bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2026–08–16
seventeen papers selected by
Mott Given



  1. J Clin Med Res. 2026 Jul;18(7): 472-488
       Background: The study aimed to develop risk stratification models for diabetic peripheral neuropathy (DPN) in patients with type 2 diabetes mellitus (T2DM) using multiple machine learning algorithms, identify the optimal model, and visualize it through a nomogram, thereby providing a clinical decision-support tool for the early identification of high-risk individuals.
    Methods: A retrospective analysis was conducted on 180 inpatients diagnosed with T2DM at the Endocrinology Department of our hospital from January 2021 to September 2024, including 88 patients with DPN (48.9%) and 92 patients without DPN (51.1%). All enrolled participants were randomly divided into a training cohort (n = 126) and an internal validation cohort (n = 54) at a 7:3 stratified ratio. Collected clinical variables covered demographic profiles (age, gender, diabetes duration), anthropometric indicators (body mass index (BMI)), and multiple laboratory biomarkers. Five predictive algorithms were adopted for model construction, namely logistic regression (LR), random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), and decision tree (DT). Model predictive efficacy was comprehensively assessed using receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and SHapley Additive exPlanations (SHAP) interpretability analysis. A visual nomogram was finally developed based on the best-performing model.
    Results: Multivariate regression analysis screened out six independent predictive factors for DPN occurrence, including diabetes duration, glycated hemoglobin (HbA1c), microalbuminuria (MAU), low-density lipoprotein cholesterol (LDL-C), neutrophil percentage (NEUT%), and BMI (all P < 0.05). Among all established models, the XGBoost algorithm yielded the best predictive outcomes in internal validation, with an area under the curve (AUC) of 0.903 (95% confidence interval (CI): 0.816-0.986), accuracy of 85.2%, sensitivity of 84.6%, and specificity of 85.7%. Its predictive efficacy was numerically superior to that of RF (AUC = 0.808), LR (AUC = 0.838), SVM (AUC = 0.884), and DT (AUC = 0.814). SHAP analysis further identified diabetes duration, HbA1c, and MAU as the most influential predictors for DPN risk. The nomogram established based on these core variables achieved a validation AUC of 0.85, with favorable calibration efficiency (Hosmer-Lemeshow P = 0.512) and positive net clinical benefit across a wide range of threshold probabilities.
    Conclusion: The XGBoost-based model shows favorable preliminary performance for cross-sectional DPN risk stratification in T2DM patients based on internal hold-out validation, outperforming traditional statistical approaches. Combined with SHAP interpretability and nomogram visualization, this model provides an exploratory clinical tool for early identification of potential high-risk individuals, requiring further external validation before clinical application.
    Keywords:  Diabetic peripheral neuropathy; Machine learning; Nomogram; Random forest; Risk stratification model; SHAP; Type 2 diabetes mellitus; XGBoost
    DOI:  https://doi.org/10.14740/jocmr6642
  2. Comput Biol Med. 2026 Aug 13. pii: S0010-4825(26)00455-5. [Epub ahead of print]214 111891
      Diabetic Retinopathy (DR) affects millions of people worldwide, but screening at population-scale is still limited by the lack of specialists and the clinical inefficiency of existing deep learning systems, which either perform binary referral or severity grading but not both within a single unified routed architecture. We present a shared EfficientNet-B4 backbone with two independently activatable heads: a binary screener (Head 1) and a joint five-class DR severity and three-class Diabetic Macular Edema (DME) risk grader (Head 2), trained by way of a five-stage progressive curriculum which prevents the gradients of one pathway from interfering with the other. Clinical routing only invokes the grader for referred cases, reducing computational demand as a function of DR prevalence. The screener achieves AUC =0.9789 on APTOS and AUC =0.8685 on zero-shot external validation (Messidor-2); a small calibration sample of 50 Messidor-2 images recovers ≥90% sensitivity, demonstrating that the AUC gap reflects threshold shift rather than a model failure. The grader achieves five-class DR severity QWK =0.7731 with simultaneous three-class DME risk output; an ordinal loss and MixUp remediation stage (Stage 5b) rebalances per-class performance, substantially improving No DR and Severe NPDR F1 at the cost of a partial reduction in PDR F1 and a decrease in Grade 1 F1 from 0.1818 to 0.1429. The principal limitation is that the grader was trained on only 413 IDRiD images, and Grade 1 F1 remains low across all variants, reflecting the scarcity of annotated grading data.
