At its core, a digital twin is a continuously updated simulation model that learns over time. Unlike conventional computational models, digital twins represent dynamical systems that can model both populations and individuals, integrating longitudinal data, physiological measurements, and clinical information to simulate real-world health processes. Many are built on mechanistic models that can even provide causal information among system variables, which means they offer explainable models allowing not only to make predictions but leading to structural interpretations. That combination of prediction and explanation is what distinguishes a true digital twin from an ordinary model, and it is why the concept, which first gained traction in engineering and manufacturing around the year 2000, is now being translated into medicine with such urgency.

To gauge how quickly the field is moving, the authors searched the Dimensions database of abstracts and citations for the term digital twin in titles and abstracts. The search yielded 5,641 articles in total, including 2,908 identified from titles alone, distributed across 21 categories of the Health Research Classification System developed by the UK Clinical Research Collaboration Partners. Most articles fall within the broad category generic health relevance, with 4,367 publications and 57,828 citations, followed by cardiovascular with 529 publications and cancer with 322. The timeline of published articles in the top five categories from 2017 to 2025 reveals a striking pattern: while the general relevance of digital twins in medicine became apparent in 2019 within the generic health relevance category, uptake in all other categories did not occur until 2023. The authors note that this delay is understandable, given the challenges of translating the digital twin concept from engineering to medical applications.

The bibliometric analysis also reveals which research fields have contributed to this development. Among all HRCS categories, Information and Computing Sciences dominates all other research fields followed by Engineering. Interestingly, while the Manufacturing Engineering field contributes a substantial number of publications across all HRCS categories, the majority are concentrated in generic health relevance, suggesting a strong foundational focus with potential for further expansion into more specialized medical applications. When the authors looked at Clinical Sciences, the most widely studied medical areas were cardiovascular, metabolic and endocrine, and cancer. Smaller categories tell their own stories: when including categories with fewer publications, renal and urogenital shows the highest citation impact across all fields with an average of 28.80 citations per article, while the inflammatory and immune system and respiratory categories also exhibit relatively high citation averages, indicating elevated interest in these areas.

Some of the most compelling proofs of concept lie outside the individual patient altogether. One study applied a simulation-based digital twin of emergency medical communication centers to evaluate organizational changes on accessibility. The digital twin models emergency call center operations under decompartmentalized scenarios, showing that more flexible call distribution improves service quality. A related study demonstrated the use of digital twins combined with discrete event simulation to optimize workflows in critical care units, employing a novel dual-layer architecture that tracks both physical and conceptual entities in real time, with cloud-based monitoring of staff-performed tasks from observation forms. These results demonstrate how digital twins can provide an objective, data-driven tool to optimize emergency response systems, enhance patient safety, operational efficiency, and precision care by aligning actual clinical practice with ideal workflows.

Digital twin models of the human heart are among the most widely studied, and the technical sophistication of these systems is remarkable. One team presented a multi-scale electro-mechanical whole heart model that includes a coupled description of electrophysiology, mechanics, and a closed-loop model of circulation. This model allows the description of membrane kinetics, excitation-contraction coupling and active tension generation in the atria and the ventricles, and it can be personalized by adjusting characteristics such as active tension or conduction velocity. To tame the enormous computational cost of such high-fidelity simulations, researchers introduced a surrogate model for cardiac digital twins based on Latent Neural Ordinary Differential Equations. The model learns pressure-volume dynamics from high-fidelity simulations while reducing computational complexity, enabling fast analyses such as sensitivity studies and parameter estimation with uncertainty quantification, and thereby significantly lowering the computational barrier for cardiac digital twins.

Oncology is the second most widely studied disease category for digital twin applications, and two recent examples illustrate the breadth of the approach. Researchers developed a quantitative systems pharmacology-based digital twin framework for TCR-engineered T cell therapy in patients with solid tumors, including metastatic HPV-associated epithelial cancers and pancreatic cancer. These digital twins capture the kinetics of infused and endogenous T cells across blood, lymph nodes, tumor, and peripheral tissues. Patient-matched digital twins reproduce observed T cell dynamics and show that stem cell-like memory T cells are key determinants of expansion and persistence, explaining inter-patient variability. In silico simulations suggest that enriching these stem cell-like memory T cells enhances persistence and enables lower dosing, providing a quantitative tool to optimize T cell therapies. Separately, another team introduced a model to predict treatment-induced toxicity in patients with acute myeloid leukemia, employing a mechanistic model of venetoclax and azacitidine therapy informed by measurements of neutrophil counts and blast percentages from blood samples. This study demonstrated that continuously updating the digital twin with new data enhances its predictive performance over time.

