DEV Community

Cover image for AI Agent to Match Clinical Trials: An Oncology Case Study
Abto Software
Abto Software

Posted on

AI Agent to Match Clinical Trials: An Oncology Case Study

This post is a quick overview of an Abto Software blog article about AI agent for clinical trials.

In this overview, we explore how AI agents can transform the way patients find and access clinical trials. These digital assistants can check a patient’s health profile against complex eligibility criteria, scan continuously for new trials, and even handle first-line communication.

The result? Less manual work, quicker results, and a win-win for both patients searching for new treatment options and researchers looking to fill their studies.

The challenges of finding clinical trials that fit

Clinical trials connect patients with innovative, experimental treatments. But finding the right trial is far from simple. The process requires reviewing years of medical history, cross-checking strict eligibility rules, and sorting through countless trial databases.

No surprise then that over 80% of clinical trials face recruitment issues, leaving many patients without access to potentially life-saving therapies.

Now consider the complexity: each patient has a unique medical history, preferences, and limitations such as travel restrictions. On the other side, every trial comes with rigid inclusion and exclusion rules. Today, the process still depends heavily on manual searches in registries and endless phone calls or emails to study sites.

For patients with aggressive cancers, every week spent searching can have a serious impact on outcomes.

The solution: a personalized clinical trial AI agent

Abto Software has designed an AI agent proof of concept (POC) that combines advanced algorithms, natural language processing (NLP), and deep clinical knowledge to automate trial matching. Studies show that AI can perform this work with near-human accuracy, giving patients much faster access to relevant clinical studies.

Here’s how it works:

1. Initial patient profile registration

The first step is building a detailed patient profile:

  • The patient provides information on medical history, cancer type and stage, prior therapies, demographics, and biomarker results.
  • They can upload electronic health records or complete a guided questionnaire.
  • The AI agent uses NLP to parse unstructured notes and extract essential details.
  • If anything is missing, the agent asks follow-up questions such as:
  1. “Have you already received immunotherapy?”
  2. “Is the EGFR mutation still present?”

This ensures no crucial data is overlooked.

2. Intelligent search and retrieval

Once the profile is ready, the AI agent begins scanning trial databases. Unlike a simple keyword search, it applies intelligent retrieval methods—for example, generating unique keyword combinations or using embeddings tailored to the patient’s medical background.

This targeted approach lets the system recall over 90% of relevant trials while only reviewing a small portion of the database, making the process far more efficient.

3. Automated eligibility screening and matching

Next, the AI agent screens each trial against the patient profile:

  • If information is missing, the agent prompts for clarification.
  • Every trial is marked as eligible, ineligible, or “needs review.”
  • Summaries explain which inclusion/exclusion rules were satisfied.

Because eligibility criteria are often long and written in technical language, the agent’s NLP capabilities translate them into clear terms—checking cancer stage, lab thresholds, or biomarker presence automatically.

4. Ranked recommendations

After filtering, the agent delivers a ranked list of trials based on:

  • Match quality
  • Trial phase
  • Distance or location
  • Urgency of enrollment

Each trial listing comes with a plain-language summary, avoiding confusing medical jargon. This allows patients to understand their options quickly and clearly.

Further support: taking action

The AI agent doesn’t just stop at listing results. It can:

  • Provide site contact information
  • Draft emails with the patient’s relevant clinical details (e.g., “Patient with metastatic lung cancer, EGFR exon 19 deletion, ECOG 1, meets eligibility for NCTXXXXX.”)

This level of support speeds up the process and makes discussions with trial investigators more productive.

Continuous monitoring and updates

After the profile is registered, the AI agent keeps working in the background:

  • It monitors for new trials that match.
  • If the patient’s health changes (such as the discovery of a new mutation), the system automatically reruns the search.

What was once a one-off search turns into ongoing trial scouting, ensuring patients never miss opportunities.

Case study: Emily’s experience

Step 1. Getting started

Emily, 55, is living with late-stage lung cancer. Standard treatments have failed, so her oncologist suggests exploring clinical trials.

She signs up on the AI platform and shares her medical details.

Step 2. Profile enrichment

The AI agent processes her data and asks clarifying questions:

“What is your current ECOG status?”

“Do you have any significant health issues such as autoimmune disorders or heart conditions?”

Emily responds that she is self-sufficient and has controlled hypertension. The agent then pulls lab results from her health records to complete her profile.

Step 3. Finding candidate trials

The AI agent scans thousands of records. Within seconds, it identifies 15 trials targeting her specific mutation. All are in advanced development phases and either located nearby or offer travel support.

What might have taken weeks manually is finished in moments.

Step 4. Eligibility screening

The agent carefully reviews all 15 trials:

  • One ALK inhibitor study matches every inclusion criterion → flagged as a top choice.
  • An immunotherapy trial excludes patients with autoimmune diseases. Emily qualifies.
  • Another excludes patients with brain metastases. The system confirms Emily has none → marked eligible.

After filtering, Emily is left with five strong trial options.

Step 5. Final results

Her top recommendation: “High match: You meet 10/10 criteria for this ALK+ targeted therapy. Location: 50 miles away.”

Another: “Moderate match: You qualify based on prior therapies. Location: 200 miles, travel support available.”

Each summary is simplified into plain language so Emily doesn’t need to decode technical jargon.

Step 6. Getting qualified

Emily clicks “Contact site for me.” The AI drafts and sends an email to the trial coordinator.

Within a week, Emily is scheduled for screening—skipping endless calls and emails. The AI has accelerated her access to a potentially life-saving therapy.

Clinical trial matching agent: the benefits and impact

For providers & clinicians

The AI agent automates pre-screening, freeing doctors to focus on patient care. It enhances—not replaces—decision-making, offering faster, more accurate options to discuss with patients.

For patients

Patients save weeks of manual searching. They receive fewer false leads and clearer information, which is crucial when time is short.

In fact, an NIH study showed that doctors using AI for trial matching spent 40% fewer work hours on screening, with no drop in accuracy.

For trial sites & sponsors

Trial sites benefit from better-qualified candidates, leading to faster recruitment and smoother study progress. This directly tackles the 80% recruitment failure rate and helps new treatments reach the market sooner.

Broader impact

The agent also learns over time, identifying common reasons for exclusion or gaps in available trials. In the long term, this can highlight unmet medical needs and support more inclusive healthcare access.

How we can help

At Abto Software, we believe AI should simplify healthcare processes, not complicate them. Our clinical trial matching agent is just one example of how automation can make an immediate difference.

Our expertise includes:

Our services:

Book a strategy call and let’s discuss how AI can support your healthcare projects.

Top comments (0)