Effectiveness and safety of AI-driven closed-loop systems in diabetes management: a systematic review and meta-analysis

Abstract

Background

Diabetes is a metabolic disease that can lead to severe cardiovascular diseases and neuropathy. The associated medical costs and complications make timely and effective management particularly important. Traditional diagnostic and management methods, like frequent glucose sampling and insulin injections, impose physical injuries on subjects. The development of artificial intelligence (AI) has opened new opportunities for diabetes management.

Methods

We conducted a meta-analysis integrating existing research, identifying a total of 1156 subjects to assess the effectiveness and safety of AI-based wearable devices, specifically closed-loop insulin delivery systems, in diabetes treatment.

Results

Compared to standard controls, AI-based closed-loop systems can analyze glucose data in real-time and automatically adjust insulin delivery, resulting in reduced time outside target glucose ranges (SMD = 0.90, 95% CI = 0.69 to 1.10, I² = 58%, P < 0.001).

Conclusion

AI-based closed-loop systems enhance the precision and convenience of diabetes treatment. This meta-analysis providing essential references for clinical treatment and policymaking in diabetes care.

Introduction

Diabetes has emerged as a significant public health issue worldwide. Currently, the number of adults with diabetes globally exceeds 800 million, with the prevalence among adults rising from 7 to 14%. Notably, nearly 450 million adults aged 30 and older (approximately 59% of all adults with diabetes) have not yet received treatment [1]. Diabetes not only results in substantial medical costs [2, 3] (at least 966 billion dollars) but also increases the risk of serious complications such as diabetic retinopathy, stroke, cardiovascular disease, diabetic nephropathy, peripheral vascular disease, diabetic neuropathy [4, 5] (Fig. 1). Therefore, timely and effective diabetes management is crucial for improving quality of life of people with diabetes and alleviating the societal healthcare burden [6, 7]. However, managing diabetes is a daily challenge, and people with diabetes need to remain highly vigilant and actively engage in self-care [8, 9], includes closely monitoring glucose levels and administering insulin injections when necessary [10]. Traditional diagnostic methods often involve frequent fingerstick glucose tests, and insulin therapy remains a cornerstone of management for many individuals [11, 12]. While these methods are clinically effective, ongoing research continues to explore ways to optimize their implementation and reduce subject burden [13, 14].

In the current context, the application of artificial intelligence (AI) in the medical field is advancing at an astonishing pace, particularly in disease management and treatment decision support [15]. AI technologies, such as machine learning (ML) and deep learning (DL), are capable of analyzing and processing vast amounts of data, revealing potential patterns and trends [16, 17]. Recent studies have demonstrated AI’s capability to predict both immunological progression in type 1 diabetes using continuous glucose monitoring (CGM) and genetic data [18], as well as the onset of diabetic complications like retinopathy using ML models [19], underscoring AI’s expanding role across the diabetes care continuum. In diabetes management, the application of AI can be divided into two aspects: those relevant to healthcare providers and those pertinent to people with diabetes. For clinicians, AI can analyze large-scale subject datasets, generating retrospective insights that assist in optimizing treatment strategies [20, 21]. These insights include trends in glucose fluctuations and individualized treatment effect analyses, thereby enhancing the accuracy and efficiency of interventions [22,23,24,25]. For people with diabetes, AI integrated into continuous monitoring systems can provide real-time decision support, aiding them in daily self-management [10, 11, 26,27,28]. The application of AI algorithms in automated insulin delivery (AID) and CGM systems [29,30,31] has created closed-loop systems resembling an artificial pancreas [32, 33]. This system can make instantaneous insulin dosing decisions based on real-time data analysis from CGM sensors, thereby reducing the need for continuous people with diabetes monitoring [34, 35]. Specifically, the establishment of closed-loop systems relies on the integration of CGM systems (such as Dexcom G6 or Freestyle Libre) with insulin pumps (such as Medtronic or Tandem). The role of AI algorithms is to optimize data processing to adjust insulin delivery strategies in real time [32, 36]. For instance, AI can analyze historical glucose data alongside current CGM readings to predict trends in glucose fluctuations and adjust insulin delivery accordingly, maintaining glucose within target ranges. This intelligent regulation helps mitigate the risks of hyperglycemia and hypoglycemia, thereby improving overall glycemic management [29, 31, 37,38,39]. This innovative approach not only enhances the quality of life for people with diabetes but also opens new avenues for the future of diabetes treatment. By combining AI algorithms with AID systems, people with diabetes can benefit from more efficient and personalized treatment plans, significantly improving their disease management outcomes.

Here, we aim to conduct a systematic meta-analysis of the included studies, such as randomized controlled trials (RCTs) and crossover trials, to assess the effectiveness and safety of AI in diabetes management. Our search databases include PubMed, Cochrane Library, and ClinicalTrials.gov, with keywords encompassing diabetes, AI, and closed-loop systems. We analyzed two categories of AI-based interventions such as CGM and AID. Effectiveness will be evaluated by changes in the time spent in the target glucose range (70–180 mg/dL) [40, 41], while safety assessments will focus on the incidence of severe hypoglycemic events and diabetic ketoacidosis (DKA) [42,43,44,45], time below range (TBR) < 70 mg/dL and < 54 mg/dL may also be considered. This meta-analysis will help identify the optimal treatment regimen by comparing the efficacy outcomes of different treatment modalities, such as sensor-based and traditional care. The integrated evidence will elucidate the role of AI in improving glucose and insulin communication, potentially providing insights for regulatory guidelines for AI-based diabetes devices and guiding future research directions.

