- Email: [email protected]
Influence of particle size on the quantitative determination of salicylic acid in a pharmaceutical ointment using FT-Raman spectroscopy
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/ejps
Influence of particle size on the quantitative determination of salicylic acid in a pharmaceutical ointment using FT-Raman spectroscopy T.R.M. De Beer a , W.R.G. Baeyens a,∗ , Y. Vander Heyden b , J.P. Remon c , C. Vervaet c , F. Verpoort d a
Laboratory of Drug Quality Control, Department of Pharmaceutical Analysis, Ghent University, Harelbekestraat 72, B-9000 Ghent, Belgium b Department of Analytical Chemistry and Pharmaceutical Technology, Pharmaceutical Institute, Vrije Universiteit Brussel (VUB), Laarbeeklaan 103, B-1090 Brussels, Belgium c Laboratory of Pharmaceutical Technology, Ghent University, Harelbekestraat 72, B-9000 Ghent, Belgium d Department of Inorganic and Physical Chemistry, Ghent University, Krijgslaan 281 (S3), B-9000 Ghent, Belgium
a r t i c l e
i n f o
a b s t r a c t
Article history:
A second order polynomial calibration model was developed and statistically validated for
Received 1 August 2006
the direct and non-destructive quantitative analysis – without sample preparation – of the
Received in revised form
active pharmaceutical ingredient (API) salicylic acid in a pharmaceutical ointment using FT-
8 November 2006
Raman spectroscopy. The calibration curve was modeled by plotting the peak intensity of
Accepted 9 November 2006
the vector normalized spectral band between 757 and 784 cm−1 against the known salicylic
Published on line 18 November 2006
acid concentrations in standards. At this band, no spectral interferences from the ointment vehiculum (white vaseline) are observed. For the validation of the polynomial model, its fit
Keywords:
and its predictive properties were evaluated. The validated model was used for the quantifi-
FT-Raman spectroscopy
cation of 25 ointments, compounded by different retail pharmacists. The same standards
HPLC
and samples were used, both for development and validation of a regression model and
Particle size
for quantitative determination by HPLC – with sample preparation – as described for the
Polynomial regression
related substances of salicylic acid in the Ph. Eur. IV. The quantification results obtained
Validation
by the FT-Raman method corresponded with the HPLC results (p = 0.22), provided that the
Ointment
particle size of salicylic acid in the standards is the same as in the analyzed samples. The non-destructive FT-Raman method is a reliable alternative for the destructive HPLC method, as it is faster and does not require sample pre-treatment procedures. © 2006 Elsevier B.V. All rights reserved.
1.
Introduction
Since the last decade, high technological improvements in Raman spectrometers (McCreery, 2000; Long, 2002) coupled with the important advantages offered by Raman spectroscopy, have made this spectroscopic technique more
∗
Corresponding author. Tel.: +32 92648097; fax: +32 92648196. E-mail address: [email protected] (W.R.G. Baeyens). 0928-0987/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ejps.2006.11.009
attractive for the qualitative and quantitative analysis of pharmaceuticals (Vankeirsbilck et al., 2002). Substantial advantages result from the fact that Raman spectroscopy is a fast and non-destructive technique and that spectra can be recorded directly inside the packaging (transparent glass, plastics and blisters) without worth mentioning interferences.
230
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
These advantages make it possible to measure APIs as well as additives as they appear in reality in drug formulations, without any prior manipulation by sample treatment procedures. As water is a very weak Raman scatterer and hence yields no significant signal in Raman spectra, Raman spectroscopy can be an appropriate tool for analysis in aqueous environments (De Beer et al., 2004). This is in contrast to infrared (IR) spectroscopy where water provides a huge absorption signal, often overwhelming the signals of interest. As Raman spectra are characterized by sharp bands, characteristic bands recorded for the API in a sample often do not overlap with the bands produced by the excipients. This makes the development of univariate calibration models possible. Near infrared (NIR) signals are mostly very wide bands often overlapping, hence making the application of more complex multivariate models necessary. Until now, mostly Raman-based quantitative analyses of pharmaceutical solid dosage forms, either or not through packaging, have been described (Compton and Compton, 1991; Tsuchihashi et al., 1997; Niemczyk et al., 1998; Skoulika and Georgiou, 2001; Yang and Irudayaraj, 2002; Vergote et al., 2002; Skoulika and Georgiou, 2003). The non-destructivity and rapidity of analyses performed on solid materials make Raman spectroscopy an excellent and promising Process Analytical Technology (PAT) tool for pharmaceutical production processes (blending, reaction monitoring, controlling the solid states of pharmaceutical crystals during production processes, etc.) (Clegg and Everall, 2003; Vankeirsbilck et al., 2002; Christopher, 1999; Vergote et al., 2004). Quantitative determination methods applying Raman spectroscopy or NIR to other pharmaceutical forms (creams, ointments, solutions, emulsions, etc.) are less frequent in literature, so far. When studying solids, it is important to bear in mind that particle size influences significantly the Raman signal (intensity). Overlooking the effect of the particle size during quantitative analysis may result in erroneous conclusions (Pellow-Jarman et al., 1996; Campbell Roberts et al., 2002). In the present work, a rapid method by FT-Raman spectroscopy for the quantitative determination of salicylic acid in ointments was developed, validated and applied to analyze ointments compounded by retail pharmacists. Herewith, the influence of the particle size differences, between the salicylic acid used to develop the calibration model and that used by the retail pharmacists, was evaluated. The non-destructive FT-Raman method was compared to the conventional HPLC method described for the related substances of salicylic acid in the Ph. Eur. IV. Salicylic acid ointments are medically used for the treatment of warts and hyperkeratosis of skin and nails. The ointments are formulated by suspending salicylic acid in white vaseline (20:30:40%, w/w) as described in the Belgian Therapeutic Magistral Formularium (TMF).
2.
Materials and methods
2.1.
Materials
The salicylic acid (particle size <180 m) and white vaseline used for the preparation of the standard oint-
ments were purchased from Alpha Pharma (Nazareth, Belgium). The compounded ointments were obtained from 25 different retail pharmacies. The origin and particle size of the salicylic acid used by the pharmacists is unknown.
2.2.
