The Continuous Patient

A prediction

Medicine has always meant making decisions with half the picture. A doctor gets fragments: a symptom you mention, a blood test taken once, a scan from a single moment. From those, they build a best guess about a body that never stops changing. Then they wait to find out whether the guess was right, which is usually later than anyone would like.

I've come to think the hardest problem in medicine was never the disease. It was the gap between check-ups. The body is always a step ahead of the doctor.

This post is my attempt to imagine what happens when that gap closes. Some of it is grounded in work that exists today. Some of it is a guess, and I'll try to be clear about which is which.

Three things coming together

I don't think the change will come from one big invention. I think it will come from three areas of work that have been developing separately and are starting to meet.

The first is continuous sensing. For most of history, a measurement was an event: you went somewhere, someone took a sample, a number got written down. That's already changing. There are sensors that power themselves from the motion of breathing and walking, so they need no battery. There are skin patches that measure glucose in sweat without a needle, sensitive enough to pick up tiny concentrations and checked against standard lab tests. There are thin, skin-like membranes that track several things at once, including lactate, sodium, potassium, and temperature, and feed the readings into software that can flag problems in real time.

Here's where I start guessing. I don't think it stops at patches and watches. I can imagine sensors so small and unobtrusive that you stop noticing you're being measured at all. At that point a measurement stops being an appointment and becomes a stream, and a body you can watch in motion is very different from one you have to reconstruct from snapshots.

The second is modelling the individual. A digital twin is a computer model of something real that keeps updating as the real thing changes. Engineers have used them for years. A twin of a jet engine can be run forward to predict when a part will fail. Doing that for a person is much harder, but there are now three broad kinds, which I wrote about in an earlier post. Grey box twins build on medical knowledge and the data hospitals already collect. Surrogate twins could follow a patient whose care is split across provinces or countries. Black box twins use deep learning to find patterns specific to one person. None of them is enough alone, but together they could describe a body in enough detail to simulate what comes next.

The third, and the one people talk about least, is simulating the health system itself. A hospital in Kansas City used a system-wide model to predict a winter surge of viral illness and got the timing right to within about a week. Emergency departments that use simulation to plan staffing report waiting times dropping by 20 to 40 percent. A study in Taiwan cut simulated outpatient queues by nearly 40 percent while also using less energy. It also showed something I find important: the best schedule isn't a fixed one. It shifts as conditions change, and only a simulation can keep up with that.

Now picture those three things meeting.

A walk-in clinic, a few years from now

This part is imagination, so take it as a sketch.

I picture a small clinic, about the size of a pharmacy or a branch library, in an ordinary neighbourhood. You walk in and stand in front of a scanner for under a minute. It isn't an X-ray, just a quick reading that fills in whatever your everyday sensors couldn't capture. Your twin, stored somewhere you control, is already up to date.

On a screen, you see a model of your own body, not as a diagram but in motion, moving forward in time. It doesn't hand you a diagnosis, because a diagnosis only describes today. It shows you where you're heading and where that path could still change.

Then it shows you options. One might be a change in habits, modelled against your physiology rather than an average from a clinical trial. Another might be a medication, run through the same model so you can see how your body specifically is likely to respond. A third might combine the two, in a particular order, with a backup plan if the first step doesn't work.

The part that usually gets left out

This is the piece I care most about. Your personal twin and the health system's model have to be the same simulation.

If they aren't, you get advice that is medically right and practically useless: a scan at a hospital with a three-week wait, or a drug that's out of stock. If they are connected, your twin knows the wait times, which specialists are available, what's in supply, and how busy each service is. It might suggest a slightly less ideal option that can start today, because starting today keeps more doors open later.

It works in the other direction too. Every twin gives the system an early look at upcoming demand, so the system can adjust before the rush arrives instead of after. Capacity doesn't become infinite, but demand becomes visible. You get sent to the hospital that can actually see you, or booked on the day there's room, or offered a remote option when the model shows it would work just as well for you.

The technical catch

I won't repeat my whole argument from the LLM post, but the short version is this. Language models are great at conversation and bad at keeping track of a patient's condition over many steps. The design that makes sense is a partnership. A mechanistic model, built on how the body actually works, holds the patient's state and runs the simulations. The language model does the talking, explaining the results in plain language. A 2025 paper describes a twin as five parts: the patient, the data connection, the computer model, the interface, and the synchronisation that keeps them in step. It also notes that a language model on its own still hallucinates. I don't read that as a reason to give up. It's a description of what has to be built.

What's standing in the way

There are three big obstacles, and I don't want to wave them away.

The infrastructure doesn't exist. After twenty years of effort, most hospitals still can't reliably send a lab result to each other. I once watched imaging get passed between two world-class cancer centres with help from a fax machine. Twins that cross borders would need data-sharing rules that haven't been written yet.

Nobody checks the work consistently. A model used in real medical decisions has to be tested, and someone has to measure how uncertain it is. A recent review found only about one in ten published "digital twins" actually meets the basic definition. The label has gotten well ahead of the substance.

Nobody agrees who owns it. If the hospital owns your twin, it doesn't travel with you. If a device company owns it, it disappears when the company does. Only a twin that belongs to you lasts a lifetime and works across systems, and building that is hard legally, technically, and politically.

I think these obstacles will be overcome, though not because I'm an optimist by nature. I think people will demand it. Once someone's twin catches a problem early, or gets them around a waiting list, they won't want to give it up, and they'll expect it to come with them wherever they go. Patients don't usually force infrastructure into existence. I suspect this will be one of the rare times they do.

Who it serves

Whoever builds this first will shape what a twin is and who it's for. Early standards tend to stick, the way the internet's first protocols still shape it today, and decisions like these are already being made, mostly by people focused on engineering rather than on who ends up included.

The technology is honestly the easier part. Sensors, models, and simulations are all within reach. What will decide whether this future is good or just efficient is who owns the twin, who can see it, who profits from it, and who gets left out.

A twin owned by a hospital will serve the hospital. One owned by an insurer will serve the insurer. One owned by a device maker will serve its shareholders. One owned by the patient will serve the patient. The version I'd want to live in has patient-owned twins, a shared public system for capacity, and open standards linking the two that no single company controls.

That's a design choice, and people will make it at some point, whether or not they think of it as one. I'd rather we made it on purpose.