bims-aukdir 2026-10-25 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2026–10–25
nineteen papers selected by
Mott Given

Artificial intelligence in diabetes: Clinical applications across the disease continuum.
Artificial intelligence in diabetic retinopathy: From deep learning to foundation models, clinical translation, and real-world deployment.
Early prediction of gestational diabetes mellitus using deep learning applied to routinely collected clinical data.
Comment on "A machine learning-based framework for predicting type 2 diabetes mellitus using hematological indices".
From validation to public health impact: implementing equitable AI-enabled diabetic retinopathy screening.
Application of machine learning prediction models for in-hospital mortality in diabetic sepsis patients: a study based on the MIMIC IV database.
Large language models as sources of diabetic retinopathy patient education: a comparative evaluation of safety, health information quality, and readability.
Fundus camera and artificial intelligence for diabetic retinopathy screening in a São Paulo shelter.
Analysis of barriers to self-monitoring of blood glucose among community-dwelling elderly patients with diabetes mellitus: a cross-sectional study integrating artificial intelligence-based risk prediction.
Detection and Risk Prediction of Diabetes Related Kidney Disease Using Retinal Images: A Systematic Review and Meta-Analysis.
Predicting visual impairment in type 2 diabetes using key performance indicators identified by the Black Widow and Whale Optimization Algorithms: A case study in Saudi Arabia.
A machine learning-driven and explainable prediction model for in-hospital mortality in patients with diabetes mellitus and congestive heart failure: integrating XGBoost and SHAP approaches.
Machine Learning with Social Determinants of Health for Cross-Sectional Diabetes Classification: Development and Temporal Validation.
Multimodal Deep Learning for Wound Healing Prediction.
Outcome-specific value of longitudinal clinical information and model complexity for dynamic prediction of type 2 diabetes-related outcomes: a pooled landmarking study.
Classification of gestational diabetes mellitus placental histopathological images based on MONAI and improved GAMatrix-YOLOv8.
Development and validation of a model using routine clinical features to improve diabetes diagnosis.
Feasibility of an Activity Control System in Patients with Type 2 Diabetes: Results from a Non-Randomised Controlled Pilot Study.
Views and conditional acceptance of telehealth and artificial intelligence for lifestyle modification among individuals with prediabetes: A qualitative descriptive study.

J Diabetes Complications. 2026 Oct 19. pii: S1056-8727(26)00183-2. [Epub ahead of print]40(12): 109438

Artificial intelligence in diabetes: Clinical applications across the disease continuum.

Serena Shih Yin, Viola Viola, Paolo Pozzilli.

  Artificial intelligence (AI) is increasingly shaping the landscape of diabetes mellitus (DM) care by enabling data-driven approaches to prediction, diagnosis, classification, and management. The growing availability of continuous biometric data from wearable technologies, combined with advances in machine learning, has created new opportunities to overcome limitations of traditional diabetes metrics and population-based care models. This review summarises current clinical applications of AI in DM, with a focus on translational relevance and, critically, on the level of validation and clinical readiness of each application. AI-based models have demonstrated improved performance in predicting DM onset and diabetes-related complications by integrating electronic health records, imaging data, and longitudinal clinical variables. In diagnosis and screening, AI offers non-invasive and scalable tools, including automated retinal screening systems and emerging digital biomarkers. Data-driven clustering approaches have further refined diabetes classification. In diabetes management, AI supports personalised glucose monitoring, insulin titration, pharmacological decision-making, and lifestyle interventions. Crucially, these applications differ markedly in maturity: autonomous retinal screening and closed-loop insulin delivery are approaching or have reached routine clinical use, whereas voice-based biomarkers and simulation-evaluated lifestyle systems remain proof-of-concept. Addressing challenges of generalisability, data quality, interpretability, and ethical governance through robust validation and explainable AI will be essential to realise the full potential of AI in precision, patient-centred diabetes care.

Keywords:  Artificial intelligence; Diabetes mellitus; Digital health technologies; Machine learning; Precision diabetes care

DOI:  https://doi.org/10.1016/j.jdiacomp.2026.109438
Surv Ophthalmol. 2026 Oct 21. pii: S0039-6257(26)00152-9. [Epub ahead of print]

Artificial intelligence in diabetic retinopathy: From deep learning to foundation models, clinical translation, and real-world deployment.

Basanta Haobijam, Tourangbam Harishore Singh, Puyam Tondonba Singh, Wei-Chi Wu, Eugene Yu-Chuan Kang, Sinam Ajitkumar Singh, Anzila Laikhuram, Cheng-Ming Huang, Wei-Che Huang, Peng-Po Chun, Zhou-You Qin.

