Process Characterization Its Not A Science Project March 1, 2017 - - PowerPoint PPT Presentation

SLIDE 1

BioProcess Technology Consultants www.bptc.com CONFIDENTIAL

Process Characterization – It’s Not A Science Project March 1, 2017

SLIDE 2

Why Do Process Characterization? What’s the Goal?

CONFIDENTIAL 2

• Standard answers from clients
• Increase process knowledge

― A correct, if esoteric answer. It begs the question, “Then what?”

• Establish Proven Acceptable Ranges (PARs)

― Better, but for what immediate and ultimate purpose?

• Design of Experiments
• Creating mathematical models relating process variables to Critical

Quality Attributes

It’s Not A Science Project!

SLIDE 3

Why Do Process Characterization? What’s the Goal?

CONFIDENTIAL 3

• From PDA Technical Report #60: Process Validation: A Life

Cycle Approach, 2013 has the answer

• From Section 3.7

― (What PC is): “Process characterization is a set of documented studies in which operational parameters are purposefully varied to determine their effect on product quality and process performance” ― (Almost there!): “The approach uses the knowledge and information from risk assessments to determine a set of process characterization studies to examine proposed ranges and acceptance criteria” ― (Here’s the Goal or Application of PC): “The resulting information is used to define the PPQ ranges and acceptance criteria”

SLIDE 4

From BPTC Standard Templates for PPQ Protocols

CONFIDENTIAL 4

PC studies needed to ensure that operation at the NORs specified in PPQ protocol (and at values somewhat outside of the NOR to account for measurement error and possible deviations) will not impact CQAs and process performance.

SLIDE 5

From BPTC Standard Templates for PPQ Protocols

CONFIDENTIAL 5

In addition to historical process data from full scale GMP manufacturing lots, PC studies are required to specify acceptable CQA limits for PPQ campaign.

SLIDE 6

From BPTC Standard Templates for PPQ Protocols

CONFIDENTIAL 6

In addition to historical process data from full scale GMP manufacturing lots, PC studies are required to specify acceptable performance for PPQ campaign.

SLIDE 7

With Goal Now Defined, How Do We Design PC Program?

CONFIDENTIAL 7

• What variables do we study?
• Those that are at the greatest risk of screwing up your PC campaign
• What variables would those be?
• Those that were ranked the highest during the risk assessment

(e.g., the RPN score from an FMEA)

• OK, we have the variables we want to study, what ranges do

we study for each variable

• Study values a little bit outside of the Normal Operating Ranges

(NORs) to account for measurement error and deviations

• How do we determine what the NORs are?
• …….

SLIDE 8

What is an NOR and How Do We Establish Them?

CONFIDENTIAL 8

• One definition of NOR
• From TR60: “a defined range, within (or equal to) the Proven

Acceptable Range, specified in the manufacturing instructions as the target and range at which a process parameter is controlled, while producing unit operation material or final product material meeting release criteria and CQAs”

― NOR < PAR (That’s the only guidance TR60 gives)

• Not a helpful definition. For example …..
• Suppose we have data to know that a temperature between 34 and

39 oC has no impact on product quality or process performance (i.e., the PAR is 34 to 39). How does this help us define NOR?

― Is it 34 – 39? 35 – 38? 36.5 – 37.5? ??????? ― Key term in NOR is “normal”. Where does the process normally run? What is a reasonable range within which to control a process parameter regardless of PAR?

SLIDE 9

Establishing Normal Operating Ranges

CONFIDENTIAL 9

• Another, more useful, definition of NOR
• “A defined range within (or equal to) the Proven Acceptable Range,

specified in the manufacturing instructions as the target and range at which a process parameter is controlled, and that can be achieved with a high degree of assurance and reproducibility”.

• One way to establish NOR is from historical data

― Use process data from historical batches to calculate 95%/99% tolerance intervals

Example: 30 batches are shown and plotted. 95%/99% tolerance interval is 6.77 to 7.22. Thus, NOR is 6.77 – 7.22 (as long as PAR is wider).

