Could AI help cancer teams find patients who might otherwise miss the chance to be considered for a clinical trial?
A collaboration involving Paradigm Health, Fujitsu and Japan’s National Cancer Center Hospital East offers an encouraging example of that idea being investigated.
The reported work uses a large language model to help review clinical information and identify records that deserve further attention. It is about supporting a difficult screening task—not allowing a computer to decide someone’s treatment.
Why Matching Patients to Trials Is Difficult
Clinical trials have detailed eligibility rules. Relevant information can be spread across medical notes, test results and a patient’s treatment history.
Finding a possible match may require identifying something that has not yet been tested, as well as recognising information already in the record.
An AI system that can help organise that material could give clinicians a more focused starting point. The important word is “help”: a shortlist still needs clinical review, up-to-date information and discussion with the patient.
What the Evaluation Reported
In its release dated 25 October, Paradigm Health describes a retrospective evaluation at a single Japanese cancer hospital.
One analysis covered 3,027 records and identified 35 patients who needed RET testing. That is a prompt for further investigation, not confirmation that all 35 had the relevant alteration or would qualify for a trial.
A separate analysis involved 540 EGFR-related patients. The system shortlisted 57, of whom oncologists prioritised 51 for further consideration.
Those are different analyses with different denominators. They should not be combined into a single success rate, nor should the 51 prioritised patients be described as 51 people enrolled in a trial.
Why This Is Still Encouraging
Finding relevant information is a genuine part of medical work. A useful research opportunity can be missed if a record is difficult to search or a possible testing need is overlooked.
The positive result here is that AI-assisted review produced a set of cases that clinicians could examine and prioritise.
That is a narrower claim than “AI finds the right cancer treatment”, but it is also a more practical one. It focuses on supporting the people already responsible for making careful decisions.
What Needs to Happen Next?
This account is based on a participating company’s release. A full independently assessed study and confirmation of the associated conference details were not available for verification for this article.
Prospective evaluation would be needed to understand how the approach performs in routine practice: whether it misses suitable patients, how much checking it requires, whether it saves time and whether more people ultimately gain appropriate access to trials.
Results from one hospital also cannot be assumed to transfer directly to another healthcare system. This is Japanese research, not evidence of an NHS rollout.
The promising direction is clear nonetheless: use AI to reduce information-searching burdens, keep clinicians in charge and judge success by what happens for patients—not by the size of a shortlist.
Primary company report: Paradigm Health, Fujitsu and National Cancer Center Hospital East: LLM-based clinical trial matching ↗ — 25 October 2026. Reported retrospective results; not an independently verified clinical-outcomes study.
Less hype. More real-world AI progress.