  Diabetic retinopathy (DR) remains a leading cause of preventable vision impairment worldwide, with its burden increasing in parallel with the global diabetes pandemic. As a progressive microvascular disorder with neurovascular unit involvement, DR is driven by retinal ischemia, inflammation, and pathological neovascularization, while diabetic macular edema (DME) continues to represent a principal cause of vision loss. This structured narrative review provides a critical synthesis of advances in diabetic retinopathy diagnosis, retinal imaging, treatment strategies, and artificial intelligence (AI)-enabled screening published from January 2020 through March 2026, while selectively incorporating earlier landmark studies to provide clinical and methodological context. We summarize contemporary screening pathways and emerging retinal imaging modalities, including ultra-widefield photography (up to ~200° field of view), optical coherence tomography (OCT), and OCT angiography, examining how complementary structural and vascular biomarkers are reshaping disease severity assessment and DME characterization. We then review the evolution of AI methodologies, highlighting a transition from task-specific convolutional neural network pipelines toward transformer-based and self-supervised representation learning, including retina-specific foundation models designed to reduce annotation burden and improve robustness and generalizability under domain shift. Beyond algorithmic performance on curated benchmarks, we emphasize key translational requirements for clinical adoption, including model calibration, uncertainty-aware triage, image quality control, and external validation across imaging devices and diverse populations. Recent prospective and implementation studies across diverse healthcare systems demonstrate that the real-world effectiveness of AI-enabled DR screening depends not only on diagnostic accuracy but also on image gradability, clinically appropriate operating thresholds, uncertainty management, external validation, referral capacity, and integration with established care pathways. Evidence from the Asia-Pacific region provides important examples of these implementation challenges in heterogeneous and resource-constrained screening environments. Specific attention is given to the identification and mitigation strategies for fundus pigmentation-related performance disparities and the governance of autonomous versus hybrid human-AI screening models within evolving regulatory frameworks. Finally, we examine emerging work in oculomics for systemic risk stratification and propose a pragmatic clinical readiness framework centered on validation hierarchy, safety monitoring, and workflow integration. Particular emphasis is placed on the translational readiness of foundation models, multimodal retinal intelligence, and emerging technologies for large-scale screening, therapeutic decision-making, and longitudinal disease monitoring.

Keywords:  Artificial intelligence; Diabetic retinopathy; Real-world deployment; Retinal imaging; clinical translation; foundation models

DOI:  https://doi.org/10.1016/j.survophthal.2026.10.008
Front Endocrinol (Lausanne). 2026 ;17 1941799

Early prediction of gestational diabetes mellitus using deep learning applied to routinely collected clinical data.

Gabriel Cubillos, Max Monckeberg, Sebastian E Illanes, Claudio A Perez.

   Background: Prediction of gestational diabetes mellitus (GDM) in early pregnancy enables timely, targeted interventions to reduce risks for both the mother and fetus. We aimed to improve early GDM prediction and clinical risk stratification using a novel deep learning (DL) approach that transforms tabular clinical data into images using a genetic algorithm.
Methods: A dataset of 1,611 pregnancies was used. The dataset contained routinely collected early-pregnancy data, including clinical measurements, blood tests, and medical history. A novel method to transform tabular data into images was applied to enable the use of deep learning models for GDM prediction. A genetic algorithm optimized the spatial arrangement of variables during image generation by determining their spatial positions. Interpretability was assessed using gradient-weighted class activation mapping (Grad-CAM) to identify variables the model preferentially attended to for positive and negative predictions.
Results: Multiple model configurations were evaluated by varying the spatial arrangement of variables within the transformation. Prioritizing high sensitivity, one model achieved an accuracy of 0.86, sensitivity of 0.82, specificity of 0.87, and an AUC-ROC of 0.83. Another model achieved an accuracy of 0.79, sensitivity of 0.92, specificity of 0.77, and an AUC-ROC of 0.85. Compared with previously published machine learning models using the same dataset, our approach demonstrated improved performance. Our best models increased specificity by an average of 13.93 percentage points compared with traditional ML models at matched sensitivity levels, thereby reducing false-positive classifications that may lead to additional clinical evaluation. Grad-CAM analysis revealed that attention patterns of the model aligned with established clinical risk factors for GDM, with first trimester fasting glycemia and parity preferentially attended for positive predictions and maternal weight for negative predictions.
Conclusions: This study demonstrates that transforming tabular clinical data into image representations, optimized using a genetic algorithm, improves early GDM prediction. Representing tabular variables as images enables more effective application of DL models, which have demonstrated enhanced pattern recognition in image-based tasks compared with traditional machine learning methods. Performance improvements over state-of-the-art approaches were achieved. Accurate DL models can be developed using a limited number of input variables, supporting early identification of women at increased risk of GDM and facilitating timely clinical management.

Keywords:  clinical decision support; convolutional neural networks; deep learning models; early risk prediction; explainable artificial intelligence; genetic algorithms; gestational diabetes mellitus; risk stratification

DOI:  https://doi.org/10.3389/fendo.2026.1941799
Am Heart J Plus. 2026 Nov;71 100926

Comment on "A machine learning-based framework for predicting type 2 diabetes mellitus using hematological indices".

Addanki Ramya, Shagam Geetha, M V N Srujan Manohar.



Keywords:  Hematological indices; Machine learning; Predictive modeling; Random Forest; Risk prediction; Type 2 diabetes mellitus

DOI:  https://doi.org/10.1016/j.ahjo.2026.100926
Front Public Health. 2026 ;14 1976572

From validation to public health impact: implementing equitable AI-enabled diabetic retinopathy screening.

Jing Chen, Bin Chen.

  Diabetic retinopathy (DR) remains a preventable cause of vision loss as diabetes prevalence rises worldwide. Artificial intelligence (AI)-enabled retinal screening has progressed from retrospective validation to prospective and real-world deployment, with recent reviews showing high diagnostic performance across fundus photography, smartphone-based systems, and autonomous platforms. Yet diagnostic accuracy alone does not ensure public health benefit. Population impact can be lost when eligible people are not reached, images are ungradable, results are delayed, referrals are incomplete, treatment capacity is insufficient, or disadvantaged groups face financial, geographic, or digital barriers. Unlike reviews centered mainly on diagnostic accuracy or economic performance, this Mini Review reframes AI-enabled DR screening as a population-health intervention spanning detection through completion of effective eye care. We synthesize evidence on diagnostic validity, implementation, economics, referral completion, care cascades, patient-centered outcomes, equity, and governance. We argue that effectiveness and cost-effectiveness are properties of the complete screening-to-treatment pathway, not the algorithm alone, and propose a four-domain framework encompassing technical validity, service integration, population impact, and equity and sustainability. Future research should prioritize comparative implementation studies and outcome sets that capture screening reach, gradability, referral completion, treatment initiation, time to treatment, avoidable vision loss, vision-related quality of life, acceptability, costs, and subgroup equity. The policy question is no longer whether AI can detect DR, but whether health systems can translate automated detection into equitable completion of effective eye care at scale.

