Oda Hospital: generative AI for nursing and physician summaries
【病院DX事例】生成AIで看護サマリー業務はどう変わる？ 織田病院の実践事例
B · Useful, limited First-hand account from the director of a regional Japanese hospital on three years of generative AI for clinical documents; credible practice detail, but the video is produced by the vendor and all figures are spoken, not documented.
Summary based on the full transcript·Original: about 36 minutes to watch
In short
A video interview published 14 October 2026 (recorded in July) on OPTiM's channel with Yoshimasa Oda, general physician and, since 1 July 2026, director of Oda Hospital in Kashima, Saga Prefecture. Since 2023 the hospital has used OPTiM's on-premises generative AI on electronic record data. The first version in early 2024 produced nursing summaries staff found unusable; the fix was rewriting how staff recorded daily notes and teaching the system how nursing plans link to records. Nursing summaries are now generated for almost every discharge, and physician summaries in the internal-medicine society's J-OSLER format followed in June 2026.
Key numbers
- 200 to nearly 300AI-generated nursing discharge summaries a month, about one per discharged patient
- 100+ to 200+physician summary generations in June versus July 2026, the first two months of the J-OSLER format
- under a minutetime to generate a physician summary from physician, nursing and rehabilitation notes with one button
- about 8%share of hospital staff who are foreign nationals, for whom the director sees AI as needed in daily communication
As reported by the source.
Japanese·Video·opens youtube.com
What it reports
- Trigger: overtime and staff turnover driven by documentation written after clinical work.
- First targets: nursing discharge summaries and physician referral letters, using an LLM that could run on-premises.
- Early 2024 trial failed; Oda told staff they were "too strict with the AI", meaning the source records were the problem. Record-writing practice was changed to be clearer and faster, alongside model changes.
- Nursing summaries: 200 to nearly 300 a month, roughly matching patient numbers.
- Physician summaries: one button, under a minute; picks up allergy, family and social history often missed in admission notes. Use rose from 100+ in June to 200+ in July.
- Oda approves most drafts with little or no editing and says the task "has disappeared" rather than become more efficient.
- Residents are kept off the tool until they can write summaries themselves and pass a check; he expects this rule to loosen within four to five years.
- Human responsibilities that remain: communication with patients and staff, and final responsibility for treatment decisions. New hires cite digital tools as a reason for joining.
How they got it
Interview with the hospital director, conducted by an OPTiM staff member. Transcript from Japanese auto-captions. Figures come from the speaker; no data or quality audit is shown.
Limits
- Vendor-produced video that closes with a lead-generation form and webinar promotion.
- No measurement of time saved, error rates in summaries, or clinical safety review.
- One hospital; usage counts are not the same as quality.
Why it matters for companies
The most transferable lesson is that the first failure was a data problem: summaries improved once the people creating the records changed how they wrote them. That applies to any AI built on operational records, from HR case notes to service tickets. The training rule for residents is a practical answer to skill erosion: hold juniors back from the tool until they can do the task unaided. On-premises deployment was the condition that made sensitive records usable at all.
Key passages in translation
Japanese original
業務が効率化したというよりはもうその業務がなくなったというような私としては認識でいます。
Rather than the work becoming more efficient, my sense is that the work itself has disappeared.
Japanese original
AIに厳しすぎって言いましたみんなに。
I told everyone they were too strict with the AI.
Japanese original
研修位の段階ではちょっと使うのやめときましょうとか、編集の中でもサマリーがもうかけるようになったね、大丈夫だねっていうわゆるこうテスト的なものをこうクリアしたら使うようにしましょう
At the resident stage, let's hold off on using it; once they can write summaries and pass a kind of test showing they are fine, then they can use it.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com