    Keywords:  Boosting; Deep learning; Diabetic macular edema; Diabetic retinopathy; EfficientNet; Fundus screening; Multiple output heads; Progressive training; Severity grading
    DOI:  https://doi.org/10.1016/j.compbiomed.2026.111891
  3. Medicine (Baltimore). 2026 Aug 14. 105(33): e50234
       BACKGROUND: As the prevalence of diabetes rises, diabetic kidney disease (DKD) has become a leading cause of end-stage renal disease. Big data analysis aids in DKD prediction, diagnosis, and personalized treatment. This bibliometric study summarizes the current research status and hotspots in big data-driven DKD research.
    METHODS: On March 26, 2025, DKD-related big data publications were retrieved from the Web of Science Core Collection. CiteSpace and VOSviewer were used for co-authorship, co-occurrence, and co-citation analyses to construct knowledge networks.
    RESULTS: Three hundred twenty documents were identified, involving 2176 authors, 695 institutions, and 51 countries/regions, published in 192 journals. Research grew gradually from 2002 to 2018 and then rapidly after 2019. Frontiers in Endocrinology (21 publications) and Journal of the American Society of Nephrology (421 citations) led in publications and citations, respectively. China (189 publications), Beijing University of Chinese Medicine (10 publications), and Donovan, Michael J (6 publications) were the most productive. Hotspots included DKD (192), machine learning (ML, 99), diabetes mellitus (82), prediction (61), risk (52), chronic kidney disease (51), biomarkers (41), progression (33), artificial intelligence (AI, 27), and expression (27). ML, AI, mechanisms, and cells may be future frontiers.
    CONCLUSION: Big data-driven DKD research is growing, with multi-omics biomarkers underpinning AI/ML models that are hotspots for risk prediction and progression assessment; AI, ML, mechanisms, and cells are the frontiers, which together provide references for DKD precision medicine.
    Keywords:  artificial intelligence; bibliometrics; big data analysis; diabetic kidney disease; machine learning
    DOI:  https://doi.org/10.1097/MD.0000000000050234
  4. Sci Rep. 2026 Aug 14. pii: 25519. [Epub ahead of print]16(1):
      Diabetes mellitus is a persistent metabolic condition that requires accurate and early diagnosis to prevent severe complications. This paper proposes an Extreme Gradient Boosting (XGBoost)-based predictive framework for multi-classification of diabetes mellitus into non-diabetic, pre-diabetic, and diabetic classes. After standardization and exclusion of non-clinical identifiers, Duplicate clinical records were removed from the original dataset, leaving 826 unique records. Comprehensive preprocessing pipeline used a stratified 70:30 train-test split and five-fold cross-validation; scaling and resampling were performed only within training partitions. Experimental results on the original dataset XGBoost achieved an accuracy of 99.60%. Both Random Over Sampling (ROS) and Syntenic minority over sampling technique (SMOTE) have also achieved 99.60% accuracy but provided improved generalization at the expense of higher computational cost. In contrast, Random Under Sampling (RUS) and Cluster Centroids (CC) reduced accuracy to 92.74% and 90.32%, respectively due to information loss. Across five folds, the original XGBoost model achieved 98.79 ± 1.13% accuracy. Benchmarking against Logistic Regression, Random Forest, Support Vector Machine, Decision Tree, and K-Nearest Neighbors showed that XGBoost provided the strongest performance. These findings highlight the effectiveness of XGBoost while emphasizing classification accuracy in multiclass diabetes prediction systems.
