Quantification of paracetamol in intact tablets ... - Pharmacy Eprints

Quantification of paracetamol in intact tablets using
near-infrared transmittance spectroscopy
Alba Eustaquio, a Paul Graham, b Roger D. Jee,* c Anthony C. Moffatt c and Andrew D.
Trafford c
a Department of Chemistry, Analytical Unit, University Autonoma of Barcelona, 08193
Bellaterra, Barcelona, Spain
b Sanofi Research Divison, Alnwick Research Centre, Willowburn Avenue, Alnwick,
UK NE66 2JH
c Centre for Pharmaceutical Analysis, School of Pharmacy, University of London, 29–39
Brunswick Square, London, UK WC1N 1AX
Received 15th June 1998, Accepted 4th September 1998
Production batch samples of paracetamol tablets and specially prepared out-of-specification batches covering the
range 90–110% of the stated amount (500 mg) were analysed by the BP official UV assay and by NIR
transmittance spectroscopy. NIR measurements were made on 20 intact tablets from each batch, scanned five
times each (10 min measurement time per batch) over the spectral range 6000–11 520 cm 21. An average spectrum
was calculated for each batch. Partial least squares (PLS) regression models were set up using a calibration set (20
batches) between the NIR response and the reference tablet paracetamol content (UV). Various pre-treatments of
the spectra were examined; the smallest relative standard error of prediction (0.73%) was obtained using the first
derivative of the absorbance over the full spectrum. Only two principal components were required for the PLS
model to give a good relationship between the spectral information and paracetamol content. Applying this model
to the validation set (15 batches) gave a mean bias of 20.08% and a mean accuracy of 0.59% with relative
standard deviations of 0.75 and 0.44%, respectively. The proposed method is non-destructive and therefore lends
itself to on-line/at-line production control purposes. The method is easy to use and does not require a knowledge
of the mass of the tablets.
The use of near-infrared (NIR) spectroscopy has grown
dramatically over the past two decades, chiefly owing to the
advances in the robustness and ease of use of the instrumentation
and the availability of high quality chemometrics with
which to process the data. In addition, NIR spectroscopy can
often be used to analyse samples directly without the need for
any preparation. These attributes have made the technique well
suited for production control purposes for many different
products, including foodstuffs, 1 textiles, 2 polymers 3 and pharmaceuticals.
4
As early as 1968, Sherken 5 of the US Food and Drugs
Administration reported the use of an NIR based assay for
meprobamate in tablet mixtures and commercially available
preparations, but it involved grinding the tablets before analysis.
The accuracy of the NIR method was comparable to that of the
official National Formulary method.
Since then, NIR spectroscopy has been growing in use in the
pharmaceutical industry both on the production floor and in the
laboratory and has been used for the analysis of powder blends,
granules and milled tablets. 6 A large step forward was the
development of methods to analyse solid dosage formulations
without grinding or extraction, and in 1987 Lodder et al. 7
analysed adulterated and unadulterated capsules using NIR
spectroscopy directly. More recently, the identification of active
drugs in tablets has been reported. 8 The potential for on-line or
at-line identification, qualification and quantification of the
final dosage form therefore became a reality.
As the use of NIR spectroscopy grew, the importance of
sample presentation was recognised, particularly in the positioning
of the dosage form in the spectrometer. Specially constructed
reflective aluminium sample cells, designed for capsules 7 or
tablets, 9 appeared for use in reflectance spectroscopy.
The use of transmittance measurements has brought the
advantage over reflectance NIR spectroscopic measurements
that it is less sensitive to the inhomogeneity of the material,
which is a problem in reflectance analyses. 10 Work in this mode
of NIR spectroscopy has been slow to develop, but the new
generation of transmittance instruments that have become
available recently offer the potential for greater accuracy than
with conventional reflectance instruments.