Standard and analyzed ointments
The salicylic acid content of 25 compounded ointments was first analyzed by FT-Raman spectroscopy, followed by HPLC analysis. Both methods were based on calibration models developed by measuring a series of seven standard ointments (standards) prepared ranging from 15 to 45% (w/w) of salicylic acid with increments of 5% (w/w). The same calibration standards were used to establish the HPLC calibration model. The standard ointments were prepared according to the procedure described in the Belgian TMF: exact amounts of white vaseline and salicylic acid in a mortar, followed by profound mixing using a pestle to homogeneously suspend salicylic acid in the ointment basis. The measuring conditions for the standards and the unknown ointments during FT-Raman spectrometry were always identical. As a consequence, the ointments were transferred to the same type of glass vials prior to Raman analysis.
2.3.
Spectroscopic conditions
A Bruker FT spectrometer Equinox 55S (Bruker Optik, Ettlingen, Germany), equipped with the Raman module FRA 106 fitted to a cooled (77 K) germanium high sensitivity detector D418-T, was used. The laser wavelength during the experiments was the 1.064 m line from a diode laser pumped Nd:YAG laser. All spectra were recorded at a resolution of 4 cm−1 and a laser power of 300 mW was used. Data collection and data transfer were automated using the Bruker OPUSTM software. A motorized positioner focused the laser beam on the sample to obtain maximum intensity of the Raman signal. During the measurements, each ointment vial (standards and unknowns) was fixed in a holder to ensure that all ointments were measured in an identical position. The recorded spectra were the results of 100 scan measurements, corresponding to a measurement time of about 2.5 min. No internal standard was necessary and intersession calibration was not done. The Raman signal was vector normalized to compensate for any change in experimental conditions (variation of the laser intensity, sample positioning and temperature) (Vergote et al., 2002).
2.4.
HPLC method
Parallel to the FT-Raman calibration model development, the same standards were used to establish the HPLC calibration model. The HPLC method was based on the assay method for related substances of salicylic acid as described in the Ph. Eur. IV. The HPLC system (Merck Hitachi, Tokyo, Japan) consisted of a pump (type L-7100), a UV-detector (type L-7400), an integrator (type D-7000) and a reversed-phase column (LiChrospher® 100 RP C18 (5 m), 4 mm × 25 cm). Approximately 1 g of ointment, accurately weighed, was transferred in a flask, followed by the addition of 100.0 ml chloroform. The flask was closed
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
with a glass stopper to avoid solvent evaporation. Next, the flask was heated to 60 ◦ C, resulting in the melting of the white vaseline and the dissolution of the salicylic acid in chloroform. After cooling the solution to room temperature, an aliquot of 20 l was injected.
2.5. Development and validation of the FT-Raman and HPLC calibration models The procedure for the development and validation of the linear regression calibration models was identical for the FTRaman and the HPLC method. The measured signals (peak intensity of the vector normalized spectral band between 757 and 784 cm−1 for the FT-Raman measurements and peak area for the HPLC measurements) were plotted against the different salicylic acid concentrations in the standard ointments. Each standard was measured 10 times for the FT-Raman model and three times for the HPLC model. The regression parameters of the selected linear calibration models through the plotted data were estimated. Applying linear regression involves three assumptions, to be checked: (i) all errors occur in the instrumental signal measurements (i.e. the errors from preparation of the standards are negligible compared to the measurement errors); (ii) the signal measurement errors or residuals (i.e. differences between the measured signal and the signal predicted by the calibration model) are normally distributed with mean zero at every x-value (concentration), which can be tested by the Kolmogorov–Smirnov test (Massart et al., 1997); (iii) variation in the residuals is the same at all standard concentrations (homoscedasticity), which can be tested by a Cochran test (Massart et al., 1997; Miller, 1991; Cuadros Rodriguez et al., 1996). Next, the fit of the chosen model was evaluated from (i) the coefficient of multiple determination (R2 ); (ii) a lack-of-fit test based on an analysis of variance (ANOVA); (iii) the root mean squared error (RMSE) of the calibration points; and (iv) analysis of the residuals. These tests are described by Vander Heyden et al. (2002). Depending on the calibration range, polynomial regression might be required to model the Raman calibration data. A higher order polynomial model will fit the data better, but makes the model more complex (Massart et al., 1997). On the other hand, as calibration models are intended for prediction, the predictive properties of the selected model have to be tested. This can be done by the root mean squared prediction error (RMSEP), when using an independent test set, or the root mean squared cross validation error (RMSECV)
fact that they fit the data better. This is known as “overfitting” of the model and is due to the fact that the experimental error also is modelled. Hence, the ideal compromise between fit and predictive properties has to be found. Finally, the confidence and prediction intervals for the estimated parameters of the selected regression model as well as for the true regression line were calculated (Massart et al., 1997) using Statgraphics® and Excel® . The FT-Raman and HPLC calibration models were validated by LOO cross validation.
2.6. Drug content determination of the unknown ointments From the validated calibration models (FT-Raman and HPLC), the salicylic acid content of 25 compounded ointments was estimated.
2.7. Determination of the particle size of the unknown ointments As the particle size influences significantly the Raman signal (intensity), overlooking the effect of the particle size during quantitative analysis may result in erroneous conclusions (Pellow-Jarman et al., 1996; Campbell Roberts et al., 2002). Therefore, the particle size of the unknown ointments was examined using microscopy (Olympus microscope SZX9) to find out if the salicylic acid used by the pharmacists had a similar particle size as the salicylic acid used for the preparation of the standards. A microscopic picture (85×) from each ointment was taken. Per ointment, the longest side of 20 randomly chosen crystals was measured. Their average and standard deviation were calculated.
3.
Results and discussion
3.1.
Development of the FT-Raman calibration model
The FT-Raman spectra of salicylic acid and white vaseline are shown in Fig. 1. As vaseline produces no signal at the spectral range of interest (100–1500 cm−1 ), the signal from the ointment in this region originates from salicylic acid. The highest
n ei 2 RMSECV =
i=1
n
where ei (= yi − yˆ −i ; where yi corresponds to the measurement of the deleted point and yˆ −i to its prediction value from the model built without (xi , yi )) represents the residuals calculated after leave-one-out (LOO) cross validation of the calibration data and n the number of LOO replicates (Vander Heyden et al., 2002). The smaller the RMSECV, the better the predictive properties are. Higher order polynomials often predict worse (higher RMSECV) than lower order polynomials, despite the
231
Fig. 1 – FT-Raman spectra of salicylic acid and white vaseline.
