bims-aukdir Biomed News
on Automated knowledge discovery in diabetes research
Issue of 2025–12–07
37 papers selected by
Mott Given
  1. Exploring Explainable Machine Learning for Predicting and Interpreting Self-Reported Diabetes among Tennessee Adults: Insights from the 2023 Behavioral Risk Factor Surveillance System (BRFSS).
  2. HyperGraph-based capsule temporal memory network for efficient and explainable diabetic retinopathy detection in retinal imaging.
  3. Knowledge distillation-based lightweight MobileNet model for diabetic retinopathy classification.
  4. PCANN: Principal Convolutional Analysis Neural Network for Block Chain based Diabetic Retinopathy Detection.
  5. Exploring the Influence of Patients' Heterogeneity on Automated Prediction of Blood Glucose Levels in Type 1 Diabetes.
  6. Graph-enhanced deep learning for diabetic retinopathy diagnosis: A quality-aware and uncertainty-driven approach.
  7. Early, unplanned readmission in diabetes inpatients: Comparing the utility and performance of machine learning and traditional prediction models to guide timely diabetes management team review.
  8. Diabetic retinopathy screening model in low and middle-income countries: a scoping review.
  9. Analysis of CGM data in pregnant women through machine learning and mechanistic models.
  10. Interpretable Machine Learning Model for Predicting 1-Year Unplanned Readmissions in Ischemic Stroke Patients with Diabetes: A Synergistic View of Inflammation and Metabolism.
  11. Multidimensional dietary assessment and interpretable machine learning models predict the risk of prediabetes/diabetes and osteoporosis comorbidity in older adults.
  12. Optimizing Feature Selection and Machine Learning Algorithms for Early Detection of Prediabetes Risk: Comparative Study.
  13. HemoGraph: A Machine Learning Method for Type II Diabetes Pre-screening Using Social and Behavioral Determinants of Health.
  14. Structured Insight from Unstructured Data: Large Language Models for SDOH-Driven Diabetes Risk Prediction.
  15. Predicting Treatment Outcome of Patients with Type 2 Diabetes on Once-Daily Basal Insulin Injections using Machine Learning.
  16. AttenGluco: Multimodal Transformer-Based Blood Glucose Forecasting on AI-READI Dataset.
  17. Prospective validation of deep-learning algorithms for diabetic retinopathy screening: A systematic review and meta-analysis.
  18. Early prediction of gestational diabetes mellitus using machine learning-integrated metabolomic and clinical features.
  19. Cataract surgery in patients with diabetes mellitus: preoperative, intraoperative, and postoperative considerations.
  20. Time Series Glucose Level Detection in fuel-cell based sensors Using Machine Learning: A Comparative Study of K-NN with Dynamic Time Warping and XGBoost.
  21. SMENN-hybrid: an efficient technique combining the synthetic minority oversampling technique with ensemble learning for diabetes prediction.
  22. Modeling chaotic diabetes systems using fully recurrent neural networks enhanced by fractional-order learning.
  23. An explainable machine learning model for predicting preterm birth in pregnant women with gestational diabetes mellitus and hypertensive disorders of pregnancy: development and external validation.
  24. Machine learning-based prediction of atherosclerotic cardiovascular disease risk in adults with diabetes or prediabetes.
  25. Evaluating the Effect of Input Features on Deep Learning Models for Blood Glucose Forecasting.
  26. Multi-Metric Subgroup Analysis for Glucose Forecasting in Type 1 Diabetes.
  27. GlucoNet: A Hybrid Learning Methodology for Non-Invasive Blood Glucose Estimation from PPG.
  28. An Investigation of Foot Temperature Deviations in Individuals with Diabetes: Insights from Wearable In-Shoe Technology.
  29. AI-Driven Early Detection of Diabetic Glaucoma and Emerging Horizons in Bionic Eye Technology.
  30. Health Management of Type 2 Diabetes Mellitus and Its Complications: A Machine Learning Algorithm-Based Retrospective Study in Chinese Communities.
  31. Identifying Hypothyroidism as Complication of Type 1 Diabetes from Continuous Glucose Monitoring Data.
  32. Identifying Markers of Cardiac Arrythmias in Patients with Diabetes Mellitus Type 2.
  33. A deep learning-derived digital biomarker of dysglycemia and its association with genetic risk of type 2 diabetes.
  34. Multi-omics integrated analysis identifies causal risk factors and therapeutic targets for diabetic retinopathy.
  35. Artificial intelligence in diabetes care: from predictive analytics to generative AI and implementation challenges.
  36. Artificial Intelligence-Enabled Mobile Health Intervention (mDiabetes) to Reduce Diabetes Risk Behaviors in Rural India: Quasi-Experimental Pre-Post Study.
  37. Diabetes Prevention Chatbots to Help Dietary Self-management for the Chinese Population.
  1. J Prim Care Community Health. 2025 Jan-Dec;16:16 21501319251400546
    Exploring Explainable Machine Learning for Predicting and Interpreting Self-Reported Diabetes among Tennessee Adults: Insights from the 2023 Behavioral Risk Factor Surveillance System (BRFSS).
    Mustapha Aliyu Muhammad, Jamilu Sani, Mohamed Mustaf Ahmed.
       BACKGROUND: Diabetes remains a major public health concern in the United States, particularly in Tennessee, where prevalence rates exceed national averages. Traditional statistical approaches may not fully capture the non-linear interactions among predictors. This study applied both traditional approaches and machine learning (ML) techniques to predict and identify key contributing factors associated with self-reported diabetes using the 2023 Behavioral Risk Factor Surveillance System (BRFSS) dataset.
    METHODS: A cross-sectional analysis was conducted on 5634 (weighted population 5 614 486) adults from the Tennessee BRFSS dataset. Sociodemographic, behavioral, and health-related variables were analyzed. Data processing, exploratory analysis, and modeling were performed in Python using Pandas, NumPy, Scikit-learn, and SHAP. Seven algorithms were tested: Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, Gradient Boosting, and XGBoost, with stratified 5-fold cross-validation. Models were evaluated using accuracy, precision, recall, balanced accuracy, F1-score, AUROC, and PR-AUC.
    RESULTS: The Gradient Boosting model demonstrated the best overall performance, achieving an accuracy of 82%, precision of 48%, recall of 32%, F1-score of 37%, AUROC of 0.80, and PR-AUC of 0.45. Key predictors included high blood pressure, high cholesterol, body mass index, comorbidity burden, and physical inactivity. SHAP analysis revealed that both clinical factors and social determinants substantially influenced diabetes risk.
    CONCLUSION: This study highlights the strong potential of machine learning, particularly Gradient Boosting, in predicting self-reported diabetes. Integrating SHAP analysis enhanced interpretability by revealing how the above factors interact to influence diabetes risk, underscoring the value of explainable AI for precision public health and targeted prevention strategies.
    Keywords:  Behavioral Risk Factor Surveillance System (BRFSS); SHAP analysis; diabetes; explainable AI; machine learning; population surveillance; public health informatics; risk prediction
    DOI:  https://doi.org/10.1177/21501319251400546
  2. Sci Rep. 2025 Dec 03.
    HyperGraph-based capsule temporal memory network for efficient and explainable diabetic retinopathy detection in retinal imaging.
    Mishmala Sushith, N Malligeswari, M Anlin Sahaya Infant Tinu, M Jaiganesh.
      Diabetic retinopathy (DR) is a chronic complication of diabetes in which the retinal damage may cause vision impairment or blindness if left untreated. The challenges in DR detection are mostly due to the morphological variations of retinal lesions, e.g., microaneurysms, hemorrhages, and exudates, and the imaging condition variability between different clinical environments. Current state of the art deep learning models like convolutional neural network (CNN), recurrent neural network (RNN) and transformer-based architectures are computationally expensive, not robust to noisy datasets and have limitation on interpretability, which makes them difficult to deploy in real world clinical settings. This research offers HyperGraph Capsule Temporal Network (HGCTN), a deep learning framework to address these limitations and to create an accurate, scalable, and interpretable DR detection. Combining hypergraph neural networks for strong modeling of higher order spatial relationships between retinal lesions, capsule networks for permitting hierarchical structuring of feature and memorizing distributed routing place into temporal capsular memory unit (TCMU) for maintaining both long term and short termed temporal dependencies we propose HGCTN, a model that integrates all the methodologies to efficaciously track disease progression. Meta learning techniques and noise injection strategies are used to improve adaptability of the model and thus make the model more resilient to real world image variations. On DRIVE and Diabetic Retinopathy datasets, HGCTN is validated experimentally, and the best accuracy is 99.0% (HDCTN) and 98.8% (ADTATC), while existing models like TAHDL (96.7%) and ADTATC (98.2%) are outperformed. Furthermore, the model has a recall of 100% and 99.8% on DRIVE and the Diabetic Retinopathy dataset, respectively, with a specificity of 99.7% and 99.6%, respectively, and thus has almost no false negatives and a high reliability in identifying DR cases. Hypergraph attention maps and capsule activation images are additionally relied on to validate the model's interpretability as they offer explainable predictions to a clinical audience. HGCTN has high classification accuracy, reduced computational complexity and better generalization than the existing models, which makes it a new benchmark for DR detection, solving the key deficiency of the existing models and laying the foundation for the real-world deployment of the automated ophthalmic diagnosis systems.
    Keywords:  Attention-Based disease classification; Automated retinal screening; Capsule-Based feature extraction; Deep learning in ophthalmology; Diabetic retinopathy detection; Hierarchical feature representation; HyperGraph neural networks; Medical image analysis; Meta-Learning for medical imaging; Temporal capsule memory unit
    DOI:  https://doi.org/10.1038/s41598-025-30128-9
  3. Sci Rep. 2025 Dec 05.
    Knowledge distillation-based lightweight MobileNet model for diabetic retinopathy classification.
