Despite continued efforts to expand participation in cancer trials, enrollment remains low. Many potentially eligible patients go unidentified or are recognized too late, while uninformed manual screening can be prohibitively resource-intensive—creating a clear opportunity for technology to advance clinical trial matching (CTM) solutions.
Early rule-based CTM technologies have helped automate aspects of patient identification, yet their reliance on structured data can limit their utility. As a result, site teams may still face a substantial review burden and may miss opportunities to engage patients at the most appropriate point in their care journey.
The emergence of artificial intelligence (AI) and large language models (LLMs) introduces new ways to improve the accuracy and timeliness of patient-trial matching. As with any AI application, transparency is essential; effective CTM solutions must achieve improved accuracy while providing a view of the documents or data that led to the determination for patient inclusion or exclusion, ensuring clinical teams have confidence to act on the results.
Building on the Strengths of Structured Data for Smarter Trial Matching
Algorithmic matching approaches that rely on structured data (such as diagnosis codes, medications, and lab results) offer speed and scale, but they often miss the nuance that matters in clinical decision-making:
- Trial inclusion and exclusion criteria include complex temporal and contextual elements that don’t translate easily to rigid rules.
- Electronic Health Records (EHRs) contain much of this necessary detail—but it exists in clinical notes, pathology reports, and imaging findings that are inaccessible to CTM solutions relying on structured data.
To address this, significant engineering and data science effort has gone into extracting and pre-structuring clinical facts from unstructured notes to make them usable for structured-data-oriented models. In fields like oncology, where many nuanced data points are relevant, this creates a significant barrier to coverage of data needed for clinical trial matching across the many different trial contexts encountered in cancer care.
As a result, many CTM solutions return too many matches, too early in the care journey, leaving site teams with a substantial number of potential patient matches to sift through.
Why AI Makes a Difference
Emerging tools like LLMs unlock a new approach to patient identification and clinical trial matching; one that goes beyond checklists and coding systems to truly “understand” the patient record.
- Instead of trying to force nuance into structured fields, AI-based systems can reason over unstructured documents just as a human reviewer would.
- AI-enhanced CTM solutions can interpret eligibility criteria in their original form and apply them to individual patient charts, considering context, timing, and clinical intent.
- When done right, this enables greater accuracy and smarter prioritization, finding patients who are much closer to full eligibility and ready to undergo final screening.
Paradigm Health’s AI-powered Approach in Action
Paradigm Health recently partnered with a large community oncology practice to evaluate the impact of two CTM approaches across four therapeutic cancer trials. The goal was to understand whether AI-native technologies could improve the efficiency and accuracy of matching patients to complex eligibility criteria when compared to a rule-based approach.
The comparison included:
- A traditional, rule-based platform that relied on structured data and basic pre-parsed document fields.
- Paradigm Health’s AI-native matching solution, which combines structured data prefilters that are enhanced with LLMs capable of reasoning over full-text clinical notes and interpretation of nuanced trial eligibility criteria.
The results were striking:
- In three trials, the AI-native platform reduced screening volumes by 31–87%, substantially reducing CRC burden by surfacing a more focused set of patients to follow for last-mile screening.
- In addition to reducing the false positive rate, in the fourth trial, the number of patients surfaced increased by 32%, yet resulted in twice as many patients deemed potentially eligible after human review.
- Research staff satisfaction was high, driven by the AI-powered solution’s ability to prioritize patients who were closer to treatment decisions—reducing wasted effort on candidates who were technically eligible but not ready for near-term enrollment—and by its improved UX/UI and transparent access to source data for result verification.
Applying LLMs to patient matching streamlines site workflows and improves both the quality and timing of patient identification. These gains in accuracy have meaningful downstream effects: research staff can focus on patients with a higher likelihood of eligibility, completing full screening more quickly within the critical time window when a clinical trial remains an option. Together, these improvements create real momentum toward the broader goal of increasing trial participation and ensuring more patients can consider an interventional study as part of their high-quality cancer care.
What this Means for the Future of Trial Matching
This is more than a software upgrade—it’s a shift in approach. AI-native clinical trial matching holds promise to overcome limitations of rule-based filters by allowing a more nuanced interpretation of I/E criteria. The goal is to reduce false positives and move from quantity to quality in the number of patient matches. Rather than replacing human judgment, these tools enhance it, delivering curated, context-aware cohorts that let research staff focus on what they do best: evaluating and engaging the right patients at the right time.
The future of trial matching will require an improved understanding of context, nuance, and clinical complexity, helping sites and trial sponsors meet patients where they are.
To explore how your organization can achieve similar results, read the official abstract presented at the ASCO Quality Care Symposium 2025 ↗.