AI + data = the winning combination for clinical trial planning

AI is becoming embedded in the business world, and the pharmaceutical industry is no exception. With its adoption, however, critical questions arise as to how it is implemented and the impact it can have. This interview looks at how AI in clinical trials can improve R&D efficiency – as well as the guard rails needed along the way.

Claire Riches (CR): It’s very real. Pharmaceutical companies that lack an AI strategy or have not yet begun implementing AI are already behind the 8-ball. As an industry, we’re leaning more on technology because we have so many disparate systems creating more burden on sites and on patients.

So can data and AI help? The short answer is yes, it absolutely can.

Across the industry, we’re still manually pulling from those disparate data sources and we’re not really sure how we can maximise them to derive really useful insights to help us move forward. AI is a gamechanger is this respect.

CR: There are many areas. Let’s focus on these: predicting clinical outcomes; optimising workflows; reducing cycle time; finding eligible patients and selecting study sites.

Feasibility is playing an ever-important role in drug development. With the cost of clinical trials skyrocketing, sponsors want to know a study’s likelihood of success upfront.  Determining – with the help of AI – what factors might impede success, and course-correcting before a trial gets underway, will save time, effort and budget. By conducting feasibility assessments early on, sponsors can avoid costly protocol amendments, screen fails and other obstacles that threaten to extend timelines.

One area that probably stands to benefit most from AI is workflows, whether it’s in the life sciences or any other industry. AI is not designed to replace staff; it is designed to help automate tedious, time-consuming tasks, thus making processes more efficient.

Locating eligible patients is the bane of study teams. This is particularly true in rare disease trials, where finding those patients is like looking for a needle in a haystack. AI can serve as the magnet that draws the needle out.

Site selection is not just about finding sites with experienced staff. It’s about the intersection of staff savvy, patient access, bandwidth and track record. That’s a lot of data to cross-reference and AI can streamline the process.

CR: Data and AI go hand-in-hand. Having access to data is one thing, but sponsors must be able to parse the data and apply the findings to inform future trials.

Although sponsors want to avoid a failed trial at all costs, the data that can be mined from such trials is valuable in planning future trials. What worked? What didn’t? Were the inclusion/exclusion (I/E) criteria too restrictive? When did patients drop off? Why? Those are some of the questions whose answers can be found thanks to the help of AI.

And I would tag on another question here. What role do humans play? The current catchphrase is ‘humans in the loop’. I’d amend that to be ‘humans in the lead’. Let’s not forget that it’s the clinical expertise that adds value. We’ve got to remember that AI is a tool. It’s not the be-all, end-all. Taking AI output and having it analysed by subject matter experts is what ultimately improves outcomes.

CR: AI is only as good as the data it is provided with. As they say, ‘garbage in, garbage out’. When we refer to rich data, we’re talking about both quantity and quality. While it’s important to have a broad data set – including timespans, geography, study sites, investigators, social determinants of health (SDOH) and patient information – it’s important that your data comes from reliable, accurate and timely sources. And you need the right tools, including AI, to wade through that mountain of data.

CR: The winning combination of data and AI not only supports protocol design, but the strategy behind components such as endpoints and I/E criteria. Assuming you have the data, AI can provide a historical look at protocols and determine which were operationally successful and those that were not. Then, based on the trends revealed, AI can help build I/E criteria that will optimise a protocol.

In terms of patient recruitment, AI’s large language models (LLMs) can be used to examine unstructured data, such as electronic health records (EHRs), and identify eligible patients without revealing protected health information (PHI). AI also can examine other data sets, like real-world data, to determine what the clinical landscape looked like and how recruitment fared when the protocol was conducted. Based on that, a sponsor can begin to forecast enrolment for a given trial and make necessary tweaks. By writing the protocol with the patient in mind, a sponsor can improve enrolment from the get-go.

You don’t have to wait until that protocol is live and you have no one to recruit or you have a really high screen-fail rate. You can pressure test that upfront now with data and AI to avoid those pesky protocol amendments.

Site selection is a huge part of the patient recruitment puzzle. AI can detect the number of active sites, whittling down those with critical experience and patient availability to ‘score’ recommended sites. Traditionally, 80% of patients have been recruited from 20% of sites. There’s an untapped wealth of available sites; sponsors simply need to know how to leverage AI to find them.

Let’s face it. Clinical trial feasibility is evolving and it’s continuing to evolve.

What we’re trying to do here is reverse the process. Instead of starting with enrolment at the end, we’re starting with it. We can use AI and real-world data to figure out where the patients are, which sites and countries should be selected, and what the patient journey looks like.

I would add a caveat that if you’re working with AI tools, it represents a considerable financial investment, but one that will pay off in the long run. That said, you must be able to demonstrate value for your efforts.

Claire Riches is Vice President of Clinical Solutions at Citeline