Introduction

Over the last decade, the use of artificial intelligence (AI) and clinical informatics in healthcare has increased dramatically, driven by advancements in computational power, increasing data availability, and the potential of these tools to transform healthcare [1].

In the last 5 years (2019–2024), the number of published articles on AI in healthcare has increased by over 600% compared to the number in the previous decade (2008–2018). Specifically, articles on digital twins (DTs), a subtype of AI, have increased by over 400%.

Despite the huge potential, not all AI applications have met expectations; IBM Watson, for example, faced challenges in delivering on its promise to revolutionize cancer treatment, which highlights the complexities of implementing AI in clinical practice [2]. However, successful AI applications, such as those in endoscopy, are more common. In 2019, a new AI tool was created to enable accurate detection, differentiation, and characterization of both neoplastic and non-neoplastic colon polyps, with an average accuracy rate of 91.5% [3]. Critical care is one of the most data-rich areas of healthcare and one of the largest areas where AI-related technologies can find fertile soil to grow and iteratively refine at a swift pace [4]. However, introducing these innovations in settings where patients are critically ill or in life-or-death situations can be challenging.

As discussions continue to intensify, providing an exploratory overview on the implementation and impact of DT technology has become increasingly relevant. By examining the role of DTs in critical care, we aim to provide an overview of how this innovative approach is transforming patient care and medical research.

1. Introduction to digital twins

DTs are virtual counterparts that replicate an actual physical entity from micro to macro levels. Some authors define DTs based on three main elements: (1) real space, where a physical entity, such as an object or a person, exists; (2) virtual space, where a digital counterpart of the physical entity is modeled; and (3) a data connector that enables bidirectional communication between the two spaces [5]. A DT prototype describes a prototypical physical artifact generated before the physical product exists. A DT instance represents an individual physical object after it is manufactured, mirroring the exact characteristics of the physical system (e.g., patient- or organ-system-level DTs) [6]. A DT environment is a logical environment in which software can interact to create an entire system [7] and can simulate scenarios at many levels, such as an intensive care unit (ICU), hospital, or the entire healthcare system. These virtual twins can be used in an in silico environment as testing beds for clinical interventions without patient harm [8]. DTs have been utilized as ambient intelligent and living models capable of optimizing processes (such as manufacturing and healthcare resource management) and predicting the future state of a model (treatment prediction models) [5]. The concept of DTs initially emerged in the engineering domain with the goal of creating a virtual representation of a physical artifact that could be constantly updated to reflect its current state. Over time, the concept of DTs has also been applied to the natural environment, including human beings [7].

2. Lifecycle of digital twins

The concept of DTs was first introduced by Dr. Michael Grieves in 2002, with the innovative concept of linking real and virtual spaces. Since then, this concept has been applied in the manufacturing sector. During the design phase, the models enable iterative optimization to facilitate predictions and problem identification before the manufacturing of real-world products. When models are built for integration with their physical counterparts, they can provide real-time monitoring, production control, and predictive maintenance. DTs provide continuous support throughout the operational phase by monitoring system behavior and predicting performance; however, maintaining accurate real-time data flow remains a challenge. Looking toward a product’s end of life, DTs are emerging as valuable tools in decision-making processes [9], and this characteristic is one of the key factors that makes them a significant promise for healthcare.

3. Digital twin implementation: from bench to bedside

One of the greatest advantages of using DTs in healthcare is the ability to make mistakes in virtual counterparts without exposing patients to avoidable harm. Early applications are being developed as learning tools for common clinical scenarios in the ICU. These tools allow users such as trainees, residents, and fellows to safely simulate clinical management and decision-making, thereby reducing the risk of harm. DTs can serve as clinical decision-support tools with high fidelity and as future validated versions. Recently, DTs have been enhanced with features from other types of AI, such as machine learning (ML) and deep learning, and are currently known as “intelligent DTs.” Intelligent DTs can provide active and continuous insights to aid human decision-making processes, making them even more valuable in healthcare practice [5]. For example, early disease prediction, identification of clinical deterioration, and triaging are areas in which intelligent DTs and other AI models are continuously being developed. The aim is to provide systems that can predict outcomes and the disease course during the early phases, providing lead times before clinical worsening (early identification or prediction of clinical deterioration) and triaging (identifying patients that need urgent attention compared to stable patients) [10].