Keywords:  artificial intelligence; diabetic retinopathy; health equity; implementation science; public health; quality of life; referral completion; screening

DOI:  https://doi.org/10.3389/fpubh.2026.1976572
BMC Med Inform Decis Mak. 2026 Oct 22. pii: 426. [Epub ahead of print]26(1):

Application of machine learning prediction models for in-hospital mortality in diabetic sepsis patients: a study based on the MIMIC IV database.

Zongxin Shen, Haijiao Zhang, Yu Wang.

   BACKGROUND: Sepsis complicated by diabetes mellitus represents a significant clinical challenge in intensive care units (ICUs), associated with substantial morbidity and mortality. Accurate prognostication of in-hospital mortality in this complex patient population is crucial for optimizing clinical decision making, resource allocation, and patient management strategies.
METHODS: We conducted a retrospective cohort study utilizing data from the Medical Information Mart for Intensive Care (MIMIC-IV) database. The study focused on adult ICU patients with concurrent diagnoses of sepsis and diabetes mellitus. Six machine learning algorithms were evaluated for their predictive performance: Logistic Regression (LR), K-Nearest Neighbors (KNN), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), XGBoost and LightGBM. Additionally, we assessed the prognostic capability of the Sequential Organ Failure Assessment (SOFA) score. Model discrimination was quantified using the area under the receiver operating characteristic curve (AUC). To enhance model interpretability, we employed SHAP (SHapley Additive exPlanations) values to elucidate the relative importance of predictor variables.
RESULTS: The study cohort comprised 7716 patients, with 1357 (17.6%) experiencing in-hospital mortality. Among the machine learning models evaluated, GBDT exhibited superior predictive performance achieving an AUC of 0.792 (95% CI: 0.769-0.814). This performance surpassed that of other machine learning algorithms and the conventional SOFA score. SHAP analysis provided valuable insights into the hierarchical importance of various clinical and demographic features contributing to the GBDT model's predictions.
CONCLUSION: Our findings demonstrate the potential of machine learning approaches in accurately predicting in-hospital mortality among ICU patients with sepsis and diabetes mellitus. These advanced analytical techniques outperform traditional scoring systems such as SOFA. The integration of SHAP values enhances model interpretability, potentially facilitating the adoption of these predictive tools in clinical practice. Further multi-center validation studies and prospective trials are warranted to confirm the generalizability of these results and to explore the clinical impact of implementing such models in diverse healthcare settings.

Keywords:  Diabetes mellitus; Intensive care unit; Machine learning; Mortality prediction; Sepsis

DOI:  https://doi.org/10.1186/s12911-026-03580-4
Front Public Health. 2026 ;14 1945318

Large language models as sources of diabetic retinopathy patient education: a comparative evaluation of safety, health information quality, and readability.

Zhiqi Chen, Wen Liu, Jing Zhang, Tian Zheng, Ying Li, Qing Bao, Pingting Ji, Hongmei Yao, Ying He.

   Background: Patients increasingly use large language models (LLMs) to seek medical information, but their performance in answering diabetic retinopathy (DR) questions remains insufficiently characterized.
Objective: To evaluate the safety, accuracy, patient-centered communication, information quality, transparency, and readability of diabetic retinopathy information generated by widely accessible large language models.
Methods: Fifty standardized DR patient questions were submitted to ChatGPT, Gemini, DeepSeek, Doubao, Copilot, and Perplexity. Responses were independently evaluated by five raters using guideline-based reference answer elements. Rater-based outcomes included Safety, Accuracy, Empathy, DISCERN, EQIP, JAMA benchmark criteria, and Global Quality Score (GQS). Readability was assessed using the Automated Readability Index, Coleman-Liau Index, Flesch-Kincaid Grade Level, Gunning Fog Index, SMOG, and Flesch Reading Ease Score. Inter-rater agreement and between-model differences were analyzed using appropriate agreement statistics and paired nonparametric tests.
Results: Inter-rater agreement was moderate for Safety (Fleiss' κ = 0.581) and good for non-binary outcomes, with ICC (2,1) values ranging from 0.806 to 0.870. Most responses were classified as safe, with safe response rates ranging from 86.0 to 94.0%, and no significant between-model difference was observed. Accuracy scores were generally high across models, although between-model differences were statistically significant with a small effect size. Empathy varied more noticeably, with Copilot scoring highest and Doubao and Perplexity scoring lower. Perplexity achieved the highest DISCERN and JAMA scores, while Copilot performed best on EQIP. GQS differences were small. Readability differed significantly across models, with Gemini generally producing the most readable responses; however, all models exceeded the recommended sixth-grade reading level, and all Flesch Reading Ease Scores were below the recommended threshold.
Conclusion: No single LLM performed best across all domains. Although the evaluated models generally provided safe and accurate responses to DR patient questions, limitations remained in safety escalation, empathy, transparency, and readability. LLM-generated DR information may support general patient education but should not replace professional ophthalmic consultation.

Keywords:  diabetic retinopathy; digital health (eHealth); health communication (MESH); health literacy; large language models (LLM); patient education; patient safety-standards

DOI:  https://doi.org/10.3389/fpubh.2026.1945318
Rev Assoc Med Bras (1992). 2026 ;pii: S0104-42302026000902228. [Epub ahead of print]72(9): e20260862

Fundus camera and artificial intelligence for diabetic retinopathy screening in a São Paulo shelter.

