Privacy settings

Utilizziamo cookies tecnici, necessari per il corretto funzionamento del sito, cookies statistici e cookies di marketing (inclusi cookies di terze parti), al fine di mostrarti offerte e pubblicità basate sui tuoi gusti e preferenze. Per maggiori informazioni puoi consultare la nostra Privacy Policy. I cookies statistici e i cookies di marketing (inclusi cookies di terze parti) sono utilizzati esclusivamente con il tuo consenso. Cliccando “Seleziona tutto e acconsenti” consenti il trattamento dei cookies statistici e di marketing (inclusi cookies di terze parti).

Per personalizzare le tue impostazioni privacy, seleziona le tue preferenze tra le caselle qui sotto e clicca “Applica la selezione”. Ti ricordiamo per puoi modificare le tue impostazioni privacy in qualsiasi momento e disattivare i cookies, seguendo le istruzioni riportate nella Privacy Policy.

Show preferences Select all and accept
IMA is loading

A Data Driven Approach to Informed Lyophilization: from Multivariate Analysis to Data & Predictive Analytics.

A History of Data usage in Aseptic Manu-facturing

Since the mid ‘80s and into the early ‘90s when the data captured using a chart recorder was considered state-of-the-art, it should be of no surprise to the reader that a significant amount of data is generated in Pharmaceutical development and manufacturing unit operations. Most chart recorders generated data on paper. The paper was passed under a pen which deflected based on the signal. While the use of chart recorders has slowly phased out with the ability to store large amounts of data inexpensively, the analysis of the data and its use in process understanding gained importance. Further, with the guidance issued by the FDA (U.S. Department of Health and Human Services Food and Drug Administration, 2004) around the importance of Process Analytical Technologies (PAT) in risk-based approach to manufacturing, came the need for data analysis and interpretation.

Figure 1. Typical operations in lyophilized drug product manufacturing

A typical freeze dryer alone, one of 8- or 9-unit operations in a typical sterile fill-finish line today can have in excess of 150 sensor tags recording data at 1s intervals. This results in about 150 billion data points over the life of a typical equipment. Not only does this data need to be archived or stored for rapid queries, it also requires periodic analysis. The interpretation of this data is becoming increasingly difficult. While univariate data analysis is useful, multivariate data analysis (MVDA) has gained in popularity over the last several years. MVDA has allowed handling of such large data sets and establishing correlations among heterogeneous data. For many, this has become the backbone of risk-based root-cause analysis. The approach has further enhanced process understanding through statistical analysis and helped decision making relevant to process development, monitoring, control, product life-cycle management and sometimes for product release through continued process verification (U.S. Department of Health and Human Services Food and Drug Administration, 2011). Of particular importance is the application of MVDA towards process optimization as applied to Quality by design (U.S. Department of Health and Human Services U.S. Food and Drug Administration, 2007).
At its core, MVDA tools are designed to provide a dashboard to look for deviations and understand the relationship between different data sets/data points. However, this typically relies on analysis performed by an advanced user. Furthermore, such data analysis is descriptive since it is based on past events. It does not predict the impact of a change in a variable. This naturally leads into predictive analytics. As the name suggests, the goal of predictive analytics is to predict the outcome of a process based not only on past data and events, but also on human feedback, closing the loop on testing and validating predicted outcomes for learning the process and equipment. One example of when such learnings are crucial are changes in a system before and after a maintenance operation, closing the feedback loop allows learning the current status and applying towards future predictions. Making this feedback loop autonomous using Artificial Intelligence techniques provides a predictive analytics framework backed by machine learning.

 


 

The Role of Process Monitoring and Process Control

Process monitoring and process control are an integral part of a pharmaceutical manufacturing process today. Both have formed the basis for ensuring a controlled state of operation and controlling batch to batch heterogeneity. For example, while a capacitance manometer-based pressure gauge is used for process control in the product chamber, a Pirani pressure gauge may be used solely for process monitoring. The PID (proportional-integral-derivative) loop is used to control the chamber pressure based on capacitance manometer reading and using a proportional control valve to bleed dry nitrogen gas into the chamber. This approach for using a process monitoring and control assumes that the system does not suffer from any unexpected departures from standard operating conditions. For example, lack of supply pressure in the dry nitrogen. The addition of statistical process control techniques to augment the use of sensors can aid in root-cause investigations. However, the premise behind such statistical control is typically a reaction to a failure, which can sometimes be catastrophic and expensive.

Figure 2 shows an example of this where incorrect tuning of this process control loop causes pressure fluctuations in chamber pressure which may adversely impact product quality and stability. Thus, the ability to predict such occurrences with enough lead time to failure is crucial in data driven manufacturing. 

