Digital twins (DTs) are an emerging concept that has recently attracted the attention of researchers and engineers. The development of new technologies, such as virtual reality, blockchain, and the Internet of Things (IoT), is considered a key factor for progress in the DT research field [1]. However, the role of the recent boost in digital transformation and exponential growth in investments by giant tech companies cannot be neglected [2]. DTs have their roots in the field of engineering [3-6] but are quickly expanding to a wide range of applications, such as city planning [7-9], energy [10,11], retail [12,13], and health care [14,15]. Although DTs are an emerging concept, they integrate well-established technologies as underlying components. The origin of the DT goes back to the 1970s when NASA (National Aeronautics and Space Administration) used the “twin” concept by creating simulated environments (ie, mirrored systems) to monitor the spacecraft in the Apollo 13 program [16,17]. Although this instance may not be considered as an advanced DT today due to the lack of data exchange to allow continuous or periodic “twinning” of the digital to the physical [16,18], it represents a very good example of the potential of DT and what DTs can empower in various industries.

A DT of the city of Zurich, Switzerland [7], is a good example of a DT application. In this project, spatial 3D data is transferred to a DT to support urban planning decision-making, for example, simulating climate change and noise issues within urban plans [7]. Another example of a DT was built by Unilever PLC to increase the flexibility and efficiency of their production process [19]. They use DT systems to predict optimal process parameters for new formulations, execute simulations to identify the best operational conditions, and investigate complex what-if scenarios. DT systems have also been applied in the field of water management in Porto, Portugal [20], by creating a virtual representation of the city water network. This DT was developed to predict water quality and flooding issues, simulate and analyze burst pipe scenarios, and ensure the resilience of water infrastructure. The initial results revealed that the developed DT reduced water-supply failures by about 30% and decreased the time needed to repair burst pipes by 8% [20].

There have been several efforts to develop DTs for various purposes in health care. Clinical decision-making (CDM) and operational decision-making (ODM) are 2 of the most important decision-making processes in health care settings. In this context, CDM refers to the decisions made by health professionals in regard to direct health care delivery (diagnosis, test, intervention, etc) based on a patient’s health status and related clinical conditions, while ODM refers to the decisions made by managers and administrators after monitoring processes related to operations, such as patient waiting times and hospital revenue [21]. In subsequent sections of this paper, we present a scoping literature review to provide an up-to-date overview of the existing literature relevant to the application of DT systems in health care and to gauge the extent of the designed DT models for CDM and ODM purposes.

There have been several literature reviews on the applications of DTs in health care from different perspectives. Ahmadi-Assalemi et al [22] reviewed recent works in the field of precision health care and discussed the key enabling technologies (eg, IoT and cloud) of DTs. Hassani et al [2] provided a literature review that discussed the value of DTs in health care and proposed some key characteristics, including dynamic, real-time, and bidirectional data connections. Armeni et al [18] identified the opportunities as well as challenges regarding DT implementations in health care (eg, security and privacy, accessibility of the technology, and data collection and management). Elkefi and Asan [23] conducted a systematic review of studies that discussed the contribution of DTs to improving user experience in health care (eg, safety management or information management). They used digital twin and health as keywords and 17 papers were included in their review. Sun et al [1] provided a systematic review to explore the progress of prominent research on DT technology in medicine. They used the following terms: digital twin, medicine, digital health, and virtual healthcare. This resulted in the inclusion of 22 papers in their study after the screening process. They showed that the application of DTs to the cardiovascular system is an attractive area. Sheng et al [24] used structural topic modelling to analyze trends in DT+healthcare and their findings show that technology integration and practical applications, for example, IoT and artificial intelligence, are the main focus areas. Katsoulakis et al [25] conducted a scoping review on DTs for health. Using digital twin and health as search keywords, the authors selected 85 papers in their analysis. They first categorized the selected papers into 8 distinct categories based on their purpose and content, for example, DTs for biomarker and drug discovery as well as DTs in biomanufacturing. Then, they discussed the challenges hindering developments in this field, such as data privacy and security, computing infrastructure, and data quality and accuracy.

