fda irvine usp eas2009 final ger[1] final
TRANSCRIPT
A Comparison of Near-Infrared (NIR) Method Development Approaches Using a Drug
Product on Different Spectrophotometers and Chemometric Software
Algorithms
New Publication
Journal of Near Infrared SpectroscopyVolume 17 Issue 5, Pages 233–245 (2009) doi: 10.1255/jnirs.854
NIR Publications An imprint of IM Publications
http://www.impublications.com/nir/abstract/J17_0233
Collaborators
Assad Kazeminy,1 Saeed Hashemi,1 Roger L. Williams,2 Gary E. Ritchie,3 Ronald Rubinovitz,4 and *Sumit Sen5
1Irvine Pharmaceutical Services, Inc., 10 Vanderbilt, Irvine, CA 92618 2United States Pharmacopeial Convention, 12601 Twinbrook Parkway Rockville, Maryland 20852
3Former United States Pharmacopeial Convention, 12601 Twinbrook Parkway Rockville, Maryland 20852
4Buchi Corporation, 19 Lukens Drive, New Castle, DE 197205United States Food and Drug Administration, 19701 Fairchild, Irvine, CA 92612
*Corresponding Author: [email protected]
Acknowledgements
USP:Darrell AbernathyRebecca AllenTodd CecilWalter HauckAndrea IwanikSteven LaneSamir WahabPatricia White
Irvine:Rudy FlachCharles PetersenHeather Coffin
FDA:William Martin
Buchi:Michael Surgeary
Integrated Technical Solutions: Cynthia Kradjel
Bruker:Verne Hebard
The authors are grateful for the contributions for this project from each of the following scientists:
Background
• Drug importation has brought the Homeland Security challenges into the forefront, due to risks involved with off-shore drug production and packaging.
• We established a validated methodology using NIRS that US FDA, US Port Authority, and Homeland Security can use as an on-line device in US ports to scan and screen bulk containers of pharmaceuticals to ensure that the drugs are not counterfeit or intentionally tainted with harmful substances.
Near-Infrared (NIR) Project for Analysis of Counterfeit and Tainted Drugs Protocol(FDA, USP, Irvine Pharmaceutical Services Protocol, 2007)
Protocol Outline
• Agreement• Approval• Summary• Objective• Method• Protocol Number• Define the Type of Protocol
(i.e. General or Site Specific)• Sponsor(s)• Sample Handling and Labeling
• Experimental Outcomes• Define the Guidance
Documents and Standard Operating Procedures Applied to Conduct of the Studies
• Study Schedule• Study Design• Study Procedures• Analytical Methods• Results• References
Protocol for Model ConstructionProposed By Scafi and Pasquini
• Based on flow chart proposed by Scafi and Pasquini to design, develop, validate and deploy spectral libraries
Steps for Multivariate Analysis
• The multivariate chemometric approach to calibrate, validate and maintain the spectral libraries is based on work proposed by Workman and Brown and embodied in ASTMI E1655
Selecting the Calibration Set
The Collection of Infrared spectra
Calculating the Mathematical Model
Validation of the Calibration Set
Application of the Model for the Analysis of Unknowns
Routine analysis and Monitoring
Transfer of Calibrations
Determination of Calibration Set
Concentration or Properties or Both
Study ObjectivesThere are four specific objectives of the study:
3. Demonstrate that using computerized algorithms of PCA, three identical or nearly identical spectral libraries can be obtained on three different instruments
2. Four formulations of Ibuprofen: two branded and two store-branded Ibuprofen Immediate Release Tablets will result in four distinct clusters in multidimensional
space.
3. The calibration models are able to analyze unknown tablets and determine if they are ibuprofen immediate release tablets from one of the four known sources of origin that constitute the libraries or are not from any of these known sources.
4. Each analysis of unknown should give precisely the same answer no matter which calibration model is used, demonstrating that the protocol can be used in any laboratory, independent of location, device, analyst or software used to perform the identification.