[IMAGE OMITTED: SEE PDF]

Methods

Study design

This review was based on the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines [46]. The protocol was prospectively registered in the PROSPERO platform (CRD42024592213).

Search methodology

A comprehensive literature review was conducted to identify relevant studies focused on AI-driven closed-loop systems for diabetes management. The databases systematically searched included PubMed, the Cochrane Library, and ClinicalTrials.gov. This search was performed on July 20, 2024, and covered studies published from January 2000 to July 2024. The review was limited to articles published in English and the detailed search strategy is outlined below:

(“type 1 diabetes” OR “type 2 diabetes” OR “diabetes mellitus”) AND (“artificial intelligence” OR “machine learning” OR “AI”) AND (“wearable devices” OR “continuous glucose monitor” OR “artificial pancreas” OR “automated insulin delivery” OR “closed loop” OR “automated insulin delivery (AID)”).

Criteria for selection

The selection criteria for this review were explicitly limited to studies that met the following requirements:

1. (1)

Studies are designed such as RCTs, parallel or crossover trials.

2. (2)

Data compares closed-loop systems with a control group.

3. (3)

Participants have been treated with insulin for at least six months before enrollment.

4. (4)

Studies report all the relevant outcomes we mentioned, including safety-related issues.

The primary outcome was the percentage of time during which glucose levels were maintained within the target range (time in range, TIR, 70–180 mg/dL). Secondary outcomes included: the percentage of time spent above the target range (time above range, TAR, subdivided into > 250 mg/dL and > 180 mg/dL), and below the target range (TBR, subdivided into < 70 mg/dL and < 54 mg/dL), mean glucose level, glycated hemoglobin (HbA1c) level, as well as the incidence of severe hypoglycemia and DKA. Severe hypoglycemia and DKA were reported regarding the number of individuals affected and the total number of events.

Data extraction

The information extracted included various study characteristics, such as the author, year of publication, study design, type of closed-loop device, therapy administered to the control group, and duration of the intervention. Additionally, subjects’ demographics were recorded, encompassing age, type of diabetes, average duration of diabetes, and baseline HbA1c level. Clinical outcomes were also extracted, including the percentage of TIR, percentage of TAR, percentage of TBR, mean glucose level, HbA1c level, and the cases related to severe hypoglycemia and DKA [47,48,49,50]. For data presented as median values with interquartile ranges (IQR), we estimated the mean and standard deviation using validated methods for skewed distributions [51, 52]. These approaches are used for meta-analysis when individual participant data are unavailable and have been applied to metabolic parameters, including glucose levels [53]. To facilitate comparisons across studies, the standard deviation (SD) was calculated using the formula SD = IQR/1.35, as recommended [52]. While this approach allows for a more uniform representation of glucose data across studies, it is acknowledged that inherent variability and potential skewness in the original data may limit the robustness of these assumptions.

Risk of bias assessment

The risk of bias in the included studies was assessed using the Cochrane Collaboration tool [54]. Each study underwent a comprehensive examination of various potential sources of bias, including selection bias, performance bias, detection bias, attrition bias, and reporting bias. Studies were categorized as having a low, high, or unclear risk of bias.

Meta-analysis

A random-effects meta-analysis was performed with Review Manager 5.4 (RevMan 5.4) software. A meta-analysis was conducted to systematically evaluate the aggregated impact of AI-driven closed-loop systems on significant clinical outcomes in diabetes management. The standardized mean difference (SMD) between AI-based wearables subjects and controls was calculated for all data, with 95% confidence intervals (95% CIs). The degree of heterogeneity among the included studies was assessed using the I2 statistics, and a random-effects model was applied where appropriate. SMD > 0.8 was classified as a large effect size, p < 0.05 was considered statistically significant, and I2 ≥ 75% was considered to indicate high heterogeneity [55]. The presence of publication bias was evaluated through funnel plots.

