Digital Twins in Health Care: Duke’s Approach to Personalized Care and Earlier Disease Detection

Interview With Amanda Randles, PhD

© 2026 HMP Global. All Rights Reserved.

Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of EP Lab Digest or HMP Global, their employees, and affiliates. 

EP LAB DIGEST. 2026;26(12).

Interview by Jodie Elrod

Amanda Randles, PhD, discusses how the Duke Digital Twin Initiative is integrating imaging, wearable sensor data, and computational modeling to advance personalized care, with heart failure among its initial applications. She explores the potential for earlier detection of deterioration and the scientific, technical, and clinical challenges of bringing digital twins into routine care.

Please introduce the Duke Digital Twin Initiative and its overarching goals.
The Duke Digital Twin Initiative brings together expertise across engineering, medicine, nursing, and data science to develop computational models that evolve with individual patients. Led by the Center for Computational and Digital Health Innovation, we are building a shared data ecosystem that connects medical imaging, wearable sensors, clinical records, physics-based modeling, and artificial intelligence (AI) to support discovery and more personalized care.

The backbone of that ecosystem is PRISM, a secure platform being developed to bring together resources such as PULSE, our wearable data acquisition app, with imaging, clinical data, and computational modeling results, enabling AI inference across these complementary sources. These resources have often been collected and analyzed within separate research efforts. We want to break down those silos so investigators can draw on the combination of data and models needed to answer a particular question.

Our initial applications span cardiovascular disease, neurological conditions, and human performance, but the infrastructure is designed to support new applications as they emerge. A team studying a new disease could access the relevant subset of this multimodal ecosystem and build on established tools and workflows, accelerating both discovery and deployment. Our overarching goal is to make it faster to translate Duke’s research strengths into tools that help clinicians identify disease earlier, personalize treatment, and monitor patients over time.

How could digital twins that continually evolve with patients change the way clinicians understand and manage health over time?
Typically, we collect detailed information about patients when they interact with the healthcare system, such as during a clinic visit, an imaging study, or a hospitalization. That gives us a snapshot, or a series of snapshots, but leaves substantial gaps in our understanding of what happens in between. Digital twins that evolve with patients could help us move toward understanding how physiology changes throughout daily life, including changes that develop well before the next appointment.

Wearable sensors give us an opportunity to observe patients beyond those encounters. Combining those measurements with patient-specific computational models allows us to investigate physiological processes that the sensors cannot measure directly. In cardiovascular health, for example, we can study how changes in heart rate and activity translate into changes in pressure, blood flow, and the stresses acting on blood vessel walls. We describe this study of physiology over time as chronophysiomics.

That longitudinal perspective could help us shift from reactive to proactive care. By understanding what is typical for an individual and how their physiology is changing, we could identify signs of deterioration before they lead to symptoms or a hospitalization. We could also track how patients respond to treatment between appointments. The goal is to give clinicians the information they need to intervene earlier and tailor care to each patient’s evolving needs.

What distinguishes the Duke Digital Twin Initiative from earlier, disease-specific digital twin projects, and how will PRISM provide a scalable foundation for this work?
Earlier disease-specific projects have established important methods and demonstrated the potential of digital twins. What we are building at Duke is an ecosystem that allows those efforts to connect and grow. Different diseases require different models, measurements, and clinical expertise, but many of the underlying needs are shared: acquiring longitudinal data, connecting those data with imaging and clinical records, running computational models, and making the results available for analysis and AI inference.

PRISM provides that shared foundation. It is being developed to securely connect resources such as PULSE for wearable data acquisition with imaging, clinical information, and physics-based modeling results. Investigators can then draw on the data and computational tools relevant to their particular question. A new disease application may need only a subset of those resources, but it can build on the existing infrastructure rather than recreate the entire pipeline.

That reuse is central to scalability. We want advances in data acquisition, modeling, and AI to benefit multiple research efforts while preserving the specialized science and validation each application requires. By building the infrastructure together, we can accelerate the path from a new research question to discovery, validation, and ultimately deployment in care.

In the heart failure pilot, how will imaging, sensor data, and computational modeling be combined to identify deterioration early enough to guide clinical intervention?
The heart failure work illustrates how imaging, physiological measurements, and computational modeling can provide complementary information. Imaging gives us the patient’s pulmonary artery anatomy, while clinical measurements establish the physiological conditions for a personalized model. Our blood flow solver, HARVEY, then uses that information to calculate pressures and flow within the pulmonary circulation.

In our published study, “Digital Twins for Noninvasively Measuring Predictive Markers of Right Heart Failure,” conducted with Duke cardiologists Drs. Marat Fudim and Manesh Patel, we developed digital twins for 20 patients with heart failure. After calibration using right heart catheterization measurements, the models showed agreement with pulmonary artery pressures recorded by implanted monitors. We also examined how much anatomical detail was needed to estimate those pressures, an important step toward making the models more efficient and scalable.

That study establishes a foundation for investigating longitudinal monitoring. Sensor data could provide updated physiological inputs between clinical encounters, allowing us to study how a patient’s hemodynamics change over time. The goal is to identify pressure changes and other markers that could serve as a “digital canary,” alerting clinicians to deterioration before a patient develops overt symptoms. Establishing how reliably those signals can be detected, and whether they provide enough lead time to guide intervention, is a central question for the next stage of this work.

What scientific, technical, and clinical barriers must be overcome to move digital twins from research into routine care and ultimately reach one million people?
One of the central scientific challenges is determining what information we need to keep a digital twin reliable as a patient’s physiology changes. We need to understand which measurements are essential, how frequently they must be updated, and when a model needs to be recalibrated. We also need to establish that the physiological changes we identify are meaningful for the clinical question we are trying to answer.

Technically, bringing data together is only part of the challenge. We need to preserve provenance—meaning where a measurement came from, how it was processed, and which model generated a prediction—and propagate uncertainty through that entire process. A clinician needs to understand how much confidence to place in a result and when additional information is needed. Building that trust requires secure data management, traceable workflows, and rigorous validation. PRISM is being designed with those needs in mind.

Scale also requires computational efficiency. High-performance computing enables us to develop detailed models and establish their numerical accuracy. Reduced-order modeling and AI offer opportunities to make those capabilities accessible more broadly, provided we demonstrate that they retain the fidelity required for their intended use.

Clinically, we need prospective studies showing that digital twins provide actionable information and that acting on that information improves outcomes. The tools must fit into clinical workflows, with clear responsibilities for reviewing results and responding to changes. Regulatory pathways, reimbursement, and sustained software support will also be essential. We must address access from the beginning, including the cost of imaging and sensors, the burden on patients, and performance across diverse populations and care settings.

Reaching one million people is an ambitious goal, but the strength and breadth of the team at Duke make us well positioned to pursue it. The close connection between our engineering programs and health system allows clinicians, engineers, nurses, and data scientists to shape this work together, grounding technical advances in clinical needs from the outset. Combined with shared infrastructure through PRISM, that collaboration gives us a strong foundation for moving from individual research demonstrations toward tools that can be validated, deployed, and sustained at scale.