Umetrics Suite Blog

Interpretation for the use of multivariate models within the FDA draft guidance for continuous manufacturing

April 24, 2019

Recently the FDA issued a new draft guidance for continuous manufacturing of small molecule drugs. With these draft guidelines the FDA wants to engage more pharmaceutical manufacturers to shift from traditional batch/start-stop processing to continuous manufacturing. The main advantages associated with these processes are more room for modularity, automation and flexibility due to a smaller footprint, but also more consistent quality of the drug product. Of course the main incentive for the FDA to promote this way of processing is that it believes that this will have a positive impact on drug prices and prevent drug shortages.


To date four marketed drugs are produced using a continuous manufacturing process, but more and more are in development. Although the now published FDA draft guidelines are aimed at small molecule drug processes, this way of processing also gets a lot of attention from the biopharmaceutical industry for all the same reasons.  

One of the enablers of continuous manufacturing was the PAT framework  within the industry set-up in 2004. In continuous manufacturing it’s key to have ongoing monitoring of the process to confirm that it remains under a state of control. The advances in the PAT field make it possible to move measurements to the point of control and assess product quantity and quality in real-time, thus facilitating Real-Time Release Testing (RTRT).

In section III.B.2 and III.B.4 of the draft guidelines it is described how models can be used for active process control and RTRT as part of the overall control strategy. Calibration models for Near Infrared Spectroscopy (NIR) are already applied in this way within the industry, where NIR is used both as In Process Control (IPC) method as well as Real Time assessment of product Critical Quality Attributes (CQA), such as water content or dissolution. 

Another application of models within RTRT is the use of Multivariate Statistical Process Control (MSPC). With this method models are used to summarize multiple univariate sensor outputs in a couple of multivariate trends. MSPC helps in this way to understand the difference between natural and impactful process variations. The use of MSPC has become routine to drive insight in development, particularly in association with continuous processing, already and also within commercial manufacturing there are many capabilities possible, such as:

  • Early detection of equipment malfunction
  • Insight into raw material impact
  • In situ assessment of process stability

A data analytics solution such as SIMCA®-online utilizes data regression models to summarize all of the individual parameters from various operations into multivariate models so they can be monitored in real time. This becomes very efficient in the control room because instead of looking at a large number of individual parameters or signals, you have a small set of summary parameters that let you monitor all the variables at the same time.

Another advantage of using a MSPC technology is that you can use correlations between variables to your advantage. You can increase the sensitivity of your system to pick up changes in the process as they happen.

Being able to tie all the data from various systems together into one single monitoring system is a great advantage.

A MSPC tool like SIMCA®-online works in real time to pull data from various data sources together and allows people in different locations to view the process. It can help you increase the awareness and visualization of the process throughout the company.

With the expanding applications of MSPC, it can with the correct approach also be used as a surrogate for release testing. In this light models are generally categorized on the possible impact on product quality (model influence) on one hand, and on the impact on the overall control strategy (or model consequence) on the other. When implementing models at any stage of the product lifecycle it’s important to consider1

  • the context of use for the model
    • What is the contribution of the model to the decision relative to other available evidence?
    • What is the significance of an adverse outcome resulting from an incorrect decision
    • Are there any limitations of the model based on the assumptions?
  • Model validation strategy
  • Model maintenance
    • MSPC models can require verification or an update upon changes in
      • Process conditions
      • Equipment
      • Material characteristics
    • Monitoring and trending of model performance will be a component of continuous process verification

This makes it a valuable exercise for drug manufacturers to partner with a solutions provider, such as Sartorius Stedim Data Analytics. Our solutions are validated software packages with transparent model management systems, appropriate level of data integrity  and validated mathematical algorithms, taking the total cost of ownership for our users down.

The implementation of MSPC models for process control also creates a shift in industry perception of the ‘Established Conditions’ (ECs) as described in ICH Q12. Where traditionally ECs were parameter based conditions (or in other words process inputs), MSPC makes it possible to move to performance based ECs, where ECs consist of both process inputs as well as process outputs (albeit assisted by models).


Figure: Example of parameter based or performance based ECs for using a spectrophotometer

Although it not stated in ICH Q12, this perception of the ECs would make it possible to ensure a state of control of product quality. To support the dialogue with the industry the FDA created initiatives such as the Emerging Technology Team. Through this open approach of the regulatory bodies, models will be able to support pharmaceutical development as well as be implemented within commercial scale continuous manufacturing processes.

The amount of data and information generated in continuous processing will keep on creating new opportunities to expand the use of Machine Learning and fully benefit of the ongoing digital transformation of the industry.

Find out more

View the recorded webinar on a case study from GSK, a find out how multivariate tools were applied to monitor the performance of a continuous API manufacturing process in real-time.

 Watch the webinar


1: ICH QIWG: POINTS TO CONSIDER (R2): ICH-Endorsed Guide for ICH Q8/Q9/Q10 Implementation; 6 December 2011

Topics: Multivariate Data Analysis, Continuous Process Monitoring

Kai Touw

Written by Kai Touw

Kai Touw is a (Bio)Pharma Market Manager at Sartorius Stedim Data Analytics. He is a driven and enthusiastic technology evangelist bridging the world of data science into pharmaceutical and biopharmaceutical processing.

Search the Blog

    Subscribe to the Blog

    View the:

    Data Analytics Glossary of Terms

    List of Webinars

    Get a free trial