    Fitsum Mesfin Dejene, Yehualashet Megersa Ayano, Degaga Wolde Feyisa, Taye Girma Debelee, Hiwot Taye Mekonnen, Girum Woldegebreal Gessesse, Zelalem Chimdesa Merga, Hasset Tamirat Molla, Destaw Mulie.
      Diabetic retinopathy (DR) stands as a leading cause of global blindness. Early identification and prompt treatment are essential to prevent vision impairment caused by DR. Manual screening of retinal fundus images is challenging and time-consuming. Additionally, in low-income countries, there is a significant gap between the number of DR patients and ophthalmologists. Currently, machine learning (ML) and deep learning (DL) are becoming a viable alternative to traditional DR screening techniques. However, DL suffers a major limitation in resource-constrained devices because of its large model size and substantial computational demands. Knowledge distillation is a prominent technique for creating lightweight models, effectively transferring knowledge from a larger, complex model to a smaller, more efficient one without significant loss in performance. Therefore, in this research, a lightweight student model is proposed, which follows the MobileNet architectural design by utilizing depthwise separable convolutions. This design ensures efficient performance suitable for edge device deployment. For binary classification, our proposed model achieved an accuracy, precision, and recall of 98.38% on the APTOS 2019 dataset, whereas the proposed model achieved an accuracy of 93.03% for ternary classification on APTOS 2019.
    Keywords:  Classification; Diabetic retinopathy; Knowledge distillation; Lightweight model; MobileNet
    DOI:  https://doi.org/10.1038/s41598-025-30893-7
  4. Curr Eye Res. 2025 Nov 30. 1-12
    PCANN: Principal Convolutional Analysis Neural Network for Block Chain based Diabetic Retinopathy Detection.
    Somasundaram Krishnamoorthy, Sivakumar Paulraj, Baskaran Periyasamy, Arun Kumar Ramamoorthy.
       PURPOSE: Diabetic retinopathy is an ophthalmic disease that impairs the retinal blood vessels. Diabetic retinopathy can lead to blindness when it is not examined in earlier phases. Adversely, the accurate diabetic retinopathy recognition phase is prominently complicated and needs experienced human analysis of fundus images. Blockchain technology helps share data by allowing users to select what information to share and control who can access it, which is important for managing electronic health records in healthcare sector. Nevertheless, the privacy of user data is compromised due to the training data, which is revealed to unauthorized users.
    METHODS: In this work, a superior module for diabetic retinopathy classification based on Blockchain using principal convolutional analysis neural network is designed. Here, the simulation of Blockchain is carried out. Here, the input image is pre-processed using the Gaussian filter. LadderNet is deployed for lesion segmentation, and the segmentation of blood vessel is done using the Sine-Net model. Moreover, feature extraction is performed with the input image, lesion-segmented image, and blood vessel-segmented image. Finally, diabetic retinopathy classification is executed utilizing the proposed principal convolutional analysis neural network, which is classified into normal, mild non-proliferative diabetic retinopathy, moderate non-proliferative diabetic retinopathy, severe non-proliferative diabetic retinopathy, and proliferative.
    RESULTS: The Blockchain enabled principal convolutional analysis neural network obtains superior values of 90.9%, 91.9%, 92.5%, 89.4%, 88.4%, and 75.5% in terms of metrics like accuracy, true positive rate, true negative rate, positive predictive value, negative predictive value, and Mathews correlation coefficient.
    CONCLUSION: The integration of principal convolutional analysis neural network with Blockchain enhances data integrity and patient privacy, making it a promising solution for early diagnosis and treatment. Also, this approach ensures accurate and efficient detection of diabetic retinopathy.
    Keywords:  Diabetic retinopathy; convolutional neural network (CNN); electronic health records (EHRs); fundus image; principal component analysis network (PCANet)
    DOI:  https://doi.org/10.1080/02713683.2025.2584214
  5. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-5
    Exploring the Influence of Patients' Heterogeneity on Automated Prediction of Blood Glucose Levels in Type 1 Diabetes.
    Oresti Banos, Maria Dolores Aviles, Alvaro Garayoa, Pablo Lopez-Ibarra, Manuel Munoz-Torres, Miguel Quesada-Charneco, Ciro Rodriguez-Leon, Claudia Villalonga.
      This work investigates the impact of patients' heterogeneity, including factors such as gender, age, and clinical conditions, on the performance of machine learning models in predicting blood glucose levels in individuals with Type 1 Diabetes. Using the T1DiabetesGranada dataset, various datasets were generated based on these heterogeneity factors. Three popular prediction models -a Linear model, an LSTM neural network, and a CNN- were applied to both the generated datasets and the complete dataset to measure prediction performance at a 30-minute prediction horizon. The preliminary results suggest that incorporating patient-specific heterogeneity factors generally improves prediction performance, highlighting the existence of bias in standard blood glucose level prediction models. Future research should explore whether these findings hold in other related datasets.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11253928
  6. PLoS Comput Biol. 2025 Dec;21(12): e1013745
    Graph-enhanced deep learning for diabetic retinopathy diagnosis: A quality-aware and uncertainty-driven approach.
    Zarin Akter, Jawad Ibn Ahad, Md Mutasim Farhan, Riasat Khan.
      Diabetic retinopathy (DR) is a leading cause of vision impairment, which significantly impacts working-class populations, necessitating accurate and early diagnosis for effective treatment. Traditional DR classification relies on Convolutional Neural Network (CNN)-based models and extensive preprocessing. In this work, we propose a novel approach leveraging pre-trained models for feature extraction, followed by Graph Convolutional Networks (GCNs) for refined embedding representation. The extracted feature vectors are structured as a graph, where GCN enhances embeddings before classification. The proposed model incorporates quality assessment by predicting a confidence score through a dedicated fully connected layer, trained to align with ground truth quality using binary cross-entropy loss. Uncertainty estimation is achieved by calculating the variance across multiple stochastic passes, providing a measure of the model's prediction reliability. We evaluate the proposed DR detection approach on APTOS2019, Messidor-2, and EyePACS datasets, achieving superior performance over state-of-the-art methods. Using MobileViT as the main feature extractor, we reached a remarkable 98.45% accuracy, 98.45% F1-Score, and 98.06% Kappa on the APTOS2019 dataset. The DenseNet-169 proved to be the best backbone of the pipeline for the Messidor-2 dataset, with an accuracy of 94.90%, F1-Score of 94.87%, and Kappa of 93.63%. Additionally, for external validation, the model demonstrated strong generalization capability on the EyePACS dataset, where DenseNet-169 achieved 97.38% accuracy, 97.37% F1-Score, and 96.72% Kappa, while MobileViT obtained 96.02% accuracy, 96.02% F1-Score, and 95.03% Kappa. Our innovative architecture incorporates uncertainty estimation and quality assessment techniques, enabling accurate confidence scores and enhancing the model's reliability in clinical environments. Furthermore, to strengthen interpretability and facilitate clinical validation, Grad-CAM heatmaps were employed to demonstrate the significance of different input regions on the model's predictions.
    DOI:  https://doi.org/10.1371/journal.pcbi.1013745
  7. Diabet Med. 2025 Dec 06. e70186
    Early, unplanned readmission in diabetes inpatients: Comparing the utility and performance of machine learning and traditional prediction models to guide timely diabetes management team review.
    Sue Mei Lau, Gordana Popovic, Ben Maslen, Barbara Depczynski.
       BACKGROUND: Inpatients with diabetes have higher early unplanned readmission (EUR) rates. Diabetes management team (DMT) review reduces EUR. While glucose-based patient selection for DMT reduces in-hospital adverse outcomes, this single criterion is a suboptimal predictor of EUR.
    AIM: We developed, compared and externally validated four EUR prediction models in diabetes inpatients using primarily early admission data to facilitate timely review. Secondarily, we investigated how combining predictive and glucose data can refine patient selection for DMT review.
    METHODS: We constructed three traditional models (classification tree, logistic group lasso, elastic net) and a neural network model using 14 routinely available variables. Models were externally validated and performance assessed by area under the curve (AUC). We analysed the overlap between high-risk patients and those with abnormal glucose (≥1 glucose level <4 or >15 mmol/L) according to pre-specified sensitivities (25%, 50%, 75%).
    RESULTS: Group lasso, elastic net and neural network performed similarly (AUC 0.722-727 test cohort, 0.653-0.667 validation), outperforming the tree (AUC 0.663 test cohort, 0.595 validation). These models identified 9%, 21%-23% and 41%-42% of admissions as 'high risk' using respective sensitivities of 25%, 50% and 75%. In the group lasso, approximately half of 'high-risk' patients also had abnormal glucose which reduced the DMT review cohort to 4.9%, 10.8% and 19.2% for sensitivities of 25%, 50% and 75%.
    CONCLUSION: EUR prediction models facilitate targeted, timely DMT review. Regularised regression models offer a feasible, practical approach for identifying high-risk patients in resource-limited hospital settings. Combining model-identified risk with abnormal glucose refines patient selection, optimising resource allocation.
    Keywords:  blood glucose; diabetes mellitus; hospital readmission; machine learning; neural networks (computer); risk assessment
    DOI:  https://doi.org/10.1111/dme.70186
  8. BMC Public Health. 2025 Dec 03. 25(1): 4210
    Diabetic retinopathy screening model in low and middle-income countries: a scoping review.
