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


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


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




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


  2   therapeutic index drugs.  We only have drugs that,


  3   with this tremendous variability that we


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


  5   have any problems.  And, those individual patients


  6   having very high levels one time, low levels the


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


  8   areas under the curve the next time get through.


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


 10   need to worry about the genetic differences in


 11   their enzymes.  It has already been shown that,


 12   yes, there are tremendous differences.  Somebody is


 13   going to have very high levels because they lack


 14   the enzyme; somebody is going to have very low


 15   levels but still they are safe and effective


 16   because they are wide therapeutic index drugs.


 17             [Slide]


 18             But it makes it very difficult, as was


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


 20   to be bioequivalent and here is my champion or what


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


 22   seen, and this is progesterone which I believe is


 23   the poster drug for highly variable variability.  A


 24   repeat measures study of the innovator's product


 25   was carried out in 12 healthy post-menopausal




  1   females and it yielded intra-subject variability in


  2   an AUC of 61 percent for the coefficient of


  3   variation and intra-subject coefficient of


  4   variation for Cmax of 98 percent.


  5             If you did the calculations, it came out


  6   that you needed 300 women just to meet the


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


  8   study that a generic company, or at least the


  9   company interested in this, could afford to carry


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


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


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


 13   out of five studies would fail just on statistical


 14   chance and you have carried out a study with 300


 15   people in it to prove that this highly variable


 16   drug is bioequivalent.  This is the issue that we


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


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


 19   variable, very safe, wide therapeutic index drugs


 20   for which we can't prove bioequivalence because of


 21   the inherent variability of the innovator product.


 22             [Slide]


 23             I appeared before this committee three and


 24   a half years ago to give the recommendations of the


 25   FDA expert panel on individual bioequivalence, and




  1   these are some of the recommendations.  One that I


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


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


  4   pleased that that has happened and congratulations


  5   to the agency.


  6             Our recommendations at that time were that


  7   sponsors may see bioequivalence approval using


  8   either average bioequivalence or individual


  9   bioequivalence, and we recommended that the subject


 10   by formulation parameter be deleted since no one


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


 12   statistically.


 13             We asked that scaling for average


 14   bioequivalence be considered, that the agency and


 15   the statistical group go into this and it be


 16   something to be followed up and presented to this


 17   advisory committee at some time in the future.


 18             We recommended at that time that if an IBE


 19   study, individual bioequivalence study, was carried


 20   out and the test product fails you could not then


 21   reanalyze with average bioequivalence because in


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


 23   other.


 24             Here is something that we recommended that


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




  1   do  with confidence.  We recommended the point


  2   estimate criteria be added, and we added this not


  3   on any scientific basis that we are going to rule


  4   out products, we said that these criteria are


  5   always met today and what we have is a conception


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


  7   products that differ by 25 percent, and that we


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


  9   point estimate criterion in addition to our


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


 11   estimate criteria and Cmax plus/minus 20 percent no


 12   matter what you do, and if you have narrow


 13   therapeutic index drugs make it even smaller, make


 14   the point estimate plus/minus 10 percent for AUC


 15   and plus/minus 15 percent for Cmax.


 16             [Slide]


 17             So, what I am suggesting here today and


 18   what I am recommending to the committee to do is


 19   ask the agency to develop methodology, and we are


 20   going to hear some, to allow approval based on


 21   weighting of average bioequivalence analytical for


 22   highly variable drugs so that we can bring some


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


 24   of the progesterone example.  Also, that the point


 25   estimate criteria be added to the criteria because,




  1   in fact, all products will pass these criteria at


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


  3   increase the confidence of those that say, you


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


  5   percent because look at what the FDA criteria say.


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


  7   written two years ago, were easily misinterpreted


  8   but that also changed two years ago and now the


  9   criteria are written in a way that no clinician can


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


 11   misinterpreted.


 12             [Laughter]


 13             They still say exactly the same thing but


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


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


 16   are my recommendations.  Thank you for listening to


 17   me.


 18             DR. KIBBE:  Questions for Dr. Benet?


 19             DR. SINGPURWALLA:  I have a comment not


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


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


 22   industry is buried under the tombstone of


 23   frequentist methods.  Such methods ignore clinical


 24   and biopharmaceutical knowledge, and it is bogged


 25   down by its own weight.




  1             DR. BENET:  I disagree.


  2             DR. SINGPURWALLA:  Why?


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


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


  5   in is safety and efficacy, and in all cases


  6   measures of safety and efficacy are more variable


  7   than any pharmacokinetic measure.  What we are


  8   really interested in, what the agency is interested


  9   in is safety and efficacy.


