What If Every Patient Had a Digital Twin? A Thought Experiment About Personalized Medicine

Fast-forward a few decades. A woman is due for a new heart-rhythm medicine. Before the first pill, her cardiologist opens a file that is, in a sense, her: a computer model of her heart built from her scans, her genome, her lab results and years of readings from a watch on her wrist. The doctor tries three doses on the model, then a different drug, and watches what the virtual heart does. Only then does anything happen to the real one.

That scene is a thought experiment, not a forecast. Nobody today has a complete digital copy of themselves, and it is genuinely unclear whether anybody ever will. But parts of it already exist, in narrow and carefully tested forms. This article separates the two: first what is real now, then an openly speculative look at what a “digital twin” for every patient would require, and where the idea strains.

Plain-English: A digital twin is a computer model of a specific real thing, such as a jet engine or a person’s heart, that is kept up to date with data from that thing and can feed information back to guide decisions about it.

What a digital twin is, and isn’t

The term is used loosely, so a definition helps. A 2024 report from the US National Academies of Sciences, Engineering, and Medicine describes a digital twin as a computational model coupled to a physical counterpart and dynamically updated through bidirectional data flows: data run from the real system into the model, and the model’s outputs inform decisions about the real system.

By that standard, many things called digital twins are really personalized models: built once from one patient’s data, used for one decision, and not continuously updated. That is not a criticism; it is where the useful work is today. The same report stresses that trustworthy twins depend on verification, validation and uncertainty quantification, roughly: is the code right, does the model match reality, and how confident should we be in its answers? It identifies medical uses as promising but short of the foundational research they need.

What already exists

Virtual hearts

Cardiology is furthest along, partly because the heart’s electrical behavior follows physical rules that can be simulated. In a 2018 study in Nature Biomedical Engineering, a team led by Natalia Trayanova at Johns Hopkins built personalized heart models from imaging of patients with ventricular tachycardia, a dangerous fast heart rhythm that can follow a heart attack. The models predicted where to apply ablation, a procedure that destroys a small patch of tissue to stop faulty electrical circuits. The work combined animal experiments, a retrospective analysis of 21 patients, and a small prospective test in 5 patients. That is proof of feasibility, not proof of better outcomes.

In 2025 Johns Hopkins reported that the group was running a clinical study of digital twins to guide ablation, and that adding an AI method called DIMON could cut some simulation times from hours to seconds on a personal computer. Speed matters: a model that takes a day to run is hard to use in a working hospital.

Virtual patients in regulatory science

Simulated patients have also entered regulatory work. The UVA/Padova type 1 diabetes simulator, with 300 virtual adults, adolescents and children, was accepted by the FDA in 2008 as a substitute for animal trials in the preclinical testing of certain insulin treatments, including closed-loop “artificial pancreas” algorithms.

The FDA’s own VICTRE project ran an in-silico trial, a trial conducted entirely in a computer, with 2,986 simulated patients to compare two breast-imaging technologies. Its results matched those of a comparable trial in real people, and the FDA concluded that such simulations can be viable sources of regulatory evidence alongside other testing. In 2023 the agency finalized guidance on judging the credibility of computer models in medical device submissions.

These are the real foundations. Notice their shape: each model answers a specific question, in a specific organ or device, and has been checked against real-world data.

The thought experiment: a twin for everyone

Everything in this section is speculative. Suppose those narrow models could be joined up and kept current for each person. What might change?

• Treatment rehearsal. Doctors could test doses or drug combinations on the model first, as in our opening scene, perhaps reducing trial-and-error prescribing.
• Surgery and procedure planning. Surgeons could rehearse on a patient-specific anatomy and physiology, as the virtual-heart work already hints.
• Earlier warnings. A twin fed by continuous data might flag drift toward disease before symptoms appear.
• Different clinical trials. Simulated patients might help design trials or, in some settings, supplement comparison groups. Our explainer inside a clinical trial shows why this would need exceptional care.

A digital twin is only as honest as its uncertainty: a model that is confidently wrong about one person is more dangerous than no model at all.

Where the thought experiment breaks

Push the idea and the strains show quickly:

1. We do not know enough biology. Hearts obey fairly well-understood physics. The immune system, metabolism and the brain involve vast networks we understand only in part. A model cannot simulate what science has not yet described.
2. Validation for one person is hard. A model can be checked against populations. Proving it is right for you, before the decision it is meant to inform, is a different problem.
3. Data are patchy and biased. Models learn from the people already studied. Groups under-represented in research could get worse twins.
4. Privacy and ownership. A continuously updated model of a person is some of the most sensitive data imaginable. Who holds it, and who may query it?
5. Responsibility. If a clinician follows a twin’s advice and harm follows, where does accountability sit?
6. Over-trust. A vivid simulation can feel more certain than it is.

Some of these are technical problems that research could solve; others are social questions that no amount of computing will settle. For related tensions, see our pieces on AI drug discovery and the real role of AI in MRI.

What would have to be true

For personal digital twins to move from thought experiment to routine care, several things would need to happen, in roughly this order: organ-specific models proven in trials to improve outcomes, not just predictions; standards for validating models against individual patients; data systems that update models safely and privately; regulatory frameworks that cover models that keep learning; and evidence that the benefit justifies the cost. The FDA’s 2025 roadmap on reducing animal testing, which names computational models among its alternatives, suggests regulators are open to simulation where it earns trust.

Explore more experiments and interactive tools in The Lab.

What to watch

• Results of prospective clinical studies using cardiac digital twins to guide ablation.
• Regulatory submissions that rely partly on in-silico evidence, and how the FDA’s credibility guidance is applied.
• Progress on models of systems beyond the heart, such as metabolism and cancer growth.