The Business Case for Building the Janssen Brand
Transfer of NIR Calibration Models Between Different Instruments for Measuring CU and in-line BU and Automated ID testing s
Sarah Nielsen, PhD | Senior Scientist, Advanced Technology COE | October 28, 2015
Jennifer Jacobs, Stowaway Jennifer is a New York based artist living with Type 1 diabetes.
Outline • Introduction • Overview • Continuous Manufacturing • •
RtRT Content Uniformity
• Calibration Transfer • • • •
Preprocessing Alignment Direct Standardization Statistical Analysis of Methods
• Rollout of Methods •
Example Workflow of Analysis
• Conclusion
2
Overview
3
Continuous Manufacturing with RtRT • Tablets - Multiple AtLine Instruments for ID and Content Uniformity • Blends – Multiple Inline Instruments for Blend ID and Blend Uniformity
4
Bench NIR testing for Tablet Content Uniformity • Central location for model development • Model transferred to multiple global locations
5
Model Transfer
Child Instruments
Parent Instrument
• Model developed on parent instrument
• Conduct parallel measurements on parent and child instruments or various child instruments • Demonstrate equivalency between instruments • Model predictions and spectral quality must meet minimum predefined specifications before model can be validated for use in child instruments
6
Child Instruments
Parent Instrument
Model Transfer Validation
Equivalent, Validated
Not-Equivalent, Transfer Fail
7
Options with Transfer Fail Calibration Transfer
Analytical Method
Development Time
Severely Increased
Minimal Increase
No Increase
Additional Sample Preparation
Many Additional Samples
Minimal Sample None Requirements
RtRT
Yes
Yes
No
In-Line Testing
Yes
Yes
No
Increased Sample Testing
Yes
Yes
No
Long Term Model Maintenance
Hard
Easy
NA
Long Term Gain
Individual Models
8
Calibration Transfer
9
Transfer Fail for RtRT Tablet CU Model Parent Instrument – Bruker MPA
RMSEP= 0.8%
Child Instrument – Bruker MPA
RMSEP= 3.2%
Preprocessing and Tools and Model Information Order Matters – Standard Normal Variant (SNV), 1st Derivative – 1st Derivative, SNV
Software Tool – Eigenvector Toolbox V 8.0.1 – Direct Standardization Algorithm
Model – PLS 2 Latent Variable Model – Models built with 5 Concentration Levels (70,85,100,115,130)
11
Transfer Fail for RtRT Tablet CU Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Child Instrument “Aligned” with parent instrument. X-Axis Based Alignment and Peak Detection
12
Transfer Fail for RtRT Tablet CU Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. New order of preprocessing improves transferability
SNV, 1st Derivative 1st Derivative, SNV
13
Transfer Fail for RtRT Tablet CU Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Calibration Transfer Calculated and Applied
Parent Child DS
14
Metrics • Standard Error of Prediction (RMSEP)
15
Q Residuals vs Hotelling T2 Before Calibration Transfer
After Calibration Transfer
Child Parent
Child Parent
Parent
Child Before
Child After Cal Trans
16
Transfer Fail for In-Line Blend Model Slight differences in powder density Slight differences in instrument performance Significant Transfer Fail
17
Transfer Fail for In-line BU Model Parent Instrument
Child Instrument
– Bruker Matrix
– Bruker Matrix
Position “1”
Position “2”
RMSEP= 4.2
%Nominal
RMSEP= 0.8
Parent Child
Transfer Fail for In-Line Blend Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Child Instrument “Aligned” with parent instrument. X-Axis Based Alignment and Peak Detection
19
Transfer Fail for In-Line Blend Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
SNV, 1st Derivative
Solution Applied 1. New order of preprocessing improves transferability
1st Derivative, SNV
20
Transfer Fail for In-Line Blend Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Direct Standarization (DS)Calibration Transfer Calculated and Applied
DS
21
Metrics • Standard Error of Prediction (RMSEP)
22
Q Residuduls Before Calibration Transfer Child Parent
After Calibration Transfer Child Parent
23
Statistical Analysis of Results – Anova Analysis 100%
130%
% Nominal
% Nominal
%Nominal
70%
Statistically Not Different Between Parent and Child Instrument with Transfer
24
Implementing the Tool
25
Potential Data Workflow
26
Conclusions • Successfully transferred models from parent to child instruments which had previously fail using a combination of preprocessing, spectral alignment and direct standardization techniques • Statistical analysis showed significant improvements in SEP.
27
Acknowledgements Olav Lyngberg, PhD Mauricio Futran, PhD Yleana Colon Jenny Vargas-Irizarry Steve Mehrman, PhD Eric Sanchez
28
Questions
Contact information [email protected] (215) 628-5005
29
Sarah Nielsen, PhD | Senior Scientist, Advanced Technology COE | October 28, 2015
Jennifer Jacobs, Stowaway Jennifer is a New York based artist living with Type 1 diabetes.