Processing NIR data can be carried out in a number of ways
to simplify the spectral information. One transformation of the
data is by using principal component analysis, which decreases
the problems created by sources of spectral noise and by the
highly correlated nature of NIR wavelength data. 11 Transformation
of the spectral data to principal axes simplifies the selection
of variables for regression models and permits variables to be
added or deleted from the regression equation without changing
the coefficients of the remaining variables. One method for
quantitative analysis based on this mathematical transformation
is partial least squares (PLS) regression. This attempts to
explain as much of the observed variation in the dependent
variables as possible using the minimum of relevant factors
contained in the spectral data. 12
The recent appearance on the market of a commercial semiautomatic
NIR transmittance spectrometer specifically designed
for the analysis of intact tablets led us to try it for the
quantification of paracetamol in tablets as a model example.
PLS was used to treat the spectral data for the reasons given
above.
Analyst, 1998, 123, 2303–2306 2303

Experimental
Reagents and samples
All reagents were of analytical-reagent grade and were obtained
from BDH (Poole, Dorset, UK).
Forty-five batches of paracetamol tablets were used for the
study, of which 35 were normal production batches of Sterwin
500 mg paracetamol tablets (nominal content 84.175% m/m
paracetamol) with a nominal diameter of 12.7 mm and thickness
of 3.80 to 4.15 mm and 10 were development batches (76–92%
m/m paracetamol, i.e. 90–110% of the nominal label content).
The development batches were of the same dimensions as the
normal production samples but were specially manufactured for
this study at a pilot scale on a small press, by altering the content
of paracetamol and the major excipient (starch).
Apparatus
NIR spectroscopy. Transmittance spectra were measured
using a NIRTAB system (Buhler, Anatec, Uzwill, Switzerland)
designed for tablet analysis. It consisted of a NIRTAB FT-NIR
spectrometer and autosampler unit linked by an optical fibre
connection. The autosampler had an adjustable multi-layer plate
system with holes around its periphery to take up to 40 samples
of varying sizes, which ensured that each sample was
reproducibly located central to the light beam. NIRCAL
software version 2.0 (Buhler, Anatec) was used to control the
system and process the data. Spectra were also transferred from
NIRCAL to Excel (Microsoft Excel V7.0a) for some data
transformations.
UV spectrophotometry. A Lambda 15 UV/VIS spectrophotometer
(Perkin-Elmer, Beaconsfield, Buckinghamshire, UK)
was used for the UV assay of the tablets.
Analytical procedures
NIR measurements. The paracetamol tablets were placed on
the adjustable plate system of the autosampler and the diameter
of the holes in the plate system was adjusted to centralise the
tablets and ensure that there was no stray light during
measurement. Tablets were always placed with the same face up
and their transmittance spectra measured automatically using a
macro in the NIRCAL software. The measurement time for each
tablet was 3 s per scan and the instrument was operated at a
resolution of 12 cm 21 between 6000 and 11 520 cm 21.
Normally 20 tablets from each of the batches were measured
using five scans for each tablet and a mean spectrum for the
batch was calculated from these 100 spectra. A sample of
Spectralon (5 mm thick) was used as the reference before
scanning of each batch. The method allowed the measurement
of a production batch in about 10 min by using the programmable
autosampler.
UV measurements. The UV assay used as the reference
method to quantify paracetamol in the tablets was that of the
British Pharmacopoeia 1993. 13
Results and discussion
Selection of number of scans
Preliminary experiments were conducted to find the optimum
number of scans for each tablet. Fig. 1 shows the results of
measuring four sets of 20 tablets (all from the same batch) using
2304 Analyst, 1998, 123, 2303–2306
1, 5, 10 and 20 scans to calculate the mean spectrum. The mean
spectrum obtained using only one scan per tablet is too noisy
(although it is the mean of 20 single tablet spectra) for
quantitative work; however, using 5, 10 or 20 scans per tablet
gave acceptable spectra. Five scans were chosen as the optimum
to obtain a smooth spectrum in as short a time as possible, so
limiting the exposure of the tablets to NIR radiation, which
might cause thermal degradation.
