Graduate School of Pharmaceutical Sciences, Duquesne University, Pittsburgh, Pennsylvania 15282.
SMDD, Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indiana 46285.
J Pharm Sci. 2019 Jun;108(6):2119-2127. doi: 10.1016/j.xphs.2019.01.023. Epub 2019 Feb 8.
This study utilized multiple modeling approaches to predict immediate release tablet dissolution profiles of 2 model drugs: theophylline and carbamazepine. Two sets of designs of experiments were applied based on individual drug characteristics to build in adequate dissolution variability. The tablets were scanned using a near-infrared (NIR) spectrometer and then subjected to in vitro dissolution test at critical time points. Because of the inherent difference in dissolution profiles, a hierarchical modeling approach was applied for theophylline data, whereas global models were constructed from carbamazepine data. The partial least squares models were trained using 3 predictor sets including (1) formulation, material, and process variables, (2) NIR spectra, and (3) a combination of both. The dependent variables of the models were the dissolution profiles, which were presented either as parameters of Weibull fitting curves or raw data. The comparison among the predictive models revealed that the incorporation of NIR spectral information in calibration reduced prediction error in the carbamazepine case but undermined the performance of theophylline models. It suggests that the modeling strategy for dissolution prediction of pharmaceutical tablets should not be universal but on a case-by-case basis.
本研究采用多种建模方法来预测两种模型药物（茶碱和卡马西平）的即释片剂溶出度。根据药物的个体特征，应用了两组实验设计来构建足够的溶出度变化。使用近红外（NIR）光谱仪对片剂进行扫描，然后在关键时间点进行体外溶出度试验。由于溶出度曲线的固有差异，茶碱数据采用分层建模方法，而卡马西平数据则采用全局模型构建。偏最小二乘模型使用包括（1）制剂、材料和工艺变量、（2）NIR 光谱和（3）两者结合的三种预测器集进行训练。模型的因变量是溶出度曲线，以 Weibull 拟合曲线的参数或原始数据表示。预测模型之间的比较表明，在卡马西平情况下，将 NIR 光谱信息纳入校准可降低预测误差，但会降低茶碱模型的性能。这表明，药物片剂溶出度预测的建模策略不应该是通用的，而应该根据具体情况而定。