 10             DR. SINGPURWALLA:  Who said that Bayesian


 11   methods do not incorporate high variability?  It is


 12   these confidence intervals and these confidence


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


 14   understand Bayesian methods.


 15             DR. BENET:  I understand Bayesian methods.


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


 17   wouldn't say this.


 18             DR. BENET:  Well, I welcome the


 19   committee's spending the time discussing this with


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


 21             [Laughter]


 22             DR. MEYER:  I think I agree with


 23   everything you have said and it embarrasses me no


 24   end to say that!


 25             [Laughter]




  1             Is there still going to be a perceived


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


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


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


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


  6   high and low can be switched in the marketplace?


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


  8   get around that.  There are always going to be


  9   people who will take the present situation and use


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


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


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


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


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


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


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


 17   They would be afraid that someone will go out there


 18   and say this product has never been tested in


 19   humans; it was approved on the basis of a


 20   dissolution.  You have confidence in this product


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


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


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


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


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




  1   we can in making it happen.


  2             DR. KIBBE:  Let me just ask about an


  3   application of one of your recommendations to your


  4   own example.  If you use methodology that is


  5   developed as a weighted average, how would that


  6   play out with progesterone?  In other words, what


  7   kind of numbers would we start to work with?


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


  9   weighting to the variability of the innovator


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


 11   the denominator that you would weight.  But there


 12   are different statistical issues that have to be


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


 14   statisticians to tell us how to approach that.  But


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


 16   the variability of the innovator product in terms


 17   of the coefficient of variation for Cmax as one


 18   criterion and for AUC as another criterion.


 19             DR. KIBBE:  I have always found


 20   intellectually attractive the concept of three ways


 21   where we could look at variability and then compare


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


 23   the numbers that we need to make these kinds of


 24   decisions?


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




  1   to be some measure of intra-subject variability.


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


  3   agency for many years to make this a requirement


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


  5   variability in humans or even in patients be


  6   included in the approval process and be included in


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


  8   measure some place.


  9             I am very encouraged, even though the


 10   agency does not require that, that we are starting


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


 12   their package insert, measures of intra-subject


 13   variability included because it is important


 14   criteria and value that clinicians want to know.


 15   What is the inherent pharmacokinetic variability so


 16   that then I can say is the pharmacodynamic


 17   variability more than this inherent pharmacokinetic


 18   variability.  If they don't know the inherent


 19   pharmacokinetic variability, then they have a tough


 20   time making any decision about whether the change


 21   in efficacy is related to pharmacokinetics or to


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


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


 24   you recommend.


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




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


  2   stringent requirement.  Is that because Cmax is


  3   more variable because we don't measure it very


  4   precisely, or is it because Cmax is less important


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


  6   enough data for the latter conclusion.


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


  8   as was initially discussed, it is confounded.  As


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


 10   the agency and many academics have spent years and


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


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


 13   confounded and, as was stated, is always more


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


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


 16   have that has any component of rate in it.


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


 18   more variable.


 19             DR. VENITZ:  Les, I agree with your


 20   additional recommendation to put constraints on the


 21   point estimates.  You mentioned one of the reasons


 22   being that the public needs to be reassured that,


 23   indeed, no matter whether it is unintelligible


 24   regulation or not, we do have generics that are


 25   bioequivalent.




  1             What I am personally not certain about is


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


  3   again, we are going to have some more presentations


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


  5   aggregating variance and mean differences, and I am


  6   not sure whether one can offset the other.  In


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


  8   can that be offset by differences in variance?


  9   When we had the discussion last time with IBE,


 10   surprisingly there were drugs out there in the


 11   database that the FDA provided us with that passed


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


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


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


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


 16   that way.  So, I still personally withhold judgment


 17   on the reference scaling but I am very much in


 18   favor of putting in additional constraints.


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


 20   think having the additional constraints solves part


 21   of the problem.


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


 23   I think the committee at that time went along with


 24   that because we were worried about the IBE not


 25   being conservative enough.  Right now you are




  1   basically breaking drugs down into two categories,


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


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


  4   be used for BE assessment.


  5             DR. BENET:  Yes.


  6             DR. VENITZ:  Can you think of additional


  7   criteria along the lines that we heard Gordon talk


  8   about, that if we understand where the variability


  9   comes from we might use different criteria?  In


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


 11   some decision tree that decides which way we are


 12   going to go?


