Best Practices in Statistical Process Monitoring ... - IBC Life Sciences

IBC Process2Product Oct 4 th , 2011
Best Practices in Statistical Process
Monitoring of Biopharmaceutical
Manufacturing Operations
Amer
Pompe disease
Jack Prior
Sr. Director, BioProcess Engineering
Technical Operations

BPOG Out of Trend/Statistical Process Control
Working Group
● Steve Jones (BPOG)
● Bill Henry (GSK)
● Kevin Legg (Genzyme)
● Brian Stamper (MedImmune)
● Janet Alvarado (Merck)
● Monica Jungen (Merck-Serono)
● Paul McCormac (Pfizer)
● Christoph Hoh (Sanofi)
● Dave Hopkins (Genzyme)
● Jarrod Medeiros (Merck-Serono)
● Michael Warncke (Bayer)
● Cenk Undey (Amgen)

Origins:
BPOG Data Analysis Working Group
● BioPhorum Operations Group (BPOG)
– 13 companies & ~300 participants
– Collaborating on biopharmaceutical MFG challenges
– “Point Share” organized on Data Analysis (October 2010, Boston, MA)
– Case studies/discussion: process monitoring, platforms/tools, organizations
● Point Share Observations
– Struggling with similar challenges
– Diverse approaches in place and contemplated
– Companies in process of defining SPC practices & standards
● Follow-up Focus on SPC and OOT
– Teleconferences
– Internal surveys
– Best practice sharing
– How can we summarize….Process2Product

Converging Drivers for Formal Statistical Process
Monitoring
Business
Drivers
Technical
Support
Mission
cGMP
Requirements
Lean
Avoid Surprises
Preempt OOS
Global supply chains
CMO/tech transfers
Biosimilars
QbD
Understand
Control Improve
Continuous
Process
Verification
ICH
Q10
Cost
2011

SPC Driver - FDA Validation Guidance:
Continued Process Verification
● Continually assure the commercial process remains in a state of control
● Establish ongoing program to analyze product and process data
Recommends…
● Data statistically trended and reviewed by trained personnel
● Statisticians/equivalent with adequate SPC training develop plans
● Quality unit review
http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM070336.pdf

SPC Driver - ICH Q10:
Pharmaceutical Quality System
● “…plan and execute system for monitoring of process performance and
product quality…”
● “Analyze parameters and attributes identified in control strategy to verify
continued operation within a state of control”
● “…continued capability of processes and controls…”
● “…data management and statistical tools”
● “…identify sources of variation for continuous improvement”
● “…enrich the design space”

Goals of Effective SPC Program
● Efficient and proactive management of process
o Enhance quality compliance and process understanding
o Preempt deviations, don’t generate them
● Balanced focus on process parameters and analytics
o Focus effort on critical areas
o Reduce effort on non-critical items
“Have the appropriate people looking at the appropriate data
analyzed in the appropriate way responding in the
appropriate manner”, Kevin Legg, Genzyme

Key Questions Raised in Working Group
on SPC implementation Strategy
1. How to decide what to formally trend
2. How to respond to OOT Signals
3. How to manage OOT Response
4. How to define “out of trend”
5. How to set (and reset) limits and on what reference data
6. How to deal with “unusual” baseline data
7. How to choose what statistical rules to apply
8. How to learn more

1. How to Choose which Variables to Trend
● Define Process Control Strategy (PCS) 1
– Break process into defined steps with defined purposes (the “why” vs. “how”)
– Classify Process Control Elements (CPP, KPP, CQA, CI, PI…)
● PCS shapes monitoring strategy & signal response
– Formally trend/monitor CQAs (IPCs, ISs, Specs) and CPPs
– Set proportionate formality/response for process and consistency indicators
1
A planned set of controls, derived from
current product and process understanding
that assures process performance and
product quality. (ICH Q10).
Paul MaCormac, Pfizer

2. How to Respond to OOT Signals:
Specification, Action, and Statistical Control Limits
Specification Limit: “Voice of Customer”
Set based on process
capability, regulatory
expectations and
product safety concerns
Action Limits:
PV Acceptance Criteria,
Batch record limits
Statistical Control Limit: “Voice of Process”
+/- 3 for n> 15 and/or PpK ≥ 1.33
Set based on
process characterization
and/or large scale data
Paul MaCormac

2. How to Respond to OOT Signals:
Potential Map from PCS Classification to Signal Response
Variable
Acceptable
Range
Normal Operating
Range
Statistical Control
Limits
Attributes
CQAs*
Specification/IS
In-Process Control (IPC)
Consistency Indicator
(later)
Process Indicator
(real-time)
Critical
Deviation
Major
Deviation
deviation
N/A
N/A
Technical review
DCS Alarm
OOT/ APR Review
Immediate alert
OOT / APR Review
Immediate alert
Technical review
Immediate alert
Aggregate as CI
Potential MVA**
Parameters
Critical Process
Parameter (CPP)*
Key process parameter
(highly controlled)
Non-KPP(nKPP)
(no step impact)
Major
Deviation
deviation
DCS Alarm
DCS Alarm
Technical review
Potential MVA
Technical review
Potential MVA
*Defined in ICH Q8
**Multivariate Analysis

