Interview
Senior emergency nurse
Walked us through real alarm handling, workflow and where settings get skipped.
01What the field told us
Three weeks at OLVG. One shift in the problem.
Observation
A&E floor & routing
Mapped how nurses move between beds, the central station and competing alarms.
Rapid feedback
Mood across the shift
"Adjusting limits for alarms" and "responding to alarms" drew the most overwhelmed responses.
👀
Nurses lean on context, not the alarm itself.
⚙️
Threshold adjustment is known — but underused.
🧠
Every extra alert adds cognitive load.
📍
Support only works if it fits the workflow.
The reframe · after the research
The problem wasn't alarm volume. It was threshold personalisation.
"How might we help nurses apply patient-specific threshold-setting at the right moment — without adding another layer of cognitive load?"
The reframed design question
02Why it happens upstream
Alarm fatigue isn't made at the alarm. It's made upstream.
🏥Hospital policy & defaultsMonitors arrive on generic factory thresholds
⚙️Thresholds stay genericThe one controllable point in the chain
🔔Volume of false alarmsMost signals aren't clinically actionable
📉Trust in the system dropsNurses learn to discount the monitor
👩⚕️Nurses rely on experiencePatient appearance & context over signals
⚠️Real alarms compete for attentionThe dangerous outcome
The leverage point
Instead of redesigning the alarm, we looked at what makes so many alarms in the first place: monitors left on generic settings.
Personalising thresholds to the patient is the one move that reduces the noise at its source — not after it has already fired.
03Design judgement
We judged every idea by the attention it would cost.
A smarter alarm is still another thing to manage. The strongest directions were the ones that asked the least of an already overloaded nurse.
AI filtering
RejectedWhy — hides clinical reasoning and risks oversimplifying overlapping conditions.
Wearables (Haptic & melodic alerts)
RejectedWhy — moves alarm fatigue into yet another channel to charge, wear and check.
Smart routing
RejectedWhy — more interruption. The root cause is too many signals, not quiet ones.
Full monitor redesign
DeferredWhy — promising at the source, but beyond the integration scope available with Philips.
04The filter we designed against
Five principles. Each one earned from evidence.
🎯
Cut alarms at the source
EvidenceMost alarms are non-actionable.
InterpretationA louder alarm won't fix noise.
DecisionSupport threshold personalisation.
🔕
No new alert channel
EvidenceNurses juggle screens, phones, talk.
InterpretationAnother signal = more switching.
DecisionAdd information, not alerts.
🧭
Protect clinical judgement
EvidenceConditions overlap and change.
InterpretationAutomation can oversimplify.
DecisionMake the action easier, not automatic.
📍
Sit at the point of action
EvidenceNurses act beside the monitor.
InterpretationA distant dashboard is too far.
DecisionStatus lives at the bedside.
📊
Make impact visible
EvidenceBehaviour repeats when results show.
InterpretationA cue alone doesn't close the loop.
DecisionFeed back the effect on alarms.
05The concept
One behavioural loop, built into the workflow nurses already have.
ILLUSTRATIVE PROTOTYPE DATAThe bedside cue supports action at the point of care; the data visualisation makes its effect visible at team level. Together they form one behavioural feedback loop.
—How Alarmease works
One loop, riding the workflow nurses already have.
Nurse workflowWhat already happens
🧑⚕️Patient arrives→
🩺Monitor connected→
⚙️Threshold review→
💓Patient monitored
Alarmease layerWhat we add
🔖Bedside cue→
✋Threshold adjustment→
🔄Status updated→
🖥️Nurse-station overview
Behaviour changeWhat it shifts
👁️Visibility→
✅Action→
📊Feedback→
🤝Shared awareness
Expected effectIllustrative
⚙️More personalised thresholds→
🔕Fewer unnecessary alarms→
🧠Reduced alarm burden
06The trigger
Making an invisible setting visible.
BeforeMonitor status is invisible. Generic or reviewed — no way to tell at a glance.
