NIR Paper 2011

AAPS PharmSciTech ( # 2011)
DOI: 10.1208/s12249-010-9580-z

Research Article

NIR Spectroscopy Applications in the Development of a Compacted
Multiparticulate System for Modified Release

Stuart L. Cantor,1 Stephen W. Hoag,1,3 Christopher D. Ellison,2 Mansoor A. Khan,2 and Robbe C. Lyon2

Received 20 January 2009; accepted 16 December 2010
Abstract. The purpose of this study was to utilize near-infrared spectroscopy and chemical imaging to
characterize extrusion-spheronized drug beads, lipid-based placebo beads, and modified release tablets
prepared from blends of these beads. The tablet drug load (10.5–19.5 mg) of theophylline (2.25 mg
increments) and cimetidine (3 mg increments) could easily be differentiated using univariate analyses. To
evaluate other tablet attributes (i.e., compression force, crushing force, content uniformity), multivariate
analyses were used. Partial least squares (PLS) models were used for prediction and principal component
analysis (PCA) was used for classification. The PLS prediction models (R2 >0.98) for content uniformity
of uncoated compacted theophylline and cimetidine beads produced the most robust models. Content
uniformity data for tablets with drug content ranging between 10.5 and 19.5 mg showed standard error of
calibration (SEC), standard error of cross-validation, and standard error of prediction (SEP) values as
0.31, 0.43, and 0.37 mg, and 0.47, 0.59, and 0.49 mg, for theophylline and cimetidine, respectively, with
SEP/SEC ratios less than 1.3. PCA could detect blend segregation during tableting for preparations using
different ratios of uncoated cimetidine beads to placebo beads (20:80, 50:50, and 80:20). Using NIR
chemical imaging, the 80:20 formulations showed the most pronounced blend segregation during the
tableting process. Furthermore, imaging was capable of quantitating the cimetidine bead content among
the different blend ratios. Segregation testing (ASTM D6940-04 method) indicated that blends of coated
cimetidine beads and placebo beads (50:50 ratio) also tended to segregate.
KEY WORDS: chemical imaging; controlled release beads; partial least squares (PLS); principal
component analysis (PCA); segregation.

INTRODUCTION NIRS has also been successfully used in a variety of
other analytical chemistry applications such as the detection
Near-infrared spectroscopy (NIRS) has been gaining of degradation products in tablets (7); studying drug moisture
widespread acceptance in the pharmaceutical industry not content over time and water mobility within the drug crystal
only for its ease of use in providing non-destructive, rapid lattice (8,9); real-time monitoring of moisture during process-
analysis of solid dosage forms, but also for its potential use as ing using fluidized bed granulation (10,11) or roller compac-
a process analytical technologies (PAT) tool. PAT can provide tion (12), measurement of particle size, and size distribution
real-time data to enable improved process understanding and (13–15); and determining the tablet drug content and content
lead to better manufacturing process control, and this uniformity (16–23).
information can ultimately reduce end-product testing and When tablets are prepared using low-dose drugs (<20 mg
its associated costs (1–5). NIRS has also found utility in the or <10% by weight of a formulation), content uniformity can
laboratory as well; a recent article discusses PAT applications become a critical issue. Recently, Ji et al. (24) prepared tablets
for the use of NIRS in monitoring polymorphic changes at- with drug contents ranging from 1 to 15 mg and studied
line during preparation of beads containing either nitro- content uniformity using NIRS. While they obtained excellent
furantoin or anhydrous theophylline (6). Knowledge of such accuracy and good agreement between the NIR predicted
phase changes is important because the chemical stability, values and HPLC reference values for doses between 5 and
manufacturing processability, and release rate of the drug can 15 mg, poorer accuracy was observed with the 1 mg (drug
be dramatically affected. load 0.76% w/w) and 2.5 mg doses. They concluded that the
calibration range should have been reduced to between 1 and
1 10 mg to enhance the accuracy at the lower concentration
School of Pharmacy, University of Maryland, 20 N. Pine Street,
levels. Thosar et al. (25) prepared tablets of anhydrous
Baltimore, Maryland 21201, USA.
2
Division of Product Quality Research, Center for Drug Evaluation theophylline with concentrations ranging from 0 to 120 mg,
and Research, Food and Drug Administration, Silver Spring, and they found improved accuracy was obtained using partial
Maryland 20993, USA. least squares (PLS) over multiple linear regression in trans-
3
To whom correspondence should be addressed. (e-mail: shoag@rx. mission mode, they did not reduce the concentration range in
umaryland.edu) their calibration dataset. The results showed that prediction

1530-9932/11/0000-0001/0 # 2011 American Association of Pharmaceutical Scientists

Cantor et al.

errors were larger for the lowest dosages; with errors of plane images forms a three-dimensional matrix of data called
17.6%, 2.3%, 1.4%, 1.1%, and 0.35% for the 3, 6, 15, 30, and a hypercube and a spectrum is collected for every pixel in a
60 mg tablet doses, respectively. single-plane image of the sample. Cogdill and Drennen (40),
While there are several articles discussing NIRS predic- Hamad et al. (41), and Lyon et al. (36) have published
tion of crushing force using tablets produced from homoge- detailed descriptions of the methods and instrumentation
nous powders or granules (17,22,26–30), to our knowledge, used in chemical imaging.
no work to date has examined NIRS to study multiparticulate In summary, all the NIR research to date has examined
tableted systems; i.e., the unique complexities that multi- traditional dosage forms like tablets and fine powders, but no
particulate tablets pose to the accurate prediction of content studies have used NIR techniques to characterize multi-
uniformity and crushing force has not been adequately particulate delivery systems. In recent years, multiparticulate
studied. These complexities include variability in light scatter- delivery systems have become more popular due in part to
ing effects, baseline shifts due to particle size differences, and their many advantages such as (42): (1) greater assurance of
the drug content of the different beads present. Baseline drug release and more reproducible plasma concentrations,
shifts are attributable in part, to subtle changes in the path (2) less likely to become lodged in the GI tract with minimal
length of light returning to the detector that results from absorption, (3) less likely to undergo dose dumping, (4)
variations in surface roughness or sample density (31). These increased bioavailability, and (5) the ability to combine
sources of variability can lead to larger standard errors during multiple release profiles in a single dosage form. Given the
calibration development and poor predictability of parame- growing significance of multiparticulate delivery systems,
ters such as content uniformity and crushing force. there is a need to develop NIR analysis techniques for these
Blend segregation of a formulation during mixing or systems. However, the internal structure of multiparticulate
tableting can occur on many levels due to differences in delivery systems is more complex than traditional dosage
particle size, true density, and particle morphology. Segrega- forms (i.e., tablets or capsules). For example, a multiparticu-
tion tendency is important because its occurrence can lead to late dosage form can contain a mixture of beads coated with
problems with content uniformity and weight uniformity of a different polymers. These polymers can deliver different
dosage form. Two novel methods for the study of segregation release profiles and be used at a variety of levels to deliver
phenomenon are the ASTM D6940-04 apparatus discussed different drug release rates. Thus, it is important to under-
recently by Xie et al. (32) and the use of NIRS. NIRS is stand how the complex internal structure of multiparticulate
ideally suited to analyze segregation tendency as it generates delivery systems affects their analysis via NIRS. In addition,
spectra containing both chemical and physical information bead segregation can be a problem for the manufacturing of
about the samples. Typically, segregation is analyzed from the multiparticulate systems and there have been very few studies
point of view of the drug, but other factors such as excipient that have examined the use of NIR in the study of bead
homogeneity and particle size distribution can also be segregation. This study builds upon previous studies by the
important and affect the segregation of a formulation. authors, Cantor et al. (43–45), and seeks to better understand
This research paper presents the combined use of NIRS the advantages and limitations of applying NIRS to multi-
with the chemometric techniques, PLS, and principal compo- particulate delivery systems.
Thus, this study has aims to: (1) to determine the
nent analysis (PCA) (3,33). Chemometrics is the extraction of
feasibility of using NIRS for the study of multiparticulate
quantitative chemical and physical information from multi-
systems, and to examine the spectral differences between the
component samples using statistics and a variety of mathemat-
use of drug and excipient powders and drug beads and
ical data processing treatments to decrease baseline shifts,
placebo beads, (2) to determine how well NIRS can predict
reduce noise, and resolve overlapping spectral peaks (34,35).
the compression force, crushing force, and content uniformity
PLS uses regression analysis to generate linear models that
of multiparticulate tablets, (3) to determine how well PCA
relate predicted variables in terms of observable variables. PCA
can discriminate between tablets of differing drug contents
is used to visualize interrelationships among the independent
and between tablets prepared individually by hand weighing
variables and is useful in identifying data outliers. A benefit of
versus those manufactured on a continuously running tablet
PCA is that it requires only spectral information; i.e., no wet
press, (4) to use NIRS and PCA to pinpoint when blend
chemical analysis is needed to determine the constituent values.
segregation has begun during the tablet press operation using
PCA has several functions including the reduction of large
different drug bead to placebo bead ratios (i.e., 20:80, 50:50,
numbers of variables contained in the spectral data down to a
and 80:20), and (5) to use NIR chemical imaging as a novel
few uncorrelated variables typically containing the relevant
tool to assess drug bead content, drug bead distribution, and
information used for calibration modeling.
segregation tendency during tableting of cimetidine bead
NIR chemical imaging technology is a fairly recent
blends containing different ratios of drug and placebo beads.
development and can be used as a tool for the pharmaceutical
industry to study heterogeneous samples. Like NIRS, it is also
a rapid, non-destructive technique and is ideally suited for a MATERIALS AND METHODS
variety of pharmaceutical development or PAT applications.
Previous research has studied issues of blend uniformity (36), Materials
content uniformity (37), impurity analysis (38), and poly-
morphs in a quality assessment of commercial pharmaceutical Fine particle ethylcellulose 7 cP viscosity grade (Ethocel
products (39). 7-FP Premium) with an ethoxyl content of 48.0–49.5% was a
A chemical image is an array of pixels which maps the gift from Dow Chemical Company (Midland, MI, USA).
chemical composition of a sample. The collection of single- Microcrystalline cellulose NF (Avicel® PH-101) was supplied

