- Email: [email protected]
Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy
Accepted Manuscript Title: Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy Authors: Lisa B. Schlegel, Manfred Schubert-Zsilavecz, Mona Abdel-Tawab PII: DOI: Reference:
S0731-7085(16)31044-5 http://dx.doi.org/doi:10.1016/j.jpba.2017.04.048 PBA 11246
To appear in:
Journal of Pharmaceutical and Biomedical Analysis
Received date: Revised date: Accepted date:
7-11-2016 26-4-2017 28-4-2017
Please cite this article as: Lisa B.Schlegel, Manfred Schubert-Zsilavecz, Mona Abdel-Tawab, Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy, Journal of Pharmaceutical and Biomedical Analysishttp://dx.doi.org/10.1016/j.jpba.2017.04.048 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy Lisa B. Schlegela*, Manfred Schubert-Zsilavecza,b, Mona Abdel-Tawaba aZentrallaboratorium
Deutscher Apotheker (Central laboratory of German pharmacists), Carl-Mannich-Str. 20, 65760 Eschborn, Germany of Pharmaceutical Chemistry, Goethe University Frankfurt, Max-von-Laue-Str. 9, 60438 Frankfurt, Germany *Corresponding author. E-Mail address: [email protected], Tel. +496196937854 bInstitute
Lisa Britta Schlegel Pharmacist Central Laboratory of German Pharmacists Carl-Mannich-street 20 65760 Eschborn Germany Email: [email protected] Phone: +49 (0) 6196 937 854 Fax: +49 (0) 6196 937 815
Highlights
Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy Quantitative NIR methods for quality control of semi-solid preparations are proposed. Accuracy strongly depends not only on the concentration range, but also on the drug. Increase in water content of the cream matrices causes a loss in accuracy. Salicylic acid and urea allow content prediction within ± 5 % deviation from the target value. Critical substances are erythromycin and low dosed metronidazole.
Abstract Near infrared (NIR) spectroscopy is increasingly gaining significance in the pharmaceutical industry for quality and in-process control. However, the potential of this method for quantitative quality control in pharmacies has long been neglected and little data is available on its application in analysis of creams and ointments. This study evaluated the applicability of NIR spectrometer with limited wavelength range (1000 – 1900 nm) for quantitative quality control of six different dermatological semi-solid pharmaceutical preparations. Each contained a frequently used active ingredient in a common concentration either in a water-free lipid base or in an aqueous cream matrix. Based on direct NIR transflectance measurements through standardized glass beakers and partial least squares (PLS) multivariate calibration, quantitative models were generated comparing several data pre-processing methods. Whereas difficulties were observed for mixtures containing 2 % (w/w) metronidazole or 4 % (w/w) erythromycin, content determination was possible with sufficient accuracy for salicylic acid (5 % (w/w)) and urea (10 % (w/w)) in hydrophilic as well as in lipophilic formulations meeting the limit of a maximum deviation of ± 5 % (relative) from the reference values. Exemplarily, one of the methods was successfully validated according to the EMA Guideline, determining several figures of merit such as specificity, linearity, accuracy, precision and robustness. Keywords: near infrared spectroscopy; ; ; ; ; , pharmacy, quality control, creams, ointments, quantification
1. Introduction In the course of the past 40 years, near infrared spectroscopy (NIRS) has experienced enormous enhancements and has become a very valuable tool in quality control in the pharmaceutical industry. Together with optical fiber probes, NIRS has significantly simplified identification and qualification processes for active pharmaceutical ingredients (APIs) and excipients. Furthermore it has gained importance in on-line or in-line determination of critical process parameters such as moisture content or blend uniformity – not at least due to the FDA PAT initiative in 2004 [1,2,3]. Usage of NIRS for API 1
quantification purposes has been described as well, but mainly for solid formulations including tablets, capsules or powder mixtures [4] and to a lesser extent for fluids [4,5,6]. Direct determination of APIs in semi-solid formulations has been described using FT-IR spectroscopy with photoacoustic detection[7], but little data for the application of NIR spectroscopy in semi-solid formulations is available [4]. Over the last decade NIR spectrometer with limited wavelength range have become affordable and, therefore, meanwhile are widely available in community and clinical pharmacies as well. Until now, the application of these instruments did not exceed API identification, but since there is a rising demand for economic and practical methods in quality control of extemporaneous mixtures, the question arises whether the mentioned spectrometers are suitable for API quantification. Topical formulations such as creams and ointments represent a large fraction of individually tailored preparations. Inherently, semisolid formulations usually come along with challenges in content determination. Mostly, the API has to be extracted from the matrix to become accessible to further analysis which requires time-consuming extraction procedures followed by liquid chromatography and cannot be carried out in pharmacies. NIRS works rapid, non-destructive and is independent from any sample preparation. Furthermore, the omission of organic solvents is environmentally friendly and reduces the exposition of employees to hazardous substances. Therefore, this time and cost saving technique could represent a promising alternative to conventional quantification methods. Accordingly, the aim of our study was to evaluate the applicability of NIRS on semi-solid preparations with different frequently used APIs in hydrophilic and lipophilic creams as well as water-free lipid formulations and to find out if quantitative quality control is possible with sufficient accuracy. Our focus lay on comparing the influence of the varying drugs as well as of the water content in the cream bases on the predictive quality of the developed models. Since NIRS is not suitable for analysis of very small drug amounts, APIs were chosen that are used in dermatological effective concentrations above 1 % (w/w). Taking into account the results of a former survey on most frequently used APIs, the keratolytic active salicylic acid and the moisturizer urea were selected for analysis, supplemented by two antimicrobial substances, namely erythromycin and metronidazole [8]. We carried out NIR transflectance measurements for six different formulations each containing one of the aforementioned drugs in common concentrations and developed quantitative NIR models for their quantification in the respective matrices. To build appropriate fitting methods with low prediction errors concerning Root Mean Square Error of Prediction (RMSEP) and Standard Error of Prediction (SEP) as well as acceptable coefficients of determination, we used partial least squares regression (PLS) and multiple spectral pretreatments. Since validation is a basic requirement for any analytical method, one of the methods was validated in accordance with the main suggestions of the EMA Guideline [9]. 2. Experimental 2.1 Material and Methods 2.1.1 Samples All ointments and creams were prepared in-house using verified compendial and standardized pharmaceutical manufacturing processes (e.g. DAC/NRF) for ensuring product quality (Table 1). All APIs were weighed directly into the jars, using analytical balances being daily calibrated and complying with GMP regulations. API and ointment or cream matrix were mixed using the electronic mixing device TOPITEC® Automatic (Wepa Apothekenbedarf, Hillscheid, Germany), following the manufactures instructions regarding stirring time and rotation speed. API and semisolid matrix were filled into the jars adhering to the recommended “sandwich-technique”, assuring that solid APIs are entirely covered in cream base before the stirring device is brought in. As the APIs and matrices were weighed directly into the jars, any risk of substance loss may be excluded.
