Amgen’s digital transformation: linking raw material data from suppliers to patients

By Cenk Undey, Executive Director, Amgen

In drug development and manufacturing, variability in raw materials presents high risks to the success of our processes and our products. When raw material variability issues arise, manufacturers and suppliers must address them immediately by analyzing data and determining the root cause. Sometimes these issues can be fixed quickly, and other times they can be more serious, creating a significant setback to your development and manufacturing timelines. This is why understanding, monitoring, and controlling raw material variability across the biopharmaceutical value chain from raw material suppliers to patients is critical. However, doing so depends on how a company’s raw material data is collected, shared, and analyzed.

At Amgen, a Supplier Relationship Excellence (SRE) program was designed to open the lines of communication with suppliers and create a feedback loop where data can be shared to better understand operational performance. In doing so, Amgen has advanced several digital and predictive technologies across the value chain that have contributed to the success of SRE and the exchange of electronic data (eData) with suppliers. This includes establishing smart contracts, developing data exchange standards, using predictive models to anticipate issues or identify improvement opportunities, and leveraging artificial intelligence tools and technologies.

While the bio/pharma industry has historically lagged when it comes to digital technologies, adopting a digital transformation such as this for the collection and analysis of raw material data, along with other critical data throughout the value chain, could modernize drug development and potentially revolutionize patient care.

Standardization of raw material eData exchange

At Amgen, it is estimated that 500 million continuous data points are generated per product from manufacturing equipment used in therapeutic protein production. In research and development, one of the robotics-based drug candidate screenings generates an additional 200 000 data points per day. Much of this data comes from raw materials, as many raw material components are used at each stage of the development process. Traditionally, this information is stored and exchanged using Excel® spreadsheets, proprietary databases that are hard to access, or other paper-based systems. When a raw material issue arises, process developers use this information to look for possible trends that might help identify where the problem exists, which is very time consuming.

Rather than continue to operate in a reactionary way to raw material variability issues, Amgen Operations decided to create an in-house, validated information system, called Raw Material Information System (RMIS), where we could document and store raw material data. While it allowed for a more organized way of collecting raw material data, we wanted to explore ways we could use the system to increase supply chain visibility and anticipate/reduce variability issues. If we could achieve that, it would set up a quick mechanism for feedback between us and our suppliers that could facilitate troubleshooting and offer continued learning opportunities. One major challenge of this system, however, is that there was no standard file format for data that would allow a seamless data exchange between suppliers and users.

To address this, our team worked with one of our suppliers to develop a data file format based on scientific data exchange technologies, which we eventually presented at an annual Pharmaceutical Process Analytics Roundtable (PPAR) meeting. PPAR includes 30 representatives from across the industry interested in the current state and future direction of process analytical technology (PAT). The PPAR group encouraged our team to initiate an industry-wide effort to create eData exchange standards that would allow for the systematic study of raw material variability. The result was a group that consisted of 12 major pharma and biopharma companies and nine suppliers, who discussed the current plan and identified areas for improvement. The format was then documented in an article that also explained how it operates. We later submitted it to the American Society of Testing and Materials (ASTM), and in less than two years, it became a published standard that can be used as a reference guide for eData exchange. With this standard in place, it provides a common data structure and ease of accessibility of all other related processes and product information. In the end, facilitating the exchange of data across an entire supply chain allows a more complete understanding between the supplier and manufacturer about the impact of raw material variance and what it means to the process or product performance.

Identify, track, and control variation (ITCV) and multivariate analysis

ITCV is a framework used by Amgen that applies statistical process control to follow trends in data over time, using either one variable at a time or multivariate analysis, and to identify possible areas of variability. If any issues are identified, they are escalated to the supplier, so it can take actions to mitigate the risk, if possible. The action(s) could be at the supplier, at Amgen, or at both. For example, our team used ITCV to measure 52 variables across 42 lots of different raw material against in-house performance data. While doing so, we noticed two statistically distinct clusters of data. When comparing them across the 52 variable measurements to see which variable was causing the separation of data, we found the clustering was related to a difference in the raw material coming from a plant in one geographic location versus another. Although the systematic differences between the two sites did not have a significant impact on Amgen’s operational performance, it demonstrates the ability of the system to detect potential variability issues in their infancy. Figure 1 shows the path of our analysis and how we came to our final conclusion.

Amgen digital transformation

Fig 1. Weak signal of a raw material variability shows differences between two sites of the same supplier; Amgen performance data does not highlight an operational difference between the raw materials from either site.