    Yeni Dwi Lestari, Ratna Sitompul, Indah Suci Widyahening, Dicky Levenus Tahapary, Prasandhya Astagiri Yusuf, Muhammad Bayu Sasongko, Pandu Riono, Yosilia Nursakina, Gratcheva Alexandra.
       BACKGROUND: Diabetic retinopathy (DR), a leading cause of blindness in working-age adults, disproportionately affects low- and middle-income countries (LMICs). While preventable through early intervention, DR screening programs are often lacking in these resource-constrained settings.
    OBJECTIVE: This scoping review examines DR screening models implemented in LMICs, identifying evidence, research gaps, and potential improvement strategies.
    METHODS: A literature search across multiple databases identified studies on DR screening in LMICs, limited to peer-reviewed articles from the past 20 years focusing on DR screening model effectiveness or implementation. Key data (study design, screening techniques, outcomes) were extracted and synthesized narratively.
    RESULTS: This review synthesized 30 studies on DR screening in LMICs, mainly from India, South Africa, Pakistan, and Bangladesh. These studies explored diverse screening methods, from traditional techniques (ophthalmoscopy, funduscopy, slit lamp exams) to telemedicine and AI. Non-mydriatic fundus photography, often with AI-assisted grading, was common, as were remote grading and store-and-forward systems. Given the resource constraints in these settings, non-mydriatic methods were often preferred. Many studies optimized resources by training non-physicians for image acquisition, followed by specialist grading. The reviewed studies highlighted the effectiveness of community-based screening programs in expanding coverage and improving patient adherence. Furthermore, they demonstrated the cost-effectiveness of smartphone-based imaging devices and AI-driven systems. However, challenges remained, including limited infrastructure, inconsistent training, and follow-up difficulties.
    CONCLUSIONS: LMIC DR screening utilizes traditional and innovative technologies, with community-based approaches, telemedicine, and AI enhancing reach and accuracy. Transitioning from case-finding to population-based screening requires stronger diabetes surveillance and integration within primary care.
    Keywords:  Diabetes; Diabetic retinopathy; Low and middle-income country (LMIC); Screening
    DOI:  https://doi.org/10.1186/s12889-025-25543-6
  9. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    Analysis of CGM data in pregnant women through machine learning and mechanistic models.
    Nilde Fera, Anna Procopio, Antonio Cutruzzola, Paolo Zaffino, Nicola Cortese, Gloria Formoso, Carlo Cosentino, Concetta Irace.
      Abstract- Accurate glucose forecasting is crucial for optimizing diabetes management, particularly in pregnant women, where gestational diabetes mellitus (GDM) poses significant maternal and fetal risks. Continuous glucose monitoring (CGM) provides valuable real-time data, but the complexity and variability of glucose dynamics make precise prediction challenging. This study compares machine learning (ML) and mathematical modeling (MM) approaches to forecast glucose levels in non-diabetic pregnant women. We introduce a novel hybrid deep learning model (GRU-LSTM) alongside a refined mechanistic model based on stochastic differential equations. Our results indicate that mechanistic models offer the highest predictive accuracy, particularly for longer prediction horizons, while ML-based approaches provide computational efficiency.Clinical relevance - This research lays the foundation for advanced hybrid ML-MM models that enhance glucose fore-casting reliability, supporting better clinical decision-making and diabetes management during pregnancy.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11254496
  10. Clin Interv Aging. 2025 ;20 2163-2175
    Interpretable Machine Learning Model for Predicting 1-Year Unplanned Readmissions in Ischemic Stroke Patients with Diabetes: A Synergistic View of Inflammation and Metabolism.
    Yue Hu, Yunhong Zhang, Peichong Lin, Xinye Hu, Yiping Zhu, Peiling Yan, Fan Fei, Qiuyan Wang, Xuelin Yao, Jingjing Ren.
       Background: To develop and validate an interpretable machine learning (ML) model integrating inflammatory and metabolic biomarkers for predicting the risk of 1-year unplanned readmission in patients with ischemic stroke (IS) and type 2 diabetes mellitus (T2DM).
    Methods: This retrospective study included IS patients with comorbid T2DM who were hospitalized between June 2022 and December 2023. A total of 49 clinical variables were extracted. Least absolute shrinkage and selection operator (LASSO) regression was used for feature selection. The dataset was randomly divided into a training set (70%) and a validation set (30%). Seven widely used ML algorithms were applied to construct predictive models, and model performance was evaluated using a validation set. No external validation was performed in this study. The best-performing model was further interpreted using Shapley Additive Explanations (SHAP), and a dynamic nomogram was developed for individualized risk assessment.
    Results: A total of 833 patients were included, with a 1-year unplanned readmission rate of 34.3%. LASSO regression identified nine key variables: age, neutrophil-to-lymphocyte ratio (NLR), homocysteine (HCY), glycated hemoglobin A1c (HbA1c), triglyceride-glucose (TyG) index, metformin use, and the presence of hyperlipidemia, pulmonary infection, and renal insufficiency. The random forest model demonstrated the best overall performance (area under the curve [AUC] = 0.78, F1 score = 0.70). SHAP analysis indicated that NLR, HCY, HbA1c, and TyG index were the most influential predictors, suggesting that chronic inflammation and metabolic dysregulation play pivotal roles in readmission risk.
    Conclusion: The ML model based on inflammatory and metabolic biomarkers effectively predicts 1-year unplanned readmission risk in IS patients with T2DM, with good interpretability and clinical potential. The dynamic nomogram enables real-time, individualized risk prediction to support early identification of high-risk patients, tailored follow-up, and targeted allocation of healthcare resources.
    Keywords:  inflammatory biomarkers; ischemic stroke; machine learning; metabolic biomarkers; type 2 diabetes mellitus; unplanned readmission
    DOI:  https://doi.org/10.2147/CIA.S544949
  11. Front Nutr. 2025 ;12 1666477
    Multidimensional dietary assessment and interpretable machine learning models predict the risk of prediabetes/diabetes and osteoporosis comorbidity in older adults.
    Yuwen ShangGuan, Kangkang Ji, Zhenhao Lin, Chenyiyi He, Young-Je Sim, Haobiao Liu, Kunyi Huang, Kunpeng Wu, Litao Yan, Kunyuan Xu, Huan Li.
       Background: The health burden of diabetes mellitus and osteoporosis (DM-OP) comorbidity in the aging population is increasing, and dietary factors are modifiable risk determinants. This study developed and validated a machine learning model to predict DM-OP comorbidity using multidimensional dietary assessment.
    Methods: This study utilized data from NHANES cycles 2005-2010, 2013-2014, and 2017-2020, ultimately including 4,678 participants aged ≥65 years. Dietary data were collected through 24-h dietary recalls, encompassing macronutrients, micronutrients, food processing classification (NOVA), and five dietary quality scores. Missing data were handled using random forest algorithm, feature selection was performed using Boruta algorithm, and SMOTE technique addressed class imbalance. Eight machine learning algorithms (XGBoost, decision tree, logistic regression, multilayer perceptron, naive Bayes, k-nearest neighbors, random forest, and support vector machine) were implemented with 10-fold cross-validation for performance evaluation.
    Results: A total of 4,678 participants were included, with 347 (7.4%) having DM-OP comorbidity (concurrent prediabetes/diabetes and osteoporosis). After feature selection, 46 variables were retained for model construction. The random forest model demonstrated superior predictive performance with the lowest error rate (0.161), highest accuracy (0.839), ROC AUC of 0.965, sensitivity of 0.827, and specificity of 0.852. SHAP analysis revealed gender as the most important predictor, with females at higher risk; BMI showed positive correlation with comorbidity risk; while carotenoid, vitamin E, magnesium, and zinc intake were negatively correlated with disease risk, suggesting potential protective associations. An online risk prediction tool was developed based on the optimized random forest model for real-time individual comorbidity risk calculation.
    Conclusion: The random forest model demonstrated excellent performance in predicting diabetes-osteoporosis comorbidity in elderly adults, with gender, BMI, and specific nutrient intake as key predictors. This model provides an effective tool for clinical early identification of high-risk populations and implementation of preventive interventions.
    Keywords:  SHAP analysis; comorbidity; diabetes mellitus; dietary nutrient intake; machine learning; older adults; osteoporosis
    DOI:  https://doi.org/10.3389/fnut.2025.1666477
  12. JMIR Bioinform Biotechnol. 2025 Jul 31. 6 e70621
    Optimizing Feature Selection and Machine Learning Algorithms for Early Detection of Prediabetes Risk: Comparative Study.
    Mahmoud B Almadhoun, M A Burhanuddin.
       Background: Prediabetes is an intermediate stage between normal glucose metabolism and diabetes and is associated with increased risk of complications like cardiovascular disease and kidney failure.
    Objective: It is crucial to recognize individuals with prediabetes early in order to apply timely intervention strategies to decelerate or prohibit diabetes development. This study aims to compare the effectiveness of machine learning (ML) algorithms in predicting prediabetes and identifying its key clinical predictors.
    Methods: Multiple ML models are evaluated in this study, including random forest, extreme gradient boosting (XGBoost), support vector machine (SVM), and k-nearest neighbors (KNNs), on a dataset of 4743 individuals. For improved performance and interpretability, key clinical features were selected using LASSO (Least Absolute Shrinkage and Selection Operator) regression and principal component analysis (PCA). To optimize model accuracy and reduce overfitting, we used hyperparameter tuning with RandomizedSearchCV for XGBoost and random forest, and GridSearchCV for SVM and KNN. SHAP (Shapley Additive Explanations) was used to assess model-agnostic feature importance. To resolve data imbalance, SMOTE (Synthetic Minority Oversampling Technique) was applied to ensure reliable classifications.