NIR Spectroscopy Applications in the Development of Compacted Multiparticulate Systems

by FMC Corp. (Princeton, NJ, USA), Sodium Starch then collected from both the Surelease® and Eudragit®
Glycolate NF (Explotab®) was supplied by JRS Pharma NE30D blends with approximately 20–23 g of the bead blend
(Patterson, NY, USA), Starch 1500 NF was donated by in each sample; of which approximately 370–375 mg of the blend
Colorcon (West Point, PA, USA), and Eudragit® RS 30D was removed, i.e., subsampled for analysis, placed in 100-mL
(Methacrylate copolymer “type B”) was supplied by Degussa low actinic volumetric flasks containing distilled water and
Pharma Polymers (Piscataway, NJ, USA). Theophylline stirred at 350 rpm for 24 h at room temperature. This same
anhydrous USP, cimetidine USP, and glycerol monostearate procedure was also followed when analyzing samples for
flakes NF were purchased from Spectrum Chemicals (New content uniformity. Concentrations of cimetidine in the first,
Brunswick, NJ, USA). Glycerol behenate NF (Compritol 888 middle, and last samples were obtained using a Spectronic
ATO) was supplied by Gattefosse (Paramus, NJ, USA). Genesys 2 UV/VIS spectrophotometer at 219 nm (Thermo
Electron Corp., Waltham, MA, USA) with quartz cuvettes of
Drug Bead and Lipid-Based Placebo Bead Composition 1-cm path length. From each sample, at least five subsamples
and Methods of Manufacture were analyzed and the average drug content was determined
from the first, middle, and last sample groups. The last/first ratio
Bead formulations containing 8.57% w/w theophylline or and the coefficient of variation (CV%) were then calculated.
cimetidine were prepared by extrusion–spheronization using The last/first ratio is used to quantitate the degree of segrega-
an extruder speed of 37 rpm and spheronization conditions of tion; a ratio equal to 1.0 indicates that no segregation occurred,
500 rpm for approximately 1 min. The beads were then oven and a greater difference from 1.0 indicates that a higher degree
dried at 50°C for 24 h and a #18/30 sieve cut was performed to of segregation has occurred, as described in references (32,46).
eliminate oversized and undersized particles. In addition to
the drug, at a level of 8.57% w/w, the beads also contained Tableting and Tablet Evaluation
microcrystalline cellulose as a filler-binder at 15.43% w/w and
ethylcellulose as a hydrophobic matrix former at a level of Tablets were either prepared individually or in continu-
58.0% w/w, Eudragit RS 30D was also used as a polymeric ously running mode on a tablet press. The initial theophylline
binder to further delay drug release and was used at a level of tablets were prepared individually by hand weighing and
18.0% w/w. loading the drug bead/placebo bead blend into the die cavity
The lipid-based placebo beads contained either 50% w/w and then turning the press on. Using a 50:50 ratio of drug beads/
glycerol monostearate (GMS) or glycerol behenate along placebo beads, the beads were blended together for 3 min in a
with 42% w/w Starch 1500 and 8% w/w sodium starch plastic bag. All cimetidine tablets (used for content uniformity
glycolate as a super-disintegrant. Placebo beads were pre- analysis) were prepared on a running tablet press where the
pared by adding the dry powders to the molten wax heated to compression force was controlled between 200 and 250 kg.
80°C and subjecting the mixture to high-shear homogeniza- A Stokes B2 rotary tablet press (operating at 30 rpm)
tion at 22,000 rpm. The mass was then hand sieved through a equipped with an instrumented eye bolt for compression
#12 screen and beads were given a final spheronization at force and ejection cam was used with a single 8.7-mm round,
550 rpm for 25 s. GMS and glycerol behenate were compared concave punch set. Beads were accurately weighed and
as candidates for the lipid-based placebo beads. Both manually filled into the die to achieve target tablet weights
appeared to perform equally well in tablets, but glycerol of 350±5 mg.
behenate has a much higher melting point (82°C) compared Tablet crushing force (hardness) was determined by
with GMS (53°C) and this made it more difficult to obtain a diametric compression using a hardness tester (Model HT-300,
high yield of the correct particle size as the glycerol behenate Key International, Inc., Englishtown, NJ, USA). All tablets were
solidified into a harder mass when cooled than the GMS and it allowed to stay at ambient temperature for 24 h before hardness
was difficult to hand sieve. A lower melting point lipid was testing to allow for elastic recovery. Eighteen tablets were
desired as it would likely be softer and more plastic and subjected to 100 rotations in a friabilator (Model TA, Erweka
potentially offer superior cushioning properties to the coated GmbH, Heusenstamm, Germany) rotating at 25 rpm following
drug beads during tableting. Based on these considerations, USP 24 Method <1216>. For the sintering studies, the tablet
GMS was selected as the placebo bead lipid for this study. The weights were increased to 400 mg and tablets were subjected to a
manufacturing methods for both drug beads and GMS-placebo heat treatment for 24 h at 50°C using an oven to improve the
beads were previously discussed in Cantor et al. (43,44). tablet crushing force range. Envelope density measurements of
tablets before and after curing were determined using a GeoPyc
Segregation Testing: Blends of Coated Cimetidine Beads 1360 (Micromeritics, Norcross, GA, USA).
and GMS-Placebo Beads
NIR Spectra and Chemometrics
Segregation testing was performed in order to examine
how blends of coated drug beads and GMS-placebo beads Near-infrared spectra were recorded in the diffuse
might behave during tablet press operation. Cimetidine beads reflectance mode on a Model 6500 monochrometer from
coated with either 15% w/w Surelease® or Eudragit® FOSS NIRSystems, Inc. (Laurel, MD, USA) that was
NE30D and GMS-placebo beads were first blended together equipped with a rapid content analyzer module and operated
in a 50:50 w/w ratio (700 g total) in a cross-flow 2-qt through the Vision™ 3.2 software also from FOSS NIRSys-
V-blender for 5 min. A segregation tester was assembled tems. The two faces of each tablet were scanned over the full
according to the ASTM D6940-04 standard (32,46) using the range of 400–2,500 nm with a resolution of 2 nm and
methods described in these references. Fifty-six samples were averaged into one scan. Each individual spectrum was the

Cantor et al.