AfterA single bedside cue shows the threshold state, right where the nurse acts.
Point of action
Low effort
High visibility
Behavioural trigger
Prototype, not productTested as a physical cue to prove the behaviour. A future version integrates the status directly into the Philips monitor — so nothing is updated twice.
07My strongest contribution
I turned emotional feedback into operational feedback.
The feedback layer had to show the effect of personalisation — without feeling like it was judging nurses who were already doing their best.
Version 1 · Emotional
A mood, per shift
Critique: emotional feedback could read as blame during a stressful shift.
→
Version 2 · Operational
Practical information
- Beds reviewed
- Alarm trend vs. baseline
- Trustworthy
IllustrativeSame data, reframed around action — beds reviewed and alarm trend against the previous shift, built for clinicians.
1User feedback
“Emotional feedback felt judgemental during a stressful shift.”
2Interpretation
Nurses need practical information, not emotional evaluation.
3Design change
An operational dashboard — bed status and alarm trends, not moods.
Bed level overview
08My role across the project
From opening the problem to owning the feedback layer.
Phase 01
Research
- Interview questions
- Observation goals
- Hypothesis building
Phase 02
Concept
- Idea exploration
- Workflow-fit filtering
- Behavioural framing
Phase 03 · led
Data visualisation
- Information hierarchy
- What data to show
- Visual language
Phase 04
Testing & iteration
- Prototype versions
- User-test synthesis
- Post-test redesign
09What testing showed
A Wizard-of-Oz test, and four clear product moves.
"Should we adjust this patient's limits?"
Visible status sparked the conversation between nurse and doctor.
Keep monitor status visible & shared.
"If I already set it on the monitor, why update it again?"
A separate object duplicated work nurses had already done.
Automate status from monitor data.
"I don't change all the limits at once."
Nurses adjust each vital independently, not as one state.
Show status per parameter (HR, SpO₂, BP, RR).
"A total count doesn't tell me where to go."
Staff needed to know which bed needs follow-up.
Add bed- & parameter-level detail.
What changed because of testing
Kept
- Visible monitor status
- Shared visibility
Changed
- Automatic updates
- Bed-level overview
Removed
- Manual status updates
- Generic personalisation state
10Where it goes next
From a tested prototype to a monitor-integrated status.
Now
Current prototype
Physical bedside cue + team visualisation, tested as one loop.
→Next
Monitor integration
Status linked to Philips monitor data — no duplicate updates.
→Then
Automatic detection
Review status read automatically, per vital parameter.
→Later
Hospital dashboard
Bed- and parameter-level overview across the unit.
→Why this matters for Philips
- Works with existing monitoring workflows
- Encourages threshold personalisation
- Visibility without another alert channel
- Integrates into existing monitoring infrastructure
11What we learned
01
Behaviour beats alerts.
02
Workflow beats engagement.
03
Visibility drives action.
The strongest intervention wasn't the most advanced one — it was the one that fit naturally into how nurses already work.
12Reflection
What changed was how I judge an idea.
Before
I often judged ideas through originality and novelty.
After working with nurses
The strongest interventions are often the least visible ones.
The most successful concept wasn't the most technologically advanced — it was the one that asked for the least additional attention.
Scope & honesty
Alarmease was a 10-week concept, not a deployed product — no alarm reduction or patient-safety outcome was measured. Evidence is directional: one senior-nurse interview, co-creation with two nurses and one physician, and formative testing with two nurses. Interface data shown is illustrative; interface elements are recreations, not proprietary screens.
Team & attribution
Team: Negin Bokaeiolmousavi, Barbora Halanová, Andyka Jonathan, Duco Boomsma. I initiated the operational data-visualisation direction and developed it with Barbora. Partners: OLVG East, Amsterdam, and Philips.
AI transparency
AI tools supported writing refinement, early visual exploration, and contextual image generation. Research synthesis, decisions, and reflection were completed by the design team. AI-generated images and illustrative data are labelled where used.