    Results: A cross-validated ROC-AUC (receiver operating characteristic area under the curve) score of 0.9117 highlighted the robustness of random forest in generalizing across datasets among the models tested. XGBoost followed closely, providing balanced accuracy in distinguishing between normal and prediabetic cases. While SVMs and KNNs performed adequately as baseline models, they exhibited limitations in sensitivity. The SHAP analysis indicated that BMI, age, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol emerged as the key predictors across models. The performance was significantly enhanced through hyperparameter tuning; for example, the ROC-AUC for SVM increased from 0.813 (default) to 0.863 (tuned). PCA kept 12 components while maintaining 95% of the variance in the dataset.
    Conclusions: It is demonstrated in this research that optimized ML models, especially random forest and XGBoost, are effective tools for assessing early prediabetes risk. Combining SHAP analysis with LASSO and PCA enhances transparency, supporting their integration in real-time clinical decision support systems. Future directions include validating these models in diverse clinical settings and integrating additional biomarkers to improve prediction accuracy, offering a promising avenue for early intervention and personalized treatment strategies in preventive health care.
    Keywords:  extreme gradient boosting; feature selection; k-nearest neighbors; machine learning; prediabetes; prediction; support vector machine
    DOI:  https://doi.org/10.2196/70621
  13. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-5
    HemoGraph: A Machine Learning Method for Type II Diabetes Pre-screening Using Social and Behavioral Determinants of Health.
    Garrett O'Brien, Zhi Tu, Qinglan Ding, Tianyi Zhang.
      Type II diabetes is a serious problem because of its increasing prevalence in both adults and younger populations worldwide. In this paper, we propose HemoGraph, a deep learning-based method that uses social and behavioral determinants of health for diabetes pre-screening. HemoGraph utilizes graph neural networks to explicitly model the relationships between input features. Adopting the Learning Under Privileged Information (LUPI) framework, HemoGraph learns from lab test results during training, which are not accessible at inference time. We train and evaluate HemoGraph on real-world data from the National Health and Nutrition Examination Survey (1999-2018). Our model reaches a recall of 71.56%, achieving a significant 22% improvement compared to the American Diabetes Association self-test tool, while attaining higher overall performance.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11252678
  14. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    Structured Insight from Unstructured Data: Large Language Models for SDOH-Driven Diabetes Risk Prediction.
    Sasha Ronaghi, Prerit Choudhary, David H Rehkopf, Bryant Lin.
      Social determinants of health (SDOH) play a critical role in Type 2 Diabetes (T2D) management but are often absent from electronic health records and risk prediction models. Most individual-level SDOH data is collected through structured screening tools, which lack the flexibility to capture the complexity of patient experiences and unique needs of a clinic's population. This study explores the use of large language models (LLMs) to extract structured SDOH information from unstructured patient life stories and evaluate the predictive value of both the extracted features and the narratives themselves for assessing diabetes control. We collected unstructured interviews from 65 T2D patients aged 65 and older, focused on their lived experiences, social context, and diabetes management. These narratives were analyzed using LLMs with retrieval-augmented generation to produce concise, actionable qualitative summaries for clinical interpretation and structured quantitative SDOH ratings for risk prediction modeling. The structured SDOH ratings were used independently and in combination with traditional laboratory biomarkers as inputs to linear and tree-based machine learning models (Ridge, Lasso, Random Forest, and XGBoost) to demonstrate how unstructured narrative data can be applied in conventional risk prediction workflows. Finally, we evaluated several LLMs on their ability to predict a patient's level of diabetes control (low, medium, high) directly from interview text with A1C values redacted. LLMs achieved 60% accuracy in predicting diabetes control levels from interview text. This work demonstrates how LLMs can translate unstructured SDOH-related data into structured insights, offering a scalable approach to augment clinical risk models and decision-making.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11254798
  15. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    Predicting Treatment Outcome of Patients with Type 2 Diabetes on Once-Daily Basal Insulin Injections using Machine Learning.
    Ali Mohebbi, Niels-Kristian Kjoller, Alexander R Johansen, Henrik Bengtsson, Morten L Jensen, Bente M Stallknecht, Morten Morup.
      Basal insulin has been and remains a common and cost-effective intensification step from insufficient oral antidiabetic drug (OAD) treatment for people with type 2 diabetes (T2D), but individualized intensification alternatives are rapidly increasing. Recently, decision support tools to assist healthcare professionals based on machine learning (ML) algorithms are becoming more popular. By means of ML, the aim of this pilot study is to explore to what extent patient characteristics and continuous glucose monitoring (CGM) data enhance the ability to predict a successful basal insulin treatment outcome beyond what can be predicted based on hemoglobin A1C (HbA1c) alone at treatment initiation. Clinical data were acquired from four different trials with a total of 222 poorly regulated (HbA1c ≥ 7% ) patients with T2D on OAD initiating basal insulin treatment. HbA1c, patient characteristics, and consensus CGM metrics (based on three days) were available and systematically added as input to three classification models, respectively, based on logistic regression and Gaussian process (GP) classification with linear and both linear and nonlinear kernels. Classification models predicted a binarized HbA1c value after six months as either acceptable (HbA1c < 7%) or suboptimal (HbA1c≥7%) using a repeated stratified cross-validation setup. The consensus metrics based on only three days of CGM show a trend towards slightly improved performance when added on top of HbA1c. However, it appears difficult to accurately predict a binarized HbA1c outcome based on the considered patient information to a satisfactory level for clinical use. Future research should consider the outlined limitations associated with this study and suggested considerations for improvement. However, this pilot study can be considered an initial attempt towards leveraging the potential of ML and CGM data for personalised and cost-effective treatment decision-support for basal insulin initiation.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11253142
  16. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    AttenGluco: Multimodal Transformer-Based Blood Glucose Forecasting on AI-READI Dataset.
    Ebrahim Farahmand, Reza Rahimi Azghan, Nooshin Taheri Chatrudi, Eric Kim, Gautham Krishna Gudur, Edison Thomaz, Giulia Pedrielli, Pavan Turaga, Hassan Ghasemzadeh.
      Diabetes is a chronic metabolic disorder characterized by persistently high blood glucose levels (BGLs), leading to severe complications such as cardiovascular disease, neuropathy, and retinopathy. Predicting BGLs enables patients to maintain glucose levels within a safe range and allows caregivers to take proactive measures through lifestyle modifications. Continuous Glucose Monitoring (CGM) systems provide real-time tracking, offering a valuable tool for monitoring BGLs. However, accurately forecasting BGLs remains challenging due to fluctuations due to physical activity, diet, and other factors. Recent deep learning models show promise in improving BGL prediction. Nonetheless, forecasting BGLs accurately from multimodal, irregularly sampled data over long prediction horizons remains a challenging research problem. In this paper, we propose AttenGluco1, a multimodal Transformer-based framework for long-term blood glucose prediction. AttenGluco employs cross-attention to effectively integrate CGM and activity data, addressing challenges in fusing data with different sampling rates. Moreover, it employs multi-scale attention to capture long-term dependencies in temporal data, enhancing forecasting accuracy. To evaluate the performance of AttenGluco, we conduct forecasting experiments on the recently released AIREADI dataset, analyzing its predictive accuracy across different subject cohorts including healthy individuals, people with prediabetes, and those with type 2 diabetes. Furthermore, we investigate its performance improvements and forgetting behavior as new cohorts are introduced. Our evaluations show that AttenGluco improves all error metrics, such as root mean square error (RMSE), mean absolute error (MAE), and correlation, compared to the multimodal LSTM model, which is widely used in state-of-the-art blood glucose prediction. AttenGluco outperforms this baseline model by about 10% and 15% in terms of RMSE and MAE, respectively.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11251776
  17. Surv Ophthalmol. 2025 Dec 02. pii: S0039-6257(25)00226-7. [Epub ahead of print]
    Prospective validation of deep-learning algorithms for diabetic retinopathy screening: A systematic review and meta-analysis.
    Ting-Wei Wang, Wei-Ting Luo, Yu-Kang Tu, Yu-Bai Chou, Yu-Te Wu.
      Deep-learning (DL) algorithms are widely promoted for diabetic-retinopathy (DR) screening, yet their prospective diagnostic accuracy is not well defined. PubMed, EMBASE and ClinicalTrials.gov were searched to April 2025 for prospective evaluations of DL systems using color-fundus images. Two reviewers screened records, extracted data, and applied QUADAS-2. Hierarchical bivariate random-effects models produced pooled sensitivity and specificity for referable and vision-threatening DR), analyzed separately at patient and eye level. Twenty-one prespecified moderators were explored with uni- and multivariate meta-regression; publication bias was assessed with Deeks' test Seventy-three studies from 23 countries (255,330 examinations) met the criteria. Pooled patient-level sensitivity was 0.94 (95 % CI 0.92-0.95) and specificity 0.90 (95 % CI 0.87-0.93); eye-level values were 0.93 (95 % CI 0.91-0.95) and 0.94 (95 % CI 0.92-0.96). DR subtype, retinal-field strategy, camera form factor, and prevalence independently explained heterogeneity (p < 0.05). Performance matched or exceeded pivotal FDA trials (IDx-DR, EyeArt). AI gradability was ≥95 % in 60 % of cohorts, including handheld and smartphone systems. DL-based DR screening achieves consistent, high accuracy across devices and care settings, enabling scalable deployment in primary care, pharmacies, and mobile clinics. Quality assurance and ongoing monitoring are essential to maximize population-level benefits.