    Keywords:  Class imbalance; Diabetes mellitus; Extreme gradient boosting; Multi-classification; Performance evaluation; Preprocessing
    DOI:  https://doi.org/10.1038/s41598-026-65844-3
  5. Exp Ther Med. 2026 Oct;32(4): 259
      Lower extremity deep vein thrombosis (DVT) is a serious complication in elderly patients with hip fracture, contributing to increased morbidity and mortality. Diabetes mellitus, with its prothrombotic state, may further elevate this risk. Early identification of high-risk patients is important for targeted thromboprophylaxis. The objective of the present study was to develop and validate machine learning (ML) models for predicting DVT using clinical variables available at admission in elderly patients with hip fracture, with a specific focus on diabetes as a key predictor. A retrospective cohort study of elderly patients (≥65 years) with hip fracture who were admitted to a tertiary academic medical center (Tianjin Hospital, China) between January 2020 and December 2025, was conducted. A total of seven ML algorithms were developed and validated using a 70-30 split with 10-fold cross-validation. Model interpretability was enhanced using Shapley Additive exPlanations (SHAP) analysis. Subgroup analyses were performed to evaluate model performance across diabetic and non-diabetic populations. DVT occurred in 16.5% (66/400) of patients during hospitalization. Diabetes was present in 32.5% of the cohort and was significantly associated with DVT (odds ratio=2.64; 95% CI: 1.56-4.48). The random forest model demonstrated an improved performance [area under the curve (AUC)=0.92; accuracy=0.87; sensitivity=0.85; specificity=0.88]. SHAP analysis identified diabetes-associated variables (hemoglobin A1c, diabetes duration and fasting glucose) among the top predictors, along with age, albumin, D-dimer and immobility on admission. Lipid parameters, including low-density lipoprotein cholesterol and triglycerides, also contributed to prediction. Model performance remained robust in diabetic (AUC=0.94) and non-diabetic (AUC=0.90) subgroups. Calibration was good (slope=1.02; intercept=-0.02; Brier score=0.09; Hosmer-Lemeshow P=0.45). Decision curve analysis determined clinical net benefit across thresholds of 15-40%. ML models, particularly random forest, accurately predicted DVT in patients with elderly hip fracture using routinely available admission data. Diabetes therefore emerged as a pivotal risk factor and its inclusion enhanced predictive accuracy. The present model holds promise for early risk stratification and individualized thromboprophylaxis, although rigorous external validation is warranted before any clinical application.
    Keywords:  deep vein thrombosis; diabetes; elderly; hip fracture; machine learning; prediction model; random forest
    DOI:  https://doi.org/10.3892/etm.2026.13253
  6. Am Heart J Plus. 2026 Sep;69 100854
       Objective: This study aimed to develop a machine learning (ML) framework to predict incident type 2 diabetes mellitus (T2DM) using routinely available hematological and renal biomarkers, and to assess their added predictive value over conventional clinical risk factors.
    Methods: We analyzed data from 6093 diabetes-free participants from the prospective Mashhad Stroke and Heart Atherosclerotic Disorder (MASHAD) cohort. Predictors included white blood cell count (WBC), red blood cell count (RBC), red cell distribution width (RDW), and other hematological/renal factors. We employed logistic regression and multiple ML models (Random Forest, XGBoost, LightGBM), optimized via grid search and cross-validation.
    Results: Multivariate logistic regression identified RBC, WBC, and RDW as independent predictors of T2DM. The Random Forest model achieved the highest performance with a ROC-AUC of 0.73, an accuracy of 0.67, and correctly identified 223 of 347 incident T2DM cases Using a probability threshold of 0.20, the Random Forest model achieved a sensitivity of 0.618, a specificity of 0.707, a positive predictive value (PPV) of 0.327, and a negative predictive value (NPV) of 0.889 on the independent test set Feature importance analysis identified metabolic syndrome, BMI, uric acid, and age as the strongest contributors, while WBC, RBC, and NLR were the most influential hematological predictors.
    Conclusion: Routine hematological indices, including RBC, WBC, and RDW, were independently associated with incident T2DM, while metabolic syndrome, BMI, uric acid, and age contributed most strongly to overall model prediction. ML provides a complementary approach for early risk stratification, although further validation is required before clinical implementation.
    Keywords:  Glomerular filtration rate (GFR); Hematologic factors; Machine learning; Type 2 diabetes mellitus
    DOI:  https://doi.org/10.1016/j.ahjo.2026.100854
  7. Front Immunol. 2026 ;17 1870438
       Background: Latent autoimmune diabetes in adults (LADA) is characterized by progressive β-cell impairment and severe glycemic lability, predisposing patients to in-hospital hypoglycemia. Few tailored risk-stratification models exist for this population. This study aimed to develop and validate an interpretable machine learning model using routine clinical data to predict in-hospital hypoglycemia in LADA inpatients.
    Methods: This multicenter retrospective study recruited participants from five Chinese tertiary hospitals between January 2019 and September 2025. Data from four centers formed the derivation cohort, and the remaining center served as the independent external validation cohort. The primary endpoint was in-hospital hypoglycemia (blood glucose < 3.9 mmol/L). Three machine learning models, including logistic regression, random forest, and XGBoost, were developed using routine clinical data and assessed for discrimination, calibration, and clinical utility. SHAP analysis was applied to improve model interpretability. Exploratory subgroup analyses in the internal validation cohort examined model performance across clinical subgroups.
