Unlabelled:Nutrient estimation from meal images by multimodal large language models can support diabetes self-management, but its robustness is limited by variability driven by both sensitivity to visual presentation and inherent model instability. Accounting for and mitigating this variability is essential for robust AI-assisted dietary assessment.
Background:Diabetic retinopathy is the leading cause of blindness in US adults. Fewer than half of patients with diabetes in rural areas complete the recommended screening that could prevent blindness. Store-and-forward teleophthalmology and mobile pop-up eye clinics have each been shown to improve screening access, but studies examining the implementation of both modalities within the same rural primary care system are lacking. Objective:This study aimed to identify multilevel facilitators and barriers to implementing teleophthalmology and pop-up eye clinics for diabetic eye care in 2 rural primary care clinics in upstate New York. Methods:We conducted a mixed methods implementation study with prospective semistructured interviews and retrospective, descriptive electronic health record data at 2 university-affiliated primary care clinics classified as "isolated rural," guided by the Consolidated Framework for Implementation Research (CFIR). Electronic health record data from 885 adults with diabetes were used to characterize patient demographics and eye care use. Patients were categorized into 3 groups: eye exam with an eye doctor (n=586), on-site primary care-based eye exam (n=91; teleophthalmology n=39 and pop-up clinic n=52), or no recent eye exam (n=208). Semistructured interviews were conducted with 20 patients (n=5 per subgroup) and 14 staff members (primary care clinicians, optometrists, and administrative personnel), purposively sampled to capture diverse perspectives across eye care pathways and roles. Interviews were coded using CFIR domains and analyzed thematically until no new themes emerged. Results:Patients in the on-site primary care eye exam group were the most socially disadvantaged, with the lowest proportions reporting no transportation needs (42/91, 46.2%), no housing instability (28/91, 30.8%), and no food insecurity (36/91, 39.6%). Over 18 months, 58.4% (52/89) of scheduled pop-up clinic appointments were completed, and 92.3% (36/39) of teleophthalmology images were gradable. Pop-up clinics detected higher rates of diabetic retinopathy (11/52, 21.2% vs 3/39, 7.7%), cataract (29/52, 55.8% vs 3/39, 7.7%), and reduced visual acuity (23/52, 44.2% vs 11/39, 28.2%) than teleophthalmology. Qualitative analysis revealed that patients viewed both modalities as a convenient "one-stop shop" and valued trust in primary care staff. Staff identified complementary barriers: training and workflow complexity for teleophthalmology, and underutilization and scheduling challenges for pop-up clinics. Implementation champions across roles proposed an integrated workflow in which same-day teleophthalmology becomes a default component of annual diabetes visits, with targeted referral to pop-up clinics for patients with abnormal findings. Conclusions:Primary care-based eye programs in rural settings preferentially reached socially vulnerable patients with diabetes and detected substantial unmet eye pathology. However, operating teleophthalmology and pop-up clinics as separate programs led to inefficiencies in both programs. A stakeholder-driven integrated model embedding teleophthalmology into routine diabetes care, with targeted pop-up clinic referrals, may improve the reach, efficiency, and sustainability of diabetic eye screening in rural populations.
Background:Continuous glucose monitors (CGMs), sensor-augmented pumps (SAPs), and automated insulin delivery (AID) systems have substantially improved glycemic outcomes for people with type 1 diabetes (T1D). However, these technologies also generate frequent alarms and alerts that may contribute to emotional burden, alarm fatigue, and maladaptive behavioral responses. Despite increasing recognition of alarm-related distress, little is known about how alarm burden differs across contemporary diabetes technologies. Objective:This study evaluated differences in alarm frequency, emotional burden, and behavioral responses to alarms and alerts among adults using CGMs alone, SAP therapy, and AID systems. Methods:We conducted a cross-sectional survey of adults (≥18 years) with T1D receiving care within a large academic health system between August 2024 and March 2025. Eligible participants completed an online survey assessing demographics, diabetes technology use, perceptions of alarm frequency and burden, and responses to alarms and alerts. Participants were categorized as CGM-only, SAP, or AID users. Differences between groups were evaluated using χ2 tests and ANOVA. Ordinal logistic regression models adjusted for age, gender, and diabetes duration were used to examine associations between diabetes technology type and alarm-related outcomes. Results:Among 838 respondents (mean age 46.0, SD 16.6 years; mean diabetes duration 23.0, SD 14.0 years; n=457, 55% women), AID users reported the highest alarm frequency, with 49% (n=252) experiencing alarms several times daily, compared with 35% (n=28) of SAP users and 32% (n=79) of CGM-only users (P<.001). AID users also reported greater alarm disruptiveness, annoyance, and unwanted attention than CGM-only users. Overcorrection of glucose levels in response to alerts was common across all technologies: 44% (n=371) and 50% (n=419) of participants reported sometimes overcorrecting low and high glucose levels, respectively, while 32% (n=265) and 18% (n=147) reported often or always overcorrecting low and high glucose levels, respectively. After adjustment, AID users had greater odds of frequent alarms (odds ratio [OR] 1.98, 95% CI 1.48-2.65), alarm disruptiveness (OR 1.67, 95% CI 1.24-2.25), annoyance (OR 1.65, 95% CI 1.23-2.19), unwanted attention (OR 1.85, 95% CI 1.38-2.47), ignoring alarms (OR 1.86, 95% CI 1.40-2.47), and overcorrecting low glucose levels (OR 1.61, 95% CI 1.20-2.15) compared with CGM-only users. Perceived effectiveness of alarms did not differ by technology type. Conclusions:Although AID systems provide important clinical benefits, they are associated with the greatest alarm and alert burden. Alarm-driven behaviors, including ignoring alerts and overcorrecting glucose levels, represent previously underrecognized consequences of diabetes technology that may diminish user experience and potentially affect glycemic management. These findings identify opportunities to improve alert algorithms while highlighting the need for individualized patient education, clinician engagement in setting appropriate alert thresholds, and routine assessment of alarm burden to optimize both glycemic outcomes and patient-centered experiences with diabetes technology.
