Impurity Monitoring in a Pharmaceutical Process

1
Impurity Monitoring in a Pharmaceutical Process

• Dr Elaine Martin and Professor Julian Morris
• Centre for Process Analytics and Control
  Technology (CPACT)
• University of Newcastle

2
Overview of Presentation

• Initial Analysis of Data from a Drug Intermediate
• Objectives
• Process Description
• Multivariate Statistical Process Performance
  Monitoring
• Multi-way Principal Component Analysis
• Generic Models
• Conclusions

3
Objectives of the Study

• To understand the factors that have the greatest
  effect on impurity formation thereby enabling the
  Company to only analyse those batches that are
  close to the recommended operating levels.
• To demonstrate the value of multivariate
  statistical techniques for data interrogation and
  process performance monitoring in batch
  pharmaceutical operations.

4
Schematic of Reactor Vessel
5
Time Series Profiles
Addition 2
Addition 1
Addition 2
6
Process Description

• Two additions are made during the batch.
• Reactant A slurry of Reactant B in ethyl
  acetate
• Reactant C above mixture
• The reaction is exothermic.

7
The Data

• Twenty nine batches were made available.
• Nine batches produced in a Stainless Steel
  reactor.
• Twenty batches produced in a Hastelloy reactor.
• The rationale for changing to the Hastelloy
  reactor was to increase the cooling capacity and
  reduce batch duration and impurity levels.
• The final set of 10 Hastelloy batches were
  produced with a larger batch size, i.e. larger
  volumes of reactant A were added to larger
  volumes of reactant B.
• The larger cooling capacity of the vessel however
  meant that these runs would not necessarily be
  longer.

8
The Data

• Five process variables - reaction temperature,
  refrigerant flow, jacket inlet temperature,
  jacket outlet temperature, agitator speed, are
  recorded on a 30 second basis.
• Two quality parameters - yield and amount of
  impurity are monitored at the end of the batch.
• According to the specifications, failed batches
  were identified as those that contained
  impurities greater than 0.14.
• However in reality a good batch is one that
  contains less than 0.1 impurity.
• Longer addition times for stainless steel batches
  may contribute to higher impurity levels

9
Impurity Levels
Hastelloy Batches
10
Limitations of Univariate Process Performance
Monitoring
99 Action Limits
7
7
3
3
2
2
Univariate in-spec zone
Acceptable range for variable 1
5
5
6
6
1
1
4
4
Variable 1
1
2
3
Multivariate in-spec zone
4
5
6
Acceptable range for variable 2
7
Variable 2
11
Multivariate Statistical Process Performance
Monitoring

• One possible approach to achieving a deeper
  understanding of the process has been through the
  application of multivariate statistical
  techniques.
• Multivariate Statistical Process Performance
  Monitoring is based on the statistical projection
  techniques of Principal Component Analysis (PCA)
  and Projection to Latent Structures (PLS).
• PCA monitors the process through a single block
  of information - the process or quality
  variables.
• PLS monitors the process through a model of the
  quality variables / chemical information
  developed from the process information.

12
Multi-way Principal Component Analysis
13
Results of Multi-way Principal Component Analysis
14
Impurity Levels
Batches 27, 28, 29
15
Batch 29 - Variables 1 to 5

• Variable 3 the main cause of separation of
  Hastelloy batches.

16
Variable 3 - Stirrer Speed

• Higher stirrer speeds observed for batches 27-29.

17
Principal Component Analysis

• PCA was applied to the 20 Hastelloy batches.