Study Design
SamplingLot Brand - Advil Brand - Motrin Generic - CVS Generic - Rite Aid
B946681 PCA189 6EE0102 P450322 x 200 = 400 tablets 3 x 100 = 300 tablets 2 x 250 = 500 tablets 1 x 500 = 500 tablets
B87154 LLA103 6HE0515 P443892 x 200 = 400 tablets 3 x 100 = 300 tablets 4 x 100 = 400 tablets 1 x 500 = 500 tablets
B94669 PCA112 7BE0119 P4244773 x 100 = 300 tablets 3 x 100 = 300 tablets 1 x 750 = 750 tablets 1 x 250 = 250 tablets
B27624 PCA226 7AE0039 P446864 x 75 = 300 tablets 10 x 24 = 240 tablets 1 x 750 = 750 tablets 1 x 250 = 250 tablets
B91364 PBA123 7CE0268 P413602 x 150 = 300 tablets 3 x 75 = 225 tablets 1 x 750 = 750 tablets 3 x 100 = 300 tablets
B73322 PBA194 7AE0699 P424762 x 150 = 300 tablets 3 x 100 = 300 tablets 1 x 500 = 500 tablets 3 x 100 = 300 tablets
B33863 PAA016 6LE0478 P424983 x 100 = 300 tablets 5 x 50 = 250 tablets 1 x 500 = 500 tablets 2 x 120 = 240 tablets
B91414 PBA186 7AE0270 P441512 x 200 = 400 tablets 5 x 50 = 200 tablets 3 x 100 = 300 tablets 5 x 50 = 250 tablets
B98483 PEA106 6GE0118 P446889 x 24 = 216 tablets 2 x 100 = 200 tablets 1 x 500 = 500 tablets 5 x 50 = 250 tablets
B91386 LLA329 7BE0606 P420582 x 100 = 200 tablets 2 x 100 = 200 tablets 1 x 500 = 500 tablets 2 x 100 = 200 tablets
1
2
3
4
9
10
5
6
7
8
Ibuprofen Tablets
Innovator A Innovator B
Generic A Generic B
Ibuprofen Formulation TableADVIL MOTRIN CVS - IBUPROFEN RITE AID - IBUPROFEN
Acetylated MonoglyceridesBeeswax
Propylene GlycolCelluloseLactose
Sodium Starch GlycolateEthoxyethanol
LecithinParabens
Pharmaceutical GlazePovidone
SimethiconeSodium Benzoate
Sodium Lauryl SulfateSucrose
Fd&C Yellow #6 FD&C Yellow #6Magnesium Stearate Magnesium Stearate
Polydextrose PolydextrosePolyethylene Glycol Polyethylene Glycol
Croscarmellose Sodium Croscarmellose SodiumMicrocrystalline Cellulose Microcrystalline CellulosePharmaceutical Shellac ShellacPregelatinized Starch Pregelatinized Starch
Carnauba Wax Carnauba Wax Carnauba WaxHypromellose Hypromellose Hypromellose
Iron Oxides Iron Oxide Iron OxidesSilicon Dioxide Colloidal Silicon Dioxide Fumed Silica Gel Colloidal Silicon Dioxide
Corn Starch Corn Starch Corn Starch Corn StarchIbuprofen Ibuprofen Ibuptofen Ibuprofen
Stearic Acid Stearic Acid Stearic Acid Stearic AcidTitanium Dioxide Titanium Dioxide Titanium Dioxide Titanium Dioxide
Product & API Raw Spectra
VISIBLE REGION NEAR INFRARED REGION
IBUPROFEN API
INNOVATOR A
(ADVIL) INNOVATOR B
(MOTRIN)NAPROXEN
FENOPROFEN
Generic B
Generic A
Observations
• Advil is the most different spectrum from other manufacturers; – note 1400 nm – 1600 nm and 2000 nm – 2200 nm regions
• Generic A & B and Naproxen and Fenoprofen spectra have similar bands and offset
• Motrin spectrum is midway in offset between Generic A & B, Naproxen, Fenoprofen group and Advil spectrum. – It has similar spectral features as the group
• The visible region between 400 nm – 800 nm show distinct bands for each tablet. – For this study, will ignore the visible region and concentrate on
the region from 1000 nm - 2500 nm.
BUCHI Raw Spectra and Standard Deviation
0
0.001
0.002
0.003
0.004
0.005
0.006
1000.00 1048.66 1102.29 1161.71 1227.90 1302.08 1385.81 1481.04 1590.33 1717.03 1865.67 2042.48 2256.32
Variables
Line Plot
FOSS Raw Spectra and Standard Deviation
Log (1/R)
0
0.0005
0.0010
0.0015
0.0020
0.0025
0.0030
0.0035
1000 1500 2000 2500 SDev
Variables
Line Plot
BRUKER Raw Spectra and Standard deviation
Wavelength
Log (1/R)
0
0.002
0.004
0.006
0.008
0.010
1000.20 1048.76 1102.27 1161.53 1227.52 1301.47 1384.90 1479.76 1588.56 1714.64 1862.45 2038.15 2250.46 SDev
Variables
Line Plot
Hypothesis
• Using a combination of spectral pretreatments, a separation in wavelength space results in distinct clusters of the four Ibuprofen formulations using principal component analysis (PCA) in principal component (PC) space.
