Technologie Servier, Chemometrics Department, 45000 Orléans, France.
J Pharm Biomed Anal. 2011 Feb 20;54(3):510-6. doi: 10.1016/j.jpba.2010.09.029. Epub 2010 Oct 1.
A near infrared (NIR) method was developed for determination of tablet potency of active pharmaceutical ingredient (API) in a complex coated tablet matrix. The calibration set contained samples from laboratory and production scale batches. The reference values were obtained by high performance liquid chromatography (HPLC) and partial least squares (PLS) regression was used to establish a model. The model was challenged by calculating tablet potency of two external test sets. Root mean square errors of prediction were respectively equal to 2.0% and 2.7%. To use this model with a second spectrometer from the production field, a calibration transfer method called piecewise direct standardisation (PDS) was used. After the transfer, the root mean square error of prediction of the first test set was 2.4% compared to 4.0% without transferring the spectra. A statistical technique using bootstrap of PLS residuals was used to estimate confidence intervals of tablet potency calculations. This method requires an optimised PLS model, selection of the bootstrap number and determination of the risk. In the case of a chemical analysis, the tablet potency value will be included within the confidence interval calculated by the bootstrap method. An easy to use graphical interface was developed to easily determine if the predictions, surrounded by minimum and maximum values, are within the specifications defined by the regulatory organisation.
建立了一种近红外（NIR）方法，用于测定复杂包衣片剂基质中药物活性成分（API）的片剂效力。校准集包含来自实验室和生产规模批次的样品。通过高效液相色谱法（HPLC）获得参考值，并使用偏最小二乘（PLS）回归建立模型。通过计算两个外部测试集的片剂效力来挑战该模型。预测的均方根误差分别等于 2.0%和 2.7%。为了在生产现场的第二个光谱仪上使用该模型，使用了一种称为分段直接标准化（PDS）的校准传递方法。在转移之后，第一测试集的预测均方根误差为 2.4%，而没有转移光谱则为 4.0%。使用 PLS 残差的自举技术的统计技术用于估计片剂效力计算的置信区间。该方法需要优化的 PLS 模型、自举数量的选择和风险的确定。在化学分析的情况下，片剂效力值将包含在自举方法计算的置信区间内。开发了一个易于使用的图形界面，以方便确定预测值是否在监管机构定义的规格范围内，预测值被最小和最大值包围。