Julia Gimenes da Silva Dias, Gabriel Costa de Andrade, João Victor Tenório Lossio de Macedo, Andrezza Soldera Graziano.

OBJECTIVE: The aim of this study was to evaluate the applicability of a smartphone-based portable fundus camera for diabetic retinopathy screening in elderly residents of a specialized shelter in São Paulo, Brazil, considering image acquisition by non-ophthalmologist professionals and remote case classification.
METHODS: Prospective pilot interventional study at the Centro de Acolhida Especial para Idosos Morada São João. Thirty-eight participants (76 eyes) underwent retinal imaging using the Eyer® portable fundus camera (Phelcom Technologies, Brazil) by trained non-ophthalmologist professionals. Images were remotely evaluated by a retinal specialist and compared to the EyerMaps® artificial intelligence system.
RESULTS: Mean age was 71.6±7.0 years; 78.9% male. Fifteen participants (39.5%) had confirmed diabetes mellitus. The specialist identified abnormalities in 70 of 76 eyes (92.1%), most frequently glaucoma suspicion, media opacity suggestive of cataract, chorioretinal changes, and age-related macular degeneration signs; no diabetic retinopathy was identified. Against the specialist's assessment of any ocular abnormality as the reference standard, the artificial intelligence system classified 22 eyes as abnormal, yielding sensitivity 25.7% (95%CI 16.9-37.0), specificity 33.3% (95%CI 9.7-70.0), positive predictive value 81.8% (95%CI 61.5-92.7), negative predictive value 3.7% (95%CI 1.0-12.5), and overall accuracy 26.3% (95%CI 17.7-37.2).
CONCLUSION: The Eyer® device demonstrated operational feasibility for ocular screening by trained non-ophthalmologists. The high prevalence of media opacities and non-diabetic retinopathy conditions compromised artificial intelligence performance, highlighting the need for broader screening strategies in elderly and socially vulnerable populations.

DOI: https://doi.org/10.1590/1806-9282.20260862
BMC Endocr Disord. 2026 Aug 10. pii: 327. [Epub ahead of print]26(1):

Analysis of barriers to self-monitoring of blood glucose among community-dwelling elderly patients with diabetes mellitus: a cross-sectional study integrating artificial intelligence-based risk prediction.

Meiyan Wang, Yue Ying, Yuan Han, Jiaxiang Zhou.

   BACKGROUND: Self-monitoring of blood glucose (SMBG) is essential for diabetes management, yet elderly patients face multiple barriers that compromise adherence and glycaemic control. This study aimed to assess barriers to SMBG among community-dwelling elderly patients with diabetes mellitus, identify associated factors, and develop an artificial intelligence (AI)-based predictive model for high-risk individuals.
METHODS: A cross-sectional study was conducted at Chun'an First People's Hospital from March to August 2025. Two hundred elderly patients (≥ 60 years) with diabetes were enrolled using convenience sampling. Data were collected using validated instruments: a sociodemographic questionnaire, SMBG Barriers Scale, Flourishing Scale (FS), and Digital Health Literacy Scale (DHLS). Statistical analyses included descriptive statistics, Pearson correlations, and multiple linear regression. For predictive modelling, the dataset was split into training (80%) and testing (20%) sets. Random Forest, XGBoost, and Support Vector Machine algorithms were developed and evaluated using accuracy, precision, recall, F1-score, and AUC-ROC. Model interpretability was assessed with SHAP values.
RESULTS: The mean SMBG Barriers score was 47.6 ± 9.8. Significant negative correlations were found between SMBG barriers and FS (r = - 0.42) and DHLS (r = - 0.51) scores. Multiple regression showed DHLS (β = -0.38), Flourishing score (β = -0.25), education, and diabetes education as protective factors, while complications and living alone were risk factors (R² = 0.462). The Random Forest model achieved the best performance (accuracy 88.5%, AUC 0.921). SHAP analysis identified digital health literacy, psychological well-being, and AI acceptance as the strongest predictors.
CONCLUSION: This study highlights the importance of psychosocial and digital factors in SMBG barriers and demonstrates the effectiveness of explainable AI for risk prediction in elderly diabetic patients. Targeted interventions addressing modifiable factors could improve self-management.
CLINICAL TRIAL REGISTRATION: This study is a cross-sectional observational study and was not registered as a clinical trial.

Keywords:  Artificial intelligence; Barriers; Diabetes mellitus; Digital health literacy; Elderly patients; Self-monitoring of blood glucose

DOI:  https://doi.org/10.1186/s12902-026-02438-0
J Diabetes. 2026 Oct;18(10): e70271

Detection and Risk Prediction of Diabetes Related Kidney Disease Using Retinal Images: A Systematic Review and Meta-Analysis.

Samuel Ato Gyasi Otabir, Teketo Kassaw Tegegne, Paul Nderitu, Jiangtao Wang, Jonathan E Shaw, David Owens, Rebecca Thomas, Robyn Jennifer Tapp, Ralph Maddison.

