1

DEPARTMENT OF HEALTH AND HUMAN SERVICES

FOOD AND DRUG ADMINISTRATION

CENTER FOR DRUG EVALUATION AND RESEARCH

 

 

 

 

 

 

PROCESS AND VALIDATION WORKING GROUP

 

 

 

 

 

 

 

 

 

 

 

Tuesday, February 26, 2002

8:30 a.m.

 

 

 

 

 

Holiday Inn Gaithersburg

Two Montgomery Village Avenue

Gaithersburg, Maryland

2

PARTICIPANTS

Thomas Layloff, Ph.D., Acting Chairperson

Kathleen Reedy, Executive Secretary

MEMBERS

Gloria L. Anderson, Ph.D.

Joseph Bloom, Ph.D.

Judy P. Boehlert, Ph.D.

Arthur H. Kibbe, Ph.D.

SGE CONSULTANT

Melvin V. Koch, Ph.D.

GOVERNMENT PARTICIPANT

William F. Koch, Ph.D.

OTHER GUESTS/SPEAKERS PARTICIPANTS

Thomas J. Hale

Leon Lachman, Ph.D.

Kenneth R. Morris, Ph.D.

G.K. Raju, Ph.D.

Eva M. Sevick-Muraca, Ph.D.

INDUSTRY GUESTS/PARTICIPANTS

Robert S. Chisholm

Rick E. Cooley

Doug Dean, Ph.D.

Steve Hammond

John C. James, Ph.D.

Ronald W. Miller, Ph.D.

David Richard Rudd, Ph.D.

John G. Shabushnig, Ph.D.

Leon Shargel, Ph.D., R.Ph.

Efraim Shek, Ph.D.

Jozef H.M.T. Timmermans, Ph.D.

Judy Wong, M.S.

Jerome (Jerry) Workman, Jr.

FDA

Yuan-yuan Chiu, Ph.D. (Sessions I, II, IV)

Douglas I. Ellsworth (Sessions I, III)

Joseph Famulare (Sessions II, III)

Ajaz S. Hussain, Ph.D. (Sessions I, II,IV)

Moheb M. Nasr, Ph.D. (Session III)

Michael C. Olson (Session IV)

3

PARTICIPANTS (continued)

Process and Analytical Validation Working Group

Leon Lachman, Acting Chair

Thomas Hale

Jozef Timmermans

Robert Chisholm

Kennedy Chibwe

Carl Anderson

John James

Sonja Sekulic

FDA:

Doug Ellsworth

Moheb Nasr

David Morley

Lucinda Buhse

Michael Olson

Joseph Famulare

4

C O N T E N T S

Page No.

Process and Analytical Validation Working

Group Discussion,

Leon Lachman, Ph.D. Acting Working Group Chair 5

5

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

2 DR. LACHMAN: Yesterday we heard various

3 approaches to be used as part of PAT programs to

4 monitor product processes and to verify that we

5 have a uniform process on a continuing monitoring

6 basis. We heard various approaches used, including

7 the NIR-IR spectroscopy. We got acoustic emission,

8 deflectance measurements, fiberoptics, amperage.

9 Then, of course, we have procedures that we have

10 been using before, temperature, vacuum, pressures.

11 Also, we had discussion of chemometrics which is an

12 application of statistics and math to the chemical

13 parameters. One area that we didn't discuss too

14 much yesterday was feedback controls. Since this

15 is going to be a continuing process you also want

16 to be able to control your process by these

17 measurements. When you get into feedback controls

18 you talk a lot about automation and computers, and

19 that is a whole new ball game and there is quite a

20 bit of work that has to be done as part of this

21 overall process.

22 Then, we still are early in this

23 development program. We have heard of one example

24 where Glaxo had this process for a drug that was

25 pretty toxic and they developed an automated

6

1 process for it. Was it Zeneca? AstraZeneca? Then

2 SmithKline Beecham talked about various approaches

3 as well for process controls that use inference

4 measurements.

5 Now the question is we have two pieces

6 here. We have the new products that are under

7 development and existing products for which

8 applications have been submitted. I think the

9 guidance needs to cover both of them. I don't

10 think we can separate one from the other readily.

11 What I would like to do is open this

12 meeting for suggestions as to how to come up with a

13 guidance that will cover the major activities

14 related to PAT, but not be specific because we are

15 not ready to put together a guidance of how to but

16 just what is expected. If anybody would like to

17 start with this -- we have a set of questions here

18 that Ajaz has put together for us. The first

19 question is what approaches should be considered

20 for validating the use of new PATs for in-process

21 control of materials during pharmaceutical

22 manufacturing?

23 I think we talked about processes that are

24 being utilized currently. I am sure there are

25 others being developed. So I don't think we want

7

1 to limit or identify specific inference

2 measurements for process controls. Really, I

3 gather Ajaz asked that we consider developing an

4 outline for the various considerations on the

5 process for PAT activities that would be brought

6 forth by the other subcommittees.

7 The first question is what approaches

8 should be considered for validating the use of new

9 PATS for in-process control? Doug, did you have

10 your hand up?

11 DR. ELLSWORTH: I am just thinking.

12 DR. LACHMAN: Oh, okay. Anybody want to

13 start that?

14 DR. ANDERSON: I would assume that

15 anything we do is going to be based in current

16 regulations, and what we are going to discuss here

17 is what sort of guidances can we put forward that

18 will make it possible for industrial applications

19 to use PATs and still meet -- perhaps the best way

20 of saying this is come up with a way to interpret

21 existing regulations and add what we need to those

22 regulations to allow the use of PATs within that

23 existing framework. Is that a correct perspective?

24 DR. LACHMAN: I think it would have to

25 currently meet the intent of the GMPs. Right now

8

1 that is still broad because GMPs are not specific

2 for APIs but they are used for APIs as an overall

3 guide. I think this is what we should use here,

4 use the current GMPs as a guide in developing the

5 validation outline.

6 DR. ANDERSON: Right. So, by way of that,

7 part of the answer to this question is that a lot

8 of the framework exists. I guess what I personally

9 would like to see out of this discussion is what

10 are the differences from the PATs and the existing

11 analytical methodology, and what do we want to say

12 about how these PATs need to be implemented with

13 respect to what we do currently for analytical

14 methodology. What exceptions do we need; what

15 extra cautions need to be taken. Those are the

16 types of things that I would like to see discussed.

17 That is where I see us as needing to go.

18 DR. LACHMAN: I think the past practice on

19 analytical methodology was to use methods that are

20 suitable for validating a process. So you had to

21 do an assay. Okay? The assay could be UV, could

22 be HPLC, could be IR, whatever was necessary to

23 show reproducibility of the process. Also in the

24 past, as a segment of the validation, I think we

25 were also considering whether the product is going

9

1 to be stable for whatever period of time the expiry

2 date was intended.

3 I think right now we should leave that

4 aside and just look at the process controls. In

5 the past, you know, we used a lot of normal

6 chemistry analytical methods and now we are

7 changing that approach. We are not using HPLC, GC

8 or UV. We are using various inference methods,

9 NIR; we are using amperage -- well, we have been

10 using amperage measurements for a while to

11 determine when the wet granulation is consistently

12 wet; a lot of these reflectance measurements and

13 the acoustic emission. So, we are using new

14 technology to follow the repeatability of the

15 process by measuring active ingredient distribution

16 or particle size. We have to have some correlation

17 so we could say that the method we are using now on

18 an on-line basis is going to provide us the

19 controls equivalent to or better than we got by

20 off-line measurements.

21 DR. NASR: I would like to make a few

22 comments. My name is Moheb Nasr and I work at the

23 Center for Drug Evaluation and Research. I would

24 like to start this discussion by at least sharing

25 with you some of my ideas of where we are and

10

1 possibly how to get to where we need to be.

2 The first comment is on the process of

3 guidance and development itself. I think from our

4 experience in the past, the agency usually has a

5 draft guidance based on experience with different

6 applications. Once we have a set of ideas, we

7 usually publish this draft and receive comments

8 before we finalize the guidance and use it, again,

9 as a guidance. It is not yet a regulation. But

10 the industry and the reviewers pretty much use it

11 as a set of rules to follow in order to approve or

12 disapprove a drug application.

13 This process here is different. It is

14 different for two reasons. Reason number one, we

15 don't have that experience. So, the agency doesn't

16 have the experience that we usually have to base

17 our guidance and development. So we are here

18 seeking your help by sharing your experience with

19 us in order for the guidance to be useful, and

20 encouraging you to submit applications in order for

21 us to work with you to bring fruition to this

22 initiative.

23 So we don't have a draft in mind. We

24 would like to leave, hopefully, today with some set

25 of ideas that we can use as a basis in drafting the

11

1 guidance. So, that is the distinction I would like

2 to make between the process we are going through

3 today and what we have done in the past. This is

4 my first comment.

5 The second comment is that I think we have

6 to be aware of the challenges before us. Some of

7 the challenges, and you heard many of these before,

8 are very much what I indicated yesterday in the

9 afternoon in my comment. That is, we would like

10 this guidance to be general enough, but not just an

11 invitation letter to invite you to submit

12 applications.

13 But, for a variety of reasons, we cannot

14 make it very specific to address every possible

15 scenario and situation that you may deal with or

16 face because we are not aware of these individual

17 cases and, also, we don't want to make it too

18 restrictive or for one technology. For example, we

19 may have more information about NIR-IR. We have

20 decided not to make this an NIR-IR guidance. So,

21 we want to make it a technology-based guidance to

22 address a variety of technologies and a variety of

23 applications. We don't want to make it product

24 specific and we don't want to make it test specific

25 or process specific. That is the challenge before

12

1 us.

2 After I share with you my third comment, I

3 think I would like, if possible, to focus on

4 finding critical process and analytical validation

5 issues that we think should be included in such a

6 guidance. Maybe we can go around the room and say,

7 you know, what you think are the critical issues

8 that need to be included in order for this guidance

9 to be useful and helpful, utilized and used.

10 My third comment is that in addition to

11 what we have in this guidance, how can we, as a

12 group, utilize the experience that you all have in

13 making this guidance, especially the validation

14 issues, different from prior guidances? What do I

15 mean by that?

16 I think if we extend validation criteria

17 that we have used for years when we were doing

18 titration and then moved on to HPLC, GC and so

19 forth, and extend that to some of the new

20 technologies, I don't think it is going to work.

21 What do I mean by that? I mean if we have the idea

22 of replicate injections and all these things, are

23 they necessary for the tests and methods used

24 before? Of course. Do we have to extend that to

25 some of these current and new technologies? I

13

1 think we need to evaluate that.

2 One of the comments made yesterday by Rick

3 Cooley, with whom I have had several discussions

4 before, I think is very useful for us. What Rick

5 said was when we start drafting this guidance,

6 especially the validation issues, it is a good idea

7 to start with a blank sheet of paper and see what

8 are the critical quality attributes that we need in

9 this particular measurement, and how can we

10 identify the tests and technology used for that

11 particular method or test, rather than carry over

12 many of the old validation criteria that Leon

13 outlined very capably yesterday, and just change

14 from an HPLC to NIR-IR, and so forth.

15 The point I am making in my third comment

16 is this, I think we have to think differently,

17 think out of the box, totally out of the box about

18 some of these critical validation criteria that we

19 need to look at when we examine these new

20 technologies rather than just copying or using old

21 validation criteria that we have used in the past.

22 I think if we do that, I don't think we will move

23 forward very much.

24 DR. LACHMAN: Yes?

25 DR. OLSON: I am Mike Olson. I am with

14

1 the Office of Regulatory Affairs. I am responsible

2 for the laboratories and am kind of on the

3 chemistry side. When I look at this, maybe I am

4 very confused but I really see a couple of

5 different issues: validating the process, which I

6 think we have a lot of experience in, and then

7 there is the term "analytical validation" which we

8 tend to think of as more methodology validating a

9 method. I am not sure the analytical method

10 validation would be any different right now. I

11 don't really see that as a tremendous issue.

12 The issue I see is really on the process

13 side. You know, you are using an analytical

14 technique, some kind of spectroscopic technique,

15 but you are in a dynamic process and I think there

16 is a lot of validation that has to go on to say

17 that you are running an NIR-IR in a spectrum in a

18 lab versus inserting it and running it in a

19 process. You are going to have to deal with

20 different types of issues. I would see this as

21 leaning more towards the process validation area.

22 Finally, naturally once you start to

23 figure out what you want to measure or look at, you

24 have to make that leap, that correlation that

25 running something in a dynamic process also equals

15

1 some type of analytical value. For example, if

2 your powder is blended properly, I don't think it

3 is a big leap of faith to say you should have good

4 contact uniformity. That is the easy part.

5 I think the first thing you might want to

6 look at is more towards the process side because I

7 think that is where the bigger challenges are going

8 to be. So, once you figure out your process side,

9 then you look at what it is you are measuring, then

10 you want to see how it correlates with analytical

11 testing.

12 I am not sure, I may be misusing the term

13 "analytical validation" but I don't really see this

14 as a method validation issue. I think we are going

15 to continue to do method validation as we have

16 always done in the laboratory. Maybe the issue is

17 how we are going to analytically validate this

18 prior laboratory technique which is now in a

19 manufacturing setting. Maybe that is what this is

20 dealing with, but I don't really see the method

21 validation as being something that would require a

22 lot of thought. I see it more on the other side,

23 the process side. DR. HALE: Tom Hale. I

24 don't disagree with that, but I think there are

25 several levels of implementation that we have heard

16

1 yesterday, and different interpretations of what

2 this means. We currently have sensors on our

3 processes but we could add more sensors and run the

4 process exactly the same way, except measure

5 something else, which is one step along the way.

6 The next step would be to run particular

7 unit operations in a more continuous process as

8 opposed to a batch process, which gets into a

9 different definition of validation because you are

10 continuously measuring and collecting more data.

11 I think the third step along the way is

12 when you are actually measuring the product itself,

13 and there are hints at measuring content uniformity

14 and you can get to the point where, in the extended

15 version of it, you could measure every dosage form

16 perhaps if and when that technology is available.

17 At that point I think product and process

18 validation completely changes because you don't

19 really validate the product any more, you do every

20 batch. So, there are stages of the process and

21 product validation that we need to divide up and

22 clarify. Just adding more complicated sensors

23 doesn't mean anything until you address the

24 different process levels.

25 MR. CHISHOLM: I think I am getting a bit

17

1 puzzled here because I am not a spectroscopist and

2 I am not a chemometrician; I am an engineer. I get

3 a bit puzzled when people talk about these

4 techniques and forget these techniques have to

5 relate to a migration path to some existing

6 knowledge base because the techniques on their own

7 mean absolutely nothing unless you can relate it to

8 something you know about. If you are using things

9 like infrared -- whatever technique it is, I don't

10 mind -- let's say you are looking for content in a

11 tablet, the only way you can actually model during

12 the modeling stage is to know the content of that

13 tablet from some frame of reference that you

14 already have, i.e., probably HPLC.

15 So I don't think we should get carried

16 away on a sort of ocean wave of these technologies

17 can exist on their own. Inferential technologies,

18 by definition, cannot exist on their own. Once you

19 have actually modeled, they can be used without any

20 further analytical results once you have validated

21 your model and I think we must keep that very

22 firmly in mind. I am just a bit worried that we

23 are taking one step too far. You have to relate it

24 to your existing databases.

25 This should keep the FDA, the MCA, etc.

18

1 very happy. It is comfortable for you because it

2 means, in fact, that we have to relate these new

3 secondary methods to our existent methods to

4 actually build the models in the first place so we

5 relate it to a known database. I think we need to

6 keep that in mind in all our discussions and not

7 get too carried away. I think that is what you

8 were actually referring to when you were talking

9 about existing analytical methodology. It is still

10 there, and it is going to be there as long as I

11 know unless there is something I don't know.

12 I would like to throw that out as an open

13 question, can we use, for instance, these systems

14 without going back to a known database? I don't

15 think so.

16 MS. SEKULIC: I am going to counter that.

17 I think there is a lot of methodology out there

18 that does need to be died to an existing knowledge

19 database, analytical primary or reference technique

20 test. That is one way of utilizing these

21 techniques.

22 There are others, one other that I am

23 going to mention, and that is model-free or

24 stand-alone techniques. It doesn't matter what

25 kind of sensor it is. For example, we saw reaction

19

1 monitoring, end-point determination and blending

2 end-point determination yesterday. These are

3 examples where you can go in and have stand-alone

4 methodologies that do not require calibration if,

5 for example, you can identify the same starting

6 point each time you run the process and use that as

7 your baseline marker, let's call it, and then all

8 you are doing is calculating standard deviation or

9 some parameter, some criterion which enables you to

10 determine when the change that you are expecting to

11 see has been accomplished. You can, actually, not

12 have to tie it back to a calibration data set or a

13 primary analytical technique.

14 In my mind, these are the two moods of

15 implementing these types of tools, whether it be

16 NIR, whether it be any other type of signal coming

17 from a sensor. I think we need to keep that in

18 mind. How you use the two approaches will enable

19 you to do different things in understanding the

20 processes. I think generally, as analytical

21 chemists, we come to the table and we think I am

22 going to replace a lab method with an on-line

23 technique, and that is a perfectly legitimate way

24 of using it. That is one option we have. But if

25 we want to have a sensor like a pair of eyes inside

20

1 all our process operations and all our vessels,

2 then we have to treat it a little bit differently

3 and provide the leeway of actually doing that.

4 I may not be using it as a release test

5 result, but do we really want to stop people from

6 understanding their processes? We have to provide

7 the mechanism to enable people to do that, and I

8 think these are the two options that we have to

9 consider in the application of these types of

10 technologies. If you look into the future, we

11 don't know what kind of sensors are going to come

12 out of academia and vendor sources. We have to be

13 able to provide the flexibility and not have to sit

14 down as a committee every year to deal with the new

15 technique that is coming out on the market. I

16 think that will enable us to do that.

