Digital System Physiology: Integrating Mechanistic Models, Artificial Intelligence, and Digital Twins

A measurement tells us something about physiology only when a model interprets it, and only when that model keeps the features that matter: organization across scales, feedback, adaptation, and coupling between organs. We use the term digital system physiology for a current version of systems physiology that links many kinds of measurement, mechanistic computation, and artificial intelligence with a clear method for reasoning from data to conclusions and a preset rule for updating. Cardiac electrophysiology is the most developed clinical example. Cardiorenal, metabolic, pulmonary, and neural models are useful but less mature. In all of these areas, the evidence is stronger for mechanistic insight, individualized simulation, and early clinical feasibility than for better patient outcomes. We separate how well a model predicts from how well it explains, and we propose working criteria in which prediction and being built to inform a decision define a twin, with the cadence of updating set by the clinical question rather than by a fixed rule. Progress from here will depend on tracking each patient's state over time, coupling organs together only where it helps, stating uncertainty openly, testing models on data they were not built from, and gathering prospective evidence that matches the intended use.