Complemental Hard Modeling in Raman spectroscopy – A case study on titanium dioxide-free coating in-line monitoring

Abstract

Tablets are coated for taste or odor modification, for modified release profiles or as a protective layer to increase the stability. Here, titanium dioxide is frequently added as a coating component due to its opaque properties. Furthermore, its Raman activity makes it an integral part of in-line monitoring models. However, due to the carcinogenic potential of titanium dioxide, calcium carbonate is utilized as a substitute, exhibiting similar opaque properties. However, calcium carbonate tends to exhibit overlapped peaks with carbon hydrates in the Raman spectrum. Consequently, new models based on e.g. hard modeling are required instead of peak integration.

Highlights

Pharmaceutical coating

Raman spectroscopy

Process analytical technology

Titanium dioxide-free coating

Spectral hard modeling

Partial least squares regression

In this study, tablets were coated with a coating including calcium carbonate. Partial Least Squares Regression (PLS) and Complemental Hard Modeling (CHM) were examined as feasible in-line monitoring approaches. Furthermore, two different measurement positions in the coater were compared, orthogonal and tangential with respect to the moving tablet bed.

Cross-validation exhibited improved CHM performance with reduced RMSECV values of about 5%. The prediction of the coating mass growth occurred comparable with RMSEP values in a similar range of 2-5%. Despite this, the CHM´s achieved improved performance with reduced training data quantity and quality. The different measurement positions indicated no process-relevant differences.

Introduction

Coating pharmaceutical tablets is an essential unit operation to improve taste, appearance and stability combined with enabling controlled release and protecting the Active Pharmaceutical Ingredient (API). Traditionally, tablet coatings contain often titanium dioxide due to its opacity properties. However, with emerging health concerns, the pharmaceutical industry is currently switching to titanium dioxide-free coatings as it is considered to be potentially carcinogenic. Consequently, the mostly separated titanium dioxide peaks in Raman spectra are no longer available for peak integration. Calcium carbonate can serve as a substitute due to its similar opaque properties. However, it tends to exhibit interfering Raman peaks with hydrocarbon expedients like lactose and cellulose. Therefore, robust in-line monitoring strategies based on titanium dioxide free prediction models are necessary to ensure the quality and regulatory compliance of pharmaceutical tablet coatings.8

These in-line monitoring models provide real-time data about the process based on the implementation of Process Analytical Technologies (PAT). The data obtained in real time can be utilized for real-time release testing and thus replace the time-consuming off-line testing of samples. Furthermore, the acquired data has the potential to be integrated into control regimes and thus contribute to the digitalization of pharmaceutical processes. Here, the spectroscopic methods that are non-invasive and non-destructive and provide real-time data with high accuracy are particularly suitable.

In this context, near infrared spectroscopy (NIR) is known for its ability to provide chemical and physical information. This makes NIR spectroscopy a suitable in-line monitoring tool for evaluating of the coating thickness and composition of the coating material. Nevertheless, NIR requires considerable high calibration effort for molecule-specific applications. Furthermore, humidity can also be monitored as an additional critical quality attribute for moisture sensitive materials Terahertz time-domain imaging provides information on physical properties and is a promising tool for precise end-point determination in tablet coating for coating thicknesses and is independent of the coating and tablet core formulation. On the other hand, optical coherence tomography (OCT), enables critical quality attributes in-line monitoring of coating thickness, variability and roughness making OCT to a valuable aid in detecting defects in tablet coating as long as the refractive index difference can be determined. Although, OCT does not work if the coating formulation contains inorganic pigments20. A combination of X-ray fluorescence spectroscopy and a Monte Carlo simulation was used to determine the coating thickness21. However, despite the advantages of each PAT technology, the specific requirements of tablet coating processes, particularly for monitoring coating mass on tablets and determining coating endpoints, Raman spectroscopy was frequently implemented. It is characterized by a molecular specificity and high sensitivity.

In order to obtain quantitative information from the spectral data, models must be developed. In this context, Partial Least Squares Regression (PLS-R) is a well-established statistic method for building predictive models based on Raman spectroscopy data. It enables the extraction of relevant information from complex data sets and links between the spectral Raman information and critical quality attributes of tablet coatings like coating thickness and coating mass growth28. However, the performance of PLS-R can be limited by multicollinearity and overfitting caused by the incorrect selection of a PLS-R rank. Furthermore, nonlinearity can be only modeled to some extent31. Therefore, an alternative will be considered in this study.

Here, spectral Hard Modeling (HM) should be considered as an alternative approach that is capable to compensate the observed limitations of PLS-R. The principle of Hard Modeling (HM) consists of the construction of mathematical models in form of phenomenological motivated non-linear models representing peak functions (i.e. pseudo-Voigt type), so called Hard Models, which are integrated, superimposed and weighted. In contrast to traditional statistical regression methods, HM directly incorporates physical and chemical principles underlying the Raman spectral data and thus contributes to a more mechanistic interpretation. Additionally, the amount of data required for calibration is significantly reduced compared to a PLS-R. Complemental Hard Modeling (CHM) extends this approach. Here, only a single spectrum of the mixture and the spectra of the pure components are required as input data for the calibration. First Hard Models are combined to develop a mixture model, which is subsequently expanded by an additional Hard Model for the unknown component in the mixture spectrum.

The objective of this study is to compare well-established PLS-R with the comparatively new Complemental Hard Modeling for in-line monitoring of titanium dioxide-free tablet coatings. In particular, the study focuses on evaluating the performance of these models for prediction of the mass growth of the coating material. In addition, the influence of the probe position on the models was investigated. For this purpose, the probe was implemented at two different measurement positions, at the nozzle arm measuring orthogonal and in the back wall of the coater measuring tangential to the tablet bed.

Tablet cores

The tablet formulation consisted of 50 wt% paracetamol (Compap L, Mallinckrodt Pharmaceuticals, Staines-upon-Thames, Ireland), 44 wt% microcrystalline cellulose and colloidal silicon dioxide pre-blend (Prosolv SMCC 90, JRS Pharma, Rosenberg, Germany), 5 wt% crospovidone (Kollidon CL, BASF, Ludwigshafen, Germany) and 1 wt% magnesium stearate (Ligamed MF-2-V, Peter Greven, Bad Münstereifel, Germany). Tableting was executed on a rotary tablet press (XL 100, Korsch, Berlin, Germany)

René Brands, Jens Bartsch, Markus Thommes, Complemental Hard Modeling in Raman spectroscopy – A case study on titanium dioxide-free coating in-line monitoring, Journal of Pharmaceutical Sciences, 2024, ISSN 0022-3549, https://doi.org/10.1016/j.xphs.2024.10.044.