Péterfi Orsolya, Kovács Béla, Casian Tibor, Tőkés Erzsébet Orsolya, Kelemen Éva Katalin, Zöldi Katalin, Nagy Zsombor Kristóf, Nagy Brigitta
Department of Organic Chemistry and Technology, Faculty of Chemical Technology and Biotechnology, Budapest University of Technology and Economics, Műegyetem rkp. 3, H-1111, Budapest, Hungary.
Department F1/Biochemistry and Chemistry of Environmental Factors, Faculty of Pharmacy, George Emil Palade University of Medicine, Pharmacy, Science and Technology of Târgu Mureș, 540142, Târgu Mureș, Romania.
AAPS J. 2025 Jul 8;27(5):118. doi: 10.1208/s12248-025-01100-2.
As process analytical technology (PAT) and real-time release testing (RTRT) are gaining momentum in the pharmaceutical industry, there is an increasing need for developing methods for the non-destructive and real-time characterization of the in vitro dissolution of pharmaceuticals. In recent years, several surrogate models relying on PAT measurements and advanced chemometric techniques have been published addressing this task. Nevertheless, methodologies for the fair comparison of the model performance and setting relevant acceptance criteria are still not well established. Therefore, this study aims to draw attention to appropriate model comparison when developing and applying surrogate dissolution models and highlight the limitations of the widely used dissolution curve comparison metrics, including the f similarity value. A set of 10 different artificial neural network (ANN) models were developed for the prediction of the dissolution profiles of clopidogrel tablets produced through hot-melt granulation and tableting. Models were fitted with diverse input data, including granulation nominal experiment settings and real recorded process parameters (e.g., air and material temperature, humidity, granulation and lubrication time, tableting pressure) and near-infrared spectra. The models' goodness was compared using the f factor, coefficient of determination (R) and root mean square error (RMSE). The results demonstrated that these measures do not sufficiently reflect the discriminating ability of the models. We proposed for the first time the use of the sum of ranking differences (SRD) method for the comparison of the prediction models, which proved to be an effective tool to assess the discriminatory power of surrogate dissolution models during model development.
随着过程分析技术（PAT）和实时放行检测（RTRT）在制药行业的应用日益广泛，开发用于药物体外溶出无损和实时表征的方法的需求也在不断增加。近年来，已经发表了几种依赖PAT测量和先进化学计量技术的替代模型来解决这一任务。然而，用于公平比较模型性能和设定相关验收标准的方法仍未得到很好的确立。因此，本研究旨在在开发和应用替代溶出模型时提请注意适当的模型比较，并强调包括f相似值在内的广泛使用的溶出曲线比较指标的局限性。开发了一组10种不同的人工神经网络（ANN）模型，用于预测通过热熔制粒和压片生产的氯吡格雷片剂的溶出曲线。模型采用多种输入数据进行拟合，包括制粒标称实验设置和实际记录的过程参数（如空气和物料温度、湿度、制粒和润滑时间、压片压力）以及近红外光谱。使用f因子、决定系数（R）和均方根误差（RMSE）比较模型的优劣。结果表明，这些指标不能充分反映模型的区分能力。我们首次提出使用排名差异总和（SRD）方法来比较预测模型，该方法被证明是在模型开发过程中评估替代溶出模型区分能力的有效工具。
