Detection of Lumps in Powder Blends by Inline NIR

This study shows that the presence of API lumps can be detected by inline NIR, and elaborates on why NIR sensor dimensions and actual measured sample volume by the NIR sensor are important variables for adequate interpretation of obtained results.
Sep 02, 2017
Volume 41, Issue 9, pg 36-44

Image is courtesy of InProcess-LSP.Submitted: March 22, 2017
Accepted: May 15, 2017

Within the pharmaceutical industry, inline near infrared (NIR) spectroscopy is one of the most commonly applied process analytical technologies to monitor content uniformity of powder blends (1). Inline NIR provides direct insight into blending operations with respect to the degree of uniformity of the ingredient of interest either for process development or control during commercial manufacturing. A quantitative calibration for a specific ingredient may be applied to predict the content of small fractions of the powder blend during the blend process, such as for every revolution in the case of a rotary blender. The actual size and sampled volume of the powder blend is an important variable for interpretation of NIR results.

For formulations in which the API content is a few percent or less, uniformity of the API becomes more critical, as the impact of non-uniformity increases dramatically with respect to efficacy and safety of the end-product. While it is known that inline NIR can provide content uniformity information during the blending process, no direct link to the cause of observed non-uniformity may be revealed. A severe cause of non-uniformity is related to presence of lumps in input materials (2, 3); these lumps may be formed during processing or storage of the API. Environmental conditions during storage and/or handling may contribute to the formation of these lumps. Another possibility is that lumps or agglomerates may be formed during the blending process. In all cases, prevention or effective breakdown of lumps is required to assure sufficient uniformity of the powder blend and the corresponding end-product.

For a specific micronized potent API, formulated at 0.7%, significant differences in content uniformity were observed by manual blend sampling and offline analysis. It was decided that inline NIR should be used to investigate content uniformity during the mixing process for different API batches. Significant differences were observed in inline NIR data between API batches, which were related to the presence of API lumps. It was revealed that lumps can be detected by inline NIR. For adequate interpretation of the presence of lumps in inline NIR data, typical sub-sampling effects of NIR sensors must be considered, which will be discussed in this article. Understanding these sub-sampling effects provides a powerful methodology to detect and further study the formation as well as the breakdown of lumps with inline NIR.

Materials and methods

Mixing process on rotary blender at 10-L scale. Mixing experiments of two API batches (A: good uniformity and B: poor uniformity) with one bulk polymer-based excipient were conducted in a 10-L scale Bohle rotary blender equipped with a SentroPAT BU TL inline NIR analyzer (Sentronic). At every revolution (6 rpm) of the blender, a NIR spectrum of the blend was automatically recorded and processed by a calibration model to predict API content. In addition to API content, uniformity of the blend was estimated in real time by calculating the relative standard deviation (RSD) of the predicted API values of the last 10 revolutions. 

Inline NIR method. To monitor blend behavior of the API of interest, a calibration model was created between 60 and 140% of the target concentration of the API. The blend consisted of 0.7% API and 99.3% of a polymer-based excipient. The API shows very specific spectral features at NIR wavelength areas for which the polymer-based excipient lacks any relevant absorbance. This result allowed a straight-forward univariate calibration model for quantification of the API in the blend. In Figure 1, the specific NIR wavelength range is shown together with the univariate calibration model.

Figure 1: Standard normal variate (SNV) normalized and 2nd derivative (Savitsky-Golay) near infrared (NIR) spectra of 0, 60, 80, 100, 120, and 140% of the API target concentration (0.7%). On the right-hand side, the API specific wavelength area is shown; intensities at 1638 nm were used to create a univariate linear calibration model (R2 = 0.993) for API quantification. [All figures are courtesy of the authors.]

Results

Comparison of API batches. Figure 2 shows the inline NIR blend curves of API for batch A and batch B. The blue dots in the blend curves represent the predicted relative API content, and the solid line represents the “moving” RSD values bases on 10 preceding revolutions. For batch A, the API content shows significant variation up to revolution 50; at this time point, the moving RSD value drops just below 5%. Between revolution 50 and 100, somewhat more variation in API content is observed compared to API content variation after revolution 100. At revolution 230 and 270, a high content spike can be observed. The API uniformity of the final blend is estimated to be approximately 2.5% based on the moving RSD values. For batch B, a significantly different pattern is observed. The overall variation of the predicted API content is larger compared to batch A, the bulk (average) API content is significantly lower than the expected 100%, and many positive content spikes are present throughout the blend curve. Overall it can be concluded that API batch B results in a significantly less uniform blend compared to batch A.