Mean NIR transmission spectra for the paracetamol batches
of 90 and 110% of the stated amount are shown in Fig. 2.
Compensation for different tablet mass
Since the NIR absorption by a tablet is both content and
pathlength dependent, the effect of different tablet masses and
heights was investigated. The diameter of the tablets for a given
batch is effectively fixed by the punch and dies used. Twenty
tablets from the same batch were weighed and the heights
measured with a micrometer to establish the relationship
between the height and mass of the tablets (Fig. 3). The
relationships between absorbance (at 9300 cm 21) and tablet
height and mass are given in Fig. 4 and 5, respectively. For
individual tablet measurements it would therefore be sufficient
simply to measure the mass of each tablet and use it to allow for
variations in tablet thickness.
In this work it was decided to use the mean spectrum from 20
tablets to remove most of the differences between individual
tablets. However, spectral corrections were considered when
developing the PLS model (see below) to take into account
batch variations by using
Fig. 1 Part of the mean NIR transmittance spectra for 20 tablets using
different numbers of scans: (a) 1; (b) 5; (c) 10; and (d) 20 scans.
Fig. 2 Mean NIR transmittance spectra for paracetamol tablets (500 mg):
(a) 90% of stated amount; and (b) 110% of stated amount.

S S m
* =
m
(1)
t
where S* is the mass corrected spectrum, S the original
spectrum, m the mean tablet mass of 20 tablets for the batch, and
mt the theoretical tablet mass for production. The correction
factor was applied to each spectral data point across the whole
spectrum.
Construction of the PLS models
The 35 batches of paracetamol tablets were divided into two
sets: calibration and validation sets. The calibration set was used
to construct the PLS regression and comprised 20 batches that
gave a good spread of values from 90 to 110% of the stated
amount. It included both production batches and the specially
Fig. 3 Relationship between tablet mass and tablet height for paracetamol
tablets. All tablets from the same batch.
Fig. 4 Relationship between NIR absorbance (at 9300 cm 21 ) and tablet
height for paracetamol tablets. All tablets from the same batch.
Fig. 5 Relationship between NIR absorbance (at 9300 cm 21) and tablet
mass for paracetamol tablets. All tablets from the same batch.
produced batches to give the largest variation in paracetamol
content and types of tablets expected in practice.
The validation set was used to evaluate the prediction
capability of the PLS model constructed from the calibration
set. Fifteen batches that were not used in the calibration set, and
lay between 95 and 105% of the stated amount, were used in this
set.
PLS regression models were calculated using centred data
and the number of PLS principal components used was selected
to give the minimum residual error sum squares for the
validation set. To compare the results from all the different
models constructed, the relative standard error of prediction
(RSEP) was used:
RSEP (%) =
n
Â
i = 1
( C - C )
n
Â
i = 1
UV NIR
C
2
UV
2
¥ 100 (2)
Where C is the percentage paracetamol content as measured by
the UV (reference) and NIR methods and n the number of
samples. The RSEP value gives an estimation of the mean
percentage error for each set and the results for all the models
constructed are given in Table 1.
Three different spectral ranges were used, viz. the full
spectrum (6000–11 520 cm 21) and the ranges 7692–8400 cm 21
and 9096–11 520 cm 21 where the spectra showed the maximum
differences due to paracetamol content (Fig. 2). The results
obtained using the two restricted spectral ranges did not show
any significant differences from or advantages over using the
full spectrum, which generally gave the smaller error of
prediction. Using the full spectrum also had the advantage that
small spectral variations have a minimal effect on the
quantification.
In an attempt to improve the predictive capacity of the
spectral information, various pretreatments were tried before
the PLS model was constructed (Table 1). These spectral
pretreatments were available using the NIRCAL software and
Table 1 Relative standard error of prediction (RSEP) for different PLS
models
Spectral range
6000– 7692– 9096–
11 520 cm 21 8400 cm 21 11 520 cm 21
Pre-treatment a Cal. b Pred. c Cal. Pred. Cal. Pred.
