Digital Twin Applications in Diabetes Management: Scoping Review

Introduction

A digital twin (DT) is a dynamic, virtual representation of a physical system—such as a patient—that is continuously updated with real-world data and computational models to support prediction, simulation, and decision-making [1]. In health care, DTs are a powerful tool for personalized medicine, providing real-time, data-driven insights tailored to individual patients [2,3].

Diabetes mellitus, encompassing both type 1 and type 2 diabetes, remains a major chronic health condition requiring highly individualized care [4,5]. The complexity of diabetes management—driven by variability in disease trajectories, treatment responses, and complication risks—requires approaches that move beyond traditional one-size-fits-all models. DTs address this need by simulating glycemic dynamics, forecasting outcomes, and supporting therapy optimization on a patient-specific basis [2,3,6,7]. These models integrate diverse data sources, such as continuous glucose monitoring (CGM), insulin dosing records, electronic health records (EHRs), wearable sensors, genomic information, and lifestyle factors [3,6,7].

Recent research highlights the potential of DTs in diabetes for applications, such as predicting disease progression, personalizing nutrition, enhancing automated insulin delivery systems, and supporting self-management [3,7-11]. For instance, DT frameworks that combine machine learning, multimodal data, and mechanistic modeling have been used to predict glycemic and complication-related outcomes in diabetes [3,7,8,12,13]. Early-phase clinical and real-world studies suggest potential improvements in glycemic control, reduced medication use, and enhanced metabolic outcomes with DT-based interventions [8,11,14-18].

However, several barriers still hinder broader adoption and clinical integration. Key challenges include data integration and model personalization [3,6,7], limited interoperability across devices and systems [2,6,7], the absence of standardized validation and regulatory pathways [2,6,7], and unresolved concerns around data privacy and ethical use [2,6].

Despite promising progress, DT research in diabetes remains fragmented and undervalidated. While some reviews have examined digital health tools in diabetes or explored DTs in general health care contexts [6], no previous review has systematically synthesized DT applications in diabetes across key dimensions such as system design, personalization, data integration, validation, and implementation. This gap limits the ability of researchers, clinicians, and developers to assess maturity levels, identify best practices, and guide future development.

To address this gap, we conducted a scoping review guided by the following research questions. The review addresses 13 research questions (RQs) grouped under 7 thematic domains to improve clarity and synthesis.

1. System design and modeling foundations:
   • RQ1: What types of DT models have been developed for diabetes care and management?
   • RQ2: What system components are included in these models?
   • RQ3: What modeling approaches are used in these systems?
2. Target conditions and use context:
   • RQ4: What types of diabetes are addressed by these DT applications?
   • RQ5: What clinical goals do these DTs aim to support?
3. Data sources and personalization mechanisms:
   • RQ6: What data sources are used to build or update DTs for diabetes?
   • RQ7: How are DTs used to enable personalized care or self-management in diabetes?
4. Intelligence and adaptability:
   • RQ8: How do the DTs handle uncertainty, real-time data updates, and model interpretability?
5. Evaluation and validation:
   • RQ9: What outcomes have been reported from applying DTs in diabetes care?
   • RQ10: What methods have been used to validate these DT systems?
6. Implementation and governance:
   • RQ11: What ethical or legal issues are raised regarding the use of DTs in diabetes care?
   • RQ12: What barriers and enablers are reported for implementing DT systems in clinical practice?
7. Research and development gaps:
   • RQ13: What gaps in knowledge or practice are identified in the literature on DTs in diabetes?

By systematically synthesizing evidence across these domains, this review provides a comprehensive overview of the current state of DT research in diabetes. The findings aim to inform researchers, clinicians, and technology developers about prevailing trends, methodological practices, and future opportunities for advancing personalized diabetes care through DT technologies [2,4,5].

Figure 1 presents a synthesized architecture of DT systems in diabetes based on the common components identified across the included studies.

‎

Methods

Overview

This scoping review was conducted in accordance with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) and JBI methodological guidance for scoping reviews [19,20].

Information Sources and Search Strategy

A comprehensive literature search was conducted through PubMed, IEEE Xplore, Scopus, and Web of Science. Studies published up to April 2025 were considered, and all databases were last searched on June 23, 2025. The search strategy combined terms related to “digital twin,” “diabetes,” and “healthcare” using Boolean operators. The detailed search strategy is provided in the Multimedia Appendix 1. Reference lists of included studies and relevant reviews were also manually screened to identify additional records.

