A comparative study of two data-driven modeling approaches to predict drug release from ER matrix tablets

Abstract
The pharmaceutical industry is striving to develop innovative and promising tools, increasingly embracing new data-driven approaches, to understand, improve and accelerate the drug product development process. While extended release (ER) oral formulations offer a number of advantages, including maintenance of therapeutic drug levels, a reduction in dosing frequency, and minimization of side effects, achieving consistent drug release profiles remains a significant challenge. As a critical attribute for drug absorption into systemic circulation, in vitro dissolution testing represents a time-consuming and complex method for the evaluation of such formulations. The main objective of this study was to develop a model for predicting drug dissolution in the quality by design (QbD)-based development of ER oral hydrophilic matrix tablets comprising polyethylene oxide (PEO). Two main modeling approaches are conducted and compared: (i) model screening to fit and compare multiple predictive machine learning (ML) models and then deploy the best model, in this case, artificial neural networks (ANN), and (ii) functional data analysis (FDA) combined with the design of experiments (DoE) that fit a smoothing model to each dissolution curve as a continuous function. A dataset comprising 91 ER matrix tablet formulations was analyzed, with the dissolution data split into training, validation, and test sets (70%, 20%, and 10%, respectively). The results demonstrated that both ANN and functional DoE (FDOE) models achieved high similarity with the experimental dissolution profiles, as indicated by f2 values ranging from 48 to 88 for the FDOE and 52 to 88 for ANN. This work highlights the potential of integrating advanced data-driven modeling techniques into ER drug development to enhance dissolution prediction accuracy and streamline the formulation process, thus reducing time and costs.
Introduction
Extended release (ER) oral formulations are widely used to provide a slower drug release and absorption rate, maintaining therapeutic drug levels for longer periods, reducing dosing frequency, enhancing patient compliance, and managing peak concentration-related side effects (Florence, 2011). Despite the advantages associated with ER oral formulations, formulation scientists in pharmaceutical companies continue to invest significant time and effort attempting to formulate ER tablets with the objective of achieving consistent and controlled drug release, mimicking the desired in vivo release pattern. In recent years, particular efforts have been made to understand the impact of raw materials and manufacturing process variability on the performance of ER dosage forms (Grund et al., 2014, Ilyés et al., 2021, Vanhoorne et al., 2016, Viridén et al., 2011, Zhang et al., 2019, Zhou et al., 2014). The release rate and kinetics of some ER formulations can be significantly affected by formulation and processing variables, and understanding the nature of these variations is key to developing a robust drug product.
A critical quality attribute (CQA) of these formulations is the in vitro dissolution profile, which reflects the rate and extent of drug release from the dosage form and acts as a surrogate to ensure consistent in vivo performance. During drug product development and manufacturing, dissolution testing is routinely employed to evaluate the performance of ER formulations and supports quality control, scale-up, bioequivalence studies, and post-approval regulatory changes (Grady et al., 2018). Given the sensitivity of dissolution behavior to formulation and processing variability, dissolution testing plays a pivotal role in drug product development and manufacturing. However, in addition to the inherent complexity of simulating in vivo conditions, conducting in vitro dissolution profiles of ER tablets presents several challenges. ER dissolution profiles are time- and cost-consuming, necessitating sophisticated mathematical modeling and a comprehensive understanding of drug release mechanisms for accurate interpretation of the resulting data (Grassi and Grassi, 2014).
To address all these challenges, the pharmaceutical industry has increasingly adopted a quality by design (QbD) approach, as outlined in ICH Q8 (R2) (2009). QbD emphasizes the development of a comprehensive understanding of drug product and manufacturing process. It is an integrated, science- and risk-based methodology characterized by a well-defined framework from the definition of quality target product profile (QTPP) to the implementation of a control strategy. The CQAs of a product are influenced by both the critical material attributes (CMAs) and the critical process parameters (CPPs). Encouraged by ICH guidelines, design of experiments (DoE) has been traditionally employed for systematically exploring the relationships between these critical factors, ensuring consistent product quality and performance (ICH Q8 (R2) 2009). In the design and development of ER matrices, classical DoE analysis evaluates drug release data at discrete time points, treating each point as an independent response variable (Sousa et al., 2023b). This approach, while effective for simpler systems, does not account for the dynamic nature of drug release kinetic of ER formulations (Ramsay and Silverman, 2005a). A more effective statistical framework, functional data analysis (FDA), has been developed and used in different research fields, for fitting complex models to time-dependent data, allowing researchers to analyze and interpret functional responses as continuous curves rather than discrete points (Kenett and Gotwalt, 2023, Ramsay and Silverman, 2005a). The integration of the FDA with the design of experiments (DoE), designated as functional DoE (FDOE), allows for a highly flexible fit and systematic exploration of the multidimensional space of formulation variables. An example of the application of FDOE is reported by Fidaleo (2020) in the food and bioprocess field to develop a dynamic design space of a milling process that accurately predicts the functional responses of fineness and energy.