    Keywords:  deep learning; diabetic retinopathy; diagnostic accuracy; fundus photography; meta-analysis; prospective validation; screening
    DOI:  https://doi.org/10.1016/j.survophthal.2025.11.012
  18. Front Endocrinol (Lausanne). 2025 ;16 1687146
    Early prediction of gestational diabetes mellitus using machine learning-integrated metabolomic and clinical features.
    Qun Ji, Lan Gao, Haiwei Liu, Xiaofang Chen, Boxia Fu, Yingbei Lin, Fei Wang.
       Background: Gestational diabetes mellitus (GDM), a prevalent metabolic disorder associated with pregnancy, which often postpones intervention until after metabolic complications have developed. This study seeks to develop an integrated predictive model that combines first trimester metabolomic signatures with established clinical risk factors to enable the early detection of high-risk pregnancies prior to the onset of irreversible metabolic damages.
    Methods: A total of 89 pregnant women [45 with GDM, 44 with normal glucose tolerance (NGT)] was recruited at Hainan Provincial People's Hospital. Serum and urine samples were subjected to untargeted metabolomic profiling employing UPLC-MS/MS. Metabolite identification was conducted using the Human Metabolome Database and Metlin databases. Bioinformatics analyses were performed on the differential metabolites. Lasso regression was employed to select the metabolites and clinical features utilized in constructing the model. The entire dataset was divided into a training set and a validation set in a 7:3 ratio. Six Machine learning models were trained to identify patients with GDM. Model performance was assessed using area under the receiver operating characteristic curve (AUC), precision, recall, and F1-score. Shapley Additive exPlanations (SHAP) analysis was used to interpret feature contributions in the optimal model.
    Results: Cases of GDM demonstrated distinct metabolic profiles in comparison to participants with NGT. A total of 528 differential metabolites were identified, and KEGG pathway analysis mapped these metabolites to 20 pathways related to metabolism and human diseases. Lasso regression identified 11 differential metabolites and 3 clinical features for training the ML models. Ultimately, the multilayer perceptron achieved the highest classification performance, with an AUC of 0.984 (95%CI: 0.866-1.000) in the validation set. SHAP analysis identified GlcCer(d18:1/16:0) and triglycerides as the most significant predictors, demonstrating positive associations with the risk of GDM.
    Conclusion: Participants with GDM and NGT show great difference in the levels of many metabolites. The ML model according to the metabolites in the first trimester and clinical feature demonstrates high accuracy for early GDM prediction. The result of this research highlighted the potential of metabolites in the prediction of GDM in the early stage of pregnancy.
    Keywords:  early prediction; gestational diabetes mellitus; machine learning; metabolites; metabolomic profiling
    DOI:  https://doi.org/10.3389/fendo.2025.1687146
  19. Curr Opin Ophthalmol. 2026 Jan 01. 37(1): 42-47
    Cataract surgery in patients with diabetes mellitus: preoperative, intraoperative, and postoperative considerations.
    Carlos R Garcia-Cabrera, Natalie A Afshari.
       PURPOSE OF REVIEW: Diabetes mellitus may influence different stages of cataract surgery, from preoperative evaluation to postoperative recovery. With the rapid increase in the global prevalence of diabetes, understanding evidence-based strategies for optimizing surgical outcomes is critical.
    RECENT FINDINGS: Studies have found that fixed glycated hemoglobin (HbA1c) thresholds alone should not be used to determine the timing of surgery. Instead, a macula-first approach, integrating retinal co-management and optical coherence tomography (OCT) for diabetic patients, allows for tailored decision-making. Key advancements include ocular surface optimization to enhance biometry accuracy, proactive perioperative anti-inflammatory regimens combining NSAIDs and corticosteroids, and selective use of intravitreal anti-VEGF or corticosteroid therapy for diabetic macular edema (DME). Lens choice should refrain from multifocal optics in eyes with diabetic macular diseases. Emerging equity frameworks underscore the need to replace systemic cut-offs with risk-based protocols to enhance access and outcomes.
    SUMMARY: Modern cataract surgery in patients with diabetes requires an individualized retina-integrated approach that emphasizes inflammation control and macular preservation. Embedding equity-driven, OCT-based pathways ensures that surgical excellence extends to populations most affected by diabetes-related vision loss.
    Keywords:  cataract surgery; diabetes mellitus; diabetic macular edema; health equity; optical coherence tomography
    DOI:  https://doi.org/10.1097/ICU.0000000000001184
  20. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-5
    Time Series Glucose Level Detection in fuel-cell based sensors Using Machine Learning: A Comparative Study of K-NN with Dynamic Time Warping and XGBoost.
    Marta Picazo-Cot, Luis Marte, Oriol Guardia-Olivella, Paula Subias-Beltran, Carolina Migliorelli.
      Fuel-cell-based printed sensors offer a self-powered solution for glucose monitoring, ideal for portable and wearable applications. However, variability in current signals from glucose oxidation complicates accurate detection. To address this issue, this study investigates the application of machine learning algorithms to enhance glucose prediction accuracy. Specifically, we compare k-Nearest Neighbors paired with Dynamic Time Warping and eXtreme Gradient Boosting, in order to determine their effectiveness in handling signal variability and improving prediction robustness. The results demonstrate the strengths of both approaches for glucose monitoring, although k-Nearest Neighbors achieved a superior performance, yielding an adjusted R2 of 0.86 and an MSE of 1.76. This improvement may be attributed to the Dynamic Time Warping's ability to effectively capture temporal variations in the glucose oxidation signal.Clinical Relevance- This study demonstrates the potential of machine learning-enhanced fuel-cell-based printed sensors for accurate glucose monitoring, offering a reliable, portable, and cost-effective solution that could improve diabetes management in clinical and wearable healthcare settings.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11253806
  21. Sci Rep. 2025 Dec 03. 15(1): 43104
    SMENN-hybrid: an efficient technique combining the synthetic minority oversampling technique with ensemble learning for diabetes prediction.
    Essam H Houssein, Ibrahim A Ibrahim, Amir Mostafa, Abdullah M Albarrak, Mina Younan.
      Diabetes Mellitus (DM) is a chronic metabolic disorder and a major global health problem, with many cases undiagnosed. Early detection and effective management are essential to prevent complications. This paper presents an efficient hybrid technique that combine the Synthetic Minority Oversampling Technique combined with Edited Nearest Neighbors (SMOTE-ENN) with ensemble learning termed (SMENN-Hybrid). Gradient Boosting was identified as the most effective ensemble method through rigorous multi-metric evaluation. The proposed approach was rigorously evaluated across five diverse datasets: PIMA India, Diabetes Prediction Dataset (DPD), Diabetes Dataset 2019, Raw Merged Dataset (RMD), and Cleaned Merged Dataset (CMD). A comprehensive multi-metric assessment considering F1-Score, ROC-AUC, and Accuracy demonstrated exceptional generalizability, with Gradient Boosting achieving a composite score of 99.93/100 and maintaining coefficients of variation below 2% across all metrics (mean F1=0.9860, ROC-AUC=0.9990, Accuracy=0.9860). 5-fold stratified cross-validation confirmed remarkable stability (overall CV < 1.65% for all metrics), while systematic ablation studies validated the essential synergy between SMOTE and ENN, showing average improvements of +16.78% in F1-Score and +29.47% in Recall over unbalanced baselines. Compared to traditional methods (Logistic Regression and Decision Tree), the proposed framework achieved consistent improvements of +2.99% average F1-Score over the best baseline, with individual dataset gains ranging from +3.25% to +3.99%. Despite 246× longer training time, inference remains practical at 2.47ms, making the approach suitable for real-time clinical deployment. The combination of high effectiveness (mean F1=0.9841), exceptional consistency (CV < 2%), and comprehensive validation across multiple datasets and evaluation dimensions positions this framework as a clinically deployable solution for diabetes detection without dataset-specific tuning, offering significant advantages for similar healthcare classification tasks.
    Keywords:  Deep learning; Diabetes Mellitus (DM); Ensemble learning; Machine learning; Synthetic Minority Oversampling Technique combined with Edited Nearest Neighbors (SMOTE-ENN)
    DOI:  https://doi.org/10.1038/s41598-025-26583-z
  22. Sci Rep. 2025 Dec 05. 15(1): 43244
    Modeling chaotic diabetes systems using fully recurrent neural networks enhanced by fractional-order learning.
    Esraa Mostafa, Tarek A Mahmoud, Mohamed A El-Brawany.
      Modeling nonlinear medical systems plays a vital role in healthcare, especially in understanding complex diseases such as diabetes, which often exhibit nonlinear and chaotic behavior. Artificial neural networks (ANNs) have been widely utilized for system identification due to their powerful function approximation capabilities. This paper presents an approach for accurately modeling chaotic diabetes systems using a Fully Recurrent Neural Network (FRNN) enhanced by a Fractional-Order (FO) learning algorithm. The integration of FO learning improves the network's modeling accuracy and convergence behavior. To ensure stability and adaptive learning, a Lyapunov-based mechanism is employed to derive online learning rates for tuning the model parameters. The proposed approach is applied to simulate the insulin-glucose regulatory system under different pathological conditions, including type 1 diabetes, type 2 diabetes, hyperinsulinemia, and hypoglycemia. Comparative studies are conducted with existing models such as FRNNs trained using gradient descent (FRNN-GD), Deep Feedforward Neural Network (DFNN, Diagonal RNNs with gradient descent (DRNN-GD), and DRNNs with FO learning (DRNN-FO). Simulation results confirm that the proposed FRNN-FO model outperforms these alternatives in terms of accuracy and robustness, making it a promising tool for modeling complex biomedical dynamics.
