1. Statistical Computation Tools in Pharmaceutical Drug Development and Manufacturing Life Cycle Fasheng Li Ke Wang Pharmaceutical Statistics Worldwide R&D, Pfizer, Inc 40th Annual MBSW Muncie, IN May 24, 2017

2. Motivation
Statisticians often accumulate software in their daily work of actively providing statistical support
Script: R, SAS, Matlab, JMP, etc.
Template: MS Excel (w/wo VBA macros) or other spread sheets
Considerations of what to do with these software pieces
Align approaches across an organization
Re-usability
Sharing
Validation
Version control
Computing resources (e.g. HPC grids) 2

3. Outline
Overview of statistical computational applications (tools) developed in support of drug development and Chemistry, Manufacturing, and Controls Regulatory Affairs (Reg CMC)
Why
When
How
Internal Web Statistical Tool Cases (R program based)
Evaluation of product stability and predicting shelf life or clinical use period
Process parameter criticality assessment
Design space exploration through designed experiment and parametric bootstrapping
Conclusions/Remarks 3

4. Statistical Computational Challenges
Statisticians in drug development and Regulatory CMC areas are facing more and more challenges:
Higher work load
Quicker turnaround
More consistency and alignment of approaches
More demanding on data modeling/mining
More system traceability/repeatability
Less man power resources 4

5. Statistical Computational Solutions
Utilize accumulated software pieces (R or other based) to develop web-based statistical computational tools
Standardized approaches
Efficient
Consistent
Centralized software
Version controlled
Easy maintenance
Validated/Verified software
Accuracy/Reproducibility
Easy to deal with regulatory queries
Utilized HPC or Cloud Computing Resource
Increased computational capability
Powerful simulation work
Much shorter computation time 5

6. Statistical Computational Tools6

7. Internal Online Application Examples7

8. Statistical Computational Tool - App 1
Evaluation of product stability and predicting shelf life or clinical use period 8
Stability

9. Statistical Computational Tool - App 1
Evaluation of product stability and predicting shelf life or clinical use period
Design and collect stability data
Predict Expected Shelf Life or Use-period
Linear Regression Model Fit 9
Stability

10. Statistical Computational Tool - App 1
An Internal Online Application was set up 10
Stability

11. Statistical Computational Tool - App 111
Stability

12. Statistical Computational Tool - App 1
Result – displayed in web browser
Result – linked downloadable PDF 12
Stability

13. Statistical Computational Tool - App 1
Evaluation of product stability and predicting shelf life or clinical use period
Align statistical analyses of stability data
Offer a quick and convenient web-based tool to generate
Scatterplots of stability data
Shelf life estimation for registration batches
Typical three batches per package configuration;
Guided by ICH Q1E
Use Period Estimation for Clinical Batches
Typically a single batch
Guided by internal SOPs
R program based
Version controlled by the platform software (EASA)
Easy to maintain for feature updates
Run on HPC or Cloud grid computers 13
Stability

14. A Statistical Tool for Assessment of Criticality of Process Parameters
Statistical Computational Tool - App 2
A Statistical Tool for Assessment of Criticality of Process Parameters 14
CPP

15. Statistical Computational Tool - App 2
A Statistical Tool (CPP) for Assessment of Criticality of Process Parameters
Ref: Wang et al., Statistical Tools to Aid in the Assessment of Critical Process Parameters. Pharmaceutical Technology (3):
Assesses Statistical and Practical Significance
Data from One-factor-at-a-time or Design of Experiments
Process:
Step 1: Process Risk Evaluation
Evaluate a multifactor data set without focusing on a single parameter
Determine how close results are to a target/specification
Z Score Assessment
Step 2: Parameter Effect Size Calculation
Uses the Fitted Statistical Model to Quantify Individual Parameter Effects
Quantifies the Effect Size - Impact of a Parameter on a Response
Compares the Effect Size to the Specification with the 20% Rule 15
CPP

16. Statistical Computational Tool - App 2
A Statistical Tool (CPP) for Assessment of Criticality of Process Parameters
Z Score Assessment
For a CQA, y, with one-sided limit, such as L, the Z score is calculated as
𝑍= |𝑚𝑒𝑎𝑛(𝑦)−𝐿| 𝑠
Decision: a threshold of 2-6 was used
If Z > 6, all of the statistical significant parameters are deemed Not Critical
If Z < 2, all of the statistical significant parameters are deemed Critical
If Z is between 2 and 6, parameter effect sizes are needed for further evaluation 16
CPP