In this paper, we present a distinctive approach that distinguishes our work from existing literature reviews. The key distinctions are discussed herein. First, the health care industry and academia use various definitions, necessitating research consolidation to establish a unified understanding and ensure future research builds on robust foundations. This study conducts a systematic literature review and thematic analysis of 86 publications on DTs and offers a comprehensive analysis of elements and characterization of the concept, which is absent in existing review papers. This explicit definition of DTs, along with the identification of their elements and characteristics, enables researchers to understand what distinguishes a DT as a unique system, separate from other known ones. Second, DTs represent a burgeoning research field, particularly in health care, prone to misconceptions and misinterpretations. A significant gap in the literature lies in distinguishing established models, such as machine learning and 2D or 3D modeling from true DTs. This paper pioneers an effort to clarify that not all papers claiming to develop DTs genuinely do so. Our analysis reveals that some are simply predictive models or simulations, branded under the DTs trend. This trend risks undermining the integrity of the research. Thus, this paper aims to educate readers on discerning authentic DTs, emphasizing the essential components required for a system to qualify as such.

We reported the review conducted in this study following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines [66] (Multimedia Appendix 1).

The search was performed through 8 scientific databases, including PubMed, MEDLINE, Scopus, Web of Science, Embase, CINAHL, Cochrane, and gray literature databases in June 2024. As DTs for clinical and ODM within the health care domain is a very new area of research, we included most of the academic databases and gray literature databases (eg, Google Scholar and OpenGrey) with no date restrictions.

To ensure relevant studies were captured in our search process, besides digital twin*, we provided a list of search terms of which at least 1 needed to be mentioned within the title, abstract, or keywords, such as patient, hospital, Intensive Care Unit or ICU, surgery, clinic, and emergency. The following search terms were used with Boolean operators and wildcards: digital twin* AND hospital* OR *patient* OR health* OR ICU* OR ward* OR emergence* OR surger* OR ambulance* OR clinic* OR general practi* OR doctor* OR nurs*. The search strings used for all databases are shown in Multimedia Appendix 2.

To identify research papers aligned with our aims, publications had to meet the following criteria: (1) they had to describe the development of a DT model for health care, (2) they had to focus on CDM or ODM in health care (excluding papers focused on medical devices), (3) they had to propose a DT model with experiments, (4) any publication date was considered, (5) only peer-reviewed original research papers were included, and (6) only articles written in English were considered. Papers proposing ideas and frameworks or describing the capabilities of DTs without applications were excluded. Papers published as literature reviews, editorials, and conference posters were excluded; duplicated papers and conference papers later published as extended journal papers were also excluded.

A web-based tool for systematic or scoping reviews, called Rayyan [67], was used to conduct all stages of the screening process. We developed a standardized data extraction form in Microsoft Excel to tabulate specific information. From the included studies (ie, after performing title, abstract, and full-text screening against exclusion and inclusion criteria), the following data were extracted: the country of the first author, publication type, DT characteristics, the environment that DTs were tested within, the sample size, the service, the target area, the main underlying method used in the DTs, the type of data collected to build DTs, the International Classification of Diseases, 10th Revision (ICD-10) codes, and a summary of included studies. Most included studies are presented in the following section, while the rest are provided in Multimedia Appendix 2.

The ICD coding consists of 22 chapters (1 to 22), each covering specific disease areas and courses of treatment. While the papers included in our review did not explicitly provide ICD-10 codes, we classified them based on the specific disease areas and corresponding courses of treatment mentioned in the studies. For each paper, we first identified the main focus (ie, target area) and then mapped it to the relevant ICD-10 code, starting from the most specific level and tracing it to the highest chapter level. For example, if a study focused on prostate cancer, it would be assigned the ICD-10 code C61 (malignant neoplasm of the prostate), which is part of chapter 2, “Neoplasms.”

In addition, in our qualitative synthesis, we systematically assessed the characteristics of DT implementations based on the developed framework, focusing on ongoing updates of virtual representations, closeness to reality, and the ability for feedback to inform decision-making. We examined the studies for discussions on the frequency and methods of data updates from real-world entities, considering various sources of data. We noted potential delays in data capture that could affect the real-time functionality of the DT. To evaluate how closely the virtual representation mirrored the actual entity, we examined the validation methods and metrics used to measure functional similarity, ensuring the DT could reliably represent real-world outcomes. We also considered the number and integration of models within each DT, as using multiple model types can enhance accuracy. We examined the mechanisms for ongoing feedback loops within the DTs, focusing on their ability to analyze historical performance and guide future decisions. This included identifying decision-support tools and user interfaces designed to help decision makers act on feedback. We also assessed whether a closed-loop system existed, where the DT actively influences decisions in the real twin. In addition, we explored how feedback is presented to various stakeholders, ensuring that insights from the DT effectively support informed decision-making in clinical and operational contexts. In assessing multifunctionality, we analyzed the range of problems the DTs were designed to address, exploring the distinct functions they served within health care. This involved evaluating the extent to which each DT tackled multiple facets of health care challenges, such as diagnosis, treatment planning, and monitoring. By adopting this structured approach, we ensured a consistent evaluation across studies, capturing the presence of these key characteristics.