232
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
Table 1 – Statistical evaluation of the residuals for the FT-Raman second order polynomial calibration model Concentration (%, w/w)
Kolmogorov–Smirnov test (˛ = 0.05)
t-test (˛ = 0.05)
Calculated Critical value value (n = 10)
Fig. 2 – Plot of measured FT-Raman signals (10 per standard) against the different ointment concentrations.
peak intensity of the vector normalized spectral band between 757 and 784 cm−1 was the signal selected for the development of the calibration model and for the quantitative analysis of the unknown ointments. Although the other Raman bands in the spectral region 100–1500 cm−1 could also have been used, the 757–784 cm−1 band was chosen. This band has the highest intensity in the ointment spectrum, which means that small spectral changes will be best seen in this band. After plotting the measured signals (10 per standard) against the different ointment concentrations, apparently a simple linear regression model (y = ax + b) does not fit the data (Fig. 2). Therefore, a second order polynomial regression model was built and tested (y = 7.648 × 10−5 x2 + 1.241 × 10−3 x + 2.917 × 10−2 ). At first, the assumptions concerning the application of linear regression were checked. (i) In cases where sample preparation is limited – as in this case –, the errors during
14.97 20.08 25.54 30.29 35.44 40.55 45.13
0.205 0.204 0.142 0.152 0.109 0.166 0.225
0.262 0.262 0.262 0.262 0.262 0.262 0.262
0.13 0.62 0.14 0.24 0.07 0.10 0.21
preparation of the standards can be considered negligible compared to the errors during measuring the instrument signals (Miller, 1991). (ii) The 10 residuals at every standard concentration are normally distributed as demonstrated by the Kolmogorov–Smirnov tests (Table 1), where the calculated values for this test are smaller than the critical tabulated value (0.262) at n = 10 and ˛ = 0.05. In addition, a t-test (Table 1) shows that there is no significant difference (p > 0.05) at every concentration level between the means of the residuals and zero for ˛ = 0.05. Based on these tests, it is concluded that the residuals are normally distributed around zero at every concentration level. (iii) Homoscedasticity of the normal distributed residuals is demonstrated by a Cochran-test (p = 0.80). As the three assumptions concerning the residuals are met, it is concluded that the second order polynomial calibration model might be applied. In a next step, it was verified if the chosen model adequately describes the relationship between the standard concentration and the peak intensity, within
Fig. 3 – Evaluation of the model fit. (a) R2 vs. polynomial order plot; (b) RMSE vs. polynomial order plot; (c) residuals plot of the second order polynomial model; (d) residuals plot of the straight line model.
233
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
Fig. 4 – Evaluation of the predictive properties: RMSECV vs. polynomial order plot.
the examined concentration range (=evaluation of the fit by R2 (Fig. 3a), RMSE (Fig. 3b) and the analysis of the residuals (Fig. 3c and d)). In Fig. 3a and b it is observed that the model fit is improved considerably from a straight line to the second order model, while for higher order models only a slight further improvement is seen. In contrast to the linear model (Fig. 3d), the residuals plot of the second order polynomial model (Fig. 3c) does not show any trend. Thus, a random distribution of positive and negative residuals is obtained, which means that the model appropriately describes the data. The RMSECV (from LOO validation) versus polynomial order plot (Fig. 4) shows that the predictive properties of the third, fourth and fifth order models are not significantly better than for the second order model. In general, it has been proposed to select the first local minimum in the RMSECV versus order plot. As the lowest possible order model should be used, the second order polynomial regression model was selected. The 95% confidence interval from the observed data (Fig. 5) shows the limits of the true response (peak intensity) at a particular x-value (concentration), which may be expected with 95% probability. The 95% prediction interval for measured peak intensities at known concentrations (Fig. 5) shows with 95% assurance within which limits the peak intensity will be for one measurement of a given suspension concentration.
3.2.
Fig. 5 – 95% prediction interval and 95% confidence interval of the estimated second order polynomial FT-Raman calibration model.
Fig. 6 – Simple linear HPLC calibration model.
calibration model was obtained: y = 8.59368 × 106 x + 25376.1 (Fig. 6).
3.3.
The recovery, repeatability, linearity and range of the Raman and HPLC analyses were validated by LOO cross-validation. An overview is given in Table 2. It is clear that the bias of both methods is similar. The repeatability of the HPLC method appears better than for the Raman method.
3.4.
Development of the HPLC calibration model
The development of the HPLC calibration model was performed in the same way as the Raman model. A simple linear
Validation of the Raman and HPLC methods
Drug analysis of the unknown ointments
The drug content of 25 ointments was determined using the same procedure as for the cross-validated ointments. The sal-
Table 2 – LOO cross-validation of the Raman and the HPLC method Concentration (%, w/w)
14.97 20.08 25.54 30.29 35.44 40.55 45.13
Raman Recovery (%) (n = 10) 96.68 101.04 101.16 101.35 98.48 99.07 100.79
Repeatability (%R.S.D.) (n = 10) 4.54 7.35 2.62 4.37 3.13 2.28 2.58
HPLC Linearity (r2 )
Recovery (%) (n = 3)
0.9907
98.52 99.89 100.93 100.57 100.04 99.86 99.70
Repeatability (%R.S.D.) (n = 3) 0.28 0.11 0.34 0.10 0.23 0.19 0.34
Linearity (r2 )
0.9997
234
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
Table 3 – FT-Raman and HPLC analyses of 25 ointments Sample code
Raman Salicylic acid found (%, w/w)
O1 O2 O3 O4 O5 O6 O7 O8 O9 O10 O11 O12 O13 O14 O15 O16 O17 O18 O19 O20 O21 O22 O23 O24 O25
20.85 19.55 21.56 20.36 20.80 20.22 19.68 19.32 20.48 24.21 22.84 21.37 15.91 24.36 21.70 19.56 22.77 23.47 20.94 20.29 22.96 22.06 21.82 16.83 20.46
HPLC Deviation from 20% (%) 4.25 2.25 7.80 1.80 4.00 1.10 1.60 3.40 2.40 21.05 14.20 6.85 20.45 21.80 8.50 2.20 13.85 17.35 4.70 1.45 14.80 10.30 9.10 15.85 2.30
icylic acid dose per ointment calculated from the FT-Raman data was significantly different from the results obtained by HPLC-analysis (Table 3) as proven by a paired t-test (p = 0.01). A possible reason for this is that some unknown ointments were prepared with salicylic acid having a different particle size than the salicylic acid (<180 m) used for the standards. Therefore, microscopic measurements were performed on the unknown ointments to determine the particle size (cf. Section 2.7). The underlined ointments (Table 3) were found to have a particle size smaller than 90 m. Performing a paired t-test with these ointments excluded (O3, O14, O17, O18, O21), indicates that the salicylic acid dose per ointment calculated from the FT-Raman data is not significant different from the results obtained by HPLC analysis (p = 0.22). The Raman results for the ointments with small particles tend to be overestimated. As the target concentration of the unknown ointments was 20% (w/w) and as the maximum allowed error is 10% (according to the TMF), three ointments (O11, O13, O24) appeared not to satisfy this requirement, which was confirmed by both the Raman and HPLC analyses. Thus, both techniques are able to identify the ointments that deviate more than allowed from the target concentration.
