HomeAIAI Science & HealthFujitsu, NCCHE and Paradigm Test AI Clinical-Trial Matching

Fujitsu, NCCHE and Paradigm Test AI Clinical-Trial Matching

Fujitsu Japan, the National Cancer Center Hospital East and the US firm Paradigm Health say they have tested an AI workflow for finding clinical-trial candidates in Japanese hospital records, in a simulated, single-centre retrospective study. Fujitsu Japan announced the work on 19 November 2026. The underlying results were presented earlier as an ESMO 2026 poster, 132P, described by Paradigm Health on 25 October 2026. The figures are company-reported. The poster is not peer-reviewed. Fujitsu’s own caveat is the one that bounds the story: “These results represent the candidates identified during this joint research and do not guarantee their actual participation in clinical trials.”

The joint research ran from June to the end of October 2026. Fujitsu says an LLM was used to anonymise and structure electronic records for “3,027 lung cancer patients out of 22,334 respiratory patients at the NCCHE.” Paradigm then assessed “540 patients among these 3,027 with EGFR gene mutations, based on five criteria,” “identifying 57 priority candidates.” Physicians, Fujitsu says, “confirmed that 51 of these individuals were promising candidates.” Paradigm puts the same fraction at 51 of 57, or 89.5%. Fujitsu also says the workflow “reduced the number of patient records requiring manual review by approximately 89%.” Those two percentages measure different things: one is a physician agreement rate on the 57 priority names, the other is a cut in how many records still needed to be read by hand.

Paradigm, in its own account, adds that “0 of the 483 patients who were deprioritized by the platform were determined to be trial appropriate,” after oncologist review, and that the platform identified 35 patients who were potentially eligible for a RET-fusion trial but had not received that test; after review, it says, all 35 were deemed a priority for clinical review and follow-up testing. Those figures are Paradigm’s. The company also says Japanese records were translated into English and converted into a structured format by LLMs, with independent checks on the machine translation, and that the US-developed platform was hosted in Japan.

Fujitsu says it will now work toward commercialising a service that supports screening through to enrolment, with a “target launch by the end of fiscal year 2027.” That is a company plan, not a launch date with a named product already in hospitals. The study itself was a retrospective simulation of a trial aimed at patients with EGFR mutations, carried out at one centre. Final eligibility, Fujitsu notes, remains a decision for physicians, based on trial requirements and the patient’s consent.

AI is already being applied to oncology in other ways that should not be mixed up with this matching test. Isomorphic Labs and Merck have started an oncology collaboration to design drug candidates, which is a discovery deal, not a trial-screening study. Closer to the patient-identification problem, a study has reported that blood tests plus CT or AI can improve lung cancer identification, and IARC has launched INTERVENER, a tool for mapping barriers to cancer screening. Neither of those is a retrospective, company-reported trial-matching poster, and this poster is not a finding that more Japanese patients have been enrolled.

A simulated poster result is not a trial enrolment

The numbers in the Fujitsu and Paradigm texts are easy to over-read because they are tidy. Of 540 patients with EGFR-related findings, the platform flagged 57 people; physicians agreed that 51 of those 57 were promising, and Paradigm says none of the 483 deprioritised patients was later judged trial-appropriate. Fujitsu reports an approximately 89% cut in records that still needed manual review. In a separate slice of the same cohort, Paradigm says 35 people looked potentially eligible for a RET-fusion trial but had not been tested. All of that is what the partners say happened in a single-centre, retrospective, simulated exercise, later shown as an ESMO poster rather than as a peer-reviewed paper.

Fujitsu’s caveat is doing real work here and should be left in place. Identifying a “priority candidate” or a “promising candidate” in a historical record is not the same as that person consenting to a trial, meeting every eligibility rule at the time of enrolment, or actually receiving an investigational drug. A planned service by the end of fiscal year 2027 does not change the status of this study. The next facts to watch are a prospective test at more than one hospital, a peer-reviewed publication, or an actual service launch. None of those is in the 19 November Fujitsu release or the October Paradigm poster notice.

How we sourced this

The patient counts, the 57 and 51 figures, the “approximately 89%” manual-review reduction and the fiscal year 2027 target are taken from Fujitsu Japan’s 19 November release. The 89.5%, 0-of-483 and 35 RET-fusion figures, and the description of machine translation, are Paradigm’s. No enrolment or regulatory claim has been added.

Sources

Related stories

Isomorphic Labs and Merck Start an Oncology Collaboration is the closest recent TSN account of an AI project aimed at oncology, and is the next read, while remaining a discovery collaboration rather than a trial-matching study.

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. House Members Release an Outline of the SAFE AI Act
    AI Policy & Safety

More in this category

Latest stories

  1. House Members Release an Outline of the SAFE AI Act
    AI Policy & Safety

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.