2
Hence the analytically weighed masses of the APIs expressed as weight percentages were used as reference values in the project. In order to build a robust NIR model, the calibration set should contain as much variation as possible. Therefore, a careful risk assessment of the variables that might represent critical quality attributes was performed. Hence raw material attributes were considered by preparing samples including APIs and semi-solid matrices from at least three different batches and, if available, from different manufacturers, whenever available. Also possible cross-correlations of different API/excipient concentrations were considered by changing the concentrations of the API over the range from 70 – 130 % (rel.). As the tested semi-solid samples were exclusively composed of the API and commercially available premanufactured ointment bases, changing the excipient/excipient ratio represents no critical influence in daily practice and may be thus neglected. Considering every possible combination between API and ointment, nine specimens were manufactured in case of salicylic acid in vaseline for each concentration level (70, 90, 100, 110 and 130 % (rel.), which resulted in a total of 45 specimens. For all other preparations three creams were manufactured for all of the fore cited concentration levels, randomly combining different batches of API and matrix, resulting in a total of 15 specimens each. Preparation 1: Salicylic acid 5 % (w/w) in vaseline DAB (NRF 11.43.) Salicylic acid trituration/concentrate 50 % (w/w) DAC (in vaseline, white DAB) * vaseline, white DAB Mixing conditions: 1000 rpm, 6:00 min *manufactured in advance, mixing conditions: 1000 rpm, 6:00 min, followed by usage of a three roll mill and another mixing cycle of 1000 rpm, 2:00 min Preparation 2: Hydrophilic salicylic acid cream 5 % (w/w) (based on NRF 11.106) Salicylic acid Nonionic hydrophilic cream (Nichtionische hydrophile Creme SR DAC NRF S.26.) Mixing conditions: 1000 rpm, 5:00 min Preparation 3: Hydrophobic urea cream 10 % (w/w) (based on NRF 11.129) Urea Hydrophobic cream (Hydrophobe Basiscreme DAC) Mixing conditions: 700 rpm, 5:00 min Preparation 4: Hydrophilic urea cream 10 % (w/w) Urea Hydrophilic cream (Basiscreme DAC) Mixing conditions: 1500 rpm, 5:00 min Preparation 5: Hydrophilic erythromycin cream 4 % (w/w) (based on NRF 11.77.) Erythromycin Hydrophilic cream (Basiscreme DAC) Mixing conditions: 1500 rpm, 5:00 min Preparation 6: Hydrophilic metronidazole cream 2 % (w/w) (based on NRF 11.91.) Metronidazole trituration/concentrate 10 % (in Nonionic hydrophilic cream and Nonionic hydrophilic liniment (lotion)) Nonionic hydrophilic liniment (lotion) (Nichtionisches hydrophiles Liniment SR DAC) Mixing conditions: 700 rpm, 4:00 min
5.0 g to 50.0 g
1.25 g to 25.0 g
2.50 g to 25.0 g
2.50 g to 25.0 g
0.80 g to 20.0 g
5.0 g to 25.0 g
Table 1 Overview of the semi-solid preparations, NRF = Neues Rezeptur Formularium (New German Formulary), DAC = Deutscher Arzneimittel-Codex (German Drug Codex), DAB = Deutsches Arzneibuch (German Pharmacopoeia), SR = Standardrezepturen (standardized preparations)
3
Salicylic acid and Erythromycin were purchased from Caesar & Loretz GmbH (Caelo), Hilden, Fagron GmbH & Co. KG, Barsbüttel and Euro OTC Pharma GmbH, Bönen. Suppliers for vaseline, white DAB were the given ones and, in addition, Bombastus Werke AG, Freital. Nonionic hydrophilic cream (Nichtionische hydrophile Creme SR DAC NRF S.26.) was purchased from Bombastus, Caelo, Euro OTC and Pharmazeutisches Kontroll- und Herstellungslabor GmbH (PKH Halle), Halle. Urea was delivered by Caelo and Fagron. Suppliers for the hydrophobic cream (Hydrophobe Basiscreme DAC) were Bombastus, Caelo, Fagron and Pharmachem Pößneck GmbH & Co. KG, Pößneck. Hydrophilic cream (Basiscreme DAC) was purchased from Bombastus, Caelo, Fagron, Paul W. Beyvers GmbH, Berlin and also Henry Lamotte OILS GmbH, Bremen. Metronidazol concentrate was provided by PKH Halle. Nonionic hydrophilic liniment (Nichtionisches hydrophiles Liniment SR DAC) was supplied from Bombastus, Caelo and PKH Halle. Provider of vaseline, yellow, hydrophilic ointment base (Hydrophile Salbe DAB) and wool wax alcohol ointment (Wollwachsalkoholsalbe DAB) was Caelo. 2.1.2 NIR spectrometer Spectra of the preparation salicylic acid 5 % (w/w) in vaseline DAB (NRF 11.43.) were recorded on a HiperScan apo-ident (Dresden, Germany). For the analysis of the remaining formulations the HiperScan finder was used, presenting an enhanced version of the apo-ident spectrometer with a heated (50 °C) and therefore temperature stabilized detector. All measurements were conducted at ambient temperature. 2.1.3 Software All spectra were acquired using the SL Predictor software (SensoLogic, Norderstedt, Germany) over a wavelength range of 1000 to 1900 nm with a total of 901 measuring points (2000 scans, resolution: 1 nm). Spectral treatments and calculations were carried out using the SL Calibration Wizard software (SensoLogic, Norderstedt, Germany). Further calculations, generation of additional graphics and data administration were done with SL Calibration Workshop and SL Utilities (SensoLogic, Norderstedt, Germany). 2.1.4 Sample presentation The samples were measured in transflectance mode, using glass beakers for sample presentation, centered on the measuring window of the spectrometer. A diffuse reflecting transflectance insert made of aluminum was put in the beaker holding the sample thus resulting in a defined layer thickness of the semi-solid preparation of 0.4 mm ± 0.05 mm. The empty beaker containing the aluminum transflectance insert was used as reference for every measurement. Six samples were taken from every single manufactured specimen, resulting in 270 spectra for salicylic acid in vaseline and in 90 spectra for all other described creams. To prevent any impact of stability issues, all analysis generally were carried out within one or two days after manufacturing of the specimen - at the maximum within one week. 2.1.5 Multivariate data analysis The spectra for each semi-solid preparation were divided into calibration and test set following the common ratio of 2:1 resulting in 180 spectra for the calibration set and 90 spectra for the test set for preparation 1 (salicylic acid in vaseline), and 60 calibration set spectra and 30 test set spectra for all other formulations (preparation 2 to 6), respectively. Partial least squares regression was carried out for each calibration set, testing different spectral treatments such as absorbance (Abs), detrend with a polynomial order of 0, 1 or 2 (Det), standard normal variate (SNV), normalization (Norm), first derivative (FD) and second derivative (SD) both with varying segments and gaps, in order to find the optimal fitted model concerning low standard errors of calibration (SEC) and prediction (SEP) and with acceptable correlation coefficients. 3. Results and discussion 4
Initially, we performed a feasibility study for salicylic acid in vaseline with a wider range of concentrations with contents of the active ingredient between 0 and 25 % (w/w). The micronized drug particles are suspended in the matrix. The application of data pretreatments such as the SNV followed by the second derivative (15/0) showed to be very effective as can be seen in Fig. 1. In the wavelength area between 1590 and 1690 nm a distinct concentration dependent variance of the signal intensity could be observed, which was promising for developing and validating a quantitative model for this preparation.
Fig.1 Second derivative and standard normal variate corrected spectra for salicylic acid 0 – 25 % (w/w) in vaseline (a: total spectrum, b: detail)
3.1 Validation of a quantitative method for salicylic acid 5 % (w/w) in vaseline Based on the EMA Guideline on the use of near infrared spectroscopy by the pharmaceutical industry and the data requirements for new submissions and variations [9] the following criteria of the quantitative NIR model were studied: Standard Error of Prediction (SEP), specificity, linearity, range, accuracy, precision and robustness. Table 2 gives an overview about the established models with the lowest errors of prediction. Wavelength range (nm) 1000 - 1900
Data pretreatments SNV
PLS latent variables 3
SEC (Cal Set) (% (w/w)) 0.079
5
SEP (Test Set) (% (w/w)) 0.084
Rprd
a1
a0
0.9965
1.01
-0.04
1.00 0.00 1000 - 1900 Norm 4 0.084 0.092 0.9958 1.01 -0.04 1000 - 1900 Abs + SNV 4 0.085 0.093 0.9958 1.00 0.01 1000 - 1900 SD(15/0) + SNV 4 0.095 0.1019 0.9948 Table 2 Comparative evaluation of different NIR models on the example of salicylic acid in vaseline: spectral range, spectral pretreatment, number of PLS factors, SEC for the calibration set, SEP for the test set, correlation coefficient of validation (test set) (Rprd) and corresponding slope (a1) and intercept (a0)
3.1.1 Standard Error of Prediction The models were tested on independent validation samples, prepared several weeks after the calibration samples using new batches of API and vaseline. The validation set consisted of five samples including API concentrations of 70, 90, 100, 110 and 130 % (rel.) of the target concentration which were measured in triplicate. The results are described in Table 3. The quantitative model based on the SNV transformation of the data showed the lowest SEP and the smallest number of PLS latent variables and was therefore chosen for further validation. Wavelength PLS latent SEP (Val Set) Rprd a1 a0 Data pretreatments range (nm) variables (% (w/w)) 0.9978 1.02 -0.12 1000 - 1900 SNV 3 0.071 0.9974 0.99 0.04 1000 - 1900 Norm 4 0.073 0.9979 1.02 -0.11 1000 - 1900 Abs + SNV 4 0.068 0.9967 1.02 -0.05 1000 - 1900 SD(15/0) + SNV 4 0.084 Table 3 Comparative evaluation of the generated NIR models on independent validation samples of salicylic acid in vaseline: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
3.1.2 Specificity Since identification of raw materials is the main application field of the applied NIR spectrometer apoident and finder in this project, appropriate qualitative spectral libraries already exist and didn´t need to be developed. Nevertheless, no qualitative identification test was carried out for the applied APIs ahead of quantification as identification of every raw material may be taken as given, being mandatory by law in German pharmacies. Based on that background the important issue of specificity was addressed by challenging the quantitative model with different out of specification (OOS) samples, such as samples containing no API, any other than the desired API or containing a wrong matrix as well as samples with API content exceeding the specification limits. Hence the predictions for samples containing no salicylic acid (placebo), salicylic acid in concentrations of 2.5 and 7.5 % (w/w) (i.e. 50 respectively 150 % (rel.) of the target concentration), as well as for samples containing other APIs and other matrices that might be mistaken in practice such as acetylic salicylic acid and urea instead of salicylic acid and hydrophilic ointment base (Hydrophile Salbe DAB), wool wax alcohol ointment (Wollwachsalkoholsalbe DAB) and yellow vaseline instead of white vaseline were studied. In agreement with common practice for quality control of individually prepared pharmaceutical preparations, we defined samples with predicted values below 90 % (rel.) and above 110 % (rel.) of the target concentration as identified OOS samples. The results for placebo samples show that, in spite of predictions not equal to zero but close to 1 % (w/w) API concentration, the placebo samples could easily be distinguished from samples containing the target concentration and would not falsely pass quality control tests. The predicted values for the API content were close to 4 % (w/w) for samples with acetylic salicylic acid due to the strong structural similarity and 0 % (w/w) for the samples containing urea. Both would also not have passed quality control tests. For the samples with exchanged matrices we observed that only 6
samples containing wool wax alcohol ointment were recognized as different with sufficient distinctness (predicted values around 4 % (w/w) for the API content). The tested hydrophilic ointment (containing more than 70 % (w/w) vaseline itself) and yellow vaseline were not recognized as “wrong” by the model – yellow vaseline being distinguishable visually for the human operator, of course. For both, predicted values between 5.26 and 5.37 % (w/w) API content were obtained. The predictions for samples holding 2.5 respectively 7.5 % (w/w) salicylic acid showed very accurate results close to the actual values, although the concentrations were not included in the range of the calibration samples. A summary of the predicted values for the OOS samples can be found in Table 4. Out of specification samples Placebo: vaseline, white (mean out of 9 samples)
Content of salicylic acid (pred.) ± SEM 1.07 % (w/w) ± 0.005 % (w/w)
False API: acetylic salicylic acid (mean out of 3 samples) False API: urea (mean out of 3 samples)
3.87 % (w/w) ± 0.004 % (w/w) 0 % (w/w)
False matrix: wool wax alcohol ointment (mean out of 3 samples)
4.1 % (w/w) ± 0.002 % (w/w)
False matrix: hydrophilic ointment base (mean out of 3 samples)
5.29 % (w/w) ± 0.001 % (w/w)
False matrix: vaseline, yellow (mean out of 3 samples)
5.32 % (w/w) ± 0.001 % (w/w)
API content 50 % (rel.): salicylic acid 2.5 % (mean out of 3 samples)
2.55 % (w/w) ± 0.001 % (w/w)
API content 150 % (rel.): salicylic acid 7.5 % (mean out of 3 samples) 7.39 % (w/w) ± 0.0003 % (w/w) Table 4 Prediction results (mean of multiple measurements) for OOS samples for salicylic acid in vaseline
Specificity of the method can also be proved by looking at the loading plots for the first latent variables shown in Fig. 2. A large contribution of the band around 1659 nm can be observed both for latent variable 1 and 2, perfectly matching the spectral properties of salicylic acid, shown in Fig. 3. Other bands contributing to the model can be assigned to vaseline.
