bims-aukdir 2026-09-06 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–09–06
sixteen papers selected by
Mott Given

A data-driven machine learning model for effective diabetes diagnosis.
Enhancing early diabetic retinopathy detection: a fine-tuned NASNet framework with robust validation and external generalization.
Artificial Intelligence for Diabetic Retinopathy.
AI-Enhanced System for Diabetic Foot Ulcer Localization and Classification Using Multi-Scale Neural CNN Model.
An explainable deep learning approach for automated detection and grading of diabetic retinopathy from fundus images.
Deep learning for blood glucose prediction: Reproducibility challenges and factors affecting differential performance.
Identification of comorbidity-based clusters and mortality risk in hospitalised patients with diabetes mellitus in Karachi, Pakistan: a machine learning analysis.
EMBC Special Issue: PhyNet: A Physiology-Constrained Monotonic Neural Network for Safe and Explainable Blood Glucose Forecasting.
Diabetic retinopathy screening using artificial intelligence: a comprehensive performance comparison on datasets from diverse populations and imaging modalities.
Prediction of carotid vulnerable plaques in patients with type 2 diabetes using interpretable machine learning models.
Machine learning models for predicting new-onset diabetes following acute pancreatitis using real-world data.
Evaluating large language models using the Type 2 Diabetes Health Education guideline: a comparative analysis of ChatGPT-4.1, Claude-4.0, DeepSeek-V3, and ERNIE Bot 4.5 Turbo.
Artificial intelligence-enhanced electrocardiography for the prediction of future type 2 diabetes mellitus: a model-development and multicentre validation study.
Development and validation of a machine learning-based diagnostic prediction model for frailty in older adults with diabetes.
V2D: Detection of type 2 diabetes using only smartphone voice recordings via spectro-temporal transformer embeddings.
User Engagement and Feature Preferences in an AI-Powered mHealth Intervention for Diabetes Prevention: Secondary Analysis of a Randomized Controlled Trial.

Sci Rep. 2026 07 28. pii: 27245. [Epub ahead of print]16(1):

A data-driven machine learning model for effective diabetes diagnosis.

T Shobha, S Pradeep, Seema Patil, J Subramanya, G B Sarthaka Mitra.

  Diabetes is a prevalent long-lasting disease marked by high blood glucose due to inadequate insulin secretion and insulin resistance may result in serious life-threatening complications. Diabetes global prevalence has raised by fourfold over the past thirty years, and is the ninth foremost disease leads to death across the globe. Meanwhile, developments in machine learning presents new opportunities for prediction and classification of disease. However, despite of numerous existing models, a need for classifying types of diabetes still remains. The objective of this study is to develop an integrated, data-driven machine learning model for predicting the occurrence and classification of diabetes in an effort to improve on these limitations and investigate the ability to differentiate between types of diabetes through machine learning approach. To evaluate the proposed framework for classification of diabetes and its subtypes, publicly accessible diabetes-related datasets were employed for model development and evaluation. Binary classification is used for detecting occurrence of diabetes, while multiclass classification was employed for subtype classification, namely Prediabetes(PD), Type 1 Diabetes(T1D), Type 2 Diabetes(T2D), and Pancreatogenic (Type 3c-T3cD) Diabetes. K-Nearest Neighbors (KNN), Logistic Regression, Naive Bayes, Random Forest, and XGBoost machine learning algorithms were implemented. The XGBoost demonstrated the highest performance among all models with an accuracy of 0.97. Its feature importance scores validated predictive accuracy and to identify key factors that distinguish types of diabetes. The proposed model is intended to serve as a decision-support system for screening and classification tool using routine clinical data to facilitate early diagnosis and treatment planning.

Keywords:  Diabetes; Diabetes prediction; Machine learning models; Pancreatogenic diabetes; Prediabetes; Prevalence

DOI:  https://doi.org/10.1038/s41598-026-63505-z
Front Ophthalmol (Lausanne). 2026 ;6 1762381

Enhancing early diabetic retinopathy detection: a fine-tuned NASNet framework with robust validation and external generalization.

Nisha R, M A P Manimekalai, K Martin Sagayam, Syed Immamul Ansarullah, Sami Alshmrany, Shafat Khan, Khair Ul Nisa.