Figure 2. Importance of Process Control highlighted through a pressure fluctuation event


 

Use of Artificial Intelligence in the Pharmaceutical Industry

The adoption of AI in Pharma has grown in the past decade in 3 broad categories, a) Drug design, b) treatment outcome and c) data modeling. One example of its use in drug design is application towards identifying disease sources from hundreds of genes and finding new therapeutic opportunities. Yet, there are examples of predicting drug outcomes based on predictive modeling making it faster and more affordable to treat specific diseases. Patient compliance is a major challenge to pharmaceutical companies. It is not uncommon to find adherence rates of about 60-70%. AI is also being piloted in the improving drug adherence among Schizophrenic patients. While evidence of such development indicates higher adoption rates of AI in drug development and pre-clinical work, there has been little adoption in fill-finish and manufacturing operations. One example of its application in fill-finish operations is in AI/deep-learning based image processing for acceptance/rejection of products/containers based on prior data/knowledge. Here we discuss the application of the technology towards lyophilization.

 


 

Current Monitoring and Maintenance Practice in Aseptic Manufacturing

Manufacturing in the aseptic pharmaceutical industry and lyophilization operations is achieved through production campaigns, each with a target output. In such an environment, efficiency and scheduling of resources become a critical aspect of maintaining the required throughput. Thus, optimal equipment operation and maintenance is important to ensure product and process performance. However, there are several challenges in maintenance and management of equipment resources.
It is common in the pharmaceutical industry to ensure healthy operating conditions in manufacturing equipment by subjecting them to scheduled preventive maintenance and calibration routines. While this satisfies the need to maintain equipment, the underlying risk of scheduled preventative maintenance practice is often over maintained systems, and sometimes, even leading to under maintained systems. Not only does preventive maintenance have a significant cost, it can also fail to account for sudden changes in equipment performance.


Operational AI in Lyophilization

Today, Industry 4.0 applications in every major sector have been driven by a business needs and technology maturity. From the lens of aseptic manufacturing and lyophilization, the progression of equipment and process understanding over the past few decades can help lay out a clear path forward. The earlier freeze-drying methods involved in the manufacture of products such as blood plasma were conducted in freeze dryers made primarily out of aluminum. Such equipment then evolved to add sterilization and stoppering. During the last two decades, we have seen the use of automatic loading systems, process analytical tools and a lot of process and product understanding go into freeze drying. The data being collected by a growing number of sensors can now be analyzed with advanced Machine Learning and AI methods. Building on top of multivariate analysis, real time algorithms can be deployed to allow for responding to maintenance and process monitoring needs.


Next Steps for Freeze Drying 4.0

Advancements to current methods for aseptic manufacturing and lyophilization should be based on an understanding of the process and equipment. Multivariate methods for data analysis have enabled process R&D engineers and scientists gain more insight into the manufacturing operation. To determine next steps, we must try to image what aseptic manufacturing of the future might look like. Imagine a control room with equipment and process experts analyzing the overall system health and providing valuable feedback to manufacturing stakeholders.

Figure 3. Evolution to Pharma 4.0 seen with respect to Freeze Drying

Figure 4. Example of an architecture for implementation of Freeze Drying 4.0

Figure 5. A step wise progression from Descriptive, Diagnostic, Predictive to Prescriptive Analytics

 

An architecture relying on performing complex calculation on a cloud-based infrastructure can be deployed. An example of such a method applied to freeze dryers is show in Figure 4.
If an ambitious goal of prescriptive operation with respect to aseptic manufacturing including freeze drying is kept in mind, then a methodical approach is required to achieve it. The first steps are rooted in an understanding of the process and equipment gained through first principle and multivariate data analysis methods. Figure 5 shows an approach laying out the next steps where ML and operational AI methods will be used for failure predictions, maintenance scheduling and pattern recognition. An ongoing collaborative effort will be required to validate and gain confidence in the outcomes of such methods following which prescriptive analytics can be enabled.
You can access further literature at lab4life.ima-pharma.com

References

[1] U.S. Department of Health and Human Services Food and Drug Admini-stration, Guidance for Industry PAT – A Framework for Innovative Pharma-ceutical Development, Manufacturing, and Quality Assurance, September 2004.
[2] U.S. Department of Health and Human Services Food and Drug Admini-stration, Pharmaceutical Quality for the 21st Century A Risk-Based Approach Progress Report, May 2007.
[3] U.S. Department of Health and Human Services Food and Drug Admini-stration, Guidance for Industry Process Validation: General Principles and Practices, January 2011.
[4] Frauendorfer, E., Wolf, A. and Hergeth, W. D.Polymerization Online Monito-ring, Chemical Engineering & Technology 33 (11), pp. 1767-78 (November 2010).
[5] Miletic, I., Quinn, S., Dudzic, M., Vaculik, V., & Champagne, M. An industrial perspective on implementing on-line applications of multivariate statistics. Journal of Process Control, 14(8), 821-836. (2004)