Digital twins in rheumatology: current status and potential

A patient digital twin (DT) is a virtual replica of a patient, physiological system or organ that represents anatomy, context and processes, and informs or influences decisions. It obtains and updates data from physical components and simulates in real time, and has predictive properties [1]. Its core attributes are dynamic and multidimensional data exchange, which differs from a digital model (involves no automatic data exchange) and a digital shadow (a system with unilateral real-time data flow from a physical component into a virtual setting). The components of a DT include the patient (organ ecosystem), data interaction (synthesis and amalgamation), virtual replica (disease, organ or patient), display panel (visualization) and synchronization (virtual model representation with data inputs) [2]. Simulation patient DTs use computational modelling to create viewable, stationary, and personalized digital replicas for predicting outcomes based on hypothetical scenarios and are used for one-time assessments only, while monitoring patient DTs are viewable, dynamic and personalized digital replicas which allow real-time continuous predictions of risk and outcome over time and provide feedback to aid the optimization of care [3]. It is envisaged that these two patient DT models could congregate to leverage their inherent properties together [3]. The concepts of DTs are increasingly emerging in medicine, particularly in precision medicine, individualized training and oncology. Its applications include hospital management, workflow optimization, decision-making, medical device design, drug discovery, in silico clinical trial design and surgical planning [4]. As the technology continues to evolve, validation studies have emerged for non-invasive functional substrate in scar-related ventricular tachycardia [5]. In rheumatology, DT applications could be used for patient simulation, treatment prediction, optimization and virtual trials to investigate surrogate outcomes. The technology can be used not only to produce a disease model but also to assess deviations from homeostasis and realign them towards an ideal baseline. One potential DT model could be a detailed virtual patient representation of a disease type based on data from multiple touchpoints, aiming to accurately predict changes in biological processes over time and in response to treatment (Fig. 1). Patient DT models integrate unique physiological and disease characteristics, multi-omics data, clinical parameters and exposomic factors to generate comprehensive patient profiles that inform precision therapeutics (Fig. 1). In rheumatology, it would enable continuous monitoring and analysis of patient health status, supporting early detection of potential health issues and timely interventions. Wearables, medical devices and ambient monitoring systems provide granular data that drive such dynamic models, which evolve in response to changes in the patient’s condition. Digital twin operational model for rheumatology with potential resultant impact and outcomes Subtle alterations in physiological parameters that precede clinical manifestations can also be identified through a continuous feedback loop, facilitating preventive measures and proactive interventions. By incorporating individual genetic variability, environmental exposures, lifestyle and predisposing factors, DTs in rheumatology can support interventions at multiple levels, including preventive, predictive, participatory and personalized approaches for each patient [6]. The simulation capabilities of DT facilitate the generation of diverse treatment scenarios without compromising patient safety, thereby enabling the optimization of treatment plans prior to clinical implementation. Rheumatologists can evaluate multiple therapeutic strategies, drug dosages and intervention timings using DT before applying them to actual patients. Virtual testing and rationalization approaches help reduce healthcare costs. Economic modelling indicates that in certain chronic conditions, hospital readmission rates may decrease by 25% through DT-driven patient monitoring, treatment planning and resource optimization [3]. Predicting resource utilization across staffing, diagnostics and patient pathways can address operational inefficiencies, leading to shorter hospital stays, reduced disability and fewer complications. The clinical decision-making process for rheumatologists can be enhanced by using DT-based clinical decision support systems. When integrated into the DT framework, machine learning algorithms and causal inference models enable the identification of patterns and correlations in patient data. This integration supports an empirical, evidence-based approach to decision-making in complex rheumatological contexts involving multiple interacting factors. Visualization within the model increases transparency and provides a clear rationale for recommendations. Rheumatological disorders often exhibit fluctuating patterns of disease stabilization and deterioration during treatment. Longitudinal monitoring and management are facilitated by maintaining a digital representation of each patient’s individualized health status over time, which enables the identification of risk trajectories, supporting the development of behavioural and preventive strategies. Additionally, episodic care events can be integrated into a continuous care model, enabling assessment of treatment consistency and long-term patient outcomes. Recruiting patients for clinical trials presents significant challenges, particularly for rare rheumatological disorders. Virtual trials utilizing DT can optimize clinical trial designs. They provide model-based estimates of individual patient outcomes that would have occurred in the control group [7]. Virtual simulations enable the determination of optimal drug dosages and predict treatment response based on each patient’s disease characteristics. Insights from virtual trials provide information on the efficacy and safety profiles of interventions, thereby increasing efficiency and cost-effectiveness. Enhanced by analytical functions, DT can provide an avenue to identify potential surrogate outcomes, and such an artificial approach has been applied in studies of RA [8]. At present, the use of DT within rheumatology is evolving rapidly. A recent study reported the creation of a multicellular virtual twin of the synovial joint in RA [9]. Key challenges in rheumatology include the acquisition and integration of disease data, as well as synchronization with multimodal data fusion, and finally, its due validation. The use of extensive patient datasets, which contain sensitive health information, is essential for DT modelling; however, this raises significant privacy and security concerns. Furthermore, the diversity of rheumatological diseases, variability in disease patterns and small relevant patient cohorts present additional obstacles to ensuring data uniformity, quality, accuracy and fairness [3]. The European Virtual Human Twins Initiative is a step towards fulfilling these unmet needs [10]. A cross-collaboration network of researchers, innovators and clinicians with access to data, assets and resources, supported by industry, researchers, academia and governmental organizations, aims to accelerate the development, integration and validation of DTs through large artificial intelligence foundational model training tailored to human health and diseases [10]. Future work on multi-scale DTs dedicated to rheumatological disorders may usher in a paradigm shift from reactive to proactive care. There are no new data associated with this article. S.K.: conceptualization, investigation, data curation, formal analysis, visualization, writing—original draft, final approval. V.R.: conceptualization, data curation, formal analysis, project administration, supervision, and writing—review & editing, final approval. No specific funding was received from any bodies in the public, commercial or not-for-profit sectors to carry out the work described in this manuscript. Disclosure statement: S.K. has declared no conflicts of interest. V.R. is a co-editor of Rheumatology (Oxford) and has not peer reviewed or made any editorial decisions regarding this article.