7
Fig. 2 Loading plots for the first and second PLS factor (a: Factor 1, b: Factor 2)
Fig. 3 NIR spectrum of salicylic acid
3.1.3. Range and linearity The equally distributed residues for the validation set illustrated in Fig. 4 suggest absence of systematic trends. Assuming linearity of the model, the correlation coefficient can be determined for a fitted straight line in the predicted versus actual values plot (Fig. 5.). For this model, it is very close to 1 (0.9978) meaning, that there is good accordance between actual and predicted values for the API content. No remarkable difference between residues in the lower and higher concentration range was observed which strengthens the linearity assumption. Moreover the RPD value was calculated. The result of 11.99 was found to be indicating the excellent predictive quality of the applied model. [10]
8
Fig. 4 Visualization of the distribution of the residues (content of salicylic acid % (w/w))
Fig. 5 Plot of predicted versus actual values for the validation samples (% (w/w))
3.1.4 Accuracy The target values for the Standard Error of Prediction were set as ≤ 0.25 % (w/w), corresponding to a variance of ± 5 % (rel.) around the target API concentration. This could be achieved and even under-run with all models mentioned above. All predictions were found to be within ± 5 % (rel.) deviation from the actual value with an averaged deviation of 1.2 % (rel.). A paired samples t-test was also performed to check whether the predicted and the actual values varied significantly. The result showed that the difference between the actual and the predicted values is not significantly different from 0 (α = 0.05, pvalue = 0.24). 3.1.5 Precision The precision of the method was evaluated by determining repeatability and intermediate precision. To evaluate repeatability six samples were taken out of a 5 % (w/w) salicylic acid vaseline one after another and measured separately on the same day. In order to estimate intermediate precision additionally six spectra were recorded for six samples taken out of the same preparation on six consecutive workdays. For both measuring series the coefficient of variation for the predicted values was ≤ 1.5 % (rel.) (1.27 % (rel.) and 1.36 % (rel.) respectively). 9
3.1.6 Robustness The robustness of the quantitative model was evaluated through tests at different sample positions, different temperatures and tests with preparations with different API particle sizes (micronized vs. powdered salicylic acid). For the evaluation of the influence of the sample position, one sample taken out of a 5 % (w/w) salicylic acid vaseline was measured six times in the same beaker, the glass being rotated piecewise prior to the next spectrum recording. Additionally, another sample was measured six times in series without moving the beaker glass. For both measuring series the coefficient of variation for the predicted values was ≤ 1 % (rel.) (0.52 % (rel.) and 0.79 % (rel.) respectively). The result of the paired t-test showed that the difference between the measuring series is not significantly different from 0 (α = 0.05, p-value = 0.18), indicating that rotating the sample had no significant influence. To test the influence of the particle size of salicylic acid, two specimens that contained 5 % (w/w) of powdered but not micronized API were prepared and analyzed in triplicate. The variance in particle size led to predicted values scattering around the target value to a bigger extent than the predictions for the validation samples with micronized salicylic acid. A variation coefficient of 3.1 % (rel.) was obtained. But still, with the exception of one value, all were found in the range of ± 5 % (rel.) around the target value. The t-test showed that there is no significant difference between the means of the values for the samples with micronized and with powdered salicylic acid (α = 0.05, p-value = 0.12), indicating that the particle size had no significant influence. Temperature is known to influence NIR measurements. Therefore, the validation sample set was tested at different temperatures. Ahead of recording the spectra, samples had been stored at the following temperatures: room temperature 20 – 25 °C, cool storage 15°C ± 3°C and cold storage 5°C ± 3°C for at least 24 hours. As a side effect of the decrease in temperature a firmer consistency of the preparations could be noticed. The measurements were conducted in triplicate and carried out immediately after removing the samples out of the fridge. As expected, increasing values for bias and RMSEP were observed for both sample series measured at other temperatures than room temperature that were not included in the calibration: room temperature RMSEP 0.074 % (w/w), bias 0.011 % (w/w); cool storage RMSEP 0.075 % (w/w), bias -0.034 % (w/w); cold storage RMSEP 0.096 % (w/w), bias -0.043 % (w/w). Nevertheless, the prediction errors were still very small. Application of the paired samples t-test showed only a significant difference comparing the values for cool samples and samples at room temperature (α = 0.05, p-value = 0.01), not between the other groups (room temperature and cold storage, α = 0.05, p = 0.054 and cool and cold storage, α = 0.05, p = 0.71). Furthermore, the influence of the state of the spectrometer was evaluated, since the validation was performed on the apo-ident device, which is not equipped with a heated detector. Hence, the time after starting the device (permanent running or just started) might possibly influence the signal intensity. After 24 hours in the turned off status the spectrometer was switched on again and immediately all five validation specimens were measured once more in triplicate. The result of the paired samples t-test showed that the means of the measuring series are significantly different from each other (α = 0.05, pvalue = 0.02). 3.2 Transfer of the analytical approach on other preparations Following the successful validation of the method, the approach was tested on different pharmaceutical preparations containing other frequently used APIs and more complex matrices especially with high percentages of water, which is known to disturb NIR measurements. To evaluate the possibility of reducing the amount of calibration specimens to be prepared in order to save costs and effort prior to new investigations, further quantitative models were developed for salicylic acid in vaseline using only one third of the originally generated spectra (60 spectra for the calibration set and 30 spectra for the test set). 10
The results for the prediction errors for the validation set were compared with those of the developed models based on the entire quantity of the spectra. As can be seen in Table 5 the standard errors of prediction revealed to be very small. Therefore, the following investigations for preparation 2 to 6 were conducted with 90 spectra from 15 manufactured creams in each case and six samples taken from every specimen. SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 0.9975 1.03 -0.11 1000 - 1900 SNV 3 0.076 0.9965 0.98 0.11 1000 - 1900 Norm 4 0.088 0.9974 1.02 -0.14 1000 - 1900 Abs + SNV 4 0.075 0.9964 0.98 0.11 1000 - 1900 SD(15/0) + SNV 4 0.089 Table 5 NIR models for salicylic acid in vaseline: spectral range, spectral pretreatment, number of PLS factors, correlation coefficient of validation, SEP for the validation set (Rprd) and corresponding slope (a1) and intercept (a0)
3.2.1 Development of a quantitative method for salicylic acid 5 % (w/w) in a hydrophilic cream Sticking to the API salicylic acid, first of all we changed the lipophilic anhydrous semi-solid matrix to a hydrophilic system with high water content of 64 % (w/w). As in case of vaseline the API exists in form of suspended particles in the hydrophilic cream as well. Comparing the spectra of salicylic acid in vaseline and in the hydrophilic cream, Fig. 6 shows the considerable contribution of the high water amount to the appearance of the spectra. Still, the influence of the API concentration can be visually detected around 1660 nm.
