The promise of AI to transform the clinical trial process is being realized in fits and starts, but early applications of the technology might be surprising.
At Fierce’s DPHARM conference in Boston, Hippocratic AI general manager for life sciences, Sohit Gatiganti, sat down with David Carruthers, vice president and head of global clinical operations and biopharmaceutical R&D at AstraZeneca, to discuss how the British Big Pharma has been deploying technology in a conversation focused on agentic-first trials.
At AZ, Carruthers said the company used a simple and tangible AI deployment to help get leadership onboard: calling Ubers for patients after trials. Not only was it a task the AI could easily handle, but it also addressed one of the major hurdles of participation. Along with challenges like childcare and taking time off work, it can be expensive for patients to travel to the large academic centers where most trials take place and book a ride.
Carruthers said increasing the use of AI needs to start with becoming familiar with working with an AI agent. “When you zoom out and look at how we use similar technology when we do things like phone the bank, we don’t speak to a human,” he said. “Most of the things we do, we speak to an agent first, and it has become normal in our day-to-day lives, but it hasn’t come to trial yet.”
After booking rides, AZ chose patient screening as another place to deploy agentic AI. Because a trial sponsor can’t recruit anyone until they have been screened, it made sense to start at the beginning of the value chain.
At the same time, screening is often the patient’s first interaction with a potential trial, and Carruthers said AZ was thinking about how to avoid the frustrating AI interactions that are ubiquitous in today’s world. “We needed to make sure this works well, smoothly, and we wanted to be patient-centered about it,” he said.
Another question Gatiganti posed was how to decide where the highest-value use of AI is, given the dozens of trials that many large biopharma companies are conducting.
The task can be especially difficult in an environment with data-centric leaders who want to see evidence of effectiveness. But with a new technology, the data to show it works doesn’t exist. The key, Carruthers said, is to be “modest about your expectations.”
“We have assumed zero benefit to this technology in our trials,” he added. “Now, that doesn’t mean there’s zero benefit. There’s a huge amount of benefit, but once you start making bold claims and setting the bar too high, you end up just changing the target rather than actually doing what makes sense.”
The conversation ended with a look ahead to what the next few years could bring as biopharma companies become more familiar with how AI can change the clinical trial process.
AZ is no stranger to deploying AI throughout the drug development process. It penned an AI licensing deal earlier this month with Owkin to help pharmaceutical teams analyze scientific, clinical and competitive data more quickly, and it also added a pact with Pathos AI to collaborate on a preclinical proteolysis-targeting chimera for patients with estrogen receptor-positive, HER2-negative breast cancer.
Carruthers said they were looking at high-volume, low-value tasks that can be taken out of the hands of humans so they can use their energy somewhere else, but he turned the question back on Gatiganti, who is a leader at a company developing generative AI healthcare agents for health systems, payers and biopharma.
His vision was clear. “I would love if we could just open up a trial for Hippocratic, plug into all of our providers and our sites, surface all the eligible patients, funnel them into your trial, call them, get them scheduled for first visits as soon as the trial opens and complete enrollment in two months.”
The company, which raised a $141 million Series B round that earned it a $1.4 billion valuation last year and this year rolled out a platform of conversational agents to interact with patients, has visions far beyond enrollment. “Once we solve enrollment, we solve the rest of the medical value chain as well, and we want to be able to run a trial end-to-end in two years,” Gatiganti added.