Joint Research Launched Using Real-World Data and Generative AI (LLM) to Improve Accuracy and Efficiency of Clinical Trial Patient Recruitment

Kinki University Hospital's Department of Oncology (hereinafter, Kinki University Hospital), Chugai Pharmaceutical Co., Ltd. (hereinafter, Chugai Pharmaceutical), NTT Corporation (hereinafter, NTT), and NTT DATA Corporation (hereinafter, NTT DATA) have launched a four-party joint research project (hereinafter, this research) in June 2026 to verify the accuracy and efficiency of the clinical trial patient recruitment process using real-world data accumulated in actual clinical practice and large language models (LLMs), an AI technology. In this research, Kinki University Hospital's electronic health record data will be used as the target, and the extraction methods combining the conventional approach with LLMs will be compared and evaluated based on the eligibility criteria defined in the clinical trial protocols formulated by Chugai Pharmaceutical. Using the judgment results of physicians and clinical research coordinators (CRCs) as a benchmark, the effectiveness in actual operation, reduction in workload, and contribution to shortening the lead time until patient enrollment will be comprehensively verified.

[Background] In the clinical development of new drugs, the period until the start of a clinical trial and the enrollment period for trial participants significantly impact the time to market. In particular, the extraction of potential clinical trial patients requires physicians and CRCs to individually review patient information based on the eligibility criteria defined in the clinical trial protocol, which has historically demanded significant time and effort. As a result, it has been pointed out that patient enrollment often does not proceed as planned, affecting the overall trial schedule. In recent years, with the increasing utilization of real-world data accumulated in actual clinical practice, the application of LLMs, which can interpret patient information, including unstructured data, across various sources, has garnered attention. By leveraging these technologies, improvements in the accuracy of potential clinical trial patient extraction and increased efficiency in the extraction process are expected.

[Overview of This Research] In this research, potential clinical trial patients will be extracted using LLMs based on the eligibility criteria defined in the clinical trial protocols formulated by Chugai Pharmaceutical, utilizing Kinki University Hospital's electronic health record data and other sources. For the extraction process, NTT DATA will conduct technical verification of patient extraction using LLMs and rule-based methods, leveraging its track record in secure information management and data operation design cultivated through the operation of the medical information utilization platform "Sen-nen Karte®," as well as its expertise in the utilization and analysis of medical data. This technical verification will utilize "tsuzumi 2," a large language model developed by NTT, which is purely domestically produced. It is characterized by its ability to handle data including sensitive information under a design that emphasizes data governance. Furthermore, verification will be conducted on "tsuzumi 2" as a medical-specialized model that has undergone continuous pre-training on medical literature and other public documents.

As extraction methods, 1 a rule-based extraction method using Python and SQL, 2 an extraction method utilizing LLMs, and 3 a combined method of 1 and 2 will be implemented. The accuracy of potential clinical trial patient extraction will be evaluated by comparing their results with the judgment results from physicians and CRCs. Concurrently, the time required for extracting potential clinical trial patients, as well as changes in the workload and content of work for physicians and CRCs, will be examined to verify whether it leads to a reduction in the lead time until patient enrollment from both the accuracy and efficiency perspectives of the extraction process.

This research is scheduled to be conducted until March 2027, following approval from the Ethics Committee of Kinki University Faculty of Medicine, etc., at Kinki University Hospital.

Overview of This Research

[Significance of This Research] This research quantitatively verifies whether the improvement in accuracy and efficiency of potential clinical trial patient extraction through LLM utilization leads to a reduction in the lead time until patient enrollment. This will contribute to shortening the overall trial period, accelerating the development speed for pharmaceutical companies, streamlining the clinical trial implementation system at medical institutions, and creating an environment where patients can access new treatment options more quickly. It is important to note that AI output results are intended to support physician judgment, and the final clinical judgment will be made by the physician.

[Roles of Each Organization]

* Kinki University Hospital: Provision of medical data, extraction of potential clinical trial patients, and comparison/evaluation of extraction accuracy and process efficiency. * Chugai Pharmaceutical: Provision of clinical trial protocols (eligibility criteria) and cooperation in evaluation. * NTT: Execution of technical verification for potential clinical trial patient extraction using LLMs. * NTT DATA: Execution of rule-based potential clinical trial patient extraction and comparison/evaluation of extraction accuracy and process efficiency.

This research is based on the knowledge gained from the technical verification announced in May 2026 by Kinki University Hospital and NTT DATA. Specifically, it evaluates the effectiveness and applicability to actual operations of potential clinical trial patient extraction using real-world data under the eligibility criteria defined in ongoing clinical trial protocols for patient enrollment. Based on the results of this research, the involved companies will consider the possibility of social implementation of a clinical trial patient recruitment platform utilizing real-world data and AI through further collaboration with medical institutions and pharmaceutical companies.

[Notes] Note 1 Real-world data: Medical data accumulated in actual clinical practice, which is increasingly being utilized for clinical trials, new drug development, and improvement of medical quality. Note 2 Large Language Model (LLM): An artificial intelligence technology that learns from vast amounts of text data to understand and generate text. It is widely used in the field of natural language processing. Note 3 Clinical trial protocol: A plan that predefines the objectives, methods, and participant criteria for a clinical trial to investigate the efficacy and safety of a new drug. Note 4 Eligibility criteria: Conditions for participation, such as age and symptoms, defined to ensure the safety and scientific reliability of a clinical trial. Note 5 Clinical Research Coordinator (CRC): A healthcare professional who coordinates between physicians, trial participants, pharmaceutical companies, and other stakeholders to support the appropriate conduct of clinical trials. Note 6 Time to market: The point at which a new drug is approved by regulatory authorities and becomes available for sale and use in medical institutions. Note 7 Sen-nen Karte: The name of a project implemented by the General Incorporated Association Life Data Initiative and NTT DATA Corporation, certified by the government, based on the Act on the Protection of Personal Information Held by Medical Institutions and the Act on the Promotion of Advanced Medical Care through the Use of Personal Health Records (Next Generation Medical Infrastructure Act). Note 8 Rule-based method: A method for extracting potential clinical trial patients by pre-programming the conditions defined in the clinical trial protocol. Note 9 tsuzumi 2: A large language model developed by NTT Corporation. It supports processing of various languages, including Japanese. Note 10 Python: A programming language widely used for data analysis and system development. Note 11 SQL (Structured Query Language): A language for operating databases. It is used for data extraction and aggregation based on conditions. Note 12 Receipt data: Data created by medical institutions to claim medical fees, containing information such as treatment details and prescribed medications. Note 13 DPC survey data: Data that organizes diagnostic names and treatment details in inpatient medical care, containing information for understanding the patient's condition and treatment overview. Note 14 Technical verification regarding clinical trial patient extraction using LLM: Verification will be conducted using technology developed by NTT Computer & Data Science Laboratories. Note 15 Technical verification announced in May 2026 by Kinki University Hospital and NTT DATA: An initiative that conducted technical verification of potential clinical trial patient extraction using real-world data and evaluated its effectiveness. (Reference: https://www.nttdata.com/global/ja/news/topics/2026/052500/)

* "Sen-nen Karte" is a registered trademark of the General Incorporated Association Life Data Initiative in Japan. * "tsuzumi" is a registered trademark of NTT Corporation. * Other product names, company names, and organization names are trademarks or registered trademarks of their respective companies.