Abstract :
[en] 1. Introduction
Near-infrared spectroscopy (NIR) is a powerful analytical tool approved by the EU and US pharmacopeias. It can provide an accurate description of the physicochemical composition of samples, and hence can be used to fingerprint a drug product. With the fast-development and miniaturization of handheld spectrophotometers, this vibrational spectroscopy technique is more and more used in a large range of research and industrial applications involving characterization, identification and quality control of drug products. These applications however require the use of accurate, robust, risk-oriented, computationally efficient decision-making tools to statistically compare high dimensional spectra to references in order to identify or control the quality of pharmaceutical products.
We propose a novel and probabilistic one-class classification strategy based on newly emerging chemometric techniques of (Bayesian) functional data analysis, for the identification and quality control of medicines. The strategy uses the concept of prediction bands as acceptance region.
2. Material and methods
A representative training set of spectra of a target product is sampled from several batches of that product using a MicroPhazir® (ThermoFisher Inc) reflection NIR spectrophotometer. Based on this set and using Bayesian (functional) principal component regression [1], a statistical prediction band is constructed so that it contains a high proportion, say at least 90% or 95% of future spectra of the product (see Figure 1 for illustration). The upper and lower limits of the band are used as critical trajectories or reference spectra that would enable testing the deviation from regular behavior or excursions out of the bands of any future unit from the product batch based on its spectrum, while controlling the risks of errors [2].
3. Results and discussion
The proposed one-class classification methodology has been applied to the identification of Dafalgan® 1g. Four other paracetamol-based drugs were used to evaluate the specificity. Spectra were measured with the handheld NIR device. The predicted trajectories of future Dafalgan® 1g spectra and their limiting behaviors (band limits) are illustrated on Figure 2A and B. The pattern of deviation from the band limits (acceptance region) are illustrated on Figure 2C and D for a Dafalgan® 1g spectrum and an Excedryn® spectrum respectively. The method compares favorably with existing methods like the SIMCA, with high sensitivity between 90% and 98% and similar specificity.
4. Conclusion
A new one-class classification method for identification and quality control of drug products is proposed, using prediction bands for NIR spectra. Compared with existing spectral matching models, the proposed approach is fully predictive, with more intuitive interpretation of classification results.
5. References
[1] J.S. Morris, Functional regression. Annual Review Statistics and its Applications 2, 2015, pp. 321-359.
[2] TH Avohou, et al. A probabilistic class-modelling method based on prediction bands for functional spectral data:Methodological approach and application to near-infrared spectroscopy. Analytica Chimica Acta 1144, 2021, pp. 130-149.
