Role of continuous moisture profile monitoring by inline NIR spectroscopy during fluid bed granulation of an Enalapril formulation

Continuous Twin-Screw wet granulation process with In-Barrel drying and NIR setup for Real-Time Moisture Monitoring

The purpose of this study was to evaluate if wet granule formation and drying could take place in a single operation by utilizing in-barrel drying. The drying kinetics of the formulation were studied in order to select appropriate processing parameters and assess feasibility with short residence times in the extruder. The 18-mm extruder was operated in a 40:1 L:D ratio with 8 zones. The first two zones were used for material feeding and wet granule formation and the remaining zones were used for drying at elevated temperature. The impact of screw configuration as well as screw speed, feed rate, and residence time were all studied to optimize the drying process. Due to limitations of temperature and residence time, vacuum was added to enable sufficient drying. In-line NIR spectroscopy was incorporated into the twin-screw wet granulation (TSWG) process to monitor the moisture content of wet granules in real-time. The set-up was optimized and a predictive model was developed for future experiments. This study demonstrated the success of this technique on a pilot-scale (18-mm) extruder for the first time. Granules were formed and dried to a target loss on drying (LOD) of less than 2 % at moderate temperatures (100 °C - 110 °C) with one single operation. Streamlining wet granulation and drying into one unit operation can have a profound impact on pharmaceutical manufacturing reducing time, footprint, and environmental exposure due to reduced product transfers.

An Enhanced Stacking Ensemble Method for Granule Moisture Prediction in Fluidized Bed Granulation

Moisture is a crucial quality property for granules in fluidized bed granulation (FBG) and accurate prediction of the granule moisture is significant for decision making. This study proposed a novel stacking ensemble method to predict the granule moisture based on granulation process parameters. The proposed method employed k-nearest neighbor (KNN), random forest (RF), light gradient boosting machine (LightGBM) and deep neural networks (DNNs) as the base learners, and ridge regression (RR) as the meta learner. To improve the diversity of the base learners, perturbations of the input variables and network structures were adopted in the proposed method, implemented by feature construction and combination of multiple DNNs with a different number of hidden layers, respectively. In the feature construction, a SHapley Additive exPlanations (SHAP) approach was innovatively utilized to construct effective synthetic features, which enhanced the prediction performance of the base learners. The cross-validation results demonstrated that the proposed stacking ensemble method outperformed other machine learning (ML) algorithms in terms of performance evaluation criteria, for which the parameters MAE, MAPE, RMSE, and Adj. R2 were 0.0596, 1.5819, 0.0844, and 0.99485, respectively.

Gray-box Soft Sensor for Water Content Monitoring in Fluidized Bed Granulation

Soft sensors are powerful tools for the implementation of process analytical technology (PAT). They are categorized into white-box (first-principle), black-box (statistical), and gray-box models. Gray-box models integrate white-box and black-box models to address each drawback, i.e., prediction accuracy and intuitiveness. Although they have been applied to various industrial processes, their applicability to water content monitoring in fluidized bed granulation has not been reported. In this study, we evaluated three types of gray-box models, i.e., parallel, serial, and combined gray-box models, in terms of prediction accuracy using real operating data on a commercial scale with two formulations. The gray-box models were constructed by integrating the heat and mass balance model (white-box model) and locally weighted partial least squares regression (LW-PLSR) model (black-box model). LW-PLSR was utilized to cope with collinearity and nonlinearity. In the serial gray-box models, LW-PLSR models adjusted the fitting parameters of the white-box model depending on the process parameters for each query. In the parallel gray-box or combined gray-box models, LW-PLSR models compensated for the output error of the white-box or serial gray-box models, respectively. The results demonstrated that all three types of gray-box models improved the prediction accuracy of the white-box models regardless of the formulation. Besides, we proposed the assessment method based on Hotelling's T² and Q residual for gray-box models using LW-PLSR, which contributes decision support to select gray-box or white-box model. The accurate and descriptive gray-box models are expected to enhance process understanding and precise quality control in fluidized bed granulation.