Figure 2: Inline near infrared (NIR) blend curves. The blue dots represent the predicted API content by NIR for each revolution of the rotary blender. The solid line represents the “moving” relative standard deviation (RSD) of the blend as an indication of blend uniformity. Average relative bulk content of 70% around revolution 202 (); high relative content spike of 140% at revolution 202 ().

The observed positive content spikes for batch B must be related to high API content measurements by NIR. A slight increase in measured API bulk content is observed during the mixing process up to revolution 300, resulting in a fairly constant content level of approximately 80% of the target content at the end of the process. The high content spikes may be related to highly concentrated domains of API such as lumps or agglomerates; every time a lump or agglomerate is within the sampling volume of the NIR sensor, a deviating high content would be expected. 

The highest API content spike reaches 140% of the target API concentration (Figure 2, batch B) at revolution 202. The sampling volume of the inline spectrometer for this blend material was determined at 250 mg, which represents the amount of sample that contributes to the NIR spectrum (more details and considerations related to sampling volume by NIR are described in the next section). If it is assumed that the sampling volume at revolution 202 contains approximately 70% API (the average level of API measured around revolution 202) and one API lump, the API lump must contribute an additional 70% of API (high content spike is 140%). In 250 mg, 100% of the target concentration corresponds to 0.7% of 250 mg, which equals 1.75 mg. The size of the expected API lump should then correspond to approximately 70% of 1.75 mg ≈ 1.2 mg. Assuming a spherical lump and a density of approximately 0.6 kg/L, the diameter of the largest observed API lump is estimated to be approximately 1.5 mm. Within the NIR blend curve, no spikes larger than 140% were detected; therefore, no lumps larger than 1.5 mm would be expected in the blend. An additional assumption for this estimation is that not more than one lump was present in the sampling volume corresponding to 140%.

Sieving of the blends. To investigate actual presence of lumps, the powder blends of both batches A and B were sieved through a 3.35-mm sieve after the mixing process. In Figure 3, the sieve fractions of the powder blends are shown. In contrast to previous calculations and assumptions, both batches contain API lumps larger than 3.35 mm, while the maximum lump size was estimated for batch B at 1.5 mm. Several lumps were isolated and identified by offline Fourier transform near infrared (FT-NIR) and NIR imaging; both analyses confirmed that the lumps obtained in the sieve fraction correspond to pure API material. Also, batch A shows some API lumps, although significantly fewer compared to batch B. The maximum lump size within batch B is approximately 10 mm diameter; the corresponding weight of this lump was 314 mg, which is approximately 250 times larger than the expected maximum lump weight of 1.2 mg. 

Figure 3: Obtained sieve fractions of powder blends with API A and B (sieve size = 3.35 mm).

Apparently, the measured NIR signal is not in agreement with the expected variation in content caused by lumps, but underestimates the size of present lumps significantly. The origin of the above discrepancy is attributed to the sampling details (NIR ‘sub-sampling’ which is discussed in the following section) of measuring inline content with a NIR sensor.

Subsampling by inline NIR sensors. Spectroscopic sensors such as NIR probes suffer from subsampling, which is described as follows. If powders are measured in reflectance, a specific amount of the sample is contributing to the NIR spectrum. This amount of sample corresponds to the NIR sample size. Because uniformity assessment is directly related to the sample size, it is mandatory to have a good estimate of the actual NIR sample size that is sub-sampled with a NIR probe (4, 5). The NIR sample size is dependent on material properties of the blend, such as particle size, density, morphology, NIR absorptivity, wavelength, and sensor dimensions. The NIR sample size may be defined as the total amount of sample that contributes to the NIR spectrum (6). By measurement of increasing amounts of sample material, the NIR sample size corresponds to the amount of sample at which the NIR spectrum becomes constant in intensity. The actual interaction of the incident NIR radiation and resulting reflectance signal complicates the definition of the actual effective sample size because parts of the sample close to the NIR interface contribute more to the NIR spectrum compared to parts of the sample positioned further away from the NIR interface. In fact, an intensity (or contribution) gradient exists, which is directly related to the properties of the measured material. This gradient can be determined by measurement of defined amounts of sample and measurement of the corresponding reflected intensity at the wavelength of interest (in this case 1638 nm). Once this gradient is determined, the contribution at different penetration depths of the sample within the NIR sample size can be obtained (7). As significant parts of the NIR sample size contribute only at a minor level to the NIR spectrum, the effective sample size is consequently smaller than the NIR sample size. Particularly when the API is not distributed evenly throughout the NIR sample size, the impact of the NIR intensity gradient should be taken into account.

