Two cardiac examples show the distinction. One study used thousands of computational hearts to investigate physiology. Another technology uses a patient’s coronary CT scan to estimate blood-flow information for a defined clinical task. They differ in purpose, validation and what their results establish.
A research population of virtual hearts
A 2025 Nature Cardiovascular Research study constructed 3,461 cardiac digital twins from UK Biobank participants and another 359 from an ischemic-heart-disease cohort. The models combined cardiac MRI and electrocardiographic data with physiological constraints.
The researchers used this population to examine conduction and repolarization. Their analysis linked observed electrical differences to anatomy and other modeled properties. It is an example of how patient-specific simulations can investigate mechanisms that are difficult to separate in clinical measurements alone.
It was not a randomized test showing that giving clinicians these models reduced heart attacks or improved survival. Nor did it establish continuous tracking of all 3,820 people. A population of models can generate useful scientific findings without being a deployed clinical service.
A defined clinical modeling product
HeartFlow FFRCT is a more specific example of patient-specific simulation in clinical use. The FDA’s 2019 clearance documentation describes a model derived from previously acquired coronary CT images that computes a quantity related to coronary blood flow. Its stated use concerns clinically stable, symptomatic patients with coronary artery disease.
The document places the output alongside the person’s history, symptoms, other tests and professional judgment. It does not describe an autonomous doctor or a whole-body twin. The input is a particular scan, not a guarantee that every physiological change is continuously reflected in the model.
Calling this technology a “digital twin” can explain the patient-specific modeling idea, but the regulatory indication remains the more useful description of what the product is meant to do.
What happened when a care pathway was tested
The FORECAST randomized trial assigned 1,400 people with stable chest pain to a CT-based strategy with selective FFRCT or standard care. The primary endpoint was cardiac costs over nine months, rather than a general measure of technological accuracy.
Invasive angiography occurred in approximately 19% of the experimental group and 25% of the standard-care group. However, the study did not find a statistically significant difference in overall cardiac costs, major adverse cardiac and cerebrovascular events, angina or quality of life.
The useful finding is therefore specific: this pathway reduced use of invasive angiography in the trial. It would overstate the evidence to turn that result into a demonstrated survival benefit or guaranteed saving for every health system. The study tested a strategy incorporating selective use, not an isolated software output applied to everyone.
The questions a digital-twin claim should answer
What is represented? An organ, a physiological process and an entire patient are different scopes. A visually detailed model is not necessarily validated for the decision being proposed.
When is it updated? A model built from one scan should not be described as a real-time replica unless that updating process actually exists and has been evaluated.
What was tested? Agreement with a measurement, an explanation of physiology, a change in clinician behavior and improved patient outcomes are separate results. Evidence for one does not establish the others.
For a hospital considering implementation, these questions belong in the purchase and monitoring process. A useful model earns a place by supporting a defined decision with evidence, not by carrying the most ambitious label.
