Timeline
2024
Healthcare Interface Design
Telehealth nurses monitor dozens of patients at once, but their tools show data as disconnected snapshots. This interface replaces them with a continuous timeline, turning alarm-chasing into decisions.
Duration: 6 weeks
Partner: Philips Healthcare
Team:
Sander Randoja
Zeynep Emiroğlu
Shelley Xiao
Working on the mobile view,
please have a look on desktop
Identifying Key Contextual Cues in Tele-ICU Care
Our research showed that telehealth nurses depend on four main event types for context:
Comments
Medical Events
Alarms
Visual records
Monitoring View for the Tele-Nurse
A dual-screen setup. The left shows all monitored patients at a glance.
The right provides a detailed timeline of whichever patient the nurse selects.
Predictive Vitals View
Legacy interfaces presented vitals and alerts as isolated snapshots. A temporal graph and predictive layer surface risks before they become critical.
Patient Context Overall View
Detailed 4-hour timeline showing alarms, medical events, and comments
Hourly summary of clinical events within each time block
Vital sign graphs over the past 4 hours
Scrollable history of past
patient context
Patient information tab: recent movements, diagnosis, treatment summary, and upcoming care plans
Short video updates of patient movement
Contextual Events View
Alerts link to the timeline, surfacing evidence rather than commands and preserving the clinician's authority over the AI.
Snapshots From the Process
Field visit to the Thorax ICU, Norrlands University Hospital
Clustering and synthesising research findings
Mapping nurse workflows from research themes
Roleplaying proposed nurse workflows
Ideating ideas for richer patient context
Reflections
In healthcare, AI's job isn't to make the call. The clinicians at Norrlands Hospital were clear about this, and we kept rediscovering it as we designed. The interesting question wasn't how to automate the decision but how to reduce the noise around it, so the nurse could give the decision her full attention.
A nurse won't rely on a predictive tool in a critical moment unless she can see why it's predicting what it's predicting. Accuracy was the price of entry; explainability was what made the system usable. Every prediction links back to the data history that produced it.
With six weeks and limited clinician access, prototyping became how we asked questions, not how we proved answers. Roleplaying low-fidelity prototypes surfaced objections we couldn't have reached in interviews. By the time we tested a high-fidelity version, we'd already abandoned two earlier directions. The rough prototypes did the real work.
