Gains:
- Ability to use artificial intelligence as a 'forgetting shield' and produce risk factor and danger sign reminder lists
- Ability to maintain that diagnosis, risk grading, value interpretation and referral decisions belong to the midwife/physician, not the artificial intelligence
- Ability to recognize the 'false trust' trap created by the confident but irresponsible response of artificial intelligence and filter each list with its own knowledge
The most critical skill of midwifery is to be able to distinguish between a pregnancy that is "going well" and a pregnancy that "requires attention." Most pregnancies proceed uneventfully; But a small group progresses to situations that endanger the lives of the mother and baby if not noticed in time. That's why in midwifery practice, risk early warning — systematically scanning for danger signs before and during the visit — saves lives. Artificial intelligence can be a valuable aid in this screening: it makes it faster to remind a danger sign, write clear warning text to the pregnant woman, and systematically list risk factors in a history.
However, in this unit we draw the strictest limit of the module:
Artificial intelligence does not make risk decisions, referral decisions, or diagnoses. The evaluation of conditions such as preeclampsia (a serious condition with high blood pressure and organ damage during pregnancy), gestational diabetes (high blood sugar during pregnancy) or the threat of premature birth is the clinical decision of the midwife and the physician. At best, AI is a “forgetting shield” — it works like a checklist so no sign is missed.
The line between "decision" and "reminder"
Let's make this distinction concrete. Imagine that a pregnant woman has a headache, flashes in the eye, and hand-face edema. The decision is to evaluate this situation in favor of preeclampsia and what to do - measurement, examination, referral to a physician; This is the midwife/doctor's job. The reminder is to put it in front of a list that says "If severe headache + vision change + edema are together, think of preeclampsia, check blood pressure and protein in urine." AI produces only the second, the reminder. The AI never makes the decision; The midwife filters each item in the reminder with her own knowledge.
Caution: An AI-generated list of risks can never be considered “complete.” He may hallucinately add a sign that does not exist or omit a real sign. The list supports the midwife's memory; It does not replace the midwife's knowledge.
Steps of early warning
- Pre-visit preparation: Have the AI list (de-identified) risk factors in that pregnant woman's history — for example, older age, multiple pregnancy, previous preeclampsia, high body mass index. This list is a reminder of what to look for when visiting.
- Systematic screening at visit: Keep the AI-generated red flag checklist on hand; Evaluate each item yourself.
- Educating the pregnant woman: Make the message "Which symptoms should you apply for immediately" expressed in plain language with AI? This allows the pregnant woman to give early warning at home.
- Clinical evaluation and decision: When the findings come together, the midwife/physician makes the evaluation and referral decision. AI does not enter this step.
- Documentation: What was observed and what was done is recorded in the midwife's note (unit 7).
In none of these steps does the AI set a “threshold.” For example, numerical limits such as "after what time should blood pressure be taken into account", "how much protein in the urine is significant", "how many weeks before is considered premature birth" are all defined in the institutional protocol and clinical evaluation. Asking the AI for these thresholds carries the risk of hallucination and pushes you to trust an out-of-protocol number. The AI just says “don't forget to check that headline”; It does not say "do this after this value". Keeping this distinction clear in your mind is the safety habit that works best in early warning of risk.
three mini cases
Case 1 — Warning text to the pregnant woman (education): A midwife prepares a "symptoms requiring urgent reference" card for high-risk pregnant women. He asks AI to write in plain language items such as severe and continuous headache, blurred vision, severe pain in the abdomen, significant decrease in baby movements, water breaking, bleeding. He reviews the card with the doctor and distributes it to 40 high-risk pregnant women. Within three months, 2 pregnant women applied early, saying "they had the symptom written on the card". The AI didn't decide; It simplified the message that allows pregnant women to apply at the right time.
Case 2 — Risk factor list (reminder): A second midwife has YZ summarize the history of a pregnant woman she has just taken over and asks for a list of "risk factors to consider". AI highlights the item “gestational diabetes in previous pregnancy”; The midwife sees this and adds sugar screening follow-up to the visit plan. The AI left out a family history that was not on the list; The midwife adds her own reading. The list was helpful, but not enough on its own.
Case 3 — False trust trap (limit): In a third example, the midwife gives AI blood pressure and complaint information and asks "preeclampsia?" He feels the urge to ask. He stops and thinks: this is a diagnostic question, the red zone. He doesn't ask the question; instead, he/she evaluates the findings with the physician. If he were to ask the AI, he would get a confident but unaccountable answer—which false confidence is the most dangerous trap.