17 MR. FAMULARE: I think a lot of what these

18 sensors will enable industry to do is to control

19 that process. I think we have to go back one step

20 from process validation. I think very often that

21 is the buzz-word that we jump to when there is a

22 new way of analyzing or sensing what is going on

23 and we want to validate it right away. But I think

24 one of the purposes of these sensors has been all

25 along, even though they haven't been used as

21

1 official release tests or even in-line tests that

2 have any recognition in regulatory filings, is that

3 what they have been doing for industry is

4 controlling the process, and what they will do in

5 the future, as opposed to using an off-line test

6 and saying, well, I guess I have blended long

7 enough now because I got this assay, you know, two

8 weeks later from the laboratory.

9 So I think before we define what the

10 elements are of process validation we should

11 probably put some consideration into what process

12 control means in this guidance. Then, if we want

13 to go the next step and think of this as now the

14 possibility of using these inference or indirect

15 methods as a "test" that could count for in-process

16 testing or could even possibly count for end

17 product releasing, either a parametric type release

18 or, as someone said yesterday, parametric doesn't

19 even capture it because you are measuring the

20 product itself whether it be an in-line point or a

21 point where it actually represents the finished

22 product.

23 DR. LACHMAN: This goes back, you still

24 have to determine the natural variability of the

25 process. How you do this, you identify the

22

1 individual process steps, the important process

2 steps; put a flow chart together; determine what

3 the quality characteristics of the product are for

4 these steps; and you have to identify them and put

5 specifications on these critical steps. Then, you

6 have to have a method that will determine that the

7 product falls within these ranges for these

8 critical steps.

9 I think you still have to do the basic

10 development work. I think that is what we talked

11 about yesterday. You have to do the basic

12 development work and then you can take an inference

13 measure to then follow the process control. But

14 you still have to do that minimal work to get that

15 baseline of information to control your process.

16 Then you can assign an inference measurement that

17 will then determine the uniformity of that process.

18 But I still think you have to do that baseline

19 work. I don't see how you can get away from it.

20 MR. FAMULARE: As we go through this

21 question sheet, I think what industry will be

22 looking for in this guidance is what hurdles do I

23 have to overcome for a new process development in a

24 regulatory filing, and am I going to have to show

25 correlation with existing analytical technologies

23

1 in a filing for a new developed product? Or, if I

2 want to convert an existing product or transform an

3 existing product to this technology, again, will I

4 have to show correlation data or can it start with

5 a blank sheet of paper? I think those are some of

6 the answers that will be looked for in this

7 guidance as opposed to one technology over another.

8 DR. LACHMAN: For an existing product you

9 should have a lot of historical data with regards

10 to the process capabilities. It is using that

11 historical data and selecting an inference

12 measurement that will indicate that the process is

13 under control, using all that background data to

14 show that, in fact, it is doing that by the

15 in-process controls and the information you have

16 indicates what is my expected product variability

17 from batch to batch. Of course, you have all that

18 data and you have to treat it statistically, and

19 then you have something for your inference

20 measurement to be used and still fall within that

21 range. You are still going to have to have some

22 correlation or show some relation to the process

23 capability. I think that is what we are trying to

24 do, show process capability on a continuing basis,

25 and that is what the inference measurement is.

24

1 MR. FAMULARE: You probably would have for

2 an existing process, as you say, all of this data

3 but then, depending upon the instruments, some of

4 this actually has to be fed to the instrument

5 because the instrument will look at this data to

6 determine if this particular, for example, tablet

7 passing through is truly ibuprofen. But, as part

8 of that, you have to have data from passing and

9 failing batches.

10 DR. LACHMAN: Well, I don't know if you

11 need ibuprofen. I don't think it needs to indicate

12 ibuprofen; it has to indicate that active

13 ingredient. That method could be used for various

14 products for in-process controls on-line.

15 MR. FAMULARE: Because we had a lot of

16 discussion yesterday on, you know, what would firms

17 have to go to in terms of manufacturing, passing

18 and failing batches --

19 DR. LACHMAN: Right.

20 MR. FAMULARE: -- but more for development

21 type work. So, again, I think these are practical

22 questions that companies are going to have in terms

23 of their regulatory filings and what would be

24 expected as documentation for validation in a GMP

25 inspection.

25

1 DR. ANDERSON: In terms of looking at the

2 overall framework I see two major classes falling

3 out, and this is what Sonja was kind of pointing

4 to. We have some methods that are going to relate

5 back to existing primary analytical methods. We

6 also have places where we want to use PAT sensors

7 but there is no primary analytical method. So that

8 is the first distinction.

9 The second distinction we are looking at

10 is the case where we want to use this for release

11 testing. I want to release my product based on

12 this in-process test. A second scenario, again, is

13 I want process information. I am not releasing

14 anything on this but I want my process in better

15 control.

16 I wonder if we shouldn't focus on what are

17 the validation guidelines necessary in each of

18 those scenarios. That is what we need. We need to

19 say within this type of measurement, not a release

20 test and with existing technology, what do we need

21 to do to validate that method for each of those

22 permutations? That is what I see as needing to

23 come up with, those guidances.

24 DR. NASR: I would like to add another

25 criterion to consider as well, and that is if this

26

1 technology will be used for an existing product

2 where we have all the information about the history

3 and where we can go back to databases, or for a new

4 product, because we has several discussions

5 yesterday and in the past as well and there are

6 some people who believe that utilization of this

7 technology may be easier to implement for new

8 products than for existing products where you have

9 to re-tool the factory and you have to add all this

10 new stuff.

11 So I think we have to possibly look at

12 different classes or different areas of validation

13 that we look into. One would be for new products;

14 one for existing products; one where we have

15 existing analytical methods; and also another area,

16 as you outlined fairly well, whether these methods

17 that we use for process monitoring and/or control

18 can be used to replace some of the release tests

19 that have been used at the end of the process as

20 well.

21 DR. HALE: I think the other category that

22 would bound this guidance would be continuous

23 in-batch processes because the way you would treat

24 a V-blender or a continuous blender would

25 presumably be different in how you would define

27

1 validation and depending on the analytical method

2 that was used.

3 DR. LACHMAN: Well, shouldn't the sampling

4 be the same, sampling for continuous or batch if

5 you are using an inference method?

6 DR. HALE: It could be inference or

7 non-inference. If you had a sensor that could

8 measure concentrations of a particular product in a

9 flowing powder stream, then you wouldn't be

10 inferring anything. You would be ding a primary

11 measurement and you wouldn't have a static powder

12 bed to sample or blend. You could have a

13 continuous stream that could flow through and you

14 could extend that to multiple unit operations that

15 are interactive. The current approach to

16 validation wouldn't necessarily fit into those

17 types of processes.

18 To me, to be able to do continuous

19 processes with this technology opens up all kinds

20 of opportunities, and the guidance would be very

21 powerful in terms of implementation if it opened up

22 that opportunity to explore getting rid of these

23 distinct unit operations and putting a continuous

24 flow together.

25 MR. OLSON: May I make a comment?

28

1 DR. LACHMAN: Yes, go ahead.

2 MR. OLSON: Something I said before is

3 that you map out the process and determine the

4 critical control points, but if I started a drug

5 tomorrow I would probably sense something that

6 would say I don't want to do that. Basically, what

7 is the task of the GMPs? It is to ensure that the

8 right drug and the right amount is in every tablet.

9 So could you actually see a process where, let's

10 face it, that is your final control. If you are in

11 the final mixing step and you have determined it is

12 homogeneous, can you basically just say that is my

13 one control point? Anything else before that you

14 can call noise in the system. I mean, you can do

15 whatever you want to do but you have to get to that

16 point where you have a good mix.

17 How you define that would be kind of your

18 own business, but I could see somebody really

19 having no critical control points and just saying

20 that is my measurement. It is measuring

21 concentration so, basically, I am saying I have the

22 right drug in the mix and the same amount in every

23 unit dose, which would be quite a departure. To

24 pose it as a question, would this allow that?

25 DR. LACHMAN: I think the way you measure

29

1 the actual concentration of the active should allow

2 that. It is when you are doing an inference, I

3 don't know if it would allow that, inference

4 testing. Yes?

5 MS. SEKULIC: Could I just go back to

6 Moheb's comment regarding the pre-filing,

7 post-filing? I think conceptually the validation

8 of a method wouldn't change. I think in reality

9 what we are dealing with is if, for example, you

10 were to apply some of these process tools,

11 techniques pre-filing you would see everything that

12 there is to see. I believe that the industry-wide

13 fear in going to existing products is that these

14 sensors have the potential to show us things that

15 we were unable to see previously.

16 I think if we really want to provide a

17 guidance that dispels that fear, we have to make

18 some comment regarding either a grandfathering

19 process, you know, some kind of ability to go back

20 to our existing products that have been proven to

21 be safe in the clinics, etc., etc. for X number of

22 years and enable the review with new techniques and

23 capabilities. I think the process itself of

24 generating a validated method doesn't change. You

25 still have to do your method development, i.e.,

30

1 putting the sensor in your vessel to try and get a

2 feel for what that is telling you; getting to the

3 point where you determine whether it is an approach

4 technique for what you are trying to do; then, if

5 necessary, correlating to some other database if

6 that needs to be a quantitative method. It may

7 not; it could stay qualitative. Then, put the

8 final sort of validation package together and then

9 over time you would prove, by collecting data,

10 that, indeed, you may be able to get to that point

11 where, with your experience of manufacturing this

12 particular product, you, indeed can show, yes, if

13 my blend uniformity data with this particular tool

14 or sensor shows me that it is homogeneous, indeed,

15 I can correlate that all the tablets that I have

16 ever made in the past are okay. You could actually

17 get to that point.

18 Now, is that what everyone is going to aim

19 for first time out the door? I don't think so. I

20 think we are a little more hesitant than that and

21 we have to build the confidence within the industry

22 ourselves because there are a lot of different

23 experiences within our industry as well, some

24 positive, some not so positive. But that is the

25 learning phase. I think if we start making a

31

1 distinction pre-filing/post-filing, I think that

2 just deters people even more because they think

3 there are different standards and I really don't

4 believe that.

5 DR. NASR: I agree with you, Sonja.

6 However, there are two reasons I made the

7 statement. Reason number one, I think the agency,

8 on more than one occasion, has indicated that

9 existing process control should meet and are

10 adequate for the indented purpose. So I don't

11 think we are going to go back and change that. The

12 existing methods are adequate for this.

13 I think you are correct, maybe a statement

14 has to be included in the guidance to that effect

15 to reassure industry that we are not looking into

16 things. However, that grandfathering would have a

17 blanket statement that, no matter what we find, it

18 will not be looked into, I don't think we can do

19 that. What if we find something that really has a

20 serious impact on public health? We are a public

21 health agency and we have a responsibility to the

22 public.

23 As far as making a distinction, I think we

24 may end up doing that and let me tell you why.

25 Because if you want to use some of this process

32

1 control and sensors, or the information we gather

2 on-line to replace some of the existing laboratory

3 testing we will rely on some of the existing

4 databases that we have. For example, if we are

5 monitoring particle sizing and we want to correlate

6 that to the solution, it is easier to do it for the

7 existing product where we have a wealth of data

8 based on the information and where we can do such

9 correlations and do this to replace existing

10 laboratory tests. If we do that for a new product

11 it may be more difficult because we don't have that

12 information; we don't have the database that we can

13 use to say this is the only test we are doing. We

14 will do that test, monitor the process to control

15 the process and to release the product. It may be

16 difficult to do.

17 MS. SEKULIC: I agree. I guess I am

18 assuming that that would be the logical process.

19 If you don't have the data you couldn't possibly

20 defend that position. So, yes, in some cases you

21 wouldn't necessarily have access to the appropriate

22 data to do all the permutations of on-line testing

23 that may be technically available. But, again, I

24 think that has to be tailored to the particular

25 case in question.

33

1 MR. CHISHOLM: I think it is clear a lot

2 of interesting things are coming out. I would like

3 to add one, and that is that somewhere we have to

4 consider whether or not a total system approach

5 gives you more confidence about compliance than

6 perhaps just an individual application replacing a

7 like for like, for instance. I think once you

8 start using these systems in terms of your whole

9 manufacturing process and look at it as a total

10 entity, certainly if I were a regulator I would

11 feel much happier with that approach than the

12 single one-off approach. I think somehow we should

13 try to reflect that as well.

14 DR. TIMMERMANS: I would agree. I think

15 it is important to realize, going back to Carl's

16 distinction of three or four different scenarios,

17 that particular for the one-for-one exchange of a

18 release test or current in-process test that you do

19 and you now measure by process analytical

20 technology really nothing changes. You know, in

21 our approach to process validation we still

22 identify our CPPs, our critical process parameters,

23 or quality attributes. We measure those whether we

24 measure those in the lab or on-line, it doesn't'

25 really matter. There is no distinction in the

34

1 approach you take to process validation.

2 So I think that area is well established

3 and really doesn't need much attention. Where we

4 do need to focus is departure from current

5 practices in that if we change our approach to the

6 way we run our processes, the way we control our

7 processes, again, thinking about how we use the

8 data, personally, I believe, and seeing this from

9 experience, that a lot of the applications that

10 have been developed are not necessarily for

11 replacing a finished product test. It is more to

12 characterize a process, to fingerprint a process to

13 determine an endpoint. That still is not such a

14 tremendous departure from what we are doing now.

15 We are just getting more information. So in that

16 respect, again, I don't see that the validation

17 aspects change much.

18 It is going to be different when you

19 ultimately have a different approach to process

20 characterization and maybe going to continuous

21 processing is where your validation approach

22 changes. That is where the guidance is needed. I

23 don't think, again, that for a conventional

24 approach to process to validation however you

25 measure your CPPs and CQAs there is going to be a

35

1 difference.

2 DR. NASR: I think you are right.

3 However, the way I see the major intention of this

4 initiative is to do the process in a different way,

5 and to have a continuous process rather than the

6 batch processing that is currently taking place. I

7 think using the technology as it is being used now

8 to monitor the process, to learn about the process

9 as an alternative test -- there are no changes

10 here. If that is the case, I don't think there was

11 a strong enough reason for us to spend a couple of

12 days discussing this.

13 But the way I see it the agency is trying

14 to work with industry to encourage the use of these

15 sensors, of this technology for a continuous

16 process and now it becomes an important validation

17 issue. Why is that? Because when we have batch

18 processing we stop the process. We collect

19 samples. If we have a problem with a test in the

20 lab we collect more samples until we are quite sure

21 that this process has been completed to some

22 regulatory standards before we move on to the next

23 step.

24 What we are doing today or what we are

25 discussing today is very different. The process is

36

1 continuous. We want to have enough assurance that

2 by monitoring a certain segment of the process that

3 it is working okay before we continue and go on,

4 rather than to stop the process completely, just

5 halting the whole thing and doing testing in the

6 lab. That is maybe where we should focus, the

7 validation when we have a continuous process rather

8 than batch processing.

9 DR. TIMMERMANS: I fully agree. That is

10 exactly where I was going.

11 DR. LACHMAN: Really what we want to take

12 a look at is we have several different types of

13 dosage problems. We have solid dosage forms and

14 for those types of products you have maybe four or

15 five critical operations. Okay? You have milling;

16 you have blending; you have granulation; you have

17 drying; you have compression or encapsulation. So,

18 that is the process stream or the system that you

19 want to control.

20 DR. HALE: Maybe.

21 DR. LACHMAN: Maybe.

22 DR. HALE: Maybe you don't. Can I just

23 take us a step back?

24 DR. LACHMAN: Yes.

25 DR. HALE: I think if this process is

37

1 taking that leap into a continuous regime, then

2 there are some pretty firmly held tenets that need

3 to be looked at. One is that you have equipment

4 set points. If you are measuring a product stream

5 then equipment set points vary all of a sudden

6 because if you have feedback control the set point

7 concept is different. Also, this idea if you are

8 actually measuring something that can infer or

9 directly measure product quality in a release

10 sense, then the idea of three batches of process

11 validation can go away too because you are actually

12 collecting a lot more data for every particular

13 batch, and at some random point in time to do three

14 batches in a row is a concept that could wander

15 into oblivion, I would hope, but that has a huge

16 impact. It is ingrained in the system and I think

17 some of those things would need to be addressed as

18 well.

19 DR. CHIBWE: My name is Kennedy Chibwe.

20 Listening to everybody here, I really feel we are

21 sort of missing a point somewhere. We have to look

22 at the whole process in a holistic manner. When

23 the whole process has been validated you are just

24 going to look at the critical steps to show

25 equivalence to the old lab techniques. I don't

38

1 think you can validate independently the whole

2 system without referring to the whole system of lab

3 QC testing.

4 So I am suggesting that we really should

5 just focus on some key steps that should be sampled

6 to show equivalence. I don't think you can just

7 say that you have a new system and this is always

8 going to run; it is validated and it is fine.

9 There really have to be critical points. I know

10 that certain techniques you can't necessarily

11 relate to the existing techniques. It is totally

12 new. It is a sensor. It is probably just

13 generating some RSDs and telling you that the

14 blending is done. But you could sample and show

15 the continuing uniformity of that particular

16 sample. So there has to be a point where we should

17 identify the key steps and show equivalence to the

18 actual existing data that we have in the lab.

19 Without that, I think we would just be working in

20 the dark and that could be very dangerous.

21 DR. LACHMAN: I think we talked about that

22 a little bit earlier, to identify individual

23 processing steps that are the most critical to

24 control that process. Then you need to develop

25 process ranges around those critical steps, what

39

1 are those ranges that you can have to result in a

2 repeatable end product. Those have to be

3 identified. Then you need to have the appropriate

4 methods of measurement to assure yourself that

5 those critical steps, when they are measured, are

6 repeatable. So, that is the baseline data that you

7 start with.

8 Now, for an existing product you would

9 expect to have that data. That has been developed

10 already. Well, for the really old ones you

11 probably don't have. That gets to be a problem.

12 So we may have to go back and do some homework on

13 those things to develop that information if we are

14 going to use the process control procedure.