Two reviewers collaborated in designing this form to capture the relevant variables effectively. Each reviewer independently extracted data from the included studies, followed by discussions to resolve any discrepancies. In instances where the reviewers could not reach an agreement, a third reviewer was consulted to extract the necessary data for those specific cases.

To synthesize the collated data, we used descriptive statistics to present frequencies and proportions for various characteristics of the included studies. Moreover, we provided a more in-depth qualitative synthesis of findings in the Results section. This qualitative synthesis highlights the nuances of the studies, such as the effectiveness of DT implementations, identified barriers and challenges, and insights into the broader implications for health care practice. The data analysis was performed using both Microsoft Excel and R software.

A total of 5537 research papers were retrieved in the first step through all the given search databases. After screening titles and abstracts and removing duplicates, 815 (15%) papers were selected for full-text review. Considering the inclusion criteria, 86 (11%) out of 815 papers were selected for further analysis (ie, CDM: 75/86, 87% and ODM: 11/86, 13%). A total of 729 (89%) papers failed to meet the criteria. Around 34% (246/729) of these papers focused on providing ideas or frameworks and described the potential or capabilities of DTs which was the main reason for their exclusion. Moreover, 49% (355/729) of the papers were not relevant to the scope of this paper, for example, privacy and ethical aspects of DTs in health care, and big data and technology infrastructure required for DT implementations. The review process and results are shown as a PRISMA-ScR diagram in Figure 2.

Papers identified through our review reflect the current upward trend of publications in the DT research area as all screened and included studies were published after 2017 as shown in Figure 3. This figure also reveals the difference in research interest among researchers between CDM (more popular) and ODM.

Analysis of the distribution of the papers by the first author’s country shows that the United States (17/86, 20%), China (11/86, 13%), and India (10/86, 12%) are the countries with the highest number of published studies. Other countries, including France, Germany, the United Kingdom, South Korea, Austria, Australia, Netherlands, Italy, and Canada contributed to DT research with at least 2 papers each. Out of the 86 papers included, 65 (76%) were published as journal papers and the rest were conference publications.

Data plays an important role within DTs as it allows the physical and virtual entities to communicate together. Besides a small number of studies (5/86, 9%) that used synthetic [40,54,68,69] or randomly generated data [50], most studies used data collected from real entities. Furthermore, analyzing the results shows that the data collected through medical images (eg, magnetic resonance imaging and computed tomography scans) are rich sources to build DT models for various purposes, for example, the His-Purkinje System [31], human vertebra [70], breast cancer [71], tibial plateau fracture [72], the tibiotalar joint [48], knee joint [52], patients with liver tumors [73,74], and the human brain [75].

For further analysis, a summary of the included papers along with their DT and health-related features is provided in Table 2. In this table, we have included characteristics that are featured in the DTs (refer to column “DT characteristics”) based on the information reported in the article.

In Table 2, the CDM-based papers were categorized based on the targeted areas of the human body using the ICD-10 codes. Textbox 2 provides the details of all the chapters in ICD-10.

Studies included in this review covered more than half (14) of all 22 chapters. Chapter 19 (“Diseases of the circulatory system”; 14/86, 16%) and chapter 21 (“Factors influencing health status and contact with health services”; 12/86, 14%) include the most studies, followed by chapter 4 (“Endocrine, nutritional, and metabolic diseases”; 10/86, 12%). These findings demonstrate the diversity of health care areas where DT models have been applied.

In terms of health care services in the included studies, for CDM, optimizing treatment (20/86, 23%), early prevention (16/86, 19%), and personalized treatment (15/86, 17%) are the 3 main service areas studied in the included papers, while for ODM, facility management (3/86, 3%) and efficiency improvement (3/86, 3%) are the most popular areas.

Regarding the main computational approach used in the proposed DTs, mathematical modeling (24/86, 28%), for example, moving least square and physiological models is the most common method followed by simulation techniques, for example, discrete event simulation (17/86, 20%) and neural network-based algorithms, for example, deep convolutional neural networks and graph neural networks, (17/86, 20%). Unsurprisingly, simulation-based models, such as discrete event simulation (9/11, 81%) are the most popular method used for ODM (as they can allow decision makers to investigate “what if” scenarios by quantifying the impact of potential changes in the operational settings [132]. The detailed distribution of disease areas, services, and computational methods in the included papers are provided in the Multimedia Appendix 3.

The following was also observed for these included studies: out of 86, only 32 (37%) studies demonstrated ongoing updates of virtual representations in their developed system, although it is the most fundamental characteristic of a DT. Among those 32 studies, 16 (50%) tested the developed system in a simulated environment, and the remainder were tested in real environments. Of the 4 defined characteristics of DTs, multifunctionality is a characteristic that was found to be lacking in most of the studies (81/86, 94%). Among the included papers, all studies missed at least one of the DT’s characteristics, and therefore, did not materialize all 4 characteristics of DT.

Although none of the studies had all the desired DT characteristics, the reported impact of the deployed DTs offers promising insight into their effectiveness. For example, a DT-based program for patients with type 2 diabetes helped all 12 insulin-dependent patients who attended the program to stop insulin injections and helped 38 out of 56 to stop taking metformin [79]. A DT technology used for assisting a surgeon with real-time performance of thermal ablation (a treatment to destroy tumor cells) provided significantly better results compared to other existing methods [73]. A DT for estimating the risk of posthepatectomy liver failure (a leading cause of postoperative death) was developed by modeling blood circulation and estimating 2 important determinants of the disease that could not be predicted before the development of the DT [84].

The main finding of this study is that while DT systems in health care have shown promising results across multiple areas, they still do not fully incorporate all the desirable characteristics of a comprehensive DT. Nonetheless, DTs are making significant advances in clinical and ODM processes, particularly in areas such as cardiovascular disease, diabetes, cancer, orthopedics, and emergency department operations. The ability to integrate multiple techniques and technologies, such as big data, cloud computing, communication, virtual reality, blockchain, IoT, simulation, prediction, and optimization helps to create a digitally enabled environment for DTs to encapsulate and provide solutions for complex and multidisciplinary problems that were difficult to deal with using traditional methods. Key examples of such solutions include real-time monitoring, treatment optimization, risk factor early intervention, efficient clinical trial design, and improving operational efficiency.

The results of this study showed that the development and deployment of DTs within health care settings still lack maturity. Our review of the literature revealed a lack of a common definition and some inconsistency in the use of the term digital twin. Consequently, the focus of this study was shifted toward defining a DT and its underlying elements to support a common understanding and suggest a framework for designing and developing DTs. Building a DT is a complex process in terms of modeling, functionality, domain specificity, and data connectivity process. Therefore, research articles on DTs need to be transparent about their purpose (ie, functionality), the virtual representation that is being created, the data (eg, the frequency of data update, what data elements need to pass from the real entity to the virtual representation and vice versa), the feedback mechanism (eg, what are the outputs, who is receiving that feedback information, etc), and the computational models used (eg, the accuracy of the models and justification for the model's selection). This transparency helps any developed DTs be understandable and their potential for translation be gauged.

Assessing the identified studies against the defined characteristics of DTs revealed that while there are emerging applications of DTs for CDM and ODM, current efforts were not mature enough to include all desired features. As shown in Table 2, most of the reviewed studies did not include ongoing updates of virtual representations in their developed systems. Although they may not materialize a major characteristic of a DT, that is, fed by live data and synchronized with real-life events, they are proof-of-concept studies that provide evidence for the appetite for DTs in health care. One of the main aspects of DTs that is lacking in the included studies is the multifunctionality and scale of the proposed DTs. Studying the input and output data of the developed DTs revealed that they are mostly designed with a single purpose although DTs are designed to create a complete virtual description of a real entity that is accurate at both the micro and macrolevels. The size of the samples used in the included studies shows that the developed DTs are still in experimental stages, and to fully embrace DT technologies, they need to move from custom expert-driven implementations to accessible robust implementations at scale [133].

As DTs are an emerging research area, especially in health care, it is not surprising to witness misconceptions, misunderstandings, and misrepresentations. It has been discussed that DTs are closely related to other research areas, such as 2D or 3D modeling, system simulation, and digital prototyping and this is a big factor in this confusion [2,26]. Reviewing some of the definitions of DTs over the years provides insight into their growth and development. Considering the fact that researchers are still trying to adopt DT concepts and technologies from manufacturing to health care and figure out how DTs can be implemented in such complex systems, it is understandable that there are still a few steps to go in designing, building, and executing a true DT system within a health care setting. Reviewing the selected papers in this study revealed that DT applications in health care are still in a transformative stage moving from offline systems, for example, digital models and simulations without any real-time connection with the real entity, toward DTs encompassing the aforementioned characteristics, and there is improved awareness of what separates DTs from similar technologies as interest in this field grows.

While none of the reviewed studies possesses all desired DT characteristics, the current state of the deployed models suggests that DTs have the potential for substantial impact once fully developed. They also showcase the versatility of DTs in CDM and ODM applications, from screening, diagnosing, detecting disease, and personal treatment to clinical trial design and optimizing hospital operations. While interest in the applications of DTs in health care is evident, the key focus area might be in understanding how health care operations could benefit from DT use. Developing DTs for designing clinical trials exemplifies how they provide personalized health care by allowing patient-specific responses to therapies to be simulated and treatment plans to be optimized [76]. Nevertheless, the impact of DTs can be extended to population health by providing insights into disease patterns and treatment efficacy across varied populations. Economically, they reduce costs related to traditional clinical trials, such as participant recruitment and logistics, and also position organizations at the forefront of innovation, attracting investment and fostering industry growth. Socially, virtual testing minimizes risks to human participants and enhances trial transparency, as well as addressing disparities by generating inclusive data that accurately represents diverse demographic groups. Thus, DTs promise significant advancements, offering broader societal and economic benefits.

This study has several limitations. First, the scoping review was limited only to papers published in English. Second, the search process was restricted only to original papers with the term digital twin included in either the title, abstract, or keywords which may have resulted in excluding other closely relevant studies that may not have used this term. Third, in the inclusion criteria, we focused on DT applications within ODM and CDM; however, this may have led to excluding a portion of DT applications in other health domains, such as medicine and drug discovery. Fourth, this study reviewed a wide range of DT applications in health care delivery without critically assessing their reporting quality. Therefore, more detailed quantitative analyses can be considered as future research activities that are beyond the scope of this work. Fifth, while our search strategy was comprehensive, some relevant studies indexed under different terminology or metadata, such as MeSH terms, may have been unintentionally missed. Finally, there are articles published between the completion of the search process (June 2024) and the publication of this study that we were not aware of and therefore were not included in this review.

Despite the challenges and limitations highlighted earlier, it is important to acknowledge the strengths of our work. This study offers a comprehensive framework for defining the key characteristics of DTs in health care, providing an essential resource for future research and practical applications. Our review not only outlines the current advancements in DT technology but also pinpoints critical gaps between present capabilities and their full potential, especially in clinical and ODM. The proposed framework serves as a guide for researchers and health care providers in developing DT systems, offering a reference for incorporating all the desirable characteristics of a comprehensive DT to effectively implement and use these technologies. Considering the rapid development of DTs, our review is both timely and highly relevant for researchers and health care practitioners. We adopted a deliberately broad scope, encompassing studies from diverse clinical settings and health care disciplines to ensure a comprehensive understanding of DT applications. By incorporating both academic and gray literature, our findings are further strengthened, offering a solid basis for those working to develop and implement DT systems.

The application of DTs in health care is exponentially growing and it is expected to play a major role in decision-making processes at both operational and clinical levels. This scoping review presents a broad overview of currently developed DT technologies, and the key findings can be summarized as follows: before this study, a common definition of DTs was missing, and researchers often confused other approaches with DTs. In this study, we provided a detailed description of DT elements and characteristics and synthesized these into a universal definition and architecture for DTs for clinical and operational decision support. Second, most prior work has not materialized a major characteristic of DTs: although bidirectional data connection between a real entity and the corresponding virtual one is a fundamental element of a DT, the ongoing update characteristic is lacking in most of the included papers. Third, the implementation of DTs is still in its infancy. Only 19% of the included studies tested the developed DT in a real environment and 95% of them were designed with a single-purpose functionality. Further study is warranted to explore the potential of DTs, for example, integrating ongoing updates into DT models, evaluating practical implementations in health care settings (ie, assessing effectiveness, feasibility, and scalability), and enhancing versatility for broader functionality and seamless integration.