Comparative Accuracy of Prediction Values of Contents of Nimodipine Given by the 4 Analytical Models for Near Infrared Spectrometric Analysis of Nimodipine Tablets

Diffuse reflection near infrared spectra (NIRS) of the nimodipine (NMDP) tablets and of its powder obtained by crushing and grinding of the tablets were collected by the integral sphere-analytical module annexed to the NIR spectrometer. Furthermore, reflection NIRS and transmission NIRS of the NMDP tablet were collected by the annexed analytical module for tablet. And then 4 quantitative analytical models (named model 1, 2, 3 and 4) were derived separately from the 4 NIRS using TQ Analyst 9.0 software. On the other hand, Fifteen samples of the NMDP tablet were collected by taking 5 samples from each of 3 batches of the drug, and analyzed for their NMDP contents by the method of HPLC given in the "Pharmacopoeia of PRC (2015)". The data obtained were used for external veritication of the prediction values of NMDP in the samples given by the 4 models. As shown by the statistical parameters and the performance indexes found by the internal and cross-verification, the model 2 was found to be best among the 4 models. By introducing the NIRS of the 15 samples into the 4 models separately, and data of prediction contents of NMDP in these samples were obtained. By comparing with the values of NMDP contents found by HPLC, the relative errors found between the prediction values given by the 4 models and the HPLC data were given in the following ranges:-2.41% to 2.68% (for model 1), 0.81% to 1.18% (for model 2), -5.03% to 4.55% (for model 3) and -5.99% to 5.56% (for model 4). It was obvious that the relative errors of model 2 were the smallest, and of model 1 were larger, and of model 3 and 4 were the largest, but quite close to each other. However, no significant differences were found between the results of HPLC analysis and prediction values given by any of the 4 models according to the results of t-test (α=0.05). Hence, any one of the 4 models could be applied to the determination of NMDP in the drug tablet. Model 2 gave prediction values with smaller relative error, but sample pretreatment was required in this case and longer time was required for the whole analytical process, model 1 could be considered to apply to the determination of NMDP in the drug tablet with poor uniformity or various shapes. Although the other 3 models gave prediction values with relatively larger relative errors, but they gave more rapid analysis since no sample pretreatment was required. Especially, model 3 and 4 could give reflection and transmission NIRS informations simultaneously and were favourable for comparative study.