Pub Date : 2025-12-27DOI: 10.1016/j.ijmedinf.2025.106245
Hengjun Liu , Tianwei Meng , Rui Qie
Objective
This narrative review synthesizes machine learning (ML) applications across the stroke and post-stroke continuum from acute imaging and diagnosis to long-term sequelae prognosis and rehabilitation.
Method
We searched PubMed, Embase, and WOS from inception to October 17, 2025, for a comprehensive review. We used a combination of search terms, including “machine learning,” “deep learning,” “post stroke.” These terms were carefully selected to capture a wide range of relevant studies and articles related to stroke and ML.
Results
ML has been successfully deployed in six core domains: Image reading, where deep learning enables automated lesion segmentation on MRI/CT and prediction of tissue fate; Diagnosis, including etiology, atrial fibrillation screening; Overall prognosis, with high-accuracy models for functional outcome, mortality, and readmission; Sequelae prediction, such as cognitive impairment, motor dysfunction, aphasia, depression, fatigue, and organ diseases; Treatment response, including outcome prediction after thrombectomy and rehabilitation; Rehabilitation monitoring, using wearable sensors and robotics for objective, granular assessment of motor recovery. A clear trend toward multimodal data integration and model interpretability was observed, enhancing both predictive power and biological plausibility.
Conclusion
ML has evolved from a research tool into a transformative force in stroke care, enabling precise, individualized prediction and monitoring across the entire post-stroke trajectory. Future efforts must prioritize prospective validation, standardized reporting, and seamless integration into clinical workflows to realize its full potential for precision medicine.
{"title":"Machine learning in stroke and its sequelae: a narrative review of clinical applications and emerging trends","authors":"Hengjun Liu , Tianwei Meng , Rui Qie","doi":"10.1016/j.ijmedinf.2025.106245","DOIUrl":"10.1016/j.ijmedinf.2025.106245","url":null,"abstract":"<div><h3>Objective</h3><div>This narrative review synthesizes machine learning (ML) applications across the stroke and post-stroke continuum from acute imaging and diagnosis to long-term sequelae prognosis and rehabilitation.</div></div><div><h3>Method</h3><div>We searched PubMed, Embase, and WOS from inception to October 17, 2025, for a comprehensive review. We used a combination of search terms, including “machine learning,” “deep learning,” “post stroke.” These terms were carefully selected to capture a wide range of relevant studies and articles related to stroke and ML.</div></div><div><h3>Results</h3><div>ML has been successfully deployed in six core domains: Image reading, where deep learning enables automated lesion segmentation on MRI/CT and prediction of tissue fate; Diagnosis, including etiology, atrial fibrillation screening; Overall prognosis, with high-accuracy models for functional outcome, mortality, and readmission; Sequelae prediction, such as cognitive impairment, motor dysfunction, aphasia, depression, fatigue, and organ diseases; Treatment response, including outcome prediction after thrombectomy and rehabilitation; Rehabilitation monitoring, using wearable sensors and robotics for objective, granular assessment of motor recovery. A clear trend toward multimodal data integration and model interpretability was observed, enhancing both predictive power and biological plausibility.</div></div><div><h3>Conclusion</h3><div>ML has evolved from a research tool into a transformative force in stroke care, enabling precise, individualized prediction and monitoring across the entire post-stroke trajectory. Future efforts must prioritize prospective validation, standardized reporting, and seamless integration into clinical workflows to realize its full potential for precision medicine.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106245"},"PeriodicalIF":4.1,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885682","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-27DOI: 10.1016/j.ijmedinf.2025.106244
Xin Jiang , Ji Li , Jingjing Ju , Hao Ding , Sufang Yang
Objective
This study aimed to create and validate a machine learning (ML) model to predict the likelihood of invasive mechanical ventilation (IMV) in patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) complicated by respiratory failure.
Methods
Data from patients diagnosed with AECOPD and respiratory failure were retrospectively extracted from the Medical Information Mart for Intensive Care-IV (MIMIC-IV). A total of 551 cases were split 7:3 into a training set (385 cases) for model construction and an internal validation set (166 cases). The IMV served as the outcome event. Features were selected with the Boruta algorithm and least absolute shrinkage and selection operator (LASSO). Eight ML algorithms—XGBoost, decision tree (DT), random forest (RF), support-vector machine (SVM), LightGBM, CatBoost, Gaussian naïve Bayes (NB) and K-nearest neighbor (NN)—were trained with 10-fold cross-validation. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, calibration curve, decision curve and clinical impact curve. An external validation cohort of 100 AECOPD-respiratory failure patients admitted to Baoying People’s Hospital between January 2020 and August 2025 was collected. The final best model was interpreted with SHapley Additive exPlanations (SHAP) to clarify feature importance and decision logic, and an interactive dynamic nomogram was plotted to increase readability.