Fig. 6 Original spectra for all calibration and test set samples of a) salicylic acid in vaseline and b) salicylic acid in a hydrophilic cream 11
SEP (Val Set) Rprd a1 a0 Data PLS latent pretreatments variables (% (w/w)) 1000 – 1900 SD(15/0) + SNV 3 0.116 0.9979 1.03 0.11 1050 – 1850 SNV + SD(10/0) 3 0.127 0.9974 1,03 -0,14 1050 – 1330, 1500 – 1850 SD(15/0) + SNV 5 0,141 0.9979 1.06 -0.33 1050 – 1850 FD(2/15) + SNV 6 0.113 0.9978 1.01 -0.19 Table 6 NIR models for salicylic acid in a hydrophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0) Wavelength range (nm)
Quantitative models were developed using either the whole spectrum or omitting the area of the band characteristic for the –OH vibrations of water (in this case about 1350 – 1500 nm). Comparable results were received for the prediction errors. Table 6 shows the best models and the errors of prediction for the independent validation set, consisting out of five samples, measured in triplicate. The validation samples included API concentrations of 50, 95, 100, 105 and 150 % (rel.) of the target concentration to simulate OOS samples and samples that should easily pass quality control.
Fig. 7 Plot of predicted versus actual values for salicylic acid 5 % (w/w) in a hydrophilic cream
The results showed that the content of the API could be predicted correctly within 95 and 105 % (rel.) of the reference value with all models. Deviations greater than 5 % (rel.) were only observed for the prediction of the samples with 50 % (rel.) of the target concentration (2.5 % (w/w) salicylic acid in this case). The plot of predicted versus actual values for the method using second derivative and standard normal variate corrected spectra and three latent variables is shown in Fig. 7. 3.2.2 Development of quantitative methods for urea 10 % (w/w) in a hydrophilic cream and for urea 10 % (w/w) in a hydrophobic cream As salicylic acid, urea is used for dermatological purposes in a wide variety of matrices in concentrations from 3 % (w/w) as moisturizer or up to 40 % (w/w) as a strong keratolytic agent. For our studies, we selected a medium concentration of 10 % (w/w) as target concentration in a hydrophilic as well as a lipophilic cream. Urea is soluble in both of them as the water content amounts to 40 % (w/w) and 65 % (w/w), respectively. The best models are shown in Table 7 for the hydrophilic cream and in Table 8 for the hydrophobic cream.
12
SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1050 – 1850 FD(2/15) + SNV 4 0.105 0.9995 1.00 -0.05 1000 – 1900 SNV 6 0.145 0.9991 0.99 0.06 1050 – 1850 SNV 5 0.132 0.9992 1.00 0.10 1050 – 1850 Det 2 + SNV 4 0.171 0.9987 1.01 -0.14 Table 7 NIR models for urea in a hydrophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0) SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1050 – 1850 SNV 6 0.202 0.9985 1.03 -0.24 1050 – 1850 Det 1 + SNV 5 0.223 0.9981 1.03 -0.16 1000 – 1900 Det 2 + SNV 5 0.265 0.9971 1.03 -0.19 1000 – 1900 Abs + SD(15/0) + SNV 4 0.281 0.9970 0.96 0.35 Table 8 NIR models for urea in a lipophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
The results suggested that all models for hydrophilic and lipophilic cream can predict the API content correctly within 95 and 105 % (rel.) of the reference value for all validation samples. Remarkably, superior quality attributes such as lower predictive errors could be observed for the hydrophilic cream, although both matrices represent cream bases with comparable properties and number of components. This is mainly attributed to the 1.5 times higher water content in the lipophilic cream, which interferes with the major absorption band of urea in the area between 1450 and 1550 nm (Fig. 8, 9).
Fig. 8 NIR spectra of urea in a hydrophilic cream (calibration set), exemplary
13
Fig. 9 NIR spectrum of urea
3.2.1 Development of a quantitative method for metronidazole 2 % (w/w) in a hydrophilic liniment The determination of metronidazole in high concentrations (target concentration 10 % (w/w)) in a hydrophilic cream for vaginal application has been described by Baratieri et al [11]. As however metronidazole is used in dermatological preparations with a maximum content of 2 % (w/w), this concentration was chosen as target concentration for our study. The API was suspended in form of metronidazole concentrate into a hydrophilic semi-solid matrix (nonionic hydrophilic liniment) with a very high water content of 82 % (w/w). Lipophilic preparations were not taken into account in this study as they are not of therapeutic interest. The dimension of the prediction errors for all developed models are summarized in Table 11 and the actual versus predicted plot for the first model in Fig. 10 indicated that the determination of metronidazole was not possible with comparable accuracy as for salicylic acid and urea, which were tested in higher concentrations. Nevertheless, most of the predicted values could be found within ± 10 % (rel.) of the reference value for the validation samples with contents of 95, 100, 105 % (rel.) API. The prediction for the validation samples with 50 and 150 % (rel.) metronidazole content showed deviations around ± 20 % (rel.), but would have been detected correctly as OOS samples during quality control. SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1050 – 1850 Norm 7 0.174 0.9788 1.21 -0.35 1050 – 1350; SNV 5 0.178 0.9805 1.23 -0.40 1550 – 1850 1050 – 1300; SD(15/0) + SNV 4 0.196 0.9701 1.22 -0.41 1500 – 1850 1050 – 1850 SD(15/0) + SNV 4 0.195 0.9601 1.12 -0.19 Table 11 NIR models for metronidazole in a hydrophilic liniment: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
14
Fig. 10 Plot of predicted versus actual values for metronidazole in a hydrophilic liniment (% (w/w))
3.2.1 Development of a quantitative method for erythromycin 4 % (w/w) in a hydrophilic cream Erythromycin represents a mixture of three substances, erythromycin A, B and C and its solubility in a hydrophilic cream base depends on its concentration. Small amounts up to 2 % (w/w) will partly be dissolved while higher percentages will be found as suspended particles in the matrix. As lipophilic preparations with erythromycin are not used in practice, they were not included in the study. The development of quantitative models for this antibiotic drug led to unsatisfactory results. An overview is given in Table 12. The prediction of the API content was only possible with prevailing deviations of more than 10 % (rel.). However, the predictive quality rises with higher concentrations of the drug, as can be seen from Fig. 11. Hence, validation samples containing 6 % (w/w) API met the acceptance criteria of ± 5 % (rel.). SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1000 – 1900 SNV 8 0,333 0,9677 1,02 -0,44 1050 – 1850 SNV 7 0,362 0,9617 0,97 -0,22 1000 – 1900 Norm 8 0,331 0,9679 0,99 -0,33 1050 – 1850 Det1 + SNV 7 0,378 0,9588 0,95 -0,18 Table 12 NIR models for erythromycin in a hydrophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
15
Fig. 11 Plot of predicted versus actual values for erythromycin in a hydrophilic cream (% (w/w))
5. Conclusions All in all our results show that the application of NIRS for quantitative quality control in individually prepared semi-solid pharmaceutical formulations is possible in general, but accuracy and variance strongly depend on the active ingredient and its concentration range. In all cases an increase in the water content of the formulation was associated with a loss in accuracy. The low dosed preparation with a content of 2 % (w/w) metronidazole and the one containing 4 % (w/w) erythromycin being composed of several subgroups and showing varying solubility depending on the concentration present difficulties in developing quantitative models with sufficient accuracy. But for the active ingredients salicylic acid and urea, successful quantification meeting the limits of a maximum deviation of ± 5 % (rel.) from the actual value could be achieved. For these preparations near-infrared spectroscopy even with limited wavelength range spectrometers could present a promising alternative to conventional quality control methods. Acknowledgements The authors thank HiperScan GmbH, Dresden, Germany for providing the NIR spectrometer and calibration software. References [1] M. Blanco, J. Coello, H. Iturriaga, S. Maspoch, C. de la Pezuela, Near-infrared spectroscopy in the pharmaceutical industry, Anal. 123 (1998) 135R–150R. [2] G. Reich, Near-infrared spectroscopy and imaging: Basic principles and pharmaceutical applications, Adv. Drug Deliv. Rev. 57 (2005) 1109–1143. [3] M. Jamrógiewicz, Application of the near-infrared spectroscopy in the pharmaceutical technology, J. Pharm. Biomed. Anal. 66 (2012) 1–10. [4] Y. Roggo, P. Chalus, L. Maurer, C. Lema-Martinez, A. Edmond, N. Jent, A review of near infrared spectroscopy and chemometrics in pharmaceutical technologies, J. Pharm. Biomed. Anal. 44 (2007) 683–700.
16
[5] M.A.M. Silva, M.H. Ferreira, J.W.B. Braga, M.M. Sena, Development and analytical validation of a multivariate calibration method for determination of amoxicillin in suspension formulations by near infrared spectroscopy, Talanta 89 (2012) 342– 351. [6] L.M.M. Lê, E. Caudron, A. Baillet-Guffroy, L. Eveleigh, Non-invasive quantification of 5 fluorouracil and gemcitabine in aqueous matrix by direct measurement through glass vials using near-infrared spectroscopy, Talanta 119 (2014) 361–366. [7] R. Neubert, B. Colin, S. Wartewig, Direct Determination of Drug Content in Semisolid Formulations Using Step-Scan FT-IR Photoacoustic Spectroscopy, Pharm. Res. 14 (1997) 946–948. [8] H. Latsch, Umfrage zur Defekturherstellung, Dtsch. Apoth. Ztg. 153 (2013) 3693. [9] European Medicines Agency: Guideline on the use of near infrared spectroscopy by the pharmaceutical industry and the data requirements for new submissions and variations. http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2014/06/WC500167967.pdf, 2014 (accessed 21.07.2016). [10] P.C. Williams, D.C. Sobering, Comparison of commercial near infrared transmittance and reflectance instruments for analysis of whole grain and seeds, J. Near Infrared Spectrosc. 1 (1993) 25–32. [11] S.C. Baratieri, J.M. Barbosa, M.P. Freitas, J.A. Martins, J. Pharm. Biomed. Anal. 40 (2006) 51–55.
