Sustainable and green solvent-free quantification of trace levothyroxine in tablets using portable and benchtop NIR spectrometers with partial least squares regression.

In this study, we developed PLSR models using both benchtop and portable NIR spectral data to quantitatively determine levothyroxine in commercial tablets containing 25-150 µg per tablet of the drug. The optimal PLSR model for benchtop NIR spectrometer was performed using a standard normal variate combined with the Savitzky-Golay first derivative and mean center data preprocessing with three latent factors. Model performance was defined by its R2 model, R2 Pearson, root mean square error of prediction (RMSEP), and bias. In this case, the values were 0.9852, 0.9585, 6.5814, and 1.6483, respectively. Contrarily, for portable NIR spectrometer, the optimal PLSR model was obtained from Savitzky-Golay first derivative combined with orthogonal signal correction and mean center data preprocessing with three latent factors. The performances of the R2 model, R2 Pearson, RMSEP, and bias were 0.9747, 0.9446, 7.2820, and -1.4825, respectively. These results highlight the potential of combining PLSR with NIR spectral data and, in particular, the potential of portable devices with performances comparable to those of benchtop devices for applications in pharmaceutical solid dosage forms with very low concentrations, while providing a greener and more sustainable alternative to solvent-based chromatographic analysis.