   AIM: This study aimed to determine the effectiveness of deep learning models for detecting and predicting diabetes related kidney disease (DRKD) using two-dimensional retinal images.
METHODS: A systematic review was conducted across eight databases from inception to July 14, 2026. Eligible cross-sectional and longitudinal studies assessed deep learning models for DRKD detection or prediction using retinal images alone, clinical data alone, or hybrid models. Data on study design, model development, validation, and performance metrics were extracted, with study quality appraised using QUADAS-2 and post hoc, PROBAST+AI. Results were synthesized narratively, with random effects meta analysis for hybrid models.
RESULTS: Of 367 studies, six met inclusion criteria. Three cross-sectional studies of hybrid models contributed to the meta-analysis for DRKD detection, yielding pooled AUC 0.88, sensitivity 0.80, and specificity 0.81, though heterogeneity was high (I2 93.7%-99.8%). Three further studies were synthesized narratively rather than pooled: two cross-sectional studies used more severe or narrower CKD case definitions (AUC 0.86 and 0.96 respectively), and only one longitudinal study predicted the 5-year risk of DRKD (AUC 0.81, sensitivity 0.70, and specificity 0.77). External validation showed hybrid models consistently outperformed retinal-only (ΔAUC 0.03-0.11), but their advantage over clinical-only models was inconsistent.
CONCLUSION: Deep learning models may be useful for detecting and predicting DRKD from retinal images, with or without clinical data. However, the small number of heterogenous studies restricts generalizability. Future work should prioritize robust model development, and validation using large, diverse populations.

Keywords:  artificial intelligence; deep learning; detection; diabetes kidney disease; fundus photographs; prediction; retinal images

DOI:  https://doi.org/10.1111/1753-0407.70271
PLoS One. 2026 ;21(10): e0361022

Predicting visual impairment in type 2 diabetes using key performance indicators identified by the Black Widow and Whale Optimization Algorithms: A case study in Saudi Arabia.

Faten Al-Hussein, Laleh Tafakori, Mali Abdollahian, Khalid Al-Shali.

Visual impairment (VI) is a severe and prevalent complication of type 2 diabetes (T2D), particularly in high-risk populations, underscoring the critical need for early predictive strategies. A retrospective cross-sectional study was conducted to develop and validate machine learning (ML) predictive models for VI among patients with T2D in Saudi Arabia. To the best of our knowledge, this is the first study to employ WOA and BWO as feature selection techniques to identify the most informative predictors of VI among Saudi patients with T2D. Predictive models were subsequently developed and compared using both clinically recommended features and algorithmically selected features derived from WOA and BWO to determine the optimal feature set for predictive performance. Medical records of 1,000 T2D patients (555 VI [55.5%]; 445 non-VI [44.5%]) from King Abdulaziz University Hospital were analyzed. Models were developed using an 80/20 stratified train-test split, with 5-fold cross-validation performed on the training set for internal validation. Patients were divided into two groups: those with HbA1c ≥ 6.5% (full dataset) and a clinically relevant high-risk subgroup with HbA1c > 8% (reduced dataset), which incorporated 19 recommended demographic, lifestyle, and lipid profile features. Models utilizing the reduced dataset consistently achieved superior performance across all algorithms. Among them, the WOA-CatBoost model demonstrated the highest performance (Accuracy = 0.95, AUC = 0.96, 95% CI: 0.94-0.98) using only 9 features, outperforming both the BWO-Random Forest (BWO-RF) model (Accuracy = 0.92, AUC = 0.90, 95% CI: 0.87-0.93) with 10 features and the baseline Multi-Layer Perceptron (MLP) model (Accuracy = 0.91, AUC = 0.94, 95% CI: 0.91-0.97) with 13 features. Bonferroni-corrected paired t-tests based on five-fold cross-validation accuracies (α = 0.017) confirmed a significant performance advantage of WOA-CatBoost over BWO-RF (t = 10.17, p < 0.001), supporting WOA-CatBoost as the overall best-performing model. Key predictors identified across all three best-performing models included Hemoglobin A1c (HbA1c), body mass index (BMI), age, occupation, triglycerides (TG), type of food, and smoking. These findings demonstrate that WOA-CatBoost provides a clinically actionable, parsimonious predictive tool for VI risk stratification in T2D patients within the Saudi Arabian healthcare context.

DOI: https://doi.org/10.1371/journal.pone.0361022
Front Public Health. 2026 ;14 1951635

A machine learning-driven and explainable prediction model for in-hospital mortality in patients with diabetes mellitus and congestive heart failure: integrating XGBoost and SHAP approaches.

Jingru Hu, Shiqiong Su, Dongdong Liu.

   Background: Patients with congestive heart failure (CHF) and type 2 diabetes mellitus (T2DM) are at elevated risk of in-hospital mortality, particularly in intensive care units (ICUs). Conventional scoring systems may not fully capture the clinical complexity of such comorbidities. This study aimed to construct an interpretable machine learning model to predict in-hospital mortality in this high-risk group.
Methods: A retrospective study included 573 ICU patients with CHF and T2DM admitted to three hospitals in Henan, China (Jan 2025-Jun 2026). Seven predictors-body mass index (BMI), myocardial infarction, mild liver disease, SOFA score, LODS score, white blood cell (WBC) count, and serum creatinine-were identified via stepwise logistic regression. XGBoost and four other machine learning algorithms (KNN, SVM, CatBoost, AdaBoost) were trained and evaluated. Model performance was assessed using AUC, accuracy, sensitivity, specificity, and F1-score. SHAP (SHapley Additive exPlanations) was used to interpret XGBoost predictions.
Results: XGBoost achieved the best performance (AUC 0.917, sensitivity 86.8%, specificity 96.6%, accuracy 74.8%). SHAP analysis indicated that mild liver disease, LODS score, BMI, WBC count, and SOFA score were the top contributors, jointly explaining over 77% of the predictive power. Myocardial infarction, liver disease, elevated WBC, and higher SOFA/LODS scores increased mortality risk, while higher BMI and serum creatinine were protective-supporting the obesity paradox and sarcopenia hypothesis.
Conclusion: An XGBoost model with SHAP interpretation provides accurate, explainable prediction of in-hospital mortality in ICU patients with CHF and T2DM, offering a valuable tool for early risk stratification and clinical decision-making.

