Digital twins enter the decision room for prolonged liver perfusion
Many donor livers are still discarded even when they may be recoverable, largely because current workflows are built to preserve an organ long enough to judge whether it can be used, not to actively rehabilitate it over several days. Multi-day normothermic ex vivo organ perfusion changes that equation. Once the organ is outside the body and no longer buffered by native homeostatic control, the central challenge shifts from keeping the circuit running to deciding what the physiology means as it evolves hour by hour.
This concept paper proposes physiologic digital twins as a way to make those decisions more defensible. The idea is straightforward in principle and demanding in practice: combine mechanistic biology with statistical modeling to estimate the state of the organ during perfusion, then use those estimates to support clinicians deciding whether the graft should continue to be rehabilitated, reassessed, or ultimately accepted for transplant. In a setting where perfusion data accumulate quickly but the organ no longer benefits from the body’s own regulatory systems, interpretation becomes the limiting step.
The paper is focused on the liver, but the logic reaches beyond a single organ. Ex vivo perfusion creates a rare therapeutic window in which ischemic injury, cellular recovery, metabolic reserve, and vascular behavior can all change over time. A static snapshot is no longer enough. A rising lactate level, shifting vascular resistance, or subtle changes in synthetic function may reflect ongoing injury in one case and delayed recovery in another; the clinical problem is separating those trajectories before a scarce graft is lost or, just as concerning, before a marginal graft is accepted without sufficient confidence.
Why prolonged perfusion changes the problem
Standard donor assessment has long favored brevity. A liver is preserved, evaluated, and either accepted or rejected within a relatively constrained window. Multi-day perfusion expands that window, but it also exposes a structural weakness in current decision-making: physiologic measurements are abundant, yet their meaning is uncertain without an integrated framework. The organ is alive, but isolated. It responds to perfusate, flow, oxygenation, temperature, and time, while lacking the endocrine, neural, immune, and renal feedback loops that would ordinarily shape recovery.
That is where a digital twin becomes attractive. By blending mechanistic models of organ biology with statistical inference, a twin can estimate hidden variables rather than relying only on what the perfusion circuit directly records. In a liver, those hidden variables may include the state of cellular metabolism, microvascular integrity, or the balance between injury and recovery. A model that updates as data accumulate could help distinguish transient instability from irreversible decline, which is exactly the distinction that matters when a transplant team is trying to decide whether the organ should remain in the system for another day.
The clinical logic is easy to see. Ex vivo perfusion is not just preservation; it is active observation under manipulated conditions. If the system can be modeled well enough, the organ’s trajectory may become more interpretable than a single lab value or a single macroscopic inspection. That could reduce the number of donor livers discarded out of caution and could also prevent overconfidence in organs that look acceptable early but deteriorate when stressed over time.
Mechanistic modeling meets statistical inference
The proposed framework is built on the fusion of two modeling traditions. Mechanistic models bring biological structure, allowing the digital twin to reflect the known relationships among perfusion parameters, metabolism, and organ state. Statistical models bring flexibility, helping the twin learn from observed data and adapt to noise, variability, and incomplete measurement. In combination, they offer a way to infer the condition of a perfused organ even when direct bedside-style judgment is not possible.
This is particularly relevant during multi-day perfusion because organ physiology is dynamic rather than binary. Recovery does not always proceed in a straight line. A graft may stabilize, drift, and recover again as edema resolves, metabolism improves, or vascular tone changes. A rigid threshold-based approach risks flattening those dynamics into yes-or-no decisions too early. A physiologic digital twin is meant to preserve the shape of the trajectory.
The appeal here is not abstract. Ex vivo perfusion platforms generate continuous streams of data, and continuous data demand continuous interpretation. A model that can assimilate those inputs in real time may help transplant teams identify when an organ is genuinely improving versus when it is merely tolerating support. The same approach could also clarify when an intervention during perfusion is doing something meaningful, rather than simply changing the numbers on the monitor.
The transplant impact may be in the timing of judgment
The most immediate consequence of this approach is not a new biomarker in isolation. It is a different way of deciding when the liver is ready for transplantation, and when it is not. Prolonged perfusion already creates the biological space for rehabilitation; the digital twin is intended to create the interpretive space for using that time well. In practice, that means more informed decisions about whether a donor liver should be kept under support, reconditioned further, or withdrawn from consideration.
The concept also speaks to a persistent problem in organ utilization: viable grafts may be lost because confidence runs out before physiology does. By giving clinicians a model-based estimate of organ state, the digital twin could support decisions that are less dependent on isolated readings and more grounded in the trajectory as a whole. For livers that were previously considered marginal, that may be the difference between discard and transplant.
At the same time, this is a proposal rather than a validation study, and the article stays within that boundary. The framework still depends on accurate models, reliable data streams, and careful clinical integration. If the inputs are incomplete or the physiology is more complex than the model allows, the twin will be only as useful as its assumptions. The next step is not conceptual enthusiasm but rigorous testing against real perfusion outcomes, because the value of any digital twin in transplantation will rest on whether it can correctly track living organ behavior when the clock is running and the margin for error is small.