Development of a method for nondestructive NIR transmittance spectroscopic analysis of acetaminophen and caffeine anhydrate in intact bilayer tablets.

Calibration models for nondestructive NIR analysis of API (active pharmaceutical ingredient) contents in two separate layers of intact bilayer tablets were established. These models will enable the use of NIR transmittance spectroscopy in bilayer tableting processes for the control of API contents in separate layers. Acetaminophen and caffeine anhydrate were used as APIs, and tablets were made by the direct compression method. Their NIR spectra were measured in the transmittance mode. The reference assay was performed by HPLC. Calibration models were generated by the partial least-squares (PLS) regression. The initial calibration generated models with insufficient linearity and accuracy because the fluctuation range of tablet thickness was excessively large and irrelevant information on the thickness fluctuation was included in the models. By narrowing the fluctuation range to determine the proper range for acceptable prediction accuracy, it was confirmed that calibration models with less irrelevant information can be generated when the range was 4.30+/-0.06 mm or narrower. Furthermore, the fluctuation range of 4.30+/-0.06 mm was considered to be empirically valid in covering the fluctuation actually observed in ordinary tableting processes. Thus, the sample tablets within this range were used to generate the final calibration models, and calibration models sufficient in linearity and accuracy were established. In addition, it was proven that controlling the irradiated side was unnecessary. Namely, it is not necessary to keep the same side of sampled tablets for the online NIR analysis during bilayer tableting. It is useful, in order to obtain adequate calibration models, to evaluate the variable factors that affect the linearity and accuracy of the generated models and restrict the range of models or use a subset of prepared samples. Loading vectors, explained variances, and correlation coefficients between components and scores are important for the evaluation of variable factors.