4.
Conclusion
The results showed that the fast and non-destructive FTRaman spectroscopic method is reliable for the quantitative determination of salicylic acid in ointments, having a similar bias compared to the HPLC method, which requires a sample preparation. The repeatability of the Raman method is less good, but acceptable. An important requirement for the
Salicylic acid found (%, w/w) 20.44 19.34 20.47 20.35 19.61 19.79 20.10 20.82 20.12 19.98 22.48 20.04 17.35 20.06 20.45 20.09 20.59 20.01 20.42 20.58 20.23 21.00 20.72 17.09 21.38
Deviation from 20% (%) 2.20 3.30 2.35 1.75 1.95 1.05 0.50 4.10 0.60 0.10 12.40 0.20 13.25 0.30 2.25 0.45 2.95 0.05 2.10 2.90 1.15 5.00 3.60 14.55 6.90
Raman method is that the particle size of salicylic acid present in the samples is similar to that used for the preparation of the standards.
references
Campbell Roberts, S.N., Williams, A.C., Grimsey, I.M., Booth, S.W., 2002. Quantitative analysis of mannitol polymorphs. FT-Raman spectroscopy. J. Pharm. Biomed. Anal. 28, 1135–1147. Christopher, F.J., 1999. Review of pharmaceutical applications of Raman spectroscopy. In: Pelletier, M.J. (Ed.), Analytical Applications of Raman Spectroscopy. London, UK, pp. 224–275. Clegg, I., Everall, N., 2003. On-line measurement of crystalline forms. Eur. Pharm. Rev. (3), 56–62. Compton, D.A.C., Compton, S.V., 1991. Examination of packaged consumer-goods by using FT-Raman spectrometry. Appl. Spectrosc. 45 (10), 1587–1589. ˜ A.M., Bosque Sendra, Cuadros Rodriguez, L., Garc´ıa Campana, J.M., 1996. Statistical estimation of linear calibration range. Anal. Lett. 29, 1231–1239. De Beer, T.R.M., Vergote, G.J., Baeyens, W.R.G., Remon, J.P., Vervaet, C., Verpoort, F., 2004. Development and validation of a direct non-destructive quantitative method for medroxyprogesterone acetate in a pharmaceutical suspension using FT-Raman spectroscopy. Eur. J. Pharm. Sci. 23, 355–362. Long, D.A., 2002. The Raman Effect: A Unified Treatment of the Theory of Raman Scattering by Molecules. Wiley–Interscience, Chichester, West Sussex, England, 597 pp. Massart, D.L., Vandeginste, B.G.M., Buydens, L.M.C., De Jong, S., Lewi, P.J., Smeyers-Verbeke, J., 1997. Handbook of Chemometrics and Qualimetrics Part A. Elsevier, Amsterdam, The Netherlands.
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
McCreery, R.L., 2000. Raman Spectroscopy for Chemical Analysis. Wiley–Interscience, New York, USA. Miller, J.N., 1991. Basis statistical methods for analytical chemistry. Part 2. Calibration and regression methods: a review. Analyst 116, 3–14. Niemczyk, T.M., Delgado-Lopez, M.M., Allen, F.S., 1998. Quantitative determination of bucindolol concentration in intact gel capsules using Raman spectroscopy. Anal. Chem. 70 (13), 2762–2765. Pellow-Jarman, M.V., Hendra, P.J., Lehnert, R.J., 1996. The dependence of the Raman signal intensity on particle size for crystal powders. Vibrat. Spectrosc. 12, 257–261. Skoulika, S.G., Georgiou, C.A., 2001. Rapid quantitative determination of ciprofloxacin in pharmaceuticals by use of solid-state FT-Raman spectroscopy. Appl. Spectrosc. 55 (9), 1259–1265. Skoulika, S.G., Georgiou, C.A., 2003. Rapid noninvasive quantitative determination of acyclovir in pharmaceutical solid dosage forms through their poly(vinyl chloride) blister by solid-state fourier transform Raman spectroscopy. Appl. Spectrosc. 57 (4), 407–412. Tsuchihashi, H., Katagi, M., Nishikawa, M., Tatsuno, M., Nishioka, H., Nara, A., Nishio, E., Petty, C., 1997. Determination of
235
methamphetamine and its related compounds using fourier transform Raman spectroscopy. Appl. Spectrosc. 51 (12), 1796–1799. Vander Heyden, Y., Popovici, S.T., Schoenmakers, P.J., 2002. Evaluation of size-exclusion chromatography and size-exclusion electrochromatography calibration curves. J. Chromatogr. A 957, 127–137. Vankeirsbilck, T., Vercauteren, A., Baeyens, W., Van der Weken, G., Verpoort, F., Vergote, G., Remon, J.P., 2002. Applications of Raman spectroscopy in pharmaceutical analysis. Trends Anal. Chem. 21 (12), 869–877. Vergote, G.J., Vervaet, C., Remon, J.P., Haemers, T., Verpoort, F., 2002. Near-infrared FT-Raman spectroscopy as a rapid analytical tool for the determination of diltiazem hydrochloride in tablets. Eur. J. Pharm. Sci. 16, 63–67. Vergote, G.J., De Beer, T.R.M., Vervaet, C., Remon, J.P., Baeyens, W.R.G., Diericx, N., Verpoort, F., 2004. In-line monitoring of a pharmaceutical blending process using FT-Raman spectroscopy. Eur. J. Pharm. Sci. 21, 479–485. Yang, H., Irudayaraj, J., 2002. Rapid determination of Vitamin C by NIR, MIR and FT-Raman techniques. J. Pharm. Pharmacol. 54 (9), 1247–1255.
