Page 1
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Statistical Questions from CPV Monitoring of Bioreactor Data
Craig Bernier
Principal StatisticianDesign to Value and Quality Engineering
Janssen Pharmaceutical Companies of Johnson and Johnson
Page 2
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
CPV Statistics
• Much support around developing procedures, teaching control charts, and providing backup to more complex questions. As per the 2011 Guidance:
“The data should be statistically trended and reviewed by trained personnel.” ….
…“We recommend that a statistician or person with adequate training in statistical process control techniques develop the data collection plan and statistical methods and procedures used in measuring and evaluating process stability and process capability.”
1
252321191715131197531
102
100
98
96
94
92
Observation
Ind
ivid
ual
Valu
e
_X=97.21
UCL=101.50
LCL=92.92
Assay Control Chart ExampleTrending of Typical CQA of a Product
Page 3
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Common Statistical Questions for CPV
• Data failed the Normality test
– ‘Do I have to transform the Data?’
– ‘How should I transform the data?’ or ‘ What transformation do I use?’
• What Shewhart tests to apply?
• Long Term vs Short Term sd
• ‘Can I use Levey Jennings Chart’?
2
Page 4
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Common Statistical Question for CPV
• For dissolution do I just chart the averages? Or Individuals?
– For individuals do I use:
• Individuals chart?
• X-bar / R, or X-bar / S?
• Should we use the I-MR R/S (Between / Within) charts?
• Do I have to calculate Process Capability?
– How often?
– Use Cpk or Ppk?
– Confidence intervals?
3
Page 5
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Statistically Related Questions
• If I see a point outside the control limits, do I have to open an investigation?
• The control limits are to 3 decimals, but the spec is one decimal?
• The data is rounded to same precision as the specification. Do I need to ask for ‘unrounded data’?
4
Page 6
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Somewhat Statistically Related Questions
• How often do I need to trend / report?
– Why?
• How to evaluate risks, prioritize actions from CPV reports?
• Do I need to trend Yields?
• Do I need to trend
– CMA (Critical Material Attributes)?
– CPP (Critical Process Parameters)?
• Is Minitab/JMP/R ‘Validated’?
5
Page 7
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
More Complex Statistical Questions
For Example:
• How to monitor longitudinal data from the bioreactor during CPV?
– Multiple CPV parameters (e.g. Viable Cell Density, Viability, IgG content, pH)
– Daily offline measurements for each batch
• Unique data profile across culture period
• Up to 60 days depending on product
6
Page 8
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Stage 2:Production by
continuous perfusion in bioreactor
Upstream Production Process
7
Stage 1:Preculture and
expansion
continuous perfusion bioreactor
Page 9
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Bioreactor Data
8
Page 10
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Problem
• How to evaluate cell growth data over time for trending purposes.
• When is a batch acting differently than usual – what tool to use to detect special cause variability?
Page 11
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Problem Continued
• Initial solution from our practitioners.
• This type of trending shows a sample average and some expected variability around that model. Essentially mean +/-3sd at each day
• A very nice solution to be able to detect unusual results and potential special cause variability
• Some challenges though…
Page 12
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Problem Continued
• We do not have the ability to monitor the process over time as we usually desire in CPV
– Is process trending up / down?
• Another challenge is that this approach allows for multiple chances for a batch to signal.
– What to do about a batch that has a single point or a couple points outside the range but in general follows a typical pattern
Page 13
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Alternate approach
•Trending of residuals from average at each day
– This approach captures the deviations from average across time as the current approach. But also allows us to then trend the batches over time potentially using typical control chart methods.
Page 14
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Plot Individuals
Autocorrelation exists due to the nature of the data
Page 15
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Plot Averages and sd
Batch 45Batch 40Batch 35Batch 30Batch 25Batch 20Batch 15Batch 10Batch 5Batch 1
2
1
0
-1
-2
Batch
Mean
Batch 45Batch 40Batch 35Batch 30Batch 25Batch 20Batch 15Batch 10Batch 5Batch 1
2.0
1.5
1.0
0.5
0.0
Batch
StD
ev
Time Series Plot of Residuals Mean
Time Series Plot of StDev
Page 16
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Charting of Average Residuals
Batch 46Batch 41Batch 36Batch 31Batch 26Batch 21Batch 16Batch 11Batch 6Batch 1
1
0
-1
-2
Batch
Ind
ivid
ual
Valu
e
_X=0.010
UCL=1.327
LCL=-1.307
Batch 46Batch 41Batch 36Batch 31Batch 26Batch 21Batch 16Batch 11Batch 6Batch 1
1.50
1.25
1.00
0.75
0.50
Batch
Sam
ple
StD
ev
_S=0.830
UCL=1.093
LCL=0.567
1
1
1
1
1
I chart of Residuals Means
Batch 31 omitted from calculations
S Chart of SRES1
*Standard X-bar chart shows variability more than within batch variability. We use I-MR R/S type approach
Page 17
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Batches of interest identified
Control Charts signaled at Batch 12, Batch 21, and Batch 31. (as well as Batch 6)
Page 18
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Trend Data by Test Date
• Potential additional benefit of evaluating the residuals is that we could also plot the data by test date.
– Are test results unusually high / low on a particular day indicating some lab variability?
918273645546372819101
0.50
0.25
0.00
-0.25
-0.50
-0.75
-1.00
-1.25
Date
Sam
ple
Mean
__X=-0.000
UCL=0.586
LCL=-0.586
1
1
1
Xbar Chart of SRES by Test Date_Qtr 20xx
17
Page 19
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
References
Montgomery, Douglas C. Introduction to Statistical Quality Control 7th edition, Wiley, 2013.
*Note: Montgomery text page 461-495 discusses various works with some examples on monitoring autocorrelated data as well as profile monitoring.
18
Page 20
© 2017 Janssen Supply Group, LLC, and/or its affiliates. All rights reserved. Any use of this material without specific permission of Janssen is strictly prohibited.
Questions
Thanks to:
Tim Overkleeft
Wendy de Wit
Kevin Pipkins
Clemens Haerder
David Enck