Moreover, while QbD provides a robust framework for developing ER tablets, multifactorial designs and non-linear responses can be challenging to model effectively. The integration of advanced statistical techniques like multivariate data analysis (MVDA) (Sousa et al., 2023a) and more sophisticated machine learning (ML) tools can further enhance the process identification and comprehension of CMAs-CPPs-CQAs relationship. ML is a method of analyzing data using algorithms and statistical models to identify relationships, patterns and make predictions from large datasets containing vast amounts of information. ML has emerged in the pharmaceutical industry in the era of data-driven innovation, revolutionized by Pharma 4.0 and the rise of big data powered by artificial intelligence (AI), exponentially improving digital transformation across the pharmaceutical lifecycle from drug discovery to post-marketing surveillance (Arden et al., 2021, Zagar and Mihelic, 2022). Regulatory authorities acknowledge the growing significance of AI/ML in the drug development lifecycle and its potential across various stages of the drug development process (Nene et al., 2024). To date, various algorithms for ML have been employed, and many applications have already been described in the development of solid oral dosage forms (Lou et al., 2021), encompassing information on drug and excipient properties (Hayashi et al., 2018, Hayashi et al., 2019, Hayashi et al., 2023, Yoo et al., 2022), formulation (Akseli et al., 2017, Djuris et al., 2021, Duranovic et al., 2021) and manufacturing processes (Akseli et al., 2017, Han et al., 2018, Mäki-Lohiluoma et al., 2021, Paul et al., 2021). Notably, relatively few studies have investigated the use of ML techniques to accelerate the development of ER tablets. As discussed in our previous review, the literature on the application of ML in the development of ER tablets indicates that artificial neural networks (ANN) are the first and most commonly used approach, with the objective of optimizing formulation and modeling release kinetics (Sousa et al., 2023b). In a recent study, Galata and colleagues investigated the potential of different ML algorithms, including ANN, to predict the in vitro dissolution profiles of ER tablets. To this end, they employed data collected from near-infrared (NIR) and Raman spectroscopy. Drug content, matrix polymer content, compression force and particle size distribution were set as the input factors (Galata et al., 2019, Galata et al., 2021). Although ANN have demonstrated significant potential for predicting drug release, thereby facilitating real-time release testing, there is still scope for further improvement in regard to the incorporation of additional formulations and process parameters to increase the accuracy and robustness of the models.
In this study, a dataset comprising 91 batches of polyethylene oxide (PEO)-based ER matrix tablets was analyzed. PEO is a water-soluble, non-ionic polymer widely used in the development of ER matrix tablets due to its versatility in tablet manufacturing and unique swelling, hydration, and gel-forming properties. Upon contact with aqueous environments, PEO rapidly forms a hydrogel layer that controls the ingress of water and subsequent drug release, predominantly through a diffusion-controlled mechanism (Li et al., 2008, Park et al., 2010, Vanza et al., 2020). The molecular weight of PEO plays a crucial role in modulating its properties, with higher molecular weight grades forming more robust gel layers that slow drug release and reduce polymer erosion. (Martin and Rajabi-Siahboomi, 2014). However, the mechanical properties of PEO-based hydrogels, particularly under in vivo conditions, may be less resistant to mechanical stress making the matrix system more susceptible to an erosion process alongside with swelling and diffusion release mechanism. Consequently, optimizing the grade and concentration of PEO is essential for achieving a balance between robustness and the desired release profile (Draksler et al., 2021).
The main aim of this work was to evaluate the effectiveness of two advanced statistical approaches for predicting drug dissolution rates from ER matrix tablets using JMP® Pro 17. First, multiple predictive models are fitted to the data and compared to identify the best predictive model using a model screening platform. In the second approach, FDA was employed to fit each dissolution profile as a smooth function, and the resulting data were subsequently used to construct a predictive model. To our knowledge, there is no prior research employing FDOE for modeling dissolution profiles. Finally, the performance of the predictive ML model is compared with that of the FDA. By systematically evaluating these two advanced statistical approaches, this innovative methodology aims to enhance the understanding of predictive modeling and expand knowledge of future data-driven approaches in the pharmaceutical development of ER matrix tablets.
Download the full article as PDF here A comparative study of two data-driven modeling approaches to predict drug release from ER matrix tablets
or read it here
Materials
The active substance under study is a synthetic model drug classified as type III according to the Biopharmaceutical Classification System (BCS). The model drug source cannot be disclosed for confidentiality reasons. Different grades of polyethylene oxide (PEO) (POLYOX™ WSR N-750 with a MW of 300,000 Da; POLYOX™ WSR 1105 with a MW of 900,000 Da; POLYOX™ WSR N-60 K with a MW of 2,000,000 Da; and POLYOX™ WSR 303 with a MW of 7,000,000 Da) were kindly donated by DuPont (Dartford, UK). Silicified microcrystalline cellulose (SMCC) was provided by JRS Pharma (PROSOLV® SMCC HD90, Rosenberg, Baden-Wurttemberg, Germany). Maltodextrin “MD-IT12” was supplied by Roquette (Glucidex® IT 12, Lestrem, France). Isomalt “G721” was kindly donated by BENEO (galenIQ™ 721, Mannheim, Baden-Wurttemberg, Germany). Magnesium stereate was supplied by UNDESA (Barcelona, Spain). All other reagents were of analytical or high-performance liquid chromatography (HPLC) grade.
Manufacturing of tablets
Tablets were prepared by direct compression. The excipients were manually passed through a sieve with a mesh size of 500 µm. Each formulation was lubricated with 1 % w/w magnesium stearate, which was previously sieved through a 250 µm mesh. Tablets were produced by a compaction simulator (STYL’One Evolution, MEDELPHARM, France), which was tooled with a standard EU-D 12.75 × 6.38 mm elliptical punch and die.
A.S. Sousa, J. Serra, C. Estevens, R. Costa, A.J. Ribeiro, A comparative study of two data-driven modeling approaches to predict drug release from ER matrix tablets, International Journal of Pharmaceutics, Volume 671, 2025, 125230, ISSN 0378-5173, https://doi.org/10.1016/j.ijpharm.2025.125230.
Read also our introduction article on Magnesium Stearate here:
