jps.21473.pdf

(435 KB) Pobierz

A Near-Infrared Spectroscopic Investigation of Relative

Density and Crushing Strength in Four-Component

Compacts

STEVEN M. SHORT,

ROBERT P. COGDILL,

PETER L.D. WILDFONG,

JAMES K. DRENNEN III,

CARL A. ANDERSON

Duquesne University Graduate School of Pharmaceutical Sciences, 410A Mellon Hall, 600 Forbes Avenue, Pittsburgh,

Pennsylvania 15282

Strategic Process Control Technologies, LLC, 306 Winter Run Lane, Mars, Pennsylvania 16046

Received 7 February 2008; revised 17 April 2008; accepted 16 May 2008

Published online 11 July 2008 in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.21473

ABSTRACT:

Near-infrared spectroscopy (NIRS) is commonly employed for the analysis

of chemical and physical attributes of intact pharmaceutical compacts. Speciﬁcally,

NIRS has proven useful in the nondestructive measurement of tablet hardness or

crushing strength. Near-infrared (NIR) reﬂectance and transmittance spectra were

acquired for 174 13-mm compacts, which were produced according to a four-constituent

mixture design (29 points) composed of anhydrous theophylline, lactose monohydrate,

microcrystalline cellulose, and soluble starch. Six compacts were produced for each

design point by compacting at multiple pressures. Physical testing and regression

analyses were used to model the effect of variation in relative density (and crushing

strength) on NIR spectra. Chemometric analyses demonstrated that the overall spectral

variance was strongly inﬂuenced by anhydrous theophylline as a result of the experi-

mental design and the component’s spectroscopic signature. The calibration for crushing

strength was more linear than the relative density model, although accuracy was poorer

in comparison to the density model due to imprecision of the reference measurements.

Based on the consideration of reﬂectance and transmittance measurements, a revised

rationalization for NIR sensitivity to compact hardness is presented.

2008 Wiley-Liss,

Inc. and the American Pharmacists Association J Pharm Sci 98:1095–1109, 2009

Keywords:

near-infrared spectroscopy; partial least squares; compaction; tableting;

multivariate analysis; chemometrics; hardness; crushing strength; relative density;

tablet

INTRODUCTION

Near-infrared spectroscopy (NIRS) has demon-

strated its utility for the analysis of intact

pharmaceutical dosage systems.

Given the quan-

titative capabilities when used in conjunction

with multivariate calibration, NIRS is frequently

Correspondence to:

Carl A. Anderson (Telephone: 412-396-

1102; Fax: 412-396-4660; E-mail: andersonca@duq.edu)

Journal of Pharmaceutical Sciences, Vol. 98, 1095–1109 (2009)

2008 Wiley-Liss, Inc. and the American Pharmacists Association

employed for the nondestructive prediction of

constituent concentrations within pharmaceuti-

cal compact and tablet matrices. It is well under-

stood, however, that near-infrared (NIR) spectra

convey information pertaining to both the chemi-

cal and physical nature of the samples.

Signals

related to physical variation (e.g., hardness or

crushing strength) are commonly treated as

interferences in composition calibration models.

Generally, variations in physical factors such

as relative density (solid fraction) result in a

characteristic baseline shift,

3–8

the effect of which

1095

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 3, MARCH 2009

1096

SHORT ET AL.

can be suppressed by mathematical treatment.

Two common chemometric algorithms used for

this purpose are standard normal variate (SNV)

scaling and multiplicative scatter correction

(MSC).

Suppression of these physical features

usually reduces the number of model factors

required to achieve optimum performance.

Beyond chemical quantiﬁcation, numerous stu-

dies have documented the modeling of NIR data

for the characterization of physical attributes.

One aspect of solid oral dosage forms that has been

examined using this technology is tablet hardness

or crushing strength. Drennen and Lodder

pioneered the use of NIRS for the measurement

of tablet hardness.

4,11

Subsequently, numerous

publications have demonstrated NIR calibrations

for crushing strength.

3–8,12–24

The majority of

these articles suggest that an increase in tablet

hardness results in a smoother tablet surface,

increasing apparent NIR absorption (presumably

because more light is lost to specular reﬂectance).

Otsuka and Yamane

took a unique approach

in which they generated calibration models to

predict the eventual hardness of tablets produced

at constant compaction pressure from powder

mixtures having varying blend times (hence,

changing the distribution of constituents). While

the authors were able to generate calibration

models having signiﬁcant correlation to tablet

hardness, they were unable to relate spectral

changes to a particular constituent. The authors

suggested that the NIR calibration was detecting

not only composition, but more subtle factors

including porosity, pore structure, and the tablet

surface and geometry.

