Grey, Surrogate, Black

There's a child I think about a lot, though I never met them.

They had medulloblastoma, a brain tumour and one of the more common cancers in children. Their surgery happened in one country. Their proton beam therapy, a precise form of radiation, happened in another. Two hospitals, two health systems, two languages, one kid.

In theory, this is exactly the case digital twins exist for. A model of that child could pull the surgical records from one hospital and the radiation data from the other and hold them together, so whoever was treating them next could see the whole picture at once. The core problem in this child's care was that the knowledge was split in half, and a twin is a tool for putting it back together.

In practice, it almost certainly won't get built. The two hospitals have no shared way to exchange data. I watched someone try to move this child's imaging from one centre to the other, and at one point, a fax machine was involved. I wish I were exaggerating.

That was when I stopped thinking of this as a modelling problem. The models are honestly the easy part. The hard part is the plumbing: the connections, agreements, and standards that let data move at all.

It helps to know there are roughly three kinds of digital twin, and they don't serve the same people.

Grey box twins blend what we know about how the body works with data a hospital already collects: notes, lab results, scans. They're the easiest to build, which means they mostly help patients who are already in well-resourced, well-connected hospitals.

Black box twins lean on large amounts of data and deep learning to find patterns no one programmed in. They can be powerful, for example in predicting neurological complications in children with cancer. But they only work where rich data exists, which again means where resources already are.

Surrogate twins are the ones that could actually follow a child like this across a border. They're also the ones that need the most infrastructure.

Put those side by side and you get an uncomfortable result: the most equitable kind of twin is the one least likely to get funded. I don't think that's an accident. Shared infrastructure that helps everyone a little is something no single hospital or payer wants to own.

The idea I keep coming back to sounds simple: the twin should belong to the patient, not the hospital. Picture a portable model, held in a data trust the patient or their family controls, that any clinician in any country could connect to and add to. Instead of being scattered across systems that don't talk, the patient would be the one thread running through all of them.

I know how naive that can sound. It's a legal problem, a trust problem, and a standards problem, and realistically a twenty-year one. Nobody gets promoted for fixing plumbing.

But I've come to think it matters more than it looks, for a reason that's more political than medical. Whoever builds the first working system for sharing health data across borders gets to define what "compatible" means for everyone who comes after, the way early internet protocols shaped the web we have now. I suspect health data is going to become a serious geopolitical issue in the coming years, with countries treating clinical data like a strategic resource. Underneath all the technical language, it's really a question of whose patients get to be seen.

I'll go out on a limb here. I think the first country to build national, patient-owned twin infrastructure will get a head start of something like a decade. It will look like a dull public works project at the time, and in hindsight it will look like one of the most important health decisions anyone made.