Evaluation Study on "CocktailAI,"
a Generative AI Solution
for Medical Documentation,
Published in International Journal
PLOS Digital Health

Fitting Cloud Inc. (hereinafter "Fitting Cloud"), a consolidated subsidiary of FINDEX Inc., conducted a clinical trial in collaboration with the Department of Ophthalmology at Kyoto University Graduate School of Medicine (led by Senior Lecturer Masahiro Miyake et al.) to evaluate the effectiveness of "CocktailAI," a generative AI solution for medical documentation. We are pleased to announce that a paper summarizing the results of this research was published online on October 27, 2026, in the international academic journal PLOS Digital Health.

PLOS Digital Health is an open-access journal published by the Public Library of Science (PLOS) that covers a wide range of research aiming to transform healthcare and medicine using digital tools and data science.

Background of Joint Research and About "CocktailAI"

Under the "Work Style Reform for Physicians" promoted by the Ministry of Health, Labour and Welfare, reducing the operational burden on physicians has become an urgent need. In particular, tasks related to creating documentation—such as electronic health records (EHRs), discharge summaries, and clinical referral letters—impose a significant time burden, and efficiency gains through generative AI are highly anticipated.

In October 2023, Fitting Cloud initiated joint research with Kyoto University Hospital regarding labor-saving solutions for documentation using generative AI.

Developed as part of this initiative, "CocktailAI" is a solution that generates draft documents from EHR entries. In addition to extracting necessary information using a proprietary template structure, it addresses the risks of fake information ("hallucinations") and factual errors—major concerns for generative AI in medical settings—by displaying side-by-side comparisons of the original EHR entries used as grounds for the generated text.

To operate safely and effectively in actual clinical settings, establishing a "Clinician-in-the-loop" workflow—where a physician performs the final verification and edits—is essential. Therefore, this clinical trial was conducted to evaluate both completion time and document quality.

Study Overview and Key Findings

This study was conducted as a pilot randomized controlled trial using simulated patient records (mock EHRs) across six cases in ophthalmology. Kazuya Okamoto, President of Fitting Cloud, and members of its R&D team (Luciano Henrique de Oliveira Santos et al.) participated as co-authors of this paper.

Targeting 21 physicians (17 completed) from the Department of Ophthalmology at Kyoto University Hospital, the trial compared the creation time and quality of discharge summaries and referral letters across three approaches: the "Clinician-in-the-loop group" (physicians edited CocktailAI drafts), the "Clinician-only group" (physicians created documents manually), and the "LLM-only group" (generated without human review).

Improvement in Quality:

In multi-faceted quality evaluations by experts, the "Clinician-in-the-loop group" (where physicians edited AI drafts) outperformed manual creation ("Clinician-only group") across most categories, including medical accuracy, linguistic expression, and overall quality.

Changes in Creation Time:

While the primary analysis showed no clear statistically significant difference, sensitivity analysis confirmed an average reduction of 59.4 seconds in creation time for discharge summaries (referral letters tended to take longer due to manual revisions).

Limitations of AI Alone:

Output from AI alone ("LLM-only group") was the fastest, but its quality scores remained at the lowest level and failed to meet quality standards.

These results demonstrate the effectiveness of incorporating active physician involvement ("Clinician-in-the-loop") rather than relying entirely on AI, thereby reducing the workload of document creation while improving document quality. In addition to co-developing CocktailAI, Fitting Cloud was responsible for building the evaluation system (a dedicated web application) for this trial.

Future Outlook

In addition to integrating with the in-hospital documentation system "DocuMaker," "CocktailAI" is expanding its lineup of system integrations, including the standalone version "CocktailBar." Moving forward, we will continue to adapt to rapidly evolving AI models and enhance implementation flexibility, contributing to reducing physician workloads and improving healthcare quality across more medical settings.

Glossary

Clinician-in-the-loop:
An operational methodology in which expert physicians verify, edit, and give final approval during the workflow rather than using AI-generated output directly. This approach is emphasized to prevent the risk of medical accidents resulting from hallucinations (false outputs).

Related News (Fitting Cloud Inc. External Site)

Launch of Joint Research on Labor-Saving Document Creation Tasks in Clinical Settings Using Generative AI (October 31, 2023)
https://fittingcloud.co.jp/news/2023/nr_20231031.html

Launch of "CocktailAI," a Generative AI Solution for Medical Documentation (June 3, 2024)
https://fittingcloud.co.jp/news/2024/nr_20240603.html

※ Please visit the Fitting Cloud website above for the full press release and product details.

Media & Contact Inquiries

Fitting Cloud Inc.
Email: info@fittingcloud.co.jp

Note: This press release has been translated from the Japanese original for reference purposes only. In the event of any discrepancy between this translated press release and the Japanese original, the original shall prevail.