Strategies for Monitoring and Troubleshooting Biopharmaceutical
Manufacturing Operations
Jack [email protected], Manufacturing Technical SupportGenzyme Corporation
Strategies for Monitoring and Troubleshooting Biopharmaceutical Manufacturing Operations
OUTLINEAssessing the Data Bottleneck for a ProcessStatistical Process Control Case StudyChromatography Troubleshooting ApproachesFive Keys to Successful Process Data Systems
Assessing the Data Bottleneck for a Process
Not Measured
Not Recorded
Process Data
Not AccessibleNot Acted On
Not Analyzed
Not Trusted
Process Understanding and Improvement
PhysicalBarriers
OrganizationalBarriers
What hinders improvement/troubleshooting most?
Process Troubleshooting Overview
Process and Data Investigation
MFG Deviation
Specification or IPC Violation
SPC Alert
Lot Disposition
Preventative Actions
Process Insight
Outlier Detection
Resolution
Process and Data Investigation
MFG Deviation
Specification or IPC Violation
SPC Alert
Lot Disposition
Preventative Actions
Process Insight
Outlier Detection
Resolution
How to Monitor Large Quantities of Data? ->Statistical Process Control Rules
Run Date
Var
iabl
e
µ -3σ
µ -2σ
µ -1σ
µ
µ +1σ
µ +2σ
µ +3σ
0.135% 2.14% 13.6% 34.1% 34.1% 13.6% 2.14% 0.135%
Western Electric Rules
1 outside 3 σ
2 out of 3 outside 2 σ
4 out of 5 outside 1 σ
8 or more same side of µ
SPC Case Study: Purification Column Recovery
Most data fall within formal alert limits
60
70
80
90
100
110
120
130
140
Date
Rec
over
y
SPC Case Study: Purification Column Recovery
60
70
80
90
100
110
120
130
140
Date
Rec
over
y Most data fell within formal alert limitsLow recovery triggered formal investigation
SPC Case Study: Purification Column Recovery
60
70
80
90
100
110
120
130
140
Date
Rec
over
y Most data fell within formal alert limitsLow recovery triggered formal investigationSPC Alert ( 8 < mean ) visible 10 runs earlier~10% of each batch went to drain for 20 batchesAllston Automated SPC System:
Immediate e-mail notificationsHyperlinked trends & lot trees
Chromatography System Overview
Pump
Column
Flowmeter
F UVpH TCond P
pH TCond P
Bubble Trap Bypass Valve
Level Sensors Bubble Trap
Vent Valve
Vent Filter
DATA • Data entry• Assay• Sampling• Sensor• Assumptions
What Can Go Wrong in Chromatography?
60 cm Chromatography Column, Skid, and Piping
EQUIPMENT • Control system• Valve leaks• Bed channeling• Bed compression• Other…
PROCESS &
MATERIALS• Mixing• Material stability• Precipitation• Parameter sensitivity• Procedural gaps• Operator errors• Raw material issues
Chromatography Investigative Options
Equipment Checks
Assay Checks Experiments
·Pressure check system·Examine valves·Check/Recalibrate sensors·Test HETP/asymmetry
·Review raw QC data·Trend assay control·Re-assay retain·Resample from process
·Buffer/Raw material testing·Scale-down experiments·Stability studies·Resin life/cleaning studies
Data Compilation·Paper batch records·Chart traces ·MRP (lot tracking & usage)·Raw material attributes
Is Data Consistent with Adjacent Operations and Batches?
Component Balances Across Adjacent
Process Operations
Historical Online Profile
Comparison
Internal Consistency
Checks
Historical Runs
Adjacent Operations Column
B
Column A
Column C
product recoveryprotein recoveryvolume balance
pHUV
pressure flow rate
conductivitytemperature
Is Data Internally Consistent?
0 10 20 30 400
5
10
15
20
25
30
35
40
y = 0.97x
R2 = 0.99
AUFC
(L)
0 50 100 150 2000
20
40
60
80
100
120
140
y = 0.78x
R2 = 0.74
Measured Total Protein (mg)
AU
C
Component Balances Across Adjacent
Process Operations
Historical Online Profile
Comparison
Internal Consistency
Checks
Elution ProfileAttributes
Total Protein vs. Area under curve
Total Flow vs. Volume
?
Systematically Choosing Investigative Steps…
SPC Alert: Unusually Low Product Recovery
Protein recovery also low?
Preceding column product recovery high?
Online data available?
YESNO
Request retest of load
YES NO
Following column product recovery high?
YES NO
Protein accounted for in
load/wash?
YES NO
Request retest of eluate
Perform experimental investigation
Check equipment
Request resample
Initial Assessment Activities
YES NO
Dig for more data
??
? ?
Allston Landing Process Data Flows
ProcessOperations
UtilitiesSIP/CIP
Media PrepBuffer PrepCell Culture
MicrofiltrationChromatographyLyophilization
Control SystemsRosemount RS3Intellution FIX
AB PLCDelta V
Manual DataQC Results
Monitoring Logs
Data HistoriansAIM
Rapid Pharma
MRPLIMS
MS Access
Manual AnalysisExcelJMP
MATLABAutomated Analysis
MATLAB Web Trends/Reports
SPC Alerts
Five Keys to Successful Data-Driven Process Improvement
#1: Create broad awareness of how the process is running
Five Keys to Successful Data-Driven Process Improvement
#2: Ensure manual data entry has both immediate and long-term benefits.
#3: Exploit Metcalfe’s Law: the value of a network grows by the square of the number of its users
Five Keys to Successful Data-Driven Process Improvement
#4: Lower barriers (activation energy) to explore ideas and confirm theories
Five Keys to Successful Data-Driven Process Improvement
#5: Create efficient cross-functional teams to drive and close-out investigation and improvement initiatives
Five Keys to Successful Data-Driven Process Improvement