Background:The American Diabetes Association and the Association of Diabetes Care & Education Specialists recommend person-first, strengths-based, and stigma-free language in diabetes care and education. However, interventions to promote stigma-free language among primary care clinicians remain limited. Objective:This mixed methods pilot study aimed to evaluate the acceptability and preliminary effects of a diabetes stigma-reduction training module for primary care clinicians, designed to increase awareness of language used in clinician-patient interactions and explore changes in attitudes and intentions toward avoiding stigmatizing language. Methods:Twenty-one primary care clinicians completed either a stigma-reduction training module (n=12) or an active control condition (n=9). Both conditions included a standardized patient encounter, an educational video, and a guided self-reflection activity. The stigma-reduction training module included diabetes-specific content on stigmatizing language and person-first, strengths-based alternatives. Theory of Planned Behavior domain measures were administered before and immediately after the assigned condition. Attitudes and intentions were assessed precondition and postcondition. Acceptability was evaluated using a postcondition survey and semistructured qualitative interviews. Repeated-measures ANOVA was used to assess between-group changes over time, and a qualitative descriptive approach was used to explore clinicians' perceptions. Results:Clinicians who received the stigma-reduction training module demonstrated significantly greater improvement in attitudes toward avoiding stigmatizing language compared with clinicians in the active control condition, with mean attitude scores increasing by 1.19 (SD 0.68) points versus 0.21 (SD 0.44) points, respectively. A significant condition × time interaction was observed (F1,19=14.27; P=.001, partial η2=0.43, Cohen d=1.67). The between-group difference in intention change was not statistically significant, with mean intention scores increasing by 0.69 (SD 1.08) points in the stigma-reduction training module condition versus 0.02 (SD 0.90) points in the active control condition (F1,19=2.33; P=.14, partial η2=0.11, Cohen d=0.67). Acceptability ratings were higher for the stigma-reduction training module than for the active control condition for clarity (mean 6.17, SD 0.72 vs mean 4.44, SD 1.74; P=.02), helpfulness (mean 6.08, SD 1.00 vs mean 4.56, SD 1.59; P=.03), and likelihood of recommendation (mean 6.25, SD 0.87 vs mean 4.56, SD 1.67; P=.02). The qualitative findings suggested that clinicians perceived the stigma-reduction training module as actionable, relevant to primary care, and useful for promoting self-reflection about habitual language use. The active control condition was generally perceived as a useful refresher but less novel and impactful. Conclusions:This pilot study suggests that a brief, theory-informed diabetes stigma-reduction training module is acceptable and associated with improved clinician attitudes toward stigmatizing language in a simulated clinical environment. Changes in intentions were descriptively in the expected direction but did not reach statistical significance. The findings should be interpreted cautiously, given the small sample size and lack of behavioral outcome assessment. Future studies should evaluate the module in larger, more diverse samples and assess whether changes in attitudes translate into sustained changes in clinician language use and patient experiences.
Background:Despite advancements in digital health coaching (DHC) for type 2 diabetes self-management, a persistent digital divide limits access among underserved populations, necessitating low-tech, telephone-based alternatives. Objective:This formative qualitative study explored the lived experiences of type 2 diabetes mellitus (T2DM) self-management challenges among adults in Alabama and elicited preferences for telephone-based DHC to inform intervention optimization. Methods:This study was part of a larger National Institutes of Health clinical trial (5R01DK129378) that examined the feasibility and needs of a telephone-based DHC intervention for self-management of T2DM. The study employed a qualitative, phenomenological approach to assess patients' needs surrounding digital coaching. Twelve patients diagnosed with T2DM were recruited between August and December 2022. Data collection involved in-depth, semistructured interviews conducted via telephone calls or secure Zoom sessions. All interviews were audio-recorded, professionally transcribed, and verified for accuracy. Two independent coders analyzed the data using NVivo software version 14 Plus. Results:Two themes emerged: (1) multifaceted barriers to diabetes self-management spanning behavioral (dietary adherence and nutritional gaps), physical/environmental (mobility limitations and weather constraints), and structural domains (medication shortages, insurance coverage limiting continuous glucose monitoring); (2) expectations for telephone-based DHC, including targeted education (meal planning, exercise adaptations, and medication effects), health coaches as accountability partners, daily reminders and check-ins via texts/calls, and incentives as extrinsic motivators to adherence. Insights from the study directly shaped the intervention components of a pilot feasibility trial (NCT05344859). Conclusions:Key challenges in day-to-day T2DM management included diet, physical activity, medication unavailability, and lack of insurance coverage for continuous glucose monitors. Expectations for a potential telephone-based DHC intervention included targeted education on diet, exercise, and medication effects; perceptions of health coaches as accountability partners; reminders via texts/calls as beneficial; and monetary incentives as extrinsic motivators for adherence to the intervention.
