Terahertz Frequency-Domain Sensing Combined with Quantitative Multivariate Analysis for Pharmaceutical Tablet Inspection

TL;DR

This study demonstrates that terahertz frequency-domain spectroscopy combined with multivariate analysis can effectively quantify both API concentration and physical tablet density, offering advantages over traditional NIR and Raman methods.

Contribution

The paper introduces the use of terahertz frequency-domain spectroscopy with multivariate analysis for simultaneous API and density measurement in pharmaceutical tablets, addressing limitations of existing techniques.

Findings

High accuracy in tablet density prediction (R²=0.97)

Moderate accuracy in API concentration prediction (R²=0.98)

Low prediction errors for density (0.7%) and API (6%)

Abstract

Near infrared (NIR) and Raman spectroscopy combined with multivariate analysis are established techniques for the identification and quantification of chemical properties of pharmaceutical tablets like the concentration of active pharmaceutical ingredients (API). However, these techniques suffer from a high sensitivity to particle size variations and are not ideal for the characterization of physical properties of tablets such as tablet density. In this work, we have explored the feasibility of terahertz frequency-domain spectroscopy, with the advantage of low scattering effects, combined with multivariate analysis to quantify API concentration and tablet density. We studied 33 tablets, consisting of Ibuprofen, Mannitol, and a lubricant with API concentration and filler particle size as the design factors. The terahertz signal was measured in transmission mode across the frequency range 750 GHz to 1.5 THz using a vector network analyzer, frequency extenders, horn antennas, and four off-axis parabolic mirrors. The attenuation spectral data were pre-processed and orthogonal partial least square (OPLS) regression was applied to the spectral data to obtain quantitative prediction models for API concentration and tablet density. The performance of the models was assessed using test sets. While a fair model was obtained for API concentration, a high-quality model was demonstrated for tablet density. The coefficient of determination for the calibration set was 0.97 for tablet density and 0.98 for API concentration, while the relative prediction errors for the test set were 0.7% and 6%for tablet density and API concentration models, respectively.