Leveraging artificial intelligence to manage clinical trials
An interview with Kim Walpole, Co-Founder and CEO of Trials.ai.
Conducted by Tatsiana Vertak
Tatsiana: What does Trials.ai do?
Kim: At Trials.ai we are on a mission to get treatments to patients faster by optimizing clinical trial protocols for speed and success. We leverage artificial intelligence to help clinical trial sponsors and CROs design better protocols by mining vast amounts of related source documents so that they can use our system to gain insights and recommendations specific to their product and therapeutic area.
Tatsiana: How do you feel about the future of incorporating AI in clinical trials? What are some of the barriers that you see with this?
Kim: Our Chief Scientist, David Fogel, recently published a paper on the reasons why clinical trials fail. In that paper he covered many opportunities for AI to address failure, in areas such as study design, site selection, recruitment, and patient burden. For me the greatest opportunity that AI presents is the ability to access data, a lot of data across therapeutic areas that no human could possibly navigate in their lifetime.
Traditionally, clinical trials have been a manual, incredibly paper laden process, so one of the biggest challenges, especially with bigger sponsors and CROs, is to aggregate the data that lives in separate files, departments and systems. Some of it isn’t even in PDF or optical character recognition (OCR)- ready, so that adds another layer of complexity.
Tatsiana: What sort of advice would you give to clinical research professionals who would like to incorporate AI in clinical trials?
Kim: I think that there are a lot of use cases for AI. Obviously, we see a great fit with trial design and site and investigator selection. We’ve got a patent pending on what we call the Patient Burden Index, which looks at a combination of factors that may slow down recruitment or lead to retention problems. The goal here is to minimize patient burden by simplifying the engagement process.
All sort of AI tools are being put to work by ourselves and others. Natural language processing (NLP) and natural language understanding (NLU) pipelines that help process information. Machine learning that surfaces insights and recommendations. Sentiment analysis. Chat bots. Technologies like smart robots that ensure compliance . It’s exciting to see the space growing and to have a vision of what the future of clinical trials could look like.
Tatsiana: Has your business model changed since you founded Trials.ai?
Kim: So much – as most startups do! We’re now focused on trial planning, specifically protocol optimization. But we started off on the execution side of clinical trials. We originally built out a platform to run clinical trials. The system navigated stakeholders (sponsors, sites and subjects) through the trials based on their role in incorporating data collection, adherence and monitoring.
The problem was that more and more customers were coming to us with clearly flawed protocols. We soon realized that we could be more effective if we focused our technology solely on optimizing clinical trial protocols. The reality is that the trial execution space is crowded, and if efforts to accelerate clinical trials could be solved via execution alone , trials wouldn’t be taking longer and costing more than ever. We had to think about the space differently. This pivot provides us with an opportunity to really change the face of clinical trials by going upstream to fix the root causes of failure.
We see Trials.ai as a support tool for research teams, rather than a tool to replace experience and critical thinking. We call it a smart protocol builder. The truth is that in the digital age, humans can’t physically consume every bit of information relevant to a given decision. In our current clinical research environment, there are vast amounts of clinical trial data housed in many different systems, on top of paper documents. There’s no way that even the most experienced thought leader in a therapeutic area could keep up! In the life sciences, this means that researchers are making important decisions and organizations are making major investments with incomplete information.
If you’ve seen a clinical trial protocol you know that it’s made up of many parts, including objectives, endpoints, inclusion/exclusion criteria, statistical analysis, and patient interactions. It’s basically the roadmap for how you’re going to run the trial. We are building our system to take things like regulatory guidelines, patient and site burden, and new technologies into consideration as our algorithms aim to provide insights and recommendations to research teams. Humans can then incorporate the feedback or not based on their context.
Tatsiana: Dr. Fogel’s paper focused on reasons why clinical trials fail. Can AI help to prevent these failures?
Kim: AI can improve even small things that lead to big savings over the long haul. One area for big gains using AI is in predicting protocol amendments. I think that the last article I read estimated a phase 3 major amendment as costing $535,000 per amendment. That’s huge in terms of cost and time wasted.
Brilliant people are constructing protocols, and there’s always logic behind why they’re doing what they do. However, in many instances they are making critical decisions in silos. The current process is manual — whether the trial sponsor is running the trial using their own internal resources or hiring out to a consultant.
I think one of the fundamental ways that AI improves clinical trials is simply by making sense out of typically unstructured data. For that we use natural language processing. We’re continuing to build our own clinical trial ontology that affords us a greater opportunity for precision in our recommendations. Once that data is codified, then in my opinion the real fun begins because machine learning algorithms can surface insights across therapeutic areas that may not have been considered before.
The bottom line is that, from our perspective, the more complex the trial is, the greater the risk, and the greater possibility of amendments, deviations, and challenges with recruitment or retention. For us, the reduction of trial complexity is one very practical way to see positive results.
Tatsiana: Thank you, Kim. What’s up next for Trials.ai?
Kim: We’re working on our web-based interface that will allow you to construct your protocols from start to finish. You can think of it as Turbo Tax for clinical trial planning. There are structured authoring tools available out there, but our goal is to optimize protocols with data driven insights and recommendations. For instance, given your context and therapeutic area these are the common inclusion and exclusion criteria. As you select and deselect those, how do they impact recruitment, cost, and time needed for the trial? We can also integrate with Transcelerate’s common protocol template in order to meet companies where they are today.