Outline • Introduction • Overview • Continuous Manufacturing • •
RtRT Content Uniformity
• Calibration Transfer • • • •
Preprocessing Alignment Direct Standardization Statistical Analysis of Methods
• Rollout of Methods •
Example Workflow of Analysis
• Conclusion
2
Overview
3
Continuous Manufacturing with RtRT • Tablets - Multiple AtLine Instruments for ID and Content Uniformity • Blends – Multiple Inline Instruments for Blend ID and Blend Uniformity
4
Bench NIR testing for Tablet Content Uniformity • Central location for model development • Model transferred to multiple global locations
5
Model Transfer
Child Instruments
Parent Instrument
• Model developed on parent instrument
• Conduct parallel measurements on parent and child instruments or various child instruments • Demonstrate equivalency between instruments • Model predictions and spectral quality must meet minimum predefined specifications before model can be validated for use in child instruments
6
Child Instruments
Parent Instrument
Model Transfer Validation
Equivalent, Validated
Not-Equivalent, Transfer Fail
7
Options with Transfer Fail Calibration Transfer
Analytical Method
Development Time
Severely Increased
Minimal Increase
No Increase
Additional Sample Preparation
Many Additional Samples
Minimal Sample None Requirements
RtRT
Yes
Yes
No
In-Line Testing
Yes
Yes
No
Increased Sample Testing
Yes
Yes
No
Long Term Model Maintenance
Hard
Easy
NA
Long Term Gain
Individual Models
8
Calibration Transfer
9
Transfer Fail for RtRT Tablet CU Model Parent Instrument – Bruker MPA
RMSEP= 0.8%
Child Instrument – Bruker MPA
RMSEP= 3.2%
Preprocessing and Tools and Model Information Order Matters – Standard Normal Variant (SNV), 1st Derivative – 1st Derivative, SNV
Software Tool – Eigenvector Toolbox V 8.0.1 – Direct Standardization Algorithm
Model – PLS 2 Latent Variable Model – Models built with 5 Concentration Levels (70,85,100,115,130)
11
Transfer Fail for RtRT Tablet CU Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Child Instrument “Aligned” with parent instrument. X-Axis Based Alignment and Peak Detection
12
Transfer Fail for RtRT Tablet CU Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. New order of preprocessing improves transferability
SNV, 1st Derivative 1st Derivative, SNV
13
Transfer Fail for RtRT Tablet CU Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Calibration Transfer Calculated and Applied
Parent Child DS
14
Metrics • Standard Error of Prediction (RMSEP)
15
Q Residuals vs Hotelling T2 Before Calibration Transfer
After Calibration Transfer
Child Parent
Child Parent
Parent
Child Before
Child After Cal Trans
16
Transfer Fail for In-Line Blend Model Slight differences in powder density Slight differences in instrument performance Significant Transfer Fail
17
Transfer Fail for In-line BU Model Parent Instrument
Child Instrument
– Bruker Matrix
– Bruker Matrix
Position “1”
Position “2”
RMSEP= 4.2
%Nominal
RMSEP= 0.8
Parent Child
Transfer Fail for In-Line Blend Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Child Instrument “Aligned” with parent instrument. X-Axis Based Alignment and Peak Detection
19
Transfer Fail for In-Line Blend Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
SNV, 1st Derivative
Solution Applied 1. New order of preprocessing improves transferability
1st Derivative, SNV
20
Transfer Fail for In-Line Blend Model Calibration Optimization 1. Spectral Alignment 2. Pre-Processing 3. Background Difference 4. Unique Instrument Variation
Solution Applied 1. Direct Standarization (DS)Calibration Transfer Calculated and Applied
DS
21
Metrics • Standard Error of Prediction (RMSEP)
22
Q Residuduls Before Calibration Transfer Child Parent
After Calibration Transfer Child Parent
23
Statistical Analysis of Results – Anova Analysis 100%
130%
% Nominal
% Nominal
%Nominal
70%
Statistically Not Different Between Parent and Child Instrument with Transfer
24
Implementing the Tool
25
Potential Data Workflow
26
Conclusions • Successfully transferred models from parent to child instruments which had previously fail using a combination of preprocessing, spectral alignment and direct standardization techniques • Statistical analysis showed significant improvements in SEP.
27
Acknowledgements Olav Lyngberg, PhD Mauricio Futran, PhD Yleana Colon Jenny Vargas-Irizarry Steve Mehrman, PhD Eric Sanchez
28
Questions
Contact information [email protected] (215) 628-5005
29