15 DR. NASR: I would like to bring the focus

16 back to the guidance issue. The reason I do that

17 is because I think people would like to have some

18 indication from the agency for how to proceed and

19 what is meant by that initiative, and hoping that

20 the guidance will provide that.

21 From what I am hearing today and

22 yesterday, correct me if I am wrong, I think this

23 technology has been used for a variety of reasons

24 over the years in some official regulatory manner

25 and some not so official regulatory manner. So,

40

1 maybe the guidance should have some language to

2 acknowledge existing applications and to encourage

3 the continuous utilization of such technology. We

4 understand that this technology has been used to

5 replace alternative testing, in development

6 process, understanding reaction monitoring, blah,

7 blah, blah. Then there would be more focus -- and

8 that is where I think we are coming from, and that

9 is why we are here today -- to encourage the

10 utilization of these technologies for a continuous

11 process and what would be the criteria needed?

12 What needs to be accomplished during process

13 development, and how can we continue to utilize

14 this as a continuous process rather than batch

15 processing? Because I don't think we are here

16 today to provide assurance that it is good to have

17 the same results from HPLC, correlate that with

18 NIR-IR and to have a correlating coefficient close

19 to 1.0. We know that. There is no reason for us

20 to do that.

21 So, what I am saying again in summary is

22 that the guidance may provide some acknowledgement

23 of existing applications; encourage the utilization

24 for these existing applications and purposes; and

25 look at ways, and focus on how we can use this

41

1 technology for a continuous process to change the

2 way of manufacture of pharmaceuticals in the U.S.

3 industry.

4 DR. LACHMAN: So, what you are saying

5 really is the purpose is to expand the use of

6 current and future technology for continuous

7 in-process controls.

8 DR. NASR: Correct. That is how I see it.

9 DR. LACHMAN: So that is the purpose now.

10 DR. CHISHOLM: Could I say something about

11 that? I am perfectly happy with it. Coming from a

12 continuous process industry before I moved into

13 pharmaceuticals, I am very comfortable with that

14 approach. But it is too big a step just to limit

15 yourself to continuous manufacturing here, for

16 sure, because a lot of companies are going to

17 continue with batch manufacturing. The very nature

18 of the equipment they have means they will continue

19 for a large number of years.

20 DR. NASR: And that is why I tried to make

21 a distinction that there will be two parts of this

22 guidance, if you wish. One part will acknowledge

23 existing utilization of this technology for

24 whatever reason we are using now, and encourage

25 continuing with such utilization. But the other

42

1 part is to move into the continuous manufacturing

2 process.

3 DR. CHISHOLM: I think there is actually

4 an in between step, and that is when you do go for

5 holistic control and you do actually trend over a

6 large number of batches you are almost

7 quasi-continuous anyway if you have this data bank.

8 I think that is a natural step and I think industry

9 will probably go in that direction.

10 MS. SEKULIC: I just wanted to throw out

11 that I agree with Bob. In listening to the

12 discussions and to some of the impediments and

13 difficulties, a lot of questions come up and I

14 think probably a lot of them can be summarized as

15 follows: "You know, I've got this great new tool.

16 I've got a pair of eyes inside my vessel now but

17 you know what, I've still got to blend the sucker

18 for three hours. So what's the point?" Okay?

19 I think that raises a very good question.

20 Are we, in this guidance, going to be endorsing the

21 fact that now that we have the capability to

22 establish when the true endpoint of a process step

23 is, is that going to be an acceptable way to

24 operate our processes and our plants? I think it

25 is a very real question in a lot of people's minds.

43

1 DR. LACHMAN: Well, you know, I think we

2 do this to some extent now. When you make wet

3 granulations you use amperage measurements on the

4 blender to determine an endpoint. So, indirectly

5 you are doing this now, and also for bed drying you

6 are using temperature and air flow and moisture.

7 That is being done right now on the process batch

8 basis.

9 DR. ANDERSON: A good example of this is

10 will the draft include language, for example, on

11 blending? Blending is a great example. We are

12 validated for three hours.

13 DR. LACHMAN: Right.

14 DR. ANDERSON: You run that thing for

15 three hours. The draft needs to say if you have a

16 validated process measurement that will indicate an

17 endpoint that meets the necessary criteria, you may

18 stop at that point.

19 DR. LACHMAN: Yes.

20 DR. NASR: Absolutely.

21 DR. ANDERSON: I think that is a critical

22 point in this whole thing. Without that feedback

23 we are spinning our wheels.

24 DR. LACHMAN: So, let's say the purpose is

25 to expand the use of current technology for

44

1 continuous and batch on-line in-process controls.

2 It is batch and continuous on-line in-process

3 controls.

4 MS. SEKULIC: I think validated variable

5 processing times, some verbiage to that effect

6 would go a long way to building confidence that

7 this is endorsed by the regulatory authorities.

8 DR. LACHMAN: Yes.

9 MR. OLSON: I would even go so far as to

10 say, you know, you make a submission that just says

11 that you will blend until well mixed.

12 [Laughter]

13 I mean, at first our heads would explode

14 but then we would have to sit back and say, okay,

15 what does that mean? Granted, you have tools to

16 determine that, but that is really going to be a

17 big issue, a big issue for reviewers and

18 investigators because that is where this will be

19 going to where, you know, you no longer use the

20 three hours, four hours, six hours. That is why I

21 was kind of thinking yesterday they were producing

22 products that are out of compliance. You wouldn't

23 do that. I mean, no one would start making tablets

24 from a poor blend. So, you almost don't end up in

25 that situation of ever having to re-process,

45

1 assuming that everything is in control and your

2 equipment is working fine.

3 DR. NASR: I really don't consider this a

4 problem at all. The reason I don't is this, if we

5 are using this for existing products obviously we

6 have enough data to make us look at your submission

7 for a possible change because you have the data to

8 back it up. For a new product you should do enough

9 process development and generate the data to

10 support your idea that blending should go on only

11 for 15 seconds rather than 15 hours but, again, you

12 should have the data to support that. So the idea

13 is that once you have the data to back up your

14 submission, I don't see that as a problem.

15 MR. ELLSWORTH: One of the things that we

16 discussed earlier, and listening to this

17 conversation, is that I think a lot of our

18 validation now uses surrogate endpoints, to use a

19 term from the clinical side, like blend time and

20 things like that, that relate back to analytical

21 results, general analytical wet chemistry because,

22 obviously, we can't do enough testing to test a

23 whole batch so we create these surrogate endpoints

24 to say that everything is okay. Whereas with this

25 inferential testing, if you have a testing endpoint

46

1 that is actually better than time or blender speed,

2 that is something that we need to encourage. As

3 you said, it is a variable thing but the

4 inferential endpoint has to be correlated with the

5 analytical data that shows, say, the blend is

6 perfectly blended and if it takes 30 minutes as

7 opposed to three hours, you hit that endpoint point

8 and you know that endpoint is good to move forward.

9 I don't know if I have explained that very well but

10 I think that is kind of an important point.

11 The other thing too is that when we talk

12 about how do we go about validating, there are a

13 lot of current principles about process validation,

14 and I hear this conversation about is something

15 changing here. When we correlate we are still

16 running the normal validation; we are taking units;

17 we are determining upper and lower control limits

18 and correlating maybe this new instrument to those

19 limits. Maybe we do a little bit additional

20 testing because we probably want to show that that

21 new instrument can actually show if something

22 fails; maybe we take it further than we have

23 before. Is there something unique about the

24 general principles across the validation that this

25 new technology poses?

47

1 I will add to that. I am not so sure that

2 continuous monitoring or the continuous process is

3 really that big an issue because we deal with

4 process validation in the area of active

5 pharmaceutical ingredients and a lot of those are

6 continuous processes; just divide up the batches as

7 time units.

8 DR. HALE: But whether it is real or not,

9 there is hesitancy to go to continuous processing.

10 I have personal experience with that. If you take

11 the blending example you can go through the stages

12 of taking a V-blender and measuring it better so

13 that you create a different endpoint. There are

14 technologies that are developed so you can get rid

15 of the blender altogether and make a continuous one

16 where it has no endpoint; the concept of endpoint

17 goes away. There has been resistance, and maybe it

18 is internal within the company itself, to go to

19 that because it is a perceived regulatory problem.

20 So, maybe, in a step-wise fashion, if you

21 implemented something like that, that would be a

22 continuous batch process because it wouldn't be

23 completely continuous, those capabilities are

24 available in this guidance as well.

25 DR. LACHMAN: What if we say to expand the

48

1 use of current and future technology of on-line

2 controls for batch and continuous product process

3 validation? Sonja, you had a concern with time.

4 DR. NASR: Can you go back to the last few

5 words in your sentence?

6 DR. LACHMAN: Expand the use of current

7 and future technology of on-line controls for batch

8 and continuous product process validation.

9 DR. NASR: Why validation? What is the

10 use of the word validation? It doesn't fit there.

11 DR. LACHMAN: Okay. MS. SEKULIC: But we

12 were trying to incorporate some kind of variable

13 processing.

14 DR. LACHMAN: You were looking for the

15 time factor there.

16 MS. SEKULIC: Yes.

17 DR. NASR: I think what you have is the

18 purpose, and that is okay. I think we can go to

19 other areas, categories or critical issues that we

20 need to address that relate to the process and

21 analytical validation.

22 DR. TIMMERMANS: One thing that, in

23 listening to the discussions, repeatedly comes to

24 mind is maybe a question that needs to be addressed

25 in the guidance. Let's take the example of a

49

1 blending process, is a dual approach of blending to

2 an endpoint through an on-line technology, as well

3 as blending for a specific time because we know

4 that after blending for, let's say, three hours we

5 have a uniform bland as determined by either the

6 on-line technology or by existing analytical

7 technology, is that acceptable, to have a dual

8 approach essentially in place?

9 DR. LACHMAN: I don't see why not. I

10 really don't see why not.

11 DR. NASR: You can do that now. You don't

12 need a guidance for that.

13 DR. LACHMAN: Yes. I don't see what the

14 problem is right now.

15 DR. TIMMERMANS: I agree that at this

16 point for a conventional approach I don't need a

17 guidance, but to allow me to blend to an endpoint

18 using on-line process analytical technology that is

19 where the guidance would apply. That is where this

20 specific guidance would apply. So I think the

21 consideration of having the two run in parallel --

22 I see this also particularly where you are doing

23 this fingerprinting and where you are learning

24 about your new processes, if you decide to

25 implement process analytical technologies where you

50

1 may not have built enough confidence in the

2 methodology through the process validation stage.

3 In order to get the confidence in your process

4 analytical technology, you may need to process for,

5 you know, ten batches. So you need some type of a

6 dual approach in that period of time so that you

7 have confidence that your product is, indeed,

8 acceptable but you have the opportunity to build

9 your process analytical technology.

10 DR. LACHMAN: You do need the numbers to

11 have a statistical confidence level. No question

12 about it. You can't do it with three batches. I

13 think you need somewhere between ten and twenty

14 batches. However, you are going to have a lot more

15 data if you are doing continuous measurement. So,

16 the number of individual measurements may

17 compensate for ten batches because you are doing

18 multiple measurements, and the tolerance levels

19 that you can estimate with 1000, 2000 points is a

20 lot better than, you know, 20 or 50 points of

21 measurements.

22 DR. TIMMERMANS: I agree, but I still need

23 to demonstrate on a number of batches that I have,

24 indeed, reached my endpoint. If I take 15 data

25 points to get to the endpoint I still need to

51

1 demonstrate that I hit that endpoint confidently.

2 DR. LACHMAN: But how many would you need

3 with all those points to accomplish that? Now that

4 you have, let's say, 1500 points per batch rather

5 than 50, if you have three batches that will give

6 you 100 points for a statistical evaluation,

7 wouldn't it?

8 MS. SEKULIC: I think that kind of falls

9 into the method development discussion. If I can

10 just say that we all seem to be in agreement that

11 this is not an issue, however, in the industry

12 there is a lot of misconception that this is not an

13 endorsed type of an approach to controlling our

14 processes. That is why we are asking for it to be

15 verbalized in this guidance.

16 DR. NASR: I am aware of that, and that is

17 why I suggested earlier that we could possibly in

18 the guidance have two areas to cover. One area is

19 to encourage the existing utilization of these

20 technologies to monitor the product, to control the

21 process. I think what you are asking for is some

22 guidelines for how we can do that and how it will

23 be accepted by the agency. That will be an area

24 that I think we should encourage. As a matter of

25 fact, I think we should focus right now on why.

52

1 Because we have enough data sets out there already

2 being used, just to make it a little bit more

3 formal, that guidance should facilitate that.

4 The other area is utilization of these

5 technologies in a continuous process. What is

6 needed during process development; how many

7 batches; how many samples; what kind of

8 correlations you need, and so forth. These are two

9 different areas. I am not trying to say old versus

10 new, or whatever, but I think we have to somehow

11 organize the guidance in a way to facilitate the

12 implementation.

13 DR. LACHMAN: Well, let's get this down

14 pat now. I think we have been going back and

15 forth, and right now I have come back to expand the

16 use of current and future technology of on-line

17 controls for batch and continuous production for

18 existing and new products.

19 MR. CHISHOLM: Can we be a little bit

20 careful about the use of the word on-line? It

21 doesn't necessarily convey what I think you are

22 after. On-line means it is actually looking at the

23 process. At-line means you are taking a sample and

24 doing it right beside the process but on a much

25 more statistical basis than doing it just at the

53

1 end of a batch. We have to watch that terminology

2 because a lot of companies will probably go to

3 at-line before they go to on-line, especially if it

4 is pharmacists perhaps doing development rather

5 than engineers.

6 MR. FAMULARE: I don't think anybody would

7 argue that we shouldn't be restrictive in any way,

8 and it is just a matter of crafting it so that it

9 is open for either one.

10 MR. CHISHOLM: Can I come back to the

11 previous point because I think there is a very

12 crucial issue here? At the moment the industry is

13 very comfortable with the three batch approach

14 because it can get its products to market very,

15 very quickly. Once you get into using statistical

16 techniques, you are perfectly right, the bigger the

17 data set you have the more confident you are. And,

18 this is a very, very critical area that I think we

19 have to address in some way. It may well be that

20 it is not old versus new. It may be different and

21 I think we have to look at that because unless we

22 oil these wheels I don't think the industry will

23 move in the direction that the FDA would like to

24 move in.

25 The other point is that once you go to

54

1 these big control systems on your processes, the

2 win-win situation issue actually is to revalidate

3 your process for every batch, and I think that is

4 absolutely crucial and it has to reflect in some

5 way as being a goal that people should be looking

6 for.

7 DR. NASR: I think Leon has done a really

8 good job of trying to state a purpose of the

9 guidance. Is it logical for us to move to the next

10 step? If there is some endorsement or agreement

11 that we are looking at two different areas here to

12 consider, maybe we can go for each one of these

13 areas and try to identify some critical issues that

14 need to be addressed. Then once we go through

15 that, if we have time today we can see what answers

16 the industry and the agency have to come up with to

17 encourage utilization of these technologies.

18 DR. LACHMAN: If we take on-line off and

19 just say expand the use of current and future

20 technology of controls for batch and continuous

21 production for existing and new products, they can

22 do it on-line or off-line. That is up to the

23 individual. Let's go to the next one then.

24 MR. FAMULARE: I just wanted to go back to

25 a key point that I think Bob brought up, and that

55

1 is that the industry now is comfortable with

2 validating three batches using existing

3 methodology, and there is the potential in the

4 development of some of these on-line or at-line

5 controls to take a great deal longer to develop

6 certain databases to put that in place.

7 DR. LACHMAN: Yes.

8 MR. FAMULARE: And, if we want to

9 encourage the use of this technology I think we

10 have to, as Moheb said, think outside the box. I

11 think we have to think of a creative way to not

12 hold up product approval while this huge database

13 is being developed, if it is needed. So, I think

14 we really have to think about those data points and

15 what is needed. Will it be some sort of dual

16 approach where conventional validation will be

17 overtaken by PAT once it can be put into place? I

18 am thinking as I am speaking here, but I think

19 obviously if it is seen as slower than the existing

20 process we will have abated our purpose, which is

21 to encourage the use.

22 DR. LACHMAN: I don't think we could get

23 away from the current approach until we have this

24 developed. So, I think the current approach will

25 still apply. You know, this will be going on

56

1 simultaneously for existing products, or for new

2 products you can start this way.

3 DR. NASR: That is correct.

4 DR. LACHMAN: But you are still going to

5 have to continue with the current approach until

6 this is accepted.

7 MR. ELLSWORTH: I guess I have a question

8 because of something that was said earlier. Really

9 we are talking about correlation between the new

10 approach and the existing approach, but you had

11 mentioned that there may be areas where we would

12 not necessarily be correlating to an existing

13 knowledge base; we would be generating a new

14 knowledge base using this technology. That I think

15 could move us to a quicker conclusion that a

16 process is validated, but I am not sure how we do

17 that if we don't correlate it to an existing

18 knowledge base.

19 MS. SEKULIC: It depends on how you define

20 correlation. Traditionally we look at some on-line

21 measurement data and we correlate that to what is

22 happening in the lab. If we stir the pot a little

23 bit and challenge that understanding, say, for

24 example, I put a probe in my blender but I don't

25 necessarily want to pull samples with a thief, take

57

1 it to the lab and run LC. What I am actually going

2 to do is correlate that data, or the endpoint or

3 some parameter of that measurement technique to

4 maybe my tablet release data, or to the next

5 processing step measurement. So, if I have

6 correlation between two steps, is that a validated

7 approach? In some cases it will be. That is

8 really what we are proposing.

9 If you look at the holistic approach to

10 what your processing steps are, you don't

11 necessarily have to have an on-line measurement in

12 each and every step. More than likely it will be

13 some permutation of on-line measurements and lab

14 testing, depending on how you define your critical

15 points and where your confidences lie and how much

16 data you have. So in that situation, yes, I have a

17 qualitative on-line measurement where I don't

18 necessarily have a database of lab data to

19 correlate to, but in the overall process definition

20 I wouldn't be making that measurement unless it is

21 meaningful. What is it meaningful to? Perhaps to

22 some other measurement technique, and that is what

23 I would consider to be another type of correlation.

24 So, that is what I was thinking about.

25 MR. ELLSWORTH: But that is important to

58

1 be in the guidance, that concept.