  Early detection of diabetic retinopathy (DR) is critical for preventing irreversible vision loss among patients with diabetes mellitus. Automated screening systems based on deep learning have demonstrated substantial promise in improving diagnostic efficiency and scalability. This study proposes a robust multiclass DR classification framework using a fine-tuned NASNetLarge architecture applied to retinal fundus images. The model leverages transfer learning from ImageNet and incorporates a structured fine-tuning protocol to adapt high-level features to DR-specific patterns. Experiments were conducted on the publicly available APTOS 2019 Blindness Detection dataset, comprising five DR severity classes. The model was evaluated using accuracy, precision, recall, F1-score, sensitivity, specificity, and receiver operating characteristic-area under the curve (AUC) metrics. To ensure robustness, experiments were repeated across five independent runs and reported as mean ± standard deviation. External validation was performed using the EyePACS dataset to assess generalisability. The fine-tuned NASNet model achieved 97.52% ± 0.42 accuracy with strong per-class AUC values (≥ 0.98). External validation yielded 93.42% accuracy and an AUC value of 0.941, confirming generalization capability. Comparative analysis against EfficientNet-B4, DenseNet121, and Vision Transformer (ViT-B16) demonstrated superior performance under identical conditions. The proposed framework shows strong potential as an automated DR screening tool for early-stage detection and large-scale population screening.

Keywords:  deep learning; diabetic retinopathy; fundus images; medical imaging; neural networks; transfer learning

DOI:  https://doi.org/10.3389/fopht.2026.1762381
JAMA Ophthalmol. 2026 Sep 03.

Artificial Intelligence for Diabetic Retinopathy.

Jennifer I Lim, Theodore Leng.

DOI: https://doi.org/10.1001/jamaophthalmol.2026.3479
Health Sci Rep. 2026 Sep;9(9): e73161

AI-Enhanced System for Diabetic Foot Ulcer Localization and Classification Using Multi-Scale Neural CNN Model.

Md Nur-A-Alam, Md Anwar Hussen Wadud, Mostofa Kamal Nasir, Khandaker Mohammad Mohi Uddin.

   Background and Aims: Diabetic foot ulcers (DFUs), which heal slowly because of inadequate blood supply, are among the serious infections and chronic foot problems that can result from diabetes mellitus (DM). The primary consequence of DFU, which can result in amputation if left untreated.
Methods: The paper proposes an intelligent and automated approach for classifying foot images as either healthy or DFU images. At the initial stage, the proposed system presents a novel dataset containing 5500 foot images collected from diverse individuals with healthy and DFU conditions. The preprocessing stage involves three steps: the Region of Interest (ROI) method removes unnecessary portions of the foot images, the RGB images are converted to grayscale, and Non-Local Means (NLM) filtering is applied to reduce noise and remove unwanted information. Two neural feature extractors, ResNet50 and Faster R-CNN, are used to independently extract features from the foot images, and the resulting feature vectors are integrated into a single fused feature. The softmax function of Faster RCNN method classifies DFU or normal image using fused features, and bounding box method regressor function localize the ulcer region from DFU image.
Results: Compared with relevant state-of-the-art methods, the proposed Faster RCNN-based deep learning approach with feature fusion demonstrates superior performance in DFU recognition, achieving a testing accuracy of 99.85%, specificity of 99.37%, and precision of 99.50%. With a detection accuracy of 98.83%, this fusion-based approach demonstrated competitive performance through the use of a generalization validation mechanism.
Conclusion: For automated diabetic foot ulcer identification and localization, the suggested fusion-based Faster R-CNN framework shows very accurate and dependable performance. These results point to its significant potential as a helpful clinical decision-making tool for the early identification and treatment of DFU.

Keywords:  diabetic foot ulcer (DFU); faster region‐convolutional neural network (RCNN); neural feature integration; non‐local means (NLM) filtering method

DOI:  https://doi.org/10.1002/hsr2.73161
Microvasc Res. 2026 Sep 03. pii: S0026-2862(26)00109-3. [Epub ahead of print] 105009

An explainable deep learning approach for automated detection and grading of diabetic retinopathy from fundus images.

Muradul Islam, Marisa Parvez Oishy, Ferdaus Anam Jibon, Utpal Kanti Das, A H M Kamal, Mohammad Mehedi Hassan, Gahangir Hossain.