Results

Study selection and characteristics

We conducted a systematic literature search, identifying 3,344 records. After excluding 1,345 duplicates and 1,929 irrelevant articles based on titles/abstracts, we evaluated 70 full texts for eligibility, ultimately selecting 8 studies for meta-analysis (Fig. 2). These studies enrolled 1,156 subjects and compared closed-loop therapy with conventional diabetes management (sensor-augmented pumps or multiple daily injections). The included trials exhibited methodological diversity, seven focused on type 1 diabetes (T1D) and one on type 2 diabetes (T2D), with participant ages spanning 3.8–69.3 years and follow-up durations ranging from 72 h to 6 months, baseline characteristics were generally balanced across groups. The closed-loop systems employed heterogeneous algorithmic approaches (such as DIAS USS, Cambridge hybrid, MPC, Control-IQ) and device configurations (such as t: slim X2, iLet bionic pancreas, Omnipod 5). Notably, while all systems utilized CGM, the specific CGM models and their integration with insulin pumps were not uniformly reported across studies. For instance, some trials explicitly mentioned Dexcom G6 or FreeStyle Navigator II use, whereas others provided limited technical details. This lack of standardization in reporting CGM-pump communication may influence the interpretation of system efficacy, particularly regarding real-time glucose data accuracy and algorithmic responsiveness. Control arms predominantly used conventional therapy with CGM, though the level of integration (such as sensor-augmented pumps vs. standalone CGM) varied. Importantly, the role of AI in these systems ranging from basic threshold alerts to advanced predictive hypoglycemia prevention was seldom explicitly discussed, despite its potential to enhance closed-loop performance through adaptive learning and personalized insulin dosing. Study designs included RCTs (n = 7) and single-arm prospective studies (n = 1), with sample sizes varying from 20 to 256 participants. Notably, Brown, S. A. stratified results by age (children [6.0-13.9 years] and adults [14.0–70.0 years]) [56], requiring separate data extraction (Table 1 and Table S1). This clinical, technological, and reporting heterogeneity, particularly in CGM specifications and AI integration, underscores the need for cautious interpretation of pooled outcomes.

[IMAGE OMITTED: SEE PDF]

[IMAGE OMITTED: SEE PDF]

Risk of bias assessment

We assessed the quality and the risk of bias in the included studies. The results showed that 75% of the studies had a low risk of bias in random sequence generation. In comparison, 25% had uncertainties, and no studies showed a high risk of bias, indicating that most studies employed appropriate randomization methods. In evaluating allocation concealment, 62.5% of the studies exhibited a low risk of bias, 25% had uncertainties, and 12.5% showed a high risk of bias. Regarding participant and personnel blinding, 25% of the studies had a low risk of bias, 25% had uncertainties, and 50% faced high risk. This high proportion of bias risk related to blinding may impact the objectivity of the results. In the assessment of outcome blinding, 25% of the studies showed a low risk of bias, 50% had uncertainties, and 25% had a high risk. Notably, all studies reported no issues with missing outcome data due to a high loss rate to follow-up or withdrawal, resulting in a 100% low loss bias risk assessment. Furthermore, all studies performed well in reporting bias, with 100% showing a low risk of reporting bias, indicating that researchers did not selectively report specific outcome measures, ensuring transparency of results. However, 12.5% of potential biases may still need to be explicitly listed, which could stem from shortcomings in study design (Fig. S1-2).

Primary outcome: glucose level in the range of 70–180 mg/dL

Eight studies investigated the TIR between AI-driven wearables and standard care, with TIR reflecting the percentage of time glucose is in the target range in subjects (Table S2). All studies defined the range as 70–180 mg/dL, except for the study by Thabit, H., which used a target range of 5.6–10.0 mmol/L (approximately 100–180 mg/dL) [60]. A higher proportion of time spent in the target range indicates glucose management that reduces symptoms of diabetes and helps to avoid complications. Glucose levels below 70 mg/dL can lead to acute symptoms such as dizziness, fatigue, and loss of consciousness, while persistent levels above 180 mg/dL increase the risk of long-term diabetes complications, including neuropathy, retinopathy, and cardiovascular disease. Although DKA can occur with severe hyperglycemia, chronic exposure to elevated glucose levels primarily contributes to progressive organ damage over time. The combined results showed that compared to conventional therapy, closed-loop therapy maintained a longer duration in the target glucose range, with no high heterogeneity present but a significant difference (SMD = 0.90, 95% CI = 0.69 to 1.10, I2 = 58%, P < 0.001), highlighting the greater effectiveness of closed-loop systems in glucose management (Fig. 3).

[IMAGE OMITTED: SEE PDF]

3.4 Secondary Outcomes: Glucose level > 250 mg/dL, Glucose level > 180 mg/dL, Glucose level < 70 mg/dL, and Glucose level < 54 mg/dL.

TAR > 250 mg/dL typically indicates a hyperglycemic state, which can lead to acute complications such as hyperglycemic hyperosmolar state (HHS), with symptoms including polyuria, thirst, and fatigue [64, 65]. A higher proportion of TAR indicates more time above the glucose goals. The combined results showed that for TAR, closed-loop therapy demonstrated a lower percentage of time than standard care (SMD = -0.74, 95% CI = -0.94 to -0.53, I2 = 60%, P < 0.001). The studies had no high heterogeneity, but significant differences were observed (Fig. 4a). TAR > 180 mg/dL is often due to postprandial hyperglycemia, and prolonged levels at this threshold increase the risk of cardiovascular disease [66,67,68]. The meta-analysis also showed similar results, with closed-loop therapy exhibiting a significantly lower percentage of time (SMD = -0.74, 95% CI = -0.95 to -0.52, I2 = 57%, P < 0.001), with no heterogeneity but significant differences, indicating statistical significance (Fig. 4b).

