Risk detection is an integral part of the pharma manufacturing process. According to the principles of Quality by Design (the US Food and Drug Administration’s quality standards), risk needs to be pushed as far away from the patient as possible. This means protecting them from any flaws in the manufacturing process or raw material. And that means finding those flaws as early as you can.
Unfortunately, this can be a labour-intensive process, requiring a number of complicated control systems in combination with manual reporting. There’s another problem too, in that it tends to take place after the fact.
“The controls exist to record deviations, but they’re all in the rear view mirror,” says Rajiv Anand, CEO of Quartic.ai. “If you look at visual inspections with cameras, inspections of vials at the end of the process, those types of inspections exist but they aren’t able to detect things early. It’s not that they’re not in place – it’s that they’re reactive and not proactive.”
It was for this reason that Quartic.ai decided to team up with Sparta Systems on a new AI-based risk detection tool. The software will identify risks early in the pharma manufacturing process and give users quality insights in real time.
“Fundamentally what you’re doing is you’re moving from quality control to quality assurance,” says Anand. “That means a greater confidence that you’re getting it right the first time.”
How the software will work
Quartic.ai, an industrial AI software company, was set up in 2017. Its flagship product, the Quartic platform, helps engineers deploy their own AI solutions without any training in data science.
Sparta Systems, meanwhile, is a leading provider of quality management system (QMS) software for the life sciences industry. Together, the two companies are creating a system that collects data ‘on the shop floor’, learns from this data to detect emerging deviations and abnormal process behavior (Quartic) and creates proactive quality assurance (Sparta).
“Our reason for choosing to work with Sparta was that they have a strategic intent to create the next-generation QMS systems,” says Anand.
“QMS systems are largely systems of record after the fact but Sparta plans to make them more proactive, and it was here that we felt we could help each other. You need intelligence about the manufacturing operations, which is driven by IIoT and manufacturing AI. So from Quartic’s point of view that’s what we’re trying to do – we want to provide AI-driven Quality by Design in real time.”
Users will continue to work with Sparta’s TrackWise digital platform, which has modern, digital QMS workflows in place. This platform will communicate with Quartic’s IIoT and AI platform, which will be trained to spot anomalous events.
Once the algorithm is fully ‘trained’ and the solution is deployed commercially, the AI will be able to flag up any deviations from the norm. If performance drift is detected, the digital QMS workflows will alert users with corrective instructions and measure the impact of corrective action with evidence from the manufacturing floor.
“We’ll take that real-time data from the unit operations, and combine it with the previous deviations that have been captured by the QMS platform – that’s how the two capabilities will come together for proactive quality assurance,” says Anand.
Quality assurance across the value chain
In terms of the kinds of anomalies the platform will flag up, Anand describes it as “anything that isn’t possible for humans to detect in real time”.
“Think of it as any non-linear multivariate relationship that humans can analyse after the fact, but cannot detect at the time,” he says.
“Every process leaves a digital fingerprint, we know that, but that fingerprint can’t be analysed with human eyes. As we bring in new detection techniques, using complex sensors, there is too much data to analyse ourselves, so it becomes a big data problem and that’s a natural fit with machine learning algorithms.”
The way things work at present, you might visually inspect your vial at the end of the production line. If it’s contaminated, you’re unlikely to know where that contamination came from just by looking at it. You might therefore add detection techniques throughout the line, via high-speed cameras and process analytics. But you wouldn’t have those insights until afterwards, and wouldn’t be able to take corrective action at the source of contamination.
The Quartic/Sparta solution, by contrast, would allow you to avert any issues immediately, and avoid wasting time, effort and potentially finished product.
“With a traditional QMS, it’s largely limited to quality professionals, but when you make it predictive and proactive, and connect it with what’s happening on the shop floor, operations and engineering teams can more directly benefit from quality data,” says Anand. “Even the original process development teams upstream can take that quality data and bring into their future product development.”
After all, the very premise of ‘quality by design’ is that quality isn’t just a function of the quality team – it’s the responsibility of everyone in value chain.
“It means engaging operations, engineering and process development scientists, and providing an opportunity for that data to be available all the time,” says Anand. “It gives you the confidence to make your supply chains leaner, meaning you carry less in-process inventory and less lost time in lost batches because you had some kind of deviation that needs investigation. What we’ve seen with Covid-19 is that supply chains need to be very agile and lean, without circumventing the quality processes.”
Future-proofing the industry
Quartic and Sparta’s work is now well underway. Both companies have assigned dedicated resources to work on the joint project, and the development process started long before the partnership was made public in July.
“The platform isn’t ready for deployment yet – we need to engage client data or actual scenarios in the design verification activities,” says Anand. “So we’re actively working with some clients to do that. We’re targeting our initial release in winter 2020.”
He thinks the solution will be particularly suitable for biologics applications, in which the interactions between different variables are harder to understand, and the consequences of poor risk detection are greater. It may also find a grateful user base as continuous manufacturing techniques become more widespread.
“Continuous manufacturing is a priority for many manufacturers and the FDA, but it will require what I call continuous quality as well,” he says. “That’s impossible with the traditional quality techniques we have, which were largely made for batch processes. Overall what we’re doing is future-proofing the industry – as the industry moves to continuous manufacturing and biologics, we’re enabling it to move forward at lower cost, higher speed, lower risk.”