Prediction Performance and Economic Efficiency of Soft Sensors for in-Line Water Content Monitoring in Fluidized Bed Granulation: PP-Based Model vs. NIRS-Based Model

Soft sensors play a crucial role as process analytical technology (PAT) tools. They are classified into physical models, statistical models, and their hybrid models. In general, statistical models are better estimators than physical models. In this study, two types of standard statistical models using process parameters (PPs) and near-infrared spectroscopy (NIRS) were investigated in terms of prediction accuracy and development cost. Locally weighted partial least squares regression (LW-PLSR), a type of nonlinear regression method, was utilized. Development cost was defined as the cost of goods required to construct an accurate model of commercial-scale equipment. Eleven granulation lots consisting of three laboratory-scale, two pilot-scale, and six commercial-scale lots were prepared. Three commercial-scale granulation lots were selected as a validation dataset, and the remaining eight granulation lots were utilized as calibration datasets. The results demonstrated that the PP-based and NIRS-based LW-PLSR models achieved high prediction accuracy without using the commercial-scale data in the calibration dataset. This practical case study clarified that the construction of accurate LW-PLSR models requires the calibration samples with the following two features: 1) located near the validation samples on the subspace spanned by principal components (PCs), and 2) having a wide range of variations in PC scores. In addition, it was confirmed that the reduction in cost and mass fraction of active pharmaceutical ingredient (API) made the PP-based models more cost-effective than the NIRS-based models. The present work supports to build accurate models efficiently and save the development cost of PAT.

Scale-Free Soft Sensor for Monitoring of Water Content in Fluid Bed Granulation Process

In-line monitoring of granule water content during fluid bed granulation is important to control drug product qualities. In this study, a practical scale-free soft sensor to predict water content was proposed to cope with the manufacturing scale changes in drug product development. The proposed method exploits two key ideas to construct a scale-free soft sensor. First, to accommodate the changes in the manufacturing scale, the process parameters (PPs) that are critical to water content at different manufacturing scales were selected as input variables. Second, to construct an accurate statistical model, locally weighted partial least squares regression (LW-PLSR), which can cope with collinearity and nonlinearity, was utilized. The soft sensor was developed using both laboratory (approx. 4 kg) data and pilot (approx. 25 kg) scale data, and the prediction accuracy in the commercial (approx. 100 kg) scale was evaluated based on the assumption that the process was scaled-up from the pilot scale to the commercial scale. The developed soft sensor exhibited a high prediction accuracy, which was equivalent to the commonly used near-infrared (NIR) spectra-based method. The proposed method requires only standard instruments; therefore, it is expected to be a cost-effective alternative to the NIR spectra-based method.

Real time stability and viability prediction of the anticancer BCG after lyophilization

To test if trehalose could be a better cryoprotectant for BCG than the usually used lactose and predict viability of BCG during shelf-life, BCG was suspended into three stabilizer systems containing 15% w/v trehalose, trehalose-gelatin mixture (in ratio, 30:1 w/w) or lactose. Each formula was lyophilized and several quality parameters were tested before and after lyophilization including sterility, safety, viability, morphology and moisture content. Samples of lyophilized formulae were tested for their reconstitution time and others were charged to stability chambers at 5°C for the performance of real time study. Shelf-life of each formula was estimated and correlation between moisture content and loss in viability was established at each time interval of the real time stability study. Sterility, safety and morphology were retained after lyophilization. Just after lyophilization, minute diminish in viability was observed in the presence of each stabilizer (0.02-0.05%). There was no significant difference in reconstitution time of the three lyophilized formulae. Lactose BCG had the highest shelf-life among the used cryoprotectants during the real time stability studies. Also, moisture content was highly correlated to viability with correlation coefficient ranged between 0.97 and 0.99 and so, the former could be used for prediction of viability throughout the vaccine shelf-life.

Non-destructive prediction of the drug content of an acetaminophen suppository by near-infrared spectroscopy and X-ray computed tomography

The purpose of this study is to develop non-destructive methods to determine the drug content of suppositories using near-infrared (NIR) spectrometry and X-ray computed tomography (XCT). The suppository samples (acetaminophen content: 0, 100, 200, 300, 400 and 500 mg/suppository) consisted of acetaminophen powder and hard fat. NIR spectra of 18 standard suppository samples were recorded, and the data were divided into two wave number ranges, 4000-10 000 cm(-1) (LR), and 4280-6650 cm(-1) (SR). The best calibration model was determined to minimize the standard error of cross-validation (SECV) by the leave-one-out method in the partial least squares regression (PLS). Sliced XCT images of the suppositories were measured, and apparent density (AD) was evaluated using the image of the sample. The NIR models gave the best correlation coefficient constant (R) values, since the results for LR and SR gave straight lines with R of 0.9274 and 0.9707, respectively. The AD of the suppositories by XCT increased with increasing drug content, and the relationship between the AD and drug content had a straight line with R of 0.9958. Both NIR and X-ray CT performed accurate measurements of suppository samples through plastic packaging.