Background:Serious digital games have been proposed as a novel approach to support diabetes education and self-management, but evidence regarding their effectiveness remains limited. Objective:This study aimed to evaluate the effects of SugarVita, a serious game for people with type 2 diabetes, on diabetes-related knowledge, self-confidence, and self-management. Secondary outcomes included hemoglobin A1c (HbA1c), engagement, and user evaluation. Methods:In this pilot randomized controlled trial, 30 adults with type 2 diabetes were randomized to SugarVita plus standard care or standard care alone for 8 weeks. Outcomes were assessed before and after the intervention using validated questionnaires and laboratory HbA1c values. Within-group changes were analyzed using Wilcoxon signed-rank tests and between-group differences using Mann-Whitney U tests. Bonferroni correction was applied for multiple primary outcomes. Results:No statistically significant between-group differences were observed for diabetes-related knowledge, self-confidence, or self-management after correction for multiple testing. Both groups showed numerical improvements over time. HbA1c decreased significantly within the intervention group (median 73.0, IQR 70.8-81.5 to median 64.5, IQR 60.8-72.0 mmol/mol; P=.007), whereas no significant change was observed in the control group. Greater total playtime was moderately associated with HbA1c reduction. User evaluations indicated high perceived educational value. Conclusions:Participants reported positive experiences with SugarVita and perceived the game as educational and user-friendly. No statistically significant between-group differences were observed for the primary outcomes. These findings support the feasibility and acceptability of SugarVita as a digital educational intervention and warrant further evaluation in larger studies.
Abstract Background The clinical use of antidiabetic drugs (ADDs) has gained prominent public visibility due to the Food and Drug Administration (FDA) expansion of glucagon-like peptide-1 receptor agonists (GLP-1 RAs) in recent years; these drugs are now widely prescribed for weight loss. This shift is reflected in online public discourse. Objective This study aimed to analyze the current online public discourse surrounding ADDs, with an emphasis on perceptions of insulin versus noninsulin therapies and the increasing prominence of weight loss–associated medications. Methods We conducted a retrospective analysis of 23,580 English-language posts from the United States in 2023, using Boolean keyword searches to examine conversations about insulin-inclusive, with or without weight loss terms, and insulin-exclusive, with or without weight loss terms. We analyzed the volume and sentiment of online conversations regarding ADDs, frequency of drug mentions, proportion of posts by self-identified physicians, and thematic analysis of patient and clinician concerns. Results A total of 26,832 initial posts were collected in 2023. After post validation, 23,580 posts remained as the study sample size. Of the 23,580 posts, 4.6% (1087/23,580) mentioned insulin, and 95.4% (n=22,493) did not mention insulin. Semaglutide-containing drugs such as Ozempic and Mounjaro were the most referenced medications, particularly in weight-loss contexts. Weight loss conversations made up the majority of online posts. Conversations about insulin were marginal compared with conversations that did not mention insulin online. Only 7% to 10% of posts came from self-identified physicians. Key themes included drug accessibility, off-label use for weight management, concerns about supply shortages, insurance coverage, and growing calls for holistic care. Notably, public perspectives favored the dual efficacy of medications like Ozempic in managing diabetes and promoting weight loss. Conclusions The discourse on ADDs is dominated by weight loss–oriented therapies, with GLP-1 RAs driving much of the engagement. This surveillance paper illustrates the current state of public health interests around antidiabetic medications associated with weight loss and raises concerns regarding equitable access for patients with diabetes. These findings underscore the need for updated clinical guidance on necessary lifestyle behaviors for antidiabetic medication use and ongoing monitoring of public opinions regarding chronic disease management medications.
