Powder blend uniformity by NIR

Ronald Rubinovitz, of Buechi Analytical Inc, highlights the benefits of using Near Infra Red (NIR) to monitor the blending process in drug production, leading to improvements in process efficiency

In the manufacture of pharmaceutical solid dosage forms such as tablets and capsules, homogeneity in blending is necessary to achieve high quality products of uniform mixture. Currently, assessment of successful blending is carried out by means of tests performed on multiple samples removed from the blender and analysed by HPLC or UV spectrometry.

The limitations of these tests lie in their inherent analysis time as well as the necessity of removing samples from the blender.

The advantages of near-infrared (NIR) spectroscopic uniformity analysis are its:

•speed (which can be less than one minute per measurement)

•capability to measure samples at a single point or at multiple locations either inside or outside the blender

•effectiveness using a variety of chemometric techniques^{1-4(and references cited therein)}.

Thus, NIR offers the manufacturer a faster, and therefore more effective way to monitor blending by signalling the end of the mixing process more quickly. Therefore NIR has the capability to improve productivity and eliminate over-blending. The purpose of this study is to illustrate how NIR can be used to provide such information.

NIR reflectance spectra of the pure components of the blend under investigation in this report (an active material and lactose) are shown in Figure 1.

As is typical for actives and excipients used in the pharmaceutical industry, one can clearly see that each of these materials exhibits distinctive absorptions. In particular, strong characteristic absorptions for the active can be seen in the 4800 and 5800 cm^-1 spectral regions. At time intervals throughout the blending process, NIR spectra were acquired.

Selected second derivative pre-treated spectra from the blending are shown in Figure 2, where it can be seen that, although the spectra vary during the blending, these changes occur in regions associated with each of the raw materials. Thus, NIR offers the capability of simultaneously monitoring multiple components in the blend.

predicting concentration

One of the methods used to monitor blending is to utilise the quantitative capabilities of NIR to predict the concentration of the components. For this study a set of 20 standards, with concentration ranging from 4-92% (wt%) active were gravimetrically prepared and measured by NIR to develop a quantitative model. The prediction of the calibration and validation sets (N=14 and 6, respectively) is plotted in figure 3, showing excellent uniformity and performance.

The resulting two factor Partial Least Squares (PLS) equation was then applied to the spectra acquired during the blending process. Results are plotted as a function of time in Figure 4. One can readily see that although the concentration value of the active is changing rapidly at the beginning of the blending process, by approximately the twelfth measurement the concentration of the active has stabilised at the target value of 15%.

quantitative model

Although, as illustrated in Figure 4, a quantitative model may be developed to determine blending end point, 'qualitative' models offer the advantage of avoiding the initial calibration period consisting of the acquisition and measurement of standards followed by model development.

One such non-quantitative method is the evaluation of the standard deviation of the acquired spectra as the blending proceeds.

Although optimal results have been obtained from sampling multiple points in the blender and comparing their variation4, an alternate technique used here is the collection of data at a single sampling point and calculating the variation between successive measurements. With either method, the entire spectrum or only those regions of corresponding to component(s) of interest can be analysed. As the blending proceeds, the variation of the spectra should decrease to a minimum as the sample reaches the blending end point.

In this study, the spectral standard deviation was calculated as follows. After spectral smoothing and second derivative mathematical pretreatment, the standard deviation at each wavenumber data point between 4600-6300 cm^-1 (142 data points) was calculated for each spectrum and its preceding three spectra. Then the average standard deviation value across the 4600-6300 cm^-1 spectral range was calculated for each spectrum.

Results are shown in Figure 4, where it can be seen that although the spectral standard deviation magnitude is initially high, as the blending nears completion the spectral standard deviation begins to fall to a steady level. It is interesting to note that in the blending profile shown in Figure 4, both the quantitative and standard deviation techniques predict blend end point by about the twelfth measurement. Objective methods to define criteria for end-point determination, using standard deviation, have included simply observing that a minimum value has been reached, as well as more involved techniques where the observations of an ongoing blend are statistically evaluated in comparison with past fully blended mixtures.

cluster method

Another technique that has been used in the evaluation of blending progress is the use of principal component analysis (PCA). Although several PCA methods are possible, in this study, results were analysed automatically through a cluster method available in NIRCal software. Using this technique, principal component 'factors' are generated by the software for the spectra representing the fully blended mix, with each spectrum having its own unique set of loading coefficients.

The entire spectral data set can then be represented by using just the first two PCA factors, and although the loadings of spectra collected early on in the mixing tend to be scattered throughout the plot, as the blending reaches end point all spectra begin to converge in the same location of the two-dimensional loadings space. As expected, all the spectra collected after the twelfth measurement fall within this cluster, confirming the results of both the quantitative and standard deviation analyses discussed earlier in this report.

It has been demonstrated that the important parameter of blend end point can be monitored by near-infrared reflectance spectroscopy using a variety of chemometric techniques. The methods surveyed in this study all offer the advantage of 'real-time' monitoring of degree of blending completion. In particular, analysis of spectral standard deviation offers the benefit of not requiring a formal calibration process.

Regardless, of the specific technique, it has been shown that it is possible to achieve significant improvement in blending process efficiency by taking advantage of the speed and sensitivity of NIR.