AI and Machine Learning: Streamlining and Focusing Clinical Trial Recruitment
Artificial intelligence (AI) and machine learning are increasingly becoming a part of drug discovery and development beginning with identifying new compounds to structuring and designing clinical trials and targeting clinical trial populations.
A recent example came out of Linköping University in Sweden. The investigators utilized an artificial neural network to create maps of biological networks based on how different genes or proteins interact with each other. They leveraged a large database with information about the expression patterns of 20,000 genes in a large group of people. The AI was then taught to find patterns of gene expression.
And in mid-February, a drug developed using AI began testing in human clinical trials. The molecule, DSP-1181, is currently in Phase I clinical trials for obsessive-compulsive disorder. The compound is a long-acting potent serotonin 5-HT1A receptor agonist developed using AI that was part of a collaboration between Japan’s Sumitomo Dainippon Pharma and the UK’s Escientia. The AI developed the compound in about 12 months, compared to a more typical five-year process.
And only a week or so later, researchers at Massachusetts Institute of Technology (MIT) used a machine-learning algorithm to identify a new antibiotic. The computer model can screen more than 100 million compounds in a few days. It was programmed to select potential antibiotics that leverage different mechanisms of action than existing drugs.
For non-tech people, the difference between AI and machine learning isn’t terribly clear. Even when looking at the definitions, the differences aren’t all that clear. Generally speaking, AI is designed to simulate natural intelligence to solve complex problems and, seeming key to the definition, make decisions.
Machine learning, on the other hand, is where a machine, i.e., computer, can learn on its own without being explicitly programmed. It is an application of AI.
Another aspect of that is that AI will attempt to find the best solution, while machine learning will just try to find a solution, optimal or not.
Although numerous biopharma companies are creating partnerships with AI companies in order to short-cut the drug development process, as the example of Escientia and Sumitomo Dainippon Pharma suggests, we’re only in the early stages.
Using AI and machine learning to restructure and inform clinical trial design and recruitment appears to be further along. Zachary Gobst, co-founder and chief executive officer of Leapcure spoke with BioSpace about the company and this aspect of AI and machine learning.
Founded in 2015 with offices in the U.S. and Europe, Leapcure partners with pharmaceutical sponsors, contract research organizations (CROs) and investigator sites to advance patient engagement in early clinical development using AI and machine learning. The company focuses on public awareness, protocol-level feasibility and site selection.
“My background prior to Leapcure was working with startups in the virtual and mobile clinical trial space. What I noticed when helping find patients for those studies was when we were able to work with advocacy groups, and helping influencers across all indications, we were getting better research results,” Gobst said. “Advocacy patients would be reachable. What they told us would be accurate. And basically, it was the most influential way to get our research done on time and well. When we started Leapcure, it to answer the question, how do we give advocacy groups a voice in clinical trials all cross the process?”
Gobst points out that traditionally, almost half of clinical trials are unable to meet specified enrollment goals and 15% of sites can’t enroll a single patient.
“The primary thing we do is help with recruitment, but we’re picking up more projects in terms of helping improve research design. We know the voice of the patients. Essentially, what we’re able to forecast in terms of patient engagement, advocacy and digital platforms, helps provide solutions that put together the right protocol design for sponsors and select sites for best patient engagement,” Gobst said.
For example, he said, they can bring in sponsor companies and tell them, based on what their protocol for recruitment is, whether 10 out of 16 people viewing recruitment information are going to say no or proceed with standard of care or find another trial. And if patients have to travel, they can help identify where the patients travel from and how much of a problem that’s going to be.
“With our platform, when we bring in the patients and digest the information of where they are from, what they are telling us about their level of interest, our platform is able to digest this in a natural way,” Gobst said. “Then, if you’re trying to target communities of patients, how do you understand the keywords to find patients that are going to continually sign up and enroll in the trials, instead of just using keywords that drive traffic to the ads.”
In other words, their AI platform gathers patient data and is able to help identify which patients are more likely to sign up for the trials and how to reach those patients using more targeted advertising.
Their AI platform allowed Leapcure, for example, to target Google AdWords on a cost-per-reversion or cost-per-referral basis instead of cost-per-click in order to get “better quality patients.”
Gobst notes how the AI and machine learning distinction works for what they do. “How do you capture data that’s actionable? That’s not machine learning, that’s AI. We have a database that takes in signal data and push it back out in a way you can communicate with sponsors to help them make tradeoffs.”
On the other hand, he said, “the retargeting we do is driven by machine learning. What are the millions by millions of things in profiles of someone who signs up and how do we predict the outcome for the next person with million-by-million profiles and say, ‘Hey, we’re going to overbid for these keywords over others.’ So on our end, AI is kind of broad, while machine learning has specific use cases for us.”
On a more concrete basis, Leapcure has 30 to 40 case studies showing savings for sponsors of $300,000 to $3 million per study. “That’s relevant, it’s real. It’s already delivered,” Gobst said.
Often, he said, Leapcure is brought in because clinical trials are behind on their timeline and may already have run over budget. “Now they’re bringing us in at the beginning to reduce how much infrastructure they need to build. In the next year or to we expect it will really be about making sure you can plan your clinical trials around real data and making sure your protocol matches that. Before our service, sponsor companies might run focus groups or talk to smart doctors, but they often end up trying to spend their way through problems instead of using patient data to drive their strategy.”
The bottom line, and by that Gobst means both the understanding of how this works as well as the companies’ bottom lines, is “When you’re talking portfolio planning and using real-world data to drive market access and other parts of the clinical trial process with advocacy models, you need a deeper understanding of your power users and making decisions according to their needs to provide better research results.”