Background:Continuous glucose monitoring (CGM) has transformed diabetes management and research by providing high-frequency data that address many of the limitations of hemoglobin A1c, enabling more precise clinical treatment targets and responsive trial endpoints. The richness and complexity of high-resolution time-series CGM data have spurred the development of numerous metrics for both clinical care and research applications. Beyond established metrics, there is a growing set of clinical, composite, and research-oriented measures that may support clinical decision support, intervention planning, risk stratification, and discovery-oriented research. This proliferation has created significant challenges in metric selection, interpretation, calculation, and standardization, particularly when metrics are applied across different devices, populations, software packages, and study designs. Objective:The objective is to map the current landscape of CGM metrics and address ongoing challenges in metric selection, clinical and research interpretation, and standardization. We further sought to distinguish between metrics primarily suited for routine clinical interpretation and those designed to explore more granular or multidimensional features of glycemia in research settings. Methods:We identified the literature focusing on the calculation, application, and interpretation of the following categories of CGM metrics: (1) standardized, (2) clinical, (3) emerging, and (4) composite. CGM metrics included in this study were identified from the 27 metrics included in the Diabetes Research Hub platform, additional published standardized and composite metrics, metrics used in established CGM analysis software, and emerging metrics identified during review. We narratively reviewed each metric's definitions, calculation methods, interpretation, clinical and research utility, and strengths and limitations. In total, 102 articles were reviewed, supporting the synthesis of 36 distinct CGM-derived metrics. Results:The review identifies a fundamental divide in the CGM metric landscape. Standardized and clinical metrics, including time in range, mean glucose, coefficient of variation, and similar, prioritize simplicity and actionability. These metrics facilitate rapid decision-making in clinical settings but potentially mask granular glycemic fluctuations, event patterns, and discordance between average glucose values and variability. Emerging and composite metrics offer deeper insights into glycemic patterns, risk, and variability. However, many rely on specialized software or complex formulas, lack standardized thresholds or clear relationships to clinical outcomes, and do not have consensus methods of calculation and interpretation, limiting their adoption and hindering cross-study comparison. Conclusions:While consensus exists for core clinical metrics, the lack of standardization for complex metrics hinders research replicability and clinical translation. Bridging this gap requires moving toward consensus metric definitions, open-science frameworks, and standardized code libraries. Metric selection should be guided by intended use. Clinical metrics should be well-established, interpretable, and actionable. Research metrics should be clearly described, reproducible, and linked to meaningful outcomes. This review provides a comprehensive resource for navigating the diverse spectrum of CGM metrics, clarifying their applications and limitations to support both research and clinical investigation.
Background:Type 2 diabetes mellitus (T2DM) affects approximately 590 million people worldwide, and its management relies heavily on patient education. With the emergence of online health information and artificial intelligence (AI) large language models, patients are increasingly sourcing medical information independently. Objective:This study compared the quality, readability, and transparency of websites and AI-generated leaflets (AIGLs) related to T2DM. Methods:Four predefined search terms ("type 2 diabetes," "type 2 diabetes mellitus," "T2DM," and "adult diabetes") were entered into 3 major search engines (Google, Yahoo, and Bing), and the top 20 search results were retrieved. AIGLs with patient information about T2DM were produced using a standardized prompt in 4 AI large language models (ChatGPT, Gemini, DeepSeek, and Grok). Information quality was assessed using the DISCERN score, calculated by 3 independent raters and ChatGPT. The Journal of the American Medical Association (JAMA) benchmarks were used to measure reliability and transparency. The Flesch-Kincaid Grade Level was used to determine readability. Results:Seventy-five websites and 4 AIGLs were evaluated. Mean author-rated DISCERN scores were 42.6 (SD 11.3) for websites and 43.9 (SD 1.74) for AIGLs, corresponding to fair quality (DISCERN 41-51). In contrast, ChatGPT-rated mean DISCERN scores were higher, with 58.5 (SD 11.5) for websites and 61.0 (SD 2.94) for AIGLs, corresponding to good quality (DISCERN 52-63). Mean JAMA benchmark scores were 2.74 (SD 0.965) for websites, indicating moderate reliability (2-3 out of 4 points), whereas all AIGLs scored 0 out of 4 points. Mean Flesch-Kincaid Grade Level scores for websites were 8.67 (SD 2.23) and 8.30 (SD 1.92) for AIGLs, corresponding to an eighth- to ninth-grade comprehension level. Spearman rank correlation demonstrated minimal variability among the 3 independent raters but showed a significant difference between ChatGPT and the 3 independent raters. Conclusions:Given the high prevalence of T2DM, both websites and AIGLs demonstrated suboptimal quality, readability, and transparency. Increasing patient reliance on digital health information calls for improved readability standards and stronger safeguards for AI-generated content. Both websites and AIGLs require an eighth- to ninth-grade comprehension level, far above the average reading age of 9 years in the United Kingdom (fourth- to fifth-grade level). This reduces the accessibility of online health information. The landscape of medical consultations is evolving, with patients increasingly presenting with preconceived notions based on online health information; hence, health care professionals should adapt to this shift.