The search strategies developed for PubMed, Web of Science, IEEE Xplore, and Scopus were imported into the Triple-A (Article Analysis Assistant) software [21]. The tool was used to integrate bibliographic metadata, automatically remove duplicate records based on DOI, and perform additional deduplication using title, publication year, and author names. Reviewer decisions were subsequently imported into the platform, and the finalized dataset was prepared for downstream analysis and thematic synthesis.

Eligibility Criteria

Eligibility criteria were established a priori to ensure consistency and reproducibility during screening.

Inclusion Criteria

Studies were included if they were original empirical research articles, including peer-reviewed journal papers, conference proceedings, or preprints. Studies were eligible if they reported on the development, validation, implementation, or clinical evaluation of DT systems for diabetes, including type 1, type 2, gestational, or related complications. Research involving patient-specific modeling, simulation, or data-driven approaches relevant to diabetes diagnosis, management, or treatment was included. Articles addressing applications in personalized or precision medicine, clinical decision support, or individualized therapy for diabetes were also included. Publications were required to be written in English, with a structured abstract and an accessible full text.

Exclusion Criteria

Studies were excluded if they were review articles, meta-analyses, editorials, commentaries, book chapters, or theoretical or conceptual papers without original data. Studies focused on DTs for diseases or systems other than diabetes, such as cardiovascular, neurological, or orthopedic applications, were excluded. Articles lacking an abstract or full text, or published in languages other than English, were also excluded.

These criteria were set before the screening process to maintain consistency and transparency in study selection. During screening, FSR, MJ, and KK independently assessed each record for eligibility using the predefined criteria. Discrepancies or uncertainties were resolved through discussion, with EB consulted when necessary.

Study Selection

The selection process involved 3 stages—identification, screening, and eligibility assessment. We initially identified 208 studies from 4 major databases—PubMed (n=47), IEEE Xplore (n=1), Scopus (n=107), and Web of Science (n=53). During identification, 85 articles were excluded due to duplication, lack of an abstract, absence of original data, or being published in a language other than English.

Following this step, 123 articles proceeded to screening. At the screening stage, 84 articles were excluded according to the predetermined exclusion criteria. As a result, 39 articles advanced to eligibility assessment, and 28 were included in the final review [3,7-18,22-36]. The 11 full-text articles excluded at the eligibility stage and the reasons for exclusion are listed in Multimedia Appendix 2.

Screening was conducted in 2 stages:

• Title and abstract screening: FSR, MJ, and KK independently assessed each record against the predefined eligibility criteria.
• Full-text screening: Articles passing the first stage were retrieved in full and assessed for final inclusion.

To ensure consistent inclusion decisions, the following screening questions were applied, reflecting the key characteristics of DTs and their application in diabetes care (Table 1). For the purposes of this review, a study was considered to describe a DT if it included a patient-specific virtual representation or individualized computational model linked to diabetes-related data and intended for prediction, simulation, monitoring, or decision support. Studies using terms such as “virtual patient” or “simulation model” were included only if these DT-defining characteristics were present. Generic population-level models without individualized representation or diabetes-specific application were excluded.

Data Extraction and Thematic Framework

Data from the 28 included studies were charted using a structured framework guided by the 13 predefined research questions introduced in the Introduction section. These research questions were organized into seven thematic domains to facilitate systematic synthesis: (1) system design and modeling foundations (RQ1, RQ2, and RQ3), (2) target conditions and use context (RQ4 and RQ5), (3) data sources and personalization mechanisms (RQ6 and RQ7), (4) intelligence and adaptability (RQ8), (5) evaluation and validation (RQ9 and RQ10), (6) implementation and governance (RQ11 and RQ12), and (7) research and development gaps (RQ13).

Each included study was analyzed systematically using this framework. Categories were not mutually exclusive, and individual studies could be charted under more than 1 category where appropriate. The full study characteristics and data charting table are provided in Multimedia Appendix 3.

Consistent with scoping review methodology, formal risk-of-bias, reporting bias, and certainty-of-evidence assessments were not performed because the aim was to map the breadth, characteristics, and gaps in a heterogeneous body of literature rather than to compare intervention effects or generate pooled estimates.