  Diabetic Retinopathy (DR) is a leading cause of preventable blindness, and it is important to accurately detect it as early as possible, as well as to measure the severity to provide early clinical indicators. Fundus image assessment in the manual mode is subjective and labor-intensive, and is hard to scale, which has encouraged automated approaches. Current deep learning methods are usually based on either convolutional neural networks (CNNs) or transformer-based ones, focusing on local lesion features or global retina context separately. In addition, relational dependencies among lesions that are clinically significant in the case of severity development are under-modeled. This paper introduces a relational, hybrid deep learning model that integrates convolutional, transformer-based, and graph attention models to classify binary and multiclass DR. A ResNet-Graph Attention Network (ResNet-GAT) is introduced to explicitly model spatial relational dependencies among regional feature descriptors and iscompared with independent stand-alone ResNet-50 and Swin Transformer models. Also, a hybrid ResNet-Swin Transformer model combines fine-grained lesion representations and hierarchical global context. Experiments with the APTOS 2019 data show that the hybrid model has an accuracy of 98.09% with a Quadratic Weighted Kappa (QWK) of 0.9618 when used to classify binary, and 94.89% with a QWK of 0.9726 when used to classify five classes of severity. Robustness is proven by cross-dataset inference on IDRiD. The interpretability analysis performed through Grad-CAM shows that the predictions are made based on clinically significant areas, such as microaneurysms, hemorrhages, and exudates. It is demonstrated that performance and interpretability are enhanced by a combination of local, global, and relational representations in automated screening of DR.

Keywords:  Classification; Diabetic retinopathy; Ensemble learning; Fundus image; Graph attention network; Hybrid deep learning

DOI:  https://doi.org/10.1016/j.mvr.2026.105009
PLOS Digit Health. 2026 Sep;5(9): e0001633

Deep learning for blood glucose prediction: Reproducibility challenges and factors affecting differential performance.

Baiying Lu, Biratal Wagle, Zhaohui Liang, Yanjun Cui, Temiloluwa Prioleau.

Blood glucose prediction is a critical component of next-generation diabetes technologies, such as artificial pancreas systems, where reliable performance is essential for safety and effectiveness. Although deep learning methods have achieved promising advances in this area, a critical gap remains in understanding the reproducibility and generalizability of these methods. To contextualize the gap, this study reviewed 67 recent papers that proposed a deep learning method for glucose prediction to identify key reproducibility challenges. Next, we adopted a standardized framework, encompassing technical, statistical, and conceptual reproducibility evaluations, to experimentally assess the reproducibility of eight representative deep learning methods. To achieve this, we reimplemented and evaluated these eight deep learning methods using over 1.36 million continuous glucose monitoring samples (5,061 days) from 128 individuals with type 1 diabetes across three public datasets: OhioT1DM, DiaTrend, and T1DEXI. We found that even though these models demonstrated good technical and statistical reproducibility, their conceptual reproducibility-the ability to generalize to datasets with different diabetes management patterns-was limited. Further analyses revealed that each model's overall prediction performance was strongly influenced by individual glycemic control, with higher prediction errors observed among participants with lower time with blood glucose in the target range (70-180 mg/dL). This study identified key reproducibility challenges associated with current blood glucose prediction methods within type 1 diabetes populations, highlighting the need for increased transparency, dataset diversity, standardized evaluation practices, and code accessibility to ensure reproducible and reliable models for blood glucose prediction.

DOI: https://doi.org/10.1371/journal.pdig.0001633
BMJ Open. 2026 Aug 31. 16(8): e106407

Identification of comorbidity-based clusters and mortality risk in hospitalised patients with diabetes mellitus in Karachi, Pakistan: a machine learning analysis.

Namra Aziz, Tariq Mahmood, Gerald S Bloomfield, Zarmeen Nasim, Zainab Samad, Hooria Imran, Rabiya Owais, Asma Ahmed, Aysha Almas.

   BACKGROUND: Diabetes has reached epidemic proportions in Pakistan. This study applied machine learning (ML) techniques to identify comorbidity-based and diabetic complications-based clusters in hospitalised patients with diabetes and examine their association with in-hospital mortality.
DESIGN: Retrospective cross-sectional study.
SETTING: Aga Khan University Hospital, Pakistan.
PARTICIPANTS: Adult patients (≥18 years) with diabetes admitted between 2008 and 2021.
METHODS: Diagnoses were extracted using International Classification of Diseases (ICD-9 and ICD-10) codes. Data was integrated from the Hospital Information Management Systems (HIMS). The K-Modes clustering algorithm was applied to categorical diagnostic data, with the optimal number of clusters determined using elbow curve and silhouette score analysis. Latent-Dirichlet Allocation and Term Frequency-Inverse Document Frequency were applied to identify frequent terms from each cluster followed by expert validation. Logistic regression was performed to assess the association between cluster membership and in-hospital mortality.
RESULTS: Among 78 271 patients, four clusters were identified with varying mortality risks: (1) cardio, tumour and tobacco (CTT), (2) renal complication cluster (RCC), (3) cardiovascular cluster (CVC) and (4) uncontrolled diabetes mellitus, hypertension, stroke and kidney (HSK). The within cluster mortality was highest in the RCC cluster (682; 18.1%), followed by CVC (543; 4.4%), HSK (1,379; 4%) and CTT (603; 2.2%). Compared with CTT, the unadjusted odds of in-hospital mortality were significantly higher in RCC (OR=9.9, 95% CI 8.9 to 11.2, p<0.001), CVC (OR=2.1, 95% CI 1.9 to 2.3, p<0.001) and HSK (OR=1.9, 95% CI 1.7 to 2.1, p<0.001).
CONCLUSION: Among hospitalised patients with diabetes, the RCC cluster demonstrated the highest mortality risk, comprising predominantly older patients with severe complications. These findings highlight the utility of unsupervised ML approaches for identifying high-risk clinical phenotypes and informing risk stratification in resource-constrained settings. Longitudinal studies are warranted to evaluate progression and long-term outcomes across clusters.

