Transforming Clinical Trial Design and Avoiding AI Wrappers: Q&A with Angela Schwab

Many promises were made about the ability of artificial intelligence (AI) to transform the pharmaceutical industry. As existing companies continue to experiment with the technology and new AI-focused companies pop up left and right, it’s time to examine where it’s having the best impact.

Trialynx is an AI-powered clinical trial platform that uses the advanced technology to help trial designers overcome common issues, predict potential failure points and find ways to fully optimize the trials. CEO Angela Schwab, MS, PMP, spoke with Pharmaceutical Executive about specific ways AI is being used to improve clinical trials.

She also discussed how the industry can evaluate AI companies and identify the ones that are bringing genuine change, while avoiding others that are less effective.

Pharmaceutical Executive: How is AI supporting clinical trial design?
Angela Schwab: We have hit a new era of clinical trial design. Up until 2024, we’ve been doing this manually by taking a protocol that somebody else wrote and saying that it’s probably pretty good.

We copy and paste and put things together. What comes with that is the failures. There’s an incredibly high failure rate in clinical trials, such as not meeting end points, going massively over the timelines and not getting patients in.

This compromises the integrity of the science and the business.

Those failures are inherent in that old manual process. Now, we have an opportunity where AI can scrape massive data sets and look for things like end points in other clinical trials in your domain, or what other tests were done. It can provide that predictive quality into protocols and planning, so you can start to understand things that will have a positive or negative impact on the development of the clinical trial.

PE: Can AI help solve the issue of high failure rates in clinical trials?
Schwab: There are many ways that clinical trials fail. Maybe it takes too long to find patients, subjects get overloaded, sites start disappearing because things are too complicated for them. The ways to make that failure rate go down are by putting more thought up front and thinking through the decisions in creating the protocol and how they impact patients, sites, the data being collected, and determining if there are ways of doing that better.

This could involve using decentralized elements that make it easier on patients and sites. Is it possible to decrease the data burden? We’ve been overcollecting data, and a lot of it isn’t actually supporting the end points and objectives. When designing end points and protocols, there are many times you don’t need it to be so complex or have as many end points.

PE: How can teams reduce patient burden without weakening their science?
Schwab: No. 1 is lean trial design. We should never put patients through tests, visits or questionnaires that do not directly support the trial’s goals.

As an example, I was working with a clinical team that wanted to do eight psychological questionnaires on their patients. That’s about two hours of questionnaires for the patients. The team said not to worry, because the patients would be sitting in the waiting room anyway.

Why are they in the waiting room for two hours? Do you think the assessment at the end of two hours of questionnaires is going to be valid? They’ll probably get angry and start filling out “A” for every answer.

We must put ourselves in the patients’ shoes and think about that journey. What’s it like? How far do they have to drive? Could we do decentralized? Could someone go to their home? Could they do a questionnaire online, etc?

It’s worth doing a dry run and having investigators go to the site, see what the drive is like, and even what parking is like. Sometimes we think about poor planning as something that can be expensive for patients and trial sites. We must be sure we’re reimbursing them correctly for their time and everything they’re putting into it.

PE: How are sites impacted by poor planning and trial design?
Schwab: The sites really bear the brunt of all of this. I worked at a site for about 14 years; I oversaw the determination of which trials to take on. A lot of those decisions weren’t necessarily based on the science, but instead on the operational impact of the trial.

We’d make the decision based on elements such as how many coordinators we’d need, how intense are the patient visits and how intense the tests would be that impacted various departments. If a trial required 30 MRIs in a month, that could adversely affect multiple departments.

We must think through these plans step by step and not just assume that trial sites can handle the whole burden of any trial. They may say yes, but later drop off due to the burden. If you have a clinical trial and sites seem excited, and all of the sudden you’re getting crickets, there’s probably a reason behind that.

PE: How can investors identify promising AI companies apart from software simply built on top of ChatGPT’s API?
Schwab: There are a few critical elements that investors should be looking for. There must be baked-in intelligence; it shouldn’t be something that looks like ChatGPT but claims to be just for one specific thing. If it’s just for creating a document or PowerPoint, that’s likely a wrapper.

Built-in intelligence and infrastructure are in the place of an actual platform that uses AI to create specific content in specific ways. It walks the users through a journey in creating the correct document.

ChatGPT could write a protocol for a specific type of research study. But with our platform, it will work through everything step by step. It will go over the objectives, the hypothesis, the end points and the procedures. All of those things stack on each other and are backed by data that the AI can pull in and help with the decision-making process.