average of 32 scans. The NIR reflectance spectra of all drug mended by the instrument/software vendor, generally, standard
and excipient powders were scanned in 15×45-mm glass vials error of prediction (SEP) should not be greater than 1.3 times
(National Scientific, Rockwood, TN, USA). PCA and PLS the standard error of calibration (SEC) and the bias should not
were performed using the NIRS data, Matlab® 7.0.4 (The be greater than 0.6 times the SEC (50). High values of SEP or
Mathworks, Natick, MA, USA) and the PLS Toolbox 3.0 bias indicate that the errors are significantly larger for the new
software (Eigenvector Research, Inc., Manson, WA, USA). cross-validation samples and that the calibration data may not
include all the necessary variability or be over fit. A perfect
correlation will yield a slope of 1.0 and a bias (average
NIR Multivariate Calibration Development difference between NIR and laboratory values) of 0. If a large
bias is present, it indicates that there are some systematic errors
For the development of a PLS calibration model, the between the calibration and prediction datasets.
spectra of both tablet faces were averaged and the average
was used for analysis. The appropriate spectral preprocessing
treatments and the best wavelengths of the spectrum were NIR Chemical Imaging
chosen next, i.e., the noisy regions of the spectrum were
omitted from the analysis. The spectral region selected was Hyperspectral images for the cimetidine bead blend
between 1,100 and 2,300 nm since it was observed that the segregation study using different drug bead to placebo bead
noise levels increased after 2,300 nm. ratios were obtained by first scanning three different randomly
At least 100 tablets were used to build calibration models chosen tablet faces from each group (first, middle, and last) of
for compression force, crushing force, and content uniformity tablets. The images were taken using a Sapphire® NIR
(CU), with CU for cimetidine tablets having the largest dataset. Chemical Imaging System (Malvern, Columbia, MD, USA) in
The samples used for this study were prepared from the diffuse reflectance mode. The equipment was calibrated using
previous studies by the authors, Cantor et al. (43–45), and one SapphireGo!®™ software according to the method of Hamad et
consequence of this is that the number of samples available for al. (41) and the manufacture’s recommendations. Using ISys®
the different tests were limited. To develop and validate the (version 3.1, Malvern, Columbia, MD, USA) image analysis
calibration models, the tablet samples were scanned using the software, each corrected spectrum was converted to absorb-
Vision™ software and then the software randomly selected the ance. Next, the spectral data was truncated to a wavelength
samples to be included in the calibration and validation datasets. region between 1,550 and 1,800 nm to reduce both file size and
Finally, the prediction samples were randomly selected as a noise levels. For preprocessing, the second derivative was used
subset of samples used to assess the robustness of the calibration and the spectra were first smoothed using the Savitzky-Golay
model developed. To build a good calibration model, it is algorithm (15 points, fourth order polynomial). The drug bead
important to have a dataset that is large enough to include all and placebo bead spectra were compared and found to be
significant sources of sample variability; and it is also important significantly different in the 1,550–1,800-nm region. Moreover,
that the subsamples have approximately equal representation in within this region, spectra were examined at every 10 nm;
the calibration dataset. To ensure instrument accuracy, perform- therefore, 26 different wavelengths were used for analysis. The
dataset was then mean centered and each spectrum normalized
ance testing was done on a regular basis to verify instrument
to unit variance, i.e., autoscaled giving each spectrum the same
noise level, NIR and visible gain, internal wavelength perform-
intensity weighting. Normalization is useful for removing
ance (wavelength position), and precision. Wavelength lineari-
variability of lighting quality arising from the use of biconvex
zation was performed daily using an internal wavelength
tablets. Tablets composed of 100% drug beads or 100% GMS-
standard. The ceramic reference was scanned at the beginning
placebo beads were used to prepare the spectral library and
of each day and this scan was repeated after every 20–30 scans.
given the same mathematical preprocessing as the tablets from
Moreover, sample orientation was carefully controlled through
the segregation study (different drug bead to placebo bead
the retractable iris on the FOSS® machine. Performing these
ratios). The ISys® software estimated the drug content in each
checks will reduce measurement variability arising from the
pixel using a PCR algorithm.
instrument drift due to measurement procedures (47); however,
spectra should also be examined periodically to see if they are
repeatable. Content Uniformity Analysis
For analysis, outliers were identified by a Mahalanobis
distance algorithm which measures how far a sample is from A 1:1,000 dilution was prepared from tablets, then an
the center of the distribution. A sample is considered an aliquot was removed, placed into 1.2-mL Eppendorf®
outlier when its probability level exceeds the 0.95 threshold centrifuge tubes and centrifuged using a Eppendorf® 5415C
value. A variety of mathematical pretreatments were first centrifuge (Brinkmann Instruments, Inc., Westbury, NY,
tested and the best algorithms were chosen based on the USA) at 13,000 rpm for 3 min. Cimetidine and theophylline
results yielding the lowest statistical errors. When combina- tablets were analyzed spectrophotometrically according to the
tions of math pretreatments were used, they are presented in USP 29/NF 24 at 219 and 268 nm, respectively (51). At least
the order in which they were performed. The compression 15 tablets from each dosage were analyzed individually for
force, crushing force, and content uniformity calibrations drug content and the standard curve had a R2 value ≥0.999.
were developed using PLS regression. To estimate the All tablets were individually labeled and scanned prior to
statistical errors, cross-validation was used (48,49). assay. In order for tablets to be included in either the
In this study, the following criteria were used before calibration or prediction datasets, their assayed drug amount
accepting any calibrations as the best-fit models. As recom- needed to fall between 85% and 115% of their true value.


Fig. 1. NIR second derivative average spectra of theophylline and excipient powders

Reference Method Analysis spectroscopy data as the reference method according to the
following equation:
In order to evaluate the NIR performance, it is sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
important to determine the measurement error of the P
ðD1 À D2 Þ2
UV spectroscopy reference method. This will yield impor- ð1Þ
n
tant information on the precision of the method as well as
determine how accurately NIR predicts the reference where D1 and D2 represent the two blind duplicates, which
values. For this purpose, ten 15-mg theophylline tablets are then squared and summed and n is the number of
(containing 50% w/w uncoated drug beads) were dissolved samples; and in this analysis, n=15.
individually in distilled water and diluted to the desired
concentration using the same procedure as for the content
uniformity assay. The samples were then split into blind Statistical Analysis
duplicates and analyzed by the UV spectroscopy reference
method. The standard error of differences (SED) is also Statistical analysis of the data was performed using
called the laboratory error (49) and was calculated for the UV analysis of variance with least significant difference as the

Fig. 2. NIR second derivative average spectra: theophylline beads and GMS-placebo beads

Cantor et al.

Fig. 3. NIR second derivative average spectra of tablets showing rank order of intensity with theophylline dose

post-hoc test. A paired t test was used to determine the showed the strongest theophylline peaks with minimal excipient
significance between laboratory and predicted constituent interference at 1,624 and 1,664, and at 2,184 nm. Neat cimetidine
values for NIRS and also between envelope density of tablets powder had a greater number of somewhat more intense peaks
before and after curing. A p value less than 0.05 was free from excipient interference as compared to theophylline
considered significant (SPSS v.12, Chicago, IL, USA). and these appeared at 1,186, 1,698, 1,854, and 2,168 nm (spectra
not shown). The peaks with the greatest overall intensity were
due to the GMS in the placebo beads and occurred at 1,214,
RESULTS AND DISCUSSION 1,728, 1,768, 2,314, and 2,350 nm.
The drug and placebo beads were also scanned and
NIR Characterization of Components overlaid in the second derivative plot (Fig. 2) revealing drug
bead peaks of somewhat greater intensity at 1,182, 1,686, and
A NIR feasibility study was first performed using the six 2,264 nm, compared with neat theophylline powder, and with
neat excipients and the drug to check for potential interferences minimal interferences from the placebo beads at those
and to identify characteristic spectral features. Without any wavelengths. The NIR spectral change in peak intensity and
mathematical pretreatment, the excipient spectra have exten- positional shift towards the longer wavelengths could be due
sive overlap. However, the second derivative spectra (Fig. 1) to physical differences in particle size or density between the

Fig. 4. Effect of curing (24 h at 50°C). Moisture loss from 400 mg tablets containing uncoated cimetidine
beads and placebo beads (50:50 ratio)


Table I. Comparison of Crushing Force of Theophylline Tablets (50:50 Ratio)

Crushing Compression
Tablet properties No. Dose (mg) force (kP)a Friability % force (kgf)