Background:Wearable technologies, including smart insoles and sensor-equipped footwear, enable continuous monitoring of key foot parameters such as plantar pressure and temperature in individuals at risk of diabetic foot ulcers (DFUs). Objective:This systematic review aimed to evaluate the technological characteristics and clinical applications of wearable devices for monitoring DFU-related parameters. Methods:This review was conducted in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Journal articles, theses, and dissertations evaluating wearable technologies for DFU prevention or monitoring were eligible if they involved human participants and were published in English or Persian. Studies focused on nonwearable or non-foot-based systems were excluded. A comprehensive search was conducted in PubMed, Embase, Web of Science, and Scopus from July 2024 to May 2025. Two reviewers (HE and SS) independently screened studies and extracted data. The methodological quality of included studies was assessed using the Mixed Methods Appraisal Tool (MMAT) 2018. Results were synthesized using descriptive synthesis. Results:A total of 1088 records were identified, of which 23 studies met the inclusion criteria. The included studies varied in design, sample size, and follow-up duration. Wearable devices included smart insoles, socks, and external sensors, primarily monitoring plantar pressure and temperature. Devices differed in sensor type, placement, number, communication protocols, and data acquisition rates. Participants typically had diabetes, and many had a history of neuropathy or prior DFUs. Conclusions:Wearable technologies show promise for monitoring DFU risk factors and supporting early detection. However, the evidence base is limited by heterogeneity in study designs, small sample sizes, and short follow-up periods. Further high-quality studies are required to evaluate their potential clinical benefits, long-term outcomes, and role in preventing DFUs and improving patient care.
Background:Type 2 diabetes mellitus, a public health challenge, disproportionately impacts low- and middle-income countries (LMICs), accounting for 73% of global cases. Due to resource constraints, these nations have adopted task-shifting strategies using community health workers (CHWs). However, evidence on the effectiveness of training CHWs in diabetes management is limited and, at most, indirect due to the limited studies, variable training methods, and complex interventions that make it difficult to isolate training effects. Objective:A systematic review was conducted to answer the question: Does training CHWs in type 2 diabetes improve the efficacy of diabetes screening and management at the community level in LMICs? Methods:A total of 2 reviewers, supervised by 2 supervisors, conducted the review following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines. They searched databases, including PubMed, MEDLINE Ovid, Scopus, EBMR, and CINAHL, for studies published between January 2000 and April 2024, including randomized and nonrandomized controlled trials and observational studies assessing CHW training in diabetes management in LMICs. The primary outcome was the mean change in glycated hemoglobin (HbA1c) percentage levels. Data were narratively synthesized for training characteristics and study outcomes, and quality was assessed using the Risk of Bias 2 and ROBINS-I tools. Results:A total of 3387 studies were screened; 69 were eligible for full-text review, and 4 studies (3 randomized controlled trials [RCTs] and 1 observational stepped-wedge study, ~1000 patients) were included for narrative analysis. One of the 3 RCTs reported a statistically significant mean HbA1c reduction of -0.24% (P=.001), but HbA1c was not the primary outcome, and most patients were normoglycemic, prediabetic, or had diabetes. Other studies reported nonsignificant HbA1c reductions. The risk of bias among RCTs was moderate (some concerns, 1 trial at high risk), and the observational study had a serious risk of bias. No meta-analysis was performed due to the limited number of RCTs. Conclusions:Training CHWs in type 2 diabetes management has shown limited and, at most, indirect effects in improving glycemic control in LMIC settings. These findings are constrained by the small number of eligible studies, heterogeneity in training methodologies, and the multicomponent nature of the included interventions, with 1 trial demonstrating a statistically significant yet small reduction in HbA1c (-0.24%). Our review included only 4 eligible studies with a small representation of CHWs and multicomponent interventions. Considering the limited number of eligible studies, the heterogeneity in training methodologies and study designs, and the multicomponent nature of the included interventions, the existing evidence remains inadequate to definitively conclude whether CHW training significantly improves diabetes management across LMICs. Therefore, strengthening and standardizing CHW training might be an effective strategy to enhance diabetes care in underserved settings. Future larger trials and implementation research can help maximize the impact of CHWs against the growing diabetes burden.
Background: Optimizing insulin dosing and predicting future glucose levels for people with type 1 diabetes is challenging due to the dynamic nature of glucose metabolism. Traditional static insulin regimens fail to adapt to individual variability in diet, physical activity, stress, and metabolic fluctuations, leading to suboptimal glycemic control. Reinforcement learning (RL) offers a promising alternative by enabling personalized, real-time insulin adjustments that improve the balance between hyperglycemia and hypoglycemia. Objective: This study aims to develop a deep Q-network (DQN)-based RL system that dynamically personalizes insulin dosing recommendations using continuous glucose monitoring data, meal intake, and physical activity levels. By leveraging real-time data, the model adapts to patients' evolving physiological states, enhancing glucose control and patient safety. Methods: We used the OhioT1DM dataset (2018 and 2020), which includes 8 weeks of continuous glucose measurements, insulin dosing records, and physical activity data for twelve people with type 1 diabetes. The RL agent was designed with a state representation consisting of recent blood glucose levels, insulin doses, and lifestyle factors over a 2-hour window. The 2-hour window was selected based on the known pharmacodynamic profile of rapid-acting insulin (peak action within 90-120 min), as well as the typical lag in glycemic response following meals or exercise. This window size captures both recent and delayed physiological effects while balancing data density and model stability. The action space included discrete insulin dose recommendations (eg, 0.5 U, 1 U, and 1.5 U). A reward function incentivized glucose levels within the target range (70-180 mg/dL) while penalizing extreme deviations. The DQN model was trained to maximize reward by learning optimal dosing strategies through iterative trial and error. Results: Performance evaluation was conducted using both qualitative and quantitative metrics. Time-series analysis compared actual and predicted glucose levels, demonstrating effective glucose regulation. The RL model achieved a mean glucose level of 80.06 mg/dL, with a reward score of 10 during evaluation, indicating that most glucose predictions were maintained within the desired clinical range. This suggests the model has learned to regulate blood glucose effectively through adaptive insulin dosing. The root mean square error (12.39 mg/dL) was slightly higher than the mean absolute error (9.85 mg/dL), indicating stable predictions. Additionally, the percentage time in target range was 64.06%, suggesting that the model-maintained glucose within the clinically safe range for a majority of the time. Conclusions: The DQN-based RL model demonstrated its effectiveness in personalized insulin dosing while minimizing the risk of hypo-and hyperglycemia. This suggests the model has learned to regulate blood glucose effectively through adaptive insulin dosing. This approach represents a significant advancement over conventional methods, offering a scalable and adaptive strategy for real-world diabetes management, along with enhancing clinical trust and transparency through explainability techniques.