• Measure impact on specificity by pretreatments, selection of spectral region, and PCA, using Ibuprofen analog (Naproxen) formulation
Model Building Cluster
-
-
+
-
Generic B
- Baseline Corrected Second
Derivative
- Baseline Corrected
First Derivative
- Second Derivative
- First Derivative
Baseline Corrected
- + Untreated Spectra
Generic A Innovator B Innovator A
Spectra Treatment (1000 nm - 2500 nm)
Cluster
+
+
Generic B
+ + Baseline Corrected Second
Derivative
+ + Baseline Corrected
First Derivative
+ + Second Derivative
+ + First Derivative
+ + + Baseline Corrected
+ +
Generic A Innovator B Innovator A
Spectra Treatment (1000 nm - 2500 nm)
* Plus sign denotes separation of the sample sets and minus sign denotes no separation.
Model Building
Cluster
+
Generic B
- - + Baseline Corrected Second
Derivative
+ Baseline Corrected
First Derivative
- - + Second Derivative
First Derivative
- Baseline Corrected
- -
Generic A Innovator B Innovator A
Spectra Treatment (1400 nm - 1500 nm)
Cluster
-
+
-
-
-
Generic B
- Baseline Corrected Second
Derivative
+ + Baseline Corrected
First Derivative
- Second Derivative
+ + First Derivative
- + + Baseline Corrected
- +
Generic A Innovator B Innovator A
Spectra Treatment
-
Cluster
-
Generic B
Baseline Corrected Second
Derivative
Baseline Corrected
First Derivative
Second Derivative
+ First Derivative
Baseline Corrected
Generic A Innovator B Innovator A
Spectra Treatment (1400 nm - 1500 nm)
Cluster
Generic B
Baseline Corrected Second
Derivative
Baseline Corrected
First Derivative
Second Derivative
+
First Derivative
Baseline Corrected
Untreated Spectra
Generic A Innovator B Innovator A
Spectra Treatment
-
* Plus sign denotes separation of the sample sets and minus sign denotes no separation.
Raw Ibuprofen Spectrum
Baseline Corrected, First Derivative
Ibuprofen Spectrum
Observation
Spectral Preprocessing Experiment 1BUCHI• Buchi NIRFlex Solids
spectrometer • NIRCal® 5.2 • Irvine Pharmaceutical Services• 1000 nm - 2500 nm
(10000 cm-1 to 4000 cm-1) • Spectral Resolution set at 8 cm-1
• 1501 Data points• Uses customizable linear filters
to manipulate the spectra for preprocessing, Savitzky-Golay
First Derivative (SG 1st Deriv) (1000 nm – 25000 nm), using a 3rd order polynomial, 21 point smoothing, PCs: 3
Spectral Preprocessing Experiment 2FOSS• FOSS XDS Rapid Content
Analyzer • Vision™ 3.4 • United States Pharmacopeia• 400 nm to 2500 nm• Spectral Resolution set at
specified bandpass of 9 nm • 4200 Data points• Savitzky-Golay First Derivative
(SG 1st Deriv) (1000 nm – 25000 nm), using a 3rd order polynomial, 21 point smoothing, PCs: 5
Spectral Preprocessing Experiment 3
BRUKER• Bruker Vector 22N FT-NIR
spectrometer • OPUS™ 5.5• United States Pharmacopeia• of 4000 cm-1 -12000 cm-1 • Spectral Resolution set at 8 cm-1 • 2075 Data points• Savitzky-Golay First Derivative
(SG 1st Deriv) (1000 nm – 25000 nm), using a 3rd order polynomial, 21 point smoothing, Factors: 6
-0.0010
-0.0005
0
0.0005
0.0010
0.0015
0.0020
1400.61nm1411.29nm1422.12nm1433.13nm1444.31nm1455.66nm1467.19nm1478.91nm1490.82nm
Variables
Line Plot
Considerations
• Baseline corrected spectra in The Unscrambler® needed minimal pretreatment to separate all four formulations in spectral space.