TBR < 70 mg/dL indicates hypoglycemic events that may lead to dizziness, palpitations, sweating, and hunger [69]. Closed-loop therapy significantly reduced the time below the target glucose range compared to conventional treatment (SMD = -0.33, 95% CI = -0.57 to -0.09, I2 = 71%, P < 0.01), with no high heterogeneity but significant differences observed. The combined results show less time below 70 mg/dL, indicating a more effective reduction in hypoglycemia (Fig. 4c). TBR < 54 mg/dL represents very low glucose levels, which may lead to seizures and loss of consciousness. If the person is conscious and able to swallow, rapidly absorbable carbohydrates should be administered. In cases of unconsciousness, glucagon (available in subcutaneous injectable or intra-nasal formulations) should be given. Although the combined results indicated that subjects using closed-loop therapy spent more time below the target glucose level than those receiving standard care, this difference was not statistically significant (SMD = -0.20, 95% CI = -0.46 to 0.07, I2 = 77%, P > 0.05) and exhibited considerable high heterogeneity (Fig. 4d). A possible reason for this is that the studies employed a stratified approach, where the TBR < 70 mg/dL showed statistical significance but with a slightly lower effect size. Consequently, formal testing was not conducted for TBR < 54 mg/dL.

[IMAGE OMITTED: SEE PDF]

Secondary outcomes: mean glucose level and HbA1c level

To comprehensively observe the overall profile of glucose management, we also extracted average glucose data. Mean glucose level refers to the average glucose measurements over a specific period (the past few days or weeks), obtained through frequent monitoring. While the mean glucose provides an estimate of overall glycemic exposure, it does not reflect glycemic variability, SD of glucose measurements is a more appropriate indicator of fluctuations in glucose levels [70]. Elevated mean glucose level is associated with increased risk of diabetes-related complications [71]. The combined results demonstrated significantly lower mean glucose levels in the closed-loop group compared to controls (SMD = -0.58, 95% CI = -0.79 to -0.37, I2 = 61%, P < 0.001), with moderate heterogeneity (Fig. 5a).

HbA1c is the form of hemoglobin bound to glucose, reflecting mean glucose levels over the preceding 2–3 months [72, 73]. It serves as an essential indicator for assessing overall glycemic management, with a target typically below 7% for most subjects with diabetes to mitigate complication risks. In the included studies, HbA1c levels were measured through standard laboratory assays. An elevation of 1% in HbA1c corresponds to an approximate 28–30 mg/dL increase in mean glucose levels. While the intervention duration in this meta-analysis (up to 6 months) demonstrates short-term glycemic improvement, sustained HbA1c reduction is required for long-term risk reduction. Thus, the closed-loop therapy group produced lower HbA1c level compared to the control group (SMD = -0.58, 95% CI = -0.73 to -0.43, I2 = 21%, P < 0.001), demonstrating improved HbA1c outcomes with AI-based closed-loop systems (Fig. 5b).

[IMAGE OMITTED: SEE PDF]

Additionally, we examined publication bias and conducted a sensitivity analysis. The funnel plot results showed a relatively symmetrical distribution of the included studies [74], indicating a high reliability of the findings (Fig. S3-S5). This symmetry suggests that our conclusions are less likely to be affected by publication bias, enhancing confidence in the effectiveness of closed-loop systems for diabetes management.

Safety outcomes: severe hypoglycemia and DKA

To assess treatment effects and safety of closed-loop systems versus standard care [75,76,77], we analyzed adverse events (severe hypoglycemia and DKA) across studies (Table S3-S4). In pooled analyses, severe hypoglycemia occurred in 17 cases (closed-loop) vs. 8 cases (control), while DKA was reported in 4 cases (closed-loop) vs. 0 cases (control) (Fig. 6a). Notably, 4 studies had unequal group sizes, which may influence incidence rates. The complete safety profile is presented in Fig. 6b, while AI-based closed-loop systems showed greater reduction in hyperglycemia duration and HbA1c levels, the incidence of adverse events was higher.

[IMAGE OMITTED: SEE PDF]

Discussion

In this meta-analysis, we compared the effectiveness of AI-based closed-loop systems with traditional therapies in diabetes management. The primary outcomes indicated that closed-loop therapy significantly improved glucose management, with participants maintaining glucose levels within the target range for longer durations. Secondary outcomes showed reductions in both TAR and TBR, strongly supporting the efficacy of this approach. Additionally, the closed-loop group demonstrated lower mean glucose levels and HbA1c levels compared to traditional therapy. However, our analysis revealed higher absolute counts of adverse events in closed-loop systems (severe hypoglycemia: 17 vs. 8, DKA: 4 vs. 0), though the overall rates remained low (3.2% vs. 1.6%). These observations may be influenced by: (1) larger closed-loop group sizes in four trials, (2) enhanced hypoglycemia detection sensitivity through continuous monitoring, (3) algorithm adaptation phases in AID. The DKA cases, while limited, suggest potential interactions between system parameters and subject behaviors (such as sensor failures) requiring further characterization.

To optimize the safety profile of closed-loop systems, we recommend three key implementation strategies: First, risk-stratified safety protocols could mitigate hypoglycemia risks during algorithm adaptation periods. Second, targeted subject education on DKA prevention (such as proper sensor use) may reduce system-related risks. Third, dynamic algorithm adjustments during initial treatment phases could balance glycemic management and safety, particularly for clinically diverse populations. These measures should be systematically evaluated in future trials.