Three other groups have

investigated the use of NIRS for the analysis of

tablet porosity; all determined that NIRS was

suitable for the measurement of tablet porosity,

reporting varying levels of success with the use of

different mathematical techniques.

16,18,24

The objectives of this work were to demonstrate

(1) characteristic absorption effects of NIR radia-

tion by compacts of varying relative density and

crushing strength, (2) the source of spectral

variability resulting from varying relative density

and crushing strength, (3) how multivariate

analysis can be used to elucidate the effects of

chemical composition upon the prediction of

physical parameters of pharmaceutical solid oral

dosage forms, and (4) how calibration experi-

mental design inﬂuences spectroscopic variance.

Finally, these data are used to present a revised

rationalization for NIR sensitivity to compact

hardness.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 3, MARCH 2009

MATERIALS AND METHODS

Compact Production

The details regarding the production of the

compacts used for this work have been described

elsewhere.

Brieﬂy, a fully-balanced, four-con-

stituent mixture design consisting of anhydrous

theophylline (Lot No. 92577, Knoll AG, Ludwig-

shafen, Germany), Lactose 316 Fast Flo NF

Monohydrate (Lot No. 8502113061, Hansen Labs,

New Berlin, WI), microcrystalline cellulose (MCC,

Avicel PH-200, Lot No. M427C, FMC BioPolymer,

Mechanicsburgh, PA), and soluble starch GR (Lot

No. 39362, EMD Chemicals Inc., Gibbstown, NJ)

was generated. The approximate median particle

size of the theophylline, lactose, MCC, and starch

(reported by documentation from their respective

suppliers), was 90, 100, 180, and 17

mm,

respec-

tively. No further analyses or operations were

performed on the raw materials to determine or

alter their particle size or distribution. Twenty-

nine design points were chosen to cover a wide

composition range and to remove any possibility of

factor aliasing (Tab. 1). The mixture covariance

matrix demonstrates that the design is balanced

in all factors, giving equal emphasis to all

constituents.

Materials for each design point mixture were

dispensed by weight (Data Range, Model No.

AX504DR, Mettler Toledo, Columbus, OH), and

subsequently transferred to 25 mL glass scintilla-

tion vials. In total, 6000 mg of material was

weighed out for each point, and the nominal

weights for all constituents were adjusted to the

observed mass data to calculate actual concentra-

tion. The vials were mixed for 5 min cycles by

placing them on the rotating drive assembly of a

Jar Mill (US Stoneware, East Palestine, OH).

After each blending period, a NIR reﬂectance

spectrum was acquired through the bottom of

each vial (FOSS NIRSystems 5000, FOSS NIR-

Systems, Inc., Laurel, MD). Once all the vials

underwent this mixing cycle and their correspond-

ing spectra were acquired, an

ad hoc

partial least-

squares II (PLS-2) calibration was constructed to

assess homogeneity. Mixtures were assumed to be

homogeneous when further mixing failed to yield

an increase in the calibration’s coefﬁcient of

determination.

The mixtures from each design point were then

subdivided and compacted at one of ﬁve pressures

(67.0, 117.3, 167.6, 217.8, and 268.1 MPa) using

a Carver Automatic Tablet Press (Model No.

DOI 10.1002/jps

NIR INVESTIGATION OF RELATIVE DENSITY IN COMPACTS

1097

Table 1.

Concentration Design for the 4-Component Compacts

Design

Point

Anhydrous

Theophylline (w/w)

0.600

0.400

0.200

0.400

0.200

0.600

0.400

0.200

0.600

0.400

0.200

0.000

0.400

0.200

0.000

0.200

0.000

0.400

0.200

0.400

0.200

0.000

0.200

0.000

0.200

0.000

0.250

Lactose

Monohydrate (w/w)

0.200

0.400

0.600

0.200

0.400

0.200

0.400

0.600

0.000

0.200

0.400

0.600

0.000

0.200

0.400

0.000

0.200

0.400

0.000

0.200

0.400

0.000

0.200

0.000

0.200

0.250

MCC

(PH-200) (w/w)

0.200

0.400

0.600

0.000

0.200

0.401

0.400

0.600

0.000

0.200

0.400

0.000

0.200

0.250

Soluble

Starch (w/w)

0.000

0.200

0.400

0.399

0.400

0.600

0.250

3887.1SD0A00, Wabash, IN) equipped with

13 mm ﬂat-faced punches and die. A dwell time

of 10 s was employed. Six compacts weighing

approximately 800 mg were produced per design

point, with the sixth tablet’s compaction pressure

pseudo-randomly selected from one of the ﬁve

possible levels, for a total of 174 compacts.