A review protocol and project materials for this scoping review were made available through the Open Science Framework (OSF) [37].

Results

Overview

Across the 28 included studies (as shown in the PRISMA [Preferred Reporting Items for Systematic Reviews and Meta-Analyses] flow diagram in Figure 2) [3,7-18,22-36], DT systems for diabetes exhibited diverse architectures, data sources, and application goals. Most models were data-driven or hybrid (artificial intelligence [AI]+mechanistic), while purely mechanistic and conceptual designs were less common. Core system components included machine learning (ML) or AI modules, decision support layers, and real-time simulation engines. The majority of DTs leveraged CGM, wearables, and lifestyle data, with increasing use of patient-specific models to enable personalized therapy, behavioral nudges, and simulation-based feedback.

‎

ML was the dominant modeling approach, while reinforcement learning, control theory, and signal processing appeared less frequently. Strategies for uncertainty management and interpretability were adopted inconsistently, with adaptive learning and explainable AI used in some studies, but with limited human-in-the-loop oversight. Reported outcomes most often focused on glycemic control (eg, hemoglobin A_1c [HbA_1c], time-in-range [TIR], and reduced hypoglycemia), alongside improvements in predictive accuracy, metabolic markers, and patient engagement. However, external clinical validation remained scarce, with most evaluations based on retrospective datasets or simulations.

Ethical considerations—mainly privacy and transparency, with occasional references to accountability and bias—were inconsistently addressed. Implementation barriers included validation limitations, data quality issues, model limitations, and workflow misalignment. Finally, the literature highlights persistent research gaps in integration with real-world systems, scalability, and methodological rigor that must be addressed to advance DT systems into clinical use.

System Design and Modeling Foundations (RQ1, RQ2, RQ3)

Overview

This section describes how DT models in diabetes are structured and modeled. It summarizes the types of models used (RQ1), the core system components included (RQ2), and the computational modeling strategies adopted (RQ3). Together, these questions cover the architectural and technical foundations of DTs in diabetes.

Model Types (RQ1)

DT models in diabetes care fall into 4 main categories—data-driven, hybrid, mechanistic, and conceptual. Data-driven models—most commonly using ML or deep learning—were used in half of the studies and focused on prediction and classification tasks. Hybrid models, which combine physiological modeling with AI, support real-time control systems, such as automated insulin delivery. Mechanistic models appeared less frequently and were primarily used in simulation studies. Conceptual frameworks were rare and largely theoretical. Table 2 summarizes the types of DT models reported in diabetes care, with representative examples from included studies.

Key findings included:

• Data-driven models (14 studies, 50%): Applied for HbA_1c forecasting, glycemic risk scoring, and behavior modeling [8,11-18,22,31,32,34,35].
• Hybrid models (10 studies, 35.7%): Enabled adaptive insulin dosing and feedback control by integrating ML with mechanistic physiology [3,7-10,23,29,30,33,36].
• Mechanistic models (5 studies, 17.9%): Focused on ODE-based glucose-insulin dynamics for simulation and metabolic exploration [24-28].
• Conceptual frameworks (1 study, 3.6%): Proposed theoretical DT architecture without implementation [28].

System Components (RQ2)

Most DT systems consisted of modular components supporting prediction, simulation, control, and user interaction. The most common modules were ML or AI components, followed by simulation engines and data integration layers. User-facing dashboards and decision support or control modules were also frequently described, while personalization layers, backend infrastructure, and rule-based systems were less common. Table 3 summarizes the system component categories reported in diabetes DT models, including their functions and representative examples.

Key findings included:

• ML or AI modules (20, 71.4% studies) were central to prediction, therapy optimization, and personalization [7-14,16-18,22,23,29-35].
• Simulation engines (17, 60.7% studies) provided physiological modeling and glucose-insulin dynamics for testing and validation [3,7,9,13,15,17,18,24-31,33,36].
• Data integration layers (12, 42.9% studies) supported real-time data collection from CGM, internet-of-things sensors, and preprocessing pipelines [3,8,10,13,14,16,22,23,32,34-36].
• User interfaces (11, 39.3% studies) enabled interaction for patients and clinicians through mobile apps or dashboards [8,13,14,16-18,23,25,31,32,36].