With regard to predictive enrichment using DT algorithms in cardiovascular care, recent advances have been made, including electrocardiogram (ECG) interpretation, enabling more personalized care, and predicting patient outcomes. For conditions such as a left bundle branch block, DTs are being developed to simulate electromechanical coupling in the heart, allowing clinicians to better understand the patient’s condition and predict how the patient might respond to specific interventions such as cardiac resynchronization therapy (CRT). Similarly, AI-enabled ECGs have shown promise in identifying individuals with reduced left ventricular ejection fraction and predicting outcomes after CRT [11]. Moreover, DTs can be used to evaluate arrhythmias and predict the risk of future arrhythmic events. They can be integrated into ECG data to simulate how different therapies would affect heart rhythm. This provides a personalized, noninvasive method for testing various treatment strategies before applying them in practice [12].

Methods

To explore the topic, we conducted a comprehensive literature search, with the assistance of the Mayo Clinic Library, using the combination of the following terms: “digital twin,” “intensive care,” “ICU,” and “critical illness.” The search included the Embase, MEDLINE, Scopus, Web of Science, and Google Scholar databases. The search yielded 42 potentially relevant articles and abstracts. After excluding 28 articles because of duplication, irrelevance, review articles, or lack of full-text availability, the remaining 14 articles were reviewed to explore the role of DTs in the critical care field (Table 1 [8,13-25]).

Digital twins in critical care

The ICU presents a major opportunity for cyber-physical-human systems (CPHSs), a system that interconnects humans, physical systems, and cyber technologies to create the productive, next-generation care required to meet the demand for improved productivity and care [4].

The field of CPHS is growing, and currently, there are no consensus definitions of what constitutes CPHS. However, what is known is that CPHSs are founded on DT technology and offer promising potential for transformative ICU care [4]. Fig. 1 presents an overview of the current applications of DTs in critical care.

1. Mechanical ventilation for respiratory failure

Our literature search identified 14 papers, of which 11 discussed the implementation of DT tools to improve decision-making and treatment in the ICU. Five studies (Caljé-van der Klei et al. [14], Chakshu and Nithiarasu [20], Montgomery et al. [22], Weaver et al. [24], and Zhou et al. [25]) discuss the development of a DT tool for respiratory support, providing patient-specific invasive and noninvasive ventilation settings to reduce the risk of ventilator-induced lung injury and helping guide the optimization of ventilatory support. However, it is important to note that these models remain largely theoretical, and existing studies have not yet provided robust empirical validation, limiting their immediate clinical applicability.

2. Neurology and neurocritical care

Additionally, our group previously discussed the potential of these tools in acute ischemic stroke [17]. In that study, we proposed establishing an expert consensus for developing a DT model to treat patients with acute ischemic stroke in the neurocritical care unit using the Delphi process.

Despite promising theoretical frameworks, these tools need to be validated in neurocritical care settings, and further research is necessary to assess their efficacy in real-world scenarios.

3. Hemorrhagic shock, cardiogenic shock, and sepsis

Cannon et al. [15] discussed the possibility of using DTs to help manage hemorrhagic shock by predicting physiological responses to hemorrhage and providing better management for fluid resuscitation. Another study by Azampanah [19] focused on cardiogenic shock and explored a tool for monitoring left ventricular contractile function in patients undergoing venoarterial extracorporeal membrane oxygenation. Although promising, this tool has only been evaluated in five patients, indicating the need for large-scale studies to validate its clinical utility.