Transmittance 0.70 d 0.88 0.66 d 0.95 0.64 d 0.86
Transmittance-m 0.65 d 0.84 0.67 d 0.97 0.64 d 0.84
Absorbance 0.78 e 0.80 0.71 d 0.84 0.79 e 0.81
Absorbance-m 0.71 d 0.79 0.62 d 0.85 0.79 d 0.79
Absorbance-D1 0.31 e 0.73 0.53 e 0.94 0.78 e 0.77
Absorbance-D1-m 0.33 e 0.75 0.53 e 0.91 0.80 e 0.74
Absorbance-D2 0.78 e 0.80 0.60 d 0.87 0.79 e 0.81
Absorbance-D2-m 0.40 e 0.92 0.44 d 1.06 0.27 f 1.22
Transmittance-S 0.64 d 0.86 0.68 d 0.95 0.64 d 0.86
Transmittance-S-m 0.65 d 0.84 0.71 d 0.97 0.65 d 0.84
Absorbance-S 0.78 e 0.80 0.63 d 0.88 0.77 d 0.81
Absorbance-S-m 0.72 d 0.79 0.65 d 0.86 0.79 d 0.80
Transmittance-MSC 0.71 e 0.75 — — — —
Transmittance-MSC-m 0.71 e 0.75 — — — —
Absorbance-MSC 0.77 e 0.75 — — — —
Absorbance-MSC-m 0.81 d 0.73 — — — —
a m = Spectra corrected for mass; D1 = first-derivative spectra;
D2 = second-derivative spectra; S = Savitzky–Golay smoothing of spectra;
MSC = multiplicative scatter correction. b Calibration data set. c Prediction
data set. e PLS model using three PLS principal components. f PLS model
using two principal components; f PLS model using four principle
components.
Analyst, 1998, 123, 2303–2306 2305

were transmittance (T), absorbance [log (1/T)], first and second
derivatives of absorbance (Savitzky–Golay, nine points, cubic
polynomial), smoothing (Savitzky–Golay, nine points, cubic
polynomial) of both transmittance and absorbance and multiplicative
scatter correction (using the full spectral range). In
addition, spectra were also corrected for the mass of the tablets
(see above). First and second derivatives of the transmittance
spectra were not used since the transformed data exhibited
excessive noise.
Selection of the best model
There were no large differences between any of the pretreatments
and almost all of them give an error of less than 1%
(Table 1). That the PLS models only used two or three
components meant that there was a good relationship between
the spectral information and paracetamol content. Correcting
for the mass of the tablets did not appear to improve the
accuracy; presumably taking the mean spectrum for 20 tablets
in a batch effectively removes physical variability between
individual tablets.
The best results were obtained using the first derivative of the
absorbance spectrum or the absorbance spectrum after correction
for multiplicative scatter and tablet mass. Both gave an
RSEP of 0.73% for the validation set. However, the former is
more robust and does not require weighing the tablets, and
consequently was the preferred quantification model.
Quantification of paracetamol tablets
Using the first derivative of NIR absorbance and a two principal
component PLS model, the amount of paracetamol in the 35
batches of tablets was calculated and compared with the results
from the UV assay (Table 2). The percentage error was low with
a maximum value of 21.18%. The bias and accuracy (with
respect to the UV value) for the validation batches (n = 15)
were calculated using
È n
˘
( NIR value - UV value)
Mean bias (%) = Í
¥ 100 ˙ / n
Í UV value
˙
(3)
ÎÍ
=Âi 1
˚˙
Mean accuracy (%) =
È n
˘
Í NIR value - UV value
¥ 100˙
/ n
Í UV value
˙
(4)
ÎÍ
=Âi 1
˚˙
The mean bias was 20.08% and the mean accuracy was 0.59%
with standard deviations of 0.75 and 0.44%, respectively.