Keywords:  DIABETES & ENDOCRINOLOGY; Diabetic foot; Diabetic retinopathy; Machine Learning

DOI:  https://doi.org/10.1136/bmjopen-2025-106407
IEEE Trans Biomed Eng. 2026 Sep 02. PP

EMBC Special Issue: PhyNet: A Physiology-Constrained Monotonic Neural Network for Safe and Explainable Blood Glucose Forecasting.

Andrea Calzavara, Francesco Prendin, Giacomo Cappon, Simone Del Favero, Andrea Facchinetti.

OBJECTIVE: Deep learning (DL) has become state-of-the-art for blood glucose (BG) forecasting in type 1 diabetes (T1D). However, its black-box nature raises safety and reliability concerns regarding its use for therapeutic decision-support. This study aims to: (1) highlight potential risks associated with standard DL-based BG forecasting, and (2) address them with PhyNet, a physiology-constrained monotonic neural network.
METHODS: Two large-scale datasets (T1DEXI and MetaboNet, 848 subjects in total) were used to develop PhyNet-which enforces physiological consistency through a multi-branch structure and weight constraints-and compare it against six DL baselines (convolutional, recurrent, and transformer-based architectures). Models predicted BG levels up to 90-minute ahead using continuous glucose monitoring (CGM) data, carbohydrate intake, and insulin dosing, and were assessed for: (i) predictive accuracy with standard metrics, and (ii) adherence to physiological principles (i.e., carbohydrates increase BG, insulin lowers it) using explainable AI. Specifically, we evaluated model-predicted responses to varying carbohydrate and insulin intakes, and generated counterfactual explanations to identify model-recommended actions for avoiding adverse events.
RESULTS: At a 30-minute horizon, predictive accuracy was similar across models (RMSE: 19.39-21.00 mg/dL; Time Gain: 10.45-13.35 min). Despite this, only PhyNet consistently captured the physiological effects of carbohydrates and insulin, yielding 0% unsafe recommendations versus up to 64.3% for baselines.
CONCLUSION: Standard DL models can achieve state-of-the-art performance while failing to respect physiology, posing clinical risk. PhyNet preserves accuracy while enhancing physiological fidelity, supporting safer integration into T1D technologies.
SIGNIFICANCE: Evaluating physiological consistency alongside predictive accuracy is essential for responsible clinical translation of DL-based BG forecasting.

DOI: https://doi.org/10.1109/TBME.2026.3730324
Physiol Meas. 2026 Sep 04.

Diabetic retinopathy screening using artificial intelligence: a comprehensive performance comparison on datasets from diverse populations and imaging modalities.

Philipp Schapotschnikow, Vladyslav Dusiak, Murat Osswald, Paolo Sandico Silva, Leo Anthony Celi, Luis Filipe Nakayama.

  ObjectiveDiabetic retinopathy (DR) is the leading cause of preventable blindness in adults and poses significant challenges in low- and middle-income regions due to limited access to skilled clinicians and diagnostic facilities. Automated screening solutions using artificial intelligence (AI) have emerged as an efficient alternative, achieving high diagnostic accuracy. However, these solutions are often developed using data from specific populations obtained using relatively expensive high-end devices. This study addresses the potential scope limitations by evaluating the AI-based screening tool retina.help across eight datasets representing diverse populations, imaging modalities, and geographic regions.ApproachThe datasets include both public and private sources, with images captured using tabletop and handheld fundus cameras. Key performance metrics for detecting binary referable DR - sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC) - were calculated on an image-by-image basis. Main resultsRetina.help demonstrated high accuracy on tabletop images, achieving AUROC values of 0.97 on the BRSET and DeepDRiD datasets. Handheld device performance was more variable, with AUROC ranging from 0.88 (Filipino dataset) to 0.99 (Finnish dataset). Sensitivity declined with increased retinal pigmentation, as evidenced by lower values for datasets from Tanzania (62.2%) and Brazil (76.7%) compared to Finland (89.9%). Images from handheld devices often yielded lower sensitivity due to challenges related to low-contrast images. Nonetheless, retina.help generalized well across diverse datasets, showcasing its robustness.SignificanceThe study highlights the impact of imaging equipment, demographics, and image quality on diagnostic performance. These findings underscore the need for benchmarking AI-based DR screening tools using standardized datasets that encompass diverse populations and imaging conditions. Such evaluations can guide the development of equitable, reliable and robust screening solutions.&#xD.