Figure 4 shows a schematic representation of an API lump with a diameter of 5000 μm. For this material, the NIR sample size was determined at 250 mg, which corresponds to a penetration depth of approximately 1000 µm. The penetration depth of the corresponding NIR sample size can be derived by dividing the sample volume (250 mg/ρ (0.6 Kg/L) = 417 mm3) by the sensor dimensions (spot size of NIR sensor = 24 mm, diameter = 452 mm2), which results in approximately 1 mm (1000 µm). The NIR intensity gradient was determined at 100, 250, 500, and 1000 μm and resulted in 35% contribution to the NIR spectrum at 100 μm, 80% at 250 μm, 95% at 500 μm, and 100% at 1000 μm. For lump sizes larger than the penetration depth of 1000 μm, parts will not be measured since parts > 1000 μm will simply not contribute to the NIR spectrum. For lump sizes around the penetration depth, the entire lump will contribute to the spectrum. However, due to the NIR intensity gradient, parts closer to the penetration depth will have a minor contribution to the NIR spectrum, while parts close to the sensor will contribute significantly. Lumps much smaller than the penetration depth will suffer less signal loss due to the NIR intensity gradient.

Figure 4: Schematic representation of a 5000 µm lump and the near infrared (NIR) sample size corresponding to a penetration depth of 1000 µm. In the zoomed picture of the lump, the NIR intensity gradient is shown. By combining the fraction of the lump included in the NIR sample size and the impact of the NIR intensity gradient, the effectively measured amount can be calculated.

The impact of the NIR intensity gradient can be estimated quantitatively by multiplying the lump volume at a defined penetration depth with the relative contribution percentage. In this case, the impact = lump volume up to 100 μm x 35% + lump volume 100-250 μm x 45% + lump volume 250-500 μm x 15% + lump volume 500-1000 μm x 5%; in a case for which the lump diameter is smaller than 1000 μm, the contribution of “layers” above the lump diameter should be ignored. Note that for all examples, it is assumed that the lumps are spherical and ideally positioned to the NIR sensor. In practice, however, lumps are neither exactly spherical nor positioned ideally to the sensor; therefore, the estimated effectively measured lump fractions should be used as indication. For the large 10000-μm diameter lump, the calculated effectively measured amount of API (0.82 mg, see Figure 4) is, however, very close to the expected lump contribution of 1.2 mg. 

Besides the intensity of the high content spikes in the blend curve, an overall decrease in content of approximately 20% was observed for batch B (see Figure 2). Given that the presence of relative large lumps will result in a lower measured content due to the discussed mechanisms, it is not surprising that a significant amount of API is not measured by NIR when a significant amount of API lumps are present in the blend. The magnitude of the decrease in content is directly related to the amount and size of lumps and the ratio to uniformly distributed API. In general, when a decrease in content is observed, and positive content spikes occur, there is a strong indication of the presence of (large) lumps. 

Discussion 

The provided examples and calculation are based on a number of assumptions known not to be exactly true. Lumps are not exactly spherical; they do show variance in shape, size, and surface; they are not positioned ideally at the NIR interface; and they will most likely not be evenly distributed throughout a blend. In addition, the empirically determined NIR intensity gradient is a relatively rough estimate of the actual loss of intensity. By more exact determination of this gradient, for instance at more penetration depth levels, and by applying a more complicated integral function, the NIR intensity gradient impact may be predicted more accurately. Whether a more accurate intensity gradient will provide relevant information for interpretation is somewhat uncertain, because the greatest uncertainty is related to the dimension of the lumps. However, in general, the presence and intensity of positive content spikes in combination with a lower overall content are strong indicators of the presence of lumps and can provide information regarding the amount and expected lump sizes in the blend. 