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


 14   there has nothing to do with science because it is


 15   easy to prove bioequivalence of NTI drugs.  It just


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


 17   it lower, because it is easy to pass.


 18             I definitely believe that as we progress


 19   we are going to have different criteria, and I


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


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


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


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


 24   progress--and I presented to the agency last


 25   November my newest concepts in terms of using BCS




  1   or some sort of variant of BCS to actually predict


  2   drug disposition, and I think we are going to


  3   progress a lot in the next few years.


  4             DR. KIBBE:  Nozer?


  5             DR. SINGPURWALLA:  Well, just a general


  6   comment.  I was pleased to hear you acknowledge


  7   that newcomers can identify things like


  8   confounding, but I also think that newcomers can


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


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


 11   more attention to alternate methods and not get


 12   committed to an old, archaic notion of confidence


 13   intervals.  These have been criticized in the


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


 15   of confidence limits, and the difficulty that


 16   confidence limits poses both to the FDA and also to


 17   the drug industry in getting their drugs approved.


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


 19   attention to alternatives and don't dismiss it.


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


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


 22   informing the committee and the agency of these


 23   approaches and the Bayesian approach, and I think


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


 25   It is important to have fresh eyes and fresh views




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


  2   to recognize that the agency's criteria are safety


  3   and efficacy, and when we have criteria that have


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


  5   that criteria to untested criteria in terms of this


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


  7   be very careful in the changes that they make.


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


  9   more speaker before the break.  Dr. Endrenyi,


 10   welcome.


 11         Bioequivalence Methods for Highly Variable Drugs


 12             DR. ENDRENYI:  Thank you.


 13             [Slide]


 14             This presentation was put together with


 15   Laszlo Tothfalusi and I would like to acknowledge


 16   that.


 17             [Slide]


 18             I would like to raise a number of


 19   questions which I believe that this committee will


 20   have to make recommendations about eventually that,


 21   certainly, the agency ought to consider.  I would


 22   like to go through the first part fairly quickly


 23   because much of that has already been considered.


 24   So, we have the usual criterion of comparing two


 25   formulations and the confidence limits for the




  1   ratio of geometric means should be between 0.8 and


  2   1.25.  This has already been stated.


  3             [Slide]


  4             It has also been stated that for highly


  5   variable drugs this presents a problem because with


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


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


  8   order to satisfy that.


  9             [Slide]


 10             For the purpose of this presentation but


 11   not necessarily as the final word at all, the


 12   coefficient of variation has been considered


 13   exceeding 30 percent for highly variable drugs.


 14             [Slide]


 15             This slide would simply ask is there an


 16   issue and this has already been asked and the


 17   answer was probably yes.  In this case, two


 18   formulations of isoptin are considered in the same


 19   subject repeatedly, and two different occasions


 20   different relationships between the two


 21   formulations were obtained.  So, it looks as though


 22   the drug is not really bioequivalent with itself


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


 24   demonstrated by Dr. DiLiberti.


 25             [Slide]




  1             This is perhaps more recent.  This was


  2   obtained from Diane Potvin, from MDS, who


  3   demonstrated that, indeed, things look reasonable


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


  5   70 percent but beyond that it is very difficult to


  6   satisfy the criteria.  There are many, many studies


  7   submitted that failed.


  8             [Slide]


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


 10   details of these highly variable drugs.  From this,


 11   one could conclude that there is a relationship


 12   between the coefficient of variation and failure


 13   rate, higher failure rate with higher coefficient


 14   of variation.  Mind you, these are all submitted


 15   studies so this analysis is still biased because


 16   the company submitted them in the hope that they


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


 18             The second conclusion is that, indeed,


 19   AUCs fail less frequently than Cmax's but they


 20   still fail with a high frequency.  So, the


 21   variation of AUCs should not be dismissed.


 22             [Slide]


 23             Study condition--perhaps I would omit this


 24   almost entirely because it is considering single


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




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


  2   administration even though it has been demonstrated


  3   and we know that frequently in steady state we get


  4   lower variation--not frequently but not always.


  5             [Slide]


  6             This is a study showing that and in the


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


  8   irrelevant.


  9             [Slide]


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


 11   2 traditional or replicate design?  It need not be


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


 13             [Slide]


 14             Now, the advantage of replicate designs


 15   includes that one gets clear estimates of


 16   within-subject variations.  Particularly the


 17   concern would be to get a clear estimate of


 18   within-subject variation for the reference product.