• Baseline algorithms significantly different for the other three software platforms not giving sufficiently similar spectral pretreatments that result in identical processed spectra
• It was decided to drop baseline correction and use the derivative / smoothing step on all calibration data.
Baseline Correction
• Using normalization in OPUS™ 5.5 to match the baseline corrected spectra from the other three software packages found an unexpected obstacle, forcing the investigators to not use a baseline correction first.
• The normalization pretreatment algorithm resides in a different module from where the calibration model resides.
• To use that pretreatment on a calibration and unknown samples, a macro is required that would preprocess the spectra and automatically load those spectra into OPUS IDENT as well analyzing an unknown sample (using offset normalization spectral preprocessing and automatically output a result).
Considerations
• Derivative transforms in Vision™ 3.4 NIRCal®
5.2 and Unscrambler® 9.7 performed using variable settings for the polynomial
• Default first derivative transformation for OPUS™ 5.5 software is a cubic polynomial equation (3rd order polynomial)
PCA Experiment 1
BUCHI• Principal Component Analysis
(PCA) by Mahalanobis distance (MD) in PC space
Innovator A
Innovator B
Generic A
Generic B
PCA Experiment 2
FOSS• Principal Component Analysis
(PCA) by Mahalanobis distance (MD) in PC space
Innovator A
Innovator B
Generic B
Generic A
PCA Experiment 3
BRUKER• Principal Component Analysis
(PCA) by Euclidian distance (ED) in spectral space
• Uses Factorization, which is a spectral distance calculation
Considerations
• Principal Component Analysis (PCA) models by Mahalanobis distance (MD) for FOSS and Buchi calculated in PC space
• Factorization Method by Euclidian distance (ED) for the Bruker calculated in spectral space
• PCA calculation is invariant in either wavelength space or PC space
Results of Model Predictions
Instrument Lot Innovator A Innovator B *Generic A Generic B
Foss 9 20 20 0 20
10 20 8 0 20
Bruker 9 20 20 0 20
10 20 8 0 20
Buchi 9 20 20 0 20
10 20 20 0 20
Twenty tablets were tested for each combination of instrument and lot except for Lot 10 of Innovator B where only 8 were tested on the Foss and Bruker instruments.
* Generic A from Lot 9 and 10 were correctly identified as not belonging to Generic A lots comprising the calibration sets for all three data sets. It was observed that these samples have a distinct banding pattern in the region from 1400 nm – 1500 nm from all other sample spectra.
Conclusions
• This study was designed to study the impact that NIR instrument hardware and software configuration have on NIR method development and model deployment across different instrument and software platforms.
• Different optical benches (grating, Interferometer, and Polarizing Interferometer), data points, and sample presentation, as well as spectral pretreatments, PC analysis (MD) or Factorization using (ED), impact NIR method development.
Conclusions
Standardizing Reporting of NIR Method Parameters:
• Analyst• Date• Time• Location• Model Name• Software, version • Instrument,Model # and Serial
ID• Samples, Origin, Type,
(Calibration Set, Training Set, Validation Set, Model Labels
• Spectral Range, Spectral Resolution, Data Points
• Preprocessing Steps• Algorithm• Number of Variables (Factors,
PCs)• Validation Results, Number of
Samples Correctly Identified, Number of Samples Incorrectly Identified
• Calibration Raw Spectrum, Preprocessed Spectrum, Scores Plot, Dendogram, Other
• Validation Raw Spectrum, Preprocessed Spectrum, Scores Plot, Dendogram, Other
Conclusions
• However, despite significant differences in hardware and software, it is possible to develop NIR methods for a common data set that can be used to give identical results for discriminate analysis across different instrument and software platforms, if these differences are accounted for during method development.
Conclusions
• If NIR and multivariate analytical methods are to be used by pharmaceutical regulators as a tool in the war against pharmaceutical counterfeiting, NIR method development on a global scale will require carefully designed protocols to minimize variability.
• Differences arising from instrument and software, human errors, as well as built in quality assurance measures for scientific and legal validation need to be understood and controlled for accurate reporting and subsequent prosecution.
Conclusions
• It is hoped that this joint study undertaken by the US FDA, USP and private industry may serve as a model for future applications that involve large sample sets, and perhaps multiple organizations using different instrument–software combinations.
• The current global pharmaceutical anti- counterfeiting initiatives underway should benefit from knowledge gained from this study.
“The computer can't tell you the emotional story. It can give you the exact mathematical design, but what's missing is the eyebrows.” American Composer 1940 - 1993