Keywords:  diabetes mellitus; heart failure; hospital mortality; inpatients; machine learning

DOI:  https://doi.org/10.3389/fpubh.2026.1951635
Int J Med Sci. 2026 ;23(11): 3323-3335

Machine Learning with Social Determinants of Health for Cross-Sectional Diabetes Classification: Development and Temporal Validation.

Woon-Man Kung, Hsien Chang Yeh, Abel Po-Hao Huang, Md Mohaimenul Islam, Chieh-Chen Wu, Yao-Chin Wang.

   Aims: Our study aimed to determine whether incorporating social determinants of health (SDOH) improves cross-sectional classification of prevalent diagnosed diabetes and to quantify the extent to which model performance is attributable to variables that may occur downstream of diabetes diagnosis.
Methods: We analyzed data from the 2023 Behavioral Risk Factor Surveillance System (BRFSS) as the development cohort and the 2024 BRFSS as an independent temporal validation cohort. Analyses were restricted to adults aged 35 years or older residing in states that administered the SDOH module. The outcome was self-reported diagnosed diabetes. Before model development, 30% of the 2023 sample was reserved as a test set. Imputation, encoding, feature processing, and hyperparameter tuning were performed within nested cross-validation. Histogram-based gradient boosting and penalized logistic regression were evaluated across six pre-specified nested predictor specifications. Final models were applied unchanged to the 2024 cohort without refitting, re-tuning, or threshold recalibration. Performance estimates incorporated BRFSS survey weights.
Results: The development cohort comprised 179,752 adults (weighted prevalence 16.97%) and the validation cohort 161,092 (18.07%). Discrimination ranged from 0.75 to 0.81 across algorithms, with gradient boosting and penalized logistic regression indistinguishable (0.811 versus 0.808). The full model achieved an area under the receiver operating characteristic curve (AUROC) of 0.811 (95% CI 0.801 - 0.822) and transported to the validation cohort with a calibration slope of 1.040. Adding nine SDOH items changed discrimination by 0.006 (95% CI 0.004 - 0.008), with maximum net benefit gain under 0.003; discrimination rose from 0.743 to 0.811 only when healthcare utilization and comorbidity measures were added. Several SDOH nonetheless showed robust adjusted associations, including food insecurity (OR 1.27 - 1.39) and supplemental nutrition assistance program receipt (OR 1.34, 95% CI 1.11 - 1.61). At a fixed threshold, sensitivity ranged from 0.62 to 0.89 across age groups.
Conclusions: SDOH were independently associated with prevalent diagnosed diabetes but provided little incremental value beyond demographic, behavioral, and healthcare access measures, a finding that was consistent across consecutive survey cycles. Most of the observed discrimination was attributable to healthcare utilization and comorbidity variables that may plausibly reflect consequences of an established diagnosis rather than characteristics preceding disease recognition.

Keywords:  database research; disease prevention; risk prediction; type 2 diabetes

DOI:  https://doi.org/10.7150/ijms.137605
Eur J Pharm Sci. 2026 Oct 22. pii: S0928-0987(26)00267-8. [Epub ahead of print] 107693

Multimodal Deep Learning for Wound Healing Prediction.

Tatu Rojalin, Aleksi Kröger, Elle Koivunotko, Benjamin Pörhö, Timo Laaksonen, Jouni Partanen, Chris Pridgeon, Mari Madetoja, Antti Airola, Jan Akmal, Marjo Yliperttula.

  Chronic wounds, particularly diabetic foot ulcers, heal unpredictably and impose substantial healthcare costs. Deep learning (DL) systems that integrate complementary data streams could support preclinical studies by assessing wound status and therapy efficacy reliably in small-n studies. We developed a three-branch neural network combining convolutional feature extraction for red-green-blue (RGB) images, infrared thermography, and wound-area measurements to classify healing stage and infer treatment response in diabetic mouse wound model. In descriptive ablation analysis, the model integrating RGB and wound-size inputs achieved higher accuracy than the RGB-only configuration (0.75 vs. 0.70), suggesting that explicit size context provided useful complementary information. Thermal images appeared qualitatively informative during early healing but contributed limited additional signal later, likely reflecting subtle inflammatory gradients or angiogenesis in healthy murine skin; this observation motivated a constrained weighted-fusion strategy. In the development-stage comparison, static fusion with tunable fixed weights yielded the highest observed performance (∼0.85 accuracy on the development test subject). Grad-CAM visualizations showed that predictions arose from biologically relevant regions rather than image artifacts. Experiments without subject-level partitioning produced inflated metrics, underscoring the need for grouping by subject to ensure generalizable evaluation. Nested Leave-One-Group-Out cross-validation with KerasTuner-based hyperparameter tuning yielded a mean accuracy of 0.71, estimating performance on unseen subjects. This resulting proof-of-concept model predicts treatment responses for wounds treated with nanofibrillated-cellulose hydrogel. Multimodal DL combining standard and thermal photography with size metrics can stage wound healing and detect divergent treatment trajectories and thus be used in preclinical efficacy evaluation of novel wound treatments. With larger, clinically annotated datasets, this approach could be applied in a clinical setting as well.

Keywords:  AI; Biomedical Imaging; Deep Learning; Diabetic Wound; Hydrogel; Nanofibrillated Cellulose

DOI:  https://doi.org/10.1016/j.ejps.2026.107693
Front Endocrinol (Lausanne). 2026 ;17 1964592

Outcome-specific value of longitudinal clinical information and model complexity for dynamic prediction of type 2 diabetes-related outcomes: a pooled landmarking study.

Meiling Fu, Xiaohong Liu, Ting Shi, Yuxuan Lu, Jie Zou, Xiaowei Huang, Qingmei Pan, Yun Chen, Yinqin Zhong.