2 DR. LACHMAN: But, Sonja, won't that have

3 to be somewhere related to an absolute measurement

4 down the line?

5 DR. NASR: Not necessarily. I would

6 challenge this some. I think in order for us to

7 have true on-line monitoring and testing without

8 relying on laboratory data, what we need to do is

9 to have confidence in on-line testing. That

10 confidence can be established by two things, number

11 one, having historical data and correlation for

12 existing products where we have enough laboratory

13 testing stability indicating HPLCs and everything

14 in the world that with some signal you get from one

15 of these devices, and we have confidence that that

16 device and that signal is meaningful to control and

17 to monitor the process. That is one thing.

18 The other thing is research and

19 publications to establish such correlation. Once

20 we have confidence in this correlation a future

21 product may rely -- may rely, I underline the words

22 "may rely" on just a signal on-line without really

23 going back to hard laboratory batch analysis of

24 every sample you can collect.

25 DR. LACHMAN: I agree but you developed

59

1 that basic correlation at one time.

2 DR. NASR: Not for every product.

3 DR. LACHMAN: No, but at one time.

4 DR. NASR: Correct; correct.

5 DR. LACHMAN: That is fine. That is what

6 I was looking for. You have at one time developed

7 that.

8 All right, so we have a purpose now.

9 Let's go into the body of the outline of the

10 document. Do we want to have this a general body

11 or do we want to relate it to particular product

12 classes? What is the feeling of the group?

13 MS. SEKULIC: I think if we keep it

14 general and high level that is probably a better

15 approach, rather than diving into specifics.

16 DR. LACHMAN: Okay.

17 DR. TIMMERMANS: I don't know if you need

18 to go back to the questions that Moheb posed

19 before, where we have the two approaches. I see

20 that at the end of the day we need to have some

21 answers to the questions that are written down

22 here. So, maybe it is a good point to focus either

23 looking at these questions or going back to the

24 questions that Moheb proposed.

25 DR. LACHMAN: What approaches should be

60

1 considered for validating the use of new PATs for

2 in-process control during manufacturing?

3 DR. NASR: I want to interject a comment

4 here. I don't think we have to go over each one of

5 these questions. Maybe our group can define the

6 questions that we need to discuss today and that we

7 would like the guidance to address. I think that

8 would be more practical than going over these

9 questions that were brought together basically to

10 make us focus on some of the critical issues.

11 DR. HALE: Yes, I agree. I think if we at

12 least initially started out with a list of issues,

13 some of which we have highlighted already and based

14 on that list of issues figure out where to go, that

15 would at least highlight what the major components

16 of this are that need to be addressed.

17 MR. FAMULARE: I think some of our

18 discussion has been about an attempt to solve these

19 issues, but I think if we name the issues we will

20 know what to write the guidance around.

21 DR. LACHMAN: Okay, since we have been

22 jumping around, let's get those issues

23 re-identified. What is the main issue here that we

24 think we need to address to start this?

25 MR. FAMULARE: I think Carl Anderson had

61

1 laid out four points that I think are very good to

2 address in the guidance, a little while ago, in

3 terms of relating it to existing tests, new tests,

4 whether we want to deal with it for finished

5 product testing or for control. You can probably

6 state it better than myself but I thought those

7 were some good frames of reference to write the

8 guidance around.

9 Then, the other issue that we brought up

10 was really a concrete example. You know, will the

11 agency accept the result of a sensor as the

12 determining point of a blend and forget the time?

13 Forget the clock? The answer is that is one of our

14 encouraging points and it makes sense, but we need

15 to at least bring up that point in the guidance.

16 MR. ELLSWORTH: As long as there is some

17 appropriate correlation with what that means.

18 MR. FAMULARE: Yes. Then, the other issue

19 is that, again, we are encouraging the use of the

20 guidance, as Leon laid out well in the purpose. I

21 think we have to think very hard about not making

22 the accumulation of the data for this longer than

23 validating three batches under the conventional

24 method.

25 DR. NASR: I think the critical theme of

62

1 the guidance, if I may suggest, is how to encourage

2 the technologies to monitor, to control the

3 process, to replace the existing tests, and doing

4 all that without sacrificing the quality of the

5 finished product. I think that will be the theme.

6 It doesn't have to be one issue that is addressed

7 in the guidance, but I think if we agree on a theme

8 and purpose to encourage these technologies for all

9 these things without sacrificing the quality but,

10 as a matter of fact, to enhance the quality of the

11 product that we manufacture.

12 DR. LACHMAN: Let's go to key topics now.

13 Let's go to that because I think we are bouncing

14 back and forth. What are the key topics that we

15 want in the body of this draft?

16 DR. ANDERSON: Within the framework of

17 different applications the key topic is the

18 elements of validation. I think that is something

19 we have to discuss, what types of things need to be

20 done in a very general sense; what needs to be

21 demonstrated such that we may use an on-line method

22 or analytical technology.

23 To clarify that, what is it we have to do

24 to show that this measurement connects to product

25 quality in the end? That correlation, whether it

63

1 is a direct chemical entity or a process signature

2 type correlation, some guidance as to what is

3 acceptable or unacceptable, or what is acceptable

4 within that framework I think is necessary.

5 MR. FAMULARE: Or, as you had brought up,

6 the correlation of one step to another. As an

7 example of what is not acceptable, just in my own

8 mind, is to just use this to substitute if you were

9 going to do a small sample of tablets under

10 conventional wet chemistry, and to just substitute

11 this for that small sample doesn't gain anything.

12 That is an obvious not acceptable use of this

13 technology. It doesn't gain anything for the

14 agency. It may allow the company to eliminate some

15 laboratory operations but it doesn't show a gain on

16 the win-win side.

17 DR. HALE: But it shows a gain if, instead

18 of doing six you do 60 or 600 --

19 MR. FAMULARE: Exactly, right.

20 DR. HALE: If you just take the six and

21 switch from HPLC to NIR, for example, it wouldn't

22 do much for anybody.

23 MS. SEKULIC: I think the over-arching

24 umbrella of this is to recognize the fact that we

25 are not changing the quality. I mean, the quality

64

1 of the final product has to stay in place. That is

2 our golden marker. That does not move.

3 DR. LACHMAN: The final quality product

4 attributes remain the same.

5 MS. SEKULIC: Correct.

6 DR. NASR: Or improved.

7 DR. LACHMAN: Or better.

8 MS. SEKULIC: What we are talking about is

9 providing a guidance that talks about the way we

10 reach that final goal. That is really what we are

11 talking about. If you take the holistic approach

12 to a process, I think the guidance needs to sort of

13 set out the fact that we are probably looking at

14 some combination of on-line measurement monitoring

15 as well as conventional testing, some combination

16 for an entire process.

17 I really wouldn't like to see the guidance

18 narrowed into a one-for-one replacement because I

19 think we have done that, you know, for more than a

20 dozen years and it hasn't really netted us where we

21 want to be. I think making that concession in the

22 guidance and saying, you know, we really see this

23 as being a combination approach -- and who knows

24 what the future holds; maybe it will all be on-line

25 but we are not quite there yet. I think for right

65

1 now to get the confidence both on the regulatory

2 side and the industry side we really need to take

3 it a step at a time, and what makes sense for the

4 process is some combination of those two

5 approaches, and I think that needs to be spelled

6 out.

7 MR. CHISHOLM: What we are actually doing

8 is improving quality assurance in terms of

9 compliance. You are more assured because you are

10 doing a much more increased level of sampling.

11 That is a very, very important distinction. Your

12 quality may stay exactly the same. The problem you

13 have at the moment is that you don't know whether

14 it is there for the three million tablets. With

15 the new systems you will be able to give a much

16 higher level of assurance that it is, and I think

17 that is something that needs to be brought out.

18 DR. NASR: That is very good coming from

19 an engineer!

20 DR. HALE: I think the other point would

21 be this idea of the ability to change the way we do

22 processing, and perhaps that could be restated.

23 Instead of continuous, the integration of unit

24 operations. Perhaps that would allow methodology

25 for advancing the process technology as well as the

66

1 sensor technology because I think there are

2 limitations. Perhaps that is just a clarification

3 because we do some of that now. We use temperature

4 to measure endpoint of drying in the same sense

5 that we could use it in other unit operations. But

6 there is baggage that has been developed around

7 blending that is an incredible pile of baggage, so

8 we could use these to clarify techniques that we

9 already use. But if we could integrate blending

10 and compressing right now, and have in-process

11 testing that we would use the tablet content

12 uniformity as well as in-line testing of powder

13 flow we could innovate the way we develop processes

14 and there is a barrier right now.

15 Again, whether it is real or perceived,

16 there is a barrier and I think this guidance would

17 serve the industry well to specify that we can do

18 some of those things, that we can integrate

19 processes and not have to stop at each unit

20 operation in a static sense and do it. It doesn't

21 have to be the whole thing but perhaps different

22 unit operations could come together and integrate,

23 and specifically say that that type of thing could

24 be done.

25 MR. FAMULARE: I think that goes to the

67

1 advantages that we saw on the slides yesterday in

2 terms of compressing the overall processing time.

3 I think we should also be able to encourage process

4 change with minimal regulatory block to that. I

5 mean, you shouldn't have to file a process

6 improvement if it is obviated by what your sensors

7 are telling you. That should be able to be covered

8 by GMP. If, by virtue of what your sensors are

9 telling you, you should increase or decrease

10 blending time or make whatever adjustment in the

11 process and you have the data at your firm, it

12 should be able to be covered under GMP so that

13 there shouldn't be regulatory delay in implementing

14 process improvements which, again, will bring up

15 the overall quality of the product. If it actually

16 changes the specifications of a product then, of

17 course, it is an important filing issue and we

18 should have ways of expediting that.

19 DR. LACHMAN: So, what we are saying here

20 is to adjust the regulatory requirements for

21 filings versus GMPs.

22 MR. FAMULARE: Well, I think the paradigm

23 exists already to do that under the GMP. Process

24 validation is already under GMP. We do it

25 post-approval. If you are going to make process

68

1 improvement, you should be able to make it as your

2 sensors tell you.

3 DR. LACHMAN: But essentially you are not

4 changing the process. The process is still the

5 same.

6 MR. FAMULARE: You are keeping closer to

7 the process because you know it now.

8 DR. LACHMAN: Right, you are controlling

9 it.

10 MR. FAMULARE: But it may change these

11 other parameters that we have been hung up on in

12 the past, such as time, etc. As long as it is

13 demonstrably the same as the key or the pivotal

14 batch, then it should be able to move forward

15 without waiting for filings. Certainly, the review

16 side, even before the CB-30s and all these other

17 types of steps to allow these continuous

18 improvements, should carry that forth in this

19 guidance.

20 DR. HALE: But you could see changes in

21 your operations of this. It may fit into the

22 current thing but you could see if that opened up

23 the ability to innovate, you could go back and

24 reassess what you are actually trying to do in

25 terms of putting unit operations together and come

69

1 up with some new processes, new pieces of equipment

2 that haven't been seen. It may fit under the same

3 concept but I would think that this guidance would

4 want to have verbiage that would encourage that and

5 provide a means for introducing new process

6 technology, as well as new sensors, to allow this

7 higher quality and better processing. So, the look

8 of pharmaceutical manufacturing could be different

9 and perhaps encouraged.

10 DR. NASR: One issue, and I don't know if

11 it is covered or not, which I think is very

12 important to be covered in this guidance is the

13 issue of what is needed to accept an on-line test

14 in replacement of a laboratory test. Let me repeat

15 what I just said, what is needed in order to accept

16 a replacement of a laboratory test by an on-line

17 test, not to use it as an alternative or as a

18 secondary but to accept it as a test. Because I

19 think without doing that the industry will be

20 reluctant to utilize technologies because it is

21 duplication of effort and an increase of resources.

22 MS. SEKULIC: I guess I am confused. We

23 spent yesterday saying that this will never been

24 accepted, these types of techniques will never be

25 accepted as stand-alones. Now I am hearing

70

1 something different. So I just want to clarify.

2 DR. NASR: No, I meant what I just said,

3 especially for existing products where we have

4 enough information and existing databases and we

5 have all the knowledge base that we talked about

6 this morning and yesterday. Is it possible to use

7 only on-line tests to replace some of the existing

8 laboratory tests? I don't see why not.

9 MR. FAMULARE: If you had the corollary

10 background there would be no reason not to be able

11 to do that. I mean, one of the examples might be

12 the control of particle size and other blend

13 characteristics, physical characteristics versus

14 dissolution. Once you show the corollary, that you

15 are able to do that, the dissolution won't go away

16 as the official test. If somebody were to go to

17 the marketplace and pull your product it would have

18 to meet USP, whatever, for dissolution but as a

19 matter of your routine release of product the GMPs

20 tell you that you have to do a test. They don't

21 say which test. As long as you test each batch for

22 appropriate specifications upon release, you should

23 be able to use physical parameters as your now

24 direct measurement of what is controlling a process

25 versus the dissolution, and that should work under

71

1 our existing regulatory paradigm.

2 DR. NASR: Exactly. I did not mean to

3 confuse you, Sonja, but I think this is a very

4 critical point. In order to utilize this

5 technology and to advance the stop further -- if we

6 cannot do that I don't think we are going to move

7 very far. So, we have to say, well, can we use the

8 databases and the knowledge base we have under

9 existing GMP? Can we replace, not just add on,

10 on-line tests in place of laboratory tests?

11 MS. SEKULIC: But that is a very

12 significant clarification from the industry

13 perspective, again, talking about addressing

14 industry fears, phobias and misconceptions, and I

15 think this falls into that realm yet again. If the

16 regulatory authorities were to endorse in a written

17 guidance or a document that conceptually this is a

18 legitimate approach, then I think we go a long way

19 to dispelling the existing phobias.

20 DR. LACHMAN: I think what we are saying

21 here is that we have to delineate what the

22 requirements are for accepting on-line testing in

23 place of off-line testing. We heard that if you

24 can show correlation or equivalent capability of

25 the on-line versus the off-line, it should be

72

1 acceptable.

2 DR. NASR: And I would like to add to

3 that, that if we are going to do that we don't have

4 to be very restricted to utilizing even the

5 old-fashioned -- not old-fashioned; old-fashioned

6 is not the right word but the old validation

7 criteria for the old test and make the same

8 restriction on the on-line test. Let me rephrase

9 this, for example, if we have stability indicating

10 it should be a method used off-line, we don't have

11 the same validation criteria for on-line tests as

12 long as we can establish the correlations.

13 DR. ANDERSON: This guidance needs to be

14 specific, and that is a very good example, as to

15 what is necessary to validate. If we are

16 validating a moisture measurement, stability

17 indicating or specificity is irrelevant. We know

18 it is specific and we know it works for moisture.

19 That is the kind of language that we need to be

20 very careful about, that you don't need to go

21 through all of the traditional validation elements

22 on every method. The validation elements covered

23 in a process analytical technology validation need

24 to be those that are relevant to that technology

25 and to that measurement only, and we don't need to

73

1 be encumbered by these other, frankly, irrelevant

2 validation elements.

3 DR. LACHMAN: I think what you are looking

4 for really in those cases is accuracy and

5 robustness of your method. That is about it,

6 period. You don't need all the other elements.

7 DR. NASR: No, an important element that

8 must be there is that the method is suitable for

9 the intended purpose.

10 [Laughter]

11 DR. LACHMAN: I have a few more items I

12 have noted here. We want to adjust the regulatory

13 requirements for improved in-process controls

14 resulting from PAT. That should be a

15 consideration, what would be required

16 regulatory-wise for firms that have improved

17 in-process controls which may combine two process

18 steps. The process controls allow for that so

19 there should be some provision for that. This is

20 something you have just recently brought up.

21 DR. NASR: I am a little confused, Tom.

22 What did you mean by integrational process versus

23 continuous process.

24 DR. LACHMAN: It is still continuous.

25 DR. HALE: I think the example is that

74

1 right now you could have a V-blender or a

2 bin-blender that is treated as a unique operation.

3 You turn it on for a while or you measure the

4 endpoint and measure the static bed, and then you

5 drop it through a pipe into your compressing

6 machine and you measure the compressing machine as

7 an independent unit operation. One interpretation

8 of a real test of the blender is if you make good

9 tablets, and you could have a system that had

10 either the existing technology or a continuous

11 blender, or some such thing, or a measurement of a

12 moving powder bed where the success of those two

13 processes would be the success of one. The moving

14 powder bed would have good results and that the

15 tablets would have good results. So, you are

16 integrating what exists now as two independent unit

17 operations into one series of unit operations or

18 one process that makes more sense together than

19 apart. You could do that as opposed to automating

20 or continuous of the whole thing for final product.

21 You could have integrated steps.

22 I think another clarification on the point

23 of implementation is that a big concern is when the

24 individual unit operation or on-line testing needs

25 to be correlated with product release type data

75

1 because that is when people really get scared. If

2 it is a matter of improving blending for the sake

3 of blending, that is fine. But if there is a

4 correlation needed with release type data, then the

5 hurdle gets a lot higher. So, in terms of the

6 implementation, that is what people get scared

7 about. In my experience, instead of taking six

8 tablets you take 120 or 200 tablets on existing

9 products that are manufactured every day, that is a

10 real hurdle to overcome and there might be some

11 clarification there that some of these things on

12 existing products, not on new products where

13 release type testing needs to be done to implement

14 change because those are different performances.

15 DR. NASR: I think what Joe is saying is

16 you are afraid that in the process of replacing

17 some of this laboratory testing to monitor the

18 process we may end up adding more tests at the end.

