The Twin Scorecard

It was eleven at night, I was on my second coffee, and I was reading a paper I'd been looking forward to. Then I hit this sentence: "The digital twin was constructed from the patient's baseline imaging and used to predict treatment response."

I put the coffee down.

If you haven't run into the term, a digital twin is supposed to be a living computer model of a specific person. It updates as their body changes, it predicts what will happen next, and it helps someone make a better decision about their care. The word that matters there is living. What this paper described was built once, from a single set of scans at the start of treatment, and never touched again. That isn't a twin. It's a photograph.

That's what has stuck with me since. The paper itself didn't bother me much; people stretch terminology all the time. What bothered me was that the field's own checks didn't catch it. Nobody's job, apparently, was to say "wait, that's not what this word means."

You might reasonably ask why a label matters. My answer is that in health, words are infrastructure. Regulators, insurers, hospitals, and patients all decide what to trust partly based on what a thing is called. If "digital twin" can describe a snapshot, it can describe anything, and a word that describes anything can't be regulated, paid for, built on, or trusted.

The strange part is that we already have a test. The National Academies set out four criteria: a twin is updated from the real person, it makes predictions, information flows both ways between the model and the person, and it informs an actual decision. You can check a paper against those in about five minutes. A 2025 review in npj Digital Medicine did exactly that across 149 studies. Only 18 of them, about 12%, fully passed. Only two mentioned verification, validation, and uncertainty quantification, which is the unglamorous work of checking whether the model is right and how sure it is. Two out of 149. I still find that hard to sit with.

So I used to think this was a definition problem. I don't anymore. Everyone in the field knows roughly what a twin is supposed to be. What's missing is anyone willing to say so out loud. It's a small world. You'll see that author at the next conference, maybe on the next grant panel, and "your twin isn't a twin" sounds petty even when it's true. So nobody says it, and the word slowly drifts.

If that's right, and I think it is, the fix can't rely on individuals being brave. It needs to be something impersonal: a checklist journals require, a badge, a registry. Something that does the correcting so no one person has to.

I'll admit I'm less sure about the timeline than the diagnosis. My guess is we have about five years before "digital twin" either settles into a specific meaning or becomes one more piece of marketing language. The FDA is loose enough to let most things through, the EMA is strict enough to slow everyone down, and neither has a real standard yet. Gaps like that tend to get filled by whoever moves first, and right now the fastest movers are press releases.