From digital twins to clinically trustworthy twins: a clinical-claim-based validation framework for personalized digital health.

Digital twins are increasingly presented as a computational foundation for personalized and preventive medicine, because they promise to integrate multimodal data into dynamic representations of patients, organs, diseases, or care pathways. Yet the translational maturity of medical digital twins remains limited. Many systems labelled as digital twins are still digital models, digital shadows, descriptive simulations, or prediction tools whose clinical claims exceed the evidence provided for calibration, uncertainty, transportability, causal validity, or real-world utility. This Perspective argues that the next bottleneck for digital twins in health is not model complexity but clinical trustworthiness. The main contribution of this Perspective is a claim-to-evidence typology that links the evidentiary burden of a digital twin to the clinical claim it makes, rather than to its computational architecture alone. This approach connects individual-level multimodal modelling with epidemiology, prediction science, causal inference, and implementation science. This article outlines a staged framework that distinguishes descriptive, predictive, counterfactual, interventional, and population-health clinical claims made by systems labelled or proposed as digital twins, each associated with a distinct evidentiary threshold. This article further proposes that validation should integrate verification, calibration, external and temporal validation, uncertainty quantification, fairness assessment, target trial emulation where causal claims are made, and post-deployment monitoring. Without such methodological discipline, digital twins may remain sophisticated but clinically fragile simulations. Conversely, population-calibrated and prospectively evaluated digital twins could become a robust infrastructure for personalized prevention, adaptive treatment, and learning health systems.