19 Is that your fear?

20 DR. HALE: No, I think that if you could

21 implement something that would blend better, that

22 if your ranges on blending were this and you could

23 get it down to here, that is going to improve your

24 process for various reasons. But if you are

25 testing the end product there are natural

76

1 distributions and if you had to do a whole bunch of

2 testing you may or may not come up with issues on

3 the product.

4 MR. FAMULARE: Well, if you are looking at

5 more end units, as you are saying, you may see more

6 variability and that may have to be taken into

7 consideration in the file specifications, just as

8 if you are looking at a formulary result in the USP

9 versus the direct assay result based on a smaller

10 sample. So, there are other considerations that

11 you will have to think of but, again, I don't think

12 we want to flip it around in terms of saying now

13 you have the capability to do 200 tablets but let's

14 just do six because a smaller sample will pass.

15 You know, we have to deal with knowing more

16 information in the process and at the end of the

17 process, and not looking at it within the sense of

18 lowering the quality of the product but within the

19 sense of statistically you have a different answer.

20 It may reflect a better quality batch than one

21 where you only looked at six tablets but you had a

22 narrower range because of the small sample.

23 DR. LACHMAN: If we said improved quality

24 assurance with the new system by combining on-line

25 and conventional testing, that will cover both.

77

1 DR. NASR: Is that what you had in mind,

2 Tom?

3 DR. LACHMAN: We are not saying exactly,

4 you know, if you have to change it, reduce it or

5 increase it. That is going to be dependent on your

6 system. I don't think we can make an across the

7 board statement that it is going to be less or more

8 for end product testing.

9 MR. FAMULARE: I guess the problem is that

10 we don't want this to come off as additional tests.

11 DR. LACHMAN: No, we are not saying that.

12 MR. FAMULARE: No, but I guess that is the

13 fear if I am stating it correctly.

14 DR. HALE: I was really responding to

15 Sonja's statement yesterday that there were these

16 hurdles to overcome, especially on existing

17 products. It was more in response to that because

18 I think that is a real fear, that if you change

19 something then what are the hurdles that you are

20 going to need to overcome and what you are going to

21 find on existing products, not on new products but

22 on existing products, especially the older you get,

23 the more that is an issue. It was more in response

24 to that than a need for guidance regulation.

25 DR. LACHMAN: I think the agency has

78

1 agreed that the current product and processes have

2 produced repeatable products of quality attributes,

3 and that is not going to change any. If you find

4 new information that indicates that you can do

5 better, it doesn't mean that what you do now isn't

6 any good because you have been doing it for however

7 long it was approved that way.

8 MR. FAMULARE: It is a matter of putting a

9 statement in the guidance. You know, "if it's not

10 broke, don't fix it." I think that is what you are

11 saying. I think we have to change our thinking,

12 "if it's not broke, break it." In other words, you

13 have to look for continuous improvement all the

14 time, and we have to set up the environment in

15 which you can do that.

16 DR. LACHMAN: That is right.

17 MR. FAMULARE: That is what we need to

18 say.

19 DR. LACHMAN: Yes, that has been a focal

20 point yesterday and today, and I think we are going

21 to have to address that somehow to make people

22 comfortable to go ahead and do this. So, it is the

23 comfort level that we are going to have to be able

24 to do here.

25 MR. DRENNAN: Hi. I am Jim Drennan. I am

79

1 not a member of the working group, but is it

2 appropriate for me to make some comments today?

3 DR. LACHMAN: Yes.

4 MR. DRENNAN: I think the major hurdle we

5 face in terms of regulatory issues as we implement

6 these process technologies certainly has to do with

7 the case where we are replacing the existing

8 methods with the process technology. We see

9 certain examples of this coming up today as we

10 attempt to develop methods like the NIR-infrared

11 applied to analyzing intact dosage forms where we

12 have a range problem. We cannot find an adequate

13 range in the product that comes off the production

14 line and so we strive to expand that range to meet

15 the current regulations in the USP of meeting the

16 85-115 percent, etc. But these problems become

17 even more serious as we think about implementing

18 these technologies on-line somewhere because there

19 is then certainly no opportunity at all to expand

20 that range in any way. That is just one example. I

21 think the regulatory hurdles are most important as

22 we try to replace existing technologies with these

23 process systems.

24 DR. LACHMAN: I think the question here is

25 you have a range of 85-115 and if your on-line

80

1 in-process controls provide you a much narrower

2 process variability and if an FDA investigator

3 comes in and sees this in-house data, are they

4 going to request that you tighten your controls for

5 the product? That is a concern.

6 DR. NASR: Are you referring to linearity

7 requirements? Are you talking about something like

8 that?

9 DR. LACHMAN: No, no, no.

10 MR. DRENNAN: Well, that is a concern as

11 well. Linearity range, those are issue that really

12 are essentially one and the same in some of these

13 cases.

14 DR. NASR: I am missing the point here.

15 DR. LACHMAN: I think this is a compliance

16 in-process control that he is raising here.

17 MR. DRENNAN: My concern is what would be

18 an appropriate specification for us to meet in

19 terms of proving content uniformity or product

20 quality in any regard.

21 DR. NASR: So you want to make it

22 narrower?

23 MR. DRENNAN: No, but I wish to have

24 freedom to prove that the method that we have is

25 certainly valuable in terms of ultimately proving

81

1 the quality of the product.

2 MR. FAMULARE: I think the question is if

3 process analytical technology gives you more

4 results, the more results you have the more

5 possibility for outliers or out of range because

6 you are looking all across the batch. Now, there

7 are two things to consider. If a field

8 investigator were coming in and saw this in a

9 process that had been validated and was

10 reproducible and was within spec under conventional

11 methods and now you are in the process of going to

12 something that gives you more information, it

13 wouldn't be appropriate at that point for the

14 investigator to see this as a violation, as a batch

15 going out of control, out of specification. It is

16 part of your process improvement process. If you

17 are seeing that, for example, at the end of the

18 batch it is becoming superpotent, and you didn't

19 know that before because of the narrower sampling

20 range, over time with this technology you should be

21 able to address that now that you are able to

22 pinpoint that information. Three years from now,

23 if we came back and you still went along the same

24 way, I mean it is just common sense that then we

25 would say, well, now that you know this why don't

82

1 you improve your process? But, obviously, the

2 initial reaction from a well controlled process

3 under conventional methods to one that now reveals

4 more results shouldn't be that this product is in

5 violation. We now know more information about it,

6 and a company that, under normal GMP processes,

7 looks at analytical results as part of its annual

8 review could examine those things that could

9 improve the process. That is the message that we

10 have to give our field investigators.

11 DR. LACHMAN: I think this is of concern

12 and it has come up again and again. I think we are

13 going to have to have some language in the guidance

14 from that point of view, that field investigators

15 would realize that this is a development phase and

16 that this is information being developed for a

17 long-term study, and it should not be considered in

18 place of the current, accepted, approved in-process

19 requirements.

20 MR. FAMULARE: It goes back to what Ajaz

21 said yesterday, that the convectional testing that

22 exist now is not being wiped off. It is not like

23 Windows 2000. You know, the old stuff is still

24 good.

25 DR. LACHMAN: But the thing is I think

83

1 people are concerned that they are going to have

2 tighter controls from a systems basis here.

3 MR. FAMULARE: Well, I think what they are

4 going to have, or the hope and the theory is that

5 they will have a more well-controlled process that

6 will eliminate recalls, out of specs and all of

7 that.

8 DR. LACHMAN: Until they get to that, if

9 an investigator comes in for an inspection, they

10 are concerned they are going to be --

11 MR. FAMULARE: If an investigator comes

12 in, as I said, this is what we have to write in the

13 guidance --

14 DR. LACHMAN: Yes, that is what I mean.

15 MR. FAMULARE: -- and this process has

16 been working for a length of time --

17 DR. LACHMAN: Right.

18 MR. FAMULARE: -- well controlled under

19 the existing paradigm. It is something in

20 development but it is not a violation.

21 DR. LACHMAN: Correct. I think that is

22 what they want.

23 DR. NASR: Ajaz just mentioned yesterday

24 something that I think is appropriate to repeat

25 today, and that is the inspection part of this

84

1 initiative will be done utilizing a team approach.

2 Doug has been working with us on the steering

3 committee at the agency. We are going to have a

4 team of reviewers and inspectors who are well

5 trained and understand these technologies. That

6 team will be utilized for inspection and review of

7 applications.

8 DR. ANDERSON: I want to submit that

9 process analytical technology is my primary method

10 for release. My method is good from 95 percent to

11 105 percent. I don't want to narrow my product

12 spec. That is going to kill me right off the bat.

13 If they think I am going to have to pull my product

14 spec down to that, it is right out.

15 The second thing is what I will do is put

16 procedures in place such that if I see a tablet

17 that is beyond my validated range I will default

18 back to traditional methodology and analyze the

19 tablet. I need language in this draft that says I

20 can proceed along those lines.

21 DR. LACHMAN: Right.

22 DR. NASR: I don't see a problem with

23 that.

24 DR. LACHMAN: Got the language?

25 DR. NASR: Well, we can put in the

85

1 language.

2 DR. ELLSWORTH: The analogy that is used

3 now is that if you have tighter specs based on a

4 PAT technology, then what you are really doing is

5 establishing alert limits. So, when you hit those

6 alert limits it means you do something else. It

7 doesn't mean the product is necessarily rejected

8 but you do something else, and maybe it is to go

9 back to some conventional testing, increased

10 testing from a conventional method to determine it

11 meets regulatory specs.

12 DR. ANDERSON: Right, and as long as there

13 is language in this draft that says using it as an

14 alert limit is an acceptable practice, that is

15 perfect. That meets our needs very, very well.

16 DR. LACHMAN: So, you have it approved and

17 then you have a range to work in to control your

18 process.

19 DR. ANDERSON: Precisely.

20 DR. NASR: So, you are talking about the

21 working range within the specification. I don't

22 see any problem with that.

23 DR. LACHMAN: That is what we have to put

24 in there.

25 MR. CHISHOLM: There is a slightly

86

1 different problem that we actually may be missing

2 here. I totally agree with all the things that

3 have just been said. One of the problems with the

4 technology is that you can't get enough outliers to

5 get across the breadth of your specification band,

6 which I think is what Jim was actually referring

7 to. That puts us in a difficult situation for new

8 products because we may have no choice but to run

9 it 97 percent to 103 percent because you can't make

10 anything outside that range when you are trying to

11 model or, if you do, you have so few examples and

12 you have no confidence levels you can justify to an

13 agency. So, there are inherent technology problems

14 in that area. How we deal with them I don't know.

15 I am just giving you the problem basically.

16 DR. NASR: And, I think recognizing this

17 was one of the reasons I made an earlier statement

18 that there will be a little bit of a distinction

19 between old versus new products. With the old

20 products we have the databases, we have the

21 validated laboratory tests that we can perform if

22 the process of the product is outside the working

23 range. But for new products we will have process

24 development but we do not have enough data points

25 if you are using calibration tablets to establish a

87

1 particular model or range to justify this. There

2 has to be some language in the guidance to utilize

3 this and to encourage the use of this.

4 DR. LACHMAN: Also, I think the current

5 FDA content uniformity or blend uniformity

6 requirements are still applicable, and those are

7 broader than you would be getting with your process

8 controls on-line. I think here, again, even for

9 new products you can use that as an alert or action

10 process for the approved specifications.

11 DR. TIMMERMANS: I think this all comes

12 back to the point I made before for having a dual

13 approach be acceptable.

14 DR. LACHMAN: Yes.

15 DR. TIMMERMANS: You need to have this

16 backup available in case you are outside the

17 calibrated range of your process analytical

18 technology.

19 DR. LACHMAN: Right.

20 DR. HALE: I think that might be the most

21 critical point of the guidance in terms of

22 implementing some of this technology, and it is

23 worth poring over that and articulating it well and

24 having it well written because none of this will

25 happen, at least on existing ones, until that

88

1 message gets out to industry, that we can have a

2 mechanism of implementing it that is not going to

3 be throwing batches away.

4 MR. ELLSWORTH: Maybe I am afraid to ask

5 this question, can you actually validate this

6 process analytical technology without creating

7 failing batches? Because that is not necessarily

8 done in current validation.

9 DR. ANDERSON: That is precisely my point.

10 I don't ever want to make failing batches, ever. I

11 am not going to be allowed to. It is just not

12 going to happen within the framework of the

13 company. The answer is yes, I can validate within

14 that narrow range and I can give you very good

15 assurance that in that 95-105 range I know

16 precisely what the values are. I can also tell you

17 when I am outside that range, but that is all I can

18 tell you. In other words, if we have a tablet come

19 through at 110 all I can tell you is that the

20 tablet strength is greater than 105. Then I have

21 to rely on some other means to give you assurance

22 that it is within the 115, the actual product spec.

23 DR. LACHMAN: I think you can do

24 statistics. There are statistics to give you a

25 measure of the probability of what percentage would

89

1 be within the tightened range, and a broad range is

2 what the official range is.

3 DR. ANDERSON: Right, but I believe the

4 nature of the question was can I provide validation

5 of my method that assures you that I do know what

6 the concentration is with only a validation set of

7 95-105. Is that a correct assessment of your

8 question?

9 MR. ELLSWORTH: Yes.

10 DR. ANDERSON: The answer is yes, we can

11 do the validation with those restrictions.

12 MS. SEKULIC: I think that defines one

13 case, the one-for-one replacement strategy. For

14 the on-line sensors, and again we have talked about

15 blending so let's stick with that one, again, shake

16 the box a little bit. So, I am starting from a

17 high variability system and measurement to a low

18 variability and tight measurement. If, by the

19 sheer differentiation of those two endpoints, you

20 consider that to be validating the methodology for

21 an on-line pair of eyes inside a vessel, the answer

22 is yes. If the insistence is on having samples

23 pulled along that time line, that is a different

24 can of worms altogether then.

25 MR. CHISHOLM: To try and give maybe a big

90

1 answer to your question, I think everybody in this

2 room would prefer to operate between 97 and 103

3 instead of 85 to 115 because that is where we

4 should be. I think that is the first point. For

5 new processes, if we get it right, we should be

6 able to control these processes much better anyway

7 and I think perhaps the industry takes a leap of

8 faith and as the confidence levels build realizes

9 it can actually operate within the tighter

10 constraints anyway. At the moment, we can, of

11 course, make designer tablets on a large scale, but

12 you can't make very many of them, and that is why I

13 use the term confidence levels all the time. When

14 you have, say, one million tablets within 97 to 103

15 and only about 10 between 103 and 115 the

16 confidence level in that area is considerably lower

17 around the specification bound. That is the

18 problem I think we have.

19 DR. TIMMERMANS: I think this is the point

20 that Sonja made before, you know, particularly

21 thinking about a lot of free approaches, and the

22 point that Carl just made. You can tell when your

23 result is different from those that you obtained to

24 establish your calibrated range or your sample set,

25 whether that is a quantitative result or whether it

91

1 is just a qualitative result. You can tell at the

2 end of a blending process, hey, this blend looks

3 different; this end result looks different than

4 what I had before. That is exactly where the alert

5 limit is a good approach. That now alerts you to

6 do something else. You know that it is different.

7 It doesn't necessarily mean that your established

8 calibration set or calibration space, let's call it

9 that, is not acceptable, not necessarily thinking

10 in quantitative terms but just in terms of the

11 calibration space. Outside of that calibration

12 space you have to do something, whatever that may

13 be.

14 MS. SEKULIC: But it does require an

15 important distinction. Okay? Because what that

16 would trigger is -- I am going to use the word but

17 very gently -- an investigation step, not an out of

18 specification investigation, which is a very

19 important distinction, but what we are trying to do

20 is to take a step further in understanding our

21 processes and our metrics. Okay? And, that is a

22 very important distinction because I think our

23 internal folks would probably have a heart attack,

24 "what do you mean, it failed?" That is exactly the

25 situation we are trying to avoid because otherwise

92

1 we will never really be in a position to keep

2 learning and incorporating this new knowledge and

3 techniques into our processes and I think it is a

4 very important distinction to make.

5 DR. TIMMERMANS: And, that is actually

6 part of the evolution of the process analytical

7 technology. If this investigation now reveals that

8 the product is, indeed, acceptable but it is just a

9 permutation of the materials that you haven't seen

10 before, you may actually consider adding that to

11 your calibration space and now your process

12 analytical method is even more robust with respect

13 to acceptable process variation.

14 DR. LACHMAN: Well, it picks up trends

15 very fast.

16 DR. TIMMERMANS: Sorry?

17 DR. LACHMAN: That will pick up the trend

18 that it is different than the prior controls that

19 you have seen.

20 DR. NASR: Do you think that this issue

21 should be addressed in the guidance, or is this an

22 issue that, depending on the drug product and the

23 nature of the process itself, should be addressed

24 by a firm?

25 DR. ANDERSON: I think this is very much

93

1 an issue that needs to be in this guidance. We can

2 provide general language that says these are

3 generically acceptable practices and, in fact, I

4 think that is something that this group must do to

5 be successful and to facilitate implementation of

6 this technology.

7 DR. NASR: So, what you are saying is that

8 the guidance should recognize the existence of

9 these issues? Or, do you have a solution that you

10 think the guidance should provide?

11 DR. ANDERSON: Well, first of all, we have

12 to recognize the situation exists, that there is a

13 limited calibration space, for lack of a better

14 word, and the guidance should allow the company to

15 be able to put in place mechanisms for dealing with

16 this in a reasonable way. I feel that both of

17 those elements need to be in the guidance and

18 specify this situation exists, and when this

19 situation exists these are possible actions a

20 company may take and still be acting in a good GMP.

21 DR. NASR: My problem is with the second

22 part. I don't have a problem with the first part,

23 that we recognize this, but I don't think the scope

24 of the guidance with be specific enough to provide

25 investigation tools and possible solutions for

94

1 every possible situation that may arise, especially

2 if the intention is to make this a very general

3 guidance.

4 DR. LACHMAN: I think if you just have

5 that the current in-process controls that are

6 approved in the application or in the compendia are

7 still applicable, and what the in-process controls

8 now are being used for inaction levels are within

9 those approved parameters. That is what it amounts

10 to really.