The compaction order was randomized to mini-

mize heteroscedastic errors. Following compac-

tion, the samples were retained in the sealed vials

for 15 days prior to spectroscopic analysis.

Compacts consisting of each pure component

were produced in a similar manner. Approxi-

mately 800 mg of each component was compacted

at nine different compaction pressures (67.0,

90.5, 117.3, 140.8, 167.6, 191.0, 217.8, 241.3,

and 268.1 MPa) using the same press and tooling.

Four additional pressures were employed to

increase the number of data points in each

compaction proﬁle. Three replicate compacts were

produced at every compaction pressure for each

DOI 10.1002/jps

constituent, yielding 27 pure compacts per mate-

rial. The manufacturer’s lot of lactose monohy-

drate used for the compaction proﬁles differed

from that used to make the 4-component compacts

(Lot No. 8505010961, Hansen Labs, New Berlin,

WI); all other materials were from the aforemen-

tioned lots.

Data Acquisition, Instrumentation, and Software

Near-infrared reﬂectance measurements were

acquired for both sides of each compact (excluding

pure component compacts) using a scanning

monochromator instrument, equipped with a

Rapid Content Sampler, over the wavelength

range of 1100–2498 nm at a 2 nm increment,

averaging 32 scans (FOSS NIRSystems 5000-II,

Vision version 2.00, FOSS NIRSystems, Inc.). Two

ad hoc

partial least-squares II (PLS-2) calibra-

tions, using the constituent concentrations as

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 3, MARCH 2009

1098

SHORT ET AL.

reference data, were constructed from reﬂectance

spectra corresponding to a speciﬁc surface of the

tablets. Since the coefﬁcients of determination did

not differ until the third decimal place, it was

decided to only consider measurements for one

compact face.

Transmittance measurements were acquired on

a scanning monochromator instrument equipped

with an InSight

Tablet Analyzer over the wave-

length range of 600–1898 nm at a 2 nm increment,

averaging 32 scans (FOSS NIRSystems 6500,

Vision version 2.00, FOSS NIRSystems, Inc.). The

wavelength range of the transmittance spectra

was truncated to 800–1400 nm due to limitations

imposed by the sample pathlength.

All spectral data were analyzed in the Matlab

environment (version 7.1, The MathWorks, Natick,

MA) using the PLS_Toolbox (version 3.0, Eigen-

vector Research, Inc., Manson, WA) and software

developed in-house.

acquired. The true density of each constituent was

estimated via helium pycnometry (Micromeritics

Accupyc, Model No. 1330, Particle & Surface

Sciences Pty. Limited, Gosford, New South Wales,

Australia). The mean of ﬁve powder subsamples

was used for each constituent (Tab. 2). The true

densities (r

true

) of each compact were estimated

using the equation

true

i¼1

true;i

(2)

where

and

true,i

are the w/w contribution and

the true density for the

ith

component within an

n-component

sample. For the work herein,

was

either 4 or 1 corresponding to the mixture and pure

component compacts, respectively. The relative

density (D) of each mixture and pure component

compact was estimated as

comp

true

(3)

Physical Testing

Following compaction, the samples were stored in

sealed glass scintillation vials and were removed

only for analysis. After an approximately 40-day

span to allow for radial expansion, the compacts

were weighed (Data Range, Model No. AX504DR,

Mettler Toledo) and their thicknesses and dia-

meters were measured using a digital micrometer

(TESA Micromaster, Model No. IP54, Brown &

Sharpe, North Kingstown, RI). Assuming a cylin-

drical geometry, these data were used to estimate

each compact’s density (r

comp

) according to the

formula

comp

pðd=2Þ

(1)

The crushing strength of the compacts, reported

in kiloponds (kp), was estimated for the mixture

compacts by a diametric crushing test (Vision

Tablet Testing System, Model No. ElizaTest 3þ,

Elizabeth-Hata International, North Hunting-

don, PA). The maximum recordable value for

this particular instrument was 55.9 kp. Twelve

calibration compacts and one test compact were

evaluated to have values of (at least) 55.9 kp;

one other test sample yielded a value of 0.0 kp. The

information pertaining to these 14 compacts was

withheld from the subsequent crushing strength

modeling but was included in the relative density

analyses.

where

m, d,

and

are compact mass, diameter, and

thickness, respectively. The pure component

tablets were allowed to relax for approximately

55 days before their masses and dimensions were

Regression Analyses

Near-infrared reﬂectance and transmittance spec-

tra were independently modeled in an identical

manner. Partial least-squares (PLS) regression

Table 2.

