De-risking Scale-up of a High Shear Wet Granulation Process Using Latent Variable Modeling and Near-Infrared Spectroscopy

Abstract

In the development of wet granulated drug products, two primary sources of variance (disturbance) include the operational scale of the high shear wet granulation (HSWG) process and active pharmaceutical ingredient (API) lot-to-lot variability, particularly for formulations containing a high drug load. This paper presents a novel Process Analytical Technology strategy using latent variable modeling with nearinfrared spectroscopy (NIRS) to reduce risk in scale-up operations of the HSWG process while simultaneously accounting for API lot-to-lot variability, even with limited manufacturing history. The process involves building a partial least square (PLS) model among the API material properties, the HSWG process parameters, and the NIRS end points of the HSWG process based on small-scale design of experiment batches. The PLS model is then used in an optimization framework to find suitable small-scale mechanical process parameters (impeller/chopper speed) that approximate a previous large-scale operation so as to keep the NIRS end point of large-scale operation constant. Prior to making additional large-scale batches with a new lot of API, NIRS end points of large-scale HSWG with the new API lot are predicted based on the PLS model developed from the smallscale operation. If the predicted NIRS end point for the HSWG using the new API lot is not within the target region, the risks associated with the scale-up operation can then be significantly reduced by modifying other HSWG process parameters such as the total amount of water added or total granulation time to achieve the target region. A case study is presented that demonstrates the effectiveness of this methodology during development and scale-up of a drug product manufactured using a HSWG process.

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

“…Besides, two distance criteria, i.e., Hotelling's T 2 and Q residual, 21) were utilized to assess the validity of the LW-PLSR model for a query. Kamohara et al 22) and Muteki et al 23) reported that T 2 and Q could test the validity of the PLSR model for the query. In the previous research, 13) we demonstrated that the assessment based on T 2 and Q was also valuable for testing whether the LW-PLSR model was valid for the query.…”

Big Data Analytics in Chemical Engineering

“…Eberle et al (117) showed that through iterative application of PLS and ANOVA analysis, the most frequent causes of yield loss could be identified and improved. To prevent drug product quality from being affected by undesired variability of incoming raw materials, Muteki et al (118) showed that the effects of raw materials could be modeled using a latent-variable approach. A subsequent optimization can then eliminate such variabilities by strategically combining raw materials in later processing steps.…”

General Framework for Latent Variable Model Inversion for the Design and Manufacturing of New Products

“…Applications of LVRM inversion have been proposed in the pharmaceutical industry for raw material acceptance, 15 process development and optimization, 19,28 and process scale-up. 29,30 Although all the above-mentioned studies used LVRM inversion to solve different kinds of problems, the objective function being minimized has often been tailored to the specific case study. In this paper we present a general framework to perform LVRM inversion, which includes several possible different cases which one may encounter in a product/process design exercise.…”