3. How to Manage OOT Response
Survey comments
● QC Lab
– If no confirmed lab error, QC triggers deviation
– If lab error, initial result invalidated after new result obtained
– Lab error sometimes only detected in CI calculation (e.g. yield outliers)
● Process Parameters & Consistency Indicators
– Manufacturing, technical, or (sometimes) validation responsibility
– Immediately enter in quality system OR trigger discussion
– OOT below a set “performance level” can be agreed as non-concern
– If uncovered during APR, investigation may be launched
● Frequency of Data Review
– Automate/alert after each discrete observation if possible
– Cross-functional notification can facilitate resolution
– Periodic data reviews for continuous data (e.g. run reviews)
– Formal evaluation may not occur until APR

3. How to Manage OOT Response
Team Structure
● Basis for Extent of Investigation
– Risk assessment & variable classification
– Potential impact to process performance and product quality
– Upfront work in Process Control Strategy (PCS) guides actions

4. How to Define “Out of Trend”
● Variation not explained by a single distribution of process output 1
● Violation of the Western Electric or Nelson trending rules 2
● Result from a process not associated with random variation
● Visual review .. .outside 3 …discretion of lab management
● Can vary by site or group within site (i.e. QC vs. manufacturing)
…Consensus on +/- 3 limits… but from what data set
1. MJ Kiemele et al. Basic Statistics: Tools for Continuous Improvement.4 th Edition
2. NIST/SEMATECH e-Handbook of Statistical Methods, http://www.itl.nist.gov/div898/handbook /
04MAY2011

5. How and When to Define Control Limits
● Choose right reference data set…
– After ~15-30 batches from stable process (Shewhart recommends 25+)
– Use interim provisional limits from process validation criteria
– Use largest dataset representative of current process & testing
● Exclude Outliers…
– Remove known root (special) causes
– Use objective outlier exclusion methods
– Use judgement -> tighter limits (conservative, defendable, appropriate)
– Show outliers even if excluded from calculations
● Revisit/adapt when needed…
– When known process or analytical change impacts mean or variability
– As APR recommendation or in change control
– Can shift mean but retain long term variance

6. How to Handle Unusual Data
Situation
“out of control”
historical data
“non-normal”
historical data
LoQ assays
(e.g. LaL, purity)
Strategies/Comments
• Exclude special cause variability
• Incorporate all common cause variability
• Short horizon reference range often too narrow
• Utilize moving range for control limit calculations.
• 3 limits often apply 1
• Avoid complex WE/Nelson rules
• Transform data prior to applying rules
• Visually inspect trends instead
• If sufficient data is available >LoQ, exclude

7. When & How to use Western Electric/Nelson Rules
Detecting Common vs. Special Cause Variation
Short Horizon
Common Cause
• Assay noise
• Raw materials
• Procedural noise
Long Horizon
Common/Special Cause
(Not visible in 25 points)
• Lab transfers
• Reagent lots
• Personnel
• Column packing
• Upstream productivity
• Plant utilization
Special Cause
• Operator errors
• Equipment failures
• Sample handling
• Data typos/calibration
• Material instability
● Data rarely “independent and identically distributed”
● Real processes trend/move - resulting in false signals
● 3 and 2/3 outside 2 rules OK - complex rules often inappropriate
● Backward looking rules impact earlier lots
● Trend signals slow in bio environment (e.g. 6 batches in a row)
● Value of well-trained eyes on right parameters

8. How to Learn More
● Regulatory Guidance
– ICH Q9
– International Conference on Harmonization (ICH) Q10: Section 3.2
– ICH Q11 (Draft): Section 3.2, Section 6.1
– New Process Validation Guidance
http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM070336.pdf
● Statistical Process Control
– DS Chambers, D Wheeler, Understanding Statistical Process Control
– DC Montgomery, Introduction to statistical quality control. 6 th Ed.
– MJ Kiemele et al, Basic Statistics: Tools for Continuous Improvement. 4th Ed.
– NIST/SEMATECH e-Handbook of Statistical Methods,
http://www.itl.nist.gov/div898/handbook / 04MAY2011
– William A. Levinson, Statistical Process Control for Real-World Applications, CRC Press

Summary:
Key Success Factors in SPC Implementation
● Infrastructure
– Data management to enable trending and ensure data integrity
– Appropriate state of control to enable decisions based on trended data
● Quality Systems & Business Processes
– Proper OOT definition, notification methods, and proportional response
– Defined responsible parties to ensure timely response to OOT
● Process Control Strategy
– Appropriate risk-based classification of parameters and attributes
– Statistical tools/methods to ensure right limits and rules applied
– Focus on preventing deviations
– Knowledge capture