350 mg, uncoated beads ambient 24 h (uncured), running press 25 15 3.5±0.8 0.3±0.05 150–250
350 mg, Surelease® coated beads ambient 24 h (uncured), running press 25 15 3.7±0.7b 1.59±0.05b 150–250
350 mg, Surelease® coated beads 50°C 24 h (cured), running press 25 15 5.7±1.0b 0.26±0.07b 150–250
400 mg, uncoated beads ambient 24 h (uncured), hand weighting 25 17.1 4.5±0.5b – 150–250
400 mg, uncoated beads 50°C 24 h (cured), running press 25 17.1 5.4±0.5b – 150–250
400 mg, uncoated beads 50°C 24 h (cured), running press 37 17.1 7.2±1.2b – 150–470
a
Mean ± standard deviation
b
Values are significantly different using Student’s t test (p<0.05)

powders and beads, which could cause the light to be beads had little effect on the crushing force but increased the
reflected or absorbed differently from the bead surface friability. The cured tablets made from coated beads had a
compared to a powder surface (2,52). significantly increased crushing force of 5.7 kP and reduced
It was also desired to evaluate whether NIRS was friability of 0.26% compared to uncured tablets made from
sensitive enough to discriminate between the different uncoated beads. The increase in crushing force and reduction
dosages of theophylline and cimetidine that would be used in friability may be due to the partial melting and re-
in this study. The different dosage strengths were achieved by crystallization of the glycerol monostearate (m.p. 53°C)
varying the ratio of drug beads to GMS-placebo beads. around the drug beads that could occur during the curing
Fifteen spectra were scanned for each dose and the average and cooling process, which may allow the drug beads to
spectrum determined. Even though the difference between become more embedded in the cushioning lipid matrix. For
the five theophylline dosages (10.5 and 19.5 mg) was only uncoated beads, increasing the tablet weight by 50 mg and
2.25 mg, the second derivative plot (Fig. 3) showed a strong curing the tablets increased the crushing force of tablets as
well-separated rank order correlation at 1,214 nm where the expected. Furthermore, when the compression force range
GMS peak decreased as the ratio of theophylline beads to was increased to 150–470 kg, the average crushing force
GMS beads increased. Cimetidine dosages (10.5–19.5 mg) values significantly increased to 7.2 kP; for cured 400 mg
were spaced 3 mg apart and these were also well separated theophylline tablets prepared from uncoated beads.
using NIRS (spectra not shown) with similar spectral relation- The crushing force of cimetidine tablets made from
ships as that seen for theophylline tablets. uncoated beads significantly increased due to the increase in
NIR spectra were used to observe the effect of curing on tablet mass from 350 to 400 mg. The crushing force values for
cimetidine tablets. As shown in Fig. 4, average spectra were the 400 mg theophylline tablets were comparable with
compared before and after curing, and the only difference cimetidine tablets prepared under the same conditions
between the treatments was a water peak at 1,932 nm (Table II).
corresponding to the O–H stretch and H–O–H bending While it was originally hypothesized that tablets contain-
vibrations. ing higher drug bead levels would yield higher crushing force
values, as shown in Table III, the results showed otherwise.
Tablet Properties—Crushing Force and Friability The tablets with the lowest cimetidine dose (10 mg or 35%
drug bead content) had the highest crushing force (7.9 kP),
For the theophylline studies, the tablets were prepared while the 19.5-mg dosage tablets had the lowest crushing
from mixtures of drug beads to GMS-placebo beads. The force (4.4 kP). Therefore, it appears that while the GMS-
350 mg tablets contained 15 mg of drug and the 400 mg placebos are softer and more plastic than the drug beads, they
tablets contained 17.1 mg of drug. As shown in Table I, tablet still play a role in improving tablet crushing force, since the
crushing force and friability were dependent on tablet weight, 10.5 mg dose had the highest percentage (65%) of the
bead coating, tablet curing, and compression force. The GMS-placebo beads. The cimetidine tablet crushing force
average crushing force was quite low (3.5 kP) for 350 mg data was not explained by the envelope density data
uncured theophylline tablets prepared from uncoated beads (Table III). As the dose increased there was only a slight
using a running press to compression forces of 150–250 kg. increase in the tablet envelope density. Further work is
Interestingly, the friability was also low (0.3%) and within needed to determine the reason for these observed differ-
USP specifications of <1.0%. For uncured tablets, coating the ences in crushing force.

Table II. Comparison of Crushing Force of Cured, 15 mg Cimetidine Tablets

Tablet properties No. Dose (mg) Crushing force (kP)a Compression force (kg f)

350 mg, uncoated beads, 50°C 24 h, running press 50 15 5.9±0.7b 200–350
400 mg, uncoated beads, 50°C 24 h, running press 50 17.1 7.8±0.8b 200–350
a
b
Values are significantly different using Student’s t test (p<0.05)

Cantor et al.

Table III. Variations in Crushing Force and Envelope Density with Dosage for Cured, 350 mg Cimetidine Tablets Prepared from Uncoated
Drug Beads on a Running Tablet Press; Compression Force 200–250 kg

Dosage 10.5 mg 13.5 mg 16.5 mg 19.5 mg

Drug bead % 35 45 55 65
Crushing force, kP (n=15) 7.4±0.6a,b 6.8±0.6b 5.9±0.6b 4.4±0.6b
Envelope density, g/cm3 (n=7) 1.45±0.02 1.46±0.0 1.49±0.0 1.47±0.02
a
b
Values in columns are significantly different using paired t test (p<0.05)

Compression Force Models—Regression Analysis Crushing Force Models—Regression Analysis

PCA was used to classify the spectral data and determine In order to obtain a calibration dataset for NIR analysis
outliers, and Mahalanobis distance in principal component having a wider range of crushing force values, the tablet weight
space with an outlier threshold of 0.95 was used as the was increased from 350 to 400 mg and the tablets were cured in
selection criteria for outliers. Initially, the compression force an oven at 50°C for 24 h. All of the same tablets that were used
range was set from 100 to 1,500 kg in order to have a wide in the compression force study were tested for crushing force
ranging calibration set. However, it became clear that high and a 50:50 ratio (corresponding to a 17.1-mg dose) for both
compression forces did not increase tablet hardness, i.e., drugs was evaluated.
tablet hardness plateaued as compression force increased, so The statistical parameters for calibration, validation, and
the compression force range was reduced to between 40 and prediction resulting from the PLS regression analysis for
450 kg. theophylline and cimetidine tablets are shown in Table V.
PLS was used to generate compression force models for Compared to the models for compression force, the model for
theophylline and cimetidine tablets. The statistical parameters predicting theophylline crushing force was better than the
for calibration, validation, and prediction are shown in model for predicting cimetidine crushing force. For theophyl-
Table IV. The PLS model regression results for calibration line tablets, the number of factors was less, the R2 values for
and prediction datasets of compression force for cimetidine calibration and prediction were significantly improved, and
tablets is shown in Fig. 5a and b. Seven factors were needed the SEP/SEC ratio was greater but still less than 1.3. For
for this analysis for both drugs which was the highest number cimetidine tablets, the number of factors was the same, the R2
overall. The R2 was the lowest for theophylline tablets at values for calibration and prediction were significantly less,
0.898, and while the standard error of cross-validation and the SEP/SEC ratio was the same (still greater than 1.3).
(SECV) values are reasonably higher than the SEC values, The first loading for both theophylline and cimetidine tablets
both the SEC and SECV values show the highest percent closely resembled the main spectral peaks for each of the
errors over the whole dataset. A paired t test showed that drug beads as well as for the GMS-placebo beads. Successive
there were significant differences (p<0.05) between the lab loadings also modeled the spectra for the drug beads, GMS-
and predicted NIR values for compression force of both placebo beads, as well as the tablet water content.
theophylline and cimetidine tablets with p=0.013 and 0.016,
respectively.
The SEP increased steadily when theophylline tablets Table IV. Results Summary of the PLS Prediction Models to
made with compression forces greater than 350 kg were Predict Compression Force of Tablets Containing Either Uncoated
included in the dataset; therefore, the prediction maximum Theophylline Beads or Uncoated Cimetidine Beads
was set at 350 kg. However, a wider range was able to be used
Dosage Theophylline Cimetidine
for cimetidine tablets. The SEP values represent approx-
imately 13% to 16% of the average compression force for Calibration tablets 54 70
cimetidine and theophylline tablets, respectively. Also, while Validation tablets 20 24
the bias appears larger for theophylline tablets, when the Preprocessing treatments SNV; SG 9-4-2a SG 9-4-2; SNV
whole dataset is considered (40–350 kg), the error was not Spectral regions (nm) 1,150–2,200 1,160–2,232
considered significant. As a general rule, the bias adjustment Factors 7 7
SEC 30.99 24.85
(for bias only correction) should be less than 0.6 times the
SECV 43.14 46.60
value of SEC and it is for both drugs; in fact, this is the case Calibration R2 0.898 0.941
for all the bias values in Tables IV, V, and VI. Furthermore, Prediction tablets 34 29
the slope adjustment should be 1.0 for a perfect 45° fit to the SEP 34.84 40.51
calibration line and a standard error close to zero. Standard Bias adjustb −14.72 1.19
errors for the slope adjustment for both drugs were approx- Prediction R2 0.907 0.872
imately 10%. Slope and bias adjustments are typically used Slope adjust 0.85±0.08c 0.96±0.10
when transferring calibrations from one NIR machine to Bias adjustd 15.98±17.70 14.97±32.56
another; however, FOSS®, the instrument manufacturer, SEP/SEC 1.12 1.63
does not recommend using slope adjustments. Furthermore, a
Processes provided in order of application
the SEP/SEC ratio should not be greater than 1.3 for good b
For suggested bias correction only
predictability. While the theophylline data met this criterion, c
d
the ratio for cimetidine tablets was 1.63. For suggested slope and bias correction


Fig. 5. PLS calibration (a) and prediction (b) regression datasets for compression force of cimetidine tablets