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/ejps
Influence of particle size on the quantitative determination of salicylic acid in a pharmaceutical ointment using FT-Raman spectroscopy T.R.M. De Beer a , W.R.G. Baeyens a,∗ , Y. Vander Heyden b , J.P. Remon c , C. Vervaet c , F. Verpoort d a
Laboratory of Drug Quality Control, Department of Pharmaceutical Analysis, Ghent University, Harelbekestraat 72, B-9000 Ghent, Belgium b Department of Analytical Chemistry and Pharmaceutical Technology, Pharmaceutical Institute, Vrije Universiteit Brussel (VUB), Laarbeeklaan 103, B-1090 Brussels, Belgium c Laboratory of Pharmaceutical Technology, Ghent University, Harelbekestraat 72, B-9000 Ghent, Belgium d Department of Inorganic and Physical Chemistry, Ghent University, Krijgslaan 281 (S3), B-9000 Ghent, Belgium
a r t i c l e
i n f o
a b s t r a c t
Article history:
A second order polynomial calibration model was developed and statistically validated for
Received 1 August 2006
the direct and non-destructive quantitative analysis – without sample preparation – of the
Received in revised form
active pharmaceutical ingredient (API) salicylic acid in a pharmaceutical ointment using FT-
8 November 2006
Raman spectroscopy. The calibration curve was modeled by plotting the peak intensity of
Accepted 9 November 2006
the vector normalized spectral band between 757 and 784 cm−1 against the known salicylic
Published on line 18 November 2006
acid concentrations in standards. At this band, no spectral interferences from the ointment vehiculum (white vaseline) are observed. For the validation of the polynomial model, its fit
Keywords:
and its predictive properties were evaluated. The validated model was used for the quantifi-
FT-Raman spectroscopy
cation of 25 ointments, compounded by different retail pharmacists. The same standards
HPLC
and samples were used, both for development and validation of a regression model and
Particle size
for quantitative determination by HPLC – with sample preparation – as described for the
Polynomial regression
related substances of salicylic acid in the Ph. Eur. IV. The quantification results obtained
Validation
by the FT-Raman method corresponded with the HPLC results (p = 0.22), provided that the
Ointment
particle size of salicylic acid in the standards is the same as in the analyzed samples. The non-destructive FT-Raman method is a reliable alternative for the destructive HPLC method, as it is faster and does not require sample pre-treatment procedures. © 2006 Elsevier B.V. All rights reserved.
1.
Introduction
Since the last decade, high technological improvements in Raman spectrometers (McCreery, 2000; Long, 2002) coupled with the important advantages offered by Raman spectroscopy, have made this spectroscopic technique more
∗
Corresponding author. Tel.: +32 92648097; fax: +32 92648196. E-mail address: [email protected] (W.R.G. Baeyens). 0928-0987/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ejps.2006.11.009
attractive for the qualitative and quantitative analysis of pharmaceuticals (Vankeirsbilck et al., 2002). Substantial advantages result from the fact that Raman spectroscopy is a fast and non-destructive technique and that spectra can be recorded directly inside the packaging (transparent glass, plastics and blisters) without worth mentioning interferences.
230
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
These advantages make it possible to measure APIs as well as additives as they appear in reality in drug formulations, without any prior manipulation by sample treatment procedures. As water is a very weak Raman scatterer and hence yields no significant signal in Raman spectra, Raman spectroscopy can be an appropriate tool for analysis in aqueous environments (De Beer et al., 2004). This is in contrast to infrared (IR) spectroscopy where water provides a huge absorption signal, often overwhelming the signals of interest. As Raman spectra are characterized by sharp bands, characteristic bands recorded for the API in a sample often do not overlap with the bands produced by the excipients. This makes the development of univariate calibration models possible. Near infrared (NIR) signals are mostly very wide bands often overlapping, hence making the application of more complex multivariate models necessary. Until now, mostly Raman-based quantitative analyses of pharmaceutical solid dosage forms, either or not through packaging, have been described (Compton and Compton, 1991; Tsuchihashi et al., 1997; Niemczyk et al., 1998; Skoulika and Georgiou, 2001; Yang and Irudayaraj, 2002; Vergote et al., 2002; Skoulika and Georgiou, 2003). The non-destructivity and rapidity of analyses performed on solid materials make Raman spectroscopy an excellent and promising Process Analytical Technology (PAT) tool for pharmaceutical production processes (blending, reaction monitoring, controlling the solid states of pharmaceutical crystals during production processes, etc.) (Clegg and Everall, 2003; Vankeirsbilck et al., 2002; Christopher, 1999; Vergote et al., 2004). Quantitative determination methods applying Raman spectroscopy or NIR to other pharmaceutical forms (creams, ointments, solutions, emulsions, etc.) are less frequent in literature, so far. When studying solids, it is important to bear in mind that particle size influences significantly the Raman signal (intensity). Overlooking the effect of the particle size during quantitative analysis may result in erroneous conclusions (Pellow-Jarman et al., 1996; Campbell Roberts et al., 2002). In the present work, a rapid method by FT-Raman spectroscopy for the quantitative determination of salicylic acid in ointments was developed, validated and applied to analyze ointments compounded by retail pharmacists. Herewith, the influence of the particle size differences, between the salicylic acid used to develop the calibration model and that used by the retail pharmacists, was evaluated. The non-destructive FT-Raman method was compared to the conventional HPLC method described for the related substances of salicylic acid in the Ph. Eur. IV. Salicylic acid ointments are medically used for the treatment of warts and hyperkeratosis of skin and nails. The ointments are formulated by suspending salicylic acid in white vaseline (20:30:40%, w/w) as described in the Belgian Therapeutic Magistral Formularium (TMF).
2.
Materials and methods
2.1.
Materials
The salicylic acid (particle size <180 m) and white vaseline used for the preparation of the standard oint-
ments were purchased from Alpha Pharma (Nazareth, Belgium). The compounded ointments were obtained from 25 different retail pharmacies. The origin and particle size of the salicylic acid used by the pharmacists is unknown.
2.2.