6.65 6.75 6.85 6.95 7.05 7.15 7.25 7.35 5 10 15 20 25 30

BioRx pH or Elution pH or ??? Run # or Sample #

99%/99% 95%/99%

SLIDE 10

Establishing Normal Operating Ranges

CONFIDENTIAL 10

• Best way to establish NORs is from historical at-scale data
• Use process data from historical batches to calculate 95%/99%

tolerance intervals

• But what if only a few historical lots have been

manufactured, how does one establish NOR?

• Same idea as previous slide – propose ranges where you suspect

process can operate 90% or 95% of the time (regardless of impact

• n process or product as that will be discovered during PC)
• Idea is to minimize the need to execute process deviations and

investigations during commercial manufacturing

• Example: for process buffer pH, given the equipment, procedures

and instrument tolerances, propose a pH range for the process buffer that can be met 90% or 95% of the time (regardless of impact on process or product as that will be discovered during PC)

SLIDE 11

We’ve Established NORs: What’s Next for PC Program?

CONFIDENTIAL 11

• We know what variables we want to study
• We know the NORs for the variables
• What ranges do we study for each variable?
• Study values a little bit outside of the Normal Operating Ranges

(NORs) to account for measurement error and deviations

• How far outside the NORs should be studied in our DOEs?
• One answer: Potentially very far outside because we want to see

an impact on CQA and process attributes so that the DOE results generate a mathematical model relating process variables to …….

It’s Not A Science Project!

SLIDE 12

Mathematical Models “Don’t Do No Harm” ………

CONFIDENTIAL 12

• The need for mathematical models reminds me of a line

from the movie Western “Unforgiven”

• Gene Hackman as Little Bill Daggett

― “Look son, being a good shot, being quick with a pistol, that don't do no

harm, but it don't mean much next to being cool-headed. A man who will keep his head and not get rattled under fire, like as not, he'll kill ya”

• Having mathematical models relating process variables to CQAs and

performance attributes “don’t do no harm”, but it is NOT the point

• f the PC program.
• The point of the PC program is to:

― Provide experimental support for the NORs specified in the PPQ protocols; ― Provide some wiggle room to operate in the event the NORs are exceeded (i.e., deviations) & to account for measurement variability

SLIDE 13

We’ve Established NORs: What’s Next for PC Program?

CONFIDENTIAL 13

• So, how far outside the NORs should be studied in our DOEs
• One answer: Potentially very far outside because we want to see

an impact on CQA and process attributes so that the DOE results generate a mathematical model relating process variables to …….

• Far enough to account for measurement variability & to provide

wiggle room to operate in the event the NORs are exceeded so as to provide valuable information for investigations to deviations

• Example: If NOR for pH of column load is 6.7 to 7.3

― Explore 6.5 to 7.5 during PC program

6.7 7.3 pH NOR 6.5 7.5

Range studied for PC to establish PAR

SLIDE 14

What is an PAR and How Do We Establish Them?

CONFIDENTIAL 14

• One definition of PAR
• From TR60: “A characterized range of a process parameter for

which the operation within this range, while keeping other parameters constant, will result in producing a material meeting relevant quality criteria.”

NOR NOR Range of acceptable performance

• “…. producing

material meeting relevant quality criteria”. What is relevant quality criteria?

• Again, can rely
• n historical

performance of clinical batches

• Graph shows

historical clinical batches

SLIDE 15

Establishing Proven Acceptable Ranges

CONFIDENTIAL 15

• Example
• Characterization data shows all experiments studied met all criteria
• f acceptable performance defined as that seen in clinical batches
• Thus, PAR is the entire range studied
• Note: No mathematical models required!