    Results: A total of 752 LADA inpatients were enrolled. The incidence of in-hospital hypoglycemia was 44.8% in the derivation cohort and 54.4% in the external validation cohort. Six core predictive factors were identified: largest amplitude of glycemic excursion, fasting C-peptide, glycated hemoglobin, sex, insulin pump use, and previous hypoglycemia. The three models yielded numerically variable discriminative performance across cohorts. Pairwise DeLong tests indicated no statistically significant differences in the AUROC among the three algorithms during external validation. All models showed comparable calibration and threshold-dependent predictive performance in the external cohort. XGBoost was selected as the final model after comprehensive evaluation. Fasting C-peptide was identified as the most influential predictor. Exploratory subgroup analyses demonstrated generally stable model performance across clinical strata. These findings are limited by small subgroup sample sizes and wide confidence intervals, and thus cannot be generalized to external populations. Sensitivity analysis suggested that model performance was not predominantly dependent on the retained glucose-derived predictor.
    Conclusions: The interpretable XGBoost model showed acceptable discrimination, calibration, and potential clinical utility for in-hospital hypoglycemia risk stratification in patients with LADA. This pragmatic predictive tool has the potential to support individualized inpatient glycemic management and facilitate targeted clinical intervention for LADA populations.
    Keywords:  extreme gradient boosting; hypoglycemia; latent autoimmune diabetes in adults; machine learning; predictive model
    DOI:  https://doi.org/10.3389/fimmu.2026.1870438
  8. Graefes Arch Clin Exp Ophthalmol. 2026 Aug 14.
       PURPOSE: Diabetic retinopathy, a major cause of blindness in working-age individuals, advances without adequate recognition and treatment from non-proliferative to proliferative diabetic retinopathy. Accurate classification of these stages is vital for timely intervention, though existing technologies still face challenges in part due to the variability of diabetic retinopathy features. This study investigates a multimodal deep learning model to automatically classify proliferative and non-proliferative diabetic retinopathy.
    METHODS: The model integrates optical coherence tomography and near-infrared imaging using a dataset of 1024 paired images from 226 patients. An advanced machine learning model, a so-called vision transformer with Masked Autoencoders, is used for multimodal learning. This method processes retinal images in segmented sequences to distinguish between structural details from coherence tomography and vascular patterns from infrared imaging. The model's effectiveness is evaluated through refined training and diagnostic sensitivity assessments, using the ROC-AUC metric to measure its accuracy in differentiating between non-proliferative and proliferative stages of diabetic retinopathy.
    RESULTS: Results indicate that the multimodal model significantly outperformed single-modality models, achieving a ROC-AUC score of 0.92, compared to 0.89 for optic coherence tomography alone and 0.87 for near-infrared images alone.
    CONCLUSION: The model's enhanced accuracy is attributed to the complementary nature of both modalities, with tomography providing structural details and near-infrared images highlighting vascular features.
    Keywords:  Automated screening; Deep-learning.; diabetic retinopathy
    DOI:  https://doi.org/10.1007/s00417-026-07387-x
  9. Front Artif Intell. 2026 ;9 1890320
      Gestational diabetes mellitus (GDM) and preeclampsia are among the most significant pregnancy complications, affecting approximately 5-15% and 2-8% of pregnancies worldwide, respectively. These disorders share overlapping metabolic, vascular, inflammatory, and placental mechanisms, highlighting the need for integrated approaches to early prediction and risk assessment. However, existing artificial intelligence (AI)-based prediction models generally address GDM and preeclampsia independently and are often limited by inadequate multimodal data integration, insufficient external validation, and limited model interpretability. This systematic review synthesizes recent advances (2020-2026) in AI-based prediction of GDM and preeclampsia, with emphasis on predictive methodologies, data modalities, validation strategies, and potential clinical applications. The review was conducted in accordance with the PRISMA 2020 guidelines, and 120 studies employing machine learning (ML), deep learning (DL), and hybrid AI approaches using clinical, biochemical, electronic health record (EHR), and multimodal data were included. Across the reviewed studies, AI-based models demonstrated promising predictive performance, with reported area under the receiver operating characteristic curve (AUC) values ranging from 0.70 to 0.95. Ensemble and deep learning approaches generally outperformed conventional statistical methods, particularly when multimodal data were integrated. Frequently identified predictive variables included maternal clinical characteristics, metabolic biomarkers, inflammatory biomarkers, and angiogenic markers such as soluble fms-like tyrosine kinase-1 (sFlt-1) and placental growth factor (PlGF). Nevertheless, important methodological challenges remain, including limited external validation, substantial data heterogeneity, insufficient model interpretability, inconsistent reporting practices, and limited integration into routine clinical workflows. Furthermore, most existing AI models predict GDM or preeclampsia independently despite their shared pathophysiological mechanisms, highlighting an important gap in current prediction research. This review provides a comprehensive synthesis of epidemiological, clinical, mechanistic, and AI-based evidence and proposes an evidence-informed conceptual framework that integrates multimodal data, mechanism-aware modeling, explainable AI, standardized validation, and clinical decision-support considerations. Rather than representing a validated predictive system, the proposed framework provides a conceptual foundation to guide future AI model development, prospective validation, and clinical evaluation. Overall, the findings highlight key opportunities and remaining challenges for developing robust, interpretable, and generalizable AI-based prediction models to support future precision maternal healthcare and improve maternal and neonatal outcomes.