Conclusions
NIR spectroscopy has been shown to be a viable alternative to
UV spectrophotometry for the assay of paracetamol in paracetamol
tablets and it takes only 20 min to analyse a batch once
the calibration model has been set up. The proposed firstderivative
absorption model is easy to use and does not need a
knowledge of the mass of the tablets. It is a non-destructive
method and thus lends itself very well for on-line/at-line
production control purposes.
We thank Mr. Keith Longdon and Buhler for the loan of
NIRTAB and NIRCAL. Andrew Trafford thanks Sanofi Lilly
and Bristol-Myers Squibb for a Research Studentship. Alba
Eustaquio acknowledges the Ministry of Education and
Sciences, Spain, for the award of an FPI grant.
2306 Analyst, 1998, 123, 2303–2306
Table 2 Comparison of the paracetamol content as determined using the
UV reference method and NIR spectroscopy (two principal component PLS
model using first-derivative absorbance)
References
Paracetamol (% of stated amount)
Sample a UV assay NIR assay Error (%)
1c 90.73 90.45 20.31
2c 92.65 92.33 20.35
3 95.37 94.49 20.92
4c 97.63 97.25 20.39
5c 99.20 99.59 0.39
6 99.72 100.14 0.42
7c 99.85 100.28 0.43
8c 99.88 100.18 0.30
9 99.88 100.08 0.20
10c 99.97 100.68 0.71
11 100.02 101.11 1.09
12c 100.03 100.37 0.34
13 100.07 100.14 0.07
14c 100.17 100.29 0.12
15 100.20 100.35 0.15
16c 100.21 100.37 0.16
17 100.26 100.50 0.24
18c 100.27 100.34 0.07
19 100.33 101.39 1.06
20c 100.41 100.41 0.00
21 100.61 101.11 0.50
22c 100.73 100.77 0.04
23 100.87 100.09 20.77
24c 101.02 100.85 20.17
25 101.06 100.10 20.95
26c 101.09 101.16 0.07
27 101.11 101.21 0.10
28 101.57 100.37 21.18
29c 101.74 101.72 20.02
30 101.78 100.62 21.14
31c 101.88 101.44 20.43
32c 102.27 102.12 20.15
33 105.17 105.11 20.06
34c 107.87 106.97 20.83
35c 110.23 110.28 0.05
a c, Calibration set; other samples, validation set.
1 B. G. Osborne, T. Fearn and P. H. Hindle, Practical NIR
Spectroscopy with Applications in Food and Beverage Analysis,
Longman, Harlow, 1986.
2 M. Blanco, J. Coello, H. Iturriaga, S. Maspoch and E. Bertran,
Analyst, 1994, 119, 1779.
3 K. A. Bunding Lee, Appl. Spectrosc. Rev., 1993, 28, 231.
4 J. K. Drennen and R. A. Lodder, in Advances in Near-Infrared
Measurements, ed. G. Patonay, Jai Press, Greenwich, CT, 1993, vol.
1, pp. 93–112.
5 S. Sherken, J. Assoc. Off. Anal. Chem., 1968, 51, 616.
6 J. D. Kirsch and J. K. Drennen, Appl. Spectrosc. Rev., 1995, 30,
139.
7 R. A. Lodder, M. Selby and G. M. Hieftje, Anal. Chem., 1987, 59,
1921.
8 P. R. Khan, R. D. Jee, R. A. Watt and A. C. Moffat, Pharm. Sci., 1997,
3, 447.
9 R. A. Lodder and G. M. Hieftje, Appl. Spectrosc., 1988, 42, 556.
10 J. Gottfries, H. Depui, M. Fransson, M. Jongeneelen, M. Josefson,
F. W. Langkilde and D. T. Witte, J. Pharm. Biomed. Anal., 1996, 14,
1495.
11 I. A. Cowe and J. W. McNichol, Appl. Spectrosc., 1985, 39, 257.
12 M. A. Sharaf, D. L. Illman and B. R. Kowalski, Chemometrics,
Wiley, New York, 1986.
13 British Pharmacopoeia 1993, HM Stationery Office, London, 1993,
vol. II, p. 1043.
Paper 8/04528C