Despite these positive findings, several limitations warrant consideration. Significant heterogeneity existed across studies in participant demographics, methodologies, and intervention protocols, potentially affecting result generalizability. One study failed to report adverse events in the control group [56], limiting comprehensive safety comparisons. The use of averaged glucose metrics in studies lacking HbA1c data may introduce interpretation bias. Furthermore, most studies showed high risk of performance bias due to unblinded designs, a methodological constraint inherent to comparative diabetes technology trials where blinding would require impractical 24/7 caregiver involvement and raise ethical concerns regarding safety monitoring. While our results align with previous evidence supporting closed-loop efficacy, variations in system architectures, algorithm models, and population characteristics likely contributed to outcome diversity.

Future research should focus on three priority areas: (1) development of advanced AI algorithms with improved glucose fluctuation prediction and personalized lifestyle recommendations, (2) carefully designed clinician alert systems that balance clinical benefits with healthcare provider workload considerations, (3) standardized reporting of safety outcomes across studies to facilitate robust meta-analyses. With continued technological refinements, AI-based closed-loop systems have significant potential to enhance both clinical outcomes and quality of life for people with diabetes.

Conclusion

In summary, this meta-analysis indicates that AI-driven closed-loop systems can effectively maintain glucose levels within target ranges while demonstrating acceptable safety. Compared to traditional therapies, closed-loop systems exhibit superior glycemic management outcomes in the studied populations, primarily among subjects with T1D under clinical trial conditions. However, the evidence is predominantly derived from high-income western countries, which may not adequately represent variations in system performance or accessibility on a global scale.

Future application strategies should be stratified based on subject characteristics (such as age, type of diabetes, and socioeconomic background) to ensure equitable implementation. It is noteworthy that the included studies employed heterogeneous AI methodologies (such as model predictive control and rule-based algorithms), suggesting that system design may influence outcomes, warranting further exploration in practical applications. Recent advancements, such as the FDA-approved Omnipod 5 and iAPS systems, reinforce the efficacy of closed-loop systems in trials but also underscore the need for more real-world evidence to align with current regulatory and clinical trends. Future implementations should prioritize system adaptability, individualized monitoring, and real-world safety data to optimize efficacy.

Data availability

No datasets were generated or analysed during the current study.

References

Worldwide trends in. Diabetes prevalence and treatment from 1990 to 2022: a pooled analysis of 1108 population-representative studies with 141 million participants. Lancet. 2024;404(10467):2077–93.

Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, et al. IDF diabetes atlas: global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract. 2022;183:109119.

Wang Y, Zhang P, Shao H, Andes LJ, Imperatore G. Medical costs associated with diabetes complications in medicare beneficiaries aged 65 years or older with type 2 diabetes. Diabetes Care. 2022;45(11):2570–6.

Cole JB, Florez JC. Genetics of diabetes mellitus and diabetes complications. Nat Rev Nephrol. 2020;16(7):377–90.

Li Y, Liu Y, Liu S, Gao M, Wang W, Chen K, et al. Diabetic vascular diseases: molecular mechanisms and therapeutic strategies. Signal Transduct Target Therapy. 2023;8(1):152.

ElSayed NA, Aleppo G, Aroda VR, Bannuru RR, Brown FM, Bruemmer D, et al. 5. Facilitating positive health behaviors and well-being to improve health outcomes: standards of care in diabetes—2023. Diabetes Care. 2023;46(Supplement1):S68–96.

ElSayed NA, Aleppo G, Aroda VR, Bannuru RR, Brown FM, Bruemmer D, et al. 1. Improving care and promoting health in populations: standards of care in diabetes—2023. Diabetes Care. 2023;46(Supplement1):S10–8.

Kingod N, Grabowski D. In a vigilant state of chronic disruption: how parents with a young child with type 1 diabetes negotiate events and moments of uncertainty. Sociol Health Illn. 2020;42(6):1473–87.

Yoon S, Ng JH, Kwan YH, Low LL. Healthcare professionals’ views of factors influencing diabetes self-management and the utility of a mHealth application and its features to support self-care. Front Endocrinol. 2022;13:793473.

Perkins BA, Sherr JL, Mathieu C. Type 1 diabetes glycemic management: insulin therapy, glucose monitoring, and automation. Science. 2021;373(6554):522–7.

Vargas E, Nandhakumar P, Ding S, Saha T, Wang J. Insulin detection in diabetes mellitus: challenges and new prospects. Nat Reviews Endocrinol. 2023;19(8):487–95.

Sperling MA, Laffel LM. Current management of glycemia in children with type 1 diabetes mellitus. N Engl J Med. 2022;386(12):1155–64.

Ma J, Fang Y, Yu H, Yi J, Ma Y, Lei P, et al. Recent advances in living algae seeding wound dressing: focusing on diabetic chronic wound healing. Adv Funct Mater. 2024;34(2):2308387.

Eghbali-Zarch M, Masoud S. Application of machine learning in affordable and accessible insulin management for type 1 and 2 diabetes: A comprehensive review. Artif Intell Med. 2024:102868.

