A global biopharmaceutical company needed to establish parameters to monitor processes at a CMO site to facilitate data analysis
for science-based decision making and greater process understanding.
The sponsor required data from the CMO to fuel the same automated observational software tools used at its own sites to monitor
large amounts of disparate data. The sponsor had limited knowledge regarding the CMO's processes and restricted access to
CMO process experts, since collaboration was not required by the sponsor-CMO contractual agreement, which only allowed for
limited data sharing.
Without a more informed starting point, preliminary parameters to monitor processes were based on educated guesses. The sponsor—with
help from its analytical consultant team—set-up nearly 400 "suspect process and outcome parameters" to start the investigation.
For example, the team monitored incoming material temperatures at a certain point in the process, under the suspicion that
this variable would affect outcomes at the next process stage. It also predicted other parameters that had the potential to
impact product quality, such as measuring the presence of synthesis-related impurities after ingredients were combined in
one process step.
If the sponsor had been working at its own manufacturing site, it might have used a typical riskassessment approach, as guided
by ICH Q9 Quality Risk Management (2). Authors from Pfizer have described this approach in detail:
"Risk assessment is the process used to prioritize parameters and attributes most likely to impact the product quality...
The focal point of a QbD risk assessment is to be able to link quality measures and process controls to the product quality
of the drug delivery system, i.e., safety, efficacy, and performance. A Quality Target Product Profile (QTPP) is an effective
tool to help identify the Critical Quality Attributes (CQA) of the manufacturing process that link to product quality" (2).
Pfizer's approach requires access to experts and upfront risk assessment prior to setting up trending. Because the case study
involved a CMO site, the team had to set up automated trend monitoring systems and then watch the data patterns to identify
what was important.
To do so, the sponsor team compiled electronic data supplied by the CMO for analysis at the sponsor's headquarter using the
sponsor's process intelligence software system. Through automated trend monitoring of the suspect parameters, the team could
eventually narrow down the 400 initial parameters to those that were critical to quality outcomes. The software was configured
to send alerts when data was trending out of specified limits, and the team also had access to data required for investigation
of potential cause-and-effect linkages. This approach was much more interactive and real-time compared with the up-front method
used in Pfizer's risk assessment approach and involved continuous process improvement using the ongoing analysis of data to
increase process understanding.
The sponsor found this monitoring by exception method to be a helpful way of compensating for the limited access to process
experts at the CMO site. It allowed the sponsor to see and learn from the data directly, and also facilitated collaboration
by using data as a communication vehicle between CMO and sponsor.
The above trend-monitoring approach was successful because the process monitoring systems were designed to monitor by exception
(see sidebar), using an automated system to alert users when they should pay more attention to a specific parameter of concern
and further analyze the data. However, the sponsor had to select suspect parameters for monitoring without knowing their criticality.
Standard monitoring procedures were conducted, such as baselining data, removing special cause variation and estimating process
capability, and control charts were developed to monitor process parameters by exception, which involved sending alerts when
control and specification limits were violated. The true value of the control charts was not observed until a few months later
when a deviation occurred at the CMO site. The sponsor had quick access to process monitoring information and could also use
the charts to better communicate with the CMO site.
Sidebar: Monitoring by exception
Following the selection of suspect parameters, the next phase for the sponsor team was narrowing these to determine the true
CPPs. The sponsor team received alerts automatically when exceptions occurred, but if one parameter revealed a deviation,
then the team knew it needed to monitor three to five additional parameters closely related to the measurement area. As a
result, these real-word deviations provided a mechanism for filtering the suspect parameters over time, helping to determine
When working with CMOs, sponsors often receive information regarding process indicator variables, but when the sponsor's analysts
realize that a process is not functioning properly, they still have to pinpoint the root cause. Unless comprehensive contractual
agreements exist with details of data access and analysis, it can be difficult for the sponsor to access the right data for
root cause analysis and statistical modeling. This case study expanded the sponsor's process understanding and allowed for
an open conversation with the CMO site regarding which additional indictors and process and outcome parameters would help
both parties monitor processes in the future. In the end, the approach described above offered numerous benefits for both
the sponsor and the CMO. Figure 1 shows the time saved through monitoring by exception.
Figure 1: Potential time saved through monitoring by exception compared with manual monitoring using spreadsheets.