Keywords:  AI fairness; AI in ophthalmology; Artificial intelligence; Screening; diabetic retinopathy

DOI:  https://doi.org/10.1088/1361-6579/aea2ee
Front Endocrinol (Lausanne). 2026 ;17 1911126

Prediction of carotid vulnerable plaques in patients with type 2 diabetes using interpretable machine learning models.

Qi Cheng, Fengjuan Zhang, Han Wang, Yi Qian, Ping Tang, Weiming Cai, Pingbo Chen, Dengke Li, Lijie Shi.

   Objective: To develop and externally validate an interpretable machine-learning framework for estimating ultrasound-defined carotid plaque vulnerability among patients with type 2 diabetes mellitus (T2DM) and established carotid plaque.
Methods: In total, 884 T2DM patients with carotid atherosclerotic plaques from two medical centers were retrospectively recruited and allocated into training, internal validation, and external validation cohorts. Demographic, clinical, biochemical, continuous glucose monitoring (CGM), and inflammatory indicators were collected. CGM was performed for 72 consecutive hours using a retrospective CGM system (Medtronic iPro2; Medtronic, Northridge, CA, USA), and TIR was calculated from valid CGM recordings. Data preprocessing included clinically reviewed missing-data handling, standardization of continuous variables, and Synthetic Minority Over-sampling Technique (SMOTE), which was fitted only within the training data and, during cross-validation, within each training fold to avoid information leakage. Core predictors were identified by integrating feature-importance rankings derived from Random Forest, linear Support Vector Machine, and Logistic Regression. Five predictive models were developed and evaluated using discrimination, calibration, and clinical-utility metrics.
Results: Six core predictors were identified: systolic blood pressure (SBP), low-density lipoprotein cholesterol (LDL-C), age, time in range (TIR), systemic immune-inflammation index (SII), and smoking status. The XGBoost model showed the best overall validation performance, attaining an AUC of 0.882 (95% CI: 0.795-0.969) in the external validation set. SHAP analysis indicated that higher SBP, higher LDL-C, older age, higher SII, and smoking were associated with a higher predicted probability of vulnerable plaques, whereas higher TIR was associated with a lower predicted probability.
Conclusion: An interpretable XGBoost model based on six clinical, laboratory, and CGM-derived indicators showed acceptable discrimination, calibration, and clinical net benefit for estimating ultrasound-defined plaque vulnerability among patients with T2DM and established carotid plaque. The model is not a substitute for carotid ultrasound; pending prospective workflow and economic evaluation, it may support research-stage prioritization for expert plaque characterization when imaging capacity or expertise is constrained.

Keywords:  carotid vulnerable plaque; continuous glucose monitoring; machine learning; shapley additive explanations; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fendo.2026.1911126
BMJ Public Health. 2026 ;4(3): e004661

Machine learning models for predicting new-onset diabetes following acute pancreatitis using real-world data.

Djibril M Ba, Alireza Vafaei Sadr, Yue Zhang, Phil A Hart, Nazia Raja-Khan, Ruizhe Zhou, Ayesha Siddiqui, Tian Qiu, Jennifer Maranki, Vernon M Chinchilli, Vida Abedi.

   Introduction: About one-quarter of patients with acute pancreatitis (AP) will develop diabetes mellitus (DM) within 3 years, but risk factors remain unclear. This study aims to determine whether machine learning models (ML) can be trained to accurately predict new-onset DM following AP and identify key clinical features using real-world data.
Methods: This retrospective cohort study used de-identified data from the TriNetX federated electronic health records (EHR) network from 1 January 2017 to 11 March 2024. A total of 58 746 patients with AP (International Classification of Diseases-10 code K85) and no prior diagnosis of DM were included. New-onset DM following AP was the main outcome of interest. Five ML models were trained across four prediction windows, including logistic regression (LR), eXtreme Gradient Boosting, Adaptive Boosting, Random Forest and support vector machine. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC).
Results: Among the 58 746 patients with AP (mean (SD) age, 50.2 (16.6) years), the LR model demonstrated the highest overall performance, with a mean accuracy of 0.72 (SD, 0.009) and an AUROC of 0.79 (SD, 0.009). Key clinical features across models included age, pancreatic necrosis, body weight, body mass index, systolic blood pressure, number of medical visits and prior AP laboratory values such as glucose, anion gap, blood urea nitrogen (BUN) and total protein.
Conclusions: In this first real-world evidence study using EHR data, we have developed and demonstrated the feasibility of using ML to predict the new onset of DM after AP. Clinical features such as age, pancreatic necrosis, prior glucose, BUN and anion gap had the highest overall importance scores and may inform tailored prevention strategies.