Standard and analyzed ointments
The salicylic acid content of 25 compounded ointments was first analyzed by FT-Raman spectroscopy, followed by HPLC analysis. Both methods were based on calibration models developed by measuring a series of seven standard ointments (standards) prepared ranging from 15 to 45% (w/w) of salicylic acid with increments of 5% (w/w). The same calibration standards were used to establish the HPLC calibration model. The standard ointments were prepared according to the procedure described in the Belgian TMF: exact amounts of white vaseline and salicylic acid in a mortar, followed by profound mixing using a pestle to homogeneously suspend salicylic acid in the ointment basis. The measuring conditions for the standards and the unknown ointments during FT-Raman spectrometry were always identical. As a consequence, the ointments were transferred to the same type of glass vials prior to Raman analysis.
2.3.
Spectroscopic conditions
A Bruker FT spectrometer Equinox 55S (Bruker Optik, Ettlingen, Germany), equipped with the Raman module FRA 106 fitted to a cooled (77 K) germanium high sensitivity detector D418-T, was used. The laser wavelength during the experiments was the 1.064 m line from a diode laser pumped Nd:YAG laser. All spectra were recorded at a resolution of 4 cm−1 and a laser power of 300 mW was used. Data collection and data transfer were automated using the Bruker OPUSTM software. A motorized positioner focused the laser beam on the sample to obtain maximum intensity of the Raman signal. During the measurements, each ointment vial (standards and unknowns) was fixed in a holder to ensure that all ointments were measured in an identical position. The recorded spectra were the results of 100 scan measurements, corresponding to a measurement time of about 2.5 min. No internal standard was necessary and intersession calibration was not done. The Raman signal was vector normalized to compensate for any change in experimental conditions (variation of the laser intensity, sample positioning and temperature) (Vergote et al., 2002).
2.4.
HPLC method
Parallel to the FT-Raman calibration model development, the same standards were used to establish the HPLC calibration model. The HPLC method was based on the assay method for related substances of salicylic acid as described in the Ph. Eur. IV. The HPLC system (Merck Hitachi, Tokyo, Japan) consisted of a pump (type L-7100), a UV-detector (type L-7400), an integrator (type D-7000) and a reversed-phase column (LiChrospher® 100 RP C18 (5 m), 4 mm × 25 cm). Approximately 1 g of ointment, accurately weighed, was transferred in a flask, followed by the addition of 100.0 ml chloroform. The flask was closed
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
with a glass stopper to avoid solvent evaporation. Next, the flask was heated to 60 ◦ C, resulting in the melting of the white vaseline and the dissolution of the salicylic acid in chloroform. After cooling the solution to room temperature, an aliquot of 20 l was injected.
2.5. Development and validation of the FT-Raman and HPLC calibration models The procedure for the development and validation of the linear regression calibration models was identical for the FTRaman and the HPLC method. The measured signals (peak intensity of the vector normalized spectral band between 757 and 784 cm−1 for the FT-Raman measurements and peak area for the HPLC measurements) were plotted against the different salicylic acid concentrations in the standard ointments. Each standard was measured 10 times for the FT-Raman model and three times for the HPLC model. The regression parameters of the selected linear calibration models through the plotted data were estimated. Applying linear regression involves three assumptions, to be checked: (i) all errors occur in the instrumental signal measurements (i.e. the errors from preparation of the standards are negligible compared to the measurement errors); (ii) the signal measurement errors or residuals (i.e. differences between the measured signal and the signal predicted by the calibration model) are normally distributed with mean zero at every x-value (concentration), which can be tested by the Kolmogorov–Smirnov test (Massart et al., 1997); (iii) variation in the residuals is the same at all standard concentrations (homoscedasticity), which can be tested by a Cochran test (Massart et al., 1997; Miller, 1991; Cuadros Rodriguez et al., 1996). Next, the fit of the chosen model was evaluated from (i) the coefficient of multiple determination (R2 ); (ii) a lack-of-fit test based on an analysis of variance (ANOVA); (iii) the root mean squared error (RMSE) of the calibration points; and (iv) analysis of the residuals. These tests are described by Vander Heyden et al. (2002). Depending on the calibration range, polynomial regression might be required to model the Raman calibration data. A higher order polynomial model will fit the data better, but makes the model more complex (Massart et al., 1997). On the other hand, as calibration models are intended for prediction, the predictive properties of the selected model have to be tested. This can be done by the root mean squared prediction error (RMSEP), when using an independent test set, or the root mean squared cross validation error (RMSECV)
fact that they fit the data better. This is known as “overfitting” of the model and is due to the fact that the experimental error also is modelled. Hence, the ideal compromise between fit and predictive properties has to be found. Finally, the confidence and prediction intervals for the estimated parameters of the selected regression model as well as for the true regression line were calculated (Massart et al., 1997) using Statgraphics® and Excel® . The FT-Raman and HPLC calibration models were validated by LOO cross validation.
2.6. Drug content determination of the unknown ointments From the validated calibration models (FT-Raman and HPLC), the salicylic acid content of 25 compounded ointments was estimated.
2.7. Determination of the particle size of the unknown ointments As the particle size influences significantly the Raman signal (intensity), overlooking the effect of the particle size during quantitative analysis may result in erroneous conclusions (Pellow-Jarman et al., 1996; Campbell Roberts et al., 2002). Therefore, the particle size of the unknown ointments was examined using microscopy (Olympus microscope SZX9) to find out if the salicylic acid used by the pharmacists had a similar particle size as the salicylic acid used for the preparation of the standards. A microscopic picture (85×) from each ointment was taken. Per ointment, the longest side of 20 randomly chosen crystals was measured. Their average and standard deviation were calculated.
3.
Results and discussion
3.1.
Development of the FT-Raman calibration model
The FT-Raman spectra of salicylic acid and white vaseline are shown in Fig. 1. As vaseline produces no signal at the spectral range of interest (100–1500 cm−1 ), the signal from the ointment in this region originates from salicylic acid. The highest
n ei 2 RMSECV =
i=1
n
where ei (= yi − yˆ −i ; where yi corresponds to the measurement of the deleted point and yˆ −i to its prediction value from the model built without (xi , yi )) represents the residuals calculated after leave-one-out (LOO) cross validation of the calibration data and n the number of LOO replicates (Vander Heyden et al., 2002). The smaller the RMSECV, the better the predictive properties are. Higher order polynomials often predict worse (higher RMSECV) than lower order polynomials, despite the
231
Fig. 1 – FT-Raman spectra of salicylic acid and white vaseline.
