004Studio Carbon · 2020
Ventilator HMI
A redesign of the ventilator user interface for critical care
Summary
Led UX for the Borven ventilator HMI redesign, from field research across three ICUs to a validated interface, cutting time-to-action by 38% and error attempts from 1.6 to 0.4 per session.
- Role
- Lead UX Designer
- Client
- Borven
- Team
- Engineering and clinical partners
- Timeline
- 6–7 months
- Platform
- Embedded HMI, touchscreen and physical controls
My role
I led the project end to end: defining scope and success criteria, mapping ICU workflows, and building the information architecture and HMI prototypes from concept through handoff.
I worked directly with engineering and clinical partners throughout, running field visits, usability tests and recognition studies to validate every major decision before it shipped.
Impact
38%
reduction in median time-to-action from standby to therapy (n=8)
32%
fewer taps and knob actions on the primary fast path
94%
recognition of mode, alarms and vitals within 3s at 1.5–2m
1.6 → 0.4
wrong-mode or wrong-parameter attempts per session
Context
Affordable ventilators, operated under pressure
The brief was an affordable, easy-to-operate ventilator whose interface makes critical tasks faster to see, understand and act on. The objective: make critical actions obvious, confirmable and quick, reducing steps, surfacing essentials at a glance, and improving bedside confidence in ICU use.
I led it end to end: defining scope and success criteria, mapping ICU workflows, and building the information architecture and HMI prototypes from concept through handoff, testing throughout with engineering and clinicians.
Success was defined up front and numerically: time-to-action, number of steps, and error attempts. Everything else was subordinate to those three.
Discovery & research
Field visits to ICUs in three cities
I visited ICU units in Hyderabad, Nawanshahr and Ahmedabad to observe setup, modes, alarms and real constraints, and flagged hard-to-find alarm settings and deep navigation during setup.
Structured interviews with clinicians of varied experience captured tasks, risks and success signals. Notes were converted into affinity clusters that surfaced themes across hardware, ergonomics and cognition, and those clusters defined the fast paths that drove the information architecture.
Alongside the field work I mapped every screen, menu and connection from the ventilator manual to expose the real navigation depth and dependencies, and tested three competitor ventilators hands-on, documenting every step. The maps made the slow paths obvious: mode changes and adding alarm settings.
With gloves on, touchscreens miss taps. Having a physical knob makes adjustments reliable.
Synthesis
What the clinicians made unavoidable
False alarms were constant and distracting, pulling nurses off care when seconds matter. Navigation needed intuitive gestures and clear visual cues to cut cognitive load. And colour and contrast behaved very differently across the lighting conditions a real ICU moves through.
The scroll wheel and physical buttons were valued precisely because they give tactile feedback, so menu layouts and button design had to follow ergonomic constraints, not screen convention.
False alarms are constant and distracting; they pull nurses off care when seconds matter.
Design
Fast paths, and confirm before commit
The information architecture organised navigation, alarms, trends and patient info to keep critical tasks one tap away, and was used explicitly to prioritise fast paths and apply a confirm-before-commit pattern to risky actions.
Low-fidelity screens iterated quickly on layout, hierarchy and interaction patterns. Clinicians valued having key vitals always visible and integrated physical buttons for frequent actions, which directly informed control placement in later designs.
High-fidelity work then validated visual hierarchy, contrast and control density in realistic scenarios, refining colour, spacing and alert states for readability under low light and alarm conditions.
Testing
Measured, not asserted
Scenario tests with six to eight clinicians across three scripted scenarios (alarm storm, parameter tweak, mode change), instrumented with a screen recorder and task timer, recording steps, time and mis-taps for each.
Alongside that, a randomised single-trial recognition test for key vitals and alarms at 1.5–2m viewing distance, computing accuracy and response time to pick the superior theme per metric.