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


 20   Midha who has worked long years and is certainly


 21   one of the foremost experts on the bioequivalence


 22   of highly variable drugs and drug products.  So,


 23   his voice ought to be respected.


 24             Secondly, on the other hand, my concern is


 25   that one can have a pooled criterion which could




  1   have better properties, pooled criterion related to


  2   the test and reference products together.


  3             There are issues that these replicate


  4   design studies can be evaluated by various


  5   procedures, and a question is whether these


  6   procedures would give the same results and,


  7   therefore, would agencies be able to check how


  8   those results would be calculated and were


  9   calculated.


 10             Another question arises, namely, is a test


 11   comparing the variations of test and reference


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


 13   estimate of these variations simply sufficient or


 14   is that needed?


 15             [Slide]


 16             Turning to the 2 X 2 crossovers, they are


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


 18   advantage is that there are many studies on file


 19   and they could be evaluated retrospectively.


 20             Another comment is that the ratio of


 21   within-subject variabilities could be estimated.


 22   There are procedures that would permit this even


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


 24   procedure suggested here by Guilbaud and Gould is


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




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


  2   then the difference of the two; plot them against


  3   each other, have a linear regression and evaluate


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


  5   which gives the ratio of the estimated variances.


  6   So, it would be possible to evaluate this ratio


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


  8   procedure have not been studied and they ought to


  9   be evaluated.


 10             [Slide]


 11             Now, various possible methods of


 12   evaluation, the usual procedure is unscaled average


 13   bioequivalence with a criterion of 0.8 to 1.25 for


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


 15   possible to apply unscaled average bioequivalence


 16   with expanded bioequivalence limits.  One way of


 17   doing it is to present these bioequivalence limits.


 18   It has been shown that some jurisdictions do this.


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


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


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


 22   bioequivalence limits flexibly depending on the


 23   estimated variation.  I shall talk more about these


 24   procedures.


 25             Another approach is the scaled average




  1   bioequivalence and, again, I shall refer to this


  2   and shall talk about this, and I also should


  3   mention scaled individual bioequivalence for


  4   comparisons only.


  5             [Slide]


  6             To talk about unscaled average


  7   bioequivalence--these scissors are supposed to be


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


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


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


 11   you see here, the ratio of geometric means should


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


 13   This is the same statement as saying that the


 14   logarithmic bioequivalence limits should be plus


 15   and minus and in between is the difference of the


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


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


 18   and 1.25 limits were arbitrary so would be any


 19   other criteria.


 20             But another concern is that whatever way


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


 22   0.75 to 1.33 is a partial solution because it may


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


 24   intra-individual variation but not those which have


 25   higher variations and 50, 60 percent would still be




  1   the cut off.


  2             [Slide]


  3             Another approach would be to expand the


  4   limits in proportion to the estimated variation.


  5   This has been suggested by Boddy and coworkers.


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


  7   other factor is the estimated standard deviation,


  8   intra-subject variation.  This procedure has the


  9   advantage that the usual testing procedure can be


 10   applied with some proviso.  The statistical power


 11   is independent of the variation and the statistical


 12   power is higher, much higher than the unscaled


 13   average bioequivalence with the usual criterion so


 14   we need fewer subjects.


 15             On the other hand, the criterion is that


 16   bioequivalence limits, as shown there, are really


 17   random variables because they include the estimated


 18   standard deviation, estimated intra-subject


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


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


 21   quite correct, however, it is becoming


 22   approximately correct with large samples.


 23             [Slide]


 24             Scaled average bioequivalence is very


 25   similar to the previous one except that the S from




  1   the bioequivalence limits, here, came over to the


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


  3   similar and we have developed and have recommended


  4   procedures for setting the bioequivalence limits.


  5             Again, the advantages are that the


  6   statistical power is independent of the variation


  7   and with the same sample size is much higher than


  8   the unscaled average bioequivalence.  I am going to


  9   demonstrate this.  There is a sensible


 10   interpretation.  The first interpretation is very


 11   similar to that applied with individual


 12   bioequivalence, namely, the expected change to


 13   switching is being compared with the expected


 14   difference between replicate administrations and


 15   one can make sense of that.


 16             A second interpretation is that the


 17   standardized effect size is being applied which is


 18   a clinical interpretation.  There are procedures to


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


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


 21   or there is a procedure recommended by Hyslop and


 22   her coworkers which is somewhat more involved but


 23   still reasonable I think.


 24             [Slide]


 25             This is a demonstration comparing the




  1   procedures and effectiveness of various approaches.


