Biogen Idec, Wyeth Biopharma Create Value from Data

Jan. 13, 2010
Notes from the Umetrics User Group, courtesy of ARC Advisory

Over the last several decades, advances in automation and instrumentation have led to an explosion of data. Advanced analytics, such as multivariate analysis (MVA), can turn mounds of data into in-formation to address manufacturing performance. These solutions allow companies to identify relationships and acceptable, multivariate “operating envelopes.” In many cases, this information is useful for developing multivariate, causal models that support prediction of performance and qual-ity, process optimization, and on-line monitoring.

Advances in automation and instrumentation create a data glut. Advanced Analytics, such as multivariate analysis (MVA), can be used to turn mounds of data into information to address manufacturing performance. In addition, MVA is well suited for process analytical technologies (PAT) and quality by design, major components of the risk-based approach promulgated by the FDA.
The Umetrics User Group brought together numerous end-users, consultants, system integrators, and industry analysts to discuss how to improve operations through data analysis techniques

The Umetrics User Group brought together numerous end-users, consultants, system integrators, and industry analysts to discuss how to improve operations through data analysis techniques. Biogen Idec has been using statistical process control (SPC) systems for commercial and clinical cam-paigns since 2004. The rollout of Umetrics’ SIMCA Batch On-Line (SBOL) allowed the company to monitor and better control its manufacturing processes in real-time. The company’s successful rollout of SBOL is, in large part, due to its company’s culture, talented employees, and feedback from its partners. From Biogen Idec’s perspective, the implementation and use involves several dimensions including support from top management, training, lifecycle management, and accountability.

Wyeth Biotech uses SBOL for process monitoring of media sterilization, fermentation, centrifuga-tion, filtration, purification, and bulk fill. Through its initial experience with SBOL, Wyeth learned that the best way to avoid conflict between different groups is to get buy-in from stakeholders at project inception. Consequently, Wyeth developed a comprehensive strategy to implement SBOL that includes clearly defining roles and responsibilities. For Wyeth, the benefits of SBOL include improving process understanding, documenting batch evolution, saving deviations, identifying batch contamination, and saving batches from termination.

Analytics Drive Improved Operations Decision Making

To succeed in a dynamic marketplace, pharmaceutical manufacturers constantly seek to design and build higher quality into their products and processes, meet or exceed performance targets, and im-plement continuous improvements. They collect data on anything that might be useful to identify problems and highlight opportunities for improvement. Proliferation of IT, automation systems, and analytical equipment has further encouraged the buildup of data on all aspects of the products, processes, equipment, and operations.

The key to success is not to distribute more data, but to analyze the data and distribute timely in-formation. Staying competitive means transforming plant data into information and making that information available in context to all personnel involved in operations. This enables them to make better decisions that generate true value. It also means adopting methods to better monitor and con-trol processes in real time.

The growing popularity of advanced analytics software reflects the buildup of plant data. The ability to graphically present and drill down into information has become an essential capability of plant floor and business applications. Advanced analytics, such as MVA, allow manufacturers to improve their operations and competitiveness by transforming data-rich systems into valuable information to glean deeper understanding, make better decisions, and improve process monitoring and control.

The major applications for MVA include process design, offline data analysis, and online process monitoring and fault detection. In addition, the FDA has adopted a risk-based approach to promote advanced manufacturing practices within the pharmaceutical industry. Process Analytical Tech-nologies (PAT) and Quality by Design (QbD) are major components of the risk-based approach promulgated by the FDA to encourage pharmaceutical companies to adopt the latest manufacturing techniques and practices. MVA is well suited for designing quality into the process using DOE and monitoring the process by making timely measurements of critical quality parameters during proc-essing to assure acceptable end product quality at the completion of the process.

2009 Users Group Forum Highlights

The Umetrics User Group brought together numerous end-users, consultants, system integrators, and industry analysts to discuss how to improve operations through data analysis techniques. Meet-ing highlights included presentations on topics ranging from offline data analysis to online measurements and closed loop control. The session concluded with a lively integration panel discus-sion by industry experts.

Analytical Measurement Proliferation Enhance PAT Initiatives

Nouna Kettaneh and Svante Wold from NNS Consulting discussed how to use online measurements in biotech PAT. Nouna indicated that advances in optics, electronics, micro-fabrication techniques, and chemometrics usher in new developments in analytical instrumentation. More affordable, ro-bust, and reliable instruments can be operated in more demanding environments without expert supervision. Many new instruments are portable and accommodate a wide range of sample types, operate online, and approach the sensitivity and selectivity of laboratory equipment.

