Method Development and Validation of an Inline Process Analytical

Subscriber access provided by TUFTS UNIV

Article

Method development and validation of an in-line PAT method for blend monitoring in the tablet feed frame using Raman spectroscopy Yi LI, Carl A Anderson, James K. Drennen, Christian Airiau, and Benoit Igne Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.8b01009 • Publication Date (Web): 15 Jun 2018 Downloaded from http://pubs.acs.org on June 19, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Method Development and Validation of an In-line PAT Method for Blend Monitoring in the Tablet Feed Frame Using Raman Spectroscopy Yi Li a, Carl A. Andersona, James K. Drennen,IIIa, Christian Airiaub, Benoît Igneb,* a. Duquesne University, Graduate School of Pharmaceutical Sciences, Pittsburgh, PA, United States b. GlaxoSmithKline, Analytical Sciences and Development, King of Prussia, PA, United States * to whom correspondence author should be addressed: [email protected]

Abstract: In-line Process Analytical Technology sensors are the key elements to enable continuous manufacturing. They facilitate real-time monitoring of critical quality attributes of both intermediate materials and finished products. The aim of this study was to demonstrate method development and validation for in-line and off-line calibration strategies to determine the blend content during tablet compression via Raman spectroscopy. An in-line principal component regression model was developed from Raman spectra collected in the feed frame. At the same time, an off-line study was conducted over a small amount of the calibration blends using an in-house moving powder set-up to simulate the environment of the feed frame. The model developed offline was able to predict the active ingredient content after a bias correction and used only a fraction of the material.

The off-line method can serve as a simple method to facilitate calibration

development when the time and access to the press is limited. The study takes into consideration the necessary components of method development and offers perspectives on the validation of an in-line process analytics method. Method testing and validation was performed for the in-line process analytical technology method. The established Raman method was demonstrated as suitable for the determination of bulk assay of the active ingredient in powders inside the feed frame for use during batch and continuous manufacturing processes.

ACS Paragon Plus Environment

Analytical Chemistry 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Keywords: Process analytical technology, Raman spectroscopy, method validation, feed frame

Introduction The use of Process Analytical Technology (PAT) throughout batch and continuous manufacturing has been steadily increasing in the pharmaceutical industry. PAT tools, such as near infrared (NIR) spectroscopy and Raman spectroscopy are key technologies enabling the understanding, monitoring and controlling of pharmaceutical manufacturing. Recently, PAT has been integrated to the tableting process to directly measure the quality attributes of products, in-process materials and process conditions.1 This development has the potential to improve efficiency in production, cost reduction and improved assurance of tablets quality in the pharmaceutical industry.2 The use of PAT tools for powder monitoring at strategic locations in the table press provides opportunities to assure tablet quality.

Tablet compaction is the last step of tablet

manufacturing if coating is not required. A blend is discharged from the hopper and flows through the feed chute to the feed frame. The paddles in the feed frame further mix and push the powder into die cavities for compression. In-line systems can be deployed to monitor incoming material properties such as particle size distribution, moisture content and blend assay.3 Several undesirable phenomena associated with the mixing elements of the press, for example segregation4 and overlubracation,5 can be detected by real-time PAT sensors. In addition, the timely measurement of the process parameters and materials attributes can support a control system to adjust to disturbances through feed-forward and feedback loops6 to trigger diversion of out-of-specification tablets or to adjust process conditions to dampen disturbances. Near-infrared spectroscopy is one of the most commonly used PAT tools for monitoring pharmaceutical processes. A large number of NIR applications are focused on on-line, at-line and off-line7,8 quantification of tablet composition, yet only a few exploited the potential of PAT to be