17
S0731-7085(16)31044-5 http://dx.doi.org/doi:10.1016/j.jpba.2017.04.048 PBA 11246
To appear in:
Journal of Pharmaceutical and Biomedical Analysis
Received date: Revised date: Accepted date:
7-11-2016 26-4-2017 28-4-2017
Please cite this article as: Lisa B.Schlegel, Manfred Schubert-Zsilavecz, Mona Abdel-Tawab, Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy, Journal of Pharmaceutical and Biomedical Analysishttp://dx.doi.org/10.1016/j.jpba.2017.04.048 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy Lisa B. Schlegela*, Manfred Schubert-Zsilavecza,b, Mona Abdel-Tawaba aZentrallaboratorium
Deutscher Apotheker (Central laboratory of German pharmacists), Carl-Mannich-Str. 20, 65760 Eschborn, Germany of Pharmaceutical Chemistry, Goethe University Frankfurt, Max-von-Laue-Str. 9, 60438 Frankfurt, Germany *Corresponding author. E-Mail address: [email protected], Tel. +496196937854 bInstitute
Lisa Britta Schlegel Pharmacist Central Laboratory of German Pharmacists Carl-Mannich-street 20 65760 Eschborn Germany Email: [email protected] Phone: +49 (0) 6196 937 854 Fax: +49 (0) 6196 937 815
Highlights
Quantification of active ingredients in semi-solid pharmaceutical formulations by near infrared spectroscopy Quantitative NIR methods for quality control of semi-solid preparations are proposed. Accuracy strongly depends not only on the concentration range, but also on the drug. Increase in water content of the cream matrices causes a loss in accuracy. Salicylic acid and urea allow content prediction within ± 5 % deviation from the target value. Critical substances are erythromycin and low dosed metronidazole.
Abstract Near infrared (NIR) spectroscopy is increasingly gaining significance in the pharmaceutical industry for quality and in-process control. However, the potential of this method for quantitative quality control in pharmacies has long been neglected and little data is available on its application in analysis of creams and ointments. This study evaluated the applicability of NIR spectrometer with limited wavelength range (1000 – 1900 nm) for quantitative quality control of six different dermatological semi-solid pharmaceutical preparations. Each contained a frequently used active ingredient in a common concentration either in a water-free lipid base or in an aqueous cream matrix. Based on direct NIR transflectance measurements through standardized glass beakers and partial least squares (PLS) multivariate calibration, quantitative models were generated comparing several data pre-processing methods. Whereas difficulties were observed for mixtures containing 2 % (w/w) metronidazole or 4 % (w/w) erythromycin, content determination was possible with sufficient accuracy for salicylic acid (5 % (w/w)) and urea (10 % (w/w)) in hydrophilic as well as in lipophilic formulations meeting the limit of a maximum deviation of ± 5 % (relative) from the reference values. Exemplarily, one of the methods was successfully validated according to the EMA Guideline, determining several figures of merit such as specificity, linearity, accuracy, precision and robustness. Keywords: near infrared spectroscopy; ; ; ; ; , pharmacy, quality control, creams, ointments, quantification
1. Introduction In the course of the past 40 years, near infrared spectroscopy (NIRS) has experienced enormous enhancements and has become a very valuable tool in quality control in the pharmaceutical industry. Together with optical fiber probes, NIRS has significantly simplified identification and qualification processes for active pharmaceutical ingredients (APIs) and excipients. Furthermore it has gained importance in on-line or in-line determination of critical process parameters such as moisture content or blend uniformity – not at least due to the FDA PAT initiative in 2004 [1,2,3]. Usage of NIRS for API 1
quantification purposes has been described as well, but mainly for solid formulations including tablets, capsules or powder mixtures [4] and to a lesser extent for fluids [4,5,6]. Direct determination of APIs in semi-solid formulations has been described using FT-IR spectroscopy with photoacoustic detection[7], but little data for the application of NIR spectroscopy in semi-solid formulations is available [4]. Over the last decade NIR spectrometer with limited wavelength range have become affordable and, therefore, meanwhile are widely available in community and clinical pharmacies as well. Until now, the application of these instruments did not exceed API identification, but since there is a rising demand for economic and practical methods in quality control of extemporaneous mixtures, the question arises whether the mentioned spectrometers are suitable for API quantification. Topical formulations such as creams and ointments represent a large fraction of individually tailored preparations. Inherently, semisolid formulations usually come along with challenges in content determination. Mostly, the API has to be extracted from the matrix to become accessible to further analysis which requires time-consuming extraction procedures followed by liquid chromatography and cannot be carried out in pharmacies. NIRS works rapid, non-destructive and is independent from any sample preparation. Furthermore, the omission of organic solvents is environmentally friendly and reduces the exposition of employees to hazardous substances. Therefore, this time and cost saving technique could represent a promising alternative to conventional quantification methods. Accordingly, the aim of our study was to evaluate the applicability of NIRS on semi-solid preparations with different frequently used APIs in hydrophilic and lipophilic creams as well as water-free lipid formulations and to find out if quantitative quality control is possible with sufficient accuracy. Our focus lay on comparing the influence of the varying drugs as well as of the water content in the cream bases on the predictive quality of the developed models. Since NIRS is not suitable for analysis of very small drug amounts, APIs were chosen that are used in dermatological effective concentrations above 1 % (w/w). Taking into account the results of a former survey on most frequently used APIs, the keratolytic active salicylic acid and the moisturizer urea were selected for analysis, supplemented by two antimicrobial substances, namely erythromycin and metronidazole [8]. We carried out NIR transflectance measurements for six different formulations each containing one of the aforementioned drugs in common concentrations and developed quantitative NIR models for their quantification in the respective matrices. To build appropriate fitting methods with low prediction errors concerning Root Mean Square Error of Prediction (RMSEP) and Standard Error of Prediction (SEP) as well as acceptable coefficients of determination, we used partial least squares regression (PLS) and multiple spectral pretreatments. Since validation is a basic requirement for any analytical method, one of the methods was validated in accordance with the main suggestions of the EMA Guideline [9]. 2. Experimental 2.1 Material and Methods 2.1.1 Samples All ointments and creams were prepared in-house using verified compendial and standardized pharmaceutical manufacturing processes (e.g. DAC/NRF) for ensuring product quality (Table 1). All APIs were weighed directly into the jars, using analytical balances being daily calibrated and complying with GMP regulations. API and ointment or cream matrix were mixed using the electronic mixing device TOPITEC® Automatic (Wepa Apothekenbedarf, Hillscheid, Germany), following the manufactures instructions regarding stirring time and rotation speed. API and semisolid matrix were filled into the jars adhering to the recommended “sandwich-technique”, assuring that solid APIs are entirely covered in cream base before the stirring device is brought in. As the APIs and matrices were weighed directly into the jars, any risk of substance loss may be excluded.