Mental health is emerging as a surprising frontier for digital twin technology, with three distinct systems already demonstrating what is possible. One team proposed a digital twin-based framework implemented as a chatbot system that analyzes the mental status of individuals and provides personalized feedback based on the severity of their condition. The system leverages pre-trained BERT models fine-tuned on the E-DAIC dataset to detect varying severity levels, achieving high accuracy and usability. By creating virtual representations of mental states, this approach addresses barriers such as stigma, limited accessibility, and cost, enabling scalable screening and early intervention. A similar concept, a digital twin-inspired mental health monitoring system called MTWIN, creates a virtual representation of a patient’s psychological state by integrating personal and medical information, wearable sensor data, environmental factors, and emotion predictions from a machine learning model. Finally, the Digital Alzheimer’s Disease Diagnosis model uses personalized biomarkers derived from non-invasive EEG recordings to reconstruct patient-specific neurodegeneration, capturing synaptic and connectivity decline. The model predicts cerebrospinal fluid biomarker positivity and clinical conversions with high accuracy, outperforming standard EEG methods and demonstrating the potential of digital twins to transform early diagnosis in Alzheimer’s disease.

Despite these promising capabilities, the authors are candid about the challenges that must be addressed before medical digital twins become applicable for widespread clinical use. Limited access to high-quality, longitudinal, and multimodal patient data is a major obstacle, as digital twins are largely data-driven and repeated measurements over time are required to establish models that can be personalized to individual patients, though data from wearables may provide valuable complementary information alongside clinical and genomic sources. Ensuring the accuracy and robustness of digital twins remains a major barrier to clinical adoption and requires rigorous, task-specific validation against conventional approaches, with standardized benchmarks still an open challenge spanning data quality, model validation, clinical utility, interoperability, and uncertainty quantification. Computational complexity poses another hurdle: high-resolution multimodal data, physics-informed neural networks, and high-fidelity organ-level simulations are computationally intensive and require large-scale computing resources, making real-time model updating challenging, though hybrid physics-machine learning surrogate models and edge-optimized architectures can reduce latency. Regulatory issues represent a further major challenge, as existing frameworks are not yet fully adapted to dynamic, data-driven models that are continuously updated with new patient data, and questions around data privacy, ownership, and informed consent must be addressed.

Looking ahead, the survey identifies several promising yet demanding directions for future research. Slowly progressing diseases may be especially well suited for digital twins, since Alzheimer’s disease and Parkinson’s disease involve gradual changes over many years, making them ideal for longitudinal modeling and long-term data integration, with wearable sensor data capturing subtle disease-induced changes. Rather than creating separate disease-specific digital twins, integrating electronic health records allows the construction of a comprehensive digital twin representing the underlying patient, capturing the interactions and progression of multiple coexisting conditions. The concept extends beyond individual patients to entire health systems: public health systems may be represented by digital twins that model expenditures across diseases, resource allocation, and patient care pathways, simulating the impact of policy changes, new interventions, or demographic shifts on healthcare costs and outcomes. And just as industrial digital twins leverage interoperability for efficiency and scalability, medical digital twins may benefit from connected frameworks in which multiple specialized twins, such as heart, liver, metabolism, and mental health, share data, models, and simulations to address complex multi-factorial conditions. Medical digital twins, the authors conclude, are an emerging paradigm with the potential to profoundly transform medicine due to their ability to adapt to a wide range of problems.

Subject of Research: Digital twins in medicine

Article Title: Mapping the future of medicine through digital twins

Article References: Emmert-Streib, F., Tripathi, S., & Yli-Harja, O. (2026). Mapping the future of medicine through digital twins. Frontiers in Molecular Medicine, 6, Article 1840371. https://doi.org/10.3389/fmmed.2026.1840371

Image Credits: AI Generated

DOI: 10.3389/fmmed.2026.1840371

Keywords: digital twins, medicine, cardiology, oncology, mental health, health systems, machine learning, personalized medicine, wearables, Alzheimer's disease, data integration, clinical validation