Quantification of paracetamol through tablet blister packages by Raman spectroscopy and multivariate curve resolution-alternating least squares

• Pharmaceutical mixtures in blister packages were quantified by Raman spectroscopy.
• A ternary mixture design and three different batches of blister packages were used.
• Variations in blister packages interfered with Raman signal and quantification.
• By novel extension the MCR-ALS correlation constraint works on multiset structures.
• The extended correlation constraint can handle variations in blister packages.

Raman spectroscopy has potential for non-invasive detection and quantification of chemical solids concealed within packages e.g. plastic bottles, bags or blister packs. In this study we demonstrate how paracetamol as part of ternary mixtures can be quantified inside blister packages using the combination of Raman spectroscopy and multivariate curve resolution-alternating least squares (MCR-ALS). An MCR-ALS model with correlation constraint could predict paracetamol concentrations in a test set with an error of 0.8% w/w in the concentration range 0–20% w/w. Additionally, a new multivariate approach to overcome sample matrix effects is proposed to handle variation in the blister packing material between different production batches. These variations interfere with the Raman signal and therefore make quantification challenging with conventional regression methods. By a novel modification of the correlation constraint in MCR-ALS, quantification despite matrix effects is made possible. Calibration models with low prediction errors of paracetamol could be obtained if a simple calibration set of unshielded samples (samples measured directly, hence without blister package) was combined with a very small set of samples measured through blister. Models with prediction errors of 1.3, 1.4 and 1.9% w/w were obtained for samples measured through three distinct blister packages. The novelty in the modification allows working with a multiset structure, making local regression models that can handle both signal interferences and matrix effects (e.g. packing material, temperature, instrumental variations).