Leveraging Historical Data to Accelerate Process Understanding
By Katie Anderson, Chief Editor, Pharmaceutical Online
When you are in the late stages of the drug process and recognize a flaw, time is of the essence. Changes to the process need to be minimal and expedient, making process understanding critical. At the ISPE Annual Meeting & Expo, I had the chance to sit in on a presentation by Philippe Cini, partner for A-Connect, who introduced his data-focused method of process understanding to recognize defects early, make process changes quickly and stay on schedule.
Rather than generating an entirely new data set, Cini’s methodology achieved quantitative process understanding by accessing historical data from development or commercial manufacturing. “We have limited data, don't have a lot of time and have constraints where we have already made a submission with the regulatory agencies and we don't want to change too many things because then it becomes even more complicated,” explained Cini.
A Data-driven Path
Cini utilized historical data to craft models, even when the data was incomplete. He found it was both faster and cheaper than generating new data. “It's readily available. And so, it's faster, it is cheaper, but it's also more complicated to analyze. However, there are ways to extract a lot of information from this type of data,” he noted.
Cini’s methodology started with a qualitative screen that involved a POC review, interviews, a process tour and maps. It then moved to a basic quantitative screen with control charts, capability indices, bivariate analysis and a measurement systems analysis. Next comes the advanced quantitative screen. If there is enough data, there is regression analysis and optimization modeling. Without enough data, there is hypothesis generation, scouting experiments, screening DOEs and optimization DOEs. Finally, there are confirmation trials. This combination of advanced data and expert knowledge was put to the test in a few examples that Cini provided.
In one example, late-stage development process deficiencies were corrected, and in another stability issues in a commercial product were solved. Cini guided attendees through these case studies, showing the data generated from each case and where the expert team recognized the issues.
A More Robust Process
“If we understand [the manufacturing process] and put the controls around it, we have the exact product with consistent specifications that have been predefined. It makes the process robust, and we achieve a product of very good quality time and again, reliably,” he noted.
Cini wanted attendees to know that there are ways to extract insight from historical data such as batch records, materials COAs, DSRM COAs, release and in process data, analytical investigation data and historian data, even when it is limited or not organized. He added, “What is changing is the amount of data that we are having access to that we didn't have access to in the past easily. And the data can be from the company related to the product, or it can be from outside the company, but still it can provide insight into the chemistry, for example, that is at play.”
He furthered that there is work to be done on the collection side, but that machine learning is leading to better models. “I'm very excited about the future because with those capabilities in AI, in the ability to digitize the data, to manage the data, we can extract a lot of information from it that we could not before. And it should make the development process faster, a lot faster, which in the end benefits the patient,” he explained.
He concluded that we can handle more data than we ever could before. Every day, we are able to manage data better to understand the process more. This will lead to not only the ability to ensure product quality, but also to reduce deficiencies by controlling process variations.