Modeling Approaches (RQ3)

The computational strategies used in diabetes DT systems reflect both the predictive and control needs of these models. While ML was the dominant method, several studies incorporated reinforcement learning, control theory, and signal processing for adaptive and real-time decision-making. Table 4 summarizes the range of modeling techniques reported across included studies, with representative examples.

Key findings included:

• ML was the most common approach (20, 71.4% studies), used for glucose prediction, patient modeling, and feature extraction [3,7-9,11-18,22,23,27,31-35].
• Statistical and probabilistic methods appeared in 9 (32.1%) studies, often applied to regression, inference, or survival analysis [3,7,8,15,23,31,33,34,36].
• Physiological modeling was reported in 8 (28.6%) studies, leveraging ordinary differential equations, compartmental models, and mechanistic representations [9,10,24-26,29,30,36].
• Control-based approaches were less frequent, with control theory (3, 10.7% studies) [10,26,27] and reinforcement learning (2, 7.1% studies) supporting adaptive insulin delivery and personalization [3,30].

Target Conditions and Use Context (RQ4, RQ5)

Overview

This section summarizes the specific types of diabetes addressed in DT studies (RQ4) and the clinical goals these models aim to support (RQ5). Together, these questions provide insight into intended use cases and patient populations for DT applications in diabetes care.

Target Conditions (RQ4)

DT studies in diabetes addressed multiple forms of the disease, with some models applicable to more than 1 type. Most studies focused on type 1 diabetes (T1D) or type 2 diabetes (T2D), whereas fewer studies targeted gestational diabetes or diabetes-related complications. Table 5 summarizes the targeted diabetes types and disease stages addressed across the included studies, with representative examples.

Key findings included:

• T2D was the most frequent target (14, 50% studies), with models supporting therapy optimization, metabolic simulation, and lifestyle interventions [3,7-9,13-18,23,28,32,34].
• T1D was addressed in 13 (46.4%) studies, primarily through closed-loop systems, real-time insulin delivery, and glucose control simulations [3,10-12,22,24-27,29,30,33,36].
• Diabetic complications were rarely considered, with 1 (3.6%) study focused on diabetic retinopathy [35].
• Gestational diabetes was examined in 1 (3.6%) study, reflecting limited application to pregnancy-related diabetes [31].

Clinical Goals (RQ5)

DT applications in diabetes addressed a broad range of clinical objectives, spanning real-time monitoring, safety, decision support, and long-term disease management. These goals were classified into primary categories reflecting their roles in clinical care. Table 6 summarizes the clinical goals of diabetes DTs, including their functions and representative examples.

Key findings included:

• Therapeutic control or intervention was the most common application (17, 60.7% studies), including insulin dosing, closed-loop control, and management of glycemic variability [3,8,10,13-18,23,25-27,29,30,33,36].
• Decision support and treatment planning were reported in 10 (35.7%) studies, covering dietary recommendations, therapy optimization, and clinician-facing guidance [3,7-9,23-25,28,33,36].
• Safety and alerting systems also appeared in 10 (35.7%) studies, emphasizing hypoglycemia warnings and proactive risk alerts [10-12,24-27,29,30,34].
• Disease prediction or forecasting was described in 9 (32.1%) studies, targeting HbA_1c trajectories, disease progression, and gestational diabetes risk [3,7-9,11,12,22,31,32].

In Table 6, “Therapeutic control or intervention” refers to systems that actively optimize or recommend treatment actions, such as insulin dosing or therapy adjustment; “monitoring or control” refers to systems focused on tracking glycemic status or physiological trends; and “decision support or treatment planning” refers to systems that inform clinician or patient decision-making without necessarily acting as real-time controllers.

Data Sources and Personalization Mechanisms (RQ6, RQ7)

Overview

This section summarizes the types of data used to construct or update DTs for diabetes (RQ6) and the mechanisms through which these models enable personalization or self-management (RQ7). These aspects reflect both the technical input and patient-centered application of DT systems.

Data Sources (RQ6)

DT models drew on a wide range of data sources to ensure an accurate representation of patient state and dynamics. These included lifestyle, sensor-derived, clinical, and synthetic datasets, with varying degrees of adoption across studies. Table 7 summarizes the data sources used in diabetes DT systems, with representative examples.