2
Hence the analytically weighed masses of the APIs expressed as weight percentages were used as reference values in the project. In order to build a robust NIR model, the calibration set should contain as much variation as possible. Therefore, a careful risk assessment of the variables that might represent critical quality attributes was performed. Hence raw material attributes were considered by preparing samples including APIs and semi-solid matrices from at least three different batches and, if available, from different manufacturers, whenever available. Also possible cross-correlations of different API/excipient concentrations were considered by changing the concentrations of the API over the range from 70 – 130 % (rel.). As the tested semi-solid samples were exclusively composed of the API and commercially available premanufactured ointment bases, changing the excipient/excipient ratio represents no critical influence in daily practice and may be thus neglected. Considering every possible combination between API and ointment, nine specimens were manufactured in case of salicylic acid in vaseline for each concentration level (70, 90, 100, 110 and 130 % (rel.), which resulted in a total of 45 specimens. For all other preparations three creams were manufactured for all of the fore cited concentration levels, randomly combining different batches of API and matrix, resulting in a total of 15 specimens each. Preparation 1: Salicylic acid 5 % (w/w) in vaseline DAB (NRF 11.43.) Salicylic acid trituration/concentrate 50 % (w/w) DAC (in vaseline, white DAB) * vaseline, white DAB Mixing conditions: 1000 rpm, 6:00 min *manufactured in advance, mixing conditions: 1000 rpm, 6:00 min, followed by usage of a three roll mill and another mixing cycle of 1000 rpm, 2:00 min Preparation 2: Hydrophilic salicylic acid cream 5 % (w/w) (based on NRF 11.106) Salicylic acid Nonionic hydrophilic cream (Nichtionische hydrophile Creme SR DAC NRF S.26.) Mixing conditions: 1000 rpm, 5:00 min Preparation 3: Hydrophobic urea cream 10 % (w/w) (based on NRF 11.129) Urea Hydrophobic cream (Hydrophobe Basiscreme DAC) Mixing conditions: 700 rpm, 5:00 min Preparation 4: Hydrophilic urea cream 10 % (w/w) Urea Hydrophilic cream (Basiscreme DAC) Mixing conditions: 1500 rpm, 5:00 min Preparation 5: Hydrophilic erythromycin cream 4 % (w/w) (based on NRF 11.77.) Erythromycin Hydrophilic cream (Basiscreme DAC) Mixing conditions: 1500 rpm, 5:00 min Preparation 6: Hydrophilic metronidazole cream 2 % (w/w) (based on NRF 11.91.) Metronidazole trituration/concentrate 10 % (in Nonionic hydrophilic cream and Nonionic hydrophilic liniment (lotion)) Nonionic hydrophilic liniment (lotion) (Nichtionisches hydrophiles Liniment SR DAC) Mixing conditions: 700 rpm, 4:00 min
5.0 g to 50.0 g
1.25 g to 25.0 g
2.50 g to 25.0 g
2.50 g to 25.0 g
0.80 g to 20.0 g
5.0 g to 25.0 g
Table 1 Overview of the semi-solid preparations, NRF = Neues Rezeptur Formularium (New German Formulary), DAC = Deutscher Arzneimittel-Codex (German Drug Codex), DAB = Deutsches Arzneibuch (German Pharmacopoeia), SR = Standardrezepturen (standardized preparations)
3
Salicylic acid and Erythromycin were purchased from Caesar & Loretz GmbH (Caelo), Hilden, Fagron GmbH & Co. KG, Barsbüttel and Euro OTC Pharma GmbH, Bönen. Suppliers for vaseline, white DAB were the given ones and, in addition, Bombastus Werke AG, Freital. Nonionic hydrophilic cream (Nichtionische hydrophile Creme SR DAC NRF S.26.) was purchased from Bombastus, Caelo, Euro OTC and Pharmazeutisches Kontroll- und Herstellungslabor GmbH (PKH Halle), Halle. Urea was delivered by Caelo and Fagron. Suppliers for the hydrophobic cream (Hydrophobe Basiscreme DAC) were Bombastus, Caelo, Fagron and Pharmachem Pößneck GmbH & Co. KG, Pößneck. Hydrophilic cream (Basiscreme DAC) was purchased from Bombastus, Caelo, Fagron, Paul W. Beyvers GmbH, Berlin and also Henry Lamotte OILS GmbH, Bremen. Metronidazol concentrate was provided by PKH Halle. Nonionic hydrophilic liniment (Nichtionisches hydrophiles Liniment SR DAC) was supplied from Bombastus, Caelo and PKH Halle. Provider of vaseline, yellow, hydrophilic ointment base (Hydrophile Salbe DAB) and wool wax alcohol ointment (Wollwachsalkoholsalbe DAB) was Caelo. 2.1.2 NIR spectrometer Spectra of the preparation salicylic acid 5 % (w/w) in vaseline DAB (NRF 11.43.) were recorded on a HiperScan apo-ident (Dresden, Germany). For the analysis of the remaining formulations the HiperScan finder was used, presenting an enhanced version of the apo-ident spectrometer with a heated (50 °C) and therefore temperature stabilized detector. All measurements were conducted at ambient temperature. 2.1.3 Software All spectra were acquired using the SL Predictor software (SensoLogic, Norderstedt, Germany) over a wavelength range of 1000 to 1900 nm with a total of 901 measuring points (2000 scans, resolution: 1 nm). Spectral treatments and calculations were carried out using the SL Calibration Wizard software (SensoLogic, Norderstedt, Germany). Further calculations, generation of additional graphics and data administration were done with SL Calibration Workshop and SL Utilities (SensoLogic, Norderstedt, Germany). 2.1.4 Sample presentation The samples were measured in transflectance mode, using glass beakers for sample presentation, centered on the measuring window of the spectrometer. A diffuse reflecting transflectance insert made of aluminum was put in the beaker holding the sample thus resulting in a defined layer thickness of the semi-solid preparation of 0.4 mm ± 0.05 mm. The empty beaker containing the aluminum transflectance insert was used as reference for every measurement. Six samples were taken from every single manufactured specimen, resulting in 270 spectra for salicylic acid in vaseline and in 90 spectra for all other described creams. To prevent any impact of stability issues, all analysis generally were carried out within one or two days after manufacturing of the specimen - at the maximum within one week. 2.1.5 Multivariate data analysis The spectra for each semi-solid preparation were divided into calibration and test set following the common ratio of 2:1 resulting in 180 spectra for the calibration set and 90 spectra for the test set for preparation 1 (salicylic acid in vaseline), and 60 calibration set spectra and 30 test set spectra for all other formulations (preparation 2 to 6), respectively. Partial least squares regression was carried out for each calibration set, testing different spectral treatments such as absorbance (Abs), detrend with a polynomial order of 0, 1 or 2 (Det), standard normal variate (SNV), normalization (Norm), first derivative (FD) and second derivative (SD) both with varying segments and gaps, in order to find the optimal fitted model concerning low standard errors of calibration (SEC) and prediction (SEP) and with acceptable correlation coefficients. 3. Results and discussion 4
Initially, we performed a feasibility study for salicylic acid in vaseline with a wider range of concentrations with contents of the active ingredient between 0 and 25 % (w/w). The micronized drug particles are suspended in the matrix. The application of data pretreatments such as the SNV followed by the second derivative (15/0) showed to be very effective as can be seen in Fig. 1. In the wavelength area between 1590 and 1690 nm a distinct concentration dependent variance of the signal intensity could be observed, which was promising for developing and validating a quantitative model for this preparation.
3.1 Validation of a quantitative method for salicylic acid 5 % (w/w) in vaseline Based on the EMA Guideline on the use of near infrared spectroscopy by the pharmaceutical industry and the data requirements for new submissions and variations [9] the following criteria of the quantitative NIR model were studied: Standard Error of Prediction (SEP), specificity, linearity, range, accuracy, precision and robustness. Table 2 gives an overview about the established models with the lowest errors of prediction. Wavelength range (nm) 1000 - 1900
Data pretreatments SNV
PLS latent variables 3
SEC (Cal Set) (% (w/w)) 0.079
5
SEP (Test Set) (% (w/w)) 0.084
Rprd
a1
a0
0.9965
1.01
-0.04
1.00 0.00 1000 - 1900 Norm 4 0.084 0.092 0.9958 1.01 -0.04 1000 - 1900 Abs + SNV 4 0.085 0.093 0.9958 1.00 0.01 1000 - 1900 SD(15/0) + SNV 4 0.095 0.1019 0.9948 Table 2 Comparative evaluation of different NIR models on the example of salicylic acid in vaseline: spectral range, spectral pretreatment, number of PLS factors, SEC for the calibration set, SEP for the test set, correlation coefficient of validation (test set) (Rprd) and corresponding slope (a1) and intercept (a0)
3.1.1 Standard Error of Prediction The models were tested on independent validation samples, prepared several weeks after the calibration samples using new batches of API and vaseline. The validation set consisted of five samples including API concentrations of 70, 90, 100, 110 and 130 % (rel.) of the target concentration which were measured in triplicate. The results are described in Table 3. The quantitative model based on the SNV transformation of the data showed the lowest SEP and the smallest number of PLS latent variables and was therefore chosen for further validation. Wavelength PLS latent SEP (Val Set) Rprd a1 a0 Data pretreatments range (nm) variables (% (w/w)) 0.9978 1.02 -0.12 1000 - 1900 SNV 3 0.071 0.9974 0.99 0.04 1000 - 1900 Norm 4 0.073 0.9979 1.02 -0.11 1000 - 1900 Abs + SNV 4 0.068 0.9967 1.02 -0.05 1000 - 1900 SD(15/0) + SNV 4 0.084 Table 3 Comparative evaluation of the generated NIR models on independent validation samples of salicylic acid in vaseline: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
3.1.2 Specificity Since identification of raw materials is the main application field of the applied NIR spectrometer apoident and finder in this project, appropriate qualitative spectral libraries already exist and didn´t need to be developed. Nevertheless, no qualitative identification test was carried out for the applied APIs ahead of quantification as identification of every raw material may be taken as given, being mandatory by law in German pharmacies. Based on that background the important issue of specificity was addressed by challenging the quantitative model with different out of specification (OOS) samples, such as samples containing no API, any other than the desired API or containing a wrong matrix as well as samples with API content exceeding the specification limits. Hence the predictions for samples containing no salicylic acid (placebo), salicylic acid in concentrations of 2.5 and 7.5 % (w/w) (i.e. 50 respectively 150 % (rel.) of the target concentration), as well as for samples containing other APIs and other matrices that might be mistaken in practice such as acetylic salicylic acid and urea instead of salicylic acid and hydrophilic ointment base (Hydrophile Salbe DAB), wool wax alcohol ointment (Wollwachsalkoholsalbe DAB) and yellow vaseline instead of white vaseline were studied. In agreement with common practice for quality control of individually prepared pharmaceutical preparations, we defined samples with predicted values below 90 % (rel.) and above 110 % (rel.) of the target concentration as identified OOS samples. The results for placebo samples show that, in spite of predictions not equal to zero but close to 1 % (w/w) API concentration, the placebo samples could easily be distinguished from samples containing the target concentration and would not falsely pass quality control tests. The predicted values for the API content were close to 4 % (w/w) for samples with acetylic salicylic acid due to the strong structural similarity and 0 % (w/w) for the samples containing urea. Both would also not have passed quality control tests. For the samples with exchanged matrices we observed that only 6
samples containing wool wax alcohol ointment were recognized as different with sufficient distinctness (predicted values around 4 % (w/w) for the API content). The tested hydrophilic ointment (containing more than 70 % (w/w) vaseline itself) and yellow vaseline were not recognized as “wrong” by the model – yellow vaseline being distinguishable visually for the human operator, of course. For both, predicted values between 5.26 and 5.37 % (w/w) API content were obtained. The predictions for samples holding 2.5 respectively 7.5 % (w/w) salicylic acid showed very accurate results close to the actual values, although the concentrations were not included in the range of the calibration samples. A summary of the predicted values for the OOS samples can be found in Table 4. Out of specification samples Placebo: vaseline, white (mean out of 9 samples)
Content of salicylic acid (pred.) ± SEM 1.07 % (w/w) ± 0.005 % (w/w)
False API: acetylic salicylic acid (mean out of 3 samples) False API: urea (mean out of 3 samples)
3.87 % (w/w) ± 0.004 % (w/w) 0 % (w/w)
False matrix: wool wax alcohol ointment (mean out of 3 samples)
4.1 % (w/w) ± 0.002 % (w/w)
False matrix: hydrophilic ointment base (mean out of 3 samples)
5.29 % (w/w) ± 0.001 % (w/w)
False matrix: vaseline, yellow (mean out of 3 samples)
5.32 % (w/w) ± 0.001 % (w/w)
API content 50 % (rel.): salicylic acid 2.5 % (mean out of 3 samples)
2.55 % (w/w) ± 0.001 % (w/w)
API content 150 % (rel.): salicylic acid 7.5 % (mean out of 3 samples) 7.39 % (w/w) ± 0.0003 % (w/w) Table 4 Prediction results (mean of multiple measurements) for OOS samples for salicylic acid in vaseline
Specificity of the method can also be proved by looking at the loading plots for the first latent variables shown in Fig. 2. A large contribution of the band around 1659 nm can be observed both for latent variable 1 and 2, perfectly matching the spectral properties of salicylic acid, shown in Fig. 3. Other bands contributing to the model can be assigned to vaseline.