Results
Boruta plus LASSO identified total calcium, partial pressure of oxygen (PO2), oxygen saturation (SpO2) and sepsis as significant predictors. XGBoost outperformed the other algorithms, achieving an internal validation accuracy of 72.2 %, sensitivity of 64.6 %, specificity of 79.8 %, F1 score of 69.7 % and AUC of 0.813 (95 % CI 0.748–0.878). The external validation accuracy reached 76.4 %, the sensitivity reached 82.6 %, the specificity reached 70.0 %, the F1 score reached 78.7 %, and the AUC reached 0.840 (95 % CI 0.801–0.879). SHAP analysis further indicated that PO2 and SpO2 were the primary drivers of model decisions. An interactive dynamic nomogram was successfully constructed.
Conclusion
IMV in AECOPD patients with respiratory failure was associated with total calcium, PO2, and SpO2 levels and sepsis. The developed XGBoost model demonstrated good predictive value for IMV in this clinical population.
目的：本研究旨在建立并验证机器学习（ML）模型，以预测慢性阻塞性肺疾病（AECOPD）急性加重期合并呼吸衰竭患者进行有创机械通气（IMV）的可能性。方法：回顾性地从重症监护医学信息市场- iv （MIMIC-IV）中提取诊断为AECOPD和呼吸衰竭的患者的资料。551个案例以7:3的比例分成用于模型构建的训练集（385例）和内部验证集（166例）。国际货币基金组织会议是最后的会议。使用Boruta算法和最小绝对收缩和选择算子（LASSO）选择特征。8种ML算法——xgboost、决策树（DT）、随机森林（RF）、支持向量机（SVM）、LightGBM、CatBoost、高斯naïve贝叶斯（NB）和k近邻（NN）——进行了10倍交叉验证的训练。通过受试者工作特征曲线下面积（AUC）、准确度、灵敏度、特异性、F1评分、校准曲线、决策曲线和临床影响曲线评价模型性能。收集2020年1月至2025年8月在宝应市人民医院住院的100例aecopd -呼吸衰竭患者的外部验证队列。利用SHapley加性解释（SHAP）对最终的最佳模型进行解释，以明确特征重要性和决策逻辑，并绘制交互式动态nomogram以提高可读性。结果：Boruta + LASSO发现总钙、氧分压（PO2）、氧饱和度（SpO2）和脓毒症是显著的预测因素。XGBoost优于其他算法，其内部验证准确率为72.2%，灵敏度为64.6%，特异性为79.8%，F1评分为69.7%，AUC为0.813 （95% CI 0.748 ~ 0.878）。外部验证准确度达76.4%，灵敏度达82.6%，特异性达70.0%，F1评分达78.7%，AUC达0.840 （95% CI 0.801 ~ 0.879）。SHAP分析进一步表明，PO2和SpO2是模型决策的主要驱动因素。成功地构造了一个交互式动态图。结论：AECOPD合并呼吸衰竭患者IMV与总钙、PO2、SpO2水平及脓毒症相关。开发的XGBoost模型在该临床人群中显示出良好的IMV预测价值。
{"title":"Construction and validation of a machine learning-based risk prediction model for invasive mechanical ventilation in AECOPD patients complicated with respiratory failure","authors":"Xin Jiang , Ji Li , Jingjing Ju , Hao Ding , Sufang Yang","doi":"10.1016/j.ijmedinf.2025.106244","DOIUrl":"10.1016/j.ijmedinf.2025.106244","url":null,"abstract":"<div><h3>Objective</h3><div>This study aimed to create and validate a machine learning (ML) model to predict the likelihood of invasive mechanical ventilation (IMV) in patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) complicated by respiratory failure.</div></div><div><h3>Methods</h3><div>Data from patients diagnosed with AECOPD and respiratory failure were retrospectively extracted from the Medical Information Mart for Intensive Care-IV (MIMIC-IV). A total of 551 cases were split 7:3 into a training set (385 cases) for model construction and an internal validation set (166 cases). The IMV served as the outcome event. Features were selected with the Boruta algorithm and least absolute shrinkage and selection operator (LASSO). Eight ML algorithms—XGBoost, decision tree (DT), random forest (RF), support-vector machine (SVM), LightGBM, CatBoost, Gaussian naïve Bayes (NB) and K-nearest neighbor (NN)—were trained with 10-fold cross-validation. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, calibration curve, decision curve and clinical impact curve. An external validation cohort of 100 AECOPD-respiratory failure patients admitted to Baoying People’s Hospital between January 2020 and August 2025 was collected. The final best model was interpreted with SHapley Additive exPlanations (SHAP) to clarify feature importance and decision logic, and an interactive dynamic nomogram was plotted to increase readability.</div></div><div><h3>Results</h3><div>Boruta plus LASSO identified total calcium, partial pressure of oxygen (PO<sub>2</sub>), oxygen saturation (SpO<sub>2</sub>) and sepsis as significant predictors. XGBoost outperformed the other algorithms, achieving an internal validation accuracy of 72.2<!--> <!-->%, sensitivity of 64.6<!--> <!-->%, specificity of 79.8 %, F1 score of 69.7<!--> <!-->% and AUC of 0.813 (95<!--> <!-->% CI 0.748–0.878). The external validation accuracy reached 76.4<!--> <!-->%, the sensitivity reached 82.6<!--> <!-->%, the specificity reached 70.0<!--> <!-->%, the F1 score reached 78.7<!--> <!-->%, and the AUC reached 0.840 (95<!--> <!-->% CI 0.801–0.879). SHAP analysis further indicated that PO<sub>2</sub> and SpO<sub>2</sub> were the primary drivers of model decisions. An interactive dynamic nomogram was successfully constructed.</div></div><div><h3>Conclusion</h3><div>IMV in AECOPD patients with respiratory failure was associated with total calcium, PO<sub>2</sub>, and SpO<sub>2</sub> levels and sepsis. The developed XGBoost model demonstrated good predictive value for IMV in this clinical population.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106244"},"PeriodicalIF":4.1,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145866463","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper explores the potential benefits and limitations of synthetic data (SD) in paediatrics, addressing the challenges of data scarcity and privacy concerns in paediatric research.
