Sample Plan Justification and Control Strategy
Implementation for Continuous Manufacturing
Kelly A. Swinney, Ph.D.
Vertex Pharmaceuticals, Inc.
AAPS 2015
Vertex’s Commitment to CM
2
Drivers include: 1. High quality, consistent product
2. Early finalization of formulation/process on commercial scale
3. “Data rich” QbD commercial design space
Implementing CM across development
portfolio
CM Rig is a comprehensive
equipment train: blending to film
coating
CM Rig is designed with PAT to allow for
IPC and Real Time Release Testing
Continuous Manufacturing Control Strategy and PAT
3
Control strategy is based on a QbD approach to process development and design space definition at commercial scale
PAT can be employed for process control, in-process control measurements, monitoring of design space, and release (RTRT)
PAT includes spectroscopic and non-spectroscopic technology
©2012 Vertex Pharmaceuticals Incorporated 4 4 ©2012 Vertex Pharmaceuticals Incorporated .
Powder In
Day 1 AM
Day 1 AM Coated Tablets Out
Vertex’s Continuous
Manufacturing Rig
Much Smaller Footprint
• Smaller scale equipment
• All unit ops in one facility
PAT Based Control Strategy => IPCs + RTRT
4
Tablet Press
PFLS
W Th H
PAT 5 b
Waste
STEP 6 , COMPRESSION
Fluid Bed Dryer
drying air in
drying air out
Granulation liquid 1
PAT 4 LHP
100 L
1000 L 1000 L 1000 L HC
10 L
Mill
PAT 3 b
PAT 3 a LHP 10 L
100 L 100 L
10 L
Deduster
MC
Bucket Lift
STEP 5 , FEEDER / BLENDER 2 STEP 1 , FEEDER / BLENDER 1
STEP 2 , TWIN SCREW GRANULATOR
PAT 1 NIR , measures material attributes during
screening of raw materials
PAT 2 , NIR , measures composition and BU
PAT 3 a , NIR , measures granule uniformity , LOD ,
solid state form and physical attributes of
granules
PAT 3 b , Laser Diffraction , measures particle size
distribution
PAT 4 , NIR , measures composition and BU
PAT 5 a , Raman , measures assay and CU
PAT 5 b , Weight , hardness , thickness
PAT 6 , Raman , measures coat thickness
STEP 7 , COATING
STEP 3 , FLUIDIZED BED DRYER
STEP 4 , MILLING
Coating Liquid
Finished Product
STEP 1 , FEEDER / BLENDER 1
PAT 5 a
• Tablet Tester
• Raman
• NIR
• Laser Diffraction
• NIR
PAT Locations Available for IPC and RTRT Measurements
5
• Loss In Weight
Feeders
• Dryer Thermocouple
5
Components of a Continuous Manufacturing Control Strategy
6
0
25
50
75
10
0
12
5
15
0
17
5
20
0
22
5
25
0
8 5
9 0
9 5
1 0 0
1 0 5
1 1 0
1 1 5
P ro d u c t K e y
IP C L im it
IP C L im it
A c tio n L im it
A c tio n L im it
Process Control System for Comprehensive Process and
Product Monitoring
7
• Control software displays all active unit operations and alarms on HMI
• IPCs, CPPs, and PPs with design space limits are monitored in real time
• Process control charts available for operator review
• Material is tracked through the system and segregated if manufactured outside of
the design space or if it does not conform to IPC acceptance criteria
• Batch summary reports: material tracking, alarms, IPCs, CPPs and PPs with
design spaces, etc. available for batch review
Example In-Process Controls for a Continuous Wet
Granulation Manufacturing Process
Step In-Process Control
IG Blending LIW feeder mass flow
Drying Granule water content
EG Blending LIW feeder mass flow
Final blend potency
Compression Core tablet weight, thickness, hardness
• IPCs: similar to what would be
expected for a batch process
• For CM: PAT employed for IPC
measurements in real time
8
At what frequency should IPCs be assessed for CM?
