AWS for Industries

[In the News] The Intelligent Era: Design, Detect, Discover

This article originally appeared on BBC StoryWorks.

How machine learning is changing the pharmaceutical industry and reshaping the way drugs are produced in the 21st century

The scale of the pharmaceutical industry is staggering. There are millions of patients to take care of, billions of data points generated by every clinical trial—and that’s before you get to navigating the web of supply chain logistics and manufacturing. Making those massive investments more efficient is why digital transformation has become a vital pursuit for global pharmaceutical companies in recent years, and why machine learning and artificial intelligence are key considerations for any pharma firm looking for a competitive edge.

“Digital transformation is critical for us,” says Bertrand Bodson, Chief Digital Officer of Novartis. For the global pharmaceutical giant who served almost 800 million patients last year, ensuring that each and every one of those customers is able to get what they need from Novartis’ manufacturing network is core to the company’s mission. “We have around 60 manufacturing facilities across the world, and we need the ability to be flexible. How do you adjust in real time and be agile enough to serve those almost 800 million patients every day?”

Machine learning has the potential to transform the global pharmaceutical supply chain by giving companies like Novartis the ability to adjust to changes in real-time and in doing so open up new, valuable efficiencies in the global market place. This digital transformation plays a critical role in making sure that drug materials are available—and providing actionable information in case they aren’t. “If our manufacturing team knows where ingredients are and how our supply facilities are behaving in real-time, we can adjust and proactively plan for issues instead of reacting to them,” says Bodson.

Bodson knew that solving for those challenges and putting Novartis on the path to true digital transformation would require a partner with the technical and strategic acumen to match the healthcare powerhouse’s global scale. “We wanted to work with one of the world’s technology leaders,” Bodson said. “And Amazon Web Services is very much at the top of that list.”

Global scale, individual needs

Transforming a global pharmaceutical company into a data-driven organization isn’t a simple task. Much like a doctor diagnosing a patient, it requires the ability to both identify pain points and suggest a course of care—although, in Novartis’ case, there are millions of patients with their own needs. “Healthcare has become a computational field because everything is now being digitized and translated into massive amounts of data,” says Shahram Ebadollahi, Global Head of Data Science and AI at Novartis. “When that happens, the issue is: How can one make sense of all of this data, and generate insights from it?”

The challenge, Ebadollahi explains, is that healthcare data is both vast and multi-modal, which is to say it takes a range of different forms. At Novartis, data can range from patient medical histories to the information on each stage of the journey from drug development through to global production and retail. It all amounts to an enormous repository of data that represents valuable raw material, and the first ingredient in becoming a more data-driven organization, according to Ebadollahi. But data alone is not enough: Finding the actionable information within it requires something more.

“You need an increased amount of computational capability to glean insights from the data—and thanks to organizations like AWS that’s become more accessible,” Ebadollahi says. “Ten years ago, it was very hard to apply machine learning or AI in a meaningful way because, even if you had the data, you needed sufficient computational power. But now that issue has been solved to a great degree thanks to organizations like AWS.”

Novartis is using AWS ML technology like Amazon SageMaker to streamline the process of training and deploying machine learning models, while being able to scale thanks to AWS’ cloud-based processing power. To tackle Novartis’ challenges, ML engineers from AWS and Novartis formed “two pizza teams” – a concept that dates back to Amazon’s start, where projects are led by small, agile groups – to move faster and focus on what needs to get done. In such projects the key to success is first guiding the client towards the right questions, explains Taha Kass-Hout, Chief Medical Officer at Amazon Web Services and Director of Machine Learning. “We want to help customers ask the right questions about what they’re trying to solve,” says Kass-Hout. “When we find those questions, we can dive incredibly deep and start simplifying on behalf of the ultimate user of that experience.”

Tailored solutions powered by machine learning

Novartis is using machine learning to develop smart manufacturing processes, including an insight center that is able to monitor manufacturing plants on an individual and network basis. That means that Novartis is able to keep an eye on prospective issues in a single facility (e.g. the low supply of a key ingredient) as well as those that might impact the supply chain as a whole (e.g. how a global pandemic impacts their manufacturing logistics). Novartis uses AI/ML to optimize supply chains and eliminate redundancies between facilities—both of which save the company money.

“We’re able to connect all of the data that impacts our manufacturing operations and processes and get a heads up on issues that might interrupt our supply chain and, by extension, our patients,” says Ebadollahi.

Novartis is able to use machine learning to learn where they can improve throughout the supply chain process as well. “We’re able to derive insights that you can apply to other manufacturing plants in order to make the processes more efficient and more rapid and more effective,” Ebadollahi continues. “And we’re able to do that at scale which is vitally important to us.”

Novartis also uses a machine learning solution called Buying Engine to streamline purchasing and drive procurement efficiency—no small feat when you consider the pharma giant buys $2B worth of lab supplies and personal protective equipment every year, from shoe covers to coveralls to x-ray inspection machines. The pharma company is targeting a 5% savings on that multi-billion dollar expense, all driven by using machine learning technology to create a more efficient and cost-effective procurement process. Buying Engine creates a knowledge graph that captures buying habits across the organization based on context and recommends purchases based on those purchasing behaviors. The solution has been deployed in Novartis facilities in Kundl, Austria and Cambridge, MA with a wider rollout coming soon, but the impacts go beyond the bottom line, says Bodson. “The idea is to transform our manual process and bring more transparency and efficiency to some of our critical processes,” he said. “It’s about simplifying operations and bringing Amazon-like experiences to our wider team.”

Finally, Novartis uses natural language processing to improve adverse event detection, a crucial part of drug safety practice that involves collecting data on potential negative reactions to pharmaceutical products. Using SageMaker, Novartis engineers built a product that monitors social media channels for mentions of potential adverse effects, assess them using sentiment analysis, and flags messages of interest to human researchers for review. Currently “AE Brain” processes around 15,000 messages per week, capturing far more data than a human team could review, and increasing the quality of Novartis’ drug monitoring overall.

“When a patient declares a potential side effect associated with one of our drugs—whether it’s in social media, whether it’s with a doctor—we have to report those within 24 hours,” says Bodson. “We’ve been working with AWS to spot when adverse events have been reported in real time so we can act on them quickly.”

Collectively, these applications of ML and AI hold transformative potential for Novartis and, by extension, the pharmaceutical industry as a whole. Through the use of fast, streamlined and affordable cloud platforms like those provided by AWS, the huge amounts of data generated by this or any other modern industry can be processed, analyzed and turned into insight and improvement.

Ultimately, for all the talk of prediction and optimization, the use of ML in healthcare will always be about better serving patients, as Novartis’ Ebadollahi explains. “The data science we do is not just for data science’s sake, it’s really enabling a chain of events such that a medicine can be produced more efficiently, faster,” he says. “It takes, on average, 12 and a half years to bring a new medicine to market, with a lot of risks and investment along the way. At Novartis we’re deeply committed to using data science and cutting edge technologies to transform the way we innovate across R&D with the ultimate aim of bringing innovative medicines to patients faster.”

Learn more about AWS in biopharma.

Kelli Jonakin, Ph.D.

Kelli Jonakin, Ph.D.

Kelli Jonakin is the Principal Marketing Manager for the Life Sciences and Genomics Industry verticals at AWS. She comes with a background in pharmaceutical research, with a special focus on development and commercialization of biologics. Kelli received her Ph.D. in Pharmacology and Systems Biology from the University of Colorado, and received an NIH post-doctoral fellowship grant to study Biochemistry at the University of Wisconsin-Madison.