Methodology
A narrative literature review was conducted, searching PubMed and Scopus databases for relevant publications up to August 2025. The review focused on studies addressing the use, development, or application of SD in paediatric healthcare settings.
Findings
Synthetic data offers numerous benefits in paediatrics, including enhancing dataset diversity, protecting patient privacy, and enabling AI model development, especially in areas with limited real datasets such as rare diseases. Applications of SD in paediatrics span various fields, including neonatology, oncology, radiology, and neurodevelopmental disorders. However, challenges persist, including potential data bias, ensuring accuracy and quality, privacy concerns, and the lack of standardized guidelines for data generation and validation.
Conclusions and future directions
While SD demonstrates potential in specific paediatric applications, such as improving AI early warning systems and augmenting datasets for rare conditions, its use requires a structured, actionable framework for evaluation. Future efforts should focus through multi-stakeholder engagement, on developing paediatric-specific guidelines, ensuring fair and safe use of SD, and addressing unique aspects of child development in data synthesis.
{"title":"Synthetic data generation in paediatrics and paediatric nursing: what, how, and why?","authors":"Elisabetta Mezzalira , Maria Paola Boaro , Giulia Reggiani , Riccardo Biondi , Gastone Castellani , Raffaella Colombatti","doi":"10.1016/j.ijmedinf.2025.106236","DOIUrl":"10.1016/j.ijmedinf.2025.106236","url":null,"abstract":"<div><h3>Introduction</h3><div>This paper explores the potential benefits and limitations of synthetic data (SD) in paediatrics, addressing the challenges of data scarcity and privacy concerns in paediatric research.</div></div><div><h3>Methodology</h3><div>A narrative literature review was conducted, searching PubMed and Scopus databases for relevant publications up to August 2025. The review focused on studies addressing the use, development, or application of SD in paediatric healthcare settings.</div></div><div><h3>Findings</h3><div>Synthetic data offers numerous benefits in paediatrics, including enhancing dataset diversity, protecting patient privacy, and enabling AI model development, especially in areas with limited real datasets such as rare diseases. Applications of SD in paediatrics span various fields, including neonatology, oncology, radiology, and neurodevelopmental disorders. However, challenges persist, including potential data bias, ensuring accuracy and quality, privacy concerns, and the lack of standardized guidelines for data generation and validation.</div></div><div><h3>Conclusions and future directions</h3><div>While SD demonstrates potential in specific paediatric applications, such as improving AI early warning systems and augmenting datasets for rare conditions, its use requires a structured, actionable framework for evaluation. Future efforts should focus through multi-stakeholder engagement, on developing paediatric-specific guidelines, ensuring fair and safe use of SD, and addressing unique aspects of child development in data synthesis.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"209 ","pages":"Article 106236"},"PeriodicalIF":4.1,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145913972","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-25DOI: 10.1016/j.ijmedinf.2025.106243
{"title":"Reviewer Acknowledgement 2025","authors":"","doi":"10.1016/j.ijmedinf.2025.106243","DOIUrl":"10.1016/j.ijmedinf.2025.106243","url":null,"abstract":"","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"207 ","pages":"Article 106243"},"PeriodicalIF":4.1,"publicationDate":"2025-12-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145866390","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-25DOI: 10.1016/j.ijmedinf.2025.106235
Wenhao Han , Xinyu Yang , Xin Li , Jiacheng Wang , Juan Liu , Wei Pang
Objective
Eye-tracking technology has been increasingly investigated as an objective approach for distinguishing individuals with Autism Spectrum Disorder (ASD) from typically developing (TD) individuals. Artificial intelligence and machine learning (ML) methods have been widely applied to support ASD diagnosis and treatment, and prior studies suggest that ML models leveraging eye-tracking data can achieve high diagnostic accuracy. This systematic review and meta-analysis aimed to evaluate the diagnostic performance of machine-learning models using eye-tracking data to distinguish children and adolescents with ASD from TD peers.