232
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
Table 1 – Statistical evaluation of the residuals for the FT-Raman second order polynomial calibration model Concentration (%, w/w)
Kolmogorov–Smirnov test (˛ = 0.05)
t-test (˛ = 0.05)
Calculated Critical value value (n = 10)
Fig. 2 – Plot of measured FT-Raman signals (10 per standard) against the different ointment concentrations.
peak intensity of the vector normalized spectral band between 757 and 784 cm−1 was the signal selected for the development of the calibration model and for the quantitative analysis of the unknown ointments. Although the other Raman bands in the spectral region 100–1500 cm−1 could also have been used, the 757–784 cm−1 band was chosen. This band has the highest intensity in the ointment spectrum, which means that small spectral changes will be best seen in this band. After plotting the measured signals (10 per standard) against the different ointment concentrations, apparently a simple linear regression model (y = ax + b) does not fit the data (Fig. 2). Therefore, a second order polynomial regression model was built and tested (y = 7.648 × 10−5 x2 + 1.241 × 10−3 x + 2.917 × 10−2 ). At first, the assumptions concerning the application of linear regression were checked. (i) In cases where sample preparation is limited – as in this case –, the errors during
14.97 20.08 25.54 30.29 35.44 40.55 45.13
0.205 0.204 0.142 0.152 0.109 0.166 0.225
0.262 0.262 0.262 0.262 0.262 0.262 0.262
0.13 0.62 0.14 0.24 0.07 0.10 0.21
preparation of the standards can be considered negligible compared to the errors during measuring the instrument signals (Miller, 1991). (ii) The 10 residuals at every standard concentration are normally distributed as demonstrated by the Kolmogorov–Smirnov tests (Table 1), where the calculated values for this test are smaller than the critical tabulated value (0.262) at n = 10 and ˛ = 0.05. In addition, a t-test (Table 1) shows that there is no significant difference (p > 0.05) at every concentration level between the means of the residuals and zero for ˛ = 0.05. Based on these tests, it is concluded that the residuals are normally distributed around zero at every concentration level. (iii) Homoscedasticity of the normal distributed residuals is demonstrated by a Cochran-test (p = 0.80). As the three assumptions concerning the residuals are met, it is concluded that the second order polynomial calibration model might be applied. In a next step, it was verified if the chosen model adequately describes the relationship between the standard concentration and the peak intensity, within
Fig. 3 – Evaluation of the model fit. (a) R2 vs. polynomial order plot; (b) RMSE vs. polynomial order plot; (c) residuals plot of the second order polynomial model; (d) residuals plot of the straight line model.
233
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
Fig. 4 – Evaluation of the predictive properties: RMSECV vs. polynomial order plot.
the examined concentration range (=evaluation of the fit by R2 (Fig. 3a), RMSE (Fig. 3b) and the analysis of the residuals (Fig. 3c and d)). In Fig. 3a and b it is observed that the model fit is improved considerably from a straight line to the second order model, while for higher order models only a slight further improvement is seen. In contrast to the linear model (Fig. 3d), the residuals plot of the second order polynomial model (Fig. 3c) does not show any trend. Thus, a random distribution of positive and negative residuals is obtained, which means that the model appropriately describes the data. The RMSECV (from LOO validation) versus polynomial order plot (Fig. 4) shows that the predictive properties of the third, fourth and fifth order models are not significantly better than for the second order model. In general, it has been proposed to select the first local minimum in the RMSECV versus order plot. As the lowest possible order model should be used, the second order polynomial regression model was selected. The 95% confidence interval from the observed data (Fig. 5) shows the limits of the true response (peak intensity) at a particular x-value (concentration), which may be expected with 95% probability. The 95% prediction interval for measured peak intensities at known concentrations (Fig. 5) shows with 95% assurance within which limits the peak intensity will be for one measurement of a given suspension concentration.
3.2.
Fig. 5 – 95% prediction interval and 95% confidence interval of the estimated second order polynomial FT-Raman calibration model.
Fig. 6 – Simple linear HPLC calibration model.
calibration model was obtained: y = 8.59368 × 106 x + 25376.1 (Fig. 6).
3.3.
The recovery, repeatability, linearity and range of the Raman and HPLC analyses were validated by LOO cross-validation. An overview is given in Table 2. It is clear that the bias of both methods is similar. The repeatability of the HPLC method appears better than for the Raman method.
3.4.
Development of the HPLC calibration model
The development of the HPLC calibration model was performed in the same way as the Raman model. A simple linear
Validation of the Raman and HPLC methods
Drug analysis of the unknown ointments
The drug content of 25 ointments was determined using the same procedure as for the cross-validated ointments. The sal-
Table 2 – LOO cross-validation of the Raman and the HPLC method Concentration (%, w/w)
14.97 20.08 25.54 30.29 35.44 40.55 45.13
Raman Recovery (%) (n = 10) 96.68 101.04 101.16 101.35 98.48 99.07 100.79
Repeatability (%R.S.D.) (n = 10) 4.54 7.35 2.62 4.37 3.13 2.28 2.58
HPLC Linearity (r2 )
Recovery (%) (n = 3)
0.9907
98.52 99.89 100.93 100.57 100.04 99.86 99.70
Repeatability (%R.S.D.) (n = 3) 0.28 0.11 0.34 0.10 0.23 0.19 0.34
Linearity (r2 )
0.9997
234
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
Table 3 – FT-Raman and HPLC analyses of 25 ointments Sample code
Raman Salicylic acid found (%, w/w)
O1 O2 O3 O4 O5 O6 O7 O8 O9 O10 O11 O12 O13 O14 O15 O16 O17 O18 O19 O20 O21 O22 O23 O24 O25
20.85 19.55 21.56 20.36 20.80 20.22 19.68 19.32 20.48 24.21 22.84 21.37 15.91 24.36 21.70 19.56 22.77 23.47 20.94 20.29 22.96 22.06 21.82 16.83 20.46
HPLC Deviation from 20% (%) 4.25 2.25 7.80 1.80 4.00 1.10 1.60 3.40 2.40 21.05 14.20 6.85 20.45 21.80 8.50 2.20 13.85 17.35 4.70 1.45 14.80 10.30 9.10 15.85 2.30
icylic acid dose per ointment calculated from the FT-Raman data was significantly different from the results obtained by HPLC-analysis (Table 3) as proven by a paired t-test (p = 0.01). A possible reason for this is that some unknown ointments were prepared with salicylic acid having a different particle size than the salicylic acid (<180 m) used for the standards. Therefore, microscopic measurements were performed on the unknown ointments to determine the particle size (cf. Section 2.7). The underlined ointments (Table 3) were found to have a particle size smaller than 90 m. Performing a paired t-test with these ointments excluded (O3, O14, O17, O18, O21), indicates that the salicylic acid dose per ointment calculated from the FT-Raman data is not significant different from the results obtained by HPLC analysis (p = 0.22). The Raman results for the ointments with small particles tend to be overestimated. As the target concentration of the unknown ointments was 20% (w/w) and as the maximum allowed error is 10% (according to the TMF), three ointments (O11, O13, O24) appeared not to satisfy this requirement, which was confirmed by both the Raman and HPLC analyses. Thus, both techniques are able to identify the ointments that deviate more than allowed from the target concentration.