    Keywords:  artificial intelligence; clinical decision support systems; deep learning; gestational diabetes mellitus; machine learning; multi-modal learning; preeclampsia
    DOI:  https://doi.org/10.3389/frai.2026.1890320
  10. PLoS One. 2026 ;21(8): e0355729
      Gestational diabetes mellitus (GDM) is a common disorder of glucose metabolism during pregnancy. Early GDM prediction is crucial for reducing adverse maternal and neonatal outcomes. This paper proposes a hybrid Mamba-Transformer architecture that aggregates clinical and genetic features for GDM prediction. First, a correlation-driven weighted fusion method for clinical and genetic features is introduced. The integrated representation not only enhances feature representation but also highlights the interactive relationship between genetic susceptibility and clinical factors. Second, a sliding window approach is applied to reconstruct the sample sequences from the preprocessed data, generating augmented instances as model input. This transforms isolated individual features into context-aware group features, enabling the effective capture of both population-level heterogeneity and individual risk. Finally, the hybrid Mamba-Transformer architecture is constructed and trained on the publicly available competition dataset (DMRPD) from the Alibaba Cloud Tianchi platform. The model employs a modular and extensible encoder-decoder structure, where the Mamba module serves as an efficient feature extractor for dependencies, while the Transformer module performs deep semantic modeling and sequence abstraction. Experimental results indicate that the proposed method achieves competitive performance compared with other representative models. Specifically, the model attained an AUC of 0.825 on the test set, with sensitivity and specificity at the optimal threshold (0.526) of 0.827 and 0.729, respectively, suggesting reliable discriminative performance on the test set. These findings suggest that the proposed method may provide a useful approach for early GDM risk prediction and could potentially support more targeted screening strategies. However, further validation using larger and independent cohorts is required to confirm its generalizability and clinical applicability.
    DOI:  https://doi.org/10.1371/journal.pone.0355729
  11. Front Endocrinol (Lausanne). 2026 ;17 1888060
       Background: Diabetic kidney disease (DKD) is a major microvascular complication of type 2 diabetes mellitus (T2DM) and the leading cause of end-stage renal disease in China. Limited disease awareness and insufficient early screening tools hinder timely intervention for DKD patients. This study aimed to develop and internally validate a non-invasive machine learning-based prediction model for early DKD risk identification among T2DM patients using routine clinical laboratory indicators.
    Methods: A retrospective cross-sectional study was conducted with 602 eligible T2DM patients (457 without DKD, 145 with DKD) recruited from Sihui People's Hospital between January 2023 and June 2024. Subjects were randomly split into an 8:2 training set (n=481) and independent test set (n=121). Baseline clinical characteristics were compared between groups. Univariate and multivariate logistic regression analyses were performed to screen independent DKD predictors. Six machine learning algorithms including logistic regression, XGBoost, random forest, AdaBoost, support vector classifier (SVC), and Gaussian naive Bayes (GNB) were constructed and comprehensively assessed via AUC, accuracy, sensitivity, specificity, calibration curves, decision curve analysis (DCA), and SHapley Additive exPlanations (SHAP) interpretability analysis.