Sepsis is also a high-yield disease in DT applications. Three studies focused on the prediction and management of sepsis. An and Cockrell et al. [13] proposed a technology that implements DTs to treat sepsis, emphasizing real-time data integration, verification, validation, and uncertainty quantification. However, this technology remains a theoretical model without empirical data to support its application. Advancing this idea, Danesh et al. [16] presented an experimental model where researchers integrated DTs with dynamic ensemble learning (DEL), an ML method, to enhance the accuracy of sepsis prediction. Finally, a novel DT tool is in the advanced phase of development at the Mayo Clinic [21] to predict the responses of critically ill patients to specific treatments during the first 24 hours of sepsis. This tool has also been studied as a valid medical education resource for internal medicine trainees as part of their ICU rotation, with reduced cognitive burden and high system usability [8].

4. Medical education

The abovementioned tool developed at the Mayo Clinic has been discussed in two other articles by Trevena et al. [23] and Rovati et al. [8], with a specific emphasis on education. Both articles introduced alternative approaches for the same tool. This DT model, based on electronic health records, utilizes clinical variables from patients in the ICU and incorporates interactions between seven major organ systems. Based on this model, an iOS application was developed to deliver critical care education to residents and fellows [8]. This model depends on expert rules and the need for continuous updating. However, it holds great potential to transform medical education by providing more realistic simulations to manage patients who are critically ill. Trevena et al. [23] also discussed the possibility of the tool serving as an educational resource and bedside decision-support tool.

5. Resource management

Another approach to the same ICU DT model discussed previously was also developed at the Mayo Clinic by Zhong et al. [18]. These researchers adapted the SEIPS (Systems Engineering Initiative for Patient Safety) 2.0 model to conceptualize the current ICU system. The model was created to simulate interactions within the ICU environment under various conditions, such as exploring the real-time allocation of medical equipment, flexible staffing, and workflow adjustments. However, these developments are still in their early stages and prospective validation is necessary to establish their effectiveness in optimizing resource management and improving ICU efficiency.

Although these studies propose that this new technology could improve ICU efficiency and patient care, they remain largely theoretical, and further validation is required to confirm their practical benefits.

Challenges, concerns, and pitfalls

1. Expectations versus reality

A key challenge in implementing DTs in healthcare is that they are likely to handle sensitive medical data. In the industrial sector, DTs are primarily used for monitoring equipment performance, optimizing manufacturing processes, and predictive maintenance, where data are generally objective and structured [6]. In contrast, DTs in healthcare should be carefully regulated before implementation to anticipate and prevent any potential harm to patient’ rights.

The U.S. Food and Drug Administration regulates AI applications under the category of ‘Software as a Medical Device.’ However, there is still a deficiency in more specific regulations regarding advanced AI models such as DTs. In 2023, the White House issued a comprehensive executive order concerning the regulation, development, and use of AI across various sectors in the United States. The document specifies that companies must provide ongoing reports to the federal government regarding their activities, ownership, and test results. It also includes healthcare-specific guidelines in which the Secretary of Health and Human Services is directed to develop strategies and regulations for using AI in healthcare delivery, drug development, and patient safety [26]. However, the document did not provide specific information or guidelines regarding DT technologies. Additionally, there is no data addressing the use of AI in specific healthcare settings, such as critical care, a highly sensitive area involving very sick patients, where the implementation of such technologies would require even more stringent oversight and validation.

Most DT models remain largely theoretical and face significant challenges in real-world applications and validation. The existing regulatory gap exacerbates these challenges. For instance, some authors have highlighted the complexity of sepsis pathophysiology and patient variability, noting that data captured from plasma do not accurately reflect tissue-level processes, and genetic and epigenetic factors impacting treatment responses are often not considered [13]. Similarly, a study using DT Technology with DEL to predict sepsis was limited by its reliance on the MIMIC-IV (Medical Information Mart for Intensive Care IV) dataset from a single institution, thus reducing generalizability and potentially overlooking important temporal dynamics by using aggregated statistical features instead of raw time-series data [16].