Background: Prediabetes is common in the United States, and adverse social determinants of health (SDOH) are known to undermine diabetes prevention efforts. Chinese Americans experience a disproportionately high prevalence of prediabetes, yet the SDOH profiles of this population remain understudied. Objective: This study assessed SDOH among Chinese Americans at risk for diabetes and examined the associations between sociodemographic characteristics and SDOH barriers. Methods: We conducted a cross-sectional analysis of baseline survey data from the Integrating Cultural Aspects into Diabetes Education (INCLUDE) study, a randomized controlled trial of a culturally and linguistically tailored mobile diabetes prevention intervention for Chinese Americans. Participants at risk for diabetes were enrolled between April 2023 and June 2024 in New York City (N=150). Measures included in the analyses were a 14-item SDOH scale (range 0-14, with higher scores indicating more barriers) and sociodemographic characteristics. Due to the small frequencies of high SDOH scores, we collapsed the outcome into 5 categories (0, 1, 2, 3, and 4-14) to improve model stability. We first used univariable logistic regression models to examine associations between each sociodemographic factor (age, sex, years of US residence, English proficiency, education, marital status, employment status, and annual household income) and the collapsed SDOH category, followed by a multivariable ordinal regression model including all sociodemographic variables. Results: A total of 150 participants had a mean age of 49.9 (SD 12.6) years. Most were female (n=124, 82.7%), born outside the United States (n=149, 99.3%), and reported speaking English less than very well (n=132, 88.0%). Among respondents to the SDOH items (n=149), the mean SDOH score was 2.4 (SD 2.3), and 81.9% (n=122) reported at least 1 SDOH barrier. The three most frequently reported barriers were (1) the need to improve English proficiency, reading skills, or educational attainment (n=77, 51.7%); (2) experiences of racial discrimination (n=49, 32.9%); and (3) adverse housing conditions (n=38, 25.5%). After collapsing the original SDOH score, 27 (18.2%) participants had a score of 0, 39 (26.2%) had a score of 1, 23 (15.4%) had a score of 2, 27 (18.2%) had a score of 3, and 33 (22.1%) had scores of 4 to 14. In the multivariable analysis, female sex (vs male) was associated with higher SDOH score categories (odds ratio 3.83, 95% CI 1.65-9.16; P=.002). Conclusions: SDOH-related barriers were prevalent among Chinese Americans at risk for diabetes. Diabetes prevention efforts should incorporate routine SDOH screening and structured resource navigation or referral pathways, with particular attention to subgroups at higher risk, such as female individuals. Trial Registration: ClinicalTrials.gov NCT05492916; https://clinicaltrials.gov/study/NCT05492916 International Registered Report Identifier (IRRID): RR2-10.2196/65455
Background: The prevalence of diabetes in the United States necessitates investigations into how to better enable adults with type 2 diabetes mellitus (T2DM) to manage their health using easy-to-access and personally adaptable technologies. The ubiquity of digital content further justifies the need to consider the impact of different digital intervention modalities in diabetes self-care activities. Objective: This study aimed to compare the impact of 2 digital diabetes self-care education programs delivered separately and in combination to adults with T2DM across various settings in Texas. Methods: We conducted a randomized controlled trial in Texas with 188 adults with T2DM to assess whether 2 different interventions alone (Virtual Making Moves with Diabetes or Technology-Based Education and Support) or in combination (Virtual Making Moves with Diabetes followed by Technology-Based Education and Support) improved multiple outcomes associated with diabetes self-management. We used several estimation techniques, including generalized estimating equations, to account for multiple factors simultaneously. Results: All 3 digital intervention modalities led to statistically significant improvements in diabetes-related confidence, distress, and self-care behaviors, with significance from baseline through 6 months and supported by moderate to strong effect sizes (Cohen d) ranging from 0.446 to 0.827 at 3-month follow-up versus baseline and from 0.538 to 0.888 at 6-month follow-up versus baseline. No statistically significant superiority was observed among the intervention modalities. Higher self-care behaviors were significantly associated with higher baseline confidence and lower distress. Those in the most disadvantaged positions (less education, less financial stability, and no health insurance) showed significantly larger improvement in self-care behaviors. Conclusions: Given the benefits associated with this study's interventions, we suggest future work to further develop digital content that can be tailored to individuals with T2DM to help them manage their chronic conditions in a cost-effective manner.