17. Statistical Computational Tool - App 2
A Statistical Tool (CPP) for Assessment of Criticality of Process Parameters
Parameter Effect Size
The effect size (Δ) of a process parameter is defined as the maximum change of the predicted CQA caused by the process parameter change when other factors are fixed at all of their possible levels
Analytically
Numerically (by grid search in R)
Decision: the 20% Rule
If the effect size (Δ) is <= 20% of specification range, the process parameter is deemed Not Critical to the CQA
If the effect size (Δ) is > 20% of specification range, the process parameter is deemed Critical to the CQA 17
CPP

18. Statistical Computational Tool - App 2
An Internal Online Application was set up
1. DOE Data File (.csv)
2. A Model File (.csv) 18
CPP

19. Statistical Computational Tool - App 2
An Internal Online Application: Further features to allow enhancement of the analyses and outputs 19
CPP

20. Statistical Computational Tool - App 2
Result – displayed in web browser
Result – linked downloadable PDF 20
CPP

21. Statistical Computational Tool - App 2
A Statistical Tool (CPP) for Assessment of Criticality of Process Parameters 21
CPP

22. Prospective Process Reliability Estimate
Statistical Computational Tool - App 3
Prospective Process Reliability Estimate
Design space exploration through designed experiment and parametric bootstrapping 22
PPRE

23. Statistical Computational Tool - App 3
Process Prospective Reliability Estimate
Driven by Quality by Design Concept - ICH Q8-Q10
Scientific, risk-based, holistic, and proactive approach to pharmaceutical development
Deliberate design effort from product conception through commercialization
Full understanding of how product attributes and process parameters relate to product performance
Visualization of Design Space
Multidimensional – many process parameters
Multiple responses – many quality attributes with specifications
Pros/Cons of Traditional Overlay Space
Limitations
Operating region is defined on average (50% probability)
The risk along the edge of the region may not be acceptable
Limited information for decision making
E.g. want a higher probability to meet specifications
Lack of flexibility
No alternatives to mean prediction of responses/attributes
Tedious/erroneous to get a summary from the software
Advantages
Readily available in commercial software (e.g. Design Expert)
No extra programing needs 23
PPRE

24. Statistical Computational Tool - App 3
Prospective Process Reliability Estimate
For a CQA (y) with a two-sided spec (Lspec and Uspec), the probability to meet specification at 𝑋 ℎ can be estimated as
P(y >= Lspec & y <= Uspec |  , 𝛴 , 𝑋 ℎ )
For multiple CQAs, the confirmatory rate can be calculated as the probability to simultaneously meet all specifications at given levels of 𝑋 ℎ
Ref: Peterson et al. A Bayesian reliability approach to multiple response optimization with seemingly unrelated regression models. Quality Technology & Quantitative Management (4): 24
PPRE

25. Statistical Computational Tool - App 3
Prospective Process Reliability Estimate
Pros/Cons of Traditional Overlay Space
Limitations
Operating region is defined on average (50% probability)
The risk along the edge of the region may not be acceptable
Limited information for decision making
E.g. want a higher probability to meet specifications
Lack of flexibility
No alternatives to mean prediction of responses/attributes
Tedious/erroneous to get a summary from the software
Advantages
Readily available in commercial software (e.g. Design Expert)
No extra programing needs
Parallel computing algorithm implemented in the R code 25
PPRE

26. Statistical Computational Tool - App 3
An Internal Online Application was set up
1. DOE Data File (.csv)
2. A Model File (.csv) 26 26
PPRE

27. Statistical Computational Tool - App 3
An Internal Online Application: Further features to allow choose of analyses and enhance outputs
Result – Displayed in browser 27
PPRE

28. Summary
Statisticians (and Scientists) in drug development and Reg CMC areas can perform routine statistical analyses with
Increased consistency
Improved efficiency
Better alignment of statistical analyses
Easily retrieved results By
Standardizing statistical approaches
Centralizing software pieces
Validating/Verifying software pieces
Utilizing high performance computer resource
Deploying web-based applications

29. Acknowledgment
Kim Vukovinsky, Senior Director, Pharmaceutical Statistics, Worldwide R&D, Pfizer Inc. 29