Keywords:  Diabetes Mellitus; Epidemiology; Preventive Medicine

DOI:  https://doi.org/10.1136/bmjph-2025-004661
Front Public Health. 2026 ;14 1900281

Evaluating large language models using the Type 2 Diabetes Health Education guideline: a comparative analysis of ChatGPT-4.1, Claude-4.0, DeepSeek-V3, and ERNIE Bot 4.5 Turbo.

Zhaoxia Huang, Yuxin Bai, Junyue Luo, Zihang Zhang, Chunlan Hu, Xinyu Hu, Liying Ruan, Jiaxin Zeng, Qianhui Jia, Luyao Deng, Meifeng Xiong, Ying Zhou, Ziyi Qi, Wenxi Lyu, Xu Li.

   Objective: To systematically evaluate the quality and readability of health information generated by four large language models (LLMs) in response to inquiries regarding type 2 diabetes mellitus (T2DM), using an authoritative Chinese clinical guideline as the reference standard.
Methods: A total of 124 standardized questions were extracted from the Chinese Type 2 Diabetes Popular Science Guidelines. Six endocrinologists and diabetes specialists conducted independent, blind evaluations using the CLEAR tool (Completeness, Lack of false Information, Evidence, Appropriateness, Relevance) and PEMAT-P (Patient Education Materials Assessment Tool for Printable materials). Response characteristics were also recorded. Between-model differences were tested using the Kruskal-Wallis H test with Bonferroni pairwise comparisons.
Results: All four models achieved total CLEAR scores within the "very good" range (19-25), with no significant differences seen between models (χ2 = 1.985, p = 0.576). No significant differences were observed in the dimensions of Lack of false information (χ2 = 7.644, p = 0.054), Evidence (χ2 = 2.309, p = 0.511), and Relevance (χ2 = 7.516, p = 0.057). However, significant differences emerged in Completeness (χ2 = 47.661, p < 0.001) and Appropriateness (χ2 = 88.360, p < 0.001). Claude-4.0 received the lowest score in Completeness (median 4.00, IQR 3.00-5.00) but achieved the highest ranking in Appropriateness (median 4.00, IQR 4.00-5.00). On the PEMAT-P, understandability differed significantly across models (χ2 = 159.120, p < 0.001), yet all models surpassed the 70% threshold, with ChatGPT-4.1 highest (median 91.91%, IQR 91.91-100.00%). However, despite significant differences among the various models (χ2 = 354.023, p < 0.001), only ERNIE Bot 4.5 Turbo (median 75.00%, IQR75.00-75.00%) surpassed the 70% threshold, with no single model demonstrating consistent superiority across all dimensions.
Conclusion: Although the four LLMs generally provide accurate and pertinent information regarding type 2 diabetes, enduring limits in actionability and inconsistencies among models in content completeness and understandability restrict their effective use in diabetic patient education. Future development should prioritize stronger step-by-step behavioral guidance and differentiated, scenario-specific model deployment to enhance their value in patient-facing diabetes self-management support.

Keywords:  health information quality; large language models; patient education; readability; type 2 diabetes mellitus

DOI:  https://doi.org/10.3389/fpubh.2026.1900281
Eur Heart J Digit Health. 2026 Aug;7(7): ztag118

Artificial intelligence-enhanced electrocardiography for the prediction of future type 2 diabetes mellitus: a model-development and multicentre validation study.

Libor Pastika, Konstantinos Patlatzoglou, Ewa Sieliwonczyk, Joseph Barker, Boroumand Zeidaabadi, Kathryn A McGurk, Sandhi M Barreto, Lidyane Camelo, Sadia Khan, William R Scott, Declan P O'Regan, Bruce B Duncan, Maria I Schmidt, James S Ware, Shivani Misra, Daniel B Kramer, Jonathan W Waks, Nicholas S Peters, Antonio Luiz Pinho Ribeiro, Arunashis Sau, Fu Siong Ng.