Methods
We systematically searched PubMed, Embase, Web of Science, IEEE Xplore, Scopus, and the Cochrane Library from inception to August 3, 2025. We included studies that applied ML methods to eye-tracking data to distinguish children with ASD from TD children. We extracted data on participant characteristics, model performance, eye-tracking protocols, and machine-learning algorithms. The review protocol was registered in PROSPERO (CRD420251162462).
Results
We identified 1,045 records, of which 25 studies were included in the meta-analysis. The included studies comprised 2,319 participants, with sample sizes ranging from 32 to 529 per study. The pooled accuracy, sensitivity, and specificity of machine-learning models using eye-tracking data to distinguish children with ASD from TD children were 85 % (95 % CI, 81–89 %), 86 % (95 % CI, 82–89 %), and 86 % (95 % CI, 79–91 %), respectively. These results suggest that eye-tracking–based machine-learning approaches have good diagnostic performance for identifying ASD.
Conclusion
Eye-tracking–based machine-learning approaches show considerable potential for distinguishing children with ASD from TD children. However, the robustness and generalizability of these findings are limited by the lack of external validation, small sample sizes, and substantial between-study heterogeneity. To establish generalizability, future research should prioritize standardized eye-tracking paradigms and large-scale, prospective, multicenter study designs with external validation. Such efforts may facilitate the translation of these models into clinical practice as objective and efficient adjunctive screening tools.
{"title":"Machine learning-based diagnosis of autism spectrum disorder in children and adolescents using eye-tracking data: a systematic review and meta-analysis","authors":"Wenhao Han , Xinyu Yang , Xin Li , Jiacheng Wang , Juan Liu , Wei Pang","doi":"10.1016/j.ijmedinf.2025.106235","DOIUrl":"10.1016/j.ijmedinf.2025.106235","url":null,"abstract":"<div><h3>Objective</h3><div>Eye-tracking technology has been increasingly investigated as an objective approach for distinguishing individuals with Autism Spectrum Disorder (ASD) from typically developing (TD) individuals. Artificial intelligence and machine learning (ML) methods have been widely applied to support ASD diagnosis and treatment, and prior studies suggest that ML models leveraging eye-tracking data can achieve high diagnostic accuracy. This systematic review and meta-analysis aimed to evaluate the diagnostic performance of machine-learning models using eye-tracking data to distinguish children and adolescents with ASD from TD peers.</div></div><div><h3>Methods</h3><div>We systematically searched PubMed, Embase, Web of Science, IEEE Xplore, Scopus, and the Cochrane Library from inception to August 3, 2025. We included studies that applied ML methods to eye-tracking data to distinguish children with ASD from TD children. We extracted data on participant characteristics, model performance, eye-tracking protocols, and machine-learning algorithms. The review protocol was registered in PROSPERO (CRD420251162462).</div></div><div><h3>Results</h3><div>We identified 1,045 records, of which 25 studies were included in the meta-analysis. The included studies comprised 2,319 participants, with sample sizes ranging from 32 to 529 per study. The pooled accuracy, sensitivity, and specificity of machine-learning models using eye-tracking data to distinguish children with ASD from TD children were 85 % (95 % CI, 81–89 %), 86 % (95 % CI, 82–89 %), and 86 % (95 % CI, 79–91 %), respectively. These results suggest that eye-tracking–based machine-learning approaches have good diagnostic performance for identifying ASD.</div></div><div><h3>Conclusion</h3><div>Eye-tracking–based machine-learning approaches show considerable potential for distinguishing children with ASD from TD children. However, the robustness and generalizability of these findings are limited by the lack of external validation, small sample sizes, and substantial between-study heterogeneity. To establish generalizability, future research should prioritize standardized eye-tracking paradigms and large-scale, prospective, multicenter study designs with external validation. Such efforts may facilitate the translation of these models into clinical practice as objective and efficient adjunctive screening tools.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106235"},"PeriodicalIF":4.1,"publicationDate":"2025-12-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885672","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-25DOI: 10.1016/j.ijmedinf.2025.106242