With the proliferation of new analytical instruments, Nouna stressed the importance of measuring as many different variables as possible to improve product quality and process understanding. This philosophical switch from using devices constructed on the basis of a priori knowledge, to using data analytical devices where information content is assessed after the data collection, leads to the discov-ery of unexpected results, unknown events, and novel properties.

Svante Wold discussed the principles behind several types of analytical equipment such as mass spec, NMR/MRI, and chromatography. He also discussed their limitations in concentration meas-urement range and speed along with their practical areas of application. Svante showed how to calibrate instruments and determine desired variable values from indirect analytical raw data measurements.

In complex processes, the measurements from each source can be used separately for a specific pur-pose such as modeling the spectroscopic or chromatographic data to screen and classify raw materials as acceptable or unacceptable. Integrating different process measurements with process variables so that a model can use all possible information creates additional value. However, as the number of variables increase and the model gets larger, it becomes more difficult to obtain and in-terpret results. It is possible to use less variables, but at the expense of accuracy.

A better alternative is to divide the variables into conceptually meaningful and more manageable blocks – and use a hierarchal modeling approach. In process modeling, the different blocks can rep-resent different steps of the process such as raw material preparation, conversion in reactors, and separating and purifying the product. Variables measured on different steps constitute natural blocks. These may be further divided according to the type of variables, e.g., temperatures, flows, pressures, concentrations, etc.

The hierarchal model consists of a super (upper level) and sub (lower level) model. The super model shows which blocks (steps in the process) dominate the model components. The sub model indicates which variables have the largest contribution to each particular step. The hierarchal modeling ap-proach has many applications including predicting important output values (such as final product quality) and online process monitoring for each block or process step. The multi-block method makes it simpler to model complex processes and interpret the data. As more analytical measure-ments become available, multi-block hierarchal modeling will play an increasingly important role in process modeling and monitoring.

Biogen Idec’s Culture Ideology Supports MSPC Realization

Biotechnology company, Biogen Idec, Inc., specializes in drugs for neurological disorders, autoim-mune disorders, and cancer. The company was formed in 2003 when Cambridge, Massachusetts-based Biogen Inc. merged with San Diego, California-based IDEC Pharmaceuticals.

Seongkyu Yoon, William Lichtman, and Sophia Koutoulas talked about their experience with Umet-rics’ software at Biogen Idec. Biogen Idec began using Umetrics’ MVA and multivariate statistical process control (MSPC) tools for clinical and commercial campaigns around 2004. Initially, scientists and engineers used the tools; however, the company realized the tools also had potential for every-day use by manufacturing personnel. MVA and MSPC have the power to provide insight into what is going on at both the equipment level and from a cell perspective.

Biogen Idec embarked on an ambitious, methodical plan to roll out Umetrics’ MVA and MSPC solution, SBOL, to nearly every process unit operation. These range from cell cultivation to cell broth clarification and purification. This type of endeavor is especially challenging in a regulated environment. Companies must have a lifecycle plan to maintain the system and update the models as well as train its people on using the system.

Before implementing SBOL in manufacturing, monitoring was done through batch records (volumi-nous) and alarming. Most SPC functions were done manually or semi-automatically using chart recorders. The process was tedious and error prone and did not provide process insight or instill as sense of ownership among operators. For example, during each shift, operators would manually record process variables for each batch at a predefined interval. If a process variable, such as dissolved oxygen, changed during operations but no alarm went off, operators assumed everything was fine since they did not have the necessary process knowledge to understand a trend leading to a process deviation prior to an alarm state.

During purification, operators and engineers would use chart recordings to analyze batches, but this was often done after production and was subject to errors due to extracting data directly off the charts using pencils and rulers. The company used to have to compare paper records of chroma-togram after the batches were made. The lack of organized electronic data made it difficult to compare process chromatograms from multiple batches and extract meaningful information.

Prior to SBOL, operators would have to go to engineering to see if the data was telling them some-thing was wrong. The rollout of SBOL to manufacturing has allowed the company to better monitor and control its manufacturing processes in real time. SBOL allows operators and manufacturing per-sonnel to improve their process understanding and empower them to take greater ownership of the process. SBOL also provides a means to compare historical and current batches to be able to make more knowledgeable decisions to improve the process.

The company’s successful rollout of SBOL is in large part due to its company’s culture, talented em-ployees, and its technology partner, Umetrics. From Biogen Idec’s perspective, the implementation and use requires several dimensions including top-level management support, lifecycle support, training, and accountability.

Top management views the system as essential and provided resources and incentives to implement and maintain the system. Management expects every organization to use the system and understand what the data means.

