NTT Data, Chugai and Kindai University Hospital Test tsuzumi 2 to Find Clinical Trial Candidates in Medical Records
NTT Data said on November 13 that it is carrying out joint research with Chugai Pharmaceutical and Kindai University Hospital on using generative AI to find patients who could take part in clinical trials, Nikkei reported. The system searches treatment records in electronic medical charts for patients who meet a trial’s entry conditions, with the aim of easing the burden on medical staff and speeding up the development of new drugs.
The research uses a medical version of tsuzumi 2, the large language model developed by NTT, to compare the participation criteria set by a drug maker with the information in patients’ charts. Earlier programs mostly just judged whether a candidate was suitable or not. The generative AI also explains in writing why it decided a patient does not meet the conditions. For a patient whose test values fall short of a threshold, for example, it suggests a treatment intervention, and where a required observation period has not yet been completed, it shows how many more months remain before the patient could qualify. The aim is to make it easier for clinicians to check a patient’s condition and judge whether they can join a trial.
What it means in practice
The partners will test the accuracy of candidate selection and measure how much working time it saves. Results are due to be compiled in February–March 2027, and NTT Data plans to turn the work into a full service. It already launched a service in July that uses programming languages such as SQL and Python to pull candidates from hospital databases. Drug makers send the trial conditions, and the service analyses test values in electronic charts as well as doctors’ notes on the course of treatment and medical history; when it finds a match, it returns the patient ID to the hospital so doctors can invite the patient. In a technical trial with Kindai University Hospital covering lung cancer trials, it analysed data on about 3,000 patients, narrowing candidates by tissue type, gene mutation, disease stage and general condition, and cut the number of patients doctors had to review in detail by 97.4% and 99.4% in two trials.
NTT Data has partnerships with 11 medical institutions and applications from three drug makers, and is targeting more than ¥1 billion in combined sales from the two services in fiscal 2029. The project joins other Japanese efforts to put language models to work for clinicians, such as Sakana AI’s Namazu model powering an evidence finder for doctors, and recalls how hospital networks are sharing playbooks for safe imaging triage. What to watch is how accurate the AI’s explanations prove in the research and how quickly hospitals and drug makers sign up once the service launches.