The SEP values of 0.61 and 0.56 kP for theophylline and clearly lower for the theophylline tablets. Although the SEP/
cimetidine tablets, respectively, are reasonable errors for SEC ratio was higher for theophylline than for cimetidine
crushing force data. These SEP values represent approx- tablets, it was still less than 1.3, indicating good CU
imately 6% to 11% of the average crushing force for predictability (48). The SEP values of 0.37 and 0.49 mg
theophylline and cimetidine tablets, respectively. The PLS represent approximately 2.5% and 4% of the average content
model regression results for calibration and prediction datasets uniformity for theophylline and cimetidine tablets,
of crushing force for cimetidine tablets is shown in Fig. 6a and b. respectively. The cimetidine PLS regression results for
A paired t test showed that there were no significant differences calibration and prediction datasets are shown in Fig. 7a and b.
(p>0.05) between the laboratory and predicted NIR values for Compared to compression force and crushing force plots,
either theophylline or cimetidine crushing force values (p=0.073 the scatter about the PLS calibration and regression lines was
and 0.086, respectively). lowest for the CU data (Figs. 5, 6, and 7). The content
uniformity dataset also showed the highest correlation
coefficients and lowest SEC and SEP values. Therefore, it
Content Uniformity Models—Regression Analysis appears that NIRS is superior at predicting chemical proper-
ties (e.g., CU) over physical (e.g., compression force or
The content uniformity datasets for theophylline con- crushing force); these findings are consistent with other
sisted of five doses ranging from 10.5 to 19.5 mg and four research (22).
doses for cimetidine also ranging from 10.5 to 19.5 mg, with
350 mg used for tablet weights. While theophylline tablets
were each prepared by hand weighing, cimetidine tablets
were manufactured on a running press. For theophylline Table V. Summary of the PLS Prediction Models for the Crushing
tablets, the mid-range dose was produced by mixing uncoated Force of Tablets Containing Either Uncoated Theophylline Beads or
drug beads with placebo beads in a 50:50 ratio, the dose was Uncoated Cimetidine Beads
15 mg, the percent recovery for this dose of uncoated beads in
Dosage Theophylline Cimetidine
distilled water was 100.9±2.8% (n=5). In order to obtain
enough variability in the calibration samples, the dosage Calibration tablets 63 75
range was ±25% of the nominal value; while opinions differ, Validation tablets 19 24
there is general consensus that this offers an acceptable level Preprocessing treatments MSC; SG 5-4-2a SG 9-4-2; SNV
of variation for the NIRS calibration dataset (26). Spectral regions (nm) 1,150–2,200 1,160–2,238
PLS was used to generate regression models for theo- Factors 2 7
SEC 0.47 0.35
phylline and cimetidine tablet drug content. The statistical
SECV 0.51 0.57
parameters for calibration, validation, and prediction are
Calibration R2 0.971 0.916
shown in Table VI. Sample selection showed only 1 PC for Prediction tablets 26 28
content uniformity explaining 96% and 98% of the data SEP 0.61 0.56
variance for theophylline and cimetidine tablets, respectively. Bias adjustb 0.19 −0.10
The mathematical pretreatments for PLS regression models Prediction R2 0.982 0.891
were different for theophylline and cimetidine and it was Slope adjust 1.01±0.04c 0.72±0.07
found that splitting the wavelength regions for theophylline Bias adjustd 0.12±0.40 1.40±0.39
gave the best results with the fewest outliers, see Table VI. SEP/SEC 1.29 1.59
Mathematical preprocessing was also simpler with the cime- a
Processes provided in order of application
tidine dataset. While both theophylline and cimetidine CU b
For suggested bias correction only
PLS results had a reasonable number of factors and excellent c
linearity (R2 >0.98), the SEC, SECV, SEP, and bias were d
For suggested slope and bias correction

Cantor et al.

Table VI. Results Summary of the PLS Prediction Models to Predict uniformity of each dose (i.e., the UV method as indicator of
Content Uniformity of Tablets Containing Either Uncoated Theophylline method precision) was generally below the level required by
Beads or Uncoated Cimetidine Beads the USP 29/NF 24 (i.e., <6%) and ranged from 5.0% to 6.2%
(51). Also, the accuracy of the different dosages of cimetidine
Dosage Theophylline Cimetidine and theophylline, as measured by the percentage error
Calibration tablets 54 95 between the actual average drug content and the theoretical
Validation tablets 20 31 drug content of tablets generally ranged between 1.3% and
Preprocessing treatments BC; SG 9-4-2; SNVa 2nd derivative 10.2% maximum.
10-0 The accuracy is expressed as how close the NIR
Spectral regions (nm) BC, (1,500) SG and 1,110–2,288 predicted values were to the “true” UV assay values. The
SNV, 1,150–1,590, ICH guidelines (53) recommend testing the accuracy by using
1,650–1,790, a minimum of nine determinations over a minimum of three
1,900–2,200
concentration levels; in this case, five doses were used for
Factors 5 6
SEC 0.31 0.47 theophylline and four doses were used for cimetidine to study
SECV 0.43 0.59 tablet content uniformity in the range of 10–20 mg of drug
Calibration R2 0.992 0.981 content per tablet. Comparing the statistical results for both
Prediction tablets 34 30 uncoated, compacted theophylline and cimetidine beads,
SEP 0.37 0.49 respectively, it was found that the standard errors of
Bias adjustb 0.09 0.15 calibration were 0.31 and 0.47 mg, standard errors of
Prediction R2 0.993 0.991 prediction were 0.37 and 0.49 mg, and the bias results were
Slope adjust 0.96±0.02c 0.77±0.31 0.09 and 0.15 mg.
Bias adjustd 0.95±0.02 0.88±0.33 However, it appears that introducing another variable
SEP/SEC 1.20 1.04
into this already complex system, namely coated beads,
a
Processes provided in order of application results in even higher SEC and SEP values as compared to
b
For suggested bias correction only theophylline tablets prepared from uncoated beads
c
Mean ± standard deviation (Table VII). Even though SEC and SEP values were close
d
For suggested slope and bias correction to 1 mg, the number of factors was low and the SEP/SEC
ratio was less than 1.3, which is the “rule of thumb”
A paired t test showed that there was no significant recommended by FOSS® NIR Systems (50).
difference (p> 0.05) between the laboratory results and The loading plots from the content uniformity data for
predicted NIR values for content uniformity of theophylline both theophylline to cimetidine tablets were examined. The
tablets (p=0.44). The standard deviation of the residuals, an first loading displayed distinct peaks at both 1,685 and
indicator of accuracy was also low at 0.35. For cimetidine 1,735 nm for both theophylline and cimetidine tablets. The
tablets, a paired t test showed that there was no significant tablet loading plots were compared with both the NIR spectra
difference (p>0.05) in content uniformity values between the of the drug beads as well as the neat drug. The prominent
laboratory method and predicted NIR values (p=0.18) and peaks in the first loading plots for both drugs were found to
the standard deviation of the residuals was 0.33. Moreover, correspond well with the location of the NIR spectral
the standard error of difference for the reference laboratory peaks of neat theophylline and cimetidine. Moreover, this
method was low and determined to be 0.20 and 0.25 for would indicate that for the tablets used in this content
uncoated and coated drug beads, respectively. The NIR uniformity study, a significant source of variability is due
statistical parameters such as the standard error of calibration to the differences in the drug content; however, other
(SEC), etc., cannot be more accurate than the laboratory factors are also important as the correlation between
reference method’s accuracy as determined by the SED. The spectra and loadings plots was not always completely only
relative standard deviation (%RSD) for the average content from one source such as the drug.