Another example of how ITCV helps us understand, reduce, and increase control of raw material variation is related to an issue discovered with a legacy product at one of our commercial manufacturing sites. A raw material was posing high variability and risks to our manufacturing performance. To mitigate that, we were historically running bench-scale experimental models using raw material samples to mimic actual commercial operation. Each sample would be run through a bench-scale model in the process development lab to determine how much, if any, risk it poses. The turnaround time for sampling, running experiments, and analyzing these materials is 80 hours. After years of amassing raw material data, our team ultimately established a sizeable database of raw material information. We were then able to build a predictive multivariate model that uses data readily available from certificates of analysis and prior to drug substance manufacturing to provide accurate expected impurity results from the pertinent step in less than an hour. With the click of a button, the model provides the same information gathered during a lengthy laboratory analysis. This gave our team more time to work on more important process improvement efforts.

Mobile handheld technologies to capture data and enable rapid analyses

To reduce the amount of time used for lot release of raw materials, some companies are using handheld Raman and NIR spectroscopy technology, which uses Raman laser or near-infrared light for rapid multi-component analysis. Our team realized that while this on-site identification of raw material is occurring, the handheld device is simultaneously gathering valuable data that could be useful for mathematical modeling and statistical trending about raw material variability. If this information is collected early in the life cycle of drug development, it creates an opportunity to produce a library of raw material variability. In addition, we also began to think, “How can we enrich this data, so we can use it for performance management and variability issues?” To be predictive in a proactive way, or as the team at Amgen calls it, “preactive,” we began to use the data from the handhelds for computational modeling to study how a raw material might impact process product performance. In some cases, the same raw material might be used for multiple products, but each product might be impacted differently. Nevertheless, with “preactive” knowledge, we can optimize the route per product for the best possible outcome.

Computational modeling

While we made strides toward monitoring and controlling raw material variability via data-driven and empirical modeling techniques, those techniques are limited to the ranges of the variables measured. We then started thinking about how we could expand our work in leveraging first principles of computational models into explaining raw material impact into our processes. Much of the current modeling techniques have roots in engineering, biology, chemistry, physics, and fluid mechanics. However, advances in computational power and scientific understanding are enabling more modeling of biopharmaceutical process development and manufacturing. The key is to form a mathematical model that will mimic the process, equipment, and raw materials and identify anything that might contribute to performance variability. If and when those signals are found, we can determine if there are any levels in our process design and control that might compensate for the disturbance. It is early for these advanced models to be effective in understanding raw material variability. However, we believe there is a significant potential to explore wider parameter ranges in silico.

Watson Explorer Content Analytics

To achieve even better insight, Amgen began to look into advanced artificial intelligence tools, such as Watson Natural Language Processing (NLP). This system is capable of searching massive amounts of data very quickly. At Amgen, we have over 21 different source systems with close to 5 million records (and increasing every day). Watson is capable of reading into these documents and providing sentiments and correlations that are non-numerical (i.e., analyzing text).

Recently, during an inspection at one of our facilities in Singapore, the regulatory inspector requested documentation justifying our decision to check for a raw material used as an excipient in the manufacturing process. We were able to use the Watson system to search across 21 source systems for that specific excipient. Within 10 minutes, 42 documents containing the two query items were made available to the inspection support team. What would have previously taken hours to accomplish was done in minutes.

This system not only allows us to understand raw material variability but also creates feedback loops from patient experiences. Any information from patients that is made available in our complaints databases is connected to the Watson system, so it can perform text analytics on those freeform sentences. It then reads the information and applies natural language understanding to return related results. We can then trace this information and determine if we see any trends that might call our attention to an issue in, for example, a specific region. Not only does this offer many benefits to the business of drug development and manufacturing, but also, most importantly, to the patients it serves.

While Amgen continues to seek ways to better understand raw material variability, it is important we share our lessons and experiences. By collaborating openly across the industry, we can minimize and control raw material risk and better manage our supply chain for the purpose of delivering the safest and most effective drugs possible.

Next steps

The future trend will be to leverage the aforementioned data exchange ASTM standard and increase investments in data infrastructure at suppliers and manufacturers to drive towards more seamless data integration. Through the use of eCoA and potentially direct connections to supplier manufacturing sites, it will be possible to have near real-time automated detection when a raw material results in process and/or product variation.

Acknowledgements

Article written by Cenk Undey, Executive Director, Amgen. The author would like to thank Amgen colleagues who have reviewed and contributed to this article. Special thanks to Patrick Gammell, Myra Coufal, Ting Wang, Tony Wang, and Sinem Oruklu.

References

Wang, T. et al. An electronic format for data exchange between raw material suppliers and end users enabling superior knowledge management. Pharmaceutical Engineering 35,71–78 (2015).