Janne Kommusaar , Silja Elunurm , Taridzo Chomutare , Mari Kangasniemi , Sanna Salanterä , Laura-Maria Peltonen
<div><h3>Background</h3><div>The rise of digital health data has expanded opportunities for data-driven innovation, yet privacy, legal and ethical barriers frame data sharing and collaborative artificial intelligence development. Federated Learning (FL) offers a privacy-preserving alternative, but current research considers mainly technical aspects. There is no end-to-end roadmap that integrates ethical, legal, technical and administrative principles tailored to FL projects in healthcare. This study addresses that gap by developing a roadmap to guide responsible and scalable FL research in the European context.</div></div><div><h3>Methods</h3><div>A multi-method participatory approach was used to develop a roadmap for scientific projects using FL on health data. The iterative process involved three phases. First, key questions were defined and existing evidence was explored through (i) a survey of domain experts (researchers, data governance specialists and infrastructure providers), (ii) a targeted literature review of FL applications in health research and (iii) systematic mapping of relevant EU-level legislation and policy frameworks. Evidence from these sources was synthesized to identify technical, organizational, legal and sustainability-related requirements for FL-based research. Second, preliminary roadmap components were refined through stakeholder engagement in an online workshop, where feasibility, scalability and sustainability considerations were explicitly discussed. Third, the roadmap was validated and iteratively refined by an expert panel through a structured group discussion, focusing on long-term sustainability, governance and transferability across research contexts. The process was carried out within a Baltic-Nordic collaboration in 2023–2025.</div></div><div><h3>Results</h3><div>The developed roadmap integrates ethical, legal, technical, administrative and sustainability-related considerations essential for applying FL to health data. It emphasizes the importance of multidisciplinary collaboration throughout the FL project lifecycle, with particular attention to long-term governance, scalability and reuse of infrastructures and practices. The process is structured into six phases: (1) Planning, (2) Execution refinement, (3) Data, (4) FL platform, (5) FL experiment and (6) Dissemination. Across these phases, sustainability is addressed through mechanisms such as regulatory alignment, shared governance models, capacity building and integration with existing research and health data infrastructures. By merging ethical, legal, technical and administrative aspects into a unified, end-to-end framework, the roadmap provides actionable, novel guidance beyond existing recommendations.</div></div><div><h3>Conclusions</h3><div>This work consolidates early lessons from FL in healthcare into a practical, step-by-step roadmap that integrates ethical, legal, technical and administrative aspects in the European context. By offering a shared
{"title":"A roadmap for federated learning projects using health data to guide sustainable artificial intelligence development in the European Union","authors":"Janne Kommusaar , Silja Elunurm , Taridzo Chomutare , Mari Kangasniemi , Sanna Salanterä , Laura-Maria Peltonen","doi":"10.1016/j.ijmedinf.2025.106242","DOIUrl":"10.1016/j.ijmedinf.2025.106242","url":null,"abstract":"<div><h3>Background</h3><div>The rise of digital health data has expanded opportunities for data-driven innovation, yet privacy, legal and ethical barriers frame data sharing and collaborative artificial intelligence development. Federated Learning (FL) offers a privacy-preserving alternative, but current research considers mainly technical aspects. There is no end-to-end roadmap that integrates ethical, legal, technical and administrative principles tailored to FL projects in healthcare. This study addresses that gap by developing a roadmap to guide responsible and scalable FL research in the European context.</div></div><div><h3>Methods</h3><div>A multi-method participatory approach was used to develop a roadmap for scientific projects using FL on health data. The iterative process involved three phases. First, key questions were defined and existing evidence was explored through (i) a survey of domain experts (researchers, data governance specialists and infrastructure providers), (ii) a targeted literature review of FL applications in health research and (iii) systematic mapping of relevant EU-level legislation and policy frameworks. Evidence from these sources was synthesized to identify technical, organizational, legal and sustainability-related requirements for FL-based research. Second, preliminary roadmap components were refined through stakeholder engagement in an online workshop, where feasibility, scalability and sustainability considerations were explicitly discussed. Third, the roadmap was validated and iteratively refined by an expert panel through a structured group discussion, focusing on long-term sustainability, governance and transferability across research contexts. The process was carried out within a Baltic-Nordic collaboration in 2023–2025.</div></div><div><h3>Results</h3><div>The developed roadmap integrates ethical, legal, technical, administrative and sustainability-related considerations essential for applying FL to health data. It emphasizes the importance of multidisciplinary collaboration throughout the FL project lifecycle, with particular attention to long-term governance, scalability and reuse of infrastructures and practices. The process is structured into six phases: (1) Planning, (2) Execution refinement, (3) Data, (4) FL platform, (5) FL experiment and (6) Dissemination. Across these phases, sustainability is addressed through mechanisms such as regulatory alignment, shared governance models, capacity building and integration with existing research and health data infrastructures. By merging ethical, legal, technical and administrative aspects into a unified, end-to-end framework, the roadmap provides actionable, novel guidance beyond existing recommendations.</div></div><div><h3>Conclusions</h3><div>This work consolidates early lessons from FL in healthcare into a practical, step-by-step roadmap that integrates ethical, legal, technical and administrative aspects in the European context. By offering a shared","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106242"},"PeriodicalIF":4.1,"publicationDate":"2025-12-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145879368","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-24DOI: 10.1016/j.ijmedinf.2025.106241
Hongbing Liu , Ying Yao , Ce Zong , Ke Zhang , Haixu Zhao , Yuan Song , Yuming Xu , Yuan Gao
Objective
A substantial proportion of patients (12 %–25 %) with recent small subcortical infarction (RSSI) suffer poor functional outcomes at 3 months. Despite the identification of prognostic factors, a significant gap exists in predictive modeling. This study aimed to develop and validate machine learning models to accurately predict 3-month functional status in this patient population.
Methods
This multicenter study prospectively enrolled 1576 patients diagnosed with RSSI. The primary cohort (n = 1126) was randomly split into a training set (70 %) and an internal validation set (30 %). An independent external cohort (n = 450) was used for further validation. The primary outcome was an unfavorable functional status at 3 months, defined as a modified Rankin Scale (mRS) score ≥3. The Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model was employed for feature selection from demographic, clinical, laboratory, and imaging variables. 8 supervised machine learning models were developed and compared. Model performance was rigorously evaluated in the validation cohorts using the Area Under the Receiver Operating Characteristic Curve (AUC) for discrimination, calibration curves for consistency, and Decision Curve Analysis (DCA) for clinical utility. The optimal model was interpreted using SHapley Additive exPlanations (SHAP).
Results
LASSO regression identified 8 features with non-zero coefficients for predicting outcomes: NIHSS, proximal RSSI (pRSSI), glucose, stress hyperglycemia ratio (SHR), neutrophil-to-lymphocyte ratio (NLR), age, systolic blood pressure (SBP), and LDL-C. Among the eight developed models, the CatBoost model demonstrated the best performance. It achieved the highest AUC in the training set (0.961), the internal validation cohort (0.940), and the external validation cohort (0.875). The CatBoost model also showed excellent calibration and provided the greatest net benefit across a wide range of threshold probabilities in DCA for both validation cohorts. SHAP analysis identified the NIHSS score as the most significant predictor of unfavorable outcomes, followed by pRSSI, Glucose, SHR, and NLR. A publicly accessible web tool based on the model is available at: https:// predictrssi.streamlit.app.
Conclusion
This study successfully developed and validated a robust CatBoost machine learning model that accurately predicts 3-month functional outcomes in patients with RSSI using eight readily accessible features. This model, which outperforms seven other machine learning algorithms, is available as a user-friendly web application to aid clinicians in early risk stratification and personalized patient management.
{"title":"Development and validation of a machine learning model to predict functional outcomes in patients with recent small subcortical infarction","authors":"Hongbing Liu , Ying Yao , Ce Zong , Ke Zhang , Haixu Zhao , Yuan Song , Yuming Xu , Yuan Gao","doi":"10.1016/j.ijmedinf.2025.106241","DOIUrl":"10.1016/j.ijmedinf.2025.106241","url":null,"abstract":"<div><h3>Objective</h3><div>A substantial proportion of patients (12 %–25 %) with recent small subcortical infarction (RSSI) suffer poor functional outcomes at 3 months. Despite the identification of prognostic factors, a significant gap exists in predictive modeling. This study aimed to develop and validate machine learning models to accurately predict 3-month functional status in this patient population.</div></div><div><h3>Methods</h3><div>This multicenter study prospectively enrolled 1576 patients diagnosed with RSSI. The primary cohort (n = 1126) was randomly split into a training set (70 %) and an internal validation set (30 %). An independent external cohort (n = 450) was used for further validation. The primary outcome was an unfavorable functional status at 3 months, defined as a modified Rankin Scale (mRS) score ≥3. The Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model was employed for feature selection from demographic, clinical, laboratory, and imaging variables. 8 supervised machine learning models were developed and compared. Model performance was rigorously evaluated in the validation cohorts using the Area Under the Receiver Operating Characteristic Curve (AUC) for discrimination, calibration curves for consistency, and Decision Curve Analysis (DCA) for clinical utility. The optimal model was interpreted using SHapley Additive exPlanations (SHAP).</div></div><div><h3>Results</h3><div>LASSO regression identified 8 features with non-zero coefficients for predicting outcomes: NIHSS, proximal RSSI (pRSSI), glucose, stress hyperglycemia ratio (SHR), neutrophil-to-lymphocyte ratio (NLR), age, systolic blood pressure (SBP), and LDL-C. Among the eight developed models, the CatBoost model demonstrated the best performance. It achieved the highest AUC in the training set (0.961), the internal validation cohort (0.940), and the external validation cohort (0.875). The CatBoost model also showed excellent calibration and provided the greatest net benefit across a wide range of threshold probabilities in DCA for both validation cohorts. SHAP analysis identified the NIHSS score as the most significant predictor of unfavorable outcomes, followed by pRSSI, Glucose, SHR, and NLR. A publicly accessible web tool based on the model is available at: <span><span>https:// predictrssi.streamlit.app</span><svg><path></path></svg></span>.</div></div><div><h3>Conclusion</h3><div>This study successfully developed and validated a robust CatBoost machine learning model that accurately predicts 3-month functional outcomes in patients with RSSI using eight readily accessible features. This model, which outperforms seven other machine learning algorithms, is available as a user-friendly web application to aid clinicians in early risk stratification and personalized patient management.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106241"},"PeriodicalIF":4.1,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145851363","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-24DOI: 10.1016/j.ijmedinf.2025.106238
Yuto Arai , Yoshiyasu Takefuji
{"title":"Limitations of SHAP-based interpretability in sepsis progression models and paths to more robust feature validation","authors":"Yuto Arai , Yoshiyasu Takefuji","doi":"10.1016/j.ijmedinf.2025.106238","DOIUrl":"10.1016/j.ijmedinf.2025.106238","url":null,"abstract":"","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106238"},"PeriodicalIF":4.1,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-23DOI: 10.1016/j.ijmedinf.2025.106232
Petra Hospodková , Jan Bruthans , Adéla Englová
Introduction
The Patient Summary (PS), a standardized subset of the electronic health record is designed to provide essential patient information for use in emergencies, unplanned care, and cross-border healthcare. While its technical development has progressed across Europe, little is known about real-world PS adoption and physician perceptions at the national level. This study explores the awareness, usage, and perceived barriers to the PS adoption among Czech physicians.
Methods
A cross-sectional online survey was distributed to all registered physicians in the Czech Republic between February and March 2025. The questionnaire assessed demographic characteristics, PS usage patterns, perceived benefits and barriers, and alignment with clinical practice. Descriptive statistics were calculated, and non-parametric tests (Wilcoxon rank-sum, Kruskal–Wallis) were used to examine differences by years of experience and medical specialty.
Results
A total of 1,739 responses were received (response rate: 4.14 %). Most respondents (66.4 %) reported not using the PS at all, and 72.1 % were unaware that their electronic medical record could be connected to the National Contact Point for eHealth. Only 1.7 % reported a current connection. There was no significant difference in PS use by years of clinical experience (P = 0.391), but a significant difference was observed across specialties (P < 0.001), with the highest usage reported in intensive care medicine and internal medicine.
Discussion and conclusion
Despite recognized benefits, PS usage remains low in the Czech Republic, largely due to limited awareness and system integration. Targeted policy measures, improved communication, and enhanced digital training are needed to support effective adoption.
{"title":"Physicians’ attitudes toward the patient summary in the Czech Republic: A national cross-sectional survey on awareness, use, and barriers","authors":"Petra Hospodková , Jan Bruthans , Adéla Englová","doi":"10.1016/j.ijmedinf.2025.106232","DOIUrl":"10.1016/j.ijmedinf.2025.106232","url":null,"abstract":"<div><h3>Introduction</h3><div>The Patient Summary (PS), a standardized subset of the electronic health record is designed to provide essential patient information for use in emergencies, unplanned care, and cross-border healthcare. While its technical development has progressed across Europe, little is known about real-world PS adoption and physician perceptions at the national level. This study explores the awareness, usage, and perceived barriers to the PS adoption among Czech physicians.</div></div><div><h3>Methods</h3><div>A cross-sectional online survey was distributed to all registered physicians in the Czech Republic between February and March 2025. The questionnaire assessed demographic characteristics, PS usage patterns, perceived benefits and barriers, and alignment with clinical practice. Descriptive statistics were calculated, and non-parametric tests (Wilcoxon rank-sum, Kruskal–Wallis) were used to examine differences by years of experience and medical specialty.</div></div><div><h3>Results</h3><div>A total of 1,739 responses were received (response rate: 4.14 %). Most respondents (66.4 %) reported not using the PS at all, and 72.1 % were unaware that their electronic medical record could be connected to the National Contact Point for eHealth. Only 1.7 % reported a current connection. There was no significant difference in PS use by years of clinical experience (P = 0.391), but a significant difference was observed across specialties (P < 0.001), with the highest usage reported in intensive care medicine and internal medicine.</div></div><div><h3>Discussion and conclusion</h3><div>Despite recognized benefits, PS usage remains low in the Czech Republic, largely due to limited awareness and system integration. Targeted policy measures, improved communication, and enhanced digital training are needed to support effective adoption.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106232"},"PeriodicalIF":4.1,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145829159","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-23DOI: 10.1016/j.ijmedinf.2025.106239
Priyadharsini Ramamurthy , Zheng Han , Dursun Delen , Zhuqi Miao , Andrew Gin , Xiao Luo , William Paiva
Background
Traumatic brain injury (TBI) is a major risk factor for neurological disorders, including post-traumatic epilepsy (PTE), a debilitating condition associated with significant long-term consequences. The prognosis of PTE occurrence remains challenging due to the complex pathophysiology of PTE and the impracticality of traditional blood biomarker- or imaging-based screening for large populations. This study proposes a graph-based deep learning approach that leverages electronic health records (EHR) to enhance the predictive assessment of PTE risk.
Methods
We utilized Oracle Real-World Data (ORWD) to construct a Heterogeneous Graph Attention Network (HeteroGAT) that contains patient and diagnosis nodes, with temporal information represented using patients-to-diagnosis edges, and comorbidity connectivity embedded using diagnosis-to-diagnosis edges. The HeteroGAT was trained on a cohort of 1,598,998 TBI-only patients and 102,687 individuals who developed epilepsy after TBI. Model performance was evaluated using sensitivity, specificity, macro F1-score, and area under the receiver operating characteristic curve (AUC-ROC), benchmarked against traditional machine learning models. Attention scores of nodes were used to evaluate node importance. The capabilities of the HeteroGATs trained to differentiate early vs late PTE patients following TBI were also assessed.
Results
HeteroGAT significantly outperformed conventional models in PTE prediction by effectively integrating demographic data and comorbidity profiles spanning from 20 to 500 distinct conditions. The model’s multi-head attention mechanisms, in combination with learned comorbidity connectivity, enhanced its ability to capture complex dependencies within EHR data. HeteroGAT achieved an AUC-ROC of 0.80, outperforming the best-performing traditional model, random forest (AUC-ROC = 0.77). HeteroGAT also demonstrated capabilities in differentiating early and late PTEs. Ranking of nodes based on attention scores also identified predictors of PTE that are clinically relevant.
Conclusion
By modeling sparse EHR data through patient encounter embeddings, HeteroGAT effectively captures temporal and relational patterns in comorbidities critical for PTE prediction. Our findings highlight the potential of graph-based deep learning models, synergized with large-scale EHR data, in advancing personalized risk assessment, ultimately addressing the urgent need for more precise and proactive management of PTE in TBI patients.
{"title":"Graph attention network with comorbidity connectivity embedding for post-traumatic epilepsy risk prediction using sparse time-series electronic health records","authors":"Priyadharsini Ramamurthy , Zheng Han , Dursun Delen , Zhuqi Miao , Andrew Gin , Xiao Luo , William Paiva","doi":"10.1016/j.ijmedinf.2025.106239","DOIUrl":"10.1016/j.ijmedinf.2025.106239","url":null,"abstract":"<div><h3>Background</h3><div>Traumatic brain injury (TBI) is a major risk factor for neurological disorders, including post-traumatic epilepsy (PTE), a debilitating condition associated with significant long-term consequences. The prognosis of PTE occurrence remains challenging due to the complex pathophysiology of PTE and the impