(446f) A Process Monitoring Approach for Multistep Continuous Flow Pharmaceutical Manufacturing: Combining Dynamic Modeling and IR Spectroscopy with Changepoint Analysis

Within the growing field of continuous manufacturing (CM) for pharmaceutical processes is the parallel application of process analytical technology (PAT) and process modeling for these continuous operations. For drug substance CM processes, the ultimate goal is to consistently produce high-quality material from experimentally determined process set-points for parameters such as concentration, molar-equivalents, flow-rate, and reactor temperature. The quality of the product stream is sensitive in varying degrees to each of these parameter setpoints and understanding these sensitivities and intertwined dependencies is a large portion of the experimental process development and in-silico modeling endeavors of the design stage. One variety of these in-silico models is the full computational fluid dynamics (CFD) analysis of the reactor, which provides spatial and temporal resolutions for selected process conditions. As a proof of concept for moving towards incorporating online CFD models for continuous pharmaceutical processes, a lightweight on-line process simulation, referred to as the dynamic model (DM), was developed and implemented offline to yield a continuous concentration prediction using four process parameter readings and reaction kinetics data. This DM provides more meaningful insight into the product quality than from reading the parameter set-points independently. A series of experiments in which these process parameters were manually varied in either step change or gradual failure fashion, and in-line IR spectra of the process outlet stream was continuously collected to yield a direct measurement of the product concentration in the stream. Additionally, changepoint analysis (CPA) was applied offline to the process parameter data and both IR and DM concentration predictions. The DM was observed to overestimate IR and HPLC concentrations, but trends from the prediction yielded changepoints that corresponded to changepoints detected in the IR predictions, as well as with changepoints detected in the process parameters.