Skip to content

Unit C-1 Protecting Patient Information: A 3-Second Check Before You Type

Related competency: C1 | Estimated time: 10 minutes

Why This Skill Matters

A resident, trying to quickly finish a discharge summary draft, was about to paste chart notes containing a patient's name, ID, and admission date directly into a generative AI. Typing into a general-purpose consumer generative AI service can feel like it stays entirely on your own screen, but in reality it is sent over the internet to the provider's server. With general-purpose consumer services, the input may be used for training the model or for quality improvement, and there is no way to fully retract content once it has been sent. Healthcare professionals are bound by a duty of confidentiality, and the handling of patient information is also governed by personal-information-protection laws and by your own institution's rules. The first checkpoint for using generative AI safely, on an ongoing basis, is the habit of asking, "Is it okay to send this text outside, before I type it in?"

Core Concepts

  • Typing is "sending." Typing text into a general-purpose consumer cloud service is the same as sending data to the provider's server. Input may be used for training or quality improvement, and it may be seen by people at the provider, for example during troubleshooting. There is no way to retract it once sent. Note that locally run models, and organization-approved services whose contracts guarantee that input is not stored or used for training, operate under different conditions -- when you need to handle real data, check which environments your institution has approved.
  • Removing the name alone is not enough. Beyond direct identifiers such as name, patient ID, and date of birth, an individual can also be identified through a combination of details such as admission date, facility name, a rare disease name, age, region, and occupation (re-identification). Replacing a name with initials is not anonymization.
  • The 3-second check before you type. Before you hit send, confirm just three things.
    1. Does it contain information that could directly identify the individual (name, ID, date, facility name, etc.)?
    2. Does it contain information that could identify someone through a combination of details (a rare disease plus age plus region, for example)?
    3. Does real data actually need to be entered at all -- could a fictional case or an abstracted description accomplish the purpose instead?
  • Three tools for processing. When you need material close to real data, dilute the information in the following order.
    • Deletion: Remove identifiers that are not needed for the purpose, such as name, ID, and facility name (delete them rather than replacing them with symbols).
    • Abstraction: Convert age into a decade (e.g., "in their 70s"), dates into relative expressions (e.g., "hospital day 3"), and place names into deletions or broader regions.
    • Fictionalization: When the purpose is teaching material or practice, build a fictional case rather than using a real patient (the exercise in this unit uses this method).

Copyable Prompt

The following is a practice prompt, using a fictional case memo, for training yourself to spot what could lead to identifying an individual. Paste it as-is into whatever generative AI you have on hand.

[Exercise: Practicing the pre-input check]
Below is a fictional case memo created for practice. Assuming this text
were entered as-is into a generative AI service, identify every part
that could lead to identifying the individual, and for each one, suggest
whether it should be deleted or how it should be abstracted. Finally,
show the full text rewritten into a form that would be safe to enter.

--- Fictional case memo (for practice) ---
On June 2, 2025, Mr. Ichiro Sato (82 years old, male, a former employee
of XX City Hall, Patient ID: 2201456) was admitted to the Department of
Gastroenterology at XX City Central Hospital with a chief complaint of
melena. He lives in the Nishimachi area of XX City with his wife, just
the two of them. Five years ago, he underwent endoscopic treatment for
early-stage gastric cancer at this same hospital. On the day after
admission, upper gastrointestinal endoscopy revealed bleeding from a
gastric ulcer, and hemostatic treatment was performed.
--- End ---

Steps

  1. Paste the prompt above into whatever generative AI you have on hand (ChatGPT, Claude, Gemini, Copilot -- any of these will do) and run it.
  2. Before reading the AI's list, first try counting how many "identifying details" you can find yourself (how many are there besides the name and ID?).
  3. Compare the AI's list with your own answer. Pay attention both to what you missed and to what the AI missed.
  4. Re-check the "rewritten full text" the AI proposed, using your own eyes and the three points of the 3-second check (direct identifiers, combinations, and whether it was even necessary).
  5. If you have time, write your own fictional case memo drawn from your own department or profession, and repeat the same exercise.

Example of Expected Output

The AI will typically return a list like the one below (a fictional example for illustration).

Parts that could lead to identifying the individual:
1. "June 2, 2025" -- the admission date. Combined with other details, this
   raises identifiability significantly. -> Replace with a relative
   expression such as "Day X" or "hospital day 1."
2. "Department of Gastroenterology at XX City Central Hospital" -- the
   facility name. -> Abstract to something like "a regional hospital," or
   delete.
3. "Mr. Ichiro Sato" and "Patient ID: 2201456" -- direct identifiers. -> Delete.
4. "82 years old" -> Abstract to "in his 80s."
5. "a former employee of XX City Hall" and "lives in the Nishimachi area
   of XX City" -- occupation and place of residence. Combined with other
   rare attributes, these raise identification risk. -> Delete.
6. "underwent endoscopic treatment for early-stage gastric cancer at this
   same hospital five years ago" -- links a treatment history to a
   specific facility. -> Reduce to simply "has a history of endoscopic
   treatment for early-stage gastric cancer."

What's easy to miss is not the "obvious" items like name and ID, but information such as occupation, place of residence, and a treatment history linked to a specific facility -- details that only become identifying once combined with other information.

Common Pitfalls and How to Handle Them

  • Being told "this is safe" even though the AI's rewrite was incomplete: The AI's own check is itself generative AI output, and it is not a guarantee. Always run the 3-second check yourself, with your own eyes, at the end. This flow of "a human verifying AI output" is exactly the theme of Unit C-2.
  • Wondering, "surely it's fine to leave this much in": When in doubt, delete it -- that is the guiding principle. In most cases, the original purpose (drafting a summary, preparing material for a study session, etc.) can still be met after removing the detail.
  • Wanting to practice with an actual case: Do not use real data, even for practice. Practicing with a fictionalized case is entirely sufficient to meet the purpose of the exercise.

Safety Note

Do not use real patient information in this exercise either

Always carry out this exercise using a fictional case. Do not enter information about a real patient into a generative AI, even as material for practicing the anonymization process. If you accidentally send patient information, do not conceal it -- promptly report and consult according to your organization's rules on handling personal information.

Self-Check Quiz

You can check your understanding of this unit. There is no pass/fail judgment, and no record is saved.

Deliverable Feedback

At the end of the conversation where you ran the exercise, paste and send the feedback prompt from the Feedback AI page, and the AI will send back a review. When you send it, begin with the line "This is my Unit C-1 output." Writing into the conversation the details you found yourself, and how they differed from the AI's list, will make the feedback more specific. This submission is optional and is not a requirement for issuing the Area C certificate.

Next Steps

Let's move on to Unit C-2: Copyright, Bias, and Verification -- Etiquette Before You Use It. Or you can go straight to the Area C quiz (the only requirement for issuing the certificate is passing the quiz).

References (Optional)

  • 香田将英, "Ethical Challenges That Can Arise When Using Generative AI" (JSME ICT Education Committee Symposium, Session 5, 2026-04-21): Slide PDF / Recorded Talk
  • 村岡千種, "University Initiatives on Student Use of Generative AI" (JSME ICT Education Committee Symposium, Session 3, 2026-01-20): Slide PDF / Recorded Talk

This unit is self-contained even without consulting these resources.