― No need to look at wider ranges to see an impact on performance

Process Characterization Results

NOR PAR

Manufacturing Scale Results

NOR PAR

SLIDE 16

Establishing Proven Acceptable Ranges

CONFIDENTIAL 16

• Another Example
• Characterization data shows some experiments failed at values less

than the lowest boundary of the NOR

• Thus, PAR is range for which parameter gave acceptable results

― Data shown could be actual data or “data” obtained from a model fit

• Lower end of NOR in danger of providing unacceptable quality

― Need more data at lower end

Process Characterization Results Manufacturing Scale Results

NOR PAR

SLIDE 17

Establishing Proven Acceptable Ranges

CONFIDENTIAL 17

• Another Example
• Characterization data shows some experiments failed at values less

than the lowest boundary of the NOR

• Thus, PAR is range for which parameter gave acceptable results

― Data shown could be actual data or “data” obtained from a model fit

• Lower end of NOR in danger of providing unacceptable quality

― Need more data at lower end; seems like critical process parameter

Process Characterization Results Manufacturing Scale Results

PAR NOR

SLIDE 18

From BPTC Standard Templates for PPQ Protocols

CONFIDENTIAL 18

PC studies needed to ensure that operation at the NORs specified in PPQ protocol (and at values somewhat outside of the NOR to account for measurement error and possible deviations) will not impact CQAs and process performance. Note: Often times PARs are presented in PPQ protocols as well. PARs provide helpful documentation in the event NORs are exceeded as the PARs in the PPQ protocol can be cited during the investigation and review of the deviation.

SLIDE 19

What’s Next?

CONFIDENTIAL 19

• We know what variables we want to study
• We know the NORs for the variables
• We know the ranges we want to study for each variable
• We have a strategy for determining PARs
• Mine historical data to determine acceptable

performance

• Execute the studies: DOEs, OFATs
• Advice regarding DOE execution
• Don’t fall in love with JMP software
• JMP say tell you can study 7 variables with 8 experiments – DON”T

BELIEVE IT!

SLIDE 20

DOE Example

CONFIDENTIAL 20

• 7 variables with 8 experiments. The design is as follows:
• Suppose the average result is 10 and the first variable has a 20%

impact and the second variable has a 12% impact and the results are ideal with no variance.

Run # Var 1 Var 2 Var 3 Var 4 Var 5 Var 6 Var 7 1 1 1 1 1 1 1 1 2

• 1

1 1

• 1
• 1

1

• 1

3 1

• 1

1

• 1
• 1
• 1

1 4

• 1
• 1

1 1 1

• 1
• 1

5 1 1

• 1
• 1

1

• 1
• 1

6

• 1

1

• 1

1

• 1
• 1

1 7 1

• 1
• 1

1

• 1

1

• 1

8

• 1
• 1
• 1
• 1

1 1 1

• If process results, effect of variable 1 would be exactly 2 and the

impact of variable 3 would be exactly 1.2. All other variables are zero.

Run # Var 1 Var 2 Var 3 Var 4 Var 5 Var 6 Var 7 Result 1 1 1 1 1 1 1 1 13.2 2

• 1

1 1

• 1
• 1

1

• 1

9.2 3 1

• 1

1

• 1
• 1
• 1

1 13.2 4

• 1
• 1

1 1 1

• 1
• 1

9.2 5 1 1

• 1
• 1

1

• 1
• 1

10.8 6

• 1

1

• 1

1

• 1
• 1

1 6.8 7 1

• 1
• 1

1

• 1

1

• 1

10.8 8

• 1
• 1
• 1
• 1

1 1 1 6.8 2 1.2

SLIDE 21

DOE Example

CONFIDENTIAL 21

• 7 variables with 8 experiments. The design is as follows:
• Now assume some random variance

Run # Var 1 Var 2 Var 3 Var 4 Var 5 Var 6 Var 7 Result 1 1 1 1 1 1 1 1 13.5 2

• 1

1 1

• 1
• 1

1

• 1

9.0 3 1

• 1

1

• 1
• 1
• 1

1 13.0 4

• 1
• 1

1 1 1

• 1
• 1

9.4 5 1 1

• 1
• 1

1

• 1
• 1

11.2 6

• 1

1

• 1

1

• 1
• 1

1 6.7 7 1

• 1
• 1

1

• 1

1

• 1

10.5 8

• 1
• 1
• 1
• 1

1 1 1 6.6 2.0625 0.1125 1.2375 0.0375 0.1875 -0.088

• 0.038
• Still get essentially the same result. The most important variables

are #1 and #3. The values for the rest are due to experimental variance.