Page 2 of 27

Page 3 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


used, in-situ, during compaction.9-11 Available publications regarding powder monitoring using NIRS have indicated the exceptional sensitivity of NIR signal to physical attributes12,13 and flow patterns of the powders.14 The physical attributes of powders include powder density, mass of powder inside the feed frame (hold-up mass) and particle size distributions. Variations in distance between the probe and the powder, paddle speed and press speed, all have impact on the dynamics of the system, and can lead to NIR baseline variations.9 Although these effects can be useful for monitoring tablet quality, such as hardness, they present challenges to the quantification of blend composition, especially for low active content formulations. Raman spectroscopy is a potential in-line PAT analyzer for API content determination in the tablet feed frame. Raman spectroscopy shows excellent molecular specificity, with sharp spectral features.15 Nevertheless, Raman spectra still contain information regarding non-analyte specific variances. These variations include sample absorption, fluorescence and scattering due to varying physical properties such as density and particle size.16 Commercial off-the-shelf Raman instruments are equipped with wide-area illumination, featuring a reduced sensitivity to sample presentation (physical properties).17 The reduced sensitivity to physical properties may also facilitate model transfer and scale-up, and life cycle requirements of Raman based analytical methods. Several challenges are associated with the use of Raman spectroscopy for in-line powder monitoring in the feed frame. Long acquisition time or multiple data collection may be required to achieve adequate signal-to-noise ratio and the effect of photo bleaching on model performance should be considered. Also, variation in the distance from the powder to the probe due to paddle wheel speed may have an effect on the model performance. 11 The aim of the study was to use in-line and off-line modeling approaches to determine the composition of blends in the tablet feed frame prior to compression using Raman spectroscopy. The target formulation of the blends was comprised of pharmaceutically relevant ingredients, including an active pharmaceutical ingredient (API) at 8.0 %w/w. In the first part of the study, an in-line



model was built incorporating only the necessary compositional and process variables in the calibration design. The developed in-line method was validated according to ICH-Q2 guidelines. In the second part of the study, two modeling approaches were tested to predict the in-line data. An off-line method was developed from measurements of powders in an in-house moving powder setup. The other approach involved measurement of static powders in the feed frame. This study presents the deployment and validation of a Raman probe in the feed frame of a tablet press for quantitative assessment of real-time tableting process. Experimental Section Materials. A direct compression formulation comprised of acetaminophen (Granules India Limited, Andhra Pradesh, India), lactose anhydrous (Supertab 21AN, DFE pharma, Paramus, NJ, USA), Microcrystalline cellulose (Avicel PH102, FMC, Philadelphia, PA, USA), croscarmellose sodium (AcDiSol, FMC, Philadelphia, PA, USA), magnesium stearate (Ligamed MF-2-V, Peter Greven GmbH & Co.KG, Bad Mü nstereifel, Germany) and colloidal silicone dioxide (Aerosil 200, Evonik Co., Parsippany, NJ, USA) was used. The active ingredient in the target formulation was 8% on a weight basis or 12 mg per tablet (150 mg target tablet weight).

Blend Preparation. Blends used in calibration, test and validation were prepared in a 5-liter Vblender (Patterson-Kelly, East Stroudsburg, PA, USA). The batch size was 2 kg. Acetaminophen preblend were prepared by mixing acetaminophen with 1.0 % of silicon dioxide at 25 rpm for 20 min. The acetaminophen pre-blend, lactose, microcrystalline cellulose (MCC), sodium croscarmellose were then blended at 25 rpm for 20 min. The lubricant magnesium stearate was finally added and mixed for another 4 min.


Page 4 of 27

Page 5 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Experimental Design. To optimize the design of experiment and minimize the number of samples to manufacture and run through the press, factors such as API concentration, moving speed of powders in the press and feed-frame were evaluated in an initial scoping study (Supporting Information Table S.1) to identify their criticality on the shape and intensity of the Raman signal. Meanwhile, a spectral investigation of the raw materials was performed. After sufficient specificity between the API and excipients was demonstrated (see results section), concentrations of excipients were fixed in the calibration set as the lack of overlap between the various components would not be affected by changing concentrations of excipients. The calibration set covered five API concentration levels (75%, 90%, 100%, 110% and 125% of label claim). An independent test set was manufactured to verify the model performance before validation. The test included three batches that varied in API concentration (90%, 100% and 110% of label claim) and MCC concentration (97%, 100% and 103% of nominal). The MCC concentration was varied by 3% to challenge the specificity of the calibration model and ensure baseline was properly corrected for by the preprocessing step. During calibration and test, the press speed was kept constant at 20 rotations per min (rpm) and the paddle speed was kept constant at 15 rpm. The validation set, independent from the calibration and test sets, consisted of three concentration levels (90%, 100% and 110% of label claim) of API to cover the range (90-110% of label claim) specified by USP monograph for acetaminophen tablets.18 The press speed was changed for each batch to investigate method robustness (18 rpm, 20 rpm and 22 rpm for each concentration level).

Raman Spectrometer. Raman spectral collection was carried out with a Kaiser RXN4®-Hybrid Raman system (Kaiser Optical System, Ann Arbor, MI, USA) coupled with a PhAT contact probe. A



The Raman spectrometer was equipped with a 400mW laser at 785 nm. The diameter of laser spot size was 6 mm. The spectra in the range of 150-1890 cm-1 were acquired with a 4 cm-1 resolution. Spectra with 3 s acquisition time and 2 co-adds were taken every 10 s to provide adequate signal compared to noise and suitable sampling rate with respect to the powder flow rate.