  2   They include the scaled individual bioequivalence,


  3   scaled average bioequivalence and unscaled average


  4   bioequivalence.  You see the probability of


  5   acceptance.  These are results of simulations.  It


  6   amounts to the probability of acceptance at various


  7   distances between the two means.  The first thing


  8   you can see is that for individual bioequivalence


  9   the range is very wide.  Ranges are much narrower


 10   with scaled average bioequivalence.  So, this wide


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


 12   observation is that scaled average bioequivalence


 13   is, indeed, much more powerful than unscaled


 14   average bioequivalence.  So, we again need fewer


 15   people.


 16             [Slide]


 17             What is the limiting variation for highly


 18   variable drugs?  This is obviously a subject of


 19   regulatory decision, as are the others.  The


 20   procedure could be that we apply unscaled average


 21   bioequivalence if the variation is less than the


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


 23   procedure appropriate for highly variable drugs if


 24   the variation is higher.


 25             Perhaps I should go down here.  This is




  1   the same kind of mixed model that was suggested for


  2   individual bioequivalence but, just as Dr. Benet


  3   suggested, it is not reasonable that a sponsor


  4   should play both ways.  The sponsor should declare


  5   the intention of using one procedure or the other


  6   in the protocol.


  7             I wouldn't necessarily dismiss these other


  8   possibilities.  For example, K. Midha recommends 25


  9   percent.  The outcome of those probabilities that


 10   you have seen on the previous slide depend on how


 11   you set these limiting variations.  Obviously, 30


 12   percent is stricter than 25 percent.  In all cases


 13   you and the agency will ask what is the practically


 14   reasonable criterion that one can live with, the


 15   agency can live with and the industry can live


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


 17   necessarily set everything on the 30 percent; do


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


 19   be.


 20             [Slide]


 21             Now, this method of the secondary


 22   criterion has arisen in connection with the


 23   features of individual bioequivalence.  So, we talk


 24   about two approaches, that of individual


 25   bioequivalence and today we are talking about




  1   highly variable drugs.  There are two very


  2   different concerns.


  3             First of all, we have already seen that


  4   for highly variable drugs the potential variation


  5   is smaller than with individual bioequivalence.  In


  6   the case of individual bioequivalence the


  7   deviations arose because the regulatory criterion


  8   was changed.  A much more liberal regulatory


  9   criterion was introduced whereas in the case of


 10   highly variable drugs it is a natural change of the


 11   variability between the two means.  You know this


 12   very well.  With the usual kind of drug the


 13   variation between the means just fluctuates


 14   slightly.  Most of the differences are probably


 15   between the two means and are within the range of


 16   10 percent.  But with highly variable drugs those


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


 18   constraint of 10-15 percent on this natural


 19   variation means that the natural fluctuation is


 20   altered so the sources of the concern are very


 21   different.  Whereas in the case of individual


 22   bioequivalence you have to deal with the criterion,


 23   here you have to deal with the natural variation.


 24             [Slide]


 25             So, I would like to raise some caution. 




  1   In addition, the imposition of the secondary


  2   criterion has serious consequences.  I present this


  3   from my life earlier when I dealt with individual


  4   bioequivalence because we had the results then; I


  5   don't have many results for average bioequivalence.


  6   But, again, here you have the results for


  7   individual bioequivalence.  This is the probability


  8   curve for the constrained criterion alone and this


  9   is then the application of the combined criterion.


 10             The combined criterion is expected and


 11   does always run below the two separate criteria.


 12   But when the GMR criterion is highly constricting,


 13   as in this case, then the combined criterion is


 14   really a GMR criterion essentially and has nothing


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


 16   criterion.  So, if you were to consider the


 17   secondary criterion, then this slide suggests to do


 18   it with great caution and after serious


 19   consideration.


 20             [Slide]


 21             Here are the questions again which I have


 22   raised for the committee's consideration and for


 23   the agency's consideration.  They certainly suggest


 24   that many of these issues require further


 25   consideration and further investigation. 




  1   Originally I wanted to end with this loose and


  2   compliant mode, however, I looked at the questions


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


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


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


  6   combined with the application of this secondary


  7   criterion.  I would like to call your attention to


  8   the fact that these are two separate questions.


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


 10   mind, the restriction criterion is much more


 11   controversial and requires thorough exploration for


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


 13   would recommend a separation of those questions.


 14             Also, I have a question about reference


 15   scaling.  I would certainly like to be an advocate


 16   for scaling, but whether the scaling ought to be


 17   reference scaling I would like again to be a


 18   subject for study.  Thank you.