Abstract BackgroundGestational diabetes mellitus (GDM) requires effective self-management to mitigate associated health risks. A comprehensive understanding of the multilevel factors influencing adherence is essential to designing effective, holistic support strategies. ObjectiveThis systematic review aimed to synthesize the obstacles and enablers related to GDM self-management across all 5 levels of the socioecological model (SEM). MethodsA systematic search was conducted across 6 databases (MEDLINE, PsycINFO, CINAHL, PubMed, Cochrane Library, and Web of Science) for literature published from January 2010 to October 2025. Thirty studies (24 qualitative, 4 quantitative, and 2 mixed methods) were included. Data on obstacles and enablers were extracted and synthesized using the SEM as an analytical framework. ResultsThe analysis identified 11 key factors across the intrapersonal (eg, knowledge and emotional response), interpersonal (eg, functional support network), organizational (eg, health care access and workplace demands), community (eg, food environment and digital information landscape), and policy levels (eg, funding and economic support). The synthesis reveals how these factors interact across levels, creating systemic challenges such as fragmentation of support and information inequity. ConclusionsSuccessful GDM self-management support requires integrated, multilevel strategies that address factors ranging from the individual level to the policy level. This review provides an evidence-based SEM framework to inform the development of comprehensive support interventions, including those leveraging digital health platforms, to improve maternal and neonatal outcomes.
Background:Diabetes mellitus is a chronic metabolic disorder marked by elevated blood glucose levels and has emerged as a global epidemic that requires management strategies for effective glycemic control through diet. In recent years, mobile apps have emerged as valuable tools for supporting self-management in chronic diseases such as diabetes, particularly for the nutritional aspects of the disease. However, the quality, accuracy, and adherence of these apps to established dietary guidelines remain underexplored and inconsistent. Objective:The study aims to evaluate the quality and adherence to guidelines of digital nutrition management apps for diabetes, with a focus on dietary guidelines from the American Diabetes Association (ADA), World Health Organization (WHO), European Association for the Study of Diabetes (EASD), and Diabetes Canada (DC). Methods:A quality evaluation was performed, involving the identification of mobile apps from the Google Play Store and Apple App Store. A total of 24 apps were selected based on predefined inclusion and exclusion criteria. Apps were analyzed for their content and features using a compliance checklist derived from official dietary guidelines for diabetes, including carbohydrate tracking, meal planning, glycemic control, fiber intake, physical activity, weight management, and education about diabetes. Additionally, the Mobile App Rating Scale was used to evaluate app quality in terms of engagement, functionality, aesthetics, and information. Intraclass correlation coefficients were used to calculate interrater reliability. Pearson correlation coefficients were also calculated. Results:Only 2 apps showed full compliance with the dietary guidelines, while most apps showed partial adherence. The Mobile App Rating Scale evaluation revealed significant variability in app quality, with mean total scores ranging from 9.46 to 17.69. This finding indicated major gaps in user engagement, functionality, educational content, and personalization. The intraclass correlation coefficient was 0.86, which indicates good interrater reliability, and the Pearson correlation coefficient was 0.88, suggesting good consistency between the authors. Conclusions:Even with the growing availability of nutrition management apps for diabetes, many lack full compliance with dietary guidelines and show room for improvement in their content quality. Collaboration between health care professionals, developers, and patients is essential for the future development of these tools to effectively support diabetes self-management. Strengthening guideline adherence and content quality can increase the effectiveness of these digital tools in promoting self-management and improving health outcomes for individuals with diabetes.
