Moving Online Blend Uniformity from Development to Production
How to identify candidates For PAT Using NIR
Scot Ellis, NIR Product Manager, Thermo Fisher Scientific
Many sub-processes and critical control points have been identified in the tablet manufacturing process as being good candidates for Process Analytical Technology using near infrared (NIR). As the philosophy shifts from process understanding to process control and further to Quality By Design (QBD), pharmaceutical manufacturers must move from investigative to practical implementation of on-line and in-line analytical technologies. This means moving analytical tools from the domain of researchers and into areas where minimal expertise and support are available, in order to operate, interpret, and support them. It also means making a migration from intellectual equity to measurable return and economic benefit. This article discusses some considerations and requirements for making this practical move, using the pharmaceutical powder blending sub-process as an illustrative example. It borrows from the experiences of both technology vendors and drug manufacturers who have worked together to start making the “practical migration”.
The broad goals of using NIR on a powder blending system are to determine when sufficient blending has been achieved and to verify potency and composition at completion. The near infrared spectrum of the powder can be used to identify and fingerprint the state of the powder passing through a point in a tumbling blender at a point in time, and by monitoring the variation in this fingerprint the rate of change in composition (from unblended to uniform) can be inferred. Through algorithmic interpretation, relative concentrations, such as that of an active pharmaceutical intermediate, can be estimated. This information can be used for active control as well as product verification either manually or automatically. The implementation of an analyzer that can accomplish all of these tasks can be viewed in terms of three layers of integration: a physical layer; a data or “brain” layer; and a linking communication channel that allows the analytics to work in harmony with process automation and quality control.
The physical integration layer
In order to take a NIR spectral measurement, the system must be able to illuminate a powder through a window in the blending bin, re-collect light scattered from the sample and route it to a detection system. A variety of schemes have been developed by analyzer suppliers to accomplish this measurement. Firstly, a NIR-transparent and robust window must be installed into the blender bin. Such windows are usually sapphire and are mounted in the lid of the blend vessel, as modification of the bin body is not often feasible from a technical or validation standpoint. The second requirement is that the analyzer must be physically attached to the blender bin. Different designs take different approaches; simplicity and mobility are highly advisable and the ability to move an analyzer easily from one blending vessel to another will create economy of implementation. Systems that are lightweight and have simple mounting schemes can also be used in research and product development, which typically use small, laboratory scale blenders, as well as larger scale industrial operations. Implementing this technology across a broad range of scale using a single process analytics platform allows for greater analytical support and knowledge utilization, and will reduce the deployment burden on process engineering and manufacturing scientists. Another consideration is the referencing or “background” ratio that spectrometers reply upon to isolate instrument effects from the sample. The ability to run this periodic setup step automatically and without removing the analyzer from its blender bin mount eliminates many logistical problems in a clean room area and effectively removes one degree of complexity frequently associated with molecular spectroscopy in general.
The Data Layer
Often considered important, but secondary to the physical integration characteristics, a blend uniformity analyzer will have a collection of firmware and software that control data acquisition, processing or interpretation, data storage, and display. Data acquisition for the online blend analyzer will have two complicating elements not seen in many other process analyzers currently deployed in solid-dosage production: wireless communications and data triggering based on a physical parameter. As with most technologies brought into an area for the first time, there exists a certain mistrust of the dependability of wireless communications, but wireless technologies can be as reliable as wired communications today. Data backup systems can be deployed on an analyzer to overcome the perceived risk but at the expense of weight, simplicity of mounting, and overall form factor. Utilization of a blend analyzer for real-time monitoring and control by either human or control system will depend upon wireless data transfer so it is generally advisable that implementation goals not be compromised. The second complicating element, data triggering, stems from the fact that the NIR measurement is only useful while collected when the powder being processed comes into contact with the measurement window, when that window is at the bottom of the rotation. Solid state electronic trigger technologies are available that require no physical calibration and allow a collection angle (of rotation) to be set in the NIR analyzer controlling software. This is a tremendous benefit where blends use a variety of fill levels as the optimal collection angle may vary with different blends.
