Issue link: https://www.e-digitaleditions.com/i/1287521
46 September 2020 Tablets & Capsules A sample set for validation was prepared at concen- tration levels of 90, 100, and 110 percent of target con- centration for both APIs. These formulations were not used in the calibration set but were instead used to eval- uate the predictive power of the models and discriminate between them, with the goal of acquiring the best model with a low root mean square error of prediction (RMSEP) and an R 2 value close to 1. NIR spectra data were collected using a Buchi NIRFlex N-500 spectrometer with the solid measurement cell and an XL adapter, which allows the measurement of irregu- lar solid samples or direct measurement with a transpar- ent plastic bag, in the wavelength range from 10,000 to 4,000 cm -1 . This instrument has a resolution of 8 cm -1 and a polarization interferometer with tellurium dioxide (TeO 2 ) wedges. Because the formulation is a suspen- sion, the spectra data was acquired using the instrument's transflectance mode. NIR models were elaborated for each API individu- ally. Blending the suspension caused scattering in the spectrum data and significant differences among the acquired spectra. Consequently, different pretreatments were applied individually or in combination to attempt to minimize scattering, baseline drift errors, and background noise and increase the signal-to-noise ratio [6, 7]. Once the spectra data had been collected (Figure 1), chemometrics software was used to apply a multivariate analysis tool such as partial least squares regression (PLS). The ibuprofen model was developed using a spectral range from 10,000 to 4,000 cm -1 , and a Savitzky-Golay smoothing and derivative filter was applied (Figure 2a), while a Savitzky-Golay differentiation filter with a first polynomial order and standard normal variate (SNV) normalization was applied to develop the caffeine model (Figure 2b). Next, a partial least squares (PLS) regression was performed. PLS regression is a method used when prediction is the goal and many highly collinear factors are present in the data set. method to determine the concentrations of ibuprofen and caffeine in softgels. To develop this method, 15 formulation samples were prepared at laboratory scale with different API concentration levels from 0 to 120 percent, with 100 percent being the target concentration for each API, as shown in Table 1. For example, a formulation might contain 50 percent of the target ibuprofen and 100 percent of the target caffeine. The calibration set also included a placebo and capsules from pilot scale to increase the model's selectivity and sensitivity. Figure 1 NIR spectra from calibration set 0% Ibu + 259.9% Caf 50% Ibu + 100% Caf 90% Ibu + 150% Caf 100% Ibu + 105% Caf 115% Ibu + 110% Caf 110% Ibu + 95% Caf 105% Ibu + 150% Caf 95% Ibu + 80% Caf 0% Ibu + 0% Caf 100% Ibu + 100% Caf 90% Ibu + 200% Caf 100% Ibu + 0% Caf 100% Ibu + 120% Caf 120% Ibu + 90% Caf 80% Ibu + 50% Caf Final product Log (1/R) 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 Wavenumber (cm -1 ) 9,000 8,000 7,000 6,000 5,000 4,000 Figure 2 90% Ibu + 150% Caf 115% Ibu + 110% Caf 105% Ibu + 150% Caf 0% Ibu + 0% Caf 90% Ibu + 200% Caf 100% Ibu + 120% Caf 0% Ibu + 259.9% Caf Final product 50% Ibu + 100% Caf 100% Ibu + 105% Caf 110% Ibu + 95% Caf 95% Ibu + 80% Caf 100% Ibu + 100% Caf 100% Ibu + 0% Caf 120% Ibu + 90% Caf 80% Ibu + 50% Caf a. NIR spectra after Savitzky-Golay smoothing and first derivative treatment Wavenumber (cm -1 ) Log (1/R) 0.04 0.02 0.00 -0.02 -0.04 -0.06 6,500 6,000 5,500 5,000 4,500 4,000 b. NIR spectra after Savitzky-Golay first derivative and SNV normalization treatment 6,500 6,000 5,500 5,000 4,500 4,000 Log (1/R) 0.03 0.02 0.01 0.00 -0.01 -0.02 -0.03 -0.04 -0.05 Wavenumber (cm -1 )