A key limitation of the DT model for coronavirus disease 2019 (COVID-19) is the technical difficulty in integrating large volumes of real-time data into a unified framework, which may affect the accuracy of the model. Additionally, establishing real-time feedback mechanisms for the quality of experience remains challenging and has yet to be validated [27]. For patients with acute hypoxemic respiratory failure, the use of DTs revealed inaccuracies in the data owing to measurement difficulties in patients who were spontaneously breathing. Based on estimations rather than direct measurements, these models may not accurately reflect actual patient responses. Moreover, even after model validation, inherent uncertainties persist, particularly with small and specific patient cohorts [24].

Delays in manual data entry may compromise reliance on real-time or sequential data. Adapting to new diseases such as COVID-19 in transfer learning processes is complex and requires careful tuning and validation to ensure accurate predictions in new patient populations. In addition, models require constant updating and maintenance with new data and evolving healthcare practices to maintain their accuracy and relevance [20]. In conclusion, validating DTs presents significant challenges and often relies on expert experience, previous studies, and cadaver analysis. Bridging the gap between the predicted outcomes and real-life scenarios remains a critical challenge.

2. Ethics

The safe integration of sensitive patient data into healthcare technology has always been a high-risk and high-reward endeavor. The Internet of Things, software, DTs in healthcare, and AI/ML algorithms should follow sound principles of biomedical ethics (patient autonomy, non-malfeasance, utility, beneficence, and distributive justice) [28]. In critical care medicine, the use of DTs raises significant privacy concerns. Many DT models utilize patient data, including real images, raising concerns about how safely these tools can store sensitive patient information. Patients must agree to the use of their data in DTs; however, the complexity of the technology may make it difficult for patients to fully understand what they are consenting to [29].

Non-malfeasance, the principle of “do no harm,” is one of the core principles of medicine and a major concern with respect to DTs in healthcare. The possibility of data breaches or the misuse of sensitive patient information represents a significant risk. If patient data are mishandled or security measures fail, the harm could extend beyond privacy concerns and include emotional, psychological, and financial repercussions. Therefore, the personnel responsible for handling these technologies in the future should be thoroughly trained not only to prevent any potential privacy harm but also to address and resolve any security failures.

Additionally, if the data used to create and analyze DTs are biased or not representative of all groups, the quality of patient care and treatment decisions will be compromised. This could undermine patient trust in healthcare. There are concerns that reliance on DTs could erode the autonomy of health professionals, shifting decision-making away from clinical judgment. Therefore, the long-term effects of introducing DTs should be carefully monitored to ensure ethical application and equitable outcomes.

Although DTs in healthcare are designed to provide assistance by offering predictive insights and supporting clinical decisions, there is a fine line between assistance and overreliance. Healthcare professionals may become overly dependent on the outputs generated by DTs, which could lead to less critical thinking, fewer second opinions, or diminished use of clinical intuition in complex cases. The goal should be to use DTs as complementary tools to assist health professionals and not to replace them. Therefore, maintaining a balance between technology and investment in personnel knowledge and capabilities is crucial.

3. Low- and middle-income countries versus high-income countries

Integrating AI into healthcare is a complex challenge that spans technical, social, fiduciary, and ethical domains. It is estimated that AI implementation could reduce healthcare costs by 5% to 10% in the United States, leading to savings of $200 to $360 billion annually. AI could also bring non-financial benefits, such as increased healthcare quality and access, and improved patient experiences and provider satisfaction [30].

In high-income countries (HICs), advanced health infrastructure, large patient datasets, and well-established regulatory frameworks support technologies such as DT applications in critical care medicine. This is a factor that makes HICs successful in improving patient outcomes and allocating ICU resources more efficiently, thereby reducing unnecessary costs. Conversely, low- and middle-income countries (LMICs) experience a disproportionately higher burden of critical illness and associated mortality than HICs [31,32]. While AI has the potential to enhance healthcare outcomes in LMICs by improving diagnosis, triaging, and patient management through affordable smartphone-based tools, these countries face the challenge of needing large financial investments to implement these technologies effectively.