Fig. 6. PLS calibration (a) and prediction (b) regression datasets for crushing force of cimetidine tablets


Fig. 7. PLS calibration (a) and prediction (b) regression datasets for content uniformity of cimetidine tablets

Blanco et al. (54) observed that the compression pressure ships of this type can be used to assess the content uniformity
used for tableting has a profound influence on the NIR spectra, of tablets containing either uncoated or coated drug beads.
with higher pressures causing upward baseline shifts and thus, Using PCA directly has a big advantage over PLS in that you
higher absorbance especially at longer wavelengths. These do not have to perform analytical testing to determine sample
authors also noted that the effect of compaction pressure on drug concentration. While there was good separation
the NIR spectra can even exceed the effect of the variability due between the 10.5, 15, and 19.5 mg theophylline doses, the
to the drug concentration itself. Thus, the compression pressure lowest dose exhibited the greatest variability in the scores
was carefully controlled between 200 and 250 kgf during plot (Fig. 9). For these theophylline spectra where the dose
cimetidine tablet production. Furthermore, Blanco and Alcala was varied, the first PC explains approximately 90% of the
(55) found that compression pressure caused a baseline shift as data variability.
the compression pressure was increased from an uncompressed PCA analysis for uncoated cimetidine beads in tablets
powder to 880 MPa and that the spectral variability of their shows good separation for four doses in the range of 10.5–
tablets was due more to the compression pressure (vis-à-vis 19.5 mg (Fig. 10); these tablets were prepared on a running
principal component 1) than from the mirtazapine concentra- tablet press. For these cimetidine spectra where the dose was
tion (vis-à-vis principal component 2). varied, the first PC explains approximately 98% of the data
variability. However, while the score plots are generally
Tableting Methods—PCA clustered around PC1, there is still some variability along
PC2 (i.e., 1.8%) for the different doses. Based on the loading
Initial studies used theophylline tablets prepared indi- plots, the first loading contains peaks characteristic of the
vidually by hand weighing, it was desired to prepare tablets in drug beads (more so than for the pure drug) and GMS-
bulk using a running press. NIR spectra were collected from placebo beads. While NIRS is used to detect differences in
tablets prepared by both methods and PCA was used to the chemical composition of samples (i.e., dose potency),
examine if these methods produced quantitatively different other factors such as light scattering effects due to the drug
spectra. Figure 8 shows that PCA was able to differentiate bead distribution, density on the tablet faces, and crushing
between tablets prepared by hand weighing and those from a force variations can shift the sample placement in the PCA
running tablet press with the first PC explaining 95% of the score plots. This would require further study as to why the
data variability. Generally, spectra from both tablet faces different dose strengths appear to be shifted along PC2 even
were more similar and tablet appearance was more uniform though PC2 is only a minor contributor to the data variability.
for tablets taken from a running press than from those tablets Also, PCA was used to examine tablets with different
prepared by hand weighing. theophylline doses prepared using Surelease®-coated beads
(Fig. 11). There is still some scatter in each dose even though
Content Uniformity—PCA these tablets were manufactured on a running press. This plot
reflects the high SEC and SEP values observed for PLS
PCA was used as a tool to determine if differences in regression analysis presented in Table VII, and discussed
spectra were related to differences in drug content; relation- above.

Table VII. Results Summary of the PLS Calibration and Prediction Models for Content Uniformity of Coated (15% w/w Surelease®)
Theophylline Beads in Tablets

Factors SEC SECV SEP Bias adjust Prediction R2 SEP/SEC

2 0.86 0.97 0.96 0.01 0.917 1.12

Cantor et al.

Fig. 8. Principal component analysis: 15 mg theophylline tablets (50% w/w uncoated drug
beads) prepared by hand weighing (n=82) versus tablets prepared using a running tablet
press (n=168). The mathematical pretreatments used were square root mean scale
followed by SG 11-4-2 (11 data points; fourth order polynomial; second derivative)

Blend Segregation Studies Using the ASTM 6940-04 beads and their blends with the GMS-placebo beads were
Segregation Tester examined in Cantor et al. (45).
Segregation was determined by comparing the drug
Cimetidine beads coated with either Surelease® (an content between the first and last samples collected. The
ethylcellulose pseudolatex) or Eudragit NE30D® (an acrylic last/first ratio should be 1.0 if no segregation had occurred.
polymeric dispersion) and blended with GMS-placebo beads The data shows some segregation has indeed occurred. The
in a 50:50 ratio were evaluated in an ASTM D6940-04 last/first ratio, based on drug content, was 0.79 and 0.89 for
segregation tester according to the procedures described in blends containing Surelease®-coated beads and Eudragit®
the MATERIALS AND METHODS section and in Cantor et NE30D-coated beads, respectively. The coefficient of varia-
al. (44) and Xie et al. (32). The reason why Surelease® and tion is a measure of the average drug content variability
Eudragit NE30D® were chosen to coat the drug beads is that between the five samples collected from each group: first,
each polymeric coating is known to possess different mechan- middle, and last. Based on this parameter, for these for-
ical properties; i.e., ethylcellulose is a more brittle polymer mulations, the blends with Eudragit® NE30D-coated beads
while the acrylic polymer NE30D is known to be more have a larger CV% than the blends with Surelease®-coated
flexible. The different mechanical properties of these coated beads, see Table VIII.

Fig. 9. Principal component analysis: content uniformity of tablets made from uncoated
theophylline beads, tablets prepared individually by hand weighing (n = 40). The
mathematical pretreatments used were mean centering followed by SG 7-4-2 (7 data
points; fourth order polynomial; second derivative)


Fig. 10. Principal component analysis: content uniformity of uncoated cimetidine beads in
tablets prepared on a running tablet press (n=97). The mathematical pretreatments used
were SNV followed by SG 11-4-2 (11 data points; fourth order polynomial; second
derivative)

Although the particle sizes for the Surelease®-coated lower sphericity of the GMS-placebo beads may also play a
beads, Eudragit® NE30D-coated beads and placebo beads role, by impeding the ﬂow of the more spherical coated
were not practically different at 854, 882, and 858 μm, beads.
respectively, (Cantor et al. (44)), it was noticed that the
Eudragit® NE30D-coated beads were appreciably sticker and Segregation Studies Using NIR Spectra and PCA
possessed a great deal more static charge as compared with
the Surelease®-coated beads. Therefore, it appears that PCA was used in a novel way to examine segregation
factors other than particle size differences between blend phenomena during tablet press operation using different
components play a role in the segregation phenomena. blends of uncoated cimetidine beads. There are many
Perhaps some surface characteristics (i.e., rugosity, smooth- advantages to using PCA, in this case, for example, the
ness, or static charge) between coated beads and the GMS- origin and extent of blend segregation phenomena during
placebo beads could lead to the segregation tendency tablet press operation can be visualized and analyzed in real
observed with the two blends. The hydrophobic nature and time. To test the feasibility of this approach, several drug

Fig. 11. Principal component analysis: content uniformity of coated theophylline beads
(15% Surelease® w/w) in tablets prepared on a running press (n=97). The mathematical
pretreatments used were square root mean scale followed by SG 11-4-2 (11 data points;
fourth order polynomial; second derivative)

Cantor et al.

bead/placebo bead ratios were used, (20:80, 50:50, and
80:20). Both the 20:80 and 80:20 extreme ratios were designed
for the purpose of observing segregation phenomena. While
the tablet press was running, tablets were removed and
placed in sequentially numbered bags and then scanned using
NIRS. Tablets were put into the first, middle, or last groups
depending on the time they were taken from the tablet press.
Based on the separation of the clusters of PCA scores for the
first, middle, and last groups, it appears that the tablets
prepared using the 80:20 ratio showed the most segregation
tendency (Fig. 12). There was a slight separation of the PCA
scores between the first and last groups for the tablets
prepared using the 20:80 and 50:50 ratios. For this analysis,
the same mathematical treatment was used to analyze the
data from both the 50:50 and 80:20 tablet samples; the
selection was based on showing the best visual effect on the
PCA plot of the 80:20 samples. To prove that PCA can
identify segregation, it must be shown that the variability in
Fig. 12 is related to cimetidine content differences; this Fig. 12. Principal component analysis. Segregation study of tablets
assumption is validated in the next section. made from blends of uncoated cimetidine beads and placebo bead.
Three drug bead: placebo bead ratios were used 80:20 blue, 50:50
green, and 20:80 red. Using a running tablet press (n=145), tablets
were collected during the first, middle, and last parts of the run. The
Segregation Studies Using Chemical Imaging
mathematical pretreatments used were SG 15-2-2 (15 data points;
second order polynomial; second derivative) followed by autoscaling
Chemical imaging was able to estimate the drug bead
content with reasonable accuracy. The measured cimetidine
content was consistently higher in all of the first samples
and steadily decreased as the tablet press operation sample and 50:50 and 80:20 ratios, middle samples). However,
progressed (Fig. 13a–c). This color change is illustrated by more tablets would need to be evaluated in order to better
the colored sidebar (Fig. 13), which indicates the relative estimate the variability of the drug bead distribution.
intensity between the drug beads and GMS-placebo beads. When examining the chemical images, it is important to
The blue color represents pixels with spectra similar to that realize that these tablet images are PCR score images; in
of the drug beads, while the red color represents spectra other words, the scores were based on a library of only two
similar to that of the GMS-placebo beads. In the NIR reference materials, uncoated cimetidine beads, and GMS-
region, the GMS had a significantly higher absorbance than placebo beads. Using this reference library, the PCR model is
the other drug or excipient peaks. The findings of decreas- built using spectral feature of a pure “cimetidine bead”
ing drug content are consistent with the ASTM 6940-04 spectrum and a pure “GMS bead” spectrum. Then, for the
Segregation Tester results that had a last/first ratio less sample images, the model converts each pixel’s spectrum to a
than 1.0. linear combination of spectra. The PCR score image is based
It was not possible to separate out the different on the coefficient values for one of the components. Values
components within the individual beads, as they were too close to 1.0 indicate a high GMS content while values
uniformly blended together. Additionally, the drug beads do around 0.4 indicate the lowest GMS content. The percent-
not seem to be evenly distributed throughout the tablets, but age values of cimetidine beads present in Fig. 13 are not
rather appear in clusters. In Fig. 13a, the percentages of drug actually the true percentages but are a reasonable
beads in the tablets from the first part of the tablet press run estimate of average drug bead coefficients for the image,
appear higher than at the later time points. In several based on a pixel’s average score values. There are many
instances, there was also more variability in the amount of factors contributing to the discrepancy between pixel score
drug beads observed on the tablet faces (i.e., 20:80 ratio, first values and actual drug concentration. For example, with
traditional spectroscopy (e.g., UV–visible) the path length
of the light is fixed. Also, the assumption of Beer’s law
Table VIII. Segregation Studies Using a 50:50 Blend of 15% w/w that absorption is directly proportional to concentration
Coated Cimetidine Beads and GMS-Placebo Beads; (A) Surelease® holds true for dilute solutions. However, with diffuse
Coated and (B) Eudragit® NE30D Coated (n=5) reflectance spectroscopy, and especially with chemical