Key findings included:

• Lifestyle data were the most widely used input (20, 71.4% studies), covering physical activity, dietary intake, sleep, and behavioral logs [8-18,22-27,30,33,36].
• Wearable devices were incorporated in 19 (67.9%) studies, capturing heart rate, insulin delivery, and blood pressure [3,10-12,15-18,22-27,29,30,33,34,36].
• CGM appeared in 18 (64.3%) studies, enabling real-time tracking, control feedback, and risk forecasting [8,10-18,22-25,29,30,33,36].
• EHRs were used in 12 (42.9%) studies, providing longitudinal medical history, laboratory results, and medication data [3,7-9,13-18,32,33].

• Synthetic and public datasets were used in 6 (21.4%) studies, often for simulation or benchmarking, such as the UVa/Padova simulator or National Health and Nutrition Examination Survey (NHANES) data [26-28,31,32,35].

Personalization Mechanisms (RQ7)

Most DT systems aimed to enable personalized care through individualized feedback, adaptive modeling, or real-time decision support. Personalization strategies varied in scope, ranging from lifestyle guidance to therapy optimization and digital coaching. Table 8 summarizes the personalization features and tailoring strategies used in diabetes DT systems, with representative examples.

Key findings:

• Personalized lifestyle recommendations were the most frequent approach (11, 39.3% studies), providing tailored nutrition, activity, and daily routine guidance [3,8,9,13-18,23,24].
• Real-time or adaptive personalization was also reported in 11 (39.3% studies), offering interventions dynamically responsive to CGM and sensor feedback [8,10,13,15,17,27,30,31,33-35].
• Personalized insulin or therapy optimization appeared in 10 (35.7% studies), focusing on precision dosing, medication planning, and adaptive therapy [3,9,10,15,18,23-25,33,36].
• Individualized simulation models were described in 6 (21.4%) studies, enabling patient-specific scenario testing and comparative evaluation [3,27,29-31,36].

Intelligence and Adaptability (RQ8)

Overview

This section explores how DT systems in diabetes manage uncertainty, real-time data updates, and interpretability. These features are central to ensuring the trustworthiness, safety, and clinical relevance of DT models in dynamic health care settings.

Handling Uncertainty, Adaptation, and Interpretability (RQ8)

Based on the 28 included studies, 5 main categories of strategies were identified. Table 9 summarizes the strategies used for handling uncertainty, dynamic adaptation, and interpretability in diabetes DT systems.

Key findings included:

• Adaptive learning was the most common capability (18, 64.3% studies), enabling dynamic personalization through feedback loop tuning, model retraining, and continuous parameter updates [3,8-10,13-18,23,25-28,30,33,34].
• Explainable AI appeared in 16 (57.1%) studies, using methods such as feature importance analysis, visual interpretability, and knowledge graphs to improve transparency [3,7,10-12,15,22,23,25,26,29,32-36].
• Real-time synchronization was reported in 15 (53.6%) studies, supporting continuous data integration from CGM and other sensors via Kalman filtering and real-time updates [3,8,10-14,16,18,23,26,27,33-35].
• Confidence scoring approaches were applied in 12 (42.9%) studies, employing cross-validation, CIs, and robustness testing to quantify uncertainty [7-9,11,12,15,22,27,28,30,32,36].

• Human-in-the-loop oversight was reported in 3 (10.7%) studies, providing physician monitoring or manual intervention in safety-critical contexts [14,18,28].

Evaluation and Validation (RQ9, RQ10)

Overview

This section summarizes reported outcomes from DT applications in diabetes (RQ9) and describes the methods used to validate these systems (RQ10). Together, these questions address the effectiveness and credibility of DT models in clinical and experimental contexts.

Reported Outcomes (RQ9)

Across the 28 included studies [3,7-18,22-36], reported outcomes varied widely depending on the DT system’s clinical target and implementation maturity. Outcomes were grouped into major categories reflecting both clinical and system-level effects. Table 10 summarizes the clinical outcomes of DTs for diabetes.