Siontis KC, Noseworthy PA, Attia ZI, Friedman PA. Artificial intelligence-enhanced electrocardiography in cardiovascular disease management. Nat Reviews Cardiol. 2021;18(7):465–78.

Theodosiou AA, Read RC. Artificial intelligence, machine learning and deep learning: potential resources for the infection clinician. J Infect. 2023.

Heidari A, Navimipour NJ, Unal M. Applications of ML/DL in the management of smart cities and societies based on new trends in information technologies: A systematic literature review. Sustainable Cities Soc. 2022;85:104089.

Montaser E, Farhy LS, Rich SS. Enhancing type 1 diabetes immunological risk prediction with continuous glucose monitoring and genetic profiling. Diabetes Technol Ther. 2025;27(4):292–300.

Montaser E, Shah VN. Prediction of incident diabetic retinopathy in adults with type 1 diabetes using machine learning approach: an exploratory study. J Diabetes Sci Technol. 2024:19322968241292369.

Komorowski M, Celi LA, Badawi O, Gordon AC, Faisal AA. The artificial intelligence clinician learns optimal treatment strategies for sepsis in intensive care. Nat Med. 2018;24(11):1716–20.

Chi EA, Chi G, Tsui CT, Jiang Y, Jarr K, Kulkarni CV, et al. Development and validation of an artificial intelligence system to optimize clinician review of patient records. JAMA Netw Open. 2021;4(7):e2117391.

Lee YB, Kim G, Jun JE, Park H, Lee WJ, Hwang YC, et al. An integrated digital health care platform for diabetes management with AI-based dietary management: 48-week results from a randomized controlled trial. Diabetes Care. 2023;46(5):959–66.

Wang SC, Nickel G, Venkatesh KP, Raza MM, Kvedar JC. AI-based diabetes care: risk prediction models and implementation concerns. NPJ Digit Med. 2024;7(1):36.

Guan Z, Li H, Liu R, Cai C, Liu Y, Li J et al. Artificial intelligence in diabetes management: advancements, opportunities, and challenges. Cell Rep Med. 2023.

Wang X, Qiu J, Fong DT. The applications of wearable devices in the rehabilitation of ankle injuries: A systematic review and meta-analysis. Med Novel Technol Devices. 2023;17:100210.

Tyler NS, Mosquera-Lopez CM, Wilson LM, Dodier RH, Branigan DL, Gabo VB, et al. An artificial intelligence decision support system for the management of type 1 diabetes. Nat Metab. 2020;2(7):612–9.

Nimri R, Battelino T, Laffel LM, Slover RH, Schatz D, Weinzimer SA, et al. Insulin dose optimization using an automated artificial intelligence-based decision support system in youths with type 1 diabetes. Nat Med. 2020;26(9):1380–4.

Sng GGR, Tung JYM, Lim DYZ, Bee YM. Potential and pitfalls of ChatGPT and natural-language artificial intelligence models for diabetes education. Diabetes Care. 2023;46(5):e103–5.

Sherr JL, Heinemann L, Fleming GA, Bergenstal RM, Bruttomesso D, Hanaire H, et al. Automated insulin delivery: benefits, challenges, and recommendations. A consensus report of the joint diabetes technology working group of the European association for the study of diabetes and the American diabetes association. Diabetes Care. 2022;45(12):3058–74.

Huang HW, You SS, Di Tizio L, Li C, Raftery E, Ehmke C, et al. An automated all-in-one system for carbohydrate tracking, glucose monitoring, and insulin delivery. J Controlled Release. 2022;343:31–42.

Phillip M, Nimri R, Bergenstal RM, Barnard-Kelly K, Danne T, Hovorka R, et al. Consensus recommendations for the use of automated insulin delivery technologies in clinical practice. Endocr Rev. 2023;44(2):254–80.

Mackenzie SC, Sainsbury CA, Wake DJ. Diabetes and artificial intelligence beyond the closed loop: a review of the landscape, promise and challenges. Diabetologia. 2024;67(2):223–35.

Nimri R, Phillip M, Kovatchev B. Closed-Loop and artificial Intelligence–Based decision support systems. Diabetes Technol Ther. 2023;25(S1):S–70.

Shaw JL, Bannuru RR, Beach L, ElSayed NA, Freckmann G, Füzéry AK et al. Consensus considerations and good practice points for use of continuous glucose monitoring systems in hospital settings. Diabetes Care. 2024:dci240073.

Lee MY, Seav SM, Ongwela L, Lee JJ, Aubyrn R, Cao FY, et al. Empowering hospitalized patients with diabetes: implementation of a Hospital-wide CGM policy with EHR-Integrated validation for dosing insulin. Diabetes Care. 2024;47(10):1838–45.

Ware J, Allen JM, Boughton CK, Wilinska ME, Hartnell S, Thankamony A, et al. Randomized trial of closed-loop control in very young children with type 1 diabetes. N Engl J Med. 2022;386(3):209–19.

Boughton CK, Hovorka R. The role of automated insulin delivery technology in diabetes. Diabetologia. 2024:1–11.