 19             DR. KIBBE:  Thank you.  Questions?


 20   Jurgen?


 21             DR. VENITZ:  I have a question about your


 22   first simulation slide where you compare the IBE to


 23   the ABE and scaled ABE.  My question basically is


 24   that you are assuming for the purposes of


 25   simulation that the COVs for both test and




  1   reference are the same, 40 percent.  Is that


  2   correct?


  3             DR. ENDRENYI:  Yes.


  4             DR. VENITZ:  What would happen if you had


  5   differences in COVs between test and reference?  In


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


  7   much less intra-individual variability than the


  8   reference, how would that affect your curves?


  9             DR. ENDRENYI:  It does affect the curves,


 10   but mainly the curve of the individual


 11   bioequivalence.  It affects little the average


 12   bioequivalence curve.


 13             DR. VENITZ:  What about the scaled average


 14   BE?


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


 16   artifact in a way because here we consider the


 17   scaling by reference product so we didn't


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


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


 20             DR. VENITZ:  Right, right.


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


 22   that is an interesting question.  It would be worth


 23   investigating.


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


 25   using then is the reference variation.




  1             DR. ENDRENYI:  That is right.


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


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


  4             DR. ENDRENYI:  No, the estimated


  5   reference.


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


  7   2 X 2 design?


  8             DR. ENDRENYI:  Well, it is a different


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


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


 11   that.


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


 13   you leaning forward?  No?  Go ahead.


 14             DR. SINGPURWALLA:  I just have a technical


 15   comment.  Somewhere in your slides you had a


 16   restricted maximum likelihood.  Right?


 17             DR. ENDRENYI:  Yes, as a possible


 18   procedure.


 19             DR. SINGPURWALLA:  As a possible


 20   procedure?


 21             DR. ENDRENYI:  Yes.


 22             DR. SINGPURWALLA:  Well, this is a


 23   technical comment, the maximum likelihood is


 24   advocated because of its asymptotic properties in


 25   the sense that it converges to the center.  You




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


  2   restrict your maximum there is no assurance that


  3   you converge, the central limit theorem.


  4   Therefore, the value of that process cannot be


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


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


  7   want to caution you.


  8             DR. ENDRENYI:  You are absolutely right,


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


 10   replicate design probably the procedure of


 11   evaluation would have to be defined very clearly


 12   and very strictly, otherwise one can go in all


 13   different directions and that will be another task


 14   if the agency goes that way.


 15             DR. KIBBE:  Go ahead.


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


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


 18   the agency went back and looked at the content


 19   uniformity criteria and published two sets of data.


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


 21   at the bioequivalence data and look and see how


 22   often it falls within certain criteria.  You have a


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


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


 25   the committee on the secondary criteria.




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


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


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


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


  5   Marv?


  6             DR. MEYER:  This is probably a


  7   statistically ignorant question but under the


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


  9   it possible to have a product with a scale


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


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


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


 13   without the point estimate constraint you have a


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


 15   to be approved.


 16             DR. ENDRENYI:  No--


 17             DR. MEYER:  Two different studies?


 18             DR. ENDRENYI:  In two different


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


 20   wouldn't envision between study variation and I


 21   don't--I doubt it very much.


 22             DR. MEYER:  Even if the test product only


 23   released 70 percent of its dose and the innovator


 24   released 100 percent of its dose the true ratio


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




  1   be looking for 1.0.  It is not possible?


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


  3   inter-study variation, then with the low variation


  4   drugs each of them would be between 0.8 and 1.25


  5   but the two in comparison with each other could be


  6   quite different.  That is equally possible but it


  7   is not likely.  If it is the same reference


  8   product, then it is not possible.


  9             DR. KIBBE:  I see no other questions.


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


 11   We will be back at 10:52.


 12             [Brief recess]


 13             DR. KIBBE:  We have open public hearing at


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


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


 16   modify our schedule to try to get everything done


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


 18   interest in what we are talking about so we might


 19   allow our speakers a little extra time and some


 20   questions and answers to go a little further.  I


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


 22   Barbara Davit.


 23                Bioequivalence of Highly Variable


 24                        Drugs Case Studies


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




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


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


  3   has been submitted to the Division of


  4   Bioequivalence.


  5             [Slide]


  6             When Dale and I were talking about putting


  7   this presentation together for the advisory


  8   committee, one of the things we thought we would


  9   consider is looking at what has been submitted to


 10   the Division of Bioequivalence and to answer the


 11   question of whether highly variability is a


 12   significant issue in these bioequivalence studies


 13   in ANDA submissions.