Sampling Plan for In-Process Controls
• Sampling plan should be defined for each IPC – Needs to take into account equipment capability
– Needs to take into account for short term & long term process variation
and process drift
– Minimum sampling that must be met for each IPC should be defined and
statistically justified
• Actual IPC sampling achieved: NLT minimum required sampling
• IPC sampling achieved >> than for typical batch process
9
Things to Consider – Sampling and Process Control
• CM, in general, does not permit reprocessing
– Non-conformance to IPC acceptance criteria results in removal of the
material from the process
– Operator / control actions limits should be set within the IPC acceptance limit
range to ensure that IPC acceptance criteria are met
– Action limits should be taken into account in defining and justifying the
sampling plan
10
Material Flow Rate (Line Rate/Throughput) and Sampling
Advantages to defining the sampling
plan as a function of mass
• Sampling frequency automatically
adjusts with material flow rate (line
rate)
• Regardless of mass flow, the same
number of samples/mass is targeted
– Sample frequency increases with
increasing material flow rate
– Sample frequency decreases with
decreasing material flow rate
11
Sampling
Point
Line Rate 10 kg/hr
PK 1
PK 2
PK 4
PK 3
PK 1
PK 2
PK 4
PK 3
30 kg/hr
Time Triggered
Sampling Frequency
PK 1
PK 2
PK 4
PK 3
PK 1
PK 2
PK 4
PK 3
10 kg/hr 30 kg/hr
Mass Triggered
Product Key (PK): unit of mass manufactured (~1 kg) and tracked through the system
Statistical Analysis for Sampling Plan Definition and
Justification
12
• How much material can be processed without exceeding a predefined
step change threshold ?
– Step change threshold change that is acceptable between IPC measurements
• Lag analysis of material attribute data generated during continuous
manufacture
– Capable of capturing short term variation, long term variation, and process drifts not
centered around target
– Calculate “Lag k Difference” for the material attribute for each of a select number of
continuously manufactured batches for lags ranging from k=1 to k=n
– The “Lag k Difference” is defined as the difference between the material attribute
results for PKs that are spaced k PKs apart
Lag k Difference Number of PKs Between the
Analyzed PKs
Lag 1 0
Lag 2 1
Lag 3 2
Lag 4 3
Lag 5 4
Lag 6 5
Lag 7 6
Statistical Analysis for Sampling Plan Definition and
Justification
13
– Pool the “lag K difference” results for each k obtained from the batches
analyzed and calculate the lower and upper 95% limits
– Determine the number of consecutively manufactured PKs that can be
manufactured between sampling points without exceeding the step change
threshold
• Defines the minimum sampling frequency
• E.g. Step change threshold = 3.5; NMT 3 consecutively manufactured
PKs can exist between sampling points (PKs) without exceeding step
change threshold
Lag k Difference Number of PKs Between
the Analyzed PKs
Lower 95% Limit Upper 95% Limit
Lag 1 0 -1.74 1.38
Lag 2 1 -2.16 1.42
Lag 3 2 -2.62 1.57
Lag 4 3 -3.34 1.78
Lag 5 4 -3.84 1.68
Lag 6 5 -4.08 1.74
Lag 7 6 -4.09 1.65
Action Limits
14
• Action Limits:
• Set inside the IPC limits and NLT the step change threshold from the IPC limit
• Used to alert the operator adjustment to the process is needed
• Ensures the process remains in control and that non-conforming PKs are
identified with NLT 97.5% confidence
0
25
50
75
10
0
12
5
15
0
17
5
20
0
22
5
25
0
8 5
9 0
9 5
1 0 0
1 0 5
1 1 0
1 1 5
P ro d u c t K e y
IP C L im it
IP C L im it
A c tio n L im it
A c tio n L im it
NLT step change threshold
NLT step change threshold
Things to Consider
• NIR systems used for IPCs
• The performance of the spectrometers should be routinely
verified during operation
– Measurement interface cleaned, the spectrometer re-zeroed, and
the baseline verified approximately every couple hours
15
When an NIR system is undergoing its on-going performance
verification, it is not available to perform IPC measurements.
How can compliance with the sampling plan be ensured?