   Objective: To evaluate the outcome-specific incremental value of longitudinal clinical history and model complexity for annual dynamic prediction of multiple type 2 diabetes-related outcomes using pooled landmarking.
Methods: We included 646 patients with complete annual records from baseline through year 3. Landmark times at years 1 and 2 were used to predict the first recorded occurrence of hypertension, coronary heart disease, diabetic kidney disease, diabetic retinopathy, or metabolic dysfunction-associated steatotic liver disease during the subsequent year. Four progressively expanded feature sets were compared using elastic-net logistic regression and Light Gradient Boosting Machine. The primary history-current estimand was a paired within-algorithm contrast with the risk set, validation splits, preprocessing, tuning, calibration, and evaluation records held constant. Internal validation used patient-level repeated nested cross-validation. We also evaluated fully nested model selection and model-development stability using 500 patient-level bootstrap replicates with out-of-bag evaluation.
Results: Pooled out-of-fold area under the receiver operating characteristic curve ranged from 0.722 to 0.952 across the five representative models. The paired history-current comparison showed higher AUC for hypertension with elastic-net (ΔAUC 0.097, 95% confidence interval 0.057 to 0.136) and LightGBM (0.118, 0.081 to 0.158), and for metabolic dysfunction-associated steatotic liver disease with elastic-net (0.022, 0.002 to 0.042) and LightGBM (0.025, 0.009 to 0.041). The other three outcomes showed no comparable improvement. Algorithm differences were modest and outcome dependent.
Conclusion: In this single-center cohort with complete annual T0-T3 follow-up, the incremental value of longitudinal information and model complexity was outcome specific. Longitudinal history showed greater benefit for selected outcomes, whereas algorithm differences were modest and outcome dependent. The smaller MASLD signal was supported primarily by the pooled L1-L2 analysis. Independent external validation is required before clinical application.

Keywords:  dynamic prediction; internal validation; landmark analysis; longitudinal information; machine learning; type 2 diabetes

DOI:  https://doi.org/10.3389/fendo.2026.1964592
Biomed Eng Online. 2026 Jul 27. pii: 118. [Epub ahead of print]25(1):

Classification of gestational diabetes mellitus placental histopathological images based on MONAI and improved GAMatrix-YOLOv8.

Zhifa Jiang, Weirui Wu, Jingwen Liu, Xiekun Chen, Ruoping Lin, Xiangyun Ye, Donghui Huang, Zhen Zhang.

   BACKGROUND: Gestational diabetes mellitus (GDM) poses a significant threat to maternal and fetal health, and placental pathology is crucial for assessing disease severity. Traditional diagnosis relies on clinicians' visual assessment, which is often inefficient and subjective.
METHODS: To address these issues, we developed a MONAI-based image augmentation strategy and integrated the GAMatrix attention module within the YOLOv8 framework. Our study used a dataset of 5,065 placental pathology images (80% for training and 20% for validation) covering five pathology categories, supplemented by an external validation set of 500 images. A MONAI-based image augmentation pipeline was used to increase training-image diversity.
RESULTS: Four model configurations were compared: YOLOv8 trained on the original dataset, YOLOv8 trained with MONAI-based augmentation, GAMatrix-YOLOv8 trained on the original dataset, and GAMatrix-YOLOv8 trained with MONAI-based augmentation. In the original comparative analysis, YOLOv8 trained on the original dataset achieved a macro-F1 score of 95.56% (95% CI, 94.86%-96.26%) in internal validation. The GAMatrix-YOLOv8 model trained with MONAI-based augmentation showed the strongest overall performance among the evaluated configurations, while external image-level evaluation indicated persistent class-specific errors, particularly for thrombotic vascular lesion and chorangiosis.
CONCLUSION: These findings suggest that attention-enhanced YOLOv8 combined with MONAI-based augmentation may support placental histopathological image classification, but the results should be interpreted as image-level findings and require patient-level, multicenter, and prospective validation.

Keywords:  Classification; GAMatrix-YOLOv8; Gestational diabetes mellitus; Pathological images

DOI:  https://doi.org/10.1186/s12938-026-01602-w
J Endocr Soc. 2026 Nov;10(11): bvag225

Development and validation of a model using routine clinical features to improve diabetes diagnosis.

Alan L Hutchison, Simar Narula, Brandon Faubert, Celeste C Thomas, Mary E Rinella, Raghavendra G Mirmira, William F Parker.

   Context: Hemoglobin A1c (A1c) and fasting plasma glucose (FPG) miss diabetes diagnosis in many individuals, but the alternative, oral glucose tolerance testing (OGTT), is difficult to obtain.
Objective: This work aimed to develop and validate a model to identify diabetes missed by A1c and FPG but detected by OGTT and associate its outcomes with mortality.
Methods: We trained machine-learning models using National Health and Nutrition Examination Survey (NHANES) data, applied them to corresponding National Center for Health Statistics mortality data, and externally validated the models with Diabetes Prevention Program Outcomes Study (DPPOS) data. We tested models including the A1c, vital signs, complete blood count, and comprehensive metabolic panel with and without the FPG. This retrospective analysis included NHANES study participants from 1999 to 2016 and DPPOS study participants from 1996 to 2020. Main outcome measures included the area under the receiver operating curve, sensitivity, and specificity of models to detect diabetes.
Results: The A1c/FPG+ model had better sensitivity for DPPOS individuals with A1c less than 6.5% (48 mmol/mol) and FPG less than 126 mg/dL compared to a logistic regression of A1c and FPG and recalibrated FPG for individuals both on (0.69 vs 0.53 vs 0.57, respectively) and off (0.75 vs 0.65 vs 0.68, respectively) medication. NHANES participants without diabetes but with A1c/FPG+-predicted diabetes had elevated risk of mortality compared to those without A1c/FPG+-predicted diabetes (HR = 1.4; P < 2 × 10-16).
Conclusion: Machine-learning models can improve the detection of individuals who would need OGTT assessment to diagnose diabetes using standard nonfasting laboratory values and vital signs with and without FPG.