Biogen Idec requires the system to be highly visible and available 24 hours a day. IT and engineering support the system and if something goes wrong, it gets fixed immediately. The company has 48 monitors scattered throughout the plant to make observing the progress of processes part of the cul-ture. Personnel can also get access to the system remotely. Every day, each shift publishes trends and reports.

It’s important to train plant personnel on how to use SBOL to ensure proper use to create value. Change management and control is also important, particularly for updating models and ensuring regulatory compliance. The company has developed SOPs on how to use and maintain SBOL.

Management expects everyone in Manufacturing to contribute to building models by identifying important variables and factors to include as well as weighting the variables in terms of importance. Management also expects manufacturing to take ownership of the process and empowers it to use the data to limit process upsets.

To date, the SBOL implementation has been highly successful and the company has plans to use other tools to design quality into the process.

Implementing SBOL in Biotech Facility at Wyeth

Jeff Doyle explained the use of SBOL at Wyeth Biotech. Wyeth is one of the largest research-driven pharmaceutical and health care product companies in the world. Several sites around the world use Umetrics’ SBOL including the site in Andover, Massachusetts. At this site, the company uses SBOL for process monitoring of media sterilization, fermentation, centrifugation, filtration, purification, and bulk fill. SBOL has demonstrated the ability to improve process understanding, document a re-cord of batch evolution, save a deviation, identify a batch contamination, and save a batch from termination.

Wyeth Biotech began developing the Andover SBOL program in early 2008, however without a for-mal implementation strategy. The Informatics group at the site was responsible for getting SBOL up and running. During the initial implementation testing of SBOL, Informatics created a highly refined fermentation model. Within the first hour of being put online, the model identified specific process issues. Unfortunately, Informatics had not yet made Manufacturing aware that SBOL was in use and had not yet put a notification or response plan in place. When Informatics informed Manufacturing of the process issues, Manufacturing got the impression that Informatics was spying on them. Manu-facturing quickly coined the term “spy-ball” for SBOL.

To rectify this situation, Informatics developed and implemented a notification and response plan using input from the Manufacturing group. Informatics would perform monitoring through the test-ing phase and continue to refine models. The company established SBOL training for Manufacturing. This included a brief introduction to multivariate statistics, plus training on how to interpret common charts, how to use the tool and interface, and what the results mean. Informatics made SBOL available to Manufacturing and installed large screen monitors to increase visibility.

Learned from its previous implementation, Wyeth developed a detailed implementation strategy for SBOL. The plan includes developing a vision, site assessment, and specific order of events. Wyeth will use SBOL to identify non-comparable conditions and mitigate the risk of a batch failure. The site assessment is necessary to determine the data storage and retrieval architecture, available network and server resources, and available resources for modeling and process monitoring.

Wyeth structured the implementation events to ensure that prior to turning the system over to Manufacturing: 1) the required architecture is in place, 2) SBOL is connected to the process correctly, 3) the models are built, and 4) functional testing and training have been performed. The strategy also defines clear responsibilities between Informatics and Manufacturing.

MPC Using Principal Component Modeling

Chris McCready talked about model predictive control (MPC) and principal component modeling development efforts at Umetrics. MPC, which was designed with the continuous process in mind, finds the best set of operating parameters that result in the desired material quality using a process model. Traditionally, MPC not well suited for batch processes and, hence, has not been utilized by the pharmaceutical industry. One issue is that MPC pushes the system into new operating regions. For pharmaceutical processes, MPC needs to be constrained to the design space. Umetrics developed a model predictive multivariate controller (MPMC) that finds the set of operating parameters within the design space that results in the desired material quality. This controller can be applied to both continuous and batch processes.

Last Word

Over the last several decades, advancements in automation and instrumentation have led to an ex-plosion of data. Advanced analytics transforms data into information that provides deeper understanding of process operations and improves decision-making.

Umetrics users create value from data in both online and offline capacities. The company’s software and solutions support continuous improvement efforts. Users in the pharmaceutical and biotech in-dustries in particular exploit the company’s software and solutions because these support PAT. In addition, the company’s MPMC is a new model-based predictive controller that is well suited for batch applications, particularly in a regulated environment. ARC believes MPMC offers significant benefits to users in the batch-oriented industries. These industries lack the robust MPC solutions commonly implemented in the continuous process industries.

sidebar: Advances in automation and instrumentation create a data glut. Advanced Analytics, such as multivariate analysis (MVA), can be used to turn mounds of data into information to address manufacturing performance. In addition, MVA is well suited for process analytical technologies (PAT) and quality by design, major components of the risk-based approach promulgated by the FDA.
The Umetrics User Group brought together numerous end-users, consultants, system integrators, and industry analysts to discuss how to improve operations through data analysis techniques.

About the Author

Tom Fiske | ARC Advisory Group