The real key to deriving commercial benefit from a NIR blend uniformity analyzer lies in collecting information in real-time about the progression of a blending process and the concentrations of the materials in the blend. Once validation of an analyzer system is complete, manufacturing personnel may not be concerned with how this information is gleaned from a triggered NIR spectral measurement, but this is the pivotal part of implementing an analyzer as something beyond a research tool. Near infrared data primarily yields chemical information but can also be influenced by physical properties. Interpreting this data involves the correlation between spectral change and real world material changes. In a complex mixture such as a pharmaceutical powder blend, this requires either multivariate calibration or statistical evaluation (or both) to yield meaningful information about a process. A good NIR blend monitoring system will have a captive method development or chemometric modeling environment that allows models to be built and ultimately embedded into a workflow or operating recipe that can be run by someone other than a scientist or statistician without extensive training, or even automatically by a control system. However, the development of interpretive models require more than software. The spectroscopy or statistical expertise often concentrated in product development may be needed, as will process knowledge for existing processes. Making use of cross-functional resources is necessary to develop and employ a suitable calibration for a particular drug product.
If data acquisition and its conversion to useful information about the process are the engines of the data layer, the operational interface that handles them and feeds the information to the operator or a data system is the vehicle that surrounds it. Implementing process analytics requires both flexibility (for example, for different processes, or for decision making during a process) and tight control (for strict controlled operation in a cGMP process). Making a move away from scientific software is a step in the right direction, but the process engineer must consider change control, adaptability for different or changing processes or operators, and the ability to check performance and qualification status.
The communications or link layer
A third major layer of practical blend analyzer implementation is the connection between the analyzer data system and any outside data systems, which might include a process data historian, a control system, or a simple local controller. This is the part of the blend uniformity analyzer system that receives the least consideration during the equipment specification stage, and in an off-the-shelf package, the analyzer supplier is assumed not to play a role in this stage. It is also this very layer that can accelerate economic return and provide ongoing measurable benefit to the operations of a company. By making the connection between the analyzer and a control system, multiple benefits are fully realized. A model that calculates a rolling standard deviation of the data on a blend can communicate blend completeness and trigger blending to stop, preventing over-blending and reducing cycle times. The same model could detect and trigger an alarm for problem blends, create incident records automatically, or dial the cell phone of a QC manager. Final concentration information can be automatically written into manufacturing system batch records for product verification that could eventually help build a case for real-time release.
But an analyzer system must be able to receive communications as well as sending them. A blend analyzer should have full input/output capabilities to synchronize runs and become an integral part of the manufacturing process, rather than an extra tool that operators must take the time to use. Any project that implements a blend analyzer should consider both hardware connections and software tools ready to communicate via standard manufacturing protocols, which might include OPC (OLE for Process Control), 4-20 mA, and digital relay communications. The absence of these tools in an analyzer program could mean a significant investment and delay to integrate analyzer software with quality and control systems.
The requirement to link the needs and functional expertise between development and manufacturing makes a strong case for a small, simple analyzer that can span the gap between lab and plant scale. Fortunately, equipment suppliers have been evolving their products in consideration of these requirements and systems are now available that meet many of the needs without customization.
In addition, the size of a project to implement a NIR analyzer for a step in pharmaceutical batch manufacturing can always be reduced by proactive risk management, part of which falls into careful design specification. In the case of blending, it is helpful to consider the implementations in layers that progressively collect and move information from the powder to the operations that are producing it. Each layer adds value to the information and subsequently to the operation, while increasing the likelihood that problems will be trapped or prevented. Combining each of these three components necessarily requires different expertise, but linking them together can expedite the process and ensure a strong chain of information. Outsourcing as much of these layers as possible from a minimal number of providers can be a good idea but requires careful selection of equipment and the consideration of needs other than the fit and performance of the spectrometer itself.