Component True Densities as Determined by Helium Pycnometry

Component

True density (g/cm

)

True density (g/cm

)

True density (g/cm

)

True density (g/cm

)

True density (g/cm

)

Average true density (g/cm

)

Standard deviation (g/cm

)

Theophylline

1.4100

1.4071

1.4034

1.4024

1.4012

1.4050

0.0036

Lactose

1.5063

1.5072

1.5076

1.5074

1.5070

0.0005

MCC

1.5084

1.508

1.5053

1.5058

1.5070

0.0015

Starch

1.4770

1.4768

1.4766

1.4767

1.4770

0.0002

DOI 10.1002/jps

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 3, MARCH 2009

NIR INVESTIGATION OF RELATIVE DENSITY IN COMPACTS

1099

was used via the SIMPLS algorithm

to relate

spectroscopic response to relative density and

crushing strength. No spectral preprocessing was

employed; data were only transformed to absorb-

ance (log(1/R) or log(1/T)) and subsequently mean-

centered. All reference data (relative density and

crushing strength) were scaled to zero mean and

unit variance prior to modeling. Given that the

calibrations were intended to model the phys-

ical variance within the spectra, preprocessing

routine(s) were not applied. Certain applications

may necessitate spectral preprocessing to sup-

press interfering signals (e.g., spectrometer drift);

however, implementation of such methods may

reduce the net analyte signal

of a feature. For

this work, preprocessing routines, including SNV

scaling, detrending, derivatives, and combina-

tions of the preceding, were tested at the outset;

all of these pretreatments reduced the ability to

predict relative density and crushing strength

from NIR reﬂectance and transmittance spectra.

The optimum model was selected based on

minimization of ‘‘batch-wise’’ cross-validation

error,

where the batches in this instance

corresponded to the 29 different concentration

levels. The root-mean-standard error (RMSE) for

cross-validation (RMSECV), calibration (RMSEC),

and testing (RMSET) were calculated using the

formula

vﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

ðy

i¼1

RMSE

(4)

where

is the measured parameter,

is the

predicted parameter, and

is the number of

samples for the data set under consideration. The

testing data set consisted of the sixth compact

from each design point, whose compaction pres-

sure was pseudo-randomly assigned one of the ﬁve

possible levels. While this does not constitute a

truly independent dataset for model validation

(i.e., for use in process control), the course of action

is suitable for exploratory analyses such as this.

RESULTS AND DISCUSSION

Optical Effects of Varying Compaction Pressure

The optical effects of varying compaction pres-

sure, which elicits change in the physical para-

meters of the samples, are difﬁcult to visualize

amongst the broad chemical variation built into

the design. Thus, Figure 1a and b displays the

characteristic baseline shifts for reﬂectance and

transmittance spectra associated with changes in

tablet density when the percent contributions of

the constituents are unchanging. The baseline

slope increases with increasing compaction pres-

sure for reﬂectance spectra while the opposite

trend is observed for transmittance measure-

ments.

The characteristic increase in measured absor-

bance as a result of an increase in compaction

pressure for NIR reﬂectance spectra (Fig. 1a) is

consistent with the results published over the

last 15 years.

3–8,12–24

An increase in compaction

Figure 1.

Raw NIR absorbance spectra illustrating the spectral effect of compaction

pressure when concentration is unchanging for reﬂectance (a) and transmittance

(b) measurements. The design point illustrated is 40% theophylline, 40% lactose,

20% MCC, and 0% starch. The intensity recorded at the ﬁrst wavelength of each scan

was subtracted from all remaining wavelengths to facilitate viewing.

DOI 10.1002/jps

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 3, MARCH 2009

Plik z chomika:

InzChemiczna

Inne pliki z tego folderu:

jps.21452.pdf (335 KB)
jps.21450.pdf (306 KB)
jps.21462.pdf (561 KB)
jps.21451.pdf (377 KB)
jps.21456.pdf (284 KB)

jps.21473.pdf

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: