bims-aukdir 2025-12-28 papers

bims-aukdir

Biomed News

on Automated knowledge discovery in diabetes research

Issue of 2025–12–28
fifteen papers selected by
Mott Given

Detection and diagnosis of diabetic retinopathy in retinal fundus images using agentic AI approaches.
Deep learning-based joint analysis of diabetic retinopathy and glaucoma in retinal fundus images.
Intelligent glucose management in hospitalized patients: Short-term glucose and adverse events prediction.
[Adaptive lesion-aware fusion network for joint grading of multiple fundus diseases].
The Technological Transformation in Diabetes Care: From Continuous Glucose Monitoring to Artificial Intelligence and Beyond.
Pathway to precision: machine learning and bioinformatics in diabetes gene expression studies.
An ethnic-sensitive hybrid framework for T2D prediction with explainable AI and weighted ensembles.
Data-driven abdominal phenotypes of type 2 diabetes in lean, overweight, and obese cohorts from computed tomography.
Relationship between cardiometabolic index and worsening renal function in T2DM: Insights from machine learning, growth mixture modelling, and development of a web-based risk prediction tool.
Identifying the Most Crucial Factors Influencing Self-Compassion Among Community-Dwelling Older Adults with Type 2 Diabetes Using Interpretable Machine Learning.
The efficacy of CGM driven machine learning algorithms in predicting hypoglycemia in patients with T1DM: a systematic review and meta-analysis.
Integrating GWAS and machine learning for disease risk prediction in the Taiwanese Hakka population.
Development of Insulin Bolus Calculators in Type 1 Diabetes using A Framework Based on Real-world Data, Digital Twins and Machine Learning.
Prediction of time to insulin initiation in gestational diabetes mellitus: a secondary analysis of the EMERGE trial.
A closed-loop microneedle-integrated physiological model for predictive glycemic management.

Sci Rep. 2025 Dec 26.

Detection and diagnosis of diabetic retinopathy in retinal fundus images using agentic AI approaches.

R Sathya, A Valaramathi.

  In today's world, Diabetic Retinopathy (DR) remains a leading cause of vision loss globally, necessitating early detection and accurate diagnosis for timely intervention. Traditional machine learning and deep learning-based approaches, while effective, often suffer from issues such as limited interpretability, static decision-making, and inadequate generalization across diverse patient data. This research introduces an Agentic-AI Driven Framework for Diabetic Retinopathy Analysis (AADR-AI), which leverages intelligent agent-based learning mechanisms to enhance decision-making autonomy, dynamic adaptability, and contextual understanding of retinal fundus images. The novelty lies in incorporating agentic intelligence principles, autonomy, reactivity, and proactivity into DR detection systems, allowing real-time analysis and adaptive feature learning based on patient-specific variations. The proposed AADR-AI framework integrates a multi-agent ensemble of convolutional and transformer-based networks, coordinated through a decision fusion layer for robust classification. Key contributions include improved classification accuracy (up to 96.7%), enhanced model efficiency with reduced computational overhead, and real-time adaptability to varying image qualities and disease progression stages. Extensive experimentation on benchmark datasets demonstrates superior performance compared to existing state-of-the-art methods. This work highlights the transformative potential of agentic AI in medical imaging, paving the way for more autonomous and interpretable clinical decision-support systems.

Keywords:  Agent AI; Deep learning; Diabetic retinopathy; Fundus imaging; Medical diagnosis; Real-Time adaptability

DOI:  https://doi.org/10.1038/s41598-025-34016-0
Sci Rep. 2025 Dec 25.

Deep learning-based joint analysis of diabetic retinopathy and glaucoma in retinal fundus images.

T Geetha, C Hema.

  In current times, Diabetic retinopathy (DR) might be more difficult to diagnose when coexisting with glaucoma, since the two diseases share retinal abnormalities. Worldwide, DR is one of the most common causes of blindness. Conventional convolutional neural network (CNN)-based approaches struggle significantly with this type of co-morbid imaging due to the inherent difficulty in understanding both coarse-grained features and global correlations. The authors of this study propose a novel deep learning architecture, Vision Transformer (ViT) with Bi-Directional Feature Fusion (BFF) (ViT-BiFusionDRNet-HGS), to address these limitations. It is fine-tuned using the HGS technique, which was created for the Hunger Games, and combines a Vision Transformer (ViT) with Bi-Directional Feature Fusion (BFF). The BFF module enables the learning of semantic features from low-level textures, while the Vision Transformer captures long-distance spatial correlations. By incorporating the Hunger Games Search (HGS) algorithm into the model, it optimizes crucial hyperparameters and fusion weights, allowing for better generalization across complex fundus images, faster convergence, and more accurate lesion localization. With a classification accuracy of 98.4% and sensitivity levels higher than those of CNN, standalone ViT, and other baseline optimizers, the model demonstrated superior performance on open-source datasets for diabetic retinopathy and glaucoma fundus images. Clinically, ViT-BiFusionDRNet-HGS shows great potential as a real-time, scalable system for automated analysis of retinal abnormalities in complex diagnostic situations.

Keywords:  Automated diagnosis; Bi-Directional feature fusion (BFF); Deep learning; Diabetic retinopathy (DR); Fundus imaging; Glaucoma; Hunger games search (HGS) algorithm; Medical image analysis; Retinal disease detection; ViT-BiFusionDRNet; Vision transformer (ViT)

DOI:  https://doi.org/10.1038/s41598-025-32991-y
PLoS One. 2025 ;20(12): e0339360

Intelligent glucose management in hospitalized patients: Short-term glucose and adverse events prediction.

Jian Xue, Huaimin Cao, Lu Yang, Jiajia Liu, Ying Zhang, Chunlei Zhang, Ke Bai, Peng Jiang, Fanchang Hao, LiNa Xu.

The management of blood glucose in hospitalized patients is confined to retrospective interventions, preventing healthcare professionals from predicting patients' blood glucose levels and potential adverse events in advance. This study employs a deep learning model, specifically a Stacked Attention-Gated Recurrent Unit (SA-GRU) network, to forecast short-term blood glucose (BG) levels and predict adverse events in hospitalized patients, assisting clinicians in making clinical decisions. We collect continuous glucose monitoring(CGM) data from 196 hospitalized patients with type 2 diabetes, and by constructing and training this deep learning model, we predict blood glucose levels and adverse events.The model's predictions are then compared with the actual CGM data, and different evaluation metrics are used to assess the predictions of blood glucose levels and adverse events. Additionally, experiments were conducted on another publicly available type 2 diabetes dataset. On our collected data, for the 30-minute prediction, the root mean square error (RMSE) and mean absolute relative difference (MARD) of blood glucose are 4.27 ± 0.31 mg/dL and 1.77% ± 0.08%, respectively, with an adverse event classification accuracy of 98.57% ± 0.11%. For the 60-minute prediction, the RMSE and MARD of blood glucose are 10.46 ± 0.55 mg/dL and 4.59% ± 0.22%, respectively, with an adverse event classification accuracy of 95.74% ± 0.33%. Similar positive results were obtained on another publicly available dataset. The proposed model demonstrates accurate predictions for blood glucose values and adverse events in the next 30 and 60 minutes.

DOI: https://doi.org/10.1371/journal.pone.0339360
Sheng Wu Yi Xue Gong Cheng Xue Za Zhi. 2025 Dec 25. 42(6): 1172-1180

[Adaptive lesion-aware fusion network for joint grading of multiple fundus diseases].

Wei Zeng, Shengwen Guo.

  Diabetic retinopathy (DR) and its complication, diabetic macular edema (DME), are major causes of visual impairment and even blindness. The occurrence of DR and DME is pathologically interconnected, and their clinical diagnoses are closely related. Joint learning can help improve the accuracy of diagnosis. This paper proposed a novel adaptive lesion-aware fusion network (ALFNet) to facilitate the joint grading of DR and DME. ALFNet employed DenseNet-121 as the backbone and incorporated an adaptive lesion attention module (ALAM) to capture the distinct lesion characteristics of DR and DME. A deep feature fusion module (DFFM) with a shared-parameter local attention mechanism was designed to learn the correlation between the two diseases. Furthermore, a four-branch composite loss function was introduced to enhance the network's multi-task learning capability. Experimental results demonstrated that ALFNet achieved superior joint grading performance on the Messidor dataset, with joint accuracy rates of 0.868 (DR 2 & DME 3), outperforming state-of-the-art methods. These results highlight the unique advantages of the proposed approach in the joint grading of DR and DME, thereby improving the efficiency and accuracy of clinical decision-making.

Keywords:  Adaptive lesion attention; Deep feature fusion; Diabetic macular edema; Diabetic retinopathy; Joint grading

DOI:  https://doi.org/10.7507/1001-5515.202411032
Endocr Pract. 2025 Dec 18. pii: S1530-891X(25)01327-8. [Epub ahead of print]

The Technological Transformation in Diabetes Care: From Continuous Glucose Monitoring to Artificial Intelligence and Beyond.

Guillermo E Umpierrez, M Cecilia Lansang.



Keywords:  Continuous glucose monitoring; artificial intelligence; automatic insulin delivery; closed-loop; glucose monitoring

DOI:  https://doi.org/10.1016/j.eprac.2025.11.016
J Diabetes Metab Disord. 2026 Jun;25(1): 3

Pathway to precision: machine learning and bioinformatics in diabetes gene expression studies.

Tahmineh Aldaghi, Ashley Chrissan, Jan Muzik, Mostafa Rezaei-Tavirani, Elham Nazari.

   Background: Diabetes mellitus is a metabolic disorder; understanding the pathogenic mechanisms underlying diabetes is crucial. Analyzing biomarkers, supported by machine learning and bioinformatics, is crucial for identifying the molecular causes of diabetes.
Objective: This study summarizes the current advances in diabetes research, highlighting significant progress in bioinformatics, gene expression analysis, and machine learning.
Methods: The search was conducted in Google Scholar, PubMed, and Scopus using two sets of keywords, including terms like diabetes, biomarkers, bioinformatics, and machine learning. The selection process followed PRISMA guidelines. The inclusion criteria targeted open-access English articles published from 2020 to 2024, resulting in a final selection of 96 articles.
Results: The findings were grouped into six categories: data acquisition methods, complications, bioinformatics techniques, machine learning methods, promotion, and evaluation. Microarrays were the most frequent technique for data collection, with 70 occurrences. The most common bioinformatics method was KEGG pathway analysis (84 instances). The most frequently used machine learning techniques were LASSO (43) and PCA (47). Prognostic applications were reported 45 times, while diagnostic applications appeared 51 times.
Conclusion: Recent studies have highlighted the importance of KEGG pathway analysis and microarray datasets in diabetes research. Newer technologies, such as RNA-seq and single-cell RNA-seq, remain underutilized despite significant advancements in their development. A greater focus on novel biological pathways, the development of biomarkers, and a more comprehensive application of machine learning techniques could all aid in personalized treatment for diabetes. Future research should aim to overcome technological limitations and integrate clinical data with bioinformatics workflows to enhance translational relevance and promote precision medicine.

Keywords:  Bioinformatics; Diabetes; Gene expression; Machine learnings; Microarray

DOI:  https://doi.org/10.1007/s40200-025-01814-2
Sci Rep. 2025 Dec 23.

An ethnic-sensitive hybrid framework for T2D prediction with explainable AI and weighted ensembles.

Karlo Abnoosian, Rahman Farnoosh, Hamidreza Noushkaran, Danial Javaheri, Hasnat Azeez Shimal, Abdullah Abdulamer Abdulkarem.

  Type 2 diabetes (T2D) is a growing global health crisis, affecting over 537 million people as of 2021. Early prediction remains particularly challenging in low- and middle-income countries due to missing data, class imbalance, and population-specific risk factors. This study presents a four-stage predictive framework- Feature-Weighted Class-Adaptive Generative Imputation Network-Weighted Classifier Aggregation Ensemble (FW-CAGIN-WCAE)-designed to address these limitations. First, Zero-Threshold Feature Removal (ZTFR) is applied to eliminate low-quality variables. Second, missing values are imputed FW-CAGIN, a novel class-aware and feature-weighted GAN model that accounts for both class and feature importance. Third, a performance-weighted ensemble of 15 machine and deep learning algorithms is constructed. Finally, SHAP analysis is used to uncover population-specific risk indicators. The proposed method was evaluated on three benchmark datasets-PIDD, FHGDD, and BDD-and their combinations, using nested five-fold cross-validation. The model achieved a peak AUC of 0.936 ± 0.018 in PIDD-BDD combination and reduced the imputation mean absolute error (MAE) from 0.8028 to 0.0033. It also lowered AUC variability by 36.3% and improved the diagnostic odds ratio (DOR) to 68.4 ± 20.5. SHAP analysis identified as a key predictive feature across both Asian and European populations. These findings demonstrate that the proposed framework offers an accurate, interpretable, and population-sensitive solution for early T2D detection, especially in resource-limited healthcare settings.

Keywords:  Ensemble learning; FW-CAGIN imputation; Feature selection; Population-specific risk; SHAP analysis; Type 2 diabetes prediction

DOI:  https://doi.org/10.1038/s41598-025-31234-4
J Med Imaging (Bellingham). 2025 Nov;12(6): 064006

Data-driven abdominal phenotypes of type 2 diabetes in lean, overweight, and obese cohorts from computed tomography.

Lucas W Remedios, Chloe Cho, Trent M Schwartz, Dingjie Su, Gaurav Rudravaram, Chenyu Gao, Aravind R Krishnan, Adam M Saunders, Michael E Kim, Shunxing Bao, Alvin C Powers, Bennett A Landman, John Virostko.

   Purpose: Although elevated body mass index (BMI) is a well-known risk factor for type 2 diabetes, the disease's presence in some lean adults and absence in others with obesity suggests that more detailed measurements of body composition may uncover abdominal phenotypes of type 2 diabetes. With artificial intelligence (AI) and computed tomography (CT), we can now leverage robust image segmentation to extract detailed measurements of size, shape, and tissue composition from abdominal organs, abdominal muscle, and abdominal fat depots in 3D clinical imaging at scale. This creates an opportunity to empirically define body composition signatures linked to type 2 diabetes risk and protection using large-scale clinical data.
Approach: We studied imaging records of 1728 de-identified patients from Vanderbilt University Medical Center with BMI collected from the electronic health record. To uncover BMI-specific diabetic abdominal patterns from clinical CT, we applied our design four times: once on the full cohort ( n=1728 ) and once on lean ( n=497 ), overweight ( n=611 ), and obese ( n=620 ) subgroups separately. Briefly, our experimental design transforms abdominal scans into collections of explainable measurements, identifies which measurements most strongly predict type 2 diabetes and how they contribute to risk or protection, groups scans by shared model decision patterns, and links those decision patterns back to interpretable abdominal phenotypes in the original explainable measurement space of the abdomen using the following steps. (1) To capture abdominal composition: we represented each scan as a collection of 88 automatically extracted measurements of the size, shape, and fat content of abdominal structures using TotalSegmentator. (2) To learn key predictors: we trained a 10-fold cross-validated random forest classifier with SHapley Additive exPlanations (SHAP) analysis to rank features and estimate their risk-versus-protective effects for type 2 diabetes. (3) To validate individual effects: for the 20 highest-ranked features, we ran univariate logistic regressions to quantify their independent associations with type 2 diabetes. (4) To identify decision-making patterns: we embedded the top-20 SHAP profiles with uniform manifold approximation and projection and applied silhouette-guided K-means to cluster the random forest's decision space. (5) To link decisions to abdominal phenotypes: we fit one-versus-rest classifiers on the original anatomical measurements from each decision cluster and applied a second SHAP analysis to explore whether the random forest's logic had identified abdominal phenotypes.
Results: Across the full, lean, overweight, and obese cohorts, the random forest classifier achieved a mean area under the receiver operating characteristic curve (AUC) of 0.72 to 0.74. SHAP highlighted shared type 2 diabetes signatures in each group-fatty skeletal muscle, older age, greater visceral and subcutaneous fat, and a smaller or fat-laden pancreas. Univariate logistic regression confirmed the direction of 14 to 18 of the top 20 predictors within each subgroup ( p<0.05 ). Clustering the model's decision space further revealed type 2 diabetes-enriched abdominal phenotypes within the lean, overweight, and obese subgroups.
Conclusions: We found similar abdominal signatures of type 2 diabetes across the separate lean, overweight, and obese groups, which suggests that the abdominal drivers of type 2 diabetes may be consistent across weight classes. Although our model had a modest AUC, the explainable components allowed for a clear interpretation of feature importance. In addition, in both lean and obese subgroups, the most important feature for identifying type 2 diabetes was fatty skeletal muscle.

Keywords:  abdomen; body composition; computed tomography; explainable artificial intelligence; pattern discovery; phenotype; type 2 diabetes

DOI:  https://doi.org/10.1117/1.JMI.12.6.064006
Diabetes Obes Metab. 2025 Dec 22.

Relationship between cardiometabolic index and worsening renal function in T2DM: Insights from machine learning, growth mixture modelling, and development of a web-based risk prediction tool.

Quan Zhang, Yuying Zhou, Liyao Fu, Zhenhua Xing, Xunxi Deng, Shengen Yi, Shi Tai.

   AIM: Worsening renal function (WRF) is a common and serious complication of type 2 diabetes mellitus (T2DM), contributing to adverse clinical outcomes. Metabolic dysregulation is considered a key driver of its onset and progression. The cardiometabolic index (CMI), a novel marker of metabolic status, has recently been proposed as a potential predictor; however, its utility in predicting WRF remains unclear.
MATERIALS AND METHODS: A total of 10 094 participants from the Action to Control Cardiovascular Risk in Diabetes trial were included in the final analysis and external validation was performed using data from an independent, single-centre Chinese cohort. WRF was defined as either a doubling of serum creatinine from baseline or a decline in estimated glomerular filtration rate (eGFR) greater than 20 mL/min/1.73 m2 during follow-up. The Boruta algorithm and random forest-based recursive feature elimination were sequentially applied for feature selection, and six machine learning algorithms were implemented to construct predictive models. SHapley Additive exPlanations (SHAP) analysis was used to enhance model interpretability. Restricted cubic splines (RCS) were applied to examine the dose-response relationship between CMI and WRF, and a growth mixture model (GMM) was used to identify distinct CMI trajectories across follow-up visits at baseline, 12, 24, 36, 48, and 60 months. Cox proportional hazards regression models and cumulative incidence curves were used to evaluate the association between CMI and WRF across subgroups. A web-based visualisation tool was further used to enhance model accessibility.
RESULTS: After two-step feature selection, six variables were incorporated into the final machine learning model: CMI, eGFR, age, fasting plasma glucose, urinary albumin (UAlb), and urinary creatinine. All six developed models showed stable and favourable performance, with the XGBoost algorithm achieving the best results. SHAP analysis indicated that higher eGFR was the most influential predictor. A web-based visualisation tool was developed to facilitate interactive exploration of model predictions at the individual level (https://t2dm-wrf-prediction.streamlit.app/). RCS curves revealed a strong positive association between CMI and WRF after multiple adjustments. Participants in the highest CMI tertile had a 16% higher risk of WRF compared with those in the lowest tertile (adjusted HR: 1.16; 95% CI: 1.09-1.24), and significant interactions were observed among female participants, older individuals, and those assigned to the standard glycemic control arm. Using the GMM, three distinct CMI trajectory groups were identified: 'low-stable,' 'moderate-stable,' and 'high-increasing.' Cumulative incidence curves further showed that both higher baseline CMI and membership in the 'high-increasing' group were associated with substantially elevated WRF risk.
CONCLUSIONS: Elevated CMI is positively associated with WRF in patients with T2DM, and its incorporation into predictive models may improve early identification of high-risk individuals.

Keywords:  cardiometabolic index; growth mixture model; machine learning; type 2 diabetes mellitus; worsening renal function

DOI:  https://doi.org/10.1111/dom.70374
Diabetes Metab Syndr Obes. 2025 ;18 4571-4586

Identifying the Most Crucial Factors Influencing Self-Compassion Among Community-Dwelling Older Adults with Type 2 Diabetes Using Interpretable Machine Learning.

Junxian Xu, Jianzhong Yang, Yuping Lu, Jieyu Yang, Chao Gu, Jiahuan Zhu, Lanni Yang.

   Objective: Managing diabetes daily can be an emotional burden for older adults. Research shows that self-compassion, which refers to the ability to be kind and understanding toward oneself, can help improve emotional well-being. This study aimed to develop a machine learning prediction model to identify the influencing factors of self-compassion among community-dwelling older adults with type 2 diabetes.
Methods: We conducted this study in Jiaxing, China, during July and August 2024. We invited community-dwelling older adults with type 2 diabetes to complete a questionnaire that measured their levels of self-compassion, depression, and anxiety. Our goal was to find which of 26 different personal and health-related factors most influenced self-compassion. To achieve this, we used several machine learning algorithms to build and compare predictive models, selecting the best-performing one. Finally, we applied a technique called SHapley Additive exPlanations (SHAP) to clearly understand and interpret how each factor impacts self-compassion.
Results: The random forest model performed the best. SHAP analysis indicated that depression, hemoglobin A1c (HbA1C), waist circumference, and anxiety were risk factors of self-compassion, while fasting blood-glucose (FBG) was a protective factor.
Conclusion: This study provides a reliable tool for identifying older adults with type 2 diabetes who may benefit from support. The findings suggest that healthcare providers should prioritize managing depression and anxiety, along with controlling HbA1c and waist circumference, to enhance self-compassion. These results can be translated into a practical risk scorecard to guide personalized care strategies in community health settings.

Keywords:  community-dwelling older adults; interpretable machine learning; self-compassion; type 2 diabetes

DOI:  https://doi.org/10.2147/DMSO.S556917
J Diabetes Metab Disord. 2026 Jun;25(1): 2

The efficacy of CGM driven machine learning algorithms in predicting hypoglycemia in patients with T1DM: a systematic review and meta-analysis.

Tameem Quader, Siddharth Pai, Yong Mong Tan, Harshal Deshmukh, Usman Malabu, Theophilus I Emeto.

   Purpose: An important function of continuous glucose monitoring (CGM) is to alert individuals with type one diabetes mellitus (T1DM) to impending hypoglycemia, however, it lacks the ability to predict episodes beyond 30 min. Machine learning (ML) algorithms incorporating other contextual data can be used to overcome this deficiency. This study aims to quantitatively evaluate the diagnostic accuracy of these algorithms.
Methods: A systematic search of databases following PRISMA guidelines identified relevant studies that trained and assessed ML algorithms (PROSPERO CRD42024588619). The set of 2 × 2 data (i.e., number of true positives, false positives, true negatives, and false negatives) was extracted and meta-analyzed using a generalized linear mixed model to calculate pooled estimates of sensitivity and specificity and construct a summary receiver operating characteristic curve. A two-sided p-value of < 0.05 was deemed significant.
Results: Of 611 studies screened, 20 met the inclusion criteria. The pooled point estimates (95% CI) were 80% (71-87%), 89% (78-95%), 7.27 (3.96-14.20) and 0.25 (0.14-0.37) for sensitivity, specificity, positive likelihood ratio (PLR) and negative likelihood ratio (NLR), respectively.
Conclusions: Current ML algorithms have a substantial ability to predict hypoglycemia in patients with T1DM according to the Users' Guide to Medical Literature on diagnostic tests where PLR should be ≥ 5 and NLR should be ≤ 0.2 for moderate reliability. The incorporation of other inputs such as insulin, carbohydrates and physical activity have enhanced prediction accuracy. The clinical utility of these algorithms, however, should be evaluated as per the patient's daily hypoglycemic risk profile due to the moderate risk of false positives.
Supplementary Information: The online version contains supplementary material available at 10.1007/s40200-025-01820-4.

Keywords:  Continuous glucose monitoring; Hypoglycemia: machine learning; Type 1 diabetes mellitus

DOI:  https://doi.org/10.1007/s40200-025-01820-4
Front Genet. 2025 ;16 1694084

Integrating GWAS and machine learning for disease risk prediction in the Taiwanese Hakka population.

Jing-Hong Xiao, Hsiao-Yen Kang, Li-Ching Wu, Tien Hsu, Chin-Pyng Wu, Li-Jen Su.

   Introduction: Genome-wide association studies (GWAS) have identified numerous loci associated with complex diseases, yet their predictive power in small or genetically homogeneous populations remains limited. Integrating machine learning with GWAS offers a path to improve risk prediction and uncover functional variants relevant to precision medicine.
Methods: DNA samples from Taiwanese Hakka individuals with type 2 diabetes, hypertension, and eye diseases were analyzed. After standard quality control, 295,589 SNPs were retained. Fourteen machine-learning algorithms were evaluated using SNPs selected through traditional GWAS filtering and refined via wrapper-based feature selection with a best-first search algorithm. Model performance was assessed by internal cross-validation and external validation using Taiwan Biobank data, and functional annotation was conducted through GTEx v10 cis-eQTL analysis.
Results: Predictive models relying solely on significant GWAS SNPs achieved moderate internal accuracy but limited generalizability. Incorporating feature-selected SNPs markedly improved performance: the Random Forest model achieved accuracies above 88% in cross-validation and above 85% in external validation, confirmed by 1,000× bootstrap resampling. eQTL analysis identified functional associations such as rs12121653-KDM5B and rs12121653-MGAT4EP, implicating pathways involved in metabolic and mitochondrial regulation.
Discussion: These findings demonstrate that integrating GWAS with machine-learning-based feature selection enables the construction of robust, population-specific disease risk models. Given the small sample size of the discovery cohort (n = 96), all predictive results should be interpreted as exploratory. We employed stringent cross-validation and 1,000× bootstrap resampling to reduce overfitting, and genomic control metrics (QQ plots and λGC values) were evaluated to ensure no major test statistic inflation. Independent large-scale validation will still be required. The approach effectively captures additive and interaction-driven genetic components and provides a scalable framework for applying precision medicine to underrepresented or isolated populations.

Keywords:  algorithmic rules; disease risk prediction; genome-wide association studies; machine learning; type 2 diabetes

DOI:  https://doi.org/10.3389/fgene.2025.1694084
IEEE Trans Biomed Eng. 2025 Dec 25. PP

Development of Insulin Bolus Calculators in Type 1 Diabetes using A Framework Based on Real-world Data, Digital Twins and Machine Learning.

Elisa Pellizzari, Giacomo Cappon, Giulia Nicolis, Giovanni Sparacino, Andrea Facchinetti.

OBJECTIVE: Precise mealtime insulin bolus (MIB) dosing is essential in type 1 diabetes (T1D) to minimize glucose excursions from carbohydrate intake. Traditional MIB formulas, based on glucose concentration at mealtime, are suboptimal and do not exploit real-time data from continuous glucose monitoring (CGM). Existing methods that incorporate CGM data often rely on empirical rules or are developed in-silico, limiting their applicability to real-world conditions. This work investigates a framework combining machine learning (ML) algorithms, digital twins (DTs) and real-world data, to improve the assessment, tuning, and development of MIB dosing algorithms.
METHODS: We utilized ReplayBG, a DT for T1D, to: i) evaluate a published linear ML model (Noaro et al.), originally developed in-silico, on real-world data from 30 free-living subjects; ii) recalibrate this model to fit the real-world dataset; and iii) train and test nonlinear gradient-boosting models (XGBoost, LightGBM) developed entirely on real data through DT simulations.
RESULTS: Progressing to DT-enhanced models, we observed improvements in glucose control. The recalibrated linear and nonlinear models increased time-in-range (up to 80.6% with LightGBM vs. 75.6% for Noaro et al.) and reduced time-above-range. Risk metrics reflecting hypo/hyperglycemia also improved.
CONCLUSIONS: These findings demonstrate that a DT-based framework grounded in real-world data supports both the refinement and development of bolus calculators, achieving performance gains beyond the original in-silico model.
SIGNIFICANCE: DTs allow the use real-world data to develop, validate and extend the domain of validity of new MIB formulas, paving the way to practical applications of ML tailored solutions for T1D.

DOI: https://doi.org/10.1109/TBME.2025.3648515
Diabetes Res Clin Pract. 2025 Dec 21. pii: S0168-8227(25)01085-X. [Epub ahead of print]231 113070

Prediction of time to insulin initiation in gestational diabetes mellitus: a secondary analysis of the EMERGE trial.

Yueyun Zhu, Alberto Alvarez-Iglesias, Aoife M Egan, Andrew Smyth, Christine Newman, David Aguilar Macias, Declan Devane, Fidelma Dunne, Martin O'Donnell, Paula O'Shea, Paddy Gillespie, Andrew J Simpkin.

   AIMS: Gestational diabetes mellitus (GDM) is a global health problem. Insulin therapy is recommended when lifestyle management fails to control blood glucose. We aim to predict the time to insulin initiation for women with GDM.
METHODS: A random survival forest (RSF) model was developed to predict the time to insulin initiation among 413 women from the EMERGE trial, analysed separately for placebo and metformin groups. Maternal characteristics and early glucose data (collected during the two weeks after randomisation) were used as predictors. Decision curve analysis was performed to assess the net benefit of the model compared with default strategies.
RESULTS: The RSF model had a concordance index (C-index) of 0.71 (95 % CI: 0.64-0.77), time-dependent AUC ≥0.70 and Brier score ≤0.2 for the placebo group, and a C-index of 0.72 (95 % CI: 0.64-0.80), time-dependent AUC ≥0.75 and Brier score ≤0.2 for the metformin group. The decision curve analysis showed the RSF provided a higher net benefit for both groups compared to default strategies across clinically relevant threshold probabilities.
CONCLUSIONS: The RSF model effectively identified women at high risk of requiring insulin. The decision curve analysis could help clinicians to balance insulin initiation decisions. However, prospective validation is needed to confirm the generalisability of the model.

Keywords:  Decision curve analysis; Gestational diabetes mellitus; Insulin therapy; Random survival forest

DOI:  https://doi.org/10.1016/j.diabres.2025.113070
Proc Natl Acad Sci U S A. 2025 Dec 30. 122(52): e2523517122

A closed-loop microneedle-integrated physiological model for predictive glycemic management.

Marco Fratus, Muhammad A Alam.

  Continuous biomarker monitoring and on-demand therapy are essential for chronic disease management, with diabetes being a key example. Although continuous glucose monitoring (CGM) and insulin pumps are moving toward a closed-loop system, it remains difficult to predict how device design influences system performance. Traditional optimization methods are inefficient and empirical: Data-driven algorithms lack interpretability, trial-and-error design is time-consuming, and first-principle models are difficult to integrate at the system level. To address these challenges, we introduce a physics-based framework which integrates microneedle (MN) designs for sensing and therapy with a physiological model of glycemic control. The framework builds compact relationships showing how material, chemical, and geometrical features affect key metrics such as response time, extraction flux, and insulin delivery rate. Specifically, we develop a theory for three sensing MN types (hollow, porous/swellable, and nanostructured MNs) and one therapeutic patch (electro-termo-mechanical device) to create a predictive model for glycemic regulation. By linking device design to system behavior across time and spatial scales, we examine the framework in three cases: How disease progression, MN design, and MN sensitivity affect plasma glucose levels and time-in-range metrics. This work establishes the foundation for a physics-based digital twin for diabetes management, complementing ML, experiments, and numerical models. More broadly, it streamlines patch design, minimizing glycemic events and trial-and-error in next-generation CGM technology.

Keywords:  closed-loop system; diabetes; microneedles; physics-based modeling; wearable devices

DOI:  https://doi.org/10.1073/pnas.2523517122