Additionally, these countries are grappling with a growing burden of chronic noncommunicable diseases that require high-quality integrated care. AI can help bridge these gaps by providing improved diagnostic capabilities and patient-centered care solutions, contributing to the global goal of reducing premature mortality [33]. To overcome these challenges, AI development in LMICs should be supported and progressively established to fully and equitably harness its potential. A comprehensive strategy should focus on improving clinical education, developing the necessary infrastructure, and gradually integrating AI technologies [34]. A summary of the challenges in implementing DTs in critical care can be found in Fig. 2.

Conclusion

While the implementation of DTs in critical care has shown promising benefits, such as improving professional productivity and patient outcomes, the challenges add an extra layer of complexity. Correctly addressing these challenges is crucial to ensure safety and maximize the potential of DTs in the medical field.

Article information

Conflicts of interest

No potential conflict of interest relevant to this article was reported.

Acknowledgment

The authors thank Keivan Naiale (PhD, Department of Anesthesiology, Mayo Clinic, Rochester, MN) for providing excellent feedback on the figures and concepts.

Funding

None.

Author contributions

Conceptualization: GAH, OK, AL; Formal analysis, Supervision: OK, AL; Methodology: MN, AL; Project administration: GAH, AL; Visualization: AL; Writing-original draft: GAH, MN, DMG; Writing-review & editing: MN, DMG, AL.

Fig. 1.

The applications of digital twin technology in critical care medicine. Patient-specific interventions refer to tailored treatments that are based on individual patient data, with the aim of improving outcomes. Clinical decision-support systems use data to assist in diagnostic and treatment decisions. Diagnostic predictive enrichment involves predicting disease progression using real-time data, allowing for early intervention. In silico models are digital representations of anatomy or physiology, used for training purposes. Personalized learning involves customized simulations designed to enhance medical training. Real-data simulations use actual patient data in training scenarios to provide realistic practice. Workflow management focuses on optimizing the delivery of care and the allocation of staff. Robotic process automation automates routine tasks, freeing up clinician time. Cost optimization involves simulating resource use to ensure that care is delivered in a cost-effective manner. Created with BioRender (by Nemet M; BioRender.com/x40j361).

Fig. 2.

The challenges, concerns, and pitfalls of digital twin (DT) application in critical care. This figure illustrates the main challenges associated with implementing DT technology in healthcare, particularly in critical and acute care settings. Real-world validation is needed for many DT applications, which are still theoretical. However, validating these applications in hospitalized patients presents risks, such as potential harm during the implementation process. Ethical concerns arise from the use of patient data for DT validation, particularly regarding patient privacy. There is a risk that personal information could be exposed during the data-sharing process, making strong safeguards essential. In terms of investment, low- and middle-income countries (LMICs) face resource shortages, which makes it difficult to implement DT technology. While DT has the potential to optimize resource allocation and reduce costs, significant investment is necessary to effectively integrate it into healthcare systems. Created with BioRender (by Gabriele AH; BioRender.com/t31t331).