In-line Measurement Set-up. The experimental set-up of the in-line measurement is illustrated in Figure.1 (A). A Korsch XL100 press (Korsch America Inc., MA, USA) was tooled with 7 mm punches and corresponding dies. A 12-finger paddle, reduced 2mm in height to allow the insertion of the probe into the feed-frame, was used. The Raman contact probe was inserted 2 mm into the feedframe, leaving 1 mm distance between the paddles and the probe. This ensured the Raman contact probe tip was directly in contact with the powder. The contact probe (hollow tube with a sapphire window at the tip) ensured the required 25cm focal distance was maintained. An effect on predicted active content linked to the amount of powder in the feed frame was previously reported and linked to the change in density of the powder as the feed-frame emptied4. To avoid this issue, a minimum fill height in the feed chute was maintained. Each tableting batch was conducted for 11 min for the calibration set, 8 min for the test set and 19 min for the validation set. 10 tablets were collected at an interval of 1 min during steady-state operation.

Off-line Measurement Setup. The off-line set-up is illustrated in Figure.1 (B). A small spinning vessel was employed to simulate circular motion of powder inside a feed frame. A total of five 20 g powder samples were taken from the five calibration blends and placed individually in the plastic vessel. The container was secured to a lab stirrer (Eurostar 100 digital, IKA-Works, Inc., Wilmington, NC, USA) and rotated at about 200 rpm. The contact Raman probe was directly immersed into the powder. The distance between the tip of the contact probe and the bottom of the


Page 6 of 27

Page 7 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


vessel was kept at 2.5 cm. Additionally, the distance between the probe and the center of the cup was kept at 1.5 cm. The Powder was rotated for 1 min for system stabilization prior to Raman data collection. After spectral acquisition (1 min), the probe was removed and about 200 mg powder was directly sampled from the region where the probe was placed. The powder was analyzed by high performance liquid chromatography (HPLC) to obtain reference values. Three measurements were taken for each calibration batch.

In-line Static Measurement. After the compaction of calibration blends, compression was halted and six spectra were collected on the static powder. A multivariate model was constructed from the spectra and the nominal concentrations. The static measurement was carried out in the feed frame to provide a means of determining the impact of powder movement on the spectroscopic measurement.

Sample mass estimation. The sample mass measured per each Raman spectrum was estimated using Equation.(1), which was modified from a method used in a continuous blending process.19 The modified equation assumes the geometric shape of the interrogated volume of the probe is close to a portion of a hemi-ellipsoid. Equation (1) is the summation of two parts. The first part of the equation takes into account the volume of the powder moving in the feed frame at an angular velocity of (0.5 per sec). The second part calculates the volume of the half hemi-ellipsoids interrogated at the beginning and the end of each scan.

= ℎ +

(1)

where is the density of the blend (0.73 g/cm3, determined from the hold-up mass divided by the volume of feed frame), r is the Raman beam radius (0.30 cm), t is the acquisition time for one



Raman spectrum (2 co-adds, each 3 sec), is the distance from the center of the probe to the center of the feed frame (6.90 cm), and h is the penetration depth (assumed 0.05 cm19) of the 785 nm laser. Therefore, one spectrum was associated with 1.54 cm3 blend and 1.13 g of powder (about 7-8 tablets). The shape of the volume of powder samples and considered in Equation (1) can be found in the Figure.S1 of the supporting information.

HPLC Reference Analysis. 10 tablets were split into two 5-tablet sub-samples and the average was reported as the reference value for a particular sampling point. Each sub-sample was dissolved in a 100 mL flask. The mobile phase was acetonitrile-water (50/50, v/v). Two aliquots were injected into an Agilent 1100 LC system (Agilent Technologies Santa Clara, CA, USA) using an injection volume of 1 μL for each sample solution. The standard error of the laboratory (SEL) was estimated from repeated HPLC analyses using a standard solution at 0.6 mg/mL. For three consecutive days, the solution was injected for three times and the pooled standard deviation was calculated to describe the SEL. The SEL was determined to be 0.825 %.

