Before you post about a case
Safe-Posting Guidance
AIiHC exists so people can learn from real experience - including things that went wrong. That is only possible if patients cannot be identified and if you are not putting yourself at professional risk.
Read this once. It takes four minutes and it will save you from the two mistakes that matter.
Your responsibility
There is no pre-publication review in most channels. Nobody at AIiHC verifies that your case is de-identified, that you had permission to share an image, or that your employer allows it. That responsibility is yours, completely, every time you post.
AIiHC will act when something is reported or noticed, but that is after the fact. By then, other members may already have seen it.
Part 1 - Making a case unidentifiable
The test: could someone who was there, or someone who knows the patient, recognise them from your post?
Not “have I removed the name?” - “could anyone work out who this is?”
Remove or change these
| Never include | Instead |
|---|---|
| Name, initials, record number, NHS or insurance number | Nothing - omit entirely |
| Date of birth, exact age if unusual | “a man in his 60s” |
| Exact dates of admission, scan, procedure | “last spring” or “earlier this year” |
| Your hospital, department, or city combined with case detail | “a district general hospital” |
| Rare diagnosis plus rare demographic plus specific location | Change what isn’t essential to the point |
| Direct quotes from a patient or family | Paraphrase, or omit |
The combination problem
This is the one people miss. No single detail identifies anyone. Four details together often do. A 34-year-old with a rare condition, at a named type of unit, in a stated month, is identifiable to anyone who works there - even with no name.
Ask yourself: if I removed the clinical point and kept only the descriptive details, how many people in the world would that describe? If the answer is “a handful,” change something.
Change what doesn’t matter
If the patient’s age isn’t relevant to what the AI tool did, don’t state it. If the specific rare tumour type isn’t the point, say “an uncommon finding.” You are allowed to alter non-essential details to protect identity - that is not dishonesty, it is standard practice.
Children and deceased patients
Rarity makes identification easier. Families may be identifiable even when the patient is not. Both are commonly assumed to be lower-risk; both are usually higher-risk.
Part 2 - Images and video
Never post
| Never | Why |
|---|---|
| DICOM files in any form | Metadata is embedded throughout the file and survives most editing. Name, ID, date, institution, scanner, accession number |
| Any file straight from PACS, a scanner, or a clinical system | Same reason - assume it carries hidden data |
| Screenshots that include the patient banner, worklist, or any part of the system chrome | The most common real-world leak. The banner sits at the top of every PACS and EHR screen |
Before you post any image or video
- Export as a flat image - JPEG or PNG. Never the original clinical file format.
- Crop to the anatomy or the finding only. Nothing else in frame. No borders, no toolbars, no side panels, no worklist.
- Check for burned-in text - patient name, ID, date, laterality markers with identifiers, institution watermarks. These are painted into the pixels and no export removes them. Cover or crop them.
- Strip metadata. Even a JPEG carries EXIF data, and a phone photo of a screen carries GPS. Screenshot the image again, or use a metadata-removal step, before uploading.
- Look at the whole frame once more. Reflections in a monitor, a badge, a corner of a chart, a colleague’s face, a wristband.
For video: the same rules apply, plus check every frame. A banner that appears for one second at the start is still a disclosure. Trim it.
Distinctive anatomy is itself an identifier. An unusual implant, a rare congenital variant, or a striking finding can identify a patient to anyone who has seen that case. If the image is remarkable enough to post, consider whether it’s remarkable enough to recognise.
The practical test: if the image were printed and left on a train, could anyone work out who it is or where it was taken?
Permission is separate from de-identification
Many organisations require consent or internal approval before a clinical image leaves the institution, even fully anonymised. Some regulators and professional bodies require the same. Check your own organisation’s policy before you post - that is your responsibility, and AIiHC cannot check it for you.
Part 3 - Protecting yourself professionally
The test: would you be comfortable if your employer, your regulator, and the vendor all read this post?
Do not post
| Never | Why |
|---|---|
| Contract terms, pricing, procurement documents | Almost always commercially confidential |
| Internal incident reports or investigation findings | Belongs in your organisation’s system, not here |
| Unpublished evaluation data your employer owns | Not yours to share |
| Named criticism of an identifiable colleague | Professional and legal risk |
| Confidential correspondence with a vendor or regulator | Not yours to share |
| Anything under an NDA | Obvious, but easily forgotten mid-story |
You can still tell the story
You do not need to disclose the document to describe what you learned from it.
| Instead of this | Write this |
|---|---|
| “Our contract with the vendor says X” | “Our contract had a clause about X - worth checking yours” |
| “The incident report found Y” | “We found Y when we looked into it” |
| “The vendor told us in writing that Z” | “We were told Z” |
Naming vendors: you may name a product and describe your experience with it factually. You may not misrepresent it, and you must disclose any affiliation. If your account is critical, stick to what you observed and can support.
Other people in your story
Nobody in an account of something that went wrong should be identifiable unless they have agreed to it.
Part 4 - The four-question check
- Could anyone identify the patient? If a colleague who was there would recognise the case, change more.
- Am I sharing something my employer owns? If it came from a document, describe the learning, not the document.
- Have I disclosed any affiliation? If a company whose product this is pays you in any way, say so in the post.
- Would I be comfortable if my employer, regulator, and the vendor read this? If no, don’t post it.
Posting is not reversible
You can delete a post, and you should if you realise something is wrong. But deletion removes it from the community - it cannot recall what someone has already read, saved, or screenshotted. This is why the check happens before you post, not after.
If you get it wrong
If you realise after posting that something identifies a patient or breaches a confidence, delete it yourself immediately and message a moderator. Acting quickly is treated well.
If a moderator finds it first, it is removed straight away - patient-identifiable material is the one thing removed before any discussion. Depending on what happened, this may be a correction, a removal, or removal with account restriction.
Repeated failures to follow this guidance will end your membership.
Need to report something?
Use the private concern route. If an AI system is causing or may cause harm now, use your organisation’s incident and escalation process immediately. AIiHC is not an emergency reporting route.