Segregation samples Coefficient of variation (CV%)

A. First/middle 10.5 Fig. 13. Chemical imaging: segregation study of tablets using blends b
A. Middle/last 16.1 of uncoated cimetidine beads at different drug bead: placebo bead
A. First/last 16.7 ratios. Each circle represents one tablet; the blue color represents
B. First/middle 37.4 pixels with spectra similar to that of the drug beads; the red color
B. Middle/last 35.3 represents spectra similar to that of the GMS-placebo beads.
B. First/last 35.3 Representative tablets collected during the first, middle, and last
parts of the run: a 20:80, b 50:50, c 80:20

Cantor et al.

imaging, the sample size for each pixel is very small. This ACKNOWLEDGMENTS
means that the path length can change with wavelength
sample density as well as with the degree and nature of surface The opinions expressed in this article are those of the
reflection, resulting in potential non-linear behavior. authors and do not necessarily reflect the views and policies
Despite these limitations, these images provide a of the US Food and Drug Administration. The authors
representative snapshot of the chemical variability that is gratefully acknowledge research funding from the Parenteral
occurring during tableting and demonstrates another Drug Association (PDA) Fellowship and the Maryland
valuable use for this equipment (studying blend segrega- Consortium for Industrial Pharmaceutics and Training
tion). Moreover, these images can provide a clear visual (CIPET) as well as the generous donation of ethylcellulose
demonstration of how tablets from the early part of from Dow Chemical Co. and Eudragit® samples from Evonik
manufacturing vary from the later tablets in terms of Röhm Degussa GmbH.
quantity of drug beads present, possibly indicating the size
of the drug beads and the distribution of the drug beads
present in the tablets.
REFERENCES

CONCLUSIONS 1. FDA, Guidance for Industry: PAT- a framework for innovative
pharmaceutical development, manufacturing and quality assur-
ance, September 2004. [cited; Available from: http://www.fda.
This study showed that NIRS and chemical imaging can gov/cder/ops/pat.htm].
be used to evaluate extrusion-spheronized drug beads, GMS- 2. Rantanen, Rasanen E, Tenhunen J, Kansakoski M, Mannermaa
placebo beads, blends composed of drug beads and placebo J, Yliruusi J. In-line moisture measurement during granulation
beads, and modified release tablets prepared from these with a four-wavelength near-infrared sensor: an evaluation of
particle size and binder effects. Eur J Pharm Biopharm.
beads. Tablet drug load could be accurately determined using 2000;50:271–6.
univariate analyses of NIR spectra. Multivariate analyses of 3. Reich G. Near-infrared spectroscopy and imaging: basic princi-
NIR spectra could successfully evaluate other tablet attributes ples and pharmaceutical applications. Adv Drug Deliv Rev.
including compaction method, compression force, crushing 2005;57:1109–43.
4. Roggo Y, Chalus P, Maurer L, Lema-Martinez C, Jent N. A
force, and content uniformity. Partial least squares models
review of near infrared spectroscopy and chemometrics in
were used for prediction and principal component analysis pharmaceutical technologies. J Pharm Biomed Anal. 2007;44
was used for classification. Robust calibration models can be (3):683–700.
generated from NIRS data (R2 >0.98) to predict content 5. Yu X, Lionberger RA, Raw AS, D'Costa R, Wu H, Hussain AS.
uniformity with reasonable accuracy. While NIRS showed Applications of process analytical technology to crystallization
processes. Adv Drug Deliv Rev. 2004;56:349–69.
that it is superior in predicting chemical properties over 6. Sandler N, Rantanen J, Heinamaki J, Romer M, Marvola M,
physical, based on SEC and SEP results for tablets, this Yliruusi J. Pellet manufacturing by extrusion-spheronization
technique still has promise for applications such as crushing using process analytical technology. AAPS PharmSciTech.
force measurement. However, higher values of SEC and SEP 2005;6(2):E174–83.
from the content uniformity dataset of complex dosage forms 7. Drennen JK, Lodder RA. Nondestructive near-infrared analysis
of intact tablets for determination of degradation products. J
like the theophylline tablets containing Surelease®-coated Pharm Sci. 1990;79:622–7.
drug beads shows that further work remains to be done to 8. Dziki W, Bauer JF, Szpylman JJ, Quick JE, Nichols BC. The use
improve the statistical parameters, constituent predictability, of near-infrared spectroscopy to monitor the mobility of water
and perhaps even sample homogeneity. within the sarafloxacin crystal lattice. J Pharm Biomed Anal.
2000;22(5):829–48.
Segregation studies of 50:50 blends of either Sure- 9. Zhou X, Hines P, Borer ME. Moisture determination in
lease®-coated or Eudragit® NE30D-coated cimetidine hygroscopic drug substances by near infrared spectroscopy. J
beads and GMS-placebo beads revealed that the Sure- Pharm Biomed Anal. 1998;17:219–25.
lease® blends showed significantly less overall segregation. 10. Frake P, Greenhalgh D, Grierson SM, Hempemstall JM, Rud
Some novel qualitative uses of PCA were evaluated DR. Process control and endpoint determination of a fluid bed
granulation by application of near infrared spectroscopy. Int J
including observing differences in the method of tablet Pharm. 1997;151:75–80.
manufacture, as well as monitoring blend segregation 11. Watano S, Takashima H, Sato Y, Yasutomo T, Miyanami K.
using different ratios of uncoated cimetidine beads/GMS- Measurement of moisture content by IR sensor in fluidized bed
placebo beads. While the PCA results showed that all granulation: effects of operating variables on the relationship
between granule moisture content and absorbance IR spectra.
ratios were distinctly different from each other, analysis of Chem Pharm Bull. 1996;44:1267–9.
NIR spectral scans from tablets collected at the first, 12. Gupta A, Peck GE, Miller RW, Morris KR. Real-time near-
middle, and last parts of the tableting run showed that the infrared monitoring of content uniformity, moisture content,
80:20 ratio displayed the most segregation propensity. compact density, tensile strength, and Young’s modulus of roller
Chemical imaging using NIRS has been shown to be a compacted powder blends. J Pharm Sci. 2005;94(7):1589–97.
13. Han SM, Faulkner PG. Determination of SB 216469-S during
useful and powerful tool to qualitatively and quantitatively tablet production using near-infrared reflectance spectroscopy. J
examine content uniformity differences (e.g., drug bead Pharm Biomed Anal. 1996;14:1681–9.
content) among tablets where blend segregation is sus- 14. O'Neil AJ, Jee RD, Moffat AC. The application of multiple
pected to have occurred during tableting. This study also linear regression to the measurement of the median particle size
of drugs and pharmaceutical excipients by near- infrared
demonstrates the potential of using NIRS data for the spectroscopy. Analyst. 1998;123:2297–302.
rapid and non-destructive prediction of content uniformity 15. O'Neil AJ, Jee RD, Moffat AC. Measurement of cumulative
in multiparticulate tableted systems. particle size distribution of microcrystalline cellulose using near