11 DR. ANDERSON: Right. I am not asking for

12 the guidance to have specified actions. That is

13 way beyond the scope for it, but just as Leon just

14 said, very general statements.

15 DR. LACHMAN: Yes, I think we can do that.

16 DR. HALE: But it is going to have to

17 drive changes in policies within the company or

18 allow for changes in policies so that it doesn't

19 drive inspections and things like that. That would

20 be current response right now and so it has to be

21 clear enough that it wouldn't be riding the

22 procedures but clear enough that it would allow the

23 internal procedures to change.

24 MR. OLSON: Maybe I am being naive, but

25 can't you do that now? I mean, it is your process.

95

1 Can't you set up a specification that says when

2 this product is out of specification it is a bad

3 product but I don't want to use it as a trigger so

4 I want to have a tighter specification for me to do

5 whatever it is I am going to do?

6 DR. ANDERSON: Conceptually, yes.

7 MS. SEKULIC: It can be done and it has

8 been done in various companies to varying degrees.

9 But I think if we are trying to provide a guidance

10 that is going to go much wider than that, I think

11 endorsing the fact that we are agreeing that it is

12 okay to have alert levels, alert limits and, in

13 fact, we recognize the fact that there is a

14 pre-ooze step which is okay, I think that goes a

15 long way again into the widespread use of this type

16 of technology because there is a fear out there

17 that, you know, the regulators may not necessarily

18 endorse a two-tier spec approach. So, again, we

19 are trying to dispel that particular myth from

20 impeding the implementation of these methods.

21 DR. ANDERSON: And to dispel the myth

22 explicitly, not just implicitly by not disallowing

23 it.

24 DR. LACHMAN: I think we can probably take

25 a break for twenty minutes. It is now about 10:25.

96

1 We can get back around 10:45. I think we have

2 accomplished a lot so far.

3 [Brief recess]

4 DR. LACHMAN: Let's see if can come to

5 some consensus. It is time for consensus building.

6 We have a flip chart and we have a flip chart

7 operator, and David is doing the power point. So,

8 we have a good combination.

9 Let's try and hone in on the major points

10 that we have been talking about this morning.

11 Let's get the purpose down first solidly here so

12 that we all agree to what the purpose of this

13 guidance is. We had played around with some words

14 earlier and we had expand the use of current and

15 future technology --

16 DR. NASR: We should start writing this.

17 DR. LACHMAN: Why don't we do that?

18 Expand the use of current and future technology

19 controlling batch and continuous product.

20 DR. NASR: Controlling batch?

21 DR. LACHMAN: Controlling for batch and

22 continuous product.

23 DR. NASR: For existing and new products?

24 DR. LACHMAN: I don't know if I want to

25 put that in there, for existing and new products.

97

1 I will leave it that way; it leaves it open. I

2 think it would be understood that this applies both

3 for existing and new products so I don't think we

4 have to be redundant and put it in, do we?

5 DR. NASR: I suggest we put it in.

6 DR. LACHMAN: You want to put it in?

7 Okay. For existing and new products.

8 DR. NASR: Then we can debate it and see

9 whether it needs to be in or not. Everyone in this

10 room understands that but the guidance will be

11 issued to people who have not been involved in all

12 these discussions.

13 DR. LACHMAN: Okay. Any changes we want

14 to make on the purpose of this guidance, the draft?

15 This is just the initial draft. So, it reads

16 expand the use of current and future technology for

17 controlling of both batch and continuous production

18 for existing and new products. Production of

19 future PAT technology. Anyone have any comments on

20 that?

21 DR. CHIBWE: Should we address transfer of

22 this process as well? Because once you have

23 validated the continuous process within the company

24 and you want to transfer to Puerto Rico, what are

25 you going to do? Should we address that issue in

98

1 this?

2 DR. NASR: That can be an issue but --

3 DR. LACHMAN: This is just the umbrella

4 purpose. We will get down to that specific point.

5 That is a good point.

6 Let's put another statement on, improve

7 quality assurance with PAT of a combination of

8 on-line and conventional testing. Purpose is one

9 statement. We have it up there.

10 DR. NASR: I don't see what this adds.

11 DR. LACHMAN: Well, we talked about

12 combining both conventional and on-line testing to

13 provide improved quality assurance. We may in

14 certain cases not just have one approach. We may

15 have a combination. Do we want to have that for

16 any purpose or not?

17 DR. HALE: It seems more like a motivation

18 as opposed to a purpose.

19 DR. LACHMAN: You want to leave it out?

20 Okay. Let's go to the regulatory area. I think

21 this is going to cause some discussion.

22 Requirements for accepting on-line testing in place

23 of off-line testing. What correlation needs to be

24 shown? Anybody from the regulatory side want to

25 comment on that?

99

1 DR. NASR: I think this issue needs to be

2 addressed by the guidance. There is no doubt about

3 that.

4 DR. LACHMAN: Okay, so that is affirmed.

5 That stays as such then.

6 DR. NASR: Is this the process of

7 identifying the issues that the guidance will

8 address now and this is one of those issues?

9 DR. LACHMAN: This should be one of the

10 guidance issues.

11 DR. NASR: So maybe we will write that as

12 number one and move on to a list of issues, and we

13 can debate later on which one will come out first,

14 which ones will stand and which ones will be

15 removed.

16 DR. LACHMAN: One that we had a good deal

17 of discussion about is current approved in-process

18 controls in the application.

19 DR. NASR: If I recall, Leon, I think the

20 issue was to encourage the current utilization of

21 PAT in existing products and processes, and how the

22 guidance can facilitate that.

23 DR. LACHMAN: Current in-process controls

24 approved in the application or in the compendia

25 apply while the PAT are being developed. We also

100

1 spoke about the PAT be considered as alert or

2 reaction levels and the approved specifications for

3 blend uniformly apply. The broader range and the

4 narrower range from the PAT will be alert or action

5 level.

6 MR. FAMULARE: I don't want to get ahead

7 of our ability to write but the other issue that

8 everybody said they wanted articulated in the

9 guidance was that the PAT test can replace the

10 conventional test in terms of whether it be the

11 official end process test or the official release

12 test. So we are not trying to set up more testing

13 than before, duplicative testing.

14 DR. NASR: Maybe we can say PAT on-line

15 test or on-line test to replace conventional

16 testing. That is what you had there. It is not

17 PAT test, it is on-line testing.

18 MR. ELLSWORTH: I think there is an

19 additional concept too, and that is that the PAT

20 testing endpoint can actually substitute for --

21 what is the best term to use? -- mechanical,

22 traditional endpoint which is things like time. I

23 think that is the point we were talking about, the

24 variable process endpoint rather than a fixed,

25 mechanical endpoint.

101

1 MR. FAMULARE: Along that same line, if

2 the sensors, the on-line, at-line sensors are

3 giving you indications that will allow for process

4 improvement the agency needs to allow quick

5 implementation to overcome any hurdles, and allow

6 implementation that would improve product quality.

7 MS. SEKULIC: Make me sure I understand

8 that. I am going to use that slide of the

9 excursion towards the end of the process as a

10 hypothetical example, and let's say that happened

11 in my process. Okay? I have now seen some things

12 that I haven't been able to see before. That end

13 bit is creating a problem. What I will do, I will

14 analyze, try and figure out how to implement a

15 better process without necessarily having to wait

16 for endorsement from a regulatory authority. Is

17 that the gist of it?

18 MR. FAMULARE: That is the gist of it, but

19 I think you even brought in another aspect that, if

20 we didn't capture it already on the flip chart, is

21 that the existing technology is accepted and, as we

22 capture more data through on-line or at-line

23 processing, we should allow for that data to be

24 captured and allow time for continual improvement

25 while the existing adequate for intended use

102

1 methods are still working so a company will not be

2 penalized.

3 DR. LACHMAN: So, this will be a phase-in

4 type of program.

5 DR. NASR: I think we should make this a

6 point by itself.

7 MR. FAMULARE: Right, that is what I am

8 saying. I think you brought up the two points,

9 which is good.

10 DR. ANDERSON: I think that last line

11 needs to be number seven.

12 DR. NASR: If I may give you language here

13 such as existing technologies are suitable to meet

14 current quality specs.

15 DR. LACHMAN: We may want to say that the

16 final quality attributes remain as approved. We

17 are not changing those.

18 DR. HALE: If this is an issue, the issue

19 is to provide a mechanism for implementation of PAT

20 -- these words aren't very good -- but allow time.

21 I mean, there are different ways we could do that

22 there, different components of it. There could be

23 a rapid approval process; there could be this team

24 work approach so that you come in and have some

25 pre-discussed plan in place that would allow the

103

1 data to be collected. There are different ways of

2 allowing the mechanism to unravel here. It is not

3 just allowing some grace period, but what we need

4 to do is to develop that ability to collect the

5 data and to implement.

6 DR. LACHMAN: Why don't we say provide for

7 regulatory consultation?

8 DR. HALE: But the issue is to develop the

9 mechanism to implement.

10 MS. BUHSE: And, Sonja, you had even said

11 permission to go ahead without asking for

12 permission. So, I guess those are the two things I

13 am hearing.

14 MS. SEKULIC: We are almost discussing a

15 self-assessing mechanism, but there is a natural

16 question at the end of that. We do need some

17 semblance of a formal -- you know, whether it be

18 the annual update or some connectivity to the

19 formal process as well. But the in between, the

20 interim phases appear to be more of a

21 self-assessing mechanism that we are discussing.

22 DR. LACHMAN: Right.

23 MR. CHISHOLM: I think that has to do with

24 continuous improvement. We have to remember the

25 very critical one that I think we discussed, which

104

1 was that when someone comes forward with a new

2 submission, it mustn't be affected by the fact that

3 they are going to use the new technology as opposed

4 to the three validations. That has to go up there

5 somewhere I think.

6 DR. LACHMAN: Any new introductions using

7 new technology does not impact on any products that

8 have been approved with the old technology.

9 MR. CHISHOLM: No, what I am saying is

10 that at the moment all you have to do is the three

11 validation batches.

12 DR. LACHMAN: Yes.

13 MR. CHISHOLM: With the new technology you

14 need to build up big data samples, which you can

15 only actually do in your facility. So, it is a

16 continuing learning process. We have to find some

17 way of getting around that. If that stops product

18 registration, none of this will fly because a year

19 to market could end up costing a company hundreds

20 of millions of pounds. So, it just won't fly. I

21 mean it is as simple as that.

22 DR. LACHMAN: I think there are process

23 validations, like critical utilities, that go on

24 for years but you still market the product after

25 two weeks or four weeks data. So, that should be a

105

1 similar approach, that you are developing more data

2 but that shouldn't prevent you from marketing if

3 you have enough data to meet the current standards.

4 MR. CHISHOLM: No, I don't think we would

5 have enough data to create models that would stand

6 up to scrutiny in the very early days. So, it may

7 very well be that you would have to have what we

8 call a two-tier approach. Perhaps we will have to

9 do three validations with the full knowledge that

10 that gets the product to the marketplace, but then

11 gradually we build up this knowledge and switch to

12 the new system. I don't know. I mean, I have not

13 thought about it before.

14 DR. LACHMAN: See, the thing is we don't

15 know how much reliance you can have on one batch

16 with this on-line or at-line information.

17 MR. CHISHOLM: No, you would use the

18 conventional means to do that.

19 DR. LACHMAN: Right now I don't see what

20 else you could do, except the conventional.

21 MS. SEKULIC: Would that be covered under

22 the self-assessing categories if, for example, we

23 are doing conventional testing for the filing and

24 we are doing develop work in parallel to that, and

25 then at some endpoint when you do have sufficient

106

1 at commercial scale, under the self-assessing

2 category you can say, okay, I am going to implement

3 this because I feel that I have sufficient data to

4 have the confidence in it and I will update at the

5 annual update?

6 MR. CHISHOLM: I don't have a problem with

7 that. The problem I have is with the actual

8 submission and regulatory approval because if we

9 don't have the data we can't expect the regulators

10 to approve. So, there has to be some initial

11 method of getting that product into the

12 marketplace. There has to be some mechanism of

13 getting that product into the marketplace in the

14 same time that it would for the three validation

15 batches. And, that does not exist with

16 statistically based systems because you don't have

17 the statistics.

18 DR. NASR: I have something that maybe I

19 can put here and it may address your point as well.

20 What I have is this, PAT on-line test meets

21 intended purpose and validation criteria will be

22 established for that purpose only. What I mean

23 here is that on-line test will be established for a

24 specific purpose and validation criteria will be

25 used for that purpose only. If we are using NIR-IR

107

1 -- I keep using NIR-IR and keep saying we should

2 not use NIR-IR but that is where we have more

3 information. If we have NIR-IR for content

4 uniformity the issue will be does the test meet

5 what we tried what to get out of it, rather than

6 validation of NIR-IR, calibration, and all that

7 stuff?

8 DR. LACHMAN: Let me ask you, would your

9 firm be satisfied with using the three batch

10 approval approach for marketing new products while

11 the PAT process controls are being developed?

12 MR. CHISHOLM: I don't know if I can

13 answer that, I am not the chief executive --

14 DR. LACHMAN: Well, how would you feel?

15 MR. CHISHOLM: That is the way we do it at

16 the moment. There has to be something that we can

17 do basically. I mean, if I put myself in the shoes

18 of a regulator for a minute, and this is a

19 submission stage but it could just as easily be the

20 inspection stage, if I was a regulator I would want

21 to understand the model. I would want to know the

22 probability of the model being right. I think that

23 is incredibly important to you because that

24 replaces yes or no effectively. Now, in these

25 early days the probabilities will carry very low

108

1 confidence levels and, as a regulator, I don't

2 think I could be happy with that, if you know what

3 I mean. I am not expressing it very well. So,

4 there has to be some early way of still getting the

5 product to the marketplace with due regard to

6 patent expiries, etc., etc. whilst taking account

7 of that. I don't know what it is but there has to

8 be some way. The only way I can think of is you

9 actually do the traditional way until you build up

10 your data sets. That is all I can think of.

11 DR. LACHMAN: Yes.

12 DR. TIMMERMANS: Correct, and I think that

13 is what Sonja was alluding to, and this may be a

14 question for the FDA or that this guidance needs to

15 address. If you have validated your process by

16 conventional means and you have implemented prior

17 to process validation or during process validation

18 a process analytical tool, and ultimately you want

19 to change over from conventional means of, let's

20 say, an endpoint determination to this process

21 analytical tool what is the mechanism for you to

22 receive approval to do so?

23 MR. FAMULARE: Just go ahead and do it.

24 You know, process validation has always been, at

25 least in the U.S., subsequent to approval.

109

1 Generally, what a company does is they will scale

2 up. They will do their first three batches and

3 then they will go to market. FDA doesn't review

4 that as an absolute in terms of before you go to

5 market. FDA will look at that as inspections

6 unless there is, you know, some particular reason

7 to go back there. So, if you are going to go from

8 approval to conventional validation to PAT, the

9 company should be able to do that as they do

10 validation now under GMP.

11 MR. ELLSWORTH: Under GMP rather than

12 filing. Is that what you are saying?

13 DR. TIMMERMANS: But I am thinking about

14 the analytical part. In some cases where you

15 change an analytical method you may need a CB-30.

16 Does this fall under the same category? Or if you

17 change equipment?

18 DR. LACHMAN: It would be a SUPAC.

19 DR. NASR: What I am hearing is that maybe

20 the guidance should remind people of is that

21 current GMP allow for process improvement. Let me

22 repeat this point. I think maybe the guidance

23 should remind people that current existing

24 regulations, existing good manufacturing practices

25 encourage and allow for process improvement.

110

1 DR. LACHMAN: I think the question here is

2 if you are using different methodology to measure

3 --

4 DR. NASR: For what? For process or

5 release?

6 MR. FAMULARE: I guess if the filed method

7 is, let's say, HPLC for a particular attribute and

8 now you are going to advance to the point where

9 your PAT has developed over whatever time frame

10 that you could replace the PAT inference, let's say

11 particle size, as a corollary to dissolution, the

12 question that industry is asking is can we do this

13 by a CB-30 or what?

14 MR. OLSON: You have changed your method.

15 Basically you have changed your method drastically.

16 MR. FAMULARE: Well, you see GMP though

17 has always allowed you that. Let's say you have an

18 official method, you could use an alternate method

19 as long as you had validated it against the

20 conventional method.

21 DR. LACHMAN: Right.

22 MR. FAMULARE: And, I would hope we could

23 apply the same logic here.

24 DR. NASR: We already have that as number

25 four. If you look at number four on the chart,

111

1 that is very much the same issue, isn't it?

2 MR. FAMULARE: PAT on-line can replace

3 conventional testing. I think we have to clarify

4 in the guidance --

5 DR. NASR: How we do that.

6 MR. FAMULARE: We may not have the answer

7 here but we have to clarify and encourage the way

8 that can be done under GMP, and if there is still

9 determined to be some need for filing we have to

10 make that as smooth as possible.

11 DR. LACHMAN: User friendly.

12 DR. TIMMERMANS: I think one of the

13 things, and Tom just mentioned this as well, is

14 that ultimately by changing your process or by

15 changing your process controls you may change a

16 specification or, as we said before, you may change

17 an analytical method.

18 MR. FAMULARE: If you are changing a spec,

19 then that certainly is a filing issue.

20 DR. TIMMERMANS: And we are not arguing,

21 it is just that we have to delineate what the

22 mechanism is to get that approved.

23 MR. FAMULARE: Whatever we can do under

24 the GMP program, we will do that and we will have

25 to try to delineate and make a clear path for those

112

1 folks involved with filing issues to make sure that

2 we put less encumbrance, for example, than there is

3 now, if there is any, or to make that as smooth as

4 possible under SUPAC or any CBE changes.

5 MR. CHISHOLM: I am obviously coming from

6 a bit of a position of ignorance here, never having

7 made a submission, but it would seem to me that

8 perhaps one of the ways forward is when you make a

9 submission that includes both methodologies so that

10 the agency is aware that this will be coming

11 forward. Does that make things easier in some way?

