Triage is a conversation before it is a prediction.

Intake and triage communication · Research direction

The problem

People contact a service in their own words, in their own language, often frightened. What they say first decides who sees them and when. Most failures at the front door are failures of listening and routing, not of clinical knowledge.

The question

How should a system gather what a person is worried about, tell them plainly what happens next, and hand a complete picture to a human, while never telling anyone they are fine?

Triage is a conversation before it is a prediction

On the front door of a health service and what it owes the person walking through it.

The loudest ideas in this area are about prediction: score the symptoms, rank the urgency, send the right person to the right place. Prediction matters. But watch what happens at a real front door and most of the work is conversation.

A person says what is wrong in their own words. They may not have the words. They may be speaking their second language. They may minimise because they do not want to be a bother, or exaggerate because they were once told to. The first job is to understand what they mean, and to say back what you understood so they can correct it.

The second job is to be honest about what happens next. People tolerate waiting far better than uncertainty. A message that says who will look at this, roughly when, and what to do if things get worse is a clinical act as well as a kindness.

The third job is the handoff. The nurse or doctor who picks this up should not need to ask the person to start again. That is a testable property. Give a clinician only the handoff and ask whether they could proceed. If not, the intake failed, however friendly it sounded.

There are things such a system must never do. It must not tell someone their symptoms are nothing. It must not decline to pass on something that sounds urgent because the model was unsure. It must treat a phrase like chest pain, thoughts of self harm or a child who will not wake as a reason to reach a human at once, in every language it claims to support, and we should test that claim language by language.

This is not a medical device and we make no claim that it is safe for use with patients. It is a design position about what such a system should be built to do, and a test plan for finding out whether it does.

Test design

• Hard rule: the system does not provide reassurance about symptoms. It collects, routes and explains process.
• Red-flag phrases always reach a human immediately, in every supported language.
• We test handoff completeness: could the nurse act without asking the person to repeat themselves?

Protocol: Handoff completeness test