    Keywords:  Chaotic systems; Diabetes modeling; Fractional order learning; Fully recurrent neural network
    DOI:  https://doi.org/10.1038/s41598-025-28637-8
  23. Front Endocrinol (Lausanne). 2025 ;16 1665935
    An explainable machine learning model for predicting preterm birth in pregnant women with gestational diabetes mellitus and hypertensive disorders of pregnancy: development and external validation.
    Landan Kang, Dan Luo, Wenchi Xie, Xiaojing Luo, Jie Mei, Jing He.
       Background: Gestational diabetes mellitus (GDM) and hypertensive disorders of pregnancy (HDP) often coexist and share pathophysiological features such as insulin resistance and endothelial dysfunction, increasing the risk of preterm birth. However, few predictive models have focused specifically on this high-risk group. This study aimed to develop and externally validate a machine learning model for this high-risk population and assess its clinical utility and interpretability.
    Methods: This retrospective dual-center study included electronic medical records from 121 and 136 pregnant women with comorbid GDM and HDP, which served as the development and external validation cohorts, respectively. Multiple machine learning algorithms, including Least Absolute Shrinkage and Selection Operator (LASSO) regression, Random Forest (RF), and Naive Bayes (NB), were applied to construct predictive models. To address class imbalance and enhance model robustness, the Synthetic Minority Over-sampling Technique (SMOTE, which generates synthetic samples for the minority class to balance imbalanced datasets) was employed. Model interpretability was further assessed using Shapley Additive Explanations (SHAP).
    Results: Thirteen variables with univariate significance were entered into Elastic Net regression, yielding five key predictors: alanine transaminase (ALT), aspartate transaminase (AST), Albumin, lactate dehydrogenase (LDH), and systolic blood pressure at 32 - 36 weeks (SBP_32_36). While the LASSO model achieved the highest area under the receiver operating characteristic curve (AUC, 0.802), the NB model demonstrated greater clinical net benefit, higher reclassification performance as measured by the Net Reclassification Improvement (NRI, which evaluates whether patients are more accurately assigned to higher- or lower-risk groups, which reflects the average improvement in distinguishing high-risk from low-risk patients) and Integrated Discrimination Improvement (IDI), and greater robustness in SMOTE-based sensitivity analyses. In the external validation cohort (n = 136), it maintained strong generalization with an AUC of 0.777 (95% confidence interval [CI]: 0.645-0.887), accuracy of 0.801 (95% CI: 0.735-0.860), sensitivity of 0.792, and specificity of 0.804, supporting its selection as the optimal model for this high-risk population.
    Conclusions: The Naive Bayes model exhibited robust predictive ability and interpretability for identifying preterm birth risk in pregnancies with comorbid GDM and HDP, and may serve as a transparent, clinically applicable tool for individualized obstetric risk management.
    Keywords:  Elastic Net regression; Shapley Additive Explanations; gestational diabetes mellitus; hypertensive disorders of pregnancy; preterm birth; risk prediction model
    DOI:  https://doi.org/10.3389/fendo.2025.1665935
  24. BMC Cardiovasc Disord. 2025 Dec 04.
    Machine learning-based prediction of atherosclerotic cardiovascular disease risk in adults with diabetes or prediabetes.
    Yang Li, Bing Wang.
      
    Keywords:  Atherosclerotic cardiovascular disease; Diabetes; Machine learning; Prediabetes; Predictive model
    DOI:  https://doi.org/10.1186/s12872-025-05247-w
  25. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    Evaluating the Effect of Input Features on Deep Learning Models for Blood Glucose Forecasting.
    Andrea Calzavara, Francesco Prendin, Giacomo Cappon, Simone Del Favero, Andrea Facchinetti.
      In type 1 diabetes (T1D), predicting future blood glucose (BG) concentration tens of minutes in advance is a key element in decision support systems and artificial pancreas devices. Recently, given the availability of large datasets, deep learning (DL) models for BG forecasting have been investigated, showing promising results. However, the impact of typical input features-such as meal, insulin, and physical activity (PA)- on their performance remains still underexplored. The aim of this study is to assess how these inputs may affect DL models performance.We trained and evaluated five DL models on four weeks of daily-life data from 497 individuals with T1D. Seven input configurations, ranging from univariate continuous glucose monitoring (CGM) models to comprehensive approaches incorporating CGM, insulin, carbohydrate (CHO) intake, heart rate (HR), and exercise data, were evaluated. Results indicate that incorporating additional features progressively enhances performance at the 30-minute prediction horizon (PH), with all models showing similar Root Mean Squared Error (RMSE) and Time Gain (TG). For the CNN-Transformer, the model showing the greatest improvement, the univariate approach achieved an RMSE of 21.02 ± 3.5 mg/dL and a TG of 10.38 ± 1.33 minutes. Incorporating all factors reduced RMSE to 18.63 ± 4.18 mg/dL and increased TG to 12.12 ± 2.64 minutes. Notably, prediction accuracy during exercise improved only when PA data were included, reducing RMSE from 28.72 ± 9.18 mg/dL to 24.7 ± 7.8 mg/dL. While these improvements are statistically significant, their potential clinical benefit remains limited due to the modest magnitude of change.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11252941
  26. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-4
    Multi-Metric Subgroup Analysis for Glucose Forecasting in Type 1 Diabetes.
    Daphne N Katsarou, Eleni I Georga, Stelios Tigas, Costas Papaloukas, Dimitrios I Fotiadis.
      Type 1 Diabetes (T1D) management remains challenging due to the complexity of glucose regulation, demanding accurate and reliable short-term glucose prediction models. This study investigates the use of population-based and personalized predictive models to enhance glucose prediction in T1D patients. We develop an XGBoost-based model, optimized using Bayesian methods, to predict glucose concentration at 15, 30, and 60-minute prediction horizons. The model was tested across various patient subgroups categorized by glucose patterns, glucose variability (GV), and other clinical factors. Results show that the population-based model consistently outperformed personalized models, achieving RMSE values of 17.39 mg/dL, 27.28 mg/dL, and 40.69 mg/dL at 15, 30, and 60 minutes, respectively. Subgroups with higher GV exhibited poorer prediction accuracy. This study highlights the potential of combining population-based models with subgroup-specific optimizations to improve glycemic control in T1D patients. Accurate glucose prediction is crucial for improving glycemic control and reducing risks, ultimately enhancing patient outcomes.Clinical relevance- This study shows that population-based models can enhance glucose prediction in Τ1D, helping clinicians optimize insulin therapy and improve glycemic control.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11254917
  27. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-6
    GlucoNet: A Hybrid Learning Methodology for Non-Invasive Blood Glucose Estimation from PPG.
    Rahul Verma, Prateek Agrawal, Rashmi Kumari, Surita Sarkar, Pabitra Das, Amit Acharyya.
      The growing prevalence of the diabetic population across the world makes it absolutely necessary to monitor the blood glucose level (BGL) on a regular basis. Although, conventional BGL monitoring methods are widely used for clinical intervention as well as individual monitoring of BGL, those are invasive, painful, and unsuitable for continuous monitoring. These limitations lead to the development of noninvasive alternative methodologies, such as photoplethysmography(PPG) based glucose level estimation. PPG has emerged as a promising method for monitoring BGL, offering several advantages over other optical techniques. In this work, a deep learning architecture, GlucoNet has been introduced that combines Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) modules for learning both spatial and temporal features from PPG to estimate blood glucose levels from PPG signals. Our model was trained and validated on the publicly available online dataset and real-world dataset to estimate its robustness. Our proposed model outperforms the state of the art methods by achieving a mean absolute error (MAE) of 2.15 mg/dL, root mean squared error (RMSE) of 3.28 mg/dL, and an R2 of 0.99. Furthermore, our model demonstrates exceptional clinical performance, with 100% of predictions falling within the clinically acceptable zones A and B of the Clarke Error Grid analysis.Clinical relevance-This study introduces a non-invasive, real-time methodology for blood glucose level estimation using only PPG signal. This model ensures painless, and continuous monitoring of BGL, and can be used as a reliable solution for early detection and proper management of diabetes and related complications.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11253359
  28. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    An Investigation of Foot Temperature Deviations in Individuals with Diabetes: Insights from Wearable In-Shoe Technology.
    Mark Borg, Stephen Mizzi, Robert Farrugia, Tiziana Mifsud, Anabelle Mizzi, Josef Bajada, Owen Falzon.
      Plantar foot temperature is a valuable indicator of diabetes-related complications, but traditional assessment methods, such as infrared thermography and contact thermometers, require unshod feet and controlled conditions, limiting their practicality for continuous monitoring. In this study, we employ a smart insole with 21 embedded temperature sensors to capture plantar temperature data from shod feet. We introduce a novel approach that leverages per-foot relative temperature values-normalized to the foot's mean-rather than absolute values or inter-foot asymmetry. Using data collected during static postures (lying, sitting, and standing), we evaluate multiple machine learning classifiers, with Random Forest achieving the highest accuracy (83.20%), alongside high sensitivity (93.75%) but moderate specificity (63.6%). To enhance explainability, we apply SHAP analysis to interpret model predictions and identify key sensor contributions. Additionally, we derive simple decision rules from the Random Forest model, finding that two medial arch sensors can achieve near-equivalent accuracy (80.38% and 79.82%) to the full model. These results suggest that deviations in plantar temperature patterns could serve as an indicator of diabetes-related foot health changes. Future work will expand this approach to ambulatory activities, integrating static and dynamic features to develop an insole-based system for continuous foot health monitoring in real-world settings.Clinical relevance- This study demonstrates that plantar temperature patterns, captured via smart insoles, can reflect changes associated with diabetes-related foot health conditions. The identification of simple, high-accuracy decision rules using medial arch sensors suggests a potential low-cost, non-invasive tool for detecting thermal irregularities linked to diabetes-related foot complications.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11254373
  29. Zhongguo Ying Yong Sheng Li Xue Za Zhi. 2025 Dec 01. 41 e20250031
    AI-Driven Early Detection of Diabetic Glaucoma and Emerging Horizons in Bionic Eye Technology.
