How LLMs Are Changing Clinical Trial Recruitment: Key Insights from the Latest Cohort Study
A new survey just dropped from Boehringer Ingelheim and the University of Tübingen and it’s the first comprehensive review of large language models (LLMs) in trial patient matching. While adoption across the industry is still in its early stages, the findings are clear. Most models today focus on a single disease or trial, with fewer than 5% utilizing genuine patient data. The benchmarks are limited, and trust still remains a significant hurdle. Even academic tools like TrialGPT show promise, but they aren't deployed at scale yet.
Here at Trially, we’re already a step ahead. Our platform is actively prescreening real patients across therapy areas, trials, and sites. We are not working from test sets, but helping research teams solve actual recruitment problems with real data. And in head-to-head use, Trially is already outperforming academic prototypes like TrialGPT in live, clinical workflows.
The takeaway is simple. LLMs are not theoretical. They are already working. And this new research helps show us why.
The Study That Has Everyone Talking
The new publication “Cohort Discovery: A Survey on LLM-Assisted Clinical Trial Recruitment,” takes a wide-angle look at both traditional recruitment methods and emerging AI-enabled tools while examining how the effects of these AI solutions are changing patient recruitment in real time. Except these researchers didn't just sit in a lab theorizing about what might work someday. They actually rolled up their sleeves and examined everything from old school recruitment methods to emerging AI-powered tools. They looked at benchmarks, dug into evaluation strategies, and tackled the frustrating challenges that are preventing sites from adopting new technology today.
What makes this study fascinating is how it bridges the gap between technical research and what's happening on the ground at clinical sites right now. The researchers set out to answer a simple and straightforward question:
How can LLMs improve the way we match patients to clinical trials?
What they uncovered goes deeper than that, though. The study identifies exactly where current systems fall short and outlines how LLMs are stepping in to address those gaps.
What the Study Got Right About Today's Recruitment Crisis
The manual chart review bottleneck is real (and expensive)
Let's start things out with the numbers that make everyone uncomfortable. The study found that over 50% of aborted clinical trials fail due to low accrual rates. If you work at a research site, this part of the study probably feels all too familiar.
But here's where it gets really sobering. The research also reveals that experts can end up spending hours on a single patient. Think about that for a second. Hours spent on a single patient, and that's before you even know if they qualify!
Unfortunately, this is a process that pulls valuable staff away from higher-level work and delays FPI or “First Patient In.” Not to mention that the cost of that delay adds up fast, for both sites and sponsors.
Trial-specific systems are holding teams back
Another key insight the study focused on is tech limitations. Many modern recruitment tools still rely on “trial-specific” rules and workflows, which means you have to build custom workflows, train staff on new processes, and essentially reinvent the wheel each time you open a new study.
It's inefficient, it's expensive, and frankly, it's why many promising recruitment tools never get adopted in the long term.
However, the study revealed how LLMs can work as knowledge aggregators, pulling information from multiple sources and applying it across different trials and patient populations. Meaning, instead of building 20 different recruitment strategies for 20 different protocols, you can use one intelligent system that adapts to whatever you throw at it.
The result? Sites can access patient data, eligibility logic, and recruitment insights all in one place instead of managing multiple disconnected processes. That means faster matching, fewer delays, and more time spent on work that actually matters.
What This Research Means for Your Site
If you have spent any time in clinical research, you already know that recruitment is one of the biggest barriers to getting treatments to patients. This challenge is not new.
What makes this study different is that it's not just emphasizing the pain points we all know exist. Instead, it confirms how LLMs are actively reshaping the way sites operate today.
But here's the critical nuance: The gap between early adopters and everyone else is widening. While some teams are still buried in manual chart reviews and siloed systems, others are using AI tools to identify eligible patients faster, lower screen failure rates, and move confidently from protocol to enrollment.
So, how can you use this research to make better decisions about your own recruitment strategy?
It starts by asking the right questions:
Does it work with both structured and unstructured patient data?
Can it surface matches in real time without manual review?
Will it help reduce screen failures and improve patient quality?
Can it explain why a patient qualifies, or does it just give you a name?
Is it secure, compliant, and built for use inside existing systems?
If the answer to most of these is not a clear yes, it might be time to rethink what your recruitment strategy needs in order to keep up.
How Trially Checks Every Box the Research Calls For
This study clearly outlines what LLM-assisted recruitment should look like in theory, but some companies are already putting these concepts into practice with measurable results. Trially for example, is one of the few platforms that is delivering measurable results closely aligned with the study’s key findings. Here’s how:
Unlocking the full value of your patient data
The research emphasizes that patient data is represented as both structured and unstructured text in EHRs, including clinical notes, medications, images, labs, and pathology. The challenge? Many tools still only process basic demographics and miss the rich information buried in physician notes and clinical documentation.
Trially addresses this by providing real-time interpretation and summaries of rich medical data, including physician notes, demographics, diagnoses, medications, imaging, labs, and pathology reports.
As one of our site customers put it, “I can’t even express how much time Trially saves me. Especially with the employee transition and short staffing. It’s allowed me to keep the pace that I’ve needed given the amount of doctors we have, and the vastness of the database to begin with...Trially is a huge help!”
Quality over quantity in patient matching
The study also emphasizes the need to select a "precise and required size of patient population" and highlights how trial patient matching should assign labels like "eligible," "ineligible," or "irrelevant" based on comprehensive criteria analysis.
Trially delivers on this with impressive results. Matches are surfaced instantly on a 0-100% scale with no waiting or review required. Our platform achieves a 73% reduction in screen failure rates because we can identify and prescreen 10x more high-quality candidates on day 1.
The impact is straightforward. More eligible patients, fewer wasted referrals, and research teams that can focus their time on patients who actually qualify instead of sifting through unsuitable candidates.
Explainable matches builds trust with research teams
One of the key themes in the research is the growing need for explainable AI systems that clinical researchers can interpret and trust. The study specifically mentions that instruction fine-tuned LLMs for generating explainable trial patient matches have received attention, with criterion-level explanations showing potential for reducing screening time.
Trially tackles this by ensuring users can see exactly why patients may or may not qualify for specific trials by triple-clicking directly into their medical report summaries. This transparency aligns perfectly with the study's emphasis on “interpretable model predictions,” helping to reduce both screening time and decision hesitation among clinical teams.
Checking the compliance boxes (and then some)
Security and compliance are non-negotiable today, especially when working with sensitive patient data. The study notes that integration and compliance challenges are often the reason new tools fail to take hold.
Trially has prioritized this from the start by fully complying with HIPAA, SOC 2, FDA 21 CFR Part 11, and ISO 27001. More importantly, we connect through site-based API integrations with any EHR, CRM, or CTMS. In other words, our tech fits into the systems you already use without forcing you to change how you work.
No EHR? No problem. Trially prescreens directly from PDF, Excel, ads, etc.
Smarter feasibility
Finally, looking toward future applications, the research mentions "Interactive Trial Design" and "Collaborative Trial Planning" as promising directions where LLMs could help bridge the semantic gap between eligibility criteria and patient data.
With Trially AI Feasibility Explorer, sites and sponsors can assess population fit before the trial even begins. That means better-aligned trials, fewer dead ends, and more confidence when deciding whether to say yes to studies where they can realistically meet enrollment targets.
Start Making AI Decisions Your Clinical Trial Can Depend On
The research proves it. Your peers are already experiencing it. And ultimately, patients worldwide are depending on it.
This study confirms what forward-thinking sites have known for months, which is that LLM-assisted recruitment is already helping sites multiply enrollment and increase site revenues and ROI.
While the tools to address the challenges outlined in this research already exist, the real question is whether your site will take advantage of them now.
Experience What AI-Powered Prescreening Looks Like Now
Ready to see how the research findings translate into measurable results for your site? Trially offers the opportunity to experience firsthand what LLM-assisted recruitment can do for your enrollment rates, screen failure reduction, and operational efficiency.
Find qualified patients faster using tools that actually fit your workflow. No disruption to your current systems. Integrate with Trially AI in as little as one month and see how you can multiply patient enrollment in your upcoming trial. Get started today!
Schedule a demo and see how AI can help your site move from reactive recruitment to proactive enrollment.