 14             By looking at these data and focusing on


 15   some case studies, we thought also we could maybe


 16   answer the questions in a limited number of cases


 17   of what is contributing to the variability or what


 18   are some of the sources of this variability.


 19             [Slide]


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


 21   there is a significant problem with highly variable


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


 23   that this obviously represents a biased sample


 24   because we receive predominantly studies that have


 25   passed the 90 percent confidence interval criteria.




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


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


  3   what percentage that is of the total number of


  4   drugs in a company's pipeline for example.


  5             But of the submissions we saw, which are


  6   passing studies, what percentage were for highly


  7   variable drugs?  Did these studies involve


  8   enrolling a large number of subjects because that


  9   has been one of the issues that has been raised


 10   today, the large number of subjects that might be


 11   necessary to show bioequivalence for these generic


 12   products of highly variable drugs?  Also, how


 13   narrow and wide are these 90 percent confidence


 14   intervals?  That goes along with how many subjects


 15   are necessary for a passing study.


 16             [Slide]


 17             We collected data from all the


 18   bioequivalence studies that were submitted to the


 19   Division of Bioequivalence in 2003.  We used the


 20   root mean square error as an estimate of


 21   intra-subject variability.  I realize this is just


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


 23   the intra-subject variability but, unfortunately,


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


 25   two-way crossover studies so the best estimate that




  1   we could get of the intra-subject variability was


  2   the root mean square error.


  3             We defined a highly variable drug as one


  4   with a root mean square error which is greater than


  5   0.3, representing 30 percent intra-subject


  6   variability.  The data that I am going to present


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


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


  9   to be presenting passed our 90 percent confidence


 10   interval criteria, but that is because for the most


 11   part we don't receive submissions of studies where


 12   the product did not pass bioequivalence criteria.


 13             [Slide]


 14             First of all from 2003, this was a total


 15   of 212 in vivo bioequivalence studies.  Of these


 16   212, looking at only those studies in which the


 17   root mean square error of AUC or Cmax was greater


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


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


 20   about 15 percent of our studies the drug would


 21   qualify as having highly variable characteristics.


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


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


 24   total only Cmax was highly variable.  There were no


 25   studies in which only AUC was highly variable.  But




  1   there were 5 studies in which both AUC and Cmax


  2   were highly variable, and this was 2.5 percent of


  3   the total.


  4             [Slide]


  5             This goes along with the previous slide


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


  7   saw a root mean square error of a particular value


  8   for Cmax.  There is an error in this particular


  9   slide in your handout but this is the correct


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


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


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


 13   earlier that 15 percent of all the studies, 15.5


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


 15   mean square error for Cmax of greater than 0.3.


 16             [Slide]


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


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


 19   part the root mean square errors were hovering


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


 21   in AUC than Cmax.


 22             [Slide]


 23             One of the questions that we wanted to ask


 24   was what is contributing to this variability.


 25   Since for a lot of products we look at




  1   bioequivalence studies in fasted subjects as well


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


  3   having on variability.  I mentioned 33 studies.


  4   This represented a total of 24 of the ANDAs that


  5   were submitted and reviewed in 2003.  Of these,


  6   both AUC or Cmax were highly variable in both the


  7   fed and fasted studies.  In 8 of these the


  8   pharmacokinetic parameters were highly variable in


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


 10   were highly variable in only the fasted study.  But


 11   this is a little bit skewed too because we have


 12   submissions, for whatever reason, which contain


 13   only a fed study and submissions that contain only


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


 15             [Slide]


 16             This shows some of our data.  I think


 17   these are all the Cmax values from the 212 studies


 18   I was talking about in which Cmax was variable in


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


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


 21   variability because of food effects.  I am not


 22   giving the names of the drugs but I have


 23   illustrated them by class.


 24             There is a variety of reasons I think for


 25   the variability.  Some of these are prodrugs.  We




  1   have a number of angiotensin converting enzyme


  2   inhibitors and most of these are prodrugs.


  3   Generally the parent is present at low


  4   concentrations so this could contribute to the


  5   variability.  A number of these drugs also are


  6   highly metabolized and this would contribute to the


  7   variability.  But, in this case, obviously there


  8   was a food effect.  The variability was observed in


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


 10   studies too the number of subjects ranged from


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


 12   terms of numbers of subjects.


