Territorial Emergency Medicine, Non-Conveyance and AI: My Conversation with JEMS

A few weeks ago I received a message on LinkedIn from Mike Brown, Development Editor at JEMS, the leading journal for North American EMS. He had read a post of mine on non-conveyance and wanted to discuss it.

Last week we recorded an episode of the JEMS Podcast together. The title JEMS published the episode under summarizes well the direction the conversation took: From Non-Transport Care to AI-Enabled Territorial Medicine.

In this article I'll try to put in writing the points I found most interesting, not so much as a summary of the episode (which you can listen to at the bottom of the page), but as a reflection on the questions this conversation left open. These are questions that, in my view, concern all EMS systems, not just mine.

An American Editor, an Aosta Valley Physician

Mike started from a concrete fact: the post on non-conveyance. "It's something that is of a significant interest in something that we're really looking at when we look at US-based EMS systems," he told me at the beginning.

The fact that a JEMS editor decides to seek out a physician from an Italian regional system to discuss this says something about the moment American EMS is experiencing. In a context where transport is still today the main reimbursement model for many services, the question "when is not transporting the right choice?" is becoming impossible to ignore. And as we formulate it, we discover that it has better answers elsewhere.

At one point in the conversation Mike commented: "It's crazy to listen to you describe all of that, because it's so similar to a lot of the things that we're experiencing here in the United States."

Two systems that are structurally opposite, one public and universal, the other mediated by private insurance; one physician-led, the other paramedic-led, are converging on the same fundamental questions. This in itself is already the first lesson I'm taking home.

The Patient I Don't Transport

The heart of the conversation started here.

In Italy the prehospital system consists of different vehicles: ambulances with only EMTs, ambulances with nurses, rapid response vehicles with physicians, helicopter EMS. Those who manage dispatch, as I do when I'm in the 118 control center, decide which vehicle to send to each call. And the question I often ask myself is: "Are we sure that the Emergency Department is the right place to bring those people?"

Because if we look at the data, in many cases the honest answer is no. Not because the hospital does something wrong, but because the clinical need of those patients doesn't coincide with a visit to the Emergency Department.

I told Mike about three references that in recent months have changed the way I look at the problem.

Estonia, national registry 2025. On Estonian ambulances operates a mix of nurses and general practitioners, an unusual combination by European standards. The system has formalized non-conveyance as a quality indicator (QI9): they don't just count how many patients stay at home, they follow what happens afterward, at 48 hours, at seven days.

Finland. Of non-transported patients, only 4% access a hospital facility within 48 hours of the call and only 1% are admitted. The vast majority received what they needed at their home.

London, Physician Response Unit (PRU). A rapid response vehicle in the spirit of Italian ones, but developed as a structured component of the system, not as a historical legacy. Equipped with point-of-care ultrasound, blood gas analysis, essential tests. About 70% of patients treated by the PRU stay at home.

Different numbers, different systems, a coherent message: non-conveyance is not an organizational shortcut. It's a clinical outcome, and when done well it's the best outcome for many patients.

In a country like Italy, with one of the oldest populations in the world, this distinction is far from academic. The elderly patient with COPD transported to the ED for suspected pneumonia often ends their night on a stretcher in the hallway. At home, I could have performed a lung ultrasound, confirmed or ruled out pneumonia, prescribed an antibiotic, organized follow-up with the family physician and returned available for another call in half an hour. If the system works, that patient received better care without leaving home.

The Gap Nobody Funds

The most uncomfortable topic we touched on is another one, and in the podcast it came out almost as a vent.

In Italy, we have practically everything. Sufentanil in the ambulance, point-of-care ultrasound, access to the regional health record from a mobile phone. We have the clinical tools. What we don't have is information continuity.

When I leave a patient at home after a thorough evaluation, the handover to the general practitioner still today depends on a phone call. Often unanswered. Followed by a chain of bounces through the operations center. In 2026. With AI watching from above.

"Sometimes they let us use sufentanil," I told Mike, "and they cut with more important things, like this connection for the patient."

It's a paradox that doesn't depend on lack of technology, but on where public healthcare chooses to invest. In a private system these choices would be more transparent: if you don't profit from the patient who stays at home, you don't invest in the infrastructure that supports them. In a public system, paradoxically, the patient who stays at home saves the system money, but the resources to make the continuity pathway sustainable don't arrive anyway.

Non-conveyance done well requires system. Not just clinical competence, not just tools. It requires that what I saw at the home arrives, at the right time, to the right person in the healthcare system. Without that continuity, non-transport becomes fragile: if something goes wrong in the following 48 hours, nobody knows where to look.

This is the gap. And it's not a gap that's solved with another guideline. It's solved with infrastructure, with platforms, with a rethinking of how clinical data travel between 118, the hospital and territorial medicine. This is where technology comes in, when it really comes in.

From "Prehospital" to "Territorial"

There's a linguistic nuance that in the podcast I wanted to emphasize, and that Mike found much denser than it seems.