Multivariate Modeling. In order to develop an in-line model to quantify the API content of the blend in the feed frame, the reference values of tablets were matched with the Raman spectra based on the time of reference collection. Raman spectra at the 4th, 5th, 6th and 7th minute of each calibration batch were used in calibration because they represented the spectra collected during steady-state operation. Spectral preprocessing and multivariate regression were performed in MATLAB R2015a. (The MathWorks Inc., Natick, MA,USA) and the PLS_toolbox 7.9 (Eigenvector Research Inc., Manson, WA, USA). The spectral range of 741-931 cm-1 contains two strong peaks that are characteristic of


Page 8 of 27

Page 9 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


the phenyl ring of acetaminophen.20 The spectra were truncated accordingly and normalization to unit area was applied to remove baseline shifts. Principal component regression (PCR) was applied on the mean-centered data for the quantification of acetaminophen content in blends. Principal component regression is a regression analysis technique that is based on principal component analysis and least-squares regression. Random subsets cross-validation with 5 splits and 20 iterations was used to determine the number of variables to include in the model. A model developed from the off-line calibration data was used to understand the suitability of such data collection system for predicting powder spectra collected in-line. Therefore, the same spectral range, pretreatments and regression method were used for the off-line, static and in-line models.

Validation. Upon development of the in-line regression model and its testing, method validation was undertaken. The validation metrics involved specificity, linearity, range, accuracy, precision and robustness. These validation metrics were tested statistically to meet the criteria established in the ICH-Q2 guidelines and USP specifications for acetaminophen tablets. According to USP , specificity of the method against the analyte of interest is based on interpreting both spectral attributes and chemometric parameters for NIR methods.21 This approach is also applicable for other spectroscopic techniques such as Raman spectroscopy. All pure components were measured by the Raman system. The correlation coefficients between the regression vector of the calibration model and the spectra of the pure components were used to evaluate specificity. The linearity evaluation was determined by constructing a plot of predicted API concentrations versus the HPLC reference values individually for the calibration, test and validation



Page 10 of 27

sets. The correlation of determination (r2), slope and intercept of the curve were calculated to determine the proportionality of the analytical function.

Linearity was established if r2 was

significant as indicated by a student’s t-test for regression at a 5% significance level. The 95% confidence intervals for the intercept and slope should encompass 0 and 1, respectively. Accuracy describes the closeness of the results obtained by the method in relation to the true value. The root mean square error of prediction (RMSEP) of the validation set, was compared with the root mean square error of calibration (RMSEC), cross-validation (RMSECV), and RMSEP of test set. The RMSEC/CV and RMSEP were calculated as follows: / = ! =

∑( )

∑( )

(2)

(3)

where "# are the predicted API concentrations, "# are the measured API concentrations by the reference method, n is the number of samples for the dataset under consideration and N is the number of factors used in the model. Typically, the accuracy of a secondary analytical method can only be as good as the reference method. An F-test for equality of variance was used to compare the Raman predictions with the reference values. Agreement between Raman predictions and reference values was also given by a two one-sided test (TOST) which was carried out using Statistica 64 (StatSoft. Inc. Tulsa, OK, USA) at a 5% significance level. Precision describes the closeness of multiple measurements of the same homogeneous sample under a prescribed condition. Repeated measurements on each of the three validation blends at 90%, 100% and 110% of label claim was performed to determine measurement precision in static and dynamic conditions. In the static condition, six spectra were taken in the same location without disturbing the powder. The standard deviation of the Raman predictions should be


Page 11 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


indicative of how precise the analytical instrument is to the same static sample. In the in-line scheme, the paddle speed was set to 15 rpm and the turret was stopped. Additionally, the feed frame was shut to prevent powder overfill and packing. A total of six spectra were taken from powder rotated by the paddles without the possibility for powder to leave or enter the feed-frame. No change in the powder property caused by this extended mixing (36s of data collection or 6 scans) was observed. As long as the homogeneity of the blend contained in the feed frame is maintained, the repeated measurements of the moving powder can be used to estimate the precision obtained in the dynamic condition. The standard deviation of the predictions was calculated for each blend. Robustness demonstrates the capability of the method to remain unaffected by variations in process, formulation or method parameters. In the validation set the press speed was varied by 10% (18 rpm, 20 rpm and 22 rpm) at each of the three concentrations (90%, 100% and 110% of label claim) to represent the variation during normal operation. The RMSEP was calculated for all experiments performed using each press speed. Multiple t-tests were used to assess the significance of the Raman predictions across different press speed at a 5% significance level.