4.
Conclusion
The results showed that the fast and non-destructive FTRaman spectroscopic method is reliable for the quantitative determination of salicylic acid in ointments, having a similar bias compared to the HPLC method, which requires a sample preparation. The repeatability of the Raman method is less good, but acceptable. An important requirement for the
Salicylic acid found (%, w/w) 20.44 19.34 20.47 20.35 19.61 19.79 20.10 20.82 20.12 19.98 22.48 20.04 17.35 20.06 20.45 20.09 20.59 20.01 20.42 20.58 20.23 21.00 20.72 17.09 21.38
Deviation from 20% (%) 2.20 3.30 2.35 1.75 1.95 1.05 0.50 4.10 0.60 0.10 12.40 0.20 13.25 0.30 2.25 0.45 2.95 0.05 2.10 2.90 1.15 5.00 3.60 14.55 6.90
Raman method is that the particle size of salicylic acid present in the samples is similar to that used for the preparation of the standards.
references
Campbell Roberts, S.N., Williams, A.C., Grimsey, I.M., Booth, S.W., 2002. Quantitative analysis of mannitol polymorphs. FT-Raman spectroscopy. J. Pharm. Biomed. Anal. 28, 1135–1147. Christopher, F.J., 1999. Review of pharmaceutical applications of Raman spectroscopy. In: Pelletier, M.J. (Ed.), Analytical Applications of Raman Spectroscopy. London, UK, pp. 224–275. Clegg, I., Everall, N., 2003. On-line measurement of crystalline forms. Eur. Pharm. Rev. (3), 56–62. Compton, D.A.C., Compton, S.V., 1991. Examination of packaged consumer-goods by using FT-Raman spectrometry. Appl. Spectrosc. 45 (10), 1587–1589. ˜ A.M., Bosque Sendra, Cuadros Rodriguez, L., Garc´ıa Campana, J.M., 1996. Statistical estimation of linear calibration range. Anal. Lett. 29, 1231–1239. De Beer, T.R.M., Vergote, G.J., Baeyens, W.R.G., Remon, J.P., Vervaet, C., Verpoort, F., 2004. Development and validation of a direct non-destructive quantitative method for medroxyprogesterone acetate in a pharmaceutical suspension using FT-Raman spectroscopy. Eur. J. Pharm. Sci. 23, 355–362. Long, D.A., 2002. The Raman Effect: A Unified Treatment of the Theory of Raman Scattering by Molecules. Wiley–Interscience, Chichester, West Sussex, England, 597 pp. Massart, D.L., Vandeginste, B.G.M., Buydens, L.M.C., De Jong, S., Lewi, P.J., Smeyers-Verbeke, J., 1997. Handbook of Chemometrics and Qualimetrics Part A. Elsevier, Amsterdam, The Netherlands.
e u r o p e a n j o u r n a l o f p h a r m a c e u t i c a l s c i e n c e s 3 0 ( 2 0 0 7 ) 229–235
McCreery, R.L., 2000. Raman Spectroscopy for Chemical Analysis. Wiley–Interscience, New York, USA. Miller, J.N., 1991. Basis statistical methods for analytical chemistry. Part 2. Calibration and regression methods: a review. Analyst 116, 3–14. Niemczyk, T.M., Delgado-Lopez, M.M., Allen, F.S., 1998. Quantitative determination of bucindolol concentration in intact gel capsules using Raman spectroscopy. Anal. Chem. 70 (13), 2762–2765. Pellow-Jarman, M.V., Hendra, P.J., Lehnert, R.J., 1996. The dependence of the Raman signal intensity on particle size for crystal powders. Vibrat. Spectrosc. 12, 257–261. Skoulika, S.G., Georgiou, C.A., 2001. Rapid quantitative determination of ciprofloxacin in pharmaceuticals by use of solid-state FT-Raman spectroscopy. Appl. Spectrosc. 55 (9), 1259–1265. Skoulika, S.G., Georgiou, C.A., 2003. Rapid noninvasive quantitative determination of acyclovir in pharmaceutical solid dosage forms through their poly(vinyl chloride) blister by solid-state fourier transform Raman spectroscopy. Appl. Spectrosc. 57 (4), 407–412. Tsuchihashi, H., Katagi, M., Nishikawa, M., Tatsuno, M., Nishioka, H., Nara, A., Nishio, E., Petty, C., 1997. Determination of
235
methamphetamine and its related compounds using fourier transform Raman spectroscopy. Appl. Spectrosc. 51 (12), 1796–1799. Vander Heyden, Y., Popovici, S.T., Schoenmakers, P.J., 2002. Evaluation of size-exclusion chromatography and size-exclusion electrochromatography calibration curves. J. Chromatogr. A 957, 127–137. Vankeirsbilck, T., Vercauteren, A., Baeyens, W., Van der Weken, G., Verpoort, F., Vergote, G., Remon, J.P., 2002. Applications of Raman spectroscopy in pharmaceutical analysis. Trends Anal. Chem. 21 (12), 869–877. Vergote, G.J., Vervaet, C., Remon, J.P., Haemers, T., Verpoort, F., 2002. Near-infrared FT-Raman spectroscopy as a rapid analytical tool for the determination of diltiazem hydrochloride in tablets. Eur. J. Pharm. Sci. 16, 63–67. Vergote, G.J., De Beer, T.R.M., Vervaet, C., Remon, J.P., Baeyens, W.R.G., Diericx, N., Verpoort, F., 2004. In-line monitoring of a pharmaceutical blending process using FT-Raman spectroscopy. Eur. J. Pharm. Sci. 21, 479–485. Yang, H., Irudayaraj, J., 2002. Rapid determination of Vitamin C by NIR, MIR and FT-Raman techniques. J. Pharm. Pharmacol. 54 (9), 1247–1255.