Near Infrared Spectroscopic Investigations of Solid Pharmaceutical Formulations
Contributors
S. Borchert, J.-O. Henck and H. W. Siesler. Germany, 2003
Abstract
In this study near-infrared (NIR) transmission and diffuse reflection spectra of tablets with two different pharmaceutical active substances in variable concentrations were recorded. The purpose of these tests was to correlate the spectra with the content of active ingredient, hardness and moisture content in order to implement a fast, non-destructive quality control of solid drug formulations in an industrial environment.
Introduction
The aim of the present study was to develop qualitative PCA and quantitative PLS calibration models which permit high correlation between the spectra and the content of active ingredient, moisture and hardness of the investigated solid drug formulations. Materials and Methods
Two different types of tablets with varying amount of active ingredient A and B, respectively, were investigated. The NIR spectra were recorded in diffuse reflection and transmission on a Bruker Vector 22N FTNIR spectrometer (Bruker Optik GmbH, Ettlingen, Germany). The spectrometer is equipped with a sample carousel for the automated measurement of 10 tablets, an InGaAs detector for transmission measurements and a PbS detector, which is integrated in an integrating sphere, for the reflection measurements. For spectra acquisition 32 scans with a spectral resolution of 8 cm-1 were accumulated. The spectra were evaluated by The Unscrambler®.
The content of the active ingredient in the A tablets was 0.1, 0.2, 0.3 and 0.8 mg. The B tablets contained 2.5, 5, 10 and 20 mg of active ingredient. The reference method for the active ingredient content of both types of tablets was the highpessure liquid chromatography (HPLC); the tablet hardness was measured by a tablet hardness tester and the moisture content was determined by Karl Fischer (KF) titration.
Qualitative Tests
One issue of this study was to identify transmission or diffuse reflection spectroscopy as the preferential method to predict qualitative and quantitative parameters of the investigated tablets. In the first step it had to be determined whether it makes a difference when the NIR beam impinges first on the front or on the back side of the coated A-tablets in the two measurement configurations and whether the corresponding NIR transmission and diffuse-reflection spectra can be discriminated by PCA. Depending on the content of active ingredient the A-tablets had a stamp on the front side and a smooth reverse side (0.1 – 0.3 mg) or different stamps on both sides (0.8 mg).
The score plot is the most important result of a PCA. It shows if a separation of the spectral variables is possible or not. A three-dimensional score plot was derived from the PCA model of the recorded reflection spectra in terms of the tablet front/reverse side. Different colours demonstrated the different tablet sides.It was thus derived that the spectra of the tablets recorded from different sides in diffuse reflection cannot be discriminated. The same result was obtained for the transmission spectra. Furthermore, it was examined if a separation of the spectra measured in transmission or reflection mode of coated tablets was possible in terms of the active ingredient content.
Each cluster corresponds to spectra of tablets with a certain content of active ingredient. A similar result was obtained for the diffuse-reflection spectra suggesting that the difference in active ingredient is significant enough to discriminate the tablets also in the diffuse-reflection mode. However, it should be pointed out, that the tablet coating contained varying amounts of Fe2O3 depending on the content of active ingredient. In order to verify the influence of the coating further investigations were performed with the spectra of uncoated A-tablets recorded in the transmission and reflection modes. The spectra were subjected to a PCA to see if a separation of uncoated tablets with reference to the content of active ingredient was possible. The result of this test was, that it was possible to separate only the transmission spectra according to the content of active ingredient whereas the PCA was not able to separate the spectra recorded in the diffuse-reflection mode. In view of the fact that in the reflection mode the NIR-beam does not penetrate into the core of the tablet the separation obtained in the score plot of the diffuse-reflection spectra was obviously caused by the different coating for the different assays. From these introductory qualitative investigations of the A-tablets is has to be assumed that only transmission measurements provide the necessary reliability for the NIR spectroscopic determination of the active ingredient. To further support this conclusion PLS models were developed with the specification values (0.1, 0.2, 0.3 and 0.8mg) of the A-tablets. While the predicted values for transmission spectra were in good agreement with the actual values, in the diffuse-reflection mode only the 0.8 mg tablets yielded acceptable prediction accuracy. Thus, the minimum limit to predict the concentration of the active ingredient in the reflection mode is reached by the 0.8 mg tablets.
Quantitative Tests
Based on this preliminary work, further quantitative evaluations where only based on transmission spectra of the two different drug formulations. For the A-tablets, PLS-1 models were developed with the HPLC and hardness values, respectively, of 120 calibration samples. In analogy, for the B-tablets PLS-1 models were developed for the active ingredient, hardness and water content (with Karl Fischer reference values) (30 tablets per concentration for the active ingredient content, 40 tablets for the hardness and 20 tablets for the moisture content). The Tables 1a and 1b summarize the best results of these PLS models for the different parameters of the two drug formulations.
The most meaningful results of the PLS are the “Actual vs. Predicted-Plots” and the “R2/RMSEP vs. Factor-Plots”. The “Actual vs. Predicted-Plot” permits the determination of the R2-value which indicates the calibration model quality to predict the actual values. Figure 6 shows an Actual vs. Predicted-Plot for four different concentrations of the active ingredient A tablets for the PLS model with vector normalization as spectral pretreatment. The R2/RMSEP vs. factor plot shows the contribution of the separate factors. The R2- value represents the correlation coefficient and the RMSEP means Root Mean Square Error of Prediction. Each factor explains its contribution of information to the model.The models were improved by outlier detection (F-test) and/or variation of the wave number range. The spectral range was chosen to allow for the inclusion of maximum information for the model. The improved spectra were tested versus new samples which were not included in the calibration model. In the prediction of concentration for the active substance A-tablets the deviation between the actual value and the prediction value must not exceed 15% (for the 0.1 – 0.8 mg active ingredient content). For the active substance B tablets the deviation for the concentration and moisture content may be 5% (for the 2.5 – 20.0 mg active ingredient content). The hardness deviation varied within 10 N (for hardness values between 110 – 170 N).
The differences between the actual values and the predicted values of the 0.1 mg active ingredient content of the active substance A tablets are unacceptable. A quantitative prediction of the active ingredient content for active substance A-tablets is only possible for concentrations > 0.2 mg. The models for the determination of the hardness were improved by the use of tablets which had the same size (0.1 mg – 0.3 mg). The prediction of the concentration of the active substance B tablets yielded numerous outliers in the 2.5 mg and 5 mg range. A new model was generated which only included tablets of two concentration types (2.5 + 5 mg, 5 + 10 mg, 10 + 20 mg). The results of these models were better than the models which contain all concentrations (2.5 – 20 mg). The calibration should contain tablets with consistent sizes and different concentrations of active ingredient. The results of the hardness of the active substance B-tablets were acceptable. Presumably the model could be improved by the use of the same tablet size in analogy to the calibration of the concentration. The modeling of the moisture content yielded acceptable results but the deviation between the actual value and the predicted values of the 2.5 and 10 mg tablets was very high. The results derived from models of reflection spectra were not acceptable. The coating of the tablets has a large influence on the spectra.
