Season · AI, Trust & Human Oversight · 7 of 16
When a nurse records their assessment, the system may also suppress further alerts.
Concept of the Week · One response, two effects
Recording an assessment and controlling later alerts are two different effects of the same response. Clinical decision support should show both effects before a nurse chooses a response. Otherwise, they must remember a rule that the response label does not explain.
Control rule: If a response suppresses further alerts, show which alerts and for how long before it is selected.
In one documented emergency-department system, a nurse opening a patient's chart sees an artificial intelligence (AI) alert for possible sepsis. They can select “No infection suspected” to record their assessment. That response also suppresses further alerts from this advisory for eight hours. The response label alone does not explain that second effect.
A clearer design would place the affected advisory and the eight-hour interval beside the response, before selection. After saving, it would keep the time when suppression ends visible. The nurse could then see why this advisory will not appear again during that period. Suppression reflects the earlier response; it does not establish that the patient's condition has improved or remove the need for clinical review.
The response label, confirmation, and alert timer must describe the same effect, so a nurse does not have to reconcile conflicting messages.
Ship this by testing response wording and suppression feedback with representative nurses during realistic chart reviews, including deterioration during suppression, then revising and retesting against defined usability and safety criteria before release.
From a nurse recording an assessment without seeing its alert consequences to the same nurse seeing the suppression interval and end time.
Asset & Resource
Design and human factors teams can use this Alert Consequence Check with clinicians when a response to an AI advisory also controls later alerts.
When several tasks compete for attention, people may check a trusted system less closely (automation complacency). Include a task where no further advisory appears, and observe how the clinician interprets that absence.
Map both effects: Record what each response saves and what it changes about later alerts.
Check the wording: Before selection, observe whether the clinician understands both effects without opening help.
Name the scope: Check whether they can identify which advisory is suppressed and for how long.
Follow the interval: After saving and reopening the chart, check whether they can tell when suppression ends.
Test the quiet period: Introduce changed clinical information and observe whether the missing advisory is mistaken for reassurance or delays a needed review.
Use the findings in the iterative evaluation above and repeat it when response wording or suppression rules change. Clinical teams must define the appropriate review actions for each task. This is a proposed design test guide, not a validated clinical checklist.
For further reading, this study of AI-supported sepsis alerts describes response options and their effect on subsequent alerts.
Light Wisdom & Reflection
“The more advanced a control system is, so the more crucial may be the contribution of the human operator.”
Where does a response in your interface do more than its label tells the person choosing it?
Mindful Practice
When opening a window at home, pause to notice the room.
What changes besides the temperature?
Which sounds or movements do you notice now?
Next time, notice how one small choice changes more than you first intended.
Best,
Andreas Walden
Share this with someone designing how clinical responses affect alerts.