Al-Sofiani ME, Alharthi S, Albunyan S, Alzaman N, Klonoff DC, Alguwaihes A. A Real-World prospective study of the effectiveness and safety of automated insulin delivery compared with other modalities of type 1 diabetes treatment during ramadan intermittent fasting. Diabetes Care. 2024;47(4):683–91.

Fuchs S, Caserto JS, Liu Q, Wang K, Shariati K, Hartquist CM et al. A Glucose-Responsive cannula for automated and Electronics‐Free insulin delivery. Adv Mater. 2024:2403594.

Advani A. Positioning time in range in diabetes management. Diabetologia. 2020;63(2):242–52.

Wei Y, Liu C, Liu Y, Zhang Z, Feng Z, Yang X, et al. The association between time in the glucose target range and abnormal ankle-brachial index: a cross-sectional analysis. Cardiovasc Diabetol. 2022;21(1):281.

Tauschmann M, Hermann JM, Freiberg C, Papsch M, Thon A, Heidtmann B, et al. Reduction in diabetic ketoacidosis and severe hypoglycemia in pediatric type 1 diabetes during the first year of continuous glucose monitoring: a multicenter analysis of 3,553 subjects from the DPV registry. Diabetes Care. 2020;43(3):e40–2.

Bogar K, Brensinger CM, Hennessy S, Flory JH, Bell ML, Shi C, et al. Climate change and ambient temperature extremes: association with serious hypoglycemia, diabetic ketoacidosis, and sudden cardiac arrest/ventricular arrhythmia in people with type 2 diabetes. Diabetes Care. 2022;45(11):e171–3.

Pasquel FJ, Tsegka K, Wang H, Cardona S, Galindo RJ, Fayfman M, et al. Clinical outcomes in patients with isolated or combined diabetic ketoacidosis and hyperosmolar hyperglycemic state: a retrospective, hospital-based cohort study. Diabetes Care. 2020;43(2):349–57.

Miyamura K, Nawa N, Nishimura H, Fushimi K, Fujiwara T. Association between heat exposure and hospitalization for diabetic ketoacidosis, hyperosmolar hyperglycemic state, and hypoglycemia in Japan. Environ Int. 2022;167:107410.

O’Dea RE, Lagisz M, Jennions MD, Koricheva J, Noble DW, Parker TH, et al. Preferred reporting items for systematic reviews and meta-analyses in ecology and evolutionary biology: a PRISMA extension. Biol Rev. 2021;96(5):1695–722.

Sanusi AA, Xue Y, McIlwraith C, Howard H, Brocato BE, Casey B, et al. Association of continuous glucose monitoring metrics with pregnancy outcomes in patients with preexisting diabetes. Diabetes Care. 2024;47(1):89–96.

Shah VN, Sakamoto C, Pyle L. Optimal sampling duration for continuous glucose monitoring for the Estimation of glycemia risk index. Diabetes Technol Ther. 2023;25(2):140–2.

Den Braber N, Vollenbroek-Hutten MM, Westerik KM, Bakker SJ, Navis G, van Beijnum B-JF, et al. Glucose regulation beyond HbA1c in type 2 diabetes treated with insulin: real-world evidence from the DIALECT-2 cohort. Diabetes Care. 2021;44(10):2238–44.

Bajaj HS, Ásbjörnsdóttir B, Carstensen L, Laugesen C, Mathieu C, Philis-Tsimikas A, et al. Continuous glucose Monitoring–Based metrics and hypoglycemia duration in insulin-Experienced individuals with Long-standing type 2 diabetes switched from a daily basal insulin to Once-Weekly insulin icodec: post hoc analysis of ONWARDS 2 and ONWARDS 4. Diabetes Care. 2024;47(4):729–38.

Luo D, Wan X, Liu J, Tong T. Optimally estimating the sample mean from the sample size, median, mid-range, and/or mid-quartile range. Stat Methods Med Res. 2018;27(6):1785–805.

Wan X, Wang W, Liu J, Tong T. Estimating the sample mean and standard deviation from the sample size, median, range and/or interquartile range. BMC Med Res Methodol. 2014;14:135.

McGrath S, Zhao X, Steele R, Thombs BD, Benedetti A. Estimating the sample mean and standard deviation from commonly reported quantiles in meta-analysis. Stat Methods Med Res. 2020;29(9):2520–37.

Yang B, Mallett S, Takwoingi Y, Davenport CF, Hyde CJ, Whiting PF, et al. QUADAS-C: a tool for assessing risk of bias in comparative diagnostic accuracy studies. Ann Intern Med. 2021;174(11):1592–9.

Ma T, Li Q, Song Y, Hua Y. Chronic ankle instability is associated with proprioception deficits: a systematic review and meta-analysis. J Sport Health Sci. 2021;10(2):182–91.

Brown SA, Forlenza GP, Bode BW, Pinsker JE, Levy CJ, Criego AB, et al. Multicenter trial of a tubeless, on-body automated insulin delivery system with customizable glycemic targets in pediatric and adult participants with type 1 diabetes. Diabetes Care. 2021;44(7):1630–40.