12 It shouldn't come as a shock, oh, by the way, we

13 are going to switch. I think perhaps you do a

14 filing which actually takes account of both

15 methods, with the clear intention to move when the

16 data sets are available. Is that some way forward

17 rather than this? I don't know.

18 DR. LACHMAN: The only problem I think,

19 Bob, is that you may not have developed the on-line

20 method to the extent that you would be in the filed

21 on-line procedure.

22 MR. CHISHOLM: I think we know enough

23 about the technology from solid dosage to say that

24 we would know exactly, and just because you say you

25 intend to move, it doesn't mean that you have to.

113

1 DR. TIMMERMANS: I don't know whether this

2 would potentially discourage people from

3 implementing post-filing. I would prefer not to

4 see this as I need to either get it in up front or

5 it is going to be more difficult.

6 DR. LACHMAN: I think SUPAC filing is

7 usually easier filing than up front. You may delay

8 the application approval if you put it up front.

9 MR. FAMULARE: I think as an agency we

10 have to make sure that we don't do that because,

11 certainly, as we have seen in the discussions

12 yesterday, the use of this technology in product

13 development would, hopefully, make the scale-up

14 smooth because you know more about the process.

15 Whereas it may not be fully developed for

16 full-scale processing, you want to be able to file,

17 get approval and get your product to market at the

18 same rate or better than you did under conventional

19 means. You certainly don't want it to be a slower

20 means. But, as Bob indicated, there will be some

21 body of knowledge but at the time of approval you

22 may not be able to indicate the full body of

23 knowledge, how this is going to control a fully

24 scaled-up process. That is consistent with

25 conventional validation. You are getting your

114

1 approval based on a limited batch or a limited

2 number of batches and you are going to validate

3 post-approval. Well, if you are going to do PAT

4 you are going to do it post-approval as well. So

5 there shouldn't be any added encumbrance in doing

6 that.

7 DR. ANDERSON: There is no added

8 encumbrance, but is there a benefit? Is it

9 beneficial to the company to submit both things at

10 the time of filing? In other words, to submit the

11 intention to do PAT as well as the traditional? Is

12 there a benefit to us?

13 DR. LACHMAN: Let me ask Carl, in-process

14 PAT, is that going to be used for a product release

15 or just in-process control?

16 DR. ANDERSON: Yes, I think it is

17 primarily product release.

18 DR. LACHMAN: Then you need to submit.

19 MR. FAMULARE: I think the question is

20 need versus benefit. I think the sooner the review

21 folks, and I am not a reviewer but I think the

22 sooner they know it the more beneficial it is. On

23 the other hand, if it is not filed until later, I

24 don't think it is an absolute requirement because

25 there may be a variety of reasons. Maybe a

115

1 decision was made later to go to that technology.

2 So, I think the agency should be able to

3 accommodate it either way. As far as will it be

4 beneficial, I guess the sooner that thinking is

5 brought up the more the reviewer will be attuned to

6 that as any further filing comes in.

7 DR. NASR: I think if you have the

8 information at the time of the filing and the data

9 to back it up, this will be the time to submit it

10 in your submission. If you don't, then you will

11 have to submit the new method or the on-line method

12 later on.

13 DR. LACHMAN: The other problem is when

14 you are doing new technology you are going to have

15 to get the reviewers trained in the new technology

16 if you are submitting two methods. I think that

17 element is going to be important, and how long that

18 will take--I am just wondering if it wouldn't be

19 more convenient to go by the customary method and

20 then come in with the new method when you are ready

21 to implement it.

22 MR. FAMULARE: I think Ajaz has said that,

23 you know, there is the commitment to get the people

24 trained to review that not only on the review side

25 but on the ORA field side, compliance side. So, we

116

1 have an obligation to bring our people trained.

2 So, as Ajaz said in his initial slide presentation,

3 we are gearing up for this but if you give it to us

4 now we will bring in the resources that we need to

5 deal with it.

6 DR. LACHMAN: But I think that should not

7 be part of our --

8 MR. FAMULARE: I think that is more how we

9 are going to do this.

10 DR. LACHMAN: Not only how you are going

11 to do it but make sure that it is addressed.

12 DR. NASR: Not in the guidance --

13 DR. LACHMAN: No.

14 DR. NASR: -- but internally it is being

15 considered and we are ready to move.

16 DR. HALE: I think the issue is if you

17 force a parallel path it is going to restrict

18 technology and implementation, and it will minimize

19 the leaps that you take. The mechanism starts in

20 the design phase --

21 DR. LACHMAN: Right.

22 DR. HALE: -- so if you are going to do

23 something that is different, that is not a

24 traditional test, there needs to be a mechanism for

25 an interaction perhaps. I don't know what the

117

1 mechanism is but by the time it gets to filing or

2 key decisions in the development process, that the

3 companies can make decisions whether they want to

4 pursue this or not --

5 MR. FAMULARE: Well, Ajaz and Yuan-Yuan

6 have said -- I don't know whether it was in this

7 forum but in other forums it has been clear that

8 the agency will be willing to meet with firms as to

9 what their approach will be in filing so that these

10 things could be discussed in advance as opposed a

11 filing hitting the door or the mail stop and having

12 to rethink or go back to scratch. So, the agency

13 is certainly open to that, and we will have

14 meetings with any firm that wishes to pursue this.

15 DR. LACHMAN: Any further clarification we

16 need on this subject? Sonja?

17 MS. SEKULIC: A couple of thoughts running

18 through my mind. I think I would really hate to

19 see us have to bucket things in pre-filing,

20 post-filing as method validation requirements.

21 Again, we make so many different varied products

22 and different processes and stuff that it becomes

23 really, really difficult to bucket. If you have

24 sufficient data and if you can convince yourself

25 that you have the approach on-line methods,

118

1 validation data at the time of filing, there should

2 be nothing to stop you from including that in the

3 filing pending discussions and whatever mechanisms

4 are in place to implement that methodology.

5 MR. FAMULARE: I agree. As we have said,

6 you are starting with a blank piece of paper and

7 going with this technology right from the

8 beginning. The agency needs to be open to

9 accepting it as such and to be able to evaluate it

10 but, at the same time, I think we want to be

11 accommodating for anyone who wants to mix in or use

12 existing technology.

13 MS. SEKULIC: Absolutely, and I would

14 suggest that under point ten CGMPs allow process

15 improvement, I think that really requires a few

16 more sub bullets -- what that means for us in

17 implementation of PATs. I think that is a critical

18 part of it, that we are acknowledging that this is

19 an ongoing process and if you don't have a complete

20 validation package at filing but you have done some

21 feasibility tests and you think this is the way

22 your process is going to go, it is almost a test in

23 training. You know, we have done some feasibility.

24 It looks like it might work for us. We are going

25 to not change the method at filing but, you know

119

1 what, when we have had a couple of campaigns and we

2 have a couple of hundred of experience we might

3 want to do that. So, I think that is what we are

4 looking for.

5 MR. FAMULARE: But right now you are doing

6 three batches and then going on forever with those

7 three batches, you know, regardless whether things

8 tell you rightfully or wrongfully. As this

9 technology develops, as we have said, you are

10 almost validation each batch. Again, validation is

11 in the realm of GMPs so we have to really think

12 about doing it that way. There are times where you

13 might, as has been suggested, change a spec. But

14 if you are going along with improving and

15 maintaining your operating principles it should

16 fall within GMP and the firm should be able to

17 implement that.

18 DR. NASR: I think what you are

19 suggesting, Sonja, is that when I suggested ten is

20 because my discussion was with my colleagues and

21 industry and I get a sense that they have a more

22 conservative interpretation of current GMP practice

23 maybe than the agency. Maybe because of internal

24 issues you have with your own firms -- it is not my

25 job to get into that -- but when I suggested number

120

1 ten here, I think what you are saying is put more

2 beef into it to make it clear to you and who you

3 deal with within your firm that you are allowed to

4 do that.

5 MS. SEKULIC: I think that is a very

6 important point.

7 DR. LACHMAN: I think the leadership at

8 the FDA may recognize this but when we get down to

9 the field that is not always disseminated formally.

10 MR. FAMULARE: I will let Doug answer

11 that.

12 [Laughter]

13 Again, we are going to have to be very

14 clear with our field force and they are going to

15 have to get the training, as Ajaz mentioned. We

16 have already done preliminary training to the

17 field, just introducing the topic; introducing how

18 it is going to be implemented. There is a monthly

19 drug investigators phone call. This has been

20 discussed for several hours on this phone call.

21 Plus, there was a presentation made at the Parklawn

22 Building --

23 DR. NASR: In January of this year.

24 MR. FAMULARE: Right, which was connected

25 to the field offices as well to keep them attuned

121

1 to what this new technology is, what it means, and

2 the beginning thinking of how we are going to have

3 to look at this regulatory-wise. So, that is

4 ongoing through this process. So, the field is in

5 from basically square one as we are going forward

6 with this.

7 MR. ELLSWORTH: Yes, there are training

8 plans specifically being developed, although we

9 have to wait until there is a guidance. But the

10 field is being sensitized to this issue. In fact,

11 Mike and I are charged with briefing the entire

12 senior management, all the district directors,

13 office directors on this technology and where we

14 are going. So, we are taking steps to make sure

15 the field is fully informed.

16 DR. LACHMAN: Good.

17 MR. FAMULARE: Actually, Leon, if they

18 don't, these are the two gentlemen responsible.

19 [Laughter]

20 DR. NASR: I want to raise another issue.

21 I just want to stir the pot a little bit. I think

22 we have covered several good points today. I think

23 our discussion has been very fruitful and very

24 useful. However, the more I think about it, I

25 think we are covering some common areas that our

122

1 colleagues in other discussion groups are doing now

2 about benefits, submissions, applications,

3 usefulness, encouragement and so forth.

4 I want to raise a question. Are we

5 missing some critical validation issues that should

6 be listed or utilized in the guidance? Because the

7 task before us is to very much look at validation

8 that the guidance needs to address.

9 DR. HALE: I think there are some that

10 come to mind and these may be specific. One issue

11 is if we drift into continuous processes, the idea

12 of batch size becomes interesting that we would

13 have to deal with, both in terms of stopping and

14 starting a batch and also in the middle. If we are

15 continuously measuring and we have good product for

16 a while and then, all of a sudden, we don't have

17 good product can we throw that away? I think there

18 are some of those types of issues that will arise

19 by implementation of this because we would have

20 positive unit dose size sampling methodologies that

21 could be used to release.

22 DR. NASR: Can you assist us by going over

23 some of these validation issues so we can have

24 them?

25 DR. HALE: Sure.

123

1 DR. NASR: When you talk about validation

2 issues is it as they pertain to continuous process

3 versus batch process?

4 DR. HALE: Yes, the effect of PAT and

5 continuous processing on definition of batch size,

6 and impact of the partial batch issues and starting

7 and stopping.

8 DR. NASR: So, it is continuous process

9 validation and batch size, and number of batches.

10 DR. LACHMAN: I think you also have the

11 feedback mechanism for continuous processing.

12 DR. HALE: And there would be the impact

13 of defining how process set points are interpreted

14 in validation because in feedback loops they are

15 treated differently; how process control

16 definitions change in the new environment, and how

17 you relate that to validation and then release.

18 MR. OLSON: How about if you wanted to,

19 say, replace an analytical test at the end with a

20 process technology? For example, the blending

21 issue and you are using NIR-infrared and you want

22 to replace your content uniformity, the question is

23 how do you validate that? I can see somebody

24 saying I am going to run the USP test and I am

25 going to do my ten tablets. You know, when the

124

1 signal is flat on the mixer I send it to the lab

2 three times and, you know, the ten tablets are

3 within spec. But we all recognize the weakness of

4 ten tablets. So, the question then would be how

5 would you validate that? What kind of test would

6 you do? How many tablets would you do? Would you

7 do tablets? If you have seen those curves that

8 kind of go to a flat line, do you have to show on

9 the curve where it actually crosses over some type

10 of peak of a violative product versus non-violative

11 product? I mean, there are a lot of questions

12 about how to validate that and I think you can come

13 up with as many schemes as there are people in a

14 room. Then the question would be, as an agency,

15 what would we want to see?

16 DR. ANDERSON: If you are trying to get as

17 specific as to say what tests should and shouldn't

18 be done, I am afraid that is probably beyond the

19 scope of this guidance. If you are looking for the

20 types of language to put in the guidance, I think

21 we need to talk about things like validate

22 appropriate parameters, and maybe we can list

23 examples of parameters, such as accuracy in your

24 blending case. You would need to demonstrate what

25 it is responding to. In the case of moisture, you

125

1 would need to demonstrate that you are, indeed,

2 responding to water. But to say you need to do

3 these tests would be very difficult. Hopefully,

4 our chemometric friends will provide us with some

5 insight as to more specifics on how to do the

6 validation, but I think it is beyond the scope

7 here. But we do need to try and delineate what

8 general types of things need to be done.

9 DR. NASR: I don't want to get into

10 chemometrics, but I think chemometrics and what

11 chemometric tools you use will have a major impact

12 on how we validate the process and our ability to

13 replace traditional laboratory tests by on-line

14 tests. Even though that is not part of our

15 discussion here, maybe we should raise it as an

16 issue when Leon makes the presentation in case it

17 is not captured by other groups. Because I

18 consider chemometrics and statistical tools to be

19 very critical if we want to proceed with these

20 technologies.

21 DR. HALE: You could also put in there the

22 integration of individual unit operations into

23 larger process steps, which I think would take into

24 account that particular issue of looking at

25 blending as a function of blending or looking at

126

1 blending as a function of some tablet property. If

2 you could integrate those not only in the process

3 control scheme but also in a validation scheme,

4 that would make a lot more sense. That would be at

5 least a new interpretation that you wouldn't have

6 to analyze each unit operation as an individual

7 step. So, you could group them together as an

8 integrated step.

9 DR. LACHMAN: You are going to have

10 integration of processing steps potentially.

11 DR. HALE: I would say integration of unit

12 operations, individual unit operations into a

13 process step, the allowance of multiple unit

14 operations in a validation component.

15 DR. LACHMAN: You know, when we talk about

16 chemometrics we are really talking about treatment

17 of the data that is coming out of the in-process

18 control devices, whatever sensing devices we are

19 using. It is really applying math and statistics

20 to the data. Essentially, this is what you are

21 doing on a feedback mechanism to control the

22 process, and how you do that is going to be

23 dependent, I think, somewhat on the sensing devices

24 you are using for whatever parameters you are

25 following.

127

1 DR. NASR: I think the difference is even

2 though we would all argue that existing computer

3 validation criteria are very stringent, this is

4 just a walk in the park when you look at some

5 chemometric tools that, without using, we will not

6 be able to use some of these new technologies. I

7 think we have to admit that. The sooner we come to

8 grips with that, the better the chance of us moving

9 into the direction of utilizing these technologies.

10 The agency will have to have that internal debate

11 and come to grips with the technologies out there,

12 and I think the guidance needs to address that

13 whether it is part of the validation of

14 chemometrics, or whatever.

15 DR. HALE: That could be a huge thing, or

16 you could limit that argument, in this case, that

17 we need to address the impact on process control,

18 computer validation issues with expanding sensors

19 and expanding chemometrics. So, it doesn't get

20 into all the other computer issues but is

21 specifically limited to process control.

22 DR. NASR: Let me throw out a thought here

23 when it comes to chemometrics and statistical

24 tools, validation and so forth. I think a point

25 was made yesterday that maybe we should treat this

128

1 no differently in our degree of confidence and

2 acceptance of software that is being marketed

3 nowadays if there is validation that has been done

4 by a third body, if you wish, and there is

5 assurance of the accuracy of the results coming out

6 of this, maybe we can accept that, most of it or

7 all of it and move on, rather than require

8 stringent validation and so forth. I don't know

9 how you guys feel about this.

10 MS. SEKULIC: I think it is going to be

11 very important to make it exactly that. I think we

12 have this habit of putting up new phrases and new

13 terminology to mean, you know, mathematical

14 treatment of a raw data signal into a metric. That

15 is really all we are saying, and we shouldn't treat

16 the area of chemometrics any differently from any

17 of the other broad data to metric calculating

18 mechanisms, whatever they might be. I mean, you

19 know, a pH meter isn't automatically -- the raw

20 signal isn't pH numbers. So, why are we treating

21 this field as something so different? I think we

22 need to keep it in perspective. There are

23 legitimate concerns for making sure the algorithms

24 do repeatedly generate the same numbers each and

25 every time. I mean, that is a given with any

129

1 software validation requirement. So, we accept

2 that. But let's not put any additional caveats to

3 it because I really don't see the need, quite

4 frankly.

5 DR. NASR: Sonja, would you like the

6 guidance to address that?

7 MS. SEKULIC: I think some comment. You

8 know, we understand the software validation aspects

9 that are required of any other software system but

10 we don't necessarily feel that any additional

11 attention or validation would be required.

12 DR. LACHMAN: I think any acceptable

13 validation.

14 MS. SEKULIC: Yes. I think method of

15 validation too will convince us that, indeed, we

16 are getting the same number. So, we have the

17 software validation aspect as well as method

18 validation. If we put the two together I think we

19 are kind of covered.

20 DR. NASR: Yes, you holistic have

21 validation of the whole process.

22 DR. LACHMAN: Then you have the CFR 21,

23 Part 11.

24 DR. ANDERSON: Another issue that might be

25 relevant to list here is that these process

130

1 analytical technologies are more likely than other

2 methods to require updating. That is something

3 that we might want to mention in the guidance, that

4 periodic change control and updating of the methods

5 is frequently necessary.

6 DR. LACHMAN: Is that going to be a

7 revalidation process at that time?

8 DR. ANDERSON: I don't know. I mean, the

9 short answer is yes, it is, in essence, a

10 revalidation.