    Gaurav Tiwari, Ankita Wal, Raghuraj Singh Suryavanshi, Rishi Shukla, Muzammil Khan, Brijesh Kumar Chaurasia.
       BACKGROUND: Diabetic glaucoma is a serious eye disorder that can lead to permanent vision loss and is increasingly seen in individuals with long-term diabetes. With its rising global incidence, there is a critical need for early and reliable methods of detection to prevent severe complications.
    OBJECTIVE: This study highlights the growing role of artificial intelligence (AI), especially deep learning technologies, in identifying diabetic glaucoma at an early stage. It also reviews progress in bionic eye technologies designed to help restore vision in affected individuals.
    METHODS: Relevant scientific literature was reviewed by searching databases including PubMed, Taylor francis, ScienceDirect, MDPI, and Bentham. Articles published up to 2025 were considered, focusing on terms such as "diabetic glaucoma,""retinal imaging,""deep learning,""AI in eye care,""bionic eye,"and "neuroprosthetics."Studies were selected based on their relevance to diagnostic innovations and vision-restoration technologies.
    RESULTS: Recent developments in AI have enabled more accurate interpretation of retinal images, such as those from fundus cameras and optical coherence tomography (OCT), aiding in early detection of structural changes linked to glaucoma. At the same time, bionic eye systems-based on neuroprosthetic implants-are showing promise in partially restoring vision in cases of severe visual impairment.
    CONCLUSION: Combining AI-powered diagnostics with emerging bionic eye technologies represents a major shift in managing diabetic glaucoma. These innovations have the potential to improve early detection and offer new options for visual rehabilitation, paving the way for more effective patient care in ophthalmology.
    Keywords:  Artificial intelligence; Bionic eye; Deep learning; Diabetic glaucoma; Retinal imaging; Visual prosthesis
    DOI:  https://doi.org/10.62958/j.cjap.2025.031
  30. Diabetes Metab Syndr Obes. 2025 ;18 4367-4384
    Health Management of Type 2 Diabetes Mellitus and Its Complications: A Machine Learning Algorithm-Based Retrospective Study in Chinese Communities.
    Xin Luo, Jingming Liang, Hong Pan, Dian Zhou, Hong Ye, Ying Zhao, Jijia Sun, An Zhang.
       Background: The global burden of diabetes mellitus (DM) and its complications is a major global public health challenge. This study aimed to improve community capacity for DM management by developing a risk prediction model for complications and providing health management recommendations using machine learning (ML).
    Methods: A retrospective analysis was conducted of 4916 type 2 diabetes (T2DM) patients from Shanghai communities. Model I was developed and compared by using the least absolute shrinkage and selection operator (Lasso) regression, support vector machine (SVM), decision tree (DT) and logistic regression (LR). A Bayesian Network (BN) model to uncover potential causal relationships. Model I was evaluated and adjusted using the receiver operating characteristic (ROC) curve, area under the curve (AUC), calibration curve, and decision curve analysis (DCA). The BN model was assessed using AUC, accuracy, specificity, and sensitivity.
    Results: Five consistent predictors were identified: disease course, diastolic blood pressure, HbA1c, urinary creatinine, and urinary microalbumin. Model I achieved AUCs of 0.695 (training) and 0.676 (validation), with decision curve analysis showing risk thresholds of 12-92% and 20-92% respectively. The calibration curves showed good calibration. The tree-augmented BN model achieved the AUC of 0.755, accuracy of 0.733, specificity of 0.802 and sensitivity of 0.519.
    Conclusion: Effective models for predicting complication risk in T2DM patients were developed. T2DM patients with chronic comorbidities, higher income, and longer disease duration as key targets for community management. We recommend prioritizing UMA as a key monitoring indicator and strengthening comprehensive interventions, including health education, dietary self-management, and family-community support.
    Keywords:  T2DM; health management; machine learning
    DOI:  https://doi.org/10.2147/DMSO.S556130
  31. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-6
    Identifying Hypothyroidism as Complication of Type 1 Diabetes from Continuous Glucose Monitoring Data.
    Agnese Piersanti, Agnese Callegari, Libera L Del Giudice, Christian Gobl, Laura Burattini, Andrea Tura, Micaela Morettini.
      Recent studies have shown that type 1 diabetes mellitus (T1DM) is an important risk factor for the development of hypothyroidism. In this regard, a timely intervention is fundamental to limit adverse effects. Providing real-time measurements of interstitial glucose, Continuous Glucose Monitoring (CGM) devices may represent a powerful source of data to feed machine-learning based algorithms for the discovery of hidden patterns related to the development of diabetes complications such as hypothyroidism. Aim of this study was to setup a machine-learning-based approach capable to identify subjects with hypothyroidism among those with T1DM, starting from CGM tracings. CGM data acquired during a period of 26 weeks and relating to 79 subjects with T1DM taken from the REPLACE-BG campaign database, of which 51 had hypothyroidism and 28 had T1DM with no other complication, were used. The CGM traces were pre-processed to handle the presence of missing data and 41 features were extracted with the use of AGATA software. The feature set was then reduced through Two-Step Decision Tree-Embedded Feature Selection (DT-EFS), leading to the inclusion of 8 final features. The best performing model was the decision tree, showing the following testing performances: area under receiver operating characteristics of 72.3%, accuracy of 71.4%, precision of 74.6%, F1 score of 70.1%, sensitivity of 71.4% and specificity of 69.5%. The 8 features identified herein describe the long-term variability of the subjects' glycemic trace which may suggests a possible connection with the presence of hypothyroidism in T1DM.Clinical Relevance-This establishes the possibility to automatically detect hypothyroidism in T1DM from clinically meaningful CGM glycemic patterns.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11252628
  32. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-4
    Identifying Markers of Cardiac Arrythmias in Patients with Diabetes Mellitus Type 2.
    Orestis D Papagiannopoulos, Eleni I Georga, Jon Eneko Idoyaga Uribarrena, Eunate Arana Arri, Laura Lopez Perez, Miguel Rujas, Christian Temporale, Gloria Cea Sanchez, Jordi de Battle, Leandro Pecchia, Giuseppe Fico, Dimitrios I Fotiadis.
      Diabetes mellitus type 2 presents a significant public health challenge globally, predisposing affected individuals to a heightened risk of cardiovascular complications including arrhythmias. This paper investigates the intricate relationship between diabetes mellitus type 2 and arrhythmias, exploring underlying mechanisms and clinical correlations, including glycemic control, cardiovascular parameters, and lifestyle factors. Additionally, the study integrates multimodal data from wearable medical devices, offering a more comprehensive assessment. The design of the clinical study allowed us to analyze the concurrent variations in home health monitoring data (cardiological and glycemic indices) along with clinical and questionnaire data, aiming at identifying prognostic factors of cardiac arrhythmias in type 2 diabetes. Ensemble and boosting machine learning algorithms were utilized to identify key factors contributing to arrhythmia risk. Feature importance analysis identified elevated low-density lipoprotein LDL levels as a potential risk factor, consistent with existing literature.Clinical Relevance- The study highlights the critical role of lipid management in mitigating cardiovascular risks for patients with type 2 diabetes mellitus. The identification of elevated LDL levels as a significant predictor of arrhythmic events underscores the importance of incorporating lipid monitoring into routine care, advocating for aggressive lipid-lowering strategies.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11251843
  33. NPJ Metab Health Dis. 2025 Dec 02. 3(1): 46
    A deep learning-derived digital biomarker of dysglycemia and its association with genetic risk of type 2 diabetes.
    Jian Shao, Ying Pan, Jingnan Xue, Haonan Pan, Jing Wang, Shaoyun Li, Zedong Nie, Yuefei Li, Zijian Tian, Yu Zhao, Huyi Feng, Kaixin Zhou.
      Type 2 diabetes is a global health burden driven by genetic and environmental factors. Continuous glucose monitoring (CGM) can effectively guide lifestyle interventions in non-diabetic. However, predefined CGM metrics fail to fully capture the dysglycemic information contained in the high-dimensional time-series CGM data. This study employed deep learning to learn dysglycemia features from CGM data associated with diabetes and derived a digital biomarker of dysglycemia, validated against traditional dysglycemic biomarkers and diabetes polygenic risk score (PRS). Output of the deep learning model, called the deep learning-score, was significantly associated with multiple existing dysglycemic biomarkers and PRS of diabetes (P = 0.007). Moreover, existing CGM metrics were not associated with prevalent diabetes after adjusting for the deep learning-score, while the deep learning-score remained significantly associated with prevalent diabetes (P < 0.001) in a regression analysis. This digital biomarker demonstrated potential for providing dynamic feedback on dysglycemia and improving long-term intervention adherence.
    DOI:  https://doi.org/10.1038/s44324-025-00089-8
  34. J Transl Med. 2025 Dec 03. 23(1): 1374
    Multi-omics integrated analysis identifies causal risk factors and therapeutic targets for diabetic retinopathy.
    Jing Xu, Shuntai Chen, Mei Sun, Xi Chen, Zhenzhen Gu, Yige Zhang, Like Xie, Xiaofeng Hao.