    Results: Baseline comparisons showed that DKD patients presented worse glycolipid, hepatic, renal, hematological and urinary protein indicators than patients with isolated T2DM, while sex, age, BMI, DBP and FPG showed no significant intergroup differences (all P>0.05). Multivariate logistic regression identified glycated hemoglobin (HbA1c), β2-microglobulin (β2MG), and urine protein (PRO) as independent risk factors, while serum albumin (ALB) and estimated glomerular filtration rate (eGFR) acted as protective factors. Single indicator ROC analysis showed PRO and β2MG achieved the highest diagnostic AUC of 0.944. All six machine learning models exhibited excellent discriminative performance with validation AUCs over 0.975. Logistic regression was selected as the optimal model, yielding a test-set AUC of 0.979, sensitivity of 89.7%, and specificity of 92.4%. The model demonstrated favorable calibration and sustained positive net clinical benefit across almost all threshold probabilities in DCA. SHAP analysis ranked HbA1c, ALB, PRO, eGFR, and β2MG as the top five predictive features, clarifying the individual risk contribution of each biomarker.
    Conclusions: The interpretable logistic regression model built on routine non-invasive clinical indicators reliably identifies DKD risk in T2DM patients. Glycemic control and renal injury biomarkers serve as core predictive factors. This tool provides convenient, low-cost early risk stratification for clinical practice, especially for primary care settings. Further external multicenter prospective validation is required to generalize its clinical application.
    Keywords:  SHAP interpretability; diabetic kidney disease; machine learning; non-invasive screening; risk prediction model; type 2 diabetes mellitus
    DOI:  https://doi.org/10.3389/fendo.2026.1888060
  12. Dialogues Health. 2026 Dec;9 100332
       Background: The rising incidence of diabetes and its complications, particularly diabetic retinopathy (DR), in sub-Saharan Africa presents a significant public health challenge, compounded by a lack of skilled human resources. Artificial intelligence (AI)-based screening offers a promising way to address this burden. Understanding the integration of such complex interventions from research to routine clinical practice is crucial for sustainability.
    Methods: A qualitative study design was employed using the Normalization Process Theory (NPT) framework, which supports the evaluation of whether an innovation will be sustainable in everyday practice by analysing how new practices become embedded into routine work. Clinical staff and patients involved in AI-based DR screening at three clinics in Kigali, Rwanda were interviewed using an NPT-based questionnaire through semi-structured interviews and focus groups. Verbal consent was obtained for recording and transcription. Interview data were thematically analysed, with codes generated to align with the four NPT constructs: coherence, cognitive participation, collective action, and reflexive monitoring.
    Findings: In total, nine clinical staff members and 67 patient participants were interviewed. Participants reported a coherent understanding of the program's purpose, value, and benefits. They valued the technology for its ability to address the gap between the scarcity of competent healthcare providers and the growing burden of DR. Challenges identified related to workload, division of labour, initial patient distrust of AI, and restrictive organizational policies regarding operator access.
    Interpretation: The NPT framework proved valuable for analysing the implementation of this complex intervention, providing insights into user perceptions and generating actionable recommendations for enhancement. While positive adoption was observed, further research is needed to fully understand the intervention's long-term impact on health outcomes such as visual preservation and treatment success.
    Funding: Fundus cameras were donated by Topcon. The company had no input into the design and/or analysis of the study.
    Keywords:  Africa; Artificial intelligence; Diabetic retinopathy; Qualitative
    DOI:  https://doi.org/10.1016/j.dialog.2026.100332
  13. Diabetes Obes Metab. 2026 Aug 11.
       AIMS: To develop and validate a machine learning model incorporating continuous glucose monitoring (CGM) metrics to predict 3-month glycemic target achievement in type 2 diabetes mellitus (T2DM) after short-term intensive insulin pump therapy followed by physician-selected maintenance treatment, and to build a web-based prediction tool.
    MATERIALS AND METHODS: We retrospectively included 1079 patients with T2DM and divided them into training, validation, and test sets (6:2:2). Eleven models were evaluated for discrimination, calibration, net benefit, and classification performance. Candidate models were compared for sensitivity to compression of correlated CGM predictors, interpretability, and suitability for web implementation. Nested logistic regression assessed the incremental value of CGM, and SHapley Additive exPlanations (SHAP) assessed feature contributions.