Run # Var 1 Var 2 Var 3 Var 4 Var 5 Var 6 Var 7 Result 1 1 1 1 1 1 1 1 13.2 2

• 1

1 1

• 1
• 1

1

• 1

9.2 3 1

• 1

1

• 1
• 1
• 1

1 13.2 4

• 1
• 1

1 1 1

• 1
• 1

9.2 5 1 1

• 1
• 1

1

• 1
• 1

10.8 6

• 1

1

• 1

1

• 1
• 1

1 6.8 7 1

• 1
• 1

1

• 1

1

• 1

10.8 8

• 1
• 1
• 1
• 1

1 1 1 6.8 2 1.2

SLIDE 22

DOE Example

CONFIDENTIAL 22

• 7 variables with 8 experiments. The design is as follows:
• Now suppose for 1 variable – just one – you get an unusually high

variance …….

Run # Var 1 Var 2 Var 3 Var 4 Var 5 Var 6 Var 7 Result 1 1 1 1 1 1 1 1 13.5 2

• 1

1 1

• 1
• 1

1

• 1

9.0 3 1

• 1

1

• 1
• 1
• 1

1 13.0 4

• 1
• 1

1 1 1

• 1
• 1

9.4 5 1 1

• 1
• 1

1

• 1
• 1

11.2 6

• 1

1

• 1

1

• 1
• 1

1 6.7 7 1

• 1
• 1

1

• 1

1

• 1

10.5 8

• 1
• 1
• 1
• 1

1 1 1 6.6 2.0625 0.1125 1.2375 0.0375 0.1875 -0.088

• 0.038
• Now the story becomes a little muddled: is variable #5 significant?

Are variables 2 and 4? The story becomes muddled all because of 1 bad point.

Run # Var 1 Var 2 Var 3 Var 4 Var 5 Var 6 Var 7 Result 1 1 1 1 1 1 1 1 13.5 2

• 1

1 1

• 1
• 1

1

• 1

9.0 3 1

• 1

1

• 1
• 1
• 1

1 10.1 4

• 1
• 1

1 1 1

• 1
• 1

9.4 5 1 1

• 1
• 1

1

• 1
• 1

11.2 6

• 1

1

• 1

1

• 1
• 1

1 6.7 7 1

• 1
• 1

1

• 1

1

• 1

10.5 8

• 1
• 1
• 1
• 1

1 1 1 6.6 1.7 0.475 0.875 0.4 0.55 0.275

• 0.4

SLIDE 23

DOE Advice

CONFIDENTIAL 23

• Even if only had 3 variables and 8 experiments, one bad

point can make it appear as if you have interacting effects when you may not

• Here is my advice:
• Critical: Ensure that your experimental systems and

assays are reliable before executing DOEs. One bad point can muddle your results and at worse, mislead you!!!

• Critical: Know your experimental variance before

executing DOEs so that you know what level of effects are significant compared to variance.

• Helpful: Run center-points along with the DOE as controls

to provide a check that experiment ran appropriately

SLIDE 24

DOE Advice

CONFIDENTIAL 24

• Here is my advice:
• Helpful: Run duplicates if you can. Even running

duplicates for just some of the runs can be helpful.

• Helpful: If the results indicates that you have many 2-

factor interactions or if all the effects are relatively close to one another and no one effect stands out, be suspicious that one or two bad points may have muddled the results

• Critical: Leave time available to re-run experiments or to

execute more experiments to explore a variable space with higher resolution.

SLIDE 25

Summary

CONFIDENTIAL 25

• From FMEA, determine what variables you want to study
• Prioritize variables with high RPNs
• Establish NORs for these variable from historical runs and/or

equipment tolerances

• Determine the ranges you wish to study
• Ranges a tad larger than the NORs
• Determine a strategy for establishing PARs
• Mine historical data to determine acceptable performance
• Execute the studies: DOEs, OFATs
• Don’t fall in love with JMP software
• Ensure your experimental systems are reliable & know the variance