Copiable templates
Role: Assistant (NOT decision maker) producing a "reminder list" for the midwife. Task: List item by item the POSSIBLE risk factors in the de-identified history below. Add a "why caution" note next to each item. Rules: DO NOT make a diagnosis, DO NOT SAY "refer/do not refer", do not give a risk RATING. Write the following at the end of the list: "This list may be incomplete or incorrect; the evaluation belongs to the midwife/physician." Story: [story]
Role: Writer of a simple warning card for pregnant women. Task: Write a draft card titled "What symptoms should you consult a healthcare provider IMMEDIATELY?" Each symptom is one line, very simple. Rules: Describe the symptoms in everyday language, not in medical terms (short term can be given in parentheses). ADD dose/medication/diagnosis. Card maximum of 10 items. Leave [BLANK] at the end for the emergency contact line.
Role: Pre-visit checklist assistant. Task: Create a general "danger sign screening" checklist (headings such as headache/vision, edema, bleeding, water breaking, baby movement, fever, urinary complaints, etc.). Each item should be marked YES/NO. DO NOT WRITE the value/threshold/decision; the midwife will fill in that column.
Role: Simplification assistant. Task: Rewrite the following medical warning text in a way that an expectant mother with limited literacy can understand. Adding/removing information. Each sentence should not exceed 12 words. Text: [text]
Weak prompt / Strong prompt
Weak: "Does this pregnant woman have preeclampsia? Her blood pressure is 145/95."
This prompt falls into the red zone; Requests diagnosis from AI. The response is confident but irresponsible and may lead the midwife into false confidence.
Strong:
Role: reminder list assistant (not decision maker). Create a checklist of danger signs that MUST BE MINDED in a pregnant woman with severe headache and hand-face edema. Diagnosis, threshold/decision making. Write "Evaluation belongs to the midwife/physician" at the end.
The strong version asks for a reminder, not a decision; It sets the limit and responsibility sentence from the beginning. The difference is placing the AI in the right role — forgetting shield.
Decision or reminder: quick table
Status
This is a...
who does
"Is this chart preeclampsia?"
diagnosis
midwife/physician
"Should I refer this pregnant woman?"
referral decision
midwife/physician
"Remind me of risky symptoms"
reminder
AI strains draft, midwife
"Write an emergency application card for the pregnant woman"
Training text
AI draft, midwife/physician approves
"Is blood pressure 145/95 abnormal?"
Value interpretation
midwife/physician
"List risk factors in history"
reminder
AI completes draft, midwife
Common mistakes
- Asking AI for diagnosis/referral: The most common and dangerous mistake; is the red zone.
- Considering the AI list "complete": The list may contain missing or fabricated items; The midwife completes it with her own knowledge.
- Leaving the threshold/value interpretation to the AI: Questions like "Is 140/90 high?" are clinical evaluations.
- Raising anxiety in the warning card: Use clear and calm language instead of language that will cause the pregnant woman to panic.
- Giving a history with identification information: De-identification is essential even in risk screening.
- Distributing risky pregnancy material without physician approval: Emergency application criteria have clinical content; must be approved.
In summary
Artificial intelligence is a "forgetting shield" in risk early warning: it lists risk factors, creates a danger sign checklist, and writes a simple warning text to the pregnant woman. But it does not make a diagnosis, rate the risk, make a referral decision, or make a value interpretation — all of these are the clinical decisions of the midwife and physician. Never consider the AI list "complete"; Filter each item with your own knowledge. The most dangerous trap is the false confidence created by the AI's confident but irresponsible response.
Application task
Choose a high-risk pregnancy that you see frequently in your clinic (anonymous). Create a draft "emergency referral symptoms" card with AI, then review and correct it with a physician. Also test yourself on the AI: consciously write a “diagnostic question,” read the response, and make a note of three reasons why you should not trust that response.
checklist
- [ ] I did not ask AI for diagnosis/referral/value-interpretation; I just wanted a reminder/text.
- [ ] I completed the risk list with my own knowledge and removed the wrong items.
- [ ] I reviewed the warning card sent to the pregnant woman with the doctor.
- [ ] The language of the warning text is simple and calm; It does not raise anxiety.
- [ ] I removed identifying information before giving the story.
- [ ] I added the note "evaluation belongs to the midwife/physician" at the end of the AI list.