Keywords:  diabetes; diagnosis; machine learning; oral glucose tolerance test

DOI:  https://doi.org/10.1210/jendso/bvag225
Endocr Pract. 2026 Oct 19. pii: S1530-891X(26)01290-5. [Epub ahead of print]

Feasibility of an Activity Control System in Patients with Type 2 Diabetes: Results from a Non-Randomised Controlled Pilot Study.

David Pérez Cruzado, Macarena Espinilla-Estévez, Pablo de Blas Zamorano, Matilde Romero López, Pedro Montagut Martínez, Nicomedes Rodríguez Rodríguez, David Díaz-Jiménez, Jesús González Lama.

   OBJECTIVE: Diabetes mellitus (DM) is a disease that requires continuous, active management by the patient, encompassing lifestyle and activities of daily living (ADL). There is limited evidence and a scarcity of devices that monitor ADL in patients with DM using artificial intelligence-based technology. This study evaluates the feasibility of the AI2EPD home activity monitoring system, focusing on therapeutic adherence in the treatment of people with DM2.
METHODS: A pilot-controlled study was conducted. Vibration and motion sensors were placed in the home, and physical activity bracelets were worn, together with an Artificial Intelligence web system. The following scales were used for evaluation: System Usability Scale (SUS), Morisky-Green Test, Charlson Comorbidity Index (CCI), Spanish version of the Summary of Diabetes Self-Care Activities Measure (SDSCA-Sp), Diabetes Knowledge Questionnaire (DKQ-24), International Physical Activity Questionnaire (IPAQ), and EuroQol 5D.
RESULTS: Twenty-one participants were included (10 in the intervention group and 11 in the control group). The clinical results were similar for both groups. The intervention group found the system easy to use and safe, indicating high acceptance. No clinical benefit attributable to the intervention was demonstrated. The only statistically significant within-group improvement in HbA1c was observed in the control group; however, given the pilot nature of the study and the multiple exploratory comparisons, this finding should be interpreted with caution.
CONCLUSIONS: The use of the AI2EPD intelligent remote monitoring system in the home was feasible and well accepted by participants. Exploratory findings regarding therapeutic adherence and clinical outcomes warrant further evaluation in adequately powered studies.

Keywords:  Activities of Daily Living; Artificial Intelligence; Diabetes Mellitus; Remote Patient Monitoring; Telehealth

DOI:  https://doi.org/10.1016/j.eprac.2026.10.007
Int J Nurs Stud Adv. 2026 Dec;11 100699

Views and conditional acceptance of telehealth and artificial intelligence for lifestyle modification among individuals with prediabetes: A qualitative descriptive study.

Arkers Kwan Ching Wong, Chung Hei Dye, Menghan Wang, Jonathan Bayuo, Aoxi Tian.

   Background: Prediabetes is associated with an increased risk of progression to type 2 diabetes, yet sustained lifestyle modification remains difficult in everyday life. Although telehealth and artificial intelligence may extend lifestyle support through remote access, monitoring, reminders and personalised feedback, little is known about how people with prediabetes perceive these technologies or the conditions under which they would consider them acceptable.
Objective: To explore how adults with prediabetes understand and manage lifestyle modification and to identify their perceived support needs and the conditions shaping acceptance of telehealth and artificial intelligence as adjuncts to lifestyle support.
Design: A qualitative descriptive study.
Settings: Hong Kong.
Participants: 10 adults with prediabetes.
Methods: Semi-structured interviews were conducted between September and October 2025. Interviews were conducted in Cantonese, audio-recorded, transcribed verbatim, translated into English, and analysed using thematic analysis. Demographic, clinical, and digital health-related information was collected to contextualise participants' accounts.
Results: Five interrelated themes were identified. Participants' perceived risk of prediabetes was shaped less by the diagnosis itself than by family history, bodily symptoms, and personal health biographies. Although participants generally understood the need for dietary change, physical activity, and weight control, they found it difficult to translate knowledge into sustained action because of work demands, food environments, fatigue, emotional distress, and limited control over daily routines. Lifestyle modification was, therefore, approached pragmatically through modest, incremental, and sustainable adjustments rather than strict behavioural transformation. Meaningful support was perceived as ongoing, human-centred, and responsive to individual circumstances. Within this context, telehealth and artificial intelligence were viewed as potentially-acceptable extensions of care, particularly for accessible information access, reminders, self-monitoring, and personalised feedback. However, acceptability was conditional on accuracy, privacy protection, cultural relevance, professional credibility, and continued clinician involvement.
Conclusions: Telehealth and artificial intelligence may help support lifestyle modification among individuals with prediabetes, but their value depends on their ability to complement rather than human care. Artificial intelligence-enabled telehealth interventions should be designed as hybrid, clinician-integrated models that translate health data into locally relevant and actionable guidance while preserving trust, accountability, and human reassurance. Such approaches may help bridge the persistent gap between knowing what lifestyle changes are needed and sustaining those changes in everyday life for individuals with prediabetes in Hong Kong.
Registration: Not registered.

Keywords:  Artificial intelligence; Hong Kong; Lifestyle modification; Prediabetes; Qualitative descriptive study; Telehealth

DOI:  https://doi.org/10.1016/j.ijnsa.2026.100699