References

• 1. Poalelungi DG, Musat CL, Fulga A, Neagu M, Neagu AI, Piraianu AI, et al. Advancing patient care: how artificial intelligence is transforming healthcare. J Pers Med 2023;13:1214.ArticlePubMedPMC
• 2. Hamilton JG, Genoff Garzon M, Westerman JS, Shuk E, Hay JL, Walters C, et al. “A tool, not a crutch”: patient perspectives about IBM Watson for Oncology trained by Memorial Sloan Kettering. J Oncol Pract 2019;15:e277–88.ArticlePubMedPMC
• 3. Bekbolatova M, Mayer J, Ong CW, Toma M. Transformative potential of AI in healthcare: definitions, applications, and navigating the ethical landscape and public perspectives. Healthcare (Basel) 2024;12:125.ArticlePubMedPMC
• 4. Chase JG, Zhou C, Knopp JL, Moeller K, Benyo B, Desaive T, et al. Digital twins and automation of care in the intensive care unit. In: Annaswamy AM, Khargonekar PP, Lamnabhi-Lagarrigue F, Spurgeon SK, editors. Cyber–physical–human systems: fundamentals and applications. Hoboken, NJ: Wiley-Academy; 2023. p. 457–89.
• 5. Gazerani P. Intelligent digital twins for personalized migraine care. J Pers Med 2023;13:1255.ArticlePubMedPMC
• 6. Lehner C, Padovano A, Zehetner C, Hackenberg G. Digital twin and digital thread within the product lifecycle management. Procedia Comput Sci 2024;232:2875–86.Article
• 7. Blair GS, Henrys PA. The role of data science in environmental digital twins: in praise of the arrows. Environmetrics 2023;34:e2789.ArticlePDF
• 8. Rovati L, Gary PJ, Cubro E, Dong Y, Kilickaya O, Schulte PJ, et al. Development and usability testing of a patient digital twin for critical care education: a mixed methods study. Front Med (Lausanne) 2023;10:1336897.ArticlePubMed
• 9. Pronost G, Mayer F, Camargo M, Dupont L. Digital twins along the product lifecycle: a systematic literature review of applications in manufacturing. Digital Twin 2024;3:3.ArticlePDF
• 10. Taneja I, Damhorst GL, Lopez-Espina C, Zhao SD, Zhu R, Khan S, et al. Diagnostic and prognostic capabilities of a biomarker and EMR-based machine learning algorithm for sepsis. Clin Transl Sci 2021;14:1578–89.ArticlePubMedPMCPDF
• 11. Chen J, Ezzeddine FM, Liu X, Del-Carpio Munoz F, Friedman P, Cha YM. Using artificial intelligence-enabled electrocardiogram to predict cardiac resynchronization therapy outcomes of left bundle branch area pacing. Europace 2023;26:euae007.ArticlePubMedPMCPDF
• 12. Thangaraj PM, Benson SH, Oikonomou EK, Asselbergs FW, Khera R. Cardiovascular care with digital twin technology in the era of generative artificial intelligence. Eur Heart J 2024 Sep 26 [Epub]. https://doi.org/10.1093/eurheartj/ehae619.Article
• 13. An G, Cockrell C. A design specification for Critical Illness Digital Twins (CIDTs) to cure sepsis: responding to the National Academies of Sciences, Engineering and Medicine Report “Foundational Research Gaps and Future Directions for Digital Twins”. ArXiv [Preprint] 2024;arXiv:2405.05301v2.
• 14. Caljé-van der Klei T, Sun Q, Chase JG, Zhou C, Tawhai MH, Knopp JL, et al. Pulmonary response prediction through personalized basis functions in a virtual patient model. Comput Methods Programs Biomed 2024;244:107988.ArticlePubMed
• 15. Cannon JW, Gruen DS, Zamora R, Brostoff N, Hurst K, Harn JH, et al. Digital twin mathematical models suggest individualized hemorrhagic shock resuscitation strategies. Commun Med (Lond) 2024;4:113.ArticlePubMedPMCPDF
• 16. Danesh A, Juraev F, El-Sappagh S, Abuhmed T. Integrating digital twin technology with dynamic ensemble learning for sepsis prediction in intensive care units. J Intell Inf Syst 2024;30:25–59.Article
• 17. Dang J, Lal A, Montgomery A, Flurin L, Litell J, Gajic O, et al. Developing DELPHI expert consensus rules for a digital twin model of acute stroke care in the neuro critical care unit. BMC Neurol 2023;23:161.ArticlePubMedPMCPDF