Redundancy
16
• Engineering Definition
– The inclusion of extra components
that are not strictly necessary to
functioning, in case of failure in other
components
• Examples
– Back-up controls, radio links, engines
and power systems in airplanes
– Extra hard drives and power supplies
in computers used in datacenters
Example In-Process Controls for a Continuous Wet Granulation
Manufacturing Process - Redundant IPC measurements
17
Step In-Process Control Method(s)
IG Blending LIW Feeder mass flow Gravimetric
Drying Granule water content Granule discharge temperature
NIR (milled granules)
EG Blending LIW Feeder mass flow Gravimetric
Final Blend Potency NIR (final blend)
NIR (milled granules) + EG LIW feeders
Compression Core tablet weight thickness, hardness PAT 5 (Kraemer)
• Method redundancy is built into the IPC strategy for CM
− However, only one method is required per IPC for manufacture
• Ensures continuous monitoring of the process
− Coverage when NIR systems are undergoing on-going performance verification
• Intention is to operate with both methods with a target sampling plan of every PK
− If discrepancies between results arise, the non-conforming result will be acted on
Decision Tree for IPC Test Results
18
Decision Tree for IPC Test Results
Redundant
IPC Result 1
Redundant
IPC Result 2Conformance
Does not
Conform
Pass
Pass
Fail
Not
Available
or None
Identified
Fail
Not
Available
or None
Identified
Not
Available
Not
Available
or None
Identified
Investigate
Pass Pass
Conforms
Conforms
Does not
Conform
Things to Consider – Failure to Meet IPC Sampling
Criteria
• Need to have an system in place to identify in real time if
the sampling criteria has not been met
– This can be a combination of procedural controls and automated
processes
– Operator should be alerted
• Impacted material should be down streamed processed to a
segregation point and removed from the process
– Procedures should be in place to investigate the event and evaluate
the impacted material
Data used to define and justify the sampling plan can be
leveraged for assessing material with “no IPC data”
19
Investigating material with “No IPC Data”
• “Lag k difference” analysis results can be leveraged during the
investigation to assess with high confidence whether or not the
material conforms to the IPC acceptance criteria
20
Lag k Difference Number of PKs Between
the Analyzed PKs
Lower 95% Limit Upper 95% Limit
Lag 1 0 -1.74 1.38
Lag 2 1 -2.16 1.42
Lag 3 2 -2.62 1.57
Lag 4 3 -3.34 1.78
Lag 5 4 -3.84 1.68
Lag 6 5 -4.08 1.74
Lag 7 6 -4.09 1.65
Number of Contiguous PKs with No IPC Results Maximum 95% Limit
1 2.16
2 2.62
3 3.34
4 3.84
5 4.08
6 4.09
Investigating material with “No IPC Data”
• In the absence of an IPC result, evaluate the IPC data obtained before and after the
absence of data
• Use the “lag K difference” analysis results to determine with NLT 97.5% confidence whether
the material conforms or does not conform to the IPC acceptance criteria
• Downstream process the material if it is confirmed that the material also conforms to all other
IPC acceptance criteria and was manufactured within the design space
21
Number of Contiguous PKs with No IPC Results Maximum 95% Limit
1 2.16
2 2.62
3 3.34
4 3.84
5 4.08
6 4.09
PK7 PK10
Example of RTRT for Continuous Manufacturing
22
• A composite sample will be taken after tablet printing for retains, stability, and any
release testing not amenable to RTRT (e.g. appearance)
• Traditional end product testing performed for appearance
Critical Quality Attribute
PAT Technology RTRT Method Material
Identification Raman Confirm presence of API Core Tablet
Assay NIR
Weight
API Content
Tablet Weight
Final Blend
Core Tablet
Dissolution
Laser Diffraction
NIR
WTH
Granule Particle Size
API Content, Water Content
Tablet Weight, Hardness, Thickness
Milled Granules
Final Blend
Core Tablet
Water Content NIR Water Content Final Blend
Content Uniformity
NIR
Weight
Variance in API Content
Variance in Tablet weight
Final Blend
Core Tablet
Physical Form Raman API physical form Core Tablet
PAT Methods Sample a Large Percentage of the Batch
23
Analysis of a hand full of tablets by Regulatory QC Methods
vs
PAT data collected repeatedly during manufacture
How much sampling is required for RTRT to appropriately
characterize the batch?