Wadwa RP, Reed ZW, Buckingham BA, DeBoer MD, Ekhlaspour L, Forlenza GP, et al. Trial of hybrid closed-loop control in young children with type 1 diabetes. N Engl J Med. 2023;388(11):991–1001.

Group BPR, Messer LH, Buckingham BA, Cogen F, Daniels M, Forlenza G, et al. Positive impact of the bionic pancreas on diabetes control in youth 6–17 years old with type 1 diabetes: a multicenter randomized trial. Diabetes Technol Ther. 2022;24(10):712–25.

Ware J, Boughton CK, Allen JM, Wilinska ME, Tauschmann M, Denvir L, et al. Cambridge hybrid closed-loop algorithm in children and adolescents with type 1 diabetes: a multicentre 6-month randomised controlled trial. Lancet Digit Health. 2022;4(4):e245–55.

Thabit H, Hartnell S, Allen JM, Lake A, Wilinska ME, Ruan Y, et al. Closed-loop insulin delivery in inpatients with type 2 diabetes: a randomised, parallel-group trial. Lancet Diabetes Endocrinol. 2017;5(2):117–24.

Breton MD, Kanapka LG, Beck RW, Ekhlaspour L, Forlenza GP, Cengiz E, et al. A randomized trial of closed-loop control in children with type 1 diabetes. N Engl J Med. 2020;383(9):836–45.

Deshpande S, Pinsker JE, Church MM, Piper M, Andre C, Massa J, et al. Randomized crossover comparison of automated insulin delivery versus conventional therapy using an unlocked smartphone with scheduled pasta and rice meal challenges in the outpatient setting. Diabetes Technol Ther. 2020;22(12):865–74.

Benhamou PY, Franc S, Reznik Y, Thivolet C, Schaepelynck P, Renard E, et al. Closed-loop insulin delivery in adults with type 1 diabetes in real-life conditions: a 12-week multicentre, open-label randomised controlled crossover trial. Lancet Digit Health. 2019;1(1):e17–25.

Laha S, Rajput A, Laha SS, Jadhav R. A concise and systematic review on non-invasive glucose monitoring for potential diabetes management. Biosensors. 2022;12(11):965.

Gregory NS. Excessive thirst hunger, and urination in diabetes. Introduction to Clinical Pharmacology: From Symptoms to Treatment. 2023:304.

Kim MY, Kim G, Park JY, Choi MS, Jun JE, Lee YB, et al. The association between continuous glucose Monitoring-Derived metrics and cardiovascular autonomic neuropathy in outpatients with type 2 diabetes. Diabetes Technol Ther. 2021;23(6):434–42.

Shah VN, Kanapka LG, Akturk HK, Polsky S, Forlenza GP, Kollman C, et al. Time in range is associated with incident diabetic retinopathy in adults with type 1 diabetes: A longitudinal study. Diabetes Technol Ther. 2024;26(4):246–51.

6. Glycemic targets: standards of medical care in Diabetes-2020. Diabetes Care. 2020;43(Suppl 1):S66–76.

Roy A, Chatterjee S. Epileptic Seizures. MIMICS of Epileptic Seizures. 2020:180.

Belsare P, Bartolome A, Stanger C, Prioleau T. Understanding Temporal changes and seasonal variations in glycemic trends using wearable data. Sci Adv. 2023;9(38):eadg2132.

Leelarathna L, Evans ML, Neupane S, Rayman G, Lumley S, Cranston I, et al. Intermittently scanned continuous glucose monitoring for type 1 diabetes. N Engl J Med. 2022;387(16):1477–87.

Klein KR, Buse JB. The trials and tribulations of determining HbA1c targets for diabetes mellitus. Nat Reviews Endocrinol. 2020;16(12):717–30.

Fabris C, Heinemann L, Beck R, Cobelli C, Kovatchev B. Estimation of hemoglobin A1c from continuous glucose monitoring data in individuals with type 1 diabetes: is time in range all we need? Diabetes Technol Ther. 2020;22(7):501–8.

Afonso J, Ramirez-Campillo R, Clemente FM, Büttner FC, Andrade R. The perils of misinterpreting and misusing publication Bias in meta-analyses: an education review on funnel plot-based methods. Sports Med. 2024;54(2):257–69.

Pei Y, Ke W, Lu J, Lin Y, Zhang Z, Peng Y, et al. Safety event outcomes and glycemic control with a hybrid closed-loop system used by Chinese adolescents and adults with type 1 diabetes mellitus. Diabetes Technol Ther. 2023;25(10):718–25.

Carlson AL, Sherr JL, Shulman DI, Garg SK, Pop-Busui R, Bode BW, et al. Safety and glycemic outcomes during the minimed™ advanced hybrid closed-loop system pivotal trial in adolescents and adults with type 1 diabetes. Diabetes Technol Ther. 2022;24(3):178–89.

Garg SK, Grunberger G, Weinstock R, Lawson ML, Hirsch IB, DiMeglio LA, et al. Improved glycemia with hybrid closed-loop versus continuous subcutaneous insulin infusion therapy: results from a randomized controlled trial. Diabetes Technol Ther. 2023;25(1):1–12.