The promise of AI’s impact on clinical trials has many experts both bewildered and skeptical.
“[Vendors say,]‘Everything is possible thanks to machine learning and artificial intelligence.’ But it’s not that easy,” Sara Holoubek, CEO of Luminary Labs, said while moderating a panel at the AI World Conference & Expo last December.
Alongside Chris Bouton, CEO of Vyasa; Ted Slater, Global Head of Scientific AI & Analytics at Cray; and Balazs Flink, Clinical Trial Analytics Lead of R&D Business Insights and Analytics at Bristol-Myers Squibb, Holoubek put the scope of AI’s contributions to clinical research in perspective.
“We don’t move fast and break things in healthcare the way other industries do,” Holoubek said.
AI excels, Bouton said, at providing narrow solutions for narrow problems.
“[AI] can play a game really well, or they can identify a certain kind of image really well, or they can identify a certain kind of pattern in data really well,” said Bouton. “They’re not taking over the world or taking our jobs… That might be something down the line humans have to grapple with, but right now I think there’s plenty of opportunity to apply narrow AI to a wide range of low-hanging fruit.”
Slater agreed, adding that with limitations in mind, companies ought to be careful about what problems they pass off to AI.
“It’s not necessarily true that you can take these types of technologies and apply them to just any old problem and expect amazing results,” Slater said. And it’s not as simple as purchasing an AI product and hoping it solves a problem.
“You have to understand the problem you’re trying to solve, you have to be very careful about the data you choose, and if you’re using deep learning you have to find a whole ton of it,” Slater said. “You have to prepare it appropriately and prepare the framework you’re going to use to train the algorithm.”
On top of that you have to make sure you train the algorithm correctly, said Slater.
So, you get the idea—approaching AI with the mindset that it will solve every problem proves to be a costly mistake.
But there’s another pricey decision, according to Balazs Flink—choosing when to use AI in the first place.
“In R&D we have to decide whether we’re going to buy a fancy solution, or if we’re going to run two more clinical trials that may give us two new patents and two new indications for a shared portfolio that’s going to generate revenue,” Flink said. “What I don’t see fully available is a good enough algorithm that’s going to have us comprehensively solve the trial development process.”
The push and pull of managing expectations in AI comes from a lack of understanding from senior leadership in clinical research, said Flink.
“They read a lot of groundbreaking news in the media and they see self-driving cars and these popular solutions, and they don’t understand why it doesn’t work in the professional environment they’re managing,” Flink said. “So they come to me and they say, ‘Give me the magic box,’ which they think will give them the best strategic R&D choices on how to run their portfolio and make them smarter than their competitors.”
Whatever disagreements arise from such discussions, a mutuality was found in a timeline for AI’s application. While Flink, Bouton, and Slater agreed AI’s true potential is years away, Bouton suggested there are already areas where AI can make an immediate impact.
“For the next 5-10 years I think there’s an incredible amount of opportunity to apply narrow AI, or narrow deep learning algorithms, to a wide range of low-hanging fruit in the clinical trial space,” said Bouton. “Everything from text analytics to image analytics and IoT analytics are places where we can apply these algorithms to great effect.”
Read the source article at Clinical Research News.