       BACKGROUND: Diabetic retinopathy (DR) is the main cause of blindness worldwide, and its prevalence rate is constantly rising. More in-depth exploration of its risk factors and pathogenic mechanisms is needed.
    METHODS: This study systematically identified potential therapeutic targets for DR by evaluating causal effects of 16,989 genes and 2,923 proteins on DR/subtypes via two-sample Mendelian randomization (MR), validated with colocalization/Summary-data-based Mendelian randomization (SMR). National Health and Nutrition Examination Survey (NHANES) 1999-2010 cross-sectional data (weighted logistic/Restricted cubic spline (RCS)) pinpointed key risk factors; MR explored their links to DR subtypes. Bioinformatics (bulk and single-cell transcriptomics) analyzed molecular mechanisms of shared targets (gene expression, immune infiltration, pathway enrichment). Machine learning selected key targets for models. Finally, two-step mediation MR examined how targets regulate DR via risk factors.
    RESULTS: This study identified 64 core targets with causal links to DR. Subtype analysis revealed 2,128 causal genes and subtype-specific targets (e.g. 52 for background DR, 66 for proliferative DR). SMR validated these findings. NHANES data highlighted body mass index (BMI), stroke, hypertension (HBP), and C-reactive protein (CRP) as key DR risk factors, confirmed by MR. Transcriptomics identified 29 differentially expressed genes associated with both risk factors and DR, linked to immune cell regulation. Machine learning selected core targets (LY9, WWP2, etc.) and built a nomogram for DR risk prediction. Functional enrichment showed these targets enriched in chemokine/cytokine and immune-inflammatory pathways. Two-step mediation MR further revealed LY9, ARHGAP1, and WWP2 influence DR subtypes via regulating BMI, CRP, and HBP.
    CONCLUSION: This study systematically elucidates the key risk factors, potential molecular mechanisms, and core regulatory targets of DR through multi-omics integration, causal inference, and bioinformatics approaches. The results indicate that inflammation, immune dysregulation, and metabolic disorders play crucial roles in the pathogenesis of DR. Key genes such as LY9, ARHGAP1, and WWP2 could serve as potential intervention targets, offering theoretical foundations and strategic support for early warning and precision treatment of DR.
    Keywords:  Biomarkers; Diabetic retinopathy; Mediation effect; Mendelian randomization; Risk factors; Transcriptomics
    DOI:  https://doi.org/10.1186/s12967-025-07353-x
  35. Front Endocrinol (Lausanne). 2025 ;16 1620132
    Artificial intelligence in diabetes care: from predictive analytics to generative AI and implementation challenges.
    Mengqi Deng, Ruiye Yang, Xiaoran Zheng, Yaoqi Deng, Junyi Jiang.
      Generative artificial intelligence (GenAI) is transforming public health and medicine as well, in the form of disease surveillance, resource allocation and clinical decision making. Interventions to improve efficiency - multimodal predictive algorithms, federated learning platforms - reveal the internal contradictions of the system between algorithmic efficiency and fairness: speed of technical innovation and regulatory deficit, data flows without borders vs. ethical values of places. We present a three-dimensional governance structure for the topic covering the technical, institutional and ethical domains. From a technology point of view, explainability solutions and culturally-aware design align transparency with cultural sensibility. From an institution point of view, privacy-protecting data platforms and risk-based regulation align innovation with accountability. From an ethical point of view, incorporating local values and disbursing AI dividends sustain equitable health outcomes. There are still challenges that demand the utmost priority, including the algorithmic prejudice, the data imperialism and the opacity in medical AI decision making. Future priorities include the development of broader measurement tools that integrate clinical impact, equity, and societal impact; the development of transnational governance institutions to mitigate concerns relating to data sovereignty; and the development of forms of participatory design between designers, practitioners, and populations. A balance between technical creativity, visionary policy-making, and caring leadership to advocate for human-centered healthcare will provide us with trusted AI ecosystems. Technical excellence alone cannot guarantee success unless fairness and accessibility, social responsiveness, and justice for future global health is guaranteed.
    Keywords:  algorithmic fairness; data colonialism; ethical machine learning; explainable AI; generative artificial intelligence; health equity; medical AI governance; public health informatics
    DOI:  https://doi.org/10.3389/fendo.2025.1620132
  36. J Med Internet Res. 2025 Dec 05. 27 e79283
    Artificial Intelligence-Enabled Mobile Health Intervention (mDiabetes) to Reduce Diabetes Risk Behaviors in Rural India: Quasi-Experimental Pre-Post Study.
    Joshua Chadwick, Nidhi Jaswal, Janani Surya, Chandru Sivamani, Varun Ramesan, Nalini Saligram.
       BACKGROUND: India faces a dual burden of diabetes and prediabetes. Although mobile health (mHealth) interventions have shown promise in promoting healthy lifestyle changes, most interventions deploy generic, "one-size-fits-all" messages that do not consider individual behavioral patterns, motivational states, or changing needs over time.
    OBJECTIVE: This formative evaluation study aimed to assess the effectiveness of an artificial intelligence (AI)-enabled, personalized mHealth messaging intervention (mDiabetes) compared to traditional, nonpersonalized mHealth messaging in promoting engagement with diabetes risk reduction behaviors among adults in Gulbarga, Karnataka, South India.
    METHODS: A quasi-experimental pre-post study was conducted among adults without diabetes (N=1048). Participants were divided into intervention and control groups. The control group received static diabetes prevention messages via WhatsApp, while the intervention group received customized messages twice a week based on individual feedback through reinforcement learning algorithms. Data on demographics, diabetes knowledge, and lifestyle behaviors were collected via home interviews. Chi-square tests and t tests were performed to assess group differences. Intervention effects were evaluated using multivariable logistic regression for binary outcomes and ANCOVA for continuous outcomes. Adjusted odds ratios (aORs) with 95% CIs were reported, and Bonferroni correction was applied for multiple comparisons.
    RESULTS: A total of 1048 (96.9%) participants (n=661, 63.1%, female) completed the 6-month follow-up. At endline, no significant between-group differences were observed for primary outcomes. Both groups had similar odds of meeting the physical activity goal (≥30 minutes/day) at endline (aOR 1.0, 95% CI 0.7-1.3, P=.74). Baseline activity (aOR 2.1, 95% CI 1.5-3.1, P<.001) and age >50 years (aOR 3.8, 95% CI 1.6-9.3, P=.003) were significant predictors of endline physical activity, while employment was associated with lower odds of physical activity (aOR 0.2, 95% CI 0.1-0.3, P<.001). Daily fruit intake was modestly higher in the intervention group (aOR 1.4, 95% CI 0.8-2.3, P=.24), and participants aged 26-35 years had higher odds of daily fruit intake (aOR 4.7, 95% CI 1.9-11.8, P=.001), while employment was associated with lower odds (aOR 0.3, 95% CI 0.1-0.8, P=.02). The mean BMI difference at endline was -0.0 kg/m² (95% CI -0.6 to 0.5, P=.95), and baseline BMI was a strong predictor of endline BMI (P<.001). Exploratory behavioral outcomes revealed no significant differences: stair use (aOR 0.9, 95% CI 0.7-1.4, P=.79), walking for chores (aOR 2.4, 95% CI 1.0-6.1, P=.06), helping with household chores (aOR 1.0, 95% CI 0.4-2.3, P=.94), and farm work (aOR 1.3, 95% CI 0.9-1.8, P=.19).
    CONCLUSIONS: Both AI-enabled and traditional mHealth interventions have similar effectiveness in promoting diabetes prevention behaviors in rural India. Simple, well-designed mHealth interventions delivered through an accessible platform like WhatsApp can achieve meaningful behavior change without the need for complex AI technology. The comparable effectiveness suggests the potential for scalable, cost-effective, equitable diabetes prevention strategies in resource-limited settings.
    Keywords:  AI; AI-enabled mHealth; India; artificial intelligence; community intervention; diabetes prevention; mHealth; mobile health; physical activity; rural health
    DOI:  https://doi.org/10.2196/79283
  37. Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025 1-7
    Diabetes Prevention Chatbots to Help Dietary Self-management for the Chinese Population.
    Sihan Jiang, Ming-Chun Huang.
      Diabetes has become an increasingly severe problem in China in recent years. Although some methods have been conducted to promote diabetes prevention before, they have certain limitations. To better activate diabetes management in China, this research proposes an AI-driven web application that offers personalized suggestions for diabetes prevention in the Chinese population. Its core is fine-tuning a large language model based on a tailored training dataset containing conversation prompts about diabetes lifestyle prevention suggestions. Its novelty includes the training dataset building on a self-defined diabetes prevention guideline, focusing on the Chinese lifestyle and, particularly, dietary habits. Therefore, it helps bridge gaps in existing prevention tools, such as the lack of personalization and cultural relevance. The system enables users to interact with AI in real time to receive advice and download chatting histories as well. This study demonstrates the feasibility of using AI to enhance early prevention strategies, contributing to advancing AI applications in healthcare. Its future implications may include expanding the app's features for broader lifestyle management, as well as integrating with a community feedback mechanism and relevant healthcare systems to improve Chinese diabetes prevention.Clinical relevance-This project might be of interest to practicing clinicians, since it had the potential to improve the quality and accessibility of current Chinese diabetes prevention. In addition, if people's diabetes prevention self-management is improved, the burden of clinicians may be reduced to some extent.
    DOI:  https://doi.org/10.1109/EMBC58623.2025.11254311
This page is being maintained by Thomas Krichel. It was last updated on 2025‒12‒07 at 0:25.