    RESULTS: SVM_RBF achieved the highest test-set area under the curve (AUC; 0.926), compared with 0.917 for XGBoost (paired DeLong p = 0.521). The models had similar test-set discrimination and probabilistic accuracy, with no consistent difference in calibration. XGBoost was selected for implementation because it was less sensitive to compression of correlated CGM predictors and supported TreeSHAP. Leading predictors included diabetes duration, age, fasting C-peptide (FCP), mean amplitude of glucose excursions (MAGE), body mass index (BMI), and time in range (TIR). Adding CGM improved continuous net reclassification improvement (0.308, p = 0.013) and integrated discrimination improvement (0.0062, p = 0.029), but not AUC or categorical net reclassification improvement.
    CONCLUSION: The XGBoost model estimates 3-month glycemic target achievement within this treatment pathway and was implemented as a web-based clinical decision-support tool.
    Keywords:  3‐month glycemic target achievement; artificial intelligence; continuous glucose monitoring; machine learning
    DOI:  https://doi.org/10.1111/dom.71224
  14. Front Digit Health. 2026 ;8 1724029
       Introduction: Pregnancy-related complications such as gestational diabetes mellitus (GDM) and preeclampsia require timely identification, as initially low-risk pregnancies may develop clinically relevant risks during follow-up. Digital Health and Telemedicine can support remote monitoring, while Digital Twin (DT) technology offers a framework for integrating longitudinal patient data with predictive models. However, DT-based perinatal systems remain exploratory and require methodological validation before clinical use.
    Methods: This study presents a proof-of-concept DT framework for perinatal monitoring. Three data roles are distinguished: public benchmark datasets for model training and testing, synthetic patient records generated with Synthea to simulate prenatal consultations, and synthetic time-stamped wearable-like records generated with Gretel.ai to emulate repeated monitoring at the data-acquisition layer. Clinical and wearable-like streams were processed independently and combined only at the interpretation stage.
    Results: The maternal risk model performed better for low- and high-risk classes than for the intermediate class, supporting its use as a screening-oriented component rather than a definitive classifier. GDM models achieved high recall for the GDM class in both clinical and wearable-like configurations. By contrast, preeclampsia prediction showed limited performance, especially with wearable-accessible blood pressure variables alone; the clinical model improved the class profile but remained insufficient for clinical deployment. The simulated patient cases illustrate the DT workflow and the limitations of wearable-only inputs for complex obstetric risk stratification.
    Discussion: The proposed framework should be interpreted as a methodological prototype for organizing clinical and time-stamped wearable-like data within a perinatal DT. The results support combining episodic clinical information with repeated physiological monitoring, but also show that model calibration, class imbalance, richer clinical features, and validation with real longitudinal cohorts are needed before clinical decision-support use. Future work will focus on real-world validation, sequence-aware modelling, and integration with clinical workflows under medical supervision.
    Keywords:  digital twin; gestational diabetes; machine learning; maternal health; preeclampsia; pregnancy; synthetic data; wearables
    DOI:  https://doi.org/10.3389/fdgth.2026.1724029
  15. Front Public Health. 2026 ;14 1871682
       Background: Long-term self-management is essential for patients with type 2 diabetes mellitus, yet exercise management remains one of the weakest components of self-care. With the advancement of digital health technologies, digital wearable devices have increasingly been used to support diabetes management. However, the pathways by which these devices facilitate exercise behavior change remain insufficiently understood.
    Objective: This scoping review aimed to systematically map the existing evidence on digital wearable devices supporting exercise self-management in patients with type 2 diabetes mellitus, identify their behavior change pathways, and use machine learning-assisted text mining to examine major research themes and hotspots, thereby informing the development of targeted interventions.
    Methods: A scoping review was conducted across PubMed, Scopus, Web of Science, Embase, CINAHL, PsycINFO, and the Cochrane Library, covering studies published from database inception to March 2026. The search strategy was constructed using terms related to type 2 diabetes mellitus, digital wearable devices, exercise self-management, physical activity, and behavior change. Machine learning-assisted literature mining was used to identify thematic patterns in the included studies. The study selection process and overall workflow were conducted in accordance with the PRISMA Extension for Scoping Reviews (PRISMA-ScR).
    Results: 11 studies were included in the review. The evidence indicates that digital wearable devices facilitate exercise self-management in patients with type 2 diabetes mellitus primarily through self-monitoring, real-time feedback, goal setting, motivational activation, social support, and enhanced self-efficacy. Machine learning-assisted text mining further showed that the literature in this field is mainly centered on exercise intervention design, behavioral regulation mechanisms, glycemic monitoring outcomes, and physical activity tracking.