Key findings:

• Improved HbA_1c or glycemic control was the most frequently reported outcome (17, 60.7% studies), showing HbA_1c reduction, increased TIR, and reduced variability [3,8-10,13-15,17,18,23-27,29,30,33].
• Other clinical benefits were described in 11 (39.3%) studies, including retinopathy or nephropathy improvement and cardiovascular risk reduction [8,15-17,22,23,27,31,32,34,35].
• Improved prediction accuracy was reported in 9 (32.1%) studies, with accurate glucose or gestational diabetes mellitus prediction and low root-mean-square error (RMSE) and mean absolute error (MAE) [3,7,9,11,12,22,31,32,35].

• Less frequently, outcomes included medication use reduction, weight or metabolic improvements, hypo- or hyperglycemia reduction, and other patient-centered measures.

Reported quantitative outcomes suggest that some DT applications were associated with clinically meaningful improvements, although results varied by study design and use case. In 1 retrospective T2D cohort, HbA_1c decreased from 8.8% to 6.9% after 90 days, corresponding to a 1.9 percentage-point reduction, together with a 56.9% reduction in homeostatic model assessment of insulin resistance, a 6.1% decrease in body weight, and 89.1% (57/64) of participants achieving time in range (70‐180 mg/dL) ≥70% after the intervention [18]. In a DT-based exercise decision support system for T1D, mean time in range improved from 80.2% to 92.3% for aerobic exercise and from 72.3% to 87.3% for resistance exercise, while time spent in low glucose decreased from 15.1% to 5.1% and from 18.2% to 6.6%, respectively [24]. A mechanistic personalized nutrition model in prediabetes predicted individual body weight and HbA_1c trajectories with mean prediction errors of 0.7 kg and 0.08 percentage points in the training dataset, and approximately 1.1% and 1.4% percentage errors, respectively, in the test dataset [30]. Some prediction-focused systems also reported strong performance metrics, including RMSE 24.96 mg/dL, MAE 17.21 mg/dL, and area under the receiver operating characteristic curve >0.85 for postprandial glucose prediction, as well as area under the curve (AUC) of 0.80‐0.82 for chronic kidney disease identification and AUC 0.86 for 3-year chronic kidney disease prediction in T2D cohorts [8,13]. In maternal-risk applications, 1 DT system reported 83.5% accuracy for maternal health risk assessment and 97.2% precision for gestational diabetes prediction [31].

Validation Methods (RQ10)

Validation approaches were grouped into 5 broad categories, reflecting how DT systems were evaluated for performance, safety, and generalizability. Table 11 summarizes the validation methods used in diabetes DT systems.

Key findings included:

• Quantitative evaluation was the most common approach (21, 75% studies), typically using accuracy metrics (eg, RMSE, MAE, and AUC) and cross-validation methods to assess performance [7-9,11-18,22,24,27-32,34,35].
• Retrospective validation was applied in 10 (35.7%) studies, using historical datasets (eg, EHRs and CGM logs) for training or testing and retrospective analysis [8,9,11,12,22,25,31,32,34,35].
• Simulation testing was reported in 9 (32.1%) studies, often leveraging tools, such as the UVa/PADOVA simulator or ReplayBG, to validate insulin control and metabolic models [3,8,10,24-27,29,30].
• Clinical and real-world evaluation was limited, with clinical evaluation reported in 4 studies (14.3%) [8,16,33,36] and real-world evaluation reported in 4 (14.3%) studies [8,13,23,33], including small-scale pilots, randomized controlled trials, or deployment in real patient settings with CGM tracking.

• Expert review was rarely used, reported in 2 (7.1%) studies, based on clinician or user feedback or case study evaluation [23,33].

Implementation and Governance (RQ11, RQ12)

Overview

This section describes how ethical, legal, and practical considerations are addressed in the implementation of DT systems for diabetes. It summarizes reported privacy and regulatory strategies (RQ11) and examines technical and workflow-related barriers to deployment (RQ12). Together, these questions assess readiness for safe, responsible, and scalable clinical integration.

Privacy, Ethical, and Regulatory Considerations (RQ11)

DT systems introduce complex ethical and legal considerations due to their reliance on sensitive health data and AI-driven decision-making. Among the 28 studies [3,7-18,22-36], 4 high-level categories were identified—data privacy, consent and transparency, accountability, and bias or fairness. Table 12 summarizes the strategies used for handling privacy, ethical, and regulatory issues in diabetes DT systems.