11 DR. LACHMAN: That is what I would think.

12 DR. ANDERSON: It is two parts though.

13 One, it is the monitoring of the method to verify

14 that it continues to perform as intended then, two,

15 when it is necessary to change, demonstration that

16 the change is appropriate.

17 DR. LACHMAN: Correct.

18 DR. TIMMERMANS: But revalidation doesn't

19 necessarily mean, in my mind, the same full

20 validation that you would perform at the onset of

21 the development of the calibration or the

22 methodology.

23 DR. LACHMAN: I agree.

24 DR. TIMMERMANS: Just to keep that in

25 mind.

131

1 DR. LACHMAN: I think revalidation

2 generally doesn't mean the whole gamut.

3 DR. HALE: So the issue is really change

4 control.

5 DR. LACHMAN: Yes, which is needed now

6 anyway for validating systems. So, the change

7 control still applies here as well.

8 MR. DRENNAN: I think this is essentially

9 a calibration transfer type of problem, thinking

10 about calibration transfer in a very broad sense.

11 What do we do to modify that calibration when some

12 component fails or some system process parameter

13 changes, how do we update that calibration to make

14 a new appropriate measurement?

15 DR. NASR: As an extension of that, I

16 think this is a validation issue and, to some

17 extent, is unique to these technologies. If you

18 are doing titration, a burette is a burette is a

19 burette, but if you are using a spectroscopic

20 imaging device and then you go and use another

21 device with another software, the method transfer

22 aspect is going to pose a very challenging

23 validation issue, and I think the guidance should

24 allow or at least recognize this as a fact because

25 you are not going to have the same device, the same

132

1 sensor with the same software when you move the

2 method from one factory floor to another, and so

3 forth.

4 DR. LACHMAN: This brings up a good point

5 because a question was raised earlier, and we put

6 it aside, about tech transfer, what if you want to

7 transfer this technology from one site to another?

8 That was your question, I believe.

9 DR. CHIBWE: That was my question because

10 if you validate a continuous process rather than a

11 batch process and you are going to transfer -- and

12 some of these products are pretty high throughput

13 and there is a lot of demand for the product -- and

14 you want a plant, for instance, in Puerto Rico,

15 what do you do? I mean, that is a point I thought

16 we should address.

17 DR. LACHMAN: Sonja?

18 MS. SEKULIC: Can we maybe learn from our

19 existing processes? What do we do with our

20 chromatographic systems? We have system

21 suitabilities. We have method transfer exercises.

22 We have method equivalence. Probably as developers

23 of these techniques, technologies and methods we

24 are probably best positioned to try and figure out

25 what that transfer exercise, what the metric

133

1 requirement would be. For example, if you do want

2 to transfer to Puerto Rico and somebody is going to

3 be buying a new piece of equipment in Puerto Rico

4 you probably, as the originator of the method,

5 would need to identify what the system suitability

6 for that would look like.

7 DR. LACHMAN: Yes.

8 DR. CHIBWE: My point was are we going to

9 do anything different from what we traditionally do

10 for tech transfer. That is really the point I was

11 asking.

12 MS. SEKULIC: Do we need to?

13 DR. LACHMAN: Why don't we treat this like

14 we treat any validation method or process transfer

15 currently?

16 DR. ANDERSON: I agree but it should be

17 spelled out explicitly that we are going to treat

18 this as we treat any other validation method.

19 DR. LACHMAN: Okay.

20 DR. NASR: As we treat other validation

21 methods. I don't think we should use the word

22 "any."

23 DR. HALE: The only caveat is if we change

24 the ideas of three lot validation, if you look at

25 SUPACs, in the procedures we do we may not do the

134

1 same type of specific validation.

2 DR. LACHMAN: I think that is where the

3 agency has to be understanding of this process.

4 MS. SEKULIC: I think it was mentioned

5 pretty early on in the discussions about, you know,

6 if a new technology walks through the door tomorrow

7 how would we approach it in terms of understanding

8 validation, etc., etc., and we talked about

9 identifying sources of significant variability --

10 DR. LACHMAN: True.

11 MS. SEKULIC: -- and I think that is the

12 way you treat it if you identify what the sources

13 of variability that you need to control. There

14 will be some that you don't have to worry about

15 because they are not pertinent to what you are

16 trying to do, but the ones that are identified as

17 being crucial sources of variability that require

18 control would be the ones that you would then have

19 to focus your method on development, validation and

20 your tech transfer exercise would really be

21 predefined by that activity. So, that means that

22 perhaps what we need to do is say that really any

23 new technology -- what would we need? We would

24 need to identify sources of significant variability

25 and that really forms the basis of going forward

135

1 with the validation exercise, tech transfer, so on

2 and so forth.

3 DR. LACHMAN: Theoretically, for that

4 process we should have done that earlier. So, we

5 just use those concerns.

6 MR. CHISHOLM: I think you have to be a

7 little bit careful here because when you are

8 transferring to one facility into another you don't

9 actually really know what is going to happen

10 because you are going to get different process

11 signature, certainly, when you are using

12 spectroscopy. You are probably expanding your data

13 set to take account of that, so you are actually

14 back into the modeling phase and you really have to

15 go through the whole damned rigmarole again of

16 model validation and then running your process. I

17 mean, that is the straight answer to it. When you

18 are working with methylate, just putting it through

19 a different process room with almost identical

20 equipment you get different process signatures, for

21 instance, so you can imagine transferring from

22 Germany to Puerto Rico is going to have some

23 difficulties. So, whether you like it or not, you

24 are into remodeling and revalidation in these sort

25 of areas if you are using spectroscopy with

136

1 chemometric modeling. It is exactly the same as

2 the calibration transfer; it is just a bigger set

3 of the problem.

4 DR. TIMMERMANS: I think the point here is

5 that there no difference than with the process

6 analytical technology than how you are doing it

7 now. There is really no difference. Your

8 suggestion is that that be spelled out specifically

9 to clarify that point.

10 DR. LACHMAN: I think we had some points

11 raised earlier where the continuous monitoring

12 on-line or at-line provides you a tighter range of

13 the process capability and, all of a sudden, you

14 get batches outside that new process control

15 limits. However, they are within the application

16 approved limits and it has no impact and it has no

17 impact on product quality and product release, but

18 it does provide an indication to the firm that

19 there is some trend taking place that is different

20 than previously and you address it accordingly. I

21 think that is what we agreed upon.

22 DR. HALE: Does that mean that we need a

23 mechanism for internal updates of in-process

24 control, and that can be developed within the

25 confines of the product specifications?

137

1 DR. LACHMAN: Yes, we have agreed to that,

2 that this is a tighter control. It is for your

3 internal use to further control your process when

4 you see some kind of trend developing that is

5 beyond normal.

6 MS. SEKULIC: I think an important

7 distinction to make is the out-of-trend as opposed

8 to out-of-specification terminology.

9 DR. LACHMAN: Yes. Those are pretty much

10 the points that I had noted. Does anybody else

11 have any additional points to put in as items for

12 the body of the document? Anything from the

13 regulatory reviewers or compliance? Any comments?

14 DR. FRANKEWICH: My name is Ray

15 Frankewich. I am a reviewer, a CMC reviewer with

16 FDA. I am also chairman of the analytical

17 methodology technical committee, which is

18 responsible for interpreting review policy

19 regarding analytical methods.

20 I have been listening to the discussion

21 here and there are a couple of things that I think

22 I might be able to add or clarify. There was some

23 discussion about what parameters need to be

24 validated for a new process analytical test

25 replacing a conventional test. I just wanted to

138

1 say that there are guidances that address those for

2 analytical methods. You have ICH Q(2)(a) and

3 Q(2)(b) and there is another guidance, and FDA

4 guidance for validation of analytical methods in

5 draft form right now.

6 I was going to say that it depends on what

7 the test is meant for. Those guidances define four

8 different categories of test, ID assay,

9 quantitative impurities, etc. and there is a

10 discussion about content uniformity tests, most of

11 which are assays. So, you basically take the assay

12 parameters or the assay profile and apply it to

13 what you test you have and validate it according to

14 the assay parameters. I was going to say that the

15 guidance you are writing may want to reference

16 those other guidances as far as guidance on how to

17 validate the test you are developing.

18 There was also a discussion about

19 revalidation for new process analytical technology

20 methods. I wanted to say that my committee is

21 currently working on a draft on analytical

22 procedures post-approval changes guidance. That is

23 in draft form right now. Basically, it would be

24 clarifying filing requirements and possibly

25 submission requirements for changes to analytical

139

1 procedures. This guidance primarily is going to be

2 referencing the ICH Q(2)(a), (2)(b) and the 1999

3 changes guidances, changes to an approved NDA.

4 I was going to suggest that we may want to

5 work together on the draft of this guidance because

6 basically I have written most of the draft and I

7 have some ideas about how to look at an analytical

8 method.

9 DR. NASR: Ray, the guidance will be

10 drafted by us and, obviously, your committee will

11 be consulted on this.

12 DR. FRANKEWICH: Right, okay. Just a

13 couple more things. There was a discussion about

14 filing requirements for the implementation of PAT

15 methods. I was just going to give a reference to

16 the changes guidance because, basically, the

17 changes guidance pretty much delineates what the

18 final requirements for changes in analytical

19 methods are for release test as well as for things

20 like in-process tests, for tests for raw materials,

21 so on and so forth.

22 DR. NASR: Ray, could you please forward

23 some of these comments to me?

24 DR. FRANKEWICH: Sure, yes.

25 DR. NASR: Because in the steering

140

1 committee we will be drafting this, and these are

2 all excellent comments. Obviously, we need to

3 reference other guidances that the agency issued or

4 is in the process of issuing. So, please forward

5 these comments and I am sure that you will be

6 consulted.

7 DR. FRANKEWICH: All right. That is

8 pretty much what I had to say. I just wanted to

9 say that in the course of writing this guidance you

10 may want to consider references to some of these

11 other guidances because they do make it pretty

12 clear what you are supposed to do. Thank you.

13 DR. LACHMAN: Thank you very much.

14 Anything different from the LDG reviewers?

15 DR. D'SA: I am Abi D'sa, with the Office

16 of Compliance. I have a question. I have been

17 introduced to NIR-IR. I just wanted to know how

18 difficult is it to fix a model for your process

19 using just, for example, NIR-IR? Are there a lot

20 of variabilities that can affect the modeling

21 aspect? Because I see some people trying to patent

22 certain modeling of their process.

23 MS. SEKULIC: I can't really comment on

24 the patenting side of things. How difficult is it

25 to model your process? It can be very simple and

141

1 straightforward; it can be very complicated. It

2 depends on the process and how many sources of

3 variability you have and how many sources of

4 variability you are trying to incorporate into that

5 model. It can be as simple as a UV absorbance or

6 as complicated as -- I don't know.

7 DR. D'SA: Is there some expectation from

8 the modeling that needs to be spelled out in this

9 guidance?

10 MS. SEKULIC: As far as software

11 validation discussions, I think we have already

12 covered those. I think we have come to an

13 agreement that the algorithms need to perform

14 consistently in translating raw data into criteria

15 or the final sort of metric that will be used. It

16 requires consistency in performance. So, I am not

17 sure whether you are referring to selection of

18 algorithms to be used in particular situations.

19 DR. D'SA: Yes, is there some care that

20 needs to be taken in the selection of algorithms?

21 MS. SEKULIC: Yes, definitely.

22 DR. D'SA: So would that be spelled out in

23 this guidance, or at least should people be

24 alerted?

25 MS. SEKULIC: It depends on how

142

1 prescriptive you would like the guidance to be.

2 Again, I think this came up yesterday, you know,

3 how much do we want to guide the individuals? For

4 example, do we want to be as specific as saying you

5 will use cluster analysis for all raw materials

6 identification? I think that level of regulation

7 would deter in the general applicability of these

8 types of techniques.

9 DR. LACHMAN: I don't know whether we want

10 that in this particular guidance, which will be a

11 general guidance. If we want to get into more

12 specifics later, that can follow.

13 DR. D'SA: But could a bad algorithm lead

14 to product failure?

15 DR. ANDERSON: If you have chosen an

16 inappropriate algorithm, an algorithm that is

17 responsive to a parameter that you are actually not

18 measuring, yes, that could cause an incorrect

19 measurement. It could either falsely fail a

20 product or falsely pass a product. So, the answer

21 to the question of is the algorithm selection

22 critical, the answer is yes. As a matter of fact,

23 if you look at the validations that are done to

24 demonstrate these methods are appropriate, they are

25 very much centered around demonstrating that the

143

1 algorithm itself is the appropriate means of

2 analyzing the data. That is at the heart of these

3 validations.

4 DR. D'SA: So, shouldn't the guidance kind

5 of shed some light on the selection of these

6 algorithms?

7 DR. ANDERSON: I think it is beyond the

8 scope of this guidance to try and prescribe the

9 types of algorithms.

10 MR. FAMULARE: Maybe as a general

11 principle as opposed to the actual details.

12 DR. D'SA: Yes, as a general principle

13 because if it is critical to product quality in the

14 end --

15 MS. SEKULIC: Just to mention, there is a

16 chemometric session that is occurring parallel to

17 this and, again to reinforce that yesterday that

18 question did come up where we talked about off-load

19 charting the decision-making process regarding what

20 mathematics and what algorithms you ended up

21 employing in your eventual method. So, I am hoping

22 the clarification regarding that issue will come

23 from the chemometric session. If not, then

24 certainly it would have been captured yesterday in

25 the discussions.

144

1 DR. LACHMAN: Thank you.

2 MR. CHISHOLM: This is a question actually

3 more than an answer, is it not true to say that

4 because we then validate the model, perhaps against

5 analytical results or whatever method it takes,

6 that you will find whether or not you have an

7 inappropriate algorithm and, therefore, the system

8 becomes fail-safe because I would get concerned if

9 it wasn't.

10 DR. HALE: I think it would also fall

11 under the concept of the computer validation

12 discussion that we had. We have models now. We

13 have PID loops that control processes and those are

14 all models that we have. We just have the

15 propensity for having more complicated models that

16 might be harder to do, but it is the same thing.

17 MS. SEKULIC: Yes, I think what threw me

18 for a second was the use of the term "bad

19 algorithm." That, to me, means an inexperienced

20 programmer that, you know, didn't put a semicolon

21 in the right place. Whereas, what I think you were

22 actually referring to is the right algorithm for

23 the intended use. So, there is a little bit of a

24 distinction there.

25 DR. LACHMAN: Algorithm selection really.

145

1 DR. CHIBWE: One of the points that I

2 wanted to bring up we have already covered in item

3 number 11, what is going to happen with the

4 chemometrics? If you are going to change a

5 different supplier for your raw material you are

6 going to need to update your algorithm. So, there

7 is going to be a more frequent change of your

8 methods. Considering this is a generic guidance, I

9 don't think we need to be specific about NIR or

10 some other chemometric method.

11 DR. LACHMAN: Right. Anyone else from the

12 regulatory group?

13 DR. TIMMERMANS: I just want to make one

14 comment on that. I don't know whether it is

15 necessary to update your model as you need to

16 verify that your model is still correct.

17 DR. LACHMAN: Yes, a new source. Anything

18 else from this group that we want to put together

19 here? I think we have done a phenomenal job in a

20 short period of time. I want to thank everybody.

21 It is around twelve o'clock and people are getting

22 hungry. We are going to have a presentation of the

23 four groups, starting at 1:30, so we will be

24 summarizing and then have further interaction.

25 DR. HALE: Do you know whether there will

146

1 be a process of digestion? You know, this is

2 stream of consciousness talking a little bit, but

3 before June will this go around in circles a bit to

4 update it and optimize it?

5 DR. LACHMAN: Could you tell us the

6 sequence after this meeting? What is going to

7 happen with this information? How is it going to

8 be fed back to us before June?

9 DR. NASR: The way I understand the

10 process is this. Let's start with today rather

11 than June. I think you will make a presentation,

12 Leon, summarizing the discussion that we had.

13 DR. LACHMAN: We are talking about after

14 that.

15 DR. NASR: Right, once we have this

16 discussion, then the agency has formed a steering

17 committee and three members of the steering

18 committee are sitting here. Joe, Doug Ellsworth

19 and myself, along with three others. The steering

20 committee is responsible for getting this

21 information, digesting the information and then

22 addressing critical issues that will be in the

23 guidance. Once the guidance or a draft guidance is

24 ready, it will be ready for discussion when we meet

25 in June. Are you guys going to have that

147

1 information before June? I hope so. I hope that

2 the working groups will have an earlier version of

3 this draft in preparation of discussions in June.

4 DR. LACHMAN: I think some of the group

5 here would like to have a chance to review the

6 comments before the June meeting.

7 MR. FAMULARE: I think you are going to be

8 working many nights and weekends, Moheb. That is

9 the general understanding. We are going to try and

10 bring this in, all these various aspects, and the

11 steering committee will bring all this information

12 internally within FDA to try and help. It is our

13 job as the steering committee to make sure that, as

14 drafts come forward internally addressing all these

15 different issues in the guidance, they are

16 consistent with everything that was brought forward

17 in terms of ideas here today.

18 DR. NASR: I know you will be eager to

19 receive an earlier version of the draft but, at the

20 same time, if you receive a premature version where

21 the agency is not in agreement, that is not going

22 to be helpful either. So, as soon as we have

23 consensus within the agency about how to address

24 some of these issues, we will share them with you

25 as soon as we can in order to have feedback prior

148

1 to the June meeting.

2 DR. HALE: This was a few hours of

3 discussion opposed to the June time frame. There

4 may be issues that come up that we didn't happen to

5 think about.

6 DR. NASR: As a member of the

7 subcommittee, Tom, you may suggest in the afternoon

8 session that the findings and reports of these

9 individual working groups be shared with everyone

10 else within the working groups and the subcommittee

11 members. I think it would be nice to have all this

12 information and that will assist in future

13 discussions.

14 DR. LACHMAN: The meeting is adjourned for

15 lunch.

16 [Whereupon, the proceedings were recessed

17 at 12:15 p.m., to reconvene at 1:35 p.m. in joint

18 working groups]

19 - - -