 13             [Slide]


 14             It is pretty unusual to only see a highly


 15   variable Cmax in the fasting study and not the fed


 16   study, and this occurred in only two cases last


 17   year.  These were both angiotensin converting


 18   enzyme inhibitors, both prodrugs.  For one of them


 19   the bioequivalence was based on measuring the


 20   parent.  For the other one the company could not


 21   measure the parent despite I guess a number of


 22   attempts.  This is actually true for pretty much


 23   everyone who has worked with this particular drug.


 24   So, the bioequivalence here is only based on the


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




  1   majority of submissions that we have the


  2   bioequivalence is based on the parent.


  3             [Slide]


  4             This table shows the Cmax data where Cmax


  5   was highly variably in both fed and fasted studies.


  6   So, for this drug product obviously there will be


  7   highly variable regardless of whether it is the fed


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


  9   products, various drug classes, various reasons for


 10   variability; some prodrugs, some highly metabolized


 11   drugs; some drugs that undergo extensive first-pass


 12   metabolism.  The number of subjects varied from I


 13   guess 18 to 57.


 14             [Slide]


 15             Finally, this table is for two-way


 16   crossover studies and shows the data for which both


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


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


 19   in yellow, for this particular product both AUC and


 20   Cmax met the highly variable criteria in both the


 21   fed and the fasting state.  For the other drugs


 22   there was high variability in the fed but not


 23   necessarily the fasted, or Cmax and not necessarily


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


 25   class.  The number of subjects that the companies




  1   used varied from 26 to 62.


  2             [Slide]


  3             In trying to explore some of the sources


  4   for this variability, we wanted to compare the


  5   intra-subject variability for the test versus the


  6   reference product.  We don't see very many


  7   replicate design studies anymore.  In this


  8   particular class of drugs we only had two


  9   submissions last year so these are the data from


 10   the two submissions.


 11              These data are a good sign because what


 12   they show is that the variability, based on the


 13   root mean square error, was comparable for the test


 14   and the reference product for both of these drug


 15   products.  That is obviously what we are looking


 16   for because we want to see people achieve a generic


 17   product that is the same as the reference product.


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


 19   comparable, test versus reference.


 20             One study used 33 subjects.  The other,


 21   this would obviously fall into a category where it


 22   necessitated a lot of subjects because this was not


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


 24   it meant that each of these 77 subjects received


 25   the drug product four times, on four occasions. 




  1   So, this was quite an extensive study.


  2             [Slide]


  3             Another question we wanted to ask was are


  4   there ever cases in which the pharmacokinetic


  5   variability is a function of the drug product as


  6   opposed to the drug substance.  We found two


  7   instances last year, two different drug products


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


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


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


 11   an extended release tablet.  Drug D was an


 12   immediate release tablet.


 13             We will look at drug C first.  In one


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


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


 16   this product would not qualify as a highly variable


 17   drug.  Notice root mean square errors of 0.18,


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


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


 20   different formulation, another company, 0.31, 0.38,


 21   0.25, 0.34.


 22             This could be due to a number of reasons.


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


 24   release dosage forms are more complex than the


 25   immediate release dosage forms and the two




  1   formulations were quite different.  So, there could


  2   have been, you know, differences in variability due


  3   to the formulation.  Also, the bioequivalence


  4   studies were done at different sites.  I looked at


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


  6   get the specifics of the extraction methods but I


  7   noticed that the two studies had different limits


  8   of quantitation and there were different doses in


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


 10   factor this was.  This was an extended release


 11   product for which I believe there were three


 12   different strengths.  One company submitted a study


 13   on the highest strength and I think used two times


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


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


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


 17   all these factors that could be contributing to the


 18   variability.  At least, those are the factors I


 19   could think of.


 20             Drug D--this was an interesting issue.


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


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


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


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


 25   product and I noticed that the formulations of




  1   these two were qualitatively identical;


  2   quantitatively there were some differences.


  3             These were done at two different sites and


  4   in this particular application the bioanaytical


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


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


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


  8   have been a contributing factor here.


  9             I would like to stress that of all the


 10   applications that we saw last year, these were the


 11   only four in which we saw that there was a


 12   difference which was possibly due to drug


 13   formulation or possibly due to where the studies


 14   were done that was contributing to the high


 15   variability.


 16             [Slide]


 17             Then we thought we would look at how many


 18   study subjects are usually enrolled in these


 19   studies.  Once again, I emphasize that this is


 20   really a biased sample because we only see the


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


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


 23   studies were done where the company just couldn't


 24   get the study to pass the confidence interval


 25   criteria so these are just the passed studies.