7
3.1.3. Range and linearity The equally distributed residues for the validation set illustrated in Fig. 4 suggest absence of systematic trends. Assuming linearity of the model, the correlation coefficient can be determined for a fitted straight line in the predicted versus actual values plot (Fig. 5.). For this model, it is very close to 1 (0.9978) meaning, that there is good accordance between actual and predicted values for the API content. No remarkable difference between residues in the lower and higher concentration range was observed which strengthens the linearity assumption. Moreover the RPD value was calculated. The result of 11.99 was found to be indicating the excellent predictive quality of the applied model. [10]
8
3.1.4 Accuracy The target values for the Standard Error of Prediction were set as ≤ 0.25 % (w/w), corresponding to a variance of ± 5 % (rel.) around the target API concentration. This could be achieved and even under-run with all models mentioned above. All predictions were found to be within ± 5 % (rel.) deviation from the actual value with an averaged deviation of 1.2 % (rel.). A paired samples t-test was also performed to check whether the predicted and the actual values varied significantly. The result showed that the difference between the actual and the predicted values is not significantly different from 0 (α = 0.05, pvalue = 0.24). 3.1.5 Precision The precision of the method was evaluated by determining repeatability and intermediate precision. To evaluate repeatability six samples were taken out of a 5 % (w/w) salicylic acid vaseline one after another and measured separately on the same day. In order to estimate intermediate precision additionally six spectra were recorded for six samples taken out of the same preparation on six consecutive workdays. For both measuring series the coefficient of variation for the predicted values was ≤ 1.5 % (rel.) (1.27 % (rel.) and 1.36 % (rel.) respectively). 9
3.1.6 Robustness The robustness of the quantitative model was evaluated through tests at different sample positions, different temperatures and tests with preparations with different API particle sizes (micronized vs. powdered salicylic acid). For the evaluation of the influence of the sample position, one sample taken out of a 5 % (w/w) salicylic acid vaseline was measured six times in the same beaker, the glass being rotated piecewise prior to the next spectrum recording. Additionally, another sample was measured six times in series without moving the beaker glass. For both measuring series the coefficient of variation for the predicted values was ≤ 1 % (rel.) (0.52 % (rel.) and 0.79 % (rel.) respectively). The result of the paired t-test showed that the difference between the measuring series is not significantly different from 0 (α = 0.05, p-value = 0.18), indicating that rotating the sample had no significant influence. To test the influence of the particle size of salicylic acid, two specimens that contained 5 % (w/w) of powdered but not micronized API were prepared and analyzed in triplicate. The variance in particle size led to predicted values scattering around the target value to a bigger extent than the predictions for the validation samples with micronized salicylic acid. A variation coefficient of 3.1 % (rel.) was obtained. But still, with the exception of one value, all were found in the range of ± 5 % (rel.) around the target value. The t-test showed that there is no significant difference between the means of the values for the samples with micronized and with powdered salicylic acid (α = 0.05, p-value = 0.12), indicating that the particle size had no significant influence. Temperature is known to influence NIR measurements. Therefore, the validation sample set was tested at different temperatures. Ahead of recording the spectra, samples had been stored at the following temperatures: room temperature 20 – 25 °C, cool storage 15°C ± 3°C and cold storage 5°C ± 3°C for at least 24 hours. As a side effect of the decrease in temperature a firmer consistency of the preparations could be noticed. The measurements were conducted in triplicate and carried out immediately after removing the samples out of the fridge. As expected, increasing values for bias and RMSEP were observed for both sample series measured at other temperatures than room temperature that were not included in the calibration: room temperature RMSEP 0.074 % (w/w), bias 0.011 % (w/w); cool storage RMSEP 0.075 % (w/w), bias -0.034 % (w/w); cold storage RMSEP 0.096 % (w/w), bias -0.043 % (w/w). Nevertheless, the prediction errors were still very small. Application of the paired samples t-test showed only a significant difference comparing the values for cool samples and samples at room temperature (α = 0.05, p-value = 0.01), not between the other groups (room temperature and cold storage, α = 0.05, p = 0.054 and cool and cold storage, α = 0.05, p = 0.71). Furthermore, the influence of the state of the spectrometer was evaluated, since the validation was performed on the apo-ident device, which is not equipped with a heated detector. Hence, the time after starting the device (permanent running or just started) might possibly influence the signal intensity. After 24 hours in the turned off status the spectrometer was switched on again and immediately all five validation specimens were measured once more in triplicate. The result of the paired samples t-test showed that the means of the measuring series are significantly different from each other (α = 0.05, pvalue = 0.02). 3.2 Transfer of the analytical approach on other preparations Following the successful validation of the method, the approach was tested on different pharmaceutical preparations containing other frequently used APIs and more complex matrices especially with high percentages of water, which is known to disturb NIR measurements. To evaluate the possibility of reducing the amount of calibration specimens to be prepared in order to save costs and effort prior to new investigations, further quantitative models were developed for salicylic acid in vaseline using only one third of the originally generated spectra (60 spectra for the calibration set and 30 spectra for the test set). 10
The results for the prediction errors for the validation set were compared with those of the developed models based on the entire quantity of the spectra. As can be seen in Table 5 the standard errors of prediction revealed to be very small. Therefore, the following investigations for preparation 2 to 6 were conducted with 90 spectra from 15 manufactured creams in each case and six samples taken from every specimen. SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 0.9975 1.03 -0.11 1000 - 1900 SNV 3 0.076 0.9965 0.98 0.11 1000 - 1900 Norm 4 0.088 0.9974 1.02 -0.14 1000 - 1900 Abs + SNV 4 0.075 0.9964 0.98 0.11 1000 - 1900 SD(15/0) + SNV 4 0.089 Table 5 NIR models for salicylic acid in vaseline: spectral range, spectral pretreatment, number of PLS factors, correlation coefficient of validation, SEP for the validation set (Rprd) and corresponding slope (a1) and intercept (a0)
3.2.1 Development of a quantitative method for salicylic acid 5 % (w/w) in a hydrophilic cream Sticking to the API salicylic acid, first of all we changed the lipophilic anhydrous semi-solid matrix to a hydrophilic system with high water content of 64 % (w/w). As in case of vaseline the API exists in form of suspended particles in the hydrophilic cream as well. Comparing the spectra of salicylic acid in vaseline and in the hydrophilic cream, Fig. 6 shows the considerable contribution of the high water amount to the appearance of the spectra. Still, the influence of the API concentration can be visually detected around 1660 nm.
SEP (Val Set) Rprd a1 a0 Data PLS latent pretreatments variables (% (w/w)) 1000 – 1900 SD(15/0) + SNV 3 0.116 0.9979 1.03 0.11 1050 – 1850 SNV + SD(10/0) 3 0.127 0.9974 1,03 -0,14 1050 – 1330, 1500 – 1850 SD(15/0) + SNV 5 0,141 0.9979 1.06 -0.33 1050 – 1850 FD(2/15) + SNV 6 0.113 0.9978 1.01 -0.19 Table 6 NIR models for salicylic acid in a hydrophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0) Wavelength range (nm)
Quantitative models were developed using either the whole spectrum or omitting the area of the band characteristic for the –OH vibrations of water (in this case about 1350 – 1500 nm). Comparable results were received for the prediction errors. Table 6 shows the best models and the errors of prediction for the independent validation set, consisting out of five samples, measured in triplicate. The validation samples included API concentrations of 50, 95, 100, 105 and 150 % (rel.) of the target concentration to simulate OOS samples and samples that should easily pass quality control.