• 18. Zhong X, Sarijaloo FB, Prakash A, Park J, Huang CY, Barwise A, et al. A multidisciplinary approach to the development of digital twin models of critical care delivery in intensive care units. Int J Prod Res 2022;60:4197–213.Article
• 19. Azampanah K. A digital twin for cardiac monitoring: in patients undergoing venoarterial extracorporeal membrane oxygenation [master’s thesis]. Enschede, Netherlands: University of Twente; 2024.
• 20. Chakshu NK, Nithiarasu P. An AI based digital-twin for prioritising pneumonia patient treatment. Proc Inst Mech Eng H 2022;236:1662–74.ArticlePubMedPMCPDF
• 21. Lal A, Li G, Cubro E, Chalmers S, Li H, Herasevich V, et al. Development and verification of a digital twin patient model to predict specific treatment response during the first 24 hours of sepsis. Crit Care Explor 2020;2:e0249.ArticlePubMedPMC
• 22. Montgomery AJ, Litell J, Dang J, Flurin L, Gajic O, Lal A, et al. Gaining consensus on expert rule statements for acute respiratory failure digital twin patient model in intensive care unit using a Delphi method. Biomol Biomed 2023;23:1108–17.ArticlePubMedPMCPDF
• 23. Trevena W, Lal A, Zec S, Cubro E, Zhong X, Dong Y, et al. Modeling of critically ill patient pathways to support intensive care delivery. IEEE Robot Autom Lett 2022;7:7287–94.ArticlePDF
• 24. Weaver L, Shamohammadi H, Saffaran S, Tonelli R, Laviola M, Laffey JG, et al. Digital twins of acute hypoxemic respiratory failure patients suggest a mechanistic basis for success and failure of noninvasive ventilation. Crit Care Med 2024;52:e473–84.ArticlePubMedPMC
• 25. Zhou C, Chase JG, Knopp J, Sun Q, Tawhai M, Möller K, et al. Virtual patients for mechanical ventilation in the intensive care unit. Comput Methods Programs Biomed 2021;199:105912.ArticlePubMed
• 26. The White House. Executive order on the safe, secure, and trustworthy development and use of artificial intelligence [Internet]. Washington D.C.: The White House; 2023 [cited 2024 Sep 11]. https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/.
• 27. Khan A, Milne-Ives M, Meinert E, Iyawa GE, Jones RB, Josephraj AN. A scoping review of digital twins in the context of the covid-19 pandemic. Biomed Eng Comput Biol 2022;13:11795972221102115.ArticlePubMedPMCPDF
• 28. Lal A, Dang J, Nabzdyk C, Gajic O, Herasevich V. Regulatory oversight and ethical concerns surrounding software as medical device (SaMD) and digital twin technology in healthcare. Ann Transl Med 2022;10:950.ArticlePubMedPMC
• 29. Bruynseels K, Santoni de Sio F, van den Hoven J. Digital twins in health care: ethical implications of an emerging engineering paradigm. Front Genet 2018;9:31.ArticlePubMedPMC
• 30. Sahni NR, Carrus B. Artificial intelligence in U.S. health care delivery. N Engl J Med 2023;389:348–58.ArticlePubMed
• 31. Bartlett ES, Lim A, Kivlehan S, Losonczy LI, Murthy S, Lowsby R, et al. Critical care delivery across health care systems in low-income and low-middle-income country settings: a systematic review. J Glob Health 2023;13:04141.ArticlePubMedPMC
• 32. Losonczy LI, Papali A, Kivlehan S, Calvello Hynes EJ, Calderon G, Laytin A, et al. White paper on early critical care services in low resource settings. Ann Glob Health 2021;87:105.ArticlePubMedPMC
• 33. Turner HC, Hao NV, Yacoub S, Hoang VM, Clifton DA, Thwaites GE, et al. Achieving affordable critical care in low-income and middle-income countries. BMJ Glob Health 2019;4:e001675.ArticlePubMedPMC
• 34. Hosny A, Aerts HJ. Artificial intelligence for global health. Science 2019;366:955–6.ArticlePubMedPMC

Figure & Data

References

Citations

Citations to this article as recorded by