Pre-hospital documentation: how paramedics use on-device AI to capture patient care without signal

Pre-hospital medicine is documentation under conditions that no office worker encounters. A paramedic attending a cardiac arrest on a rural road, in the dark, in January, with no signal on their phone, has two tasks: keep the patient alive and record what happens accurately enough that the emergency department can pick up where the crew left off.

The second task has always been the harder one to do well. Pre-hospital documentation is written under time pressure, in poor conditions, from memory, after the fact — and errors in that documentation can affect the handover that determines the next stage of the patient's treatment.

AI changes this. Not by replacing the paramedic's judgment, but by handling the transcription and structuring of what was said and done, so the paramedic can do medicine.

The pre-hospital documentation problem

Paramedics produce two critical documents from each call: the Patient Clinical Record (PCR) and the verbal SBAR handover at the emergency department.

The PCR is a legal document. It records the mechanism of injury or illness onset, the patient's presenting history, observations, interventions, and the crew's clinical reasoning. Inaccuracies in the PCR become inaccuracies in the patient's medical record. Missing information in the handover means the receiving team is working with an incomplete picture at exactly the moment it matters most.

Pre-hospital crews write these documents retrospectively. The call may have lasted 45 minutes. The crew drives to the hospital, hands over, and then documents — sometimes hours after the call, sometimes in the middle of the next one, sometimes at the station at the end of a shift when the details of five calls are competing for the same working memory.

The conditions in which pre-hospital care happens compound this:

No signal in rural areas. A large proportion of pre-hospital work happens in rural and semi-rural areas where mobile data is unreliable or absent. Any documentation tool that requires internet connectivity is a tool that fails on the calls that are most isolated and often most critical.

Physical conditions. A patient in cardiac arrest needs CPR. A polytrauma patient needs assessment. The crew's hands are on the patient. They cannot simultaneously type. Documentation happens after the immediate clinical work, when the moment has passed.

Time pressure. On-scene time targets exist for good clinical reasons. The time available for documentation is not "whatever's needed" — it is whatever is left after treatment and packaging.

Multiple simultaneous information streams. The paramedic is talking to the patient, talking to a colleague, monitoring vitals, making clinical decisions, and receiving information from multiple sources at once. A verbal record of what's happening is the only practically achievable contemporaneous account.

What an on-device AI workflow changes

Record the scene. A paramedic attending a patient can speak the assessment aloud: "Patient is a 72-year-old male, unresponsive, bystander CPR in progress on arrival, initial rhythm VF, one shock delivered at 04:23, ROSC at 04:31, GCS 6 on arrival at hospital." Kuulo records this on an iPhone, offline, and structures it into an ABCDE assessment note on the way to hospital.

This is documentation that happens in real time, at the point of care, without internet, without a separate device, and without interrupting clinical work. The paramedic speaks the assessment as they perform it — a practice that experienced pre-hospital practitioners often use anyway as an aid to systematic thinking — and the AI structures what was said.

On the way to hospital: the crew can review the AI-structured note, correct anything that was misspoken, and arrive at the ED with a complete, structured account ready. The verbal SBAR handover can be given from the note rather than from memory.

At the station: the PCR can be completed from the AI note rather than from recall. The contemporaneous record is more accurate than retrospective recall, particularly at the end of a busy shift.

The ABCDE framework on-device

Pre-hospital assessment uses the ABCDE structure: Airway, Breathing, Circulation, Disability, Exposure. A structured AI note from a pre-hospital recording should follow this structure.

Kuulo's Clinical Assessment template produces structured output across: primary complaint and mechanism, ABCDE findings, interventions performed and timing, vital signs trends, and clinical decision-making. The template is applicable to both trauma and medical calls.

For a road traffic collision: Airway — patent, no intervention; Breathing — bilateral air entry, rate 24, SpO₂ 91%; Circulation — radial pulse absent, femoral present, BP not obtainable, haemorrhage control applied right thigh; Disability — GCS 10 (E3V3M4), pupils equal and reactive; Exposure — open femur fracture right leg, no other injuries identified.

That structured note, generated from the paramedic's spoken assessment, is the PCR in draft. It is also the basis for the SBAR handover.

Rural and no-signal operation

Kuulo runs entirely on-device. The iPhone running Kuulo in a rural ambulance, on a hillside, in a car park, or in a derelict building has the same transcription capability as one with full 5G signal. The AI model runs on the Neural Engine of the iPhone — not on a server that needs a connection to reach.

This is not a technical detail. For pre-hospital work, it is the functional requirement that determines whether a tool is usable at all. A documentation tool that fails without signal is a documentation tool that fails on exactly the calls in which documentation is most critical and most difficult.

SBAR handover for the ED

The structured pre-hospital note maps directly onto the SBAR format required for emergency department handover:

Situation: what happened, what the patient's presenting problem is.

Background: relevant history, medications, allergies.

Assessment: what the crew found, what they did, the patient's current status.

Recommendation: what the receiving team needs to do next.

A paramedic arriving at the ED with a Kuulo-generated note has the material for an SBAR handover already structured. The handover can be given from the note, consistently and completely, rather than improvised from memory under the time pressure of a busy resus bay.

Data governance in pre-hospital care

Patient data recorded pre-hospital is subject to the same GDPR protections as data recorded in hospital. The patient's presenting information — medical history, vitals, injury or illness — is health data under Article 9, requiring the highest level of protection.

On-device processing means patient audio stays on the paramedic's device. No audio is transmitted to a cloud server. No DPIA is required for the use of the recording tool itself. The data governance question is about the device and the practitioner's obligations — not about a third-party cloud processor accessing patient audio.

This matters in pre-hospital contexts specifically because the informal, real-time nature of pre-hospital documentation creates vulnerabilities that don't exist in controlled clinical environments. A recording made at a road traffic collision that is later transmitted to a cloud service for transcription is a recording of patient data in an environment where data governance is already complicated. Keeping that recording on-device simplifies the governance significantly.

Documentation and the PCR

The Patient Clinical Record is a legal document with medicolegal consequences. Inaccurate PCRs create problems for the patient's ongoing care, for the service's governance processes, and potentially for the crew if documentation is challenged.

The most common source of PCR inaccuracy is not dishonesty — it is the retrospective nature of the documentation. A crew documenting a call two hours after it happened, from memory, after three subsequent calls, will not produce as accurate a record as a crew documenting from a contemporaneous spoken account.

Kuulo doesn't write the PCR. The paramedic writes the PCR. What Kuulo provides is a contemporaneous structured record that the PCR can be written from — more accurate than retrospective memory, available immediately after the call, and structured in a format that maps onto the PCR fields.

The quality of pre-hospital documentation improves when documentation is less dependent on retrospective recall. That is what on-device AI capture provides.