The results showed that the content of the API could be predicted correctly within 95 and 105 % (rel.) of the reference value with all models. Deviations greater than 5 % (rel.) were only observed for the prediction of the samples with 50 % (rel.) of the target concentration (2.5 % (w/w) salicylic acid in this case). The plot of predicted versus actual values for the method using second derivative and standard normal variate corrected spectra and three latent variables is shown in Fig. 7. 3.2.2 Development of quantitative methods for urea 10 % (w/w) in a hydrophilic cream and for urea 10 % (w/w) in a hydrophobic cream As salicylic acid, urea is used for dermatological purposes in a wide variety of matrices in concentrations from 3 % (w/w) as moisturizer or up to 40 % (w/w) as a strong keratolytic agent. For our studies, we selected a medium concentration of 10 % (w/w) as target concentration in a hydrophilic as well as a lipophilic cream. Urea is soluble in both of them as the water content amounts to 40 % (w/w) and 65 % (w/w), respectively. The best models are shown in Table 7 for the hydrophilic cream and in Table 8 for the hydrophobic cream.
12
SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1050 – 1850 FD(2/15) + SNV 4 0.105 0.9995 1.00 -0.05 1000 – 1900 SNV 6 0.145 0.9991 0.99 0.06 1050 – 1850 SNV 5 0.132 0.9992 1.00 0.10 1050 – 1850 Det 2 + SNV 4 0.171 0.9987 1.01 -0.14 Table 7 NIR models for urea in a hydrophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0) SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1050 – 1850 SNV 6 0.202 0.9985 1.03 -0.24 1050 – 1850 Det 1 + SNV 5 0.223 0.9981 1.03 -0.16 1000 – 1900 Det 2 + SNV 5 0.265 0.9971 1.03 -0.19 1000 – 1900 Abs + SD(15/0) + SNV 4 0.281 0.9970 0.96 0.35 Table 8 NIR models for urea in a lipophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
The results suggested that all models for hydrophilic and lipophilic cream can predict the API content correctly within 95 and 105 % (rel.) of the reference value for all validation samples. Remarkably, superior quality attributes such as lower predictive errors could be observed for the hydrophilic cream, although both matrices represent cream bases with comparable properties and number of components. This is mainly attributed to the 1.5 times higher water content in the lipophilic cream, which interferes with the major absorption band of urea in the area between 1450 and 1550 nm (Fig. 8, 9).
13
3.2.1 Development of a quantitative method for metronidazole 2 % (w/w) in a hydrophilic liniment The determination of metronidazole in high concentrations (target concentration 10 % (w/w)) in a hydrophilic cream for vaginal application has been described by Baratieri et al [11]. As however metronidazole is used in dermatological preparations with a maximum content of 2 % (w/w), this concentration was chosen as target concentration for our study. The API was suspended in form of metronidazole concentrate into a hydrophilic semi-solid matrix (nonionic hydrophilic liniment) with a very high water content of 82 % (w/w). Lipophilic preparations were not taken into account in this study as they are not of therapeutic interest. The dimension of the prediction errors for all developed models are summarized in Table 11 and the actual versus predicted plot for the first model in Fig. 10 indicated that the determination of metronidazole was not possible with comparable accuracy as for salicylic acid and urea, which were tested in higher concentrations. Nevertheless, most of the predicted values could be found within ± 10 % (rel.) of the reference value for the validation samples with contents of 95, 100, 105 % (rel.) API. The prediction for the validation samples with 50 and 150 % (rel.) metronidazole content showed deviations around ± 20 % (rel.), but would have been detected correctly as OOS samples during quality control. SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1050 – 1850 Norm 7 0.174 0.9788 1.21 -0.35 1050 – 1350; SNV 5 0.178 0.9805 1.23 -0.40 1550 – 1850 1050 – 1300; SD(15/0) + SNV 4 0.196 0.9701 1.22 -0.41 1500 – 1850 1050 – 1850 SD(15/0) + SNV 4 0.195 0.9601 1.12 -0.19 Table 11 NIR models for metronidazole in a hydrophilic liniment: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
14
3.2.1 Development of a quantitative method for erythromycin 4 % (w/w) in a hydrophilic cream Erythromycin represents a mixture of three substances, erythromycin A, B and C and its solubility in a hydrophilic cream base depends on its concentration. Small amounts up to 2 % (w/w) will partly be dissolved while higher percentages will be found as suspended particles in the matrix. As lipophilic preparations with erythromycin are not used in practice, they were not included in the study. The development of quantitative models for this antibiotic drug led to unsatisfactory results. An overview is given in Table 12. The prediction of the API content was only possible with prevailing deviations of more than 10 % (rel.). However, the predictive quality rises with higher concentrations of the drug, as can be seen from Fig. 11. Hence, validation samples containing 6 % (w/w) API met the acceptance criteria of ± 5 % (rel.). SEP (Val Set) Rprd a1 a0 Wavelength PLS latent Data pretreatments range (nm) variables (% (w/w)) 1000 – 1900 SNV 8 0,333 0,9677 1,02 -0,44 1050 – 1850 SNV 7 0,362 0,9617 0,97 -0,22 1000 – 1900 Norm 8 0,331 0,9679 0,99 -0,33 1050 – 1850 Det1 + SNV 7 0,378 0,9588 0,95 -0,18 Table 12 NIR models for erythromycin in a hydrophilic cream: spectral range, spectral pretreatment, number of PLS factors, SEP for the validation set, correlation coefficient of validation (Rprd) and corresponding slope (a1) and intercept (a0)
15
5. Conclusions All in all our results show that the application of NIRS for quantitative quality control in individually prepared semi-solid pharmaceutical formulations is possible in general, but accuracy and variance strongly depend on the active ingredient and its concentration range. In all cases an increase in the water content of the formulation was associated with a loss in accuracy. The low dosed preparation with a content of 2 % (w/w) metronidazole and the one containing 4 % (w/w) erythromycin being composed of several subgroups and showing varying solubility depending on the concentration present difficulties in developing quantitative models with sufficient accuracy. But for the active ingredients salicylic acid and urea, successful quantification meeting the limits of a maximum deviation of ± 5 % (rel.) from the actual value could be achieved. For these preparations near-infrared spectroscopy even with limited wavelength range spectrometers could present a promising alternative to conventional quality control methods. Acknowledgements The authors thank HiperScan GmbH, Dresden, Germany for providing the NIR spectrometer and calibration software. References [1] M. Blanco, J. Coello, H. Iturriaga, S. Maspoch, C. de la Pezuela, Near-infrared spectroscopy in the pharmaceutical industry, Anal. 123 (1998) 135R–150R. [2] G. Reich, Near-infrared spectroscopy and imaging: Basic principles and pharmaceutical applications, Adv. Drug Deliv. Rev. 57 (2005) 1109–1143. [3] M. Jamrógiewicz, Application of the near-infrared spectroscopy in the pharmaceutical technology, J. Pharm. Biomed. Anal. 66 (2012) 1–10. [4] Y. Roggo, P. Chalus, L. Maurer, C. Lema-Martinez, A. Edmond, N. Jent, A review of near infrared spectroscopy and chemometrics in pharmaceutical technologies, J. Pharm. Biomed. Anal. 44 (2007) 683–700.
16
[5] M.A.M. Silva, M.H. Ferreira, J.W.B. Braga, M.M. Sena, Development and analytical validation of a multivariate calibration method for determination of amoxicillin in suspension formulations by near infrared spectroscopy, Talanta 89 (2012) 342– 351. [6] L.M.M. Lê, E. Caudron, A. Baillet-Guffroy, L. Eveleigh, Non-invasive quantification of 5 fluorouracil and gemcitabine in aqueous matrix by direct measurement through glass vials using near-infrared spectroscopy, Talanta 119 (2014) 361–366. [7] R. Neubert, B. Colin, S. Wartewig, Direct Determination of Drug Content in Semisolid Formulations Using Step-Scan FT-IR Photoacoustic Spectroscopy, Pharm. Res. 14 (1997) 946–948. [8] H. Latsch, Umfrage zur Defekturherstellung, Dtsch. Apoth. Ztg. 153 (2013) 3693. [9] European Medicines Agency: Guideline on the use of near infrared spectroscopy by the pharmaceutical industry and the data requirements for new submissions and variations. http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2014/06/WC500167967.pdf, 2014 (accessed 21.07.2016). [10] P.C. Williams, D.C. Sobering, Comparison of commercial near infrared transmittance and reflectance instruments for analysis of whole grain and seeds, J. Near Infrared Spectrosc. 1 (1993) 25–32. [11] S.C. Baratieri, J.M. Barbosa, M.P. Freitas, J.A. Martins, J. Pharm. Biomed. Anal. 40 (2006) 51–55.
17