Background: Individuals living with type 1 diabetes (T1D) are at an increased risk of experiencing psychological distress; however, there remains a scarcity of scalable and widely accessible support services, particularly for adolescents and young adults. To address this gap, digital mental health interventions are becoming an increasingly important area of innovation in diabetes care. Objective: This study aimed to explore qualitative feedback regarding the "Lift: Thriving with Diabetes" (Lift) well-being app, designed to support emotional well-being among adolescents and young adults with T1D, which was recently tested in a 12-week feasibility trial conducted in New Zealand and the United States. Methods: Of the 59 adolescents and young adults and 22 support people who participated in the main Lift feasibility trial, 13 agreed to participate in this secondary qualitative study. Participants attended a virtual focus group or 1-on-1 interview to discuss their experiences using the app and to explore their perspectives on the app's engagement, functionality, and perceived impact on well-being and diabetes-related coping. Transcribed audio recordings were analyzed using directed content analysis, guided by the Mobile Application Rating Scale end-user framework (with topics of engagement, functionality, aesthetics, and information quality) and interpreted from a realist theoretical position. Results: In total, 9 adolescents and young adults (mean age 21.5, SD 2.06 years; n=5, 56% men) and 4 support people (2 fathers, 1 friend, and 1 partner; mean age 31.3, SD 18.92 years; n=2, 50% men) completed interviews. Overall, participants viewed Lift as engaging, easy to use, and emotionally impactful. The most positive feedback focused on the app's interactive features, particularly a well-being tree that "grew" with increased engagement, and its "calming" visual aesthetics. Users also reported meaningful emotional or behavioral impact, particularly in promoting connection, self-awareness, and practical coping strategies in living with T1D. However, user feedback also highlighted areas for improvement, including the need for improved content pacing, personalization, connection with other digital health tools, and greater gamification to sustain long-term engagement. Participants consistently expressed a desire for content tailored to their age, role (eg, support person vs young person), and personal preferences (eg, voice, pace, tone, and interactivity). Conclusions: Findings underscore the potential of user-driven, emotionally intelligent digital tools to enhance well-being and connection for young people with T1D, as well as their support people. These insights can inform the refinement of Lift and the development of broader digital health interventions aimed at promoting well-being and fostering meaningful, sustained impact.
Background:The triglyceride-glucose (TyG) index has demonstrated promising predictive capability in clinical studies, but its distribution characteristics across different age and sex groups in the Chinese population have not been fully characterized. Objective:This study aimed to describe the population-based distribution of the TyG index. Methods:A total of 4621 participants aged 20-80 years from the China National Health Survey were included in this study. The TyG index was calculated from fasting blood glucose and triglycerides. The age- and sex-specific distribution values of the TyG index were obtained using the percentiles method. Results:Males had higher BMI (23.32, SD 2.63 vs 22.72, SD 2.60 kg/m2), triglyceride (133.78, SD 78.07 vs 111.58, SD 61.18 mg/dL), fasting glucose (97.05, SD 11.95 vs 94.34, SD 10.46 mg/dL), and TyG index values (8.63, SD 0.54 vs 8.45, SD 0.50) than females. The TyG index of males reached its peak value at approximately 40 years of age. The lower limit percentile values for females exceeded that of males around age 50. After the age of 60, the upper limit of the distribution values in females was higher than in males. Conclusions:This study characterized the age- and sex-specific distribution of the TyG index among Chinese adults aged 20-80 years. The results of this study contribute to a more precise assessment of glucolipid metabolism.
BackgroundThe rate of treatment failure with sodium-glucose cotransporter-2 inhibitors (SGLT2i) is high among individuals with type 2 diabetes (T2D). Accurately predicting SGLT2i treatment failure is important for improving the clinical management of T2D. ObjectiveThe study aimed to use machine learning (ML) models to identify factors predicting treatment failure with SGLT2i in T2D and to evaluate model performance. MethodsThis retrospective observational cohort study included adults with T2D treated with SGLT2i (2016-2024). The primary outcome was overall treatment failure with SGLT2i during follow-up (≥180 days after SGLT2i initiation). The secondary outcome was subtypes of treatment failure with SGLT2i (treatment discontinuation, failure with action, and inertial failure) or nonfailure, which was defined as not meeting the definition for one of the failure subtypes. Variables potentially associated with treatment failure were assessed during the year before SGLT2i treatment initiation (analysis 1) and the year before SGLT2i treatment failure (analysis 2). Using these variables, ML models—logistic regression (LR), multilayer perceptron (MLP), extreme gradient boosting (XGBoost), and Transformer—were used to identify significant predictors of the outcomes. Model performance metrics (accuracy, area under the curve, precision, recall, and F1-score) were calculated. Using Shapley Additive Explanations methodology, key features were identified based on their impact on model predictions. LR and Transformer models using key features were further evaluated for their potential to support the development of a risk score for predicting treatment failure with SGLT2i. ResultsAmong all individuals in the study (N=62,222), 71% (n=44,156) had treatment failure with SGLT2i. Across subtypes, failure with action (n=23,839, 38.3%) was more common than treatment discontinuation (n=16,449, 26.4%) and inertial failure (n=3868, 6.2%). Model performance was moderate in both analyses. In analysis 1, the accuracy ranged from 0.72 to 0.73 for predicting overall treatment failure and from 0.56 to 0.57 for predicting the subtype of treatment failure. In analysis 2, the accuracy ranged from 0.74 to 0.75 for predicting overall treatment failure and from 0.61 to 0.63 for predicting the subtype of treatment failure. XGBoost, MLP, and Transformer models showed small improvements compared with LR. Using the top 9 key features identified from the Shapley Additive Explanations analysis, the Transformer model performed similarly in accuracy and area under the curve to its counterpart using the full feature set. ConclusionsPerformance across the LR, MLP, XGBoost, and Transformer models was moderate. The advanced ML models performed slightly better than LR. Overall, the results suggest that further model advancements and increased data availability are needed to better predict treatment failure with SGLT2i. The LR coefficients from the key features model may inform the development of a risk score to predict SGLT2i treatment failure. Accurate prediction could inform individualized treatment planning for individuals with T2D.