    Conclusion: Digital wearable devices appear to support exercise self-management in people with type 2 diabetes through multiple, interacting behavior-support pathways rather than through any single function alone. This review identifies recurring patterns in the literature rather than making causal inferences, and may inform the development of personalized digital interventions for exercise self-management based on behavior change theory.
    Systematic review registration: https://doi.org/10.17605/OSF.IO/PBG8W.
    Keywords:  digital wearable devices; exercise self-management; machine learning; text mining; type 2 diabetes mellitus
    DOI:  https://doi.org/10.3389/fpubh.2026.1871682
  16. J Clin Med. 2026 Aug 04. pii: 6062. [Epub ahead of print]15(15):
      Background: Functional impairment is common among older adults with diabetes, but multidimensional machine learning classification in this population remains underexplored. This study compared machine learning algorithms for identifying concurrent functional impairment among older adults with self-reported physician-diagnosed diabetes. Methods: This cross-sectional study included 1213 adults aged ≥ 65 years from the nationally sampled 2015 China Health and Retirement Longitudinal Study cohort. Functional impairment was defined as limitations in ≥2 activities of daily living or instrumental activities of daily living. Training set feature selection yielded 18 input features. Eight algorithms were evaluated in a held-out internal test set. Repeated stratified nested 10-fold cross-validation with five repeats assessed stability conditional on the locked feature set. SHAP assessed feature contributions and ranking stability. Results: Functional impairment was present in 34.21% of participants. Random forest achieved the numerically highest test-set AUC (0.774; 95% CI, 0.714-0.833), with sensitivity of 0.512 and a Brier score of 0.181. Its advantage over other models was modest. Under repeated nested cross-validation, random forest achieved a mean AUC of 0.808 ± 0.043 and a mean Brier score of 0.166 ± 0.016. SHAP rankings were stable (Kendall's W = 0.901); leading contributors included depressive symptoms, self-rated health, 2.5-m walking test completion time, distance vision, history of falls, executive function, and bilateral grip strength. Excluding direct physical performance inputs retained most discrimination but modestly reduced threshold-dependent performance. Conclusions: Random forest provided the most favorable overall internal performance for classifying concurrent functional impairment, although sensitivity remained modest. External and prospective validation is required before clinical implementation.
    Keywords:  CHARLS; SHAP; cross-sectional classification; diabetes; functional impairment; machine learning; older adults
    DOI:  https://doi.org/10.3390/jcm15156062
  17. Osong Public Health Res Perspect. 2026 Aug 14.
       Objectives: Medication adherence remains a major challenge in the management of type 2 diabetes (T2D), especially in middle-income countries such as Morocco. With the rapid development of artificial intelligence, large language models, including ChatGPT, may offer new opportunities for clinical research through simulated patient profiles. This study examined the feasibility of using ChatGPT to generate culturally contextualized virtual profiles of Moroccan patients with T2D and to apply the validated Moroccan Arabic dialect version of the general medication adherence scale (GMAS).
    Methods: This exploratory pilot simulation study used ChatGPT-4 to generate 11 virtual profiles based on representative sociodemographic and clinical characteristics of Moroccan patients with T2D. Each profile was administered the validated Moroccan Arabic dialect version of the GMAS.
    Results: Four profiles (36.4%) showed high adherence, 6 (54.5%) displayed moderate adherence, and 1 (9.1%) exhibited low adherence. Expert evaluations were consistent, with all ratings ≥3 on a 4-point scale, supporting the credibility and cultural appropriateness of the profiles. The simulated patterns reflected real-world trends, with higher adherence among profiles treated with oral antidiabetic drugs, covered by health insurance, and having shorter disease duration, and lower adherence among profiles using insulin, lacking insurance, and having longer disease duration.
    Conclusion: ChatGPT generated credible, culturally adapted patient profiles that enabled application of a validated psychometric tool in a simulated environment. This rapid, low-cost approach may support exploratory research, healthcare training, and intervention design. Comparative studies using real patient data are needed to confirm validity.
    Keywords:  Artificial intelligence; Diabetes mellitus, type 2; Feasibility studies; Medication adherence; Morocco; Psychometrics
    DOI:  https://doi.org/10.24171/j.phrp.2025.0396