Abstract
Data from X-ray powder diffraction (XRD) were subjected to a partial least-squares regression analysis (PLS) to build a calibration model for predicting the polymorphic content of carbamazepine (CBZ). The effectiveness of the PLS method in the construction of calibration models was analyzed by a scientific approach based on a regulation vector. CBZ forms I and III were characterized by differential scanning calorimetry (DSC) and XRD. Powder mixtures of forms I and III at various ratios (0 - 100% w/w; form III) were subjected to XRD. Five diffraction peaks were used for the peak-area method to compare with PLS. The results obtained by PLS had a better predictive accuracy compared to those of the peak-area method. The XRD-PLS method was established as a non-destructive, non-contact way to avoid the particle orientation effect based on statistical theory.