   Aims: A significant proportion of type 2 diabetes cases remain undiagnosed despite screening advances, carrying substantial cardiometabolic risk. Artificial intelligence-enhanced electrocardiography (AI-ECG) detects subtle ECG changes in subclinical disease, potentially enabling opportunistic screening.
Methods and results: We developed AI-ECG Risk Estimator for Diabetes Mellitus (AIRE-DM), a convolutional neural network with discrete-time survival loss, for diagnosis of prevalent and prediction of incident type 2 diabetes. It was trained on 1 163 401 ECGs from 189 537 individuals from Beth Israel Deaconess Medical Center (BIDMC) and externally validated in UK Biobank (UKB; n = 65 606) and ELSA-Brasil (n = 13 739). AI-ECG Risk Estimator for Diabetes Mellitus demonstrated moderate discrimination for prevalent type 2 diabetes (area under the receiver operating characteristic curve: BIDMC 0.724, UKB 0.733, ELSA-Brasil 0.706) and incident type 2 diabetes (C-index: BIDMC 0.667, UKB 0.688, ELSA-Brasil 0.625). The highest AIRE-DM risk quartile had elevated incident diabetes risk vs. the lowest (hazard ratio: BIDMC 4.75, UKB 7.52, ELSA-Brasil 3.96). AI-ECG Risk Estimator for Diabetes Mellitus was non-inferior to the American Diabetes Association Diabetes Risk Test in BIDMC, with improved predictive accuracy when combined. In normoglycaemic patients, AIRE-DM was superior to glycated haemoglobin (HbA1c) for predicting incident diabetes in BIDMC and non-inferior in ELSA-Brasil. The highest risk quartile reached 5% cumulative type 2 diabetes mellitus incidence 5.4 years (BIDMC) and 4.8 years (ELSA-Brasil) earlier than the lowest risk quartile, after adjusting for HbA1c, age, and sex. Phenome- and genome-wide association studies revealed biologically plausible associations with glucose regulation, cardiac morphology, diastolic dysfunction, arterial stiffness, and lipid metabolism.
Conclusion: AI-ECG Risk Estimator for Diabetes Mellitus detects prevalent type 2 diabetes and predicts incident disease, uniquely identifying high-risk individuals within the normoglycaemic range. Combined with clinical scores or biomarkers, it enhances risk stratification, enabling earlier intervention.

Keywords:  AI-ECG; AI-enhanced ECG; AIRE platform; Artificial intelligence (AI); Diabetes prediction; Electrocardiography (ECG); Pre-diabetes; Risk stratification; Type 2 diabetes mellitus (T2DM)

DOI:  https://doi.org/10.1093/ehjdh/ztag118
Geriatr Nurs. 2026 Sep 03. pii: S0197-4572(26)00555-0. [Epub ahead of print]74(Pt A): 104350

Development and validation of a machine learning-based diagnostic prediction model for frailty in older adults with diabetes.

Yaqing Liu, Yuanhong Sun, Longhan Zhang, Xiaoyun Li, Yangsong Zhang, Li Liao, Jing Yang.

   BACKGROUND: Frailty is a significant risk factor for death and disability in older adults with diabetes. Early identification of frailty in this population is crucial for implementing timely interventions.
METHODS: We analyzed national longitudinal data from the China Health and Retirement Longitudinal Study. Frailty status was defined using a modified Frailty Phenotype, and candidate predictors were selected based on published systematic reviews and meta-analyses. Six machine learning models were developed, including logistic regression, support vector machine, random forest, adaptive boosting, gradient boosting decision tree, and gradient boosting. Hyperparameters were tuned using grid search, and model validation was conducted via leave-one-out cross-validation and temporal validation. Model performance was evaluated using the area under the receiver operating characteristic curve, specificity, recall, precision, negative predictive value, accuracy, F1 score, decision curve analysis, calibration curve, and Brier score. The model's decision-making mechanism was interpreted using SHapley Additive exPlanations.
RESULTS: A total of 2366 older adults with diabetes were included in the analysis, of whom 1613 were assigned to the training set and 753 to the test set. The random forest model performed best in estimating probability of frailty (training set: AUC = 0.988, 95% CI: 0.983-0.991; test set: AUC = 0.987, 95% CI: 0.982-0.993). Model interpretation identified depression as the most influential predictor. The model has been deployed on a web-based platform (https://ff.magvel.top/).
CONCLUSIONS: This study developed and validated a machine learning-based diagnostic prediction model to estimate the probability of frailty in older adults with diabetes. The random forest model demonstrated promising performance, with good accuracy and potential clinical utility.

