Enabling a Higher Assurance of Control for Drug Product Manufacturing

Ian Leavesley, BSE

Over the past 6 years, Lilly has transitioned continuous manufacturing from an investigation of business case and core technologies into our preferred platform for development of new chemical entities.  This is enabled by three identical equipment / automation kits including two GMP kits.  These units have provided multiple batches across multiple molecules including primary stability and clinical trial material. Over 10 molecules have been tested on this system to develop a process that is robust to range of material properties, formulations, and processing conditions.   This case study will focus on the utilization of CM to produce cGMP drug product for late-stage small molecule clinical assets.

The current Lilly process is based on a direct compression platform of feeders, mixer, and tablet press.  We have found the continuous version of direct compression to be much more robust than the same process implemented in a batch mode (i.e. with batch mixing).  Direct compression also enables small scale modeling to be much more predictive which has led to a virtuous cycle of improvement in development. This has enabled rapid scalability to meet the dramatically changing approach to clinical trials to bring products to market sooner.

  • How did you define the boundaries of the material that is subjected to a release decision?

The current Lilly molecules are relatively small commercial volume products.  Because of this we do not attempt to use residence time modeling as part of lot boundaries.  Rather, we use conventional batch approaches to provide traceability of starting materials.  We have developed predictive models for use DURING a product run to make collect / reject decisions using a soft sensor based on the system of powder feeders.  This creates time-based boundaries for start of rejection and collection based on feeder disturbances

  • How do you identify material that has questionable quality (potentially non-conforming material) and how do you deal with it?

Lilly has a multi-layer control strategy based on feeder data, ratio control of the feeders, and optionally system models and/or powder spectroscopy to ensure quality product is produced.  All of these are capable of triggering isolation through the tablet reject gate at the appropriate time.   This step rarely triggers for our robust process.

  • How do you justify that the proposed sample rate of the CPPs or the CQAs and IPC reflects the inherent material characteristics and process dynamics is adequate

It is important to clarify that the word sample refers not only to a physical sample of a tablet but also to process and spectroscopic measurements.  The critical concept to determine sampling rates was the understanding that each sample point is not independent and therefore the challenge is not to detect individual unique occurrences but to detect a trend before it exceeds collection criteria.  Several orthogonal modeling and experimental approaches were taken to determine the fastest rate of change that could occur in the system.  Based on this we were able to determine sampling rates and averaging periods for several process and spectroscopic parameters to ensure we were able to trigger the tablet reject gate in a timely fashion with a minimum of false rejects.  The process capability of the system provides insight on whether the primary risk is special cause variation, common cause variation, or both.  This knowledge provides insight into the level of risk.         

  • How do you assure that materials are collected while the process in a state of control? How do you track the state of the process?

Importantly, Lilly does not equate state of control with steady state operation or collection decisions.  Collection decisions are made on statistically derived collection limits.   The values for the collection limits are different than the batch average limits to accommodate the higher sampling frequency while assuring compliance with USP <905>, if tested, as well as compliance with current batch average potency regulations.  Lilly’s primary means of control are the individual feeders and the active ratio control of those feeders. This is a level 1 control strategy with the real-time modelled concentration and the level 1 control strategy tablet weight as the controlled variables.  Other parameters are manipulated to achieve this.  Hence “variations in these (manipulated) parameters would not necessarily represent a departure from a state of control.” *  There may be slight start-up of potency which would not meet the traditional definition of statistical process control.  However, the collections limits determine collect reject decisions. 

  • How can the operator stay informed, in real time, about the state of the process?

The entire system, including an inline powder probe and an integrated at-line tablet tester, is controlled with an integrated distributed control system (DCS).  Alerts and alarms are established to warn operators of abnormal parameters.  Visualization is provided by multiple monitors showing both present state values and parameter trends.  A control room adjacent to the process suite allows scientists and engineers to track and monitor the process and oversee the tablet tester. While we have operated the process remotely from the control room, initially we plan to have operators in the process suite.  Because of the highly automated nature of the process and the need for instant decisions, most of the decision making is done by the automation system.

  • How do you assess and demonstrate the reliability of the process for the intended commercial runtime?

We have demonstrated the NCEs in development at approximately ½ of expected commercial run time.  We know tablet presses can run for long periods of time.  Since direct compression is a dry, non-thermal process, we have identified material accumulation as the largest risk area depending on the material properties of that component.  We have seen with other molecules that where material accumulation is a risk, it happens relatively quickly so shorter runs can be predictive.  The inline spectroscopy is located in the feedframe which is intrinsically self-cleaning. 


* Modernizing Pharmaceutical Manufacturing: From Batch to Continuous Production, Lee,S.L. et. al.