A car crash sends data before it sends a patient. EmergX turns crash sensors into a briefing responders can read in the eight seconds they actually have.
The ambulance doesn't need more data. It needs the four facts that change the first decision, and it needs them before the doors open.
Crash data already exists in the car. None of it reaches the people treating the patient.
Responders arrive with a phone call: a location, maybe a car count. Blood type, medication, consciousness and impact severity are all discovered on scene.
Every handover between ambulance and hospital loses information again, four clinicians in four countries described the same gap.
On-site at Autoliv, plus a literature review, to isolate a workable slice: post-crash, the first minute, not pre-crash prevention.
Four semi-structured interviews with clinicians across Sweden, Iran, Bangladesh and Spain; journey maps for each role.
Four concepts (VR triage, thermal seat sensors, car black box, mobile app) mind-mapped and scored in a Pugh chart.
Brand, user flow, wireframes, then two role-based hi-fi interfaces plus a car HMI companion.
Moderated testing with a physician, a witness and general users; the interface changed on every round.
The brief was 'injury prediction'. I narrowed it to the post-crash minute (where prediction changes treatment) and went to the people who work in it.
| Role | Tools used | Focus |
|---|---|---|
| Urgent care physician (Sweden) | Handover notes, clinic records | Incomplete handovers, language barriers, referral delays |
| Ambulance paramedic (Iran) | Radio dispatch, ambulance kit | Wrong scene information, unknown casualty count |
| Field doctor, IRC (Bangladesh) | Emergency calls, portable kit | No medical history, no risk picture before arrival |
| Orthopaedic trauma surgeon (Spain) | Imaging, medical records | Injury detail lost between ambulance and hospital |
“Incomplete information in the handover delays care exactly where speed matters, post-crash trauma.”
“To intervene properly I need to know the medical background of the people in the crash, and the exact time and force of it.”
Every clinician described arriving under-informed rather than under-equipped. The decisive facts (consciousness, blood type, medication, impact force) exist somewhere, just never with the responder.
Design need · Push a structured briefing ahead of the patient, not a phone call.
Ambulance to hospital is a verbal, non-standard transfer. Orthopaedic detail in particular was routinely lost, forcing re-assessment on arrival.
Design need · One record that travels with the patient, readable by both roles.
In testing, a physician rejected generic 'light / moderate / severe' labels. Severity is read by body region and by whether trauma is external.
Design need · Classify by body region plus external-trauma flag, and state consciousness first.
If every passenger's phone auto-dials emergency services, one crash becomes five reports and dispatch loses time triaging duplicates.
Design need · Route through the car's HMI so one crash emits one authoritative call.
VR triage training, thermal seat sensors, a car black box, and a connected mobile app. A Pugh chart against life-saving value, data reliability, privacy exposure and feasibility pushed the mobile app plus car-HMI pairing to the front: it reuses sensors cars already ship and carries the one thing sensors can't infer, the occupant's own medical profile.
Seat sensors count occupants, seatbelt and biometric sensors read restraint and vitals, cameras reconstruct the impact in 3D, and a resilience indicator estimates severity. Each input maps to one fact a responder asked for, nothing was collected because it was available.
The general user owns and consents to the data; the healthcare user consumes it. Splitting the interface at login kept the driver flow calm and the responder flow dense, without maintaining two products.
A physician said treatment order starts with whether the patient responds. It became the first line of the patient card, above identity.
A leg fracture with external trauma and a head impact are not the same 'moderate'. The card states region and external-trauma state, and the AI ranking explains itself.
The HMI arbitrates. Passenger phones contribute medical profiles; the vehicle emits the single emergency call.
The original consent page was a wall of text with a redundant 'manage options' button. Testers skipped it entirely, so it was cut to a short, scoped statement, release only on a confirmed crash.
Photos show the car. A short reconstruction shows the direction and force of impact, which is what predicts internal injury.
Both users read the screen at night, roadside, in a moving vehicle. Deep navy with green status accents holds contrast under headlights and keeps the SOS action unmistakable.
Three moderated rounds with a medical professional, an accident witness and general users. Every round removed something.
| Task | First version | After testing | What I learned |
|---|---|---|---|
| Read and accept data consent | Long legal page, redundant 'manage options' | Short scoped statement, one action | Every tester scrolled past the original without reading it. |
| Connect a vehicle | Two competing selection paths, no way to unpair | One list by name and plate, explicit disconnect | Testers assumed they were still connected to an old car. |
| Add medical information | No visible entry point on profile | Dedicated add button per field group | Users read the profile as read-only. |
| Report an accident as a witness | Map in nav bar, contacts irrelevant to the case | Map inside the emergency call, nav simplified | A witness reporting for a stranger doesn't need her own contacts. |
| Assess an incoming patient | Severity as a single generic label | Consciousness, region-based severity, treatment order | Physician-driven change; also added airway resuscitation to the responder home. |
Two role-based interfaces and a car HMI concept, validated against the four clinical journeys that started the project.
Every element on the responder patient card traces to a stated need from an interview or a test session.
SWOT set the honest limits: the concept depends on working sensors and connectivity, and privacy control must stay with the user.
Next steps if built: instrument real dispatch timings, and validate the AI severity ranking against trauma registry data before any clinical claim.
Responders get a patient briefing before arrival: identity, blood type, medication, consciousness, severity and a 3D reconstruction of the impact.
Consent is explicit and scoped, medical data is released only by a confirmed crash event.
Routing the call through the car's HMI removed the flood of duplicate emergency calls the concept would otherwise have caused.
Delivered to Autoliv as a validated concept prototype; not shipped to production.
The brief said injury prediction. The research said the prediction is worthless unless it lands as a decision, in the first minute, in the responder's hand, so I designed the delivery, not the model.
Designing for two roles from one dataset was the leverage point. Splitting at login cost one screen and saved an entire second product.
If I ran it again I would test the responder card with paramedics under time pressure, not seated. Everything I learned about density came from a calm room.
I'm available for product design roles and consulting projects.
contact@lidatohidi.com