infrared reflectance spectroscopy. Analyst. 1999;124:33–6. 35. Workman JJ. NIR spectroscopy calibration basics. In: Burns DA,
16. Broad NW, Jee RD, Moffat AC, Smith MR. Application of Ciurczak EW, editors. Handbook of near-infrared analysis. New
transmission near- infrared spectroscopy to uniformity of content York: Marcel Dekker Inc; 2001.
testing of intact steroid tablets. Analyst. 2001;126(12):2207–11. 36. Lyon RC, Lester DS, Lewis EN, Lee E, Yu LX, Jefferson EH, et
17. Chen Y, Thosar SS, Forbess RA, Kemper MS, Rubinovitz RL, al. Near- infrared spectral imaging for quality assurance of
Shukla AJ. Prediction of drug content and hardness of intact pharmaceutical products: analysis of tablets to assess powder
tablets using artificial neural network and near-infrared spectro- blend homogeneity. AAPS PharmSciTech. 2002;3(3):1–15.
scopy. Drug Dev Ind Pharm. 2001;27:623–31. 37. Lee E, Huang WX, Chen P, Lewis EN. High-throughput analysis
18. Cogdill RP, Anderson CA, Delgado M, Chisholm R, Bolton R, of pharmaceutical tablet content uniformity by near-infrared
Herkert T, et al. Process analytical technology case study: part II. chemical imaging. Spectroscopy. 2006;21:24–32.
Development and validation of quantitative near-infrared cali- 38. Lyon RC, Jefferson EH, Ellison CD, Buhse LF, Spencer JA,
brations in support of a process analytical technology application Nasr MM, et al. Exploring pharmaceutical applications of near-
for real-time release. AAPS PharmSciTech. 2005;6(2):E273–83. infrared technology. Am Pharm Rev. 2003;6:62–70.
19. Laasonen M, Harmia-Pulkkinen T, Simard C, Rasanen M, 39. Westenberger BJ, Ellison CD, Fussner AS, Jenny S, Kolinski
Vuorela H. Development and validation of a near-infrared RE, Lipe TG, et al. Quality assessment of internet pharmaceut-
method for the quantitation of caffeine in intact single tablets. ical products using traditional and non-traditional analytical
Anal Chem. 2003;75(4):754–60. techniques. Int J Pharm. 2005;306:56–70.
20. Li T, Donner AD, Choi CY, Frunzi GP, Morris KR. A statistical 40. Cogdill RP, Drennen JK. Spectroscopy of pharmaceutical solids.
support for using spectroscopic methods to validate the content In: Brittain HG, editor. Near-infrared spectroscopy. New York:
uniformity of solid dosage forms. J Pharm Sci. 2003;92:1526–30. Taylor and Francis; 2006. p. 313–412.
21. Trafford AD, Jee RD, Moffat AC, Graham P. A rapid 41. Hamad ML, Ellison CD, Khan MA, Lyon RC. Drug product
quantitative assay of intact paracetamol tablets by reflectance characterized by macropixel analysis of chemical images. J
near-infrared spectroscopy. Analyst. 1999;124(2):163–7. Pharm Sci. 2007;96(12):3390–401.
22. Tatavarti A, Fahmy R, Wu H, Hussain AS, Marnane W, 42. Torrado JJ, Augsburger LL. Tableting of multiparticulate
Bensley D, et al. Assessment of NIR spectroscopy for modified release systems. In: Hoag SW, Augsburger LL, editors.
nondestructive analysis of physical and chemical attributes Pharmaceutical dosage forms: tablets volume 2. 3rd ed. New
of sulfamethazine bolus dosage forms. AAPS PharmSciTech. York: Informa Healthcare; 2008.
2005;6(1):E91–9. 43. Cantor SL, Hoag SW, Augsburger LL. Formulation and
23. Gottfries J, Depui H, Fransson M, Jongeneelen M, Josefson M, characterization of a compacted multiparticulate system for
Langkilde FW, et al. Vibrational spectroscopy for the assessment modified release of water-soluble drugs—part I acetaminophen.
of active substance in metroprolol tablets: a comparison between Drug Dev Ind Pharm. 2009;35(3):337–51.
transmission and diffuse reflectance near-infrared spectrometry. 44. Cantor SL, Hoag SW, Augsburger LL. Formulation and
J Pharm Biomed Anal. 1996;14:1495–503. characterization of a compacted multiparticulate system for
24. Ji Q, Tumuluri V, Wang W, Teelucksingh J, Mecadon M, Vegesna modified release of water-soluble drugs—part II theophylline
R. Rapid content uniformity determination of low dose TCH346 and cimetidine. Drug Dev Ind Pharm. 2009;35(5):568–82.
tablets by NIR. Am Pharm Rev. 2006;9(5):20–6. 45. Cantor SL, Hoag SW, Augsburger LL. Evaluation of the
25. Thosar SS, Forbess RA, Ebube NK, Chen Y, Rubinovitz RL, mechanical properties of extrusion-spheronized beads and multi-
Kemper MS, et al. A comparison of reflectance and trans- particulate systems. Drug Dev Ind Pharm. 2009;35(6):683–93.
mittance near- infrared spectroscopic techniques in determin- 46. ASTM Standard D6940-04 (2004). Standard practice for meas-
ing drug content in intact tablets. Pharm Dev Technol. 2001;6 uring sifting segregation tendencies of bulk solids. West Con-
(1):19–29. shohocken, PA.
26. Morisseau KM, Rhodes CT. Near-infrared spectroscopy as a 47. Beebe KR, Pell RJ, Seasholtz MB. Chemometrics: a practical
nondestructive alternative to conventional tablet hardness test- guide. New York: Wiley; 1998.
ing. Pharm Res. 1997;14:108–11. 48. Savitzky A, Golay MJE. Smoothing and differentiation of data
27. Otsuka M, Yamane I. Prediction of tablet hardness based on by simplified least squares procedures. Anal Chem. 1964;36:
near infrared spectra of raw mixed powders by chemometrics. J 1627–39.
Pharm Sci. 2006;95(7):1425–33. 49. Shenk JS, Workman JJ, Westerhaus MO. Application of NIR
28. Kirsch JD, Drennen JK. Nondestructive tablet hardness testing spectroscopy to agricultural products. In: Burns DA, Ciurczak
by near-infrared spectroscopy: a new and robust spectral best-fit EW, editors. Handbook of near infrared analysis. New York:
algorithm. J Pharm Biomed Anal. 1999;19:351–62. Marcel Dekker Inc; 2001. p. 419–74.
29. Donoso M, Kildsig DO, Ghaly ES. Prediction of tablet hardness 50. Personal Communication, November 2006, FOSS NIRSystems,
and porosity using near-infrared diffuse reflectance spectroscopy as Laurel, Maryland
a nondestructive method. Pharm Dev Technol. 2003;8(4):357–66. 51. United States Pharmacopeia 29/NF 24. 2009, Rockville, MD:
30. Guo J, Skinner GW, Harcum WW, Malone JP, Weyer LG. United States Pharmacopeial Convention.
Application of near- infrared spectroscopy in the pharmaceutical 52. Short SM, Cogdill RP, Wildfong PLD, Drennen III JK,
dosage form. Drug Dev Ind Pharm. 1999;25(12):1267–70. Anderson CA. A near-infrared spectroscopic investigation of
31. Norris KH, Williams PC. Optimization of mathematical treat- relative density and crushing strength in four-component com-
ments of raw near- infrared signal in the measurement of protein pacts. J Pharm Sci. 2009;98(3):1095–109.
in hard red spring wheat. I. Influence of particle size. Cereal 53. FDA, ICH Quality Guidelines for Method Validation, Q2A, Text
Chem. 1984;61:158–65. on Validation of Analytical Procedures, and Q2B, Validation of
32. Xie L, Wu H, Shen M, Augsburger LL, Lyon RC, Khan MA, et Analytical Procedures: Methodology, November 1996, [cited;
al. Quality-by-design: effects of testing parameters and formula- Available from:http://www.ifpma.org/ich5.html].
tion variables on the segregation tendency of pharmaceutical 54. Blanco M, Alcala M, Gonzalez JP, Torras E. A process analytical
powder measured by the ASTM D 6940–04 segregation tester. J technology approach based on near infrared spectroscopy: tablet
Pharm Sci. 2007;97(10):4485–97. hardness, content uniformity, and dissolution test measurements
33. Bokobza L. Near infrared spectroscopy. J Near Infrared of intact tablets. J Pharm Sci. 2006;95(10):2137–44.
Spectrosc. 1998;6:3–17. 55. Blanco M, Alcala M. Content uniformity and tablet hardness
34. Miller CE. Chemometrics in process analytical chemistry. In: testing of intact pharmaceutical tablets by near infrared spectro-
Bakeev KA, editor. Process analytical technology. Oxford: scopy: a contribution to process analytical technologies. Anal
Blackwell Publishing; 2005. p. 226–328. Chim Acta. 2006;557(1–2):353–9.

NIR Paper 2011

More Related Content

What's hot (20)

Similar to NIR Paper 2011 (20)

NIR Paper 2011