In Italian the official term for my specialty is not "prehospital medicine", it's territorial emergency-urgency medicine.

"Prehospital" is a spatial definition: it describes where you operate, that is, outside the hospital. "Territorial" is a functional definition: it describes what you take care of, that is, a territory and the people who inhabit it.

It's a shift in perspective that has direct operational consequences. If I'm "prehospital", the right question is how do I better transport this patient to the hospital. If I'm "territorial", the right question is how do I ensure this person receives the right care, wherever the place to receive it is. Whether it's the ED, the specialty clinic, home with scheduled follow-up, hospice, a long-term care facility. Or, when appropriate, a thorough evaluation at home with return to their own bed and a prescribed antibiotic.

It's not just a semantic debate. It changes what you measure as a system, how you train personnel, how you design management software and, what interests me most in this context, what kind of AI you design. An AI designed for prehospital medicine understood as medicalized transport will be one type of tool. An AI designed for territorial medicine will be very different. Much less spectacular, much more systemic.

Why the AI I Need Isn't the One That Reads ECGs Better Than a Cardiologist

When the conversation reached AI, Mike gave me space for a reflection I've been holding for some time.

AI is not a religion, it's science. It's a tool that's already among us, for better or worse. The point isn't to be "pro" or "against", the point is to understand what it's really for.

I'm not particularly interested in an AI that reads an ECG better than a cardiologist. Studies that compare ChatGPT with an ED physician interest me little, beyond academic curiosity. I told Mike almost verbatim: "Even if AI can become better than me, why should I push on that idea? Why should I work in order to put away my nice job that I love?"

It's not a rhetorical question. It's a design choice, even before it's a professional choice.

The AI I need in my work has two precise tasks, both in service of the clinician, not in their place.

Before the Intervention: Cognitive Unloading

While the crew travels toward a call, a few minutes are enough to mentally review a protocol, to check pediatric doses for a two-year-old child, to reread the pathophysiology of a condition seen three times a year. A focused clinical assistant, trained on serious sources (not someone's Reddit, not YouTube, not unverified forums), that responds in thirty seconds instead of making you waste two minutes navigating a PDF on the ambulance tablet.

The principle is simple: unload the cognitive burden where you can, to be fully present where it really matters, that is, at the moment of arrival at the patient.

After the Intervention: Continuity and Memory

After an evaluation, the AI should be able to read what I wrote in the report, cross-reference it with the patient's medical history, and structure a precise communication for the family physician, for the hospital specialist, for whoever will take charge of the downstream pathway. Not because I don't know how to do it, but because while I'm doing it well for this patient, another crew is already calling from the operating room for the next emergency.

The metaphor I used with Mike, and that seems most honest to me, is this: the AI I want is like having a good specialist beside me, and together with them a librarian who has read the entire patient chart and knows how to tell me what's relevant now.

The librarian doesn't decide in my place. They just know that the appendectomy at sixty probably doesn't count for the problem I'm facing today, while last February's pneumonia maybe does. This type of filter, in a setting where I have few minutes and much information, is what changes the quality of every decision.

Building EMSy has been for me, and continues to be, the practical exercise of trying to translate this principle into a real product. This isn't the place to describe how. The point, for the purposes of this article, is another: the design of clinical AI is an ethical choice before it's a technical choice. You decide what you want AI to do in place of the human being, and what not, even before writing the first line of code.

Mike summarized the concept better than I could have: "You're looking at it more so to group the information and give you the tools to more quickly make clinical decisions, not replace the clinical decision maker."

Exactly. My favorite word, when I talk about AI in emergency medicine, is supportive. Not autonomous.

What I'm Taking Home

I'll close with three points that in the days after the recording kept turning in my head.

The first is that the EU/US convergence on the topic of non-conveyance is deeper than I thought. Completely different systems, with opposite structural incentives, arrive at the same questions when they look honestly at their data. Not all patients who call the emergency number need an ED, and more and more systems are trying to give structure to this evidence. The question is no longer whether to do it, but how to fund it, measure it and govern it.

The second is that the right technology is not the most spectacular, it's the most systemic. An AI that connects 118, the family physician and the hospital, transmitting an accurate clinical summary at the right time, is less photogenic than an AI that "makes diagnoses on its own in the ambulance". But it's infinitely more useful. Territorial medicine happens or doesn't happen on this infrastructure, not on public demos.

The third is that, as Mike said closing the podcast, EMS is becoming "the gateway to the healthcare system." An entry point. This changes everything. It changes what we expect from prehospital professionals. It changes what we expect from their tools. And it changes, not least, what we expect from the companies that build those tools.

I remain convinced that the only honest way to design clinical AI is to do it from within clinical practice, with the humility of those who know that the patient remains at the center and the clinician remains necessary. The rest, almost always, is marketing.

Listen to the Conversation

The complete episode of the JEMS Podcast with Mike Brown is available below. The original JEMS article can be found HERE.

I thank Mike Brown and the JEMS editorial team for their hospitality and for the quality of the conversation.