The extraction of quantitative information regarding lumps from NIR blend curves enables studying the behavior of lumps in a real-life situation, without the need for manual sampling and disturbance of the process.  By applying different shear forces (rotation speed, fill grades, etc.), an effective breakdown of lumps can be studied dynamically and used for process optimization, setting specifications, or troubleshooting. Advanced statistical analysis of obtained NIR content predictions within a blend curve may be applied next to the qualitative interpretation of the NIR data to enhance and quantify lump detection and related non-uniformity. If lumps are present in a blend, normality of the predicted content distribution will decrease; by applying advanced normality tests to the inline data in real time, the sensitivity of lump detection can be further increased. 

Conclusion

In cases where spectral properties allow quantification of an ingredient, inline NIR can be a powerful tool to detect lumps. It was shown that inline NIR is capable of detecting API lumps in low-dosed powder mixtures. It is, however, mandatory that the ingredient of interest shows clear and specific spectral features in NIR because a content of approximately 1% or lower is often difficult to quantify. For higher doses, the risk of non-uniformity decreases while it becomes easier to quantify by NIR. 

Positive content spikes and a lower overall content will be observed in inline NIR if relatively large lumps are present in a blend. The lower the content, the higher the level of detection (if the NIR signal allows adequate quantification of the ingredient) for lumps. High content spikes of lumps are due to a NIR sub-sampling effect that is less intense than expected. For lumps of 5000 μm-10000 µm, less than 1% of the lump mass is effectively measured as the penetration depth of the NIR radiation is effectively approximately 500 µm. The reflected intensity and contribution to the NIR spectrum decreases rapidly as a function of the penetration depth. Sample information at >250 μm distance to the NIR sensor already contributes significantly less to the NIR spectrum compared to sample information < 250 μm. The intensity of the high content spikes caused by lumps is also related to material properties and dimensions of the NIR analyzer. Determination of the NIR intensity gradient and estimation of the effective sample size allows adequate interpretation of inline NIR data. In combination with either qualitative or more advanced statistical tools, lumps can be detected effectively in low-dosed powder blends. Besides the well-known content uniformity monitoring applications, inline NIR may, therefore, also be applied as an effective tool to detect lumps in low-dosed powder blends. 

References

1. D.A. Burns and E.W. Ciurczak, Handbook of Near-Infrared Analysis (CRC Press, Taylor & Francis Group, 3rd ed., 2007).
2. W. Li et al., International Journal of Pharmaceutics, 350 (1-2) 369-373 (2008).
3. H. Ma and C.A. Anderson, Journal of Pharmaceutical Sciences, 
97 (8) 3305-3320 (2008).
4. FDA, Draft Guidance for Industry, Development and Submission of NIR Analytical Procedures, (Rockville, MD, March 2015).
5. EMA, Guideline on the Use of Near Infrared Spectroscopy by the Pharmaceutical Industry and the Data Requirements for New Submissions and Variations (London, January 2014).
6. O. Berntsson et al., Analytical Chemistry, 71 (3) 617-623 (1999).
7. F.C. Clarke et al., Applied Spectroscopy, 56 (11) 1475-1483 (2002).

Article Details

Pharmaceutical Technology
Vol. 41, No. 9
September 2017
Pages: 36-44

Citation

When referring to this article, please cite it as A. Gerich, et al., “Detection of Lumps in Powder Blends by Inline NIR,” Pharmaceutical Technology 41 (9) 2017.

About the Authors

Ad Gerich is managing director, InProcess-LSP, The Netherlands; Michiel Damen is application specialist, InProcess-LSP, The Netherlands; Willy Verhoeven is senior scientist, Polymeric Dosage Forms, MSD Oss, The Netherlands; Jorina Verhoog is section head, Polymeric Dosage Forms, MSD Oss, The Netherlands; Sai Prasanth Chamarthy is associate director, Respiratory Product Development, Merck & Co, Rahway, NJ USA; and Rut Besseling is director, Science and Technology, InProcess-LSP, The Netherlands.

native1_300x100
lorem ipsum