18
Impurity Levels
25 (0.11)
14 (0.09)
19
Scores Contribution Plot

• Variable contributions to PC2
• Normal Batch
• Batch 14

• Variable contributions to PC2
• Abnormal Batch
• Batch 25

20
Hastelloy Batches 25 and 14
25 (0.11)
25 (0.11)
14 (0.09)
14 (0.09)
Jacket Outlet Temperature
Jacket Inlet Temperature
Batch 25 (Solid line) Batch 14 (
Dotted line)
21
Hastelloy Batches 25 and 14
25 (0.11)
25 (0.11)
14 (0.09)
14 (0.09)
Jacket Coolant Flow
Reactor Temperature
Batch 25 (Solid line) Batch 14 (Dotted
line)
22
Interpretation of the Change in Process Operation

• The jacket inlet temperature and jacket outlet
  temperature of batch 25 (0.11 impurity) starts
  to deviate from the trajectory of batch 14 during
  the first addition.
• The reactor temperature for batch 25 is seen to
  be much less well controlled during the first
  addition.
• For batch 14 (0.09 impurity), the reactor
  temperature is better controlled with a more
  constant jacket inlet temperature.

23
Interpretation of the Change in Process Operation

• A characteristic of batch operation is the
  closing of the coolant control valve at this
  time, allowing the reactor temperature to rise
  rapidly to its desired 21ºC.
• Stopping the coolant flow results in the jacket
  coolant temperature measurements being unreliable
  during this period.
• Following a period of erratic coolant flow
  conditions, good control is regained in batch 14
  in contrast to that in batch 25.
• This results in a clear difference in the
  responses of the jacket temperatures between the
  two batches in the subsequent four hour stir.

24
Interpretation of the Change in Process Operation

• During the stir, the reactor temperature of batch
  25 is higher than the required desired 21ºC.
• Discussions with plant personnel highlighted
  issues relating to other plant coolant demands
  and coolant control issues as possible causes for
  increased impurity levels.

25
Multi-Group (Generic) Model

• Products manufactured infrequently / small
  amounts.
• Impractical to generate a model for every product
  type.
• Method for monitoring a number of different
  processes using a single model.
• Different product types.
• Different ways of manufacturing a product.

26
Generic Modelling

• Weighted average of individual covariance
  matrices.
• Covariance matrices of each group.
• Pooled covariance matrix.
• Common principal component loadings.

27
Generic Modelling

• Model generated from 20 Hastelloy batches (2
  group model).
• Movement of high speed batches towards the
  centre.

Batch 25
Batch 26
28
Impurity Levels
25 (0.11)
26 (0.07)
29
Scores Contribution Plots

• Tracing the cause of abnormal batches 25
  (impurity level 0.11) and 26 (impurity level
  0.07) from the contributions to PC2.
• Variables 1 and 2 have the highest contributions.

30
Cause of Abnormality
Batch 25 - impurity level 0.11
Batch 26 - impurity level 0.07
Batch 25
Batch 26
Batch 25
Batch 26
Jacket Inlet Temperature
Jacket Outlet Temperature
31
Interpretation of the Change in Process Operation

• A deviation in the jacket inlet temperature, for
  batch 25, can be seen during the first addition.
  
• Also despite the inlet temperature being
  reasonably well controlled until the second
  addition, the jacket outlet temperature for batch
  25 drifts upwards.
• Following the second addition, the jacket
  temperatures are kept much lower for the low
  impurity batch (batch 26).
• The multi-group study confirmed the previous
  findings using standard multi-way PCA and data
  augmentation.

32
Conclusions

• Multi-way principal component analysis was
  capable of identifying batches with high and
  low levels of impurity.
• Poor reactor temperature control during the first
  addition appears to lead to increased impurity
  levels. In addition poor regulation at the end of
  the batch also results in temperature offsets.
• Generic modelling helped remove differences
  caused by different reactor speeds.
• The applicability and power of multivariate
  statistical techniques have been clearly
  demonstrated.

33
Acknowledgements

• Lane, S., Martin, E. B., Kooijmans, R. and
  Morris, A. J. (2001) Performance Monitoring of a
  Multi-product Semi-Batch Process Journal of
  Process Control, 11, 1-11.
• Conlin, A., Martin, E. B. and Morris, A. J.
  (1998), Data Augmentation An Alternative
  Approach to the Analysis of Spectroscopic Data,
  Chemometrics and Intelligent Lab. Systems, 44,
  161-173.
• J. Murtagh - Background process information and
  data.
• Dr. Y. Bissesseur, CPACT Research Associate and
  Ms. C. Mueller, MEng Student