Keywords:  Diabetes; Frailty; Machine learning; Prediction model

DOI:  https://doi.org/10.1016/j.gerinurse.2026.104350
IEEE J Biomed Health Inform. 2026 Sep 03. PP

V2D: Detection of type 2 diabetes using only smartphone voice recordings via spectro-temporal transformer embeddings.

Boujemaa Guermazi, Alexandre Logerais, Erica Y Huynh, Jessica Oreskovic, Adam Callanan, Yan Fossat.

Voice-based screening offers a noninvasive, scalable avenue for early detection of type 2 diabetes using everyday smartphone recordings and acoustic features alone. We present V2D (Voice2Diabetes), a novel application of spectrogram-transformer embeddings derived exclusively from short speech segments for patient-level diabetes classification, without requiring any clinical measures or demographic variables. Adults (n=461; 157 female, 304 male) each completed multiple smartphone recordings while reading randomly selected sentences on their own smartphones. Mel-spectrograms were encoded with a pretrained Audio Spectrogram Transformer (AST) to train sex-stratified patient-level classifiers under fivefold nested cross-validation with a held-out calibration set; predictions were aggregated across recordings per participant. Using acoustic features alone, the models achieved patient-level balanced accuracy of 0.724 ± 0.023 (males) and 0.713 ± 0.021 (females), with area under the receiver operating characteristic curve (AUC) of 0.779 ± 0.019 and 0.788 ± 0.018, respectively, averaged over five independent random seeds. The pipeline incorporated probability calibration, sensitivity-first threshold optimization, and optional interpretable acoustic anchors (e.g., fundamental frequency, harmonic-to-noise ratio) to support clinical interpretation. These results provide a rigorous technical validation of AST-derived embeddings for acoustic-only, sex-stratified T2D classification in smartphone recordings and motivate prospective external validation in broader populations.

DOI: https://doi.org/10.1109/JBHI.2026.3730729
JMIR Mhealth Uhealth. 2026 Sep 02. 14 e92981

User Engagement and Feature Preferences in an AI-Powered mHealth Intervention for Diabetes Prevention: Secondary Analysis of a Randomized Controlled Trial.

AI-DPP Study Group

   BACKGROUND: Prediabetes is highly prevalent and increasing globally, yet lifestyle interventions remain underused. AI-driven mobile health (mHealth) tools can help scale diabetes prevention efforts, but the key factors driving their success are not well understood.
OBJECTIVE: This post hoc secondary analysis of a randomized controlled trial (RCT) aimed to characterize the most valued features and the role of user engagement in outcomes of a fully automated mHealth intervention for diabetes prevention.
METHODS: Data from 151 participants with prediabetes and overweight or obesity who were assigned to an AI-based diabetes prevention program (Sweetch) in a parent RCT (NCT05056376) were analyzed. Engagement (defined as the total number of days the app was used) was categorized into tertiles (low, medium, and high). Baseline characteristics were compared across engagement groups using ANOVA, Kruskal-Wallis, and chi-square tests, and regression models assessed the association between engagement and achievement of diabetes risk reduction outcomes (≥5% weight loss, ≥4% weight loss with ≥150 min/week of physical activity, or ≥0.2 percentage point reduction in hemoglobin A1c [HbA1c] at 12 months). Perceived usefulness of intervention features was surveyed at 12 months.
RESULTS: Median engagement was 98 (IQR 34-232) days. Older age (P<.001) and lower baseline BMI (P=.04) were significantly associated with higher engagement. Compared with low engagement, high engagement was associated with greater odds of achieving the composite diabetes risk reduction outcome (odds ratio [OR] 2.59, 95% CI 1.11-6.01; P=.03), ≥5% weight loss (OR 3.31, 95% CI 1.16-9.42; P=.03), and ≥0.2 percentage point reduction in HbA1c (OR 3.57, 95% CI 1.19-10.75; P=.02). Participants most frequently rated weight tracking, physical activity tracking, and the digital body weight scale as the features that were most helpful for achieving their health goals.
CONCLUSIONS: Higher engagement with an AI-driven intervention requiring no human intervention was associated with improved diabetes risk reduction. Contrary to concerns about lower digital literacy, older adults engaged with the intervention more than younger adults. Features related to weight and physical activity tracking were most valued by patients in the program.
TRIAL REGISTRATION: ClinicalTrials.gov NCT05056376; https://clinicaltrials.gov/study/NCT05056376.

Keywords:  artificial intelligence; diabetes prevention; digital health; lifestyle intervention; mobile app; obesity; overweight; prediabetes; reinforcement learning; user engagement

DOI:  https://doi.org/10.2196/92981