Key findings included:

• Data privacy was the most frequently discussed (8, 28.6% studies), typically through anonymization, encryption, and compliance with HIPAA (Health Insurance Portability and Accountability Act) or GDPR (General Data Protection Regulation) [3,8,9,28,33-36].
• Accountability appeared in 6 (21.4%) studies, including the use of audit trails, traceability, and regulatory compliance mechanisms [7-9,33-35].
• Consent and transparency were also reported in 6 (21.4%) studies, covering informed consent procedures, institutional review board approvals, and patient-facing disclosures [7,8,33-36].

• Bias and fairness were noted in only 1 (3.6%) study, reflecting a critical underexplored gap in addressing algorithmic inequity [3].

Implementation Barriers and Enablers (RQ12)

Although many DT systems demonstrated technical feasibility, real-world implementation remains constrained by several recurring challenges. These were grouped into 4 main categories—data quality and availability, model limitations, validation limitations, and workflow or interoperability barriers. Table 13 summarizes the implementation barriers that exist in diabetes DT systems.

Key findings included:

• Validation limitations were the most common barrier (16, 57.1% studies), reflecting reliance on synthetic datasets, short follow-up durations, and lack of external clinical evaluation [8,10-18,25-27,30,32,33].
• Data quality and availability issues were reported in 14 (50%) studies, including missing data, unreliable sensors, and burdensome data collection procedures [7,11,15,17,18,22,25,26,29-32,35,36].
• Model limitations were described in 11 (39.3%) studies, such as limited personalization, oversimplified physiological modeling, or small training datasets [3,7,10,13,14,16,26,27,29,31,32].

• Workflow and interoperability barriers appeared in 8 (28.6%) studies, emphasizing difficulties integrating DTs into clinical workflows, EHR systems, and device ecosystems [9,10,23,28,33-36].

Research and Development Gaps (RQ13)

Although DT systems for diabetes are showing technical feasibility, multiple areas require further investigation and refinement. From the 28 reviewed studies [3,7-18,22-36], seven major gap categories that emerged were (1) limited scope of application, (2) integration challenges, (3) lack of longitudinal data, (4) data quality and availability, (5) methodological limitations, (6) need for clinical validation, and (7) scalability or usability concerns. Table 14 summarizes the reported research and development gaps in diabetes DT systems.

Key findings included:

• Need for clinical validation was the most frequently cited gap (15, 53.6% studies), reflecting the lack of randomized trials, subgroup evaluations, and real-world testing [3,9,13-16,18,23,25,27,29,31,33,34,36].
• Limited scope of application was reported in 14 (50%) studies, with DTs often targeting narrow use cases and failing to generalize across diverse populations or multimorbidity contexts [3,7-10,13,15-17,25-28,31].
• Integration challenges were noted in 11 (39.3%) studies, underscoring difficulties with EHR interoperability, real-time deployment, and multidevice environments [7,11,12,22,25,27,30,32-35].
• Usability and real-world adoption also appeared in 11 (39.3%) studies, pointing to the need for personalization, support for multiple daily injection users, and strategies for broader adoption in routine care [8,10,12,16,24,26,30,32,34-36].

Discussion

This PRISMA-ScR–compliant scoping review maps the current state of DT systems in diabetes, addressing 13 structured research questions across 7 thematic domains.

System Design and Modeling Foundations (RQ1, RQ2, RQ3)

DT systems for diabetes use a wide range of modeling techniques, most commonly ML (eg, long short-term memory, gradient boosting, and reinforcement learning) and physiological simulation. Simulation engines and predictive ML modules were often integrated into layered architectures that also included personalization modules, decision support, and user-facing dashboards. Statistical and probabilistic methods (eg, regression and Bayesian inference) were also used in several studies, although less prominently. Few systems incorporated mechanistic control theory or signal-processing models. The inclusion of key components, such as simulation engines, control-feedback modules, and data integration pipelines, reflects a growing maturity in system design.

Target Conditions and Use Context (RQ4 and RQ5)

Most DTs targeted T1D or T2D, with limited applications in gestational diabetes or diabetes-related complications, such as retinopathy. Primary clinical goals included glycemic prediction, insulin-dose optimization, lifestyle guidance, and therapeutic planning. Several systems also addressed the diagnosis of complications or risk stratification for comorbidities. The breadth of clinical use cases suggests that DTs are evolving from simple simulators into multifunctional clinical-support tools.