Measurement Uncertainty. Typically, the observed prediction error can be decomposed into two parts: measurement error for a spectroscopic method and standard error of laboratory i.e., the reference method.22 This method assumes the samples for estimating the two types of error are free of bias, which means only random measurement errors were considered. In the current study, the measurement uncertainty for Raman method was estimated based on the following error decomposition scheme: 23 !$%&% = √! − )


(4)


Page 12 of 27

where !$%&% is the standard error attributed to the Raman method, SEL is the standard error of reference method, in this case the HPLC reference error and, and SEP is the bias corrected standard error of prediction. The SEP was defined as: ! = *

. /- 01 +∑,2

∑( )

3

(5)

where "# are the predicted API concentrations, "# are the measured API concentrations by the reference method, and n is the number of samples for the dataset under consideration.

Figure.1 Experiment set-up for (A) in-line measurement; (B) off-line measurement. The distance l (1.5 cm) between the probe and the center of the spinning cup is shown in the enlarged figure.

Results and Discussion


Page 13 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


In-line Method Development. Generally, independence between the chemical components is required in the calibration design for multivariate modeling to separate the signal obtained from mixtures. However, the concentrations of excipients were fixed in the calibration set of the study, as there was no impact of the signal of the excipients on the signal of the active ingredient except for baseline effects that were removed by preprocessing. Thus, whether a complex design of experiment was used where the major excipients were orthogonal to the distribution of the active ingredient or a simple design was utilized where all the components were correlated would not have affected the final model performance, as the model would be uniquely specific to the active ingredient in both cases. The normalized spectra of the five calibration batches (6 spectra per min, 11 min per batch) are depicted in Figure.2 (A) together with the three batches of the test set (6 spectra per min, 8 min per batch) and three batches of the validation set (6 spectra per min, 19 min per batch). Normalization to unit area within the selected region (741-931 cm-1) helped to minimize baseline offsets due to differences in the fluorescence background, changes in excipient concentration and variations in path length of the illumination. The spectra were further mean centered and analyzed by principal component analysis (PCA) to reveal any possible correlation of the spectral variance with the analyte of interest. The plot of the principal component scores is presented in Figure.2 (B). The five blends with increasing content in API were distributed along the axis of the first principal component (PC1) which explained the majority of the variance (93.02%) in the calibration spectra. PC2 only described 1.48% of the total variance and the distribution of scores was independent of the API content. Four Raman spectra were selected from every batch for PCR modeling based on the time at which tablets were collected. The rationale of using PCR method is that PCR models are simpler and they often provide better model interpretability compared to partial least-squares (PLS) models.



Additionally, PLS method and PCR method led to comparable model performance using the data in the current study. The reference values were subsequently matched with the selected Raman spectra. The variance of response incorporated by this model was 99.13%. Limited additional variance (0.47%) in the response was provided by including the second component into the model. The first PC1 had strong correlation with the API content and it was used for modeling moving forward.

Figure.2 (A) Normalized Raman spectra of the calibration, test and validation sets obtained with the in-line set-up. (B) PCA scores of the in-line calibration, test and validation sets. (C) The measured versus predicted API content for in-line data. (D) Hotelling’s T2 and Q-residuals of the test and validation sets compared to that of the calibration set. The


Page 14 of 27

Page 15 of 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


composition of the samples from the test set is indicated by numbers in brackets, e.g. (90, 97) denotes 90 % LC API samples with 97 % (of nominal) MCC. The subsampling strategy for Figure.2.(C) and (D) is provided in the footnote.a

In-line method validation. Figure.2 (C) shows the predicted versus measured API content of the test and validation sets. A slightly larger RMSEP of 2.03% LC in validation may be related to the inclusion of changing press speed. The RMSEP of the validation samples using the same press speed (20 rpm) as in the calibration was 1.96 % LC, which was similar to the cross validation error. Noticeably, there were 4 samples with higher than expected Q-residuals. They corresponded to the test samples with change in MCC concentration of 3%, a source of variability not included in the calibration set and detected as unmodeled variance. However, Figure.2 (D) shows the Hotelling’s T2 of these samples were inside the calibration range, suggesting the model was capable of capturing variance that corresponded to the analyte of interest. In Figure.2 (C), the predictions of the test set were accurate regardless of minor variation in the excipient concentration. A two one-sided test (TOST), performed at a 5% significance level, showed the predicted API concentrations were equivalent to the HPLC reference measurements. The RMSEPs and bias for the validation set are reported in the Table.1. Table.1 Accuracy and precision parameters of the developed in-line Raman method calculated from the validation set

Batch content (label claim)

RMSEP

F-test

(n=9)

Var (NIR) / Var (Ref)

90%

2.33

1.46

Method Development and Validation of an Inline Process Analytical

Recommend Documents