T im e
% T
arg
et
A P I C o n te n t - F in a l B le n d (% T a rg e t )
T a b le t W e ig h t (% T a rg e t )
U p p e r L im it
L o w e r L im it
23
Target Sampling Rate Per PAT Measurement
• A target sampling rate is defined for each PAT measurement used in a
RTRT calculation for an end product CQA
24
PAT Technology Material Measured Attribute Target Rate
NIR (PAT 3) Milled granules
Water content in granules
DS content in granules
Once per minute
Laser Diffraction (PAT 3) Milled granules Particle size distribution Once per minute
NIR (PAT 4) Final blend DS content
Water Content Once per minute
Raman (PAT 5) Core tablet DS identification
DS physical form Once per 2 minutes
WTH (PAT 5) Core tablet
Weight
Thickness
Hardness
Once per 2 minutes
Example RTRT Minimal Sample Size
• A minimum sample size is defined for each PAT point used in a RTRT calculation
• Achieved sample size is a metric that can be trended as a performance indicator
25
Critical Quality Attribute
PAT Technology In-Process Material Attribute Minimum Sample Size
Identification Raman API identification in core tablets NLT 10 Tablets
Assay NIR Final blend API content NLT 50%*
Weight Core tablet weight NLT 50%*
Dissolution
Laser Diffraction Milled granule particle size distribution NLT 50%*
NIR Final blend API and water content NLT 50% *
WTH Core tablet weight, thickness, hardness NLT 50%*
Water Content NIR Final blend water content NLT 50%*
Content Uniformity
NIR Final blend API content NLT 50%*
Weight Core tablet weight NLT 50%*
Physical Form Raman API physical form in core tablets NLT 10 Tablets
NValid Results = Number of valid results obtained
NTarget Results = Theoretical number of results at the target measurement rate
𝐴𝑐𝑡𝑢𝑎𝑙 𝑆𝑎𝑚𝑝𝑙𝑒 𝑆𝑖𝑧𝑒 = 𝑁𝑉𝑎𝑙𝑖𝑑 𝑅𝑒𝑠𝑢𝑙𝑡𝑠
𝑁𝑇𝑎𝑟𝑔𝑒𝑡 𝑅𝑒𝑠𝑢𝑙𝑡𝑠× 100
Actual Sample Size for Assay, Dissolution, Water Content, and Content Uniformity
* NLT 50% target sample size achieved + NMT 15% in a contiguous segment not sampled
Statistical Analysis for Justifying Minimum Sample Size for
RTRT Calculation
26
Bootstrap analysis to demonstrate that there is no appreciable difference in the attribute
result calculated when 100% target sampling is achieved versus the minimum
• Bootstrap Analysis • Randomly select a 15% contiguous segment and remove it
from the data set
• Randomly remove 35% of the remaining data
• Calculate the mean and standard deviation
• Repeat for 100,000 iterations
• Determine the mean and upper and lower 95% limits for both
the mean and standard deviation results
• Compare the bootstrapping results (mean and standard
deviation) with the results obtained with 100% sampling (true
mean)
• Demonstrate that the difference between the 95% limits and
the true mean is also acceptable
15% contiguous segment removed
35% of remaining data removed
Batch
Mean (% Target) Standard Deviation
(% Target)
True
Mean
Minimum
Sample
Size
Absolute
Difference
True Std
Dev
Minimum
Sample
Size
Absolute
Difference
1 97.23 97.22 0.010 0.79 0.78 0.016
2 96.99 97.00 0.009 0.74 0.74 0.002
3 98.78 98.77 0.007 1.22 1.21 0.007
4 101.07 101.01 0.058 1.29 1.27 0.020
5 101.12 101.12 0.002 0.61 0.62 0.005
6 102.96 102.95 0.014 0.79 0.77 0.018
Conclusions
• CM control strategy implementation requires a sophisticated process
control system
• In-process controls require real time PAT measurements (spectroscopic
and/or non-spectroscopic) that can be acted on in real time
• Redundancy in the IPC methodology ensures continuous monitoring of
the process and quality throughout the batch
– However, process control system must also be able to assess
conformance to the sampling plan in real time and take the appropriate
actions
• Definition and justification of the IPC sampling plan requires an
approach that takes into account short term and long term process
variation as well as process drift from target
• For RTRT, the minimum sample size required to appropriately calculate
the RTRT results should be understood
– Achieved sample size is a metric that can be trended as a performance
indicator
27
Acknowledgements
28
• Team at Vertex
– Pharmaceutical Development
– Technical Operations
– Supply Chain Management
– Quality
– CMC Regulatory
– Facilities
• Equipment manufacturers
• Our CMOs, suppliers, and research collaborators
• FDA, EMA, MHRA
Top Related