Data Sources and Personalization Mechanisms (RQ6 and RQ7)

Lifestyle data, wearable devices, and CGM were the dominant inputs, with hybrid combinations being common. EHRs and synthetic datasets were also widely used to provide historical or simulated information. Personalization was achieved through mechanisms such as real-time adaptation, individual model tuning, behavior-driven feedback (eg, nudges), and insulin titration. However, persistent challenges remain in data quality, sensor integration, and dataset heterogeneity.

Intelligence and Adaptability (RQ8)

Managing uncertainty and real-time updates is crucial for clinical reliability. Studies implemented adaptive learning, feedback loops, and explainable-AI methods (eg, attention mechanisms and knowledge graphs) to improve transparency and adaptability. Real-time CGM synchronization and, in some cases, human-in-the-loop oversight were used to enhance model responsiveness and safety.

Evaluation and Validation (RQ9 and RQ10)

Quantitative validation (eg, RMSE and AUC) was common, but real-world clinical trials were rare. Most studies validated systems via retrospective datasets or simulations. Reported clinical outcomes included improved TIR, fewer hypoglycemic events, and, in some cases, T2D remission. However, evidence on long-term effectiveness, generalizability, and cost-effectiveness remains limited.

A notable finding across the included studies is the mismatch between technical sophistication and clinical maturity. Although many DT systems incorporated adaptive learning, individualized simulation, and multimodal data integration, most were evaluated using retrospective datasets or in silico simulations rather than prospective clinical deployment. This likely reflects the high implementation burden of DTs in diabetes, including the need for reliable real-time data streams, safety safeguards, interoperability with devices and clinical systems, and acceptable workflow integration. It also reflects the regulatory complexity of systems that may influence insulin dosing or therapeutic decision-making.

Implementation and Governance (RQ11, RQ12)

Privacy and ethical considerations were addressed inconsistently, often limited to brief compliance mentions (eg, GDPR and HIPAA). A smaller subset of studies explicitly discussed accountability (eg, audit trails and governance mechanisms) or algorithmic bias and fairness, highlighting underexplored areas of governance. Implementation enablers included real-time feedback and sensor integration, whereas barriers included poor data quality, system complexity, lack of clinical workflow alignment, and limited scalability.

Another important finding is the limited attention to algorithmic bias and fairness. Despite the increasing use of AI-driven modeling and decision-support approaches, only a small subset of studies explicitly discussed bias, representativeness, or equity-related concerns. This suggests that the field is still focused primarily on technical feasibility and predictive performance rather than equitable deployment across diverse patient populations.

Research and Development Gaps (RQ13)

Key gaps include limited clinical validation, insufficient longitudinal data, a lack of standardized model architectures, and limited generalizability to diverse populations. Many studies emphasized the need for integration with EHRs, real-world testing, and regulatory alignment. Addressing these gaps will be essential to enable scalable, equitable, and clinically robust DT systems for diabetes management.

Summary and Implications

This review offers a panoramic view of the evolving DT landscape in diabetes. While notable technical advances are evident—particularly in data integration and personalization—the field remains formative, with substantial work needed in clinical validation, ethical governance, and system interoperability. Future research should emphasize not only algorithmic sophistication but also real-world applicability, safety, and equity to support the scalable and responsible deployment of DTs in diabetes care.

Taken together, the literature suggests that DT research in diabetes is progressing from conceptual and simulation-based work toward more clinically relevant systems, but the field remains early in real-world maturity. Future studies should prioritize prospective validation, broader demographic and clinical representation, transparent reporting, interoperability with routine care systems, and governance frameworks that address privacy, accountability, and fairness.

Limitations

This scoping review has several limitations. First, only English-language studies with accessible full text were included, and gray literature was excluded, which may have led to the omission of some relevant studies. Second, formal risk-of-bias and certainty-of-evidence assessments were not performed because the aim was to map a heterogeneous body of literature rather than evaluate intervention effects. Third, the included studies differed substantially in design, terminology, validation methods, and outcomes, limiting direct comparison. Finally, many studies were early-phase, retrospective, or simulation-based, which limits conclusions about clinical effectiveness and real-world implementation.