Development of Near Infrared Spectroscopy-based Process Monitoring Methodology for Pharmaceutical Continuous Manufacturing Using an Offline Calibration Approach

Abstract

A near-infrared (NIR) calibration was developed using an efficient offline approach to enable a quantitative partial least-squares (PLS) chemometric model to measure and monitor the concentration of active pharmaceutical ingredients (API) in powder blends in the feed frame (FF) of a tablet press. The approach leveraged an offline "feed frame table," which was designed to mimic the full process from a NIR measurement perspective, thereby facilitating a more robust model by allowing more sources of variability to be included in the calibration by minimizing the consumption of API and other raw materials. The design of experiment (DOE) for the calibration was established by an initial risk assessment and included anticipated variability from factors related to formulation, process, environment, and instrumentation. A test set collected on the feed frame table was used to refine the PLS model. Additional fully independent test sets collected from the continuous drug product manufacturing process not only demonstrated the accuracy and precision of the model but also illustrated its robustness to material variability and process variability including mass flow rate and feed frame paddle speed. Further, it demonstrated that a calibration can be generated on the offline feed frame table and then successfully implemented on the full process equipment in a robust manner. Additional benefits of using the feed frame table include streamline model monitoring and maintenance activities in a manufacturing setting. The real-time monitoring enabled by this offline calibration approach can be useful as a key component of the control strategy for continuous manufacturing processes for drug products, including detecting special cause variations such as transient disturbances and enabling product collection/rejection based upon predetermined concentration limits, and may play an important role in enabling real-time release testing (RTRt) for manufactured pharmaceutical products.

Estimation of mass‐based composition in powder mixtures using Extended Iterative Optimization Technology (EIOT)

“…Both results are converted to percent of nominal concentration for the drug product (%). The method calibration for “Product A” by the PLS approach was reported elsewhere . In brief, a supersaturated design via Nearly Orthogonal Arrays was used to generate a 10‐blend DOE and calibration spectra from these blends were collected on an offline device to conserve the total amount of API used in the calibration.…”

“…Although both PLS and EIOT were found to estimate API concentration equivalently well during steady‐state operation, it is worth pointing out the different levels of effort required to establish the two models. The PLS model required significantly more effort than EIOT, as it consumes the time and resources necessary (e.g., personnel and material) to generate a representative calibration set based upon the aforementioned 5 kg of API and 10‐blend DOE, despite the use of offline device to conserve the amount of API used . However, EIOT only required two sets of nominal and placebo blends at two dose strengths, which does not require additional consumption of API other than steady‐state operation.…”

“…In this figure, each box represents the 25th and 75th percentile of samples. The boxes contrast the performance of a PLS model that was fitted using a complete experimental design (as discussed in Hetrick et al) with the EIOT model trained with data from two batches, at two concentration levels. As shown in Figure , the overall bias from laboratory results (obtained using HPLC) and the standard deviation at steady state for EIOT and PLS are comparable and no method is (in this case) particularly superior to the other.…”

“…The last free volume where mixing occurs is the feed‐frame of the tablet press; the mixing of the powders is incomplete before this point. Studies have shown that the concentration of drug as observed in the feed‐frame is practically the same as in the tablet . Hence, the development of PAT solutions to estimate the concentration of drug in the feed‐frame has received special attention from researchers.…”

“…This is a particularly difficult aspect as the development of a model that would potentially require less maintenance will have to account for variations and disturbances not typically seen at the development scale. Introducing these sources of variation experimentally may require large quantities of drug and incurred cost for use of a large‐scale facility; eventually raising the cost of development for the model. This high cost of model development is of course invested at risk before the drug becomes approved for commercial manufacture, and increasing overall cost of development for a new drug is opposite to current trends in the business.…”