Chugai and partners evaluate AI-assisted clinical trial patient identification using RWD

Chugai Pharmaceutical Co., Ltd. has launched a collaborative research project with Kindai University Hospital, NTT Inc., and NTT DATA to evaluate whether large language models can improve the identification of eligible clinical trial participants using real-world clinical data. The study will compare AI-assisted and conventional approaches against clinician assessments.

The Baseline

• Chugai and its partners will compare LLM-based, rule-based, and hybrid approaches for identifying eligible clinical trial participants.
• The evaluation will measure both identification accuracy and operational efficiency using physician and clinical research coordinator assessments as the reference standard.
• The findings could inform the responsible use of AI and real-world data to support clinical development while maintaining clinician oversight.

Chugai Pharmaceutical Co., Ltd. and Kindai University Hospital have announced the launch of a joint research project with NTT Inc., and NTT DATA to evaluate whether large language models (LLMs) can improve the identification of eligible participants for clinical trials using real-world data (RWD) generated during routine clinical care.

The research, which began in June 2026 and is scheduled to continue through March 2027, will assess whether AI can support one of the most resource-intensive stages of clinical development: matching patients to the eligibility criteria defined in a clinical trial protocol. The project has been approved by the Research Ethics Committee of the Kindai University Faculty of Medicine.

Identifying suitable trial participants typically requires physicians and clinical research coordinators (CRCs) to manually review electronic medical records against detailed inclusion and exclusion criteria. This process can be labor intensive and contribute to delays in participant enrollment, affecting the overall timeline for clinical development.

Within the collaboration, Chugai is providing the clinical trial protocol and eligibility criteria used throughout the evaluation, allowing the researchers to assess AI-assisted patient identification within the context of an active clinical development program. Kindai University Hospital is providing electronic medical records and other clinical data, while physicians and CRCs will independently identify eligible patients to establish the reference standard for comparison.

Using the protocol criteria supplied by Chugai, the researchers will evaluate three approaches:

1. A conventional rule-based method using Python and SQL
2. An LLM-based approach
3. A hybrid method combining both techniques

The outputs from each approach will be compared against clinician assessments to measure the accuracy of patient identification. 

Beyond accuracy, the study will examine the time required to identify eligible participants, changes in physician and CRC workload, and whether AI-assisted approaches can shorten the lead time to trial enrollment. NTT DATA is responsible for the technical evaluation of the rule-based and comparative approaches, while NTT will deploy its Japanese-developed LLM, tsuzumi 2, including a healthcare-specific version that has undergone additional pre-training using published medical literature. According to the collaborators, the model has been designed to support applications involving sensitive healthcare information while emphasizing data governance.

The research builds on technical verification work announced earlier in 2026 by Kindai University Hospital and NTT DATA. Unlike the earlier evaluation, the current study examines AI-assisted patient identification using real clinical trial eligibility criteria and operational clinical workflows.

For clinical development, the study reflects growing interest in applying AI to improve trial recruitment while maintaining clinician oversight. The researchers emphasized that AI-generated recommendations are intended to support physician decision-making rather than replace it, with final eligibility decisions remaining the responsibility of clinicians. If successful, the findings could provide further evidence on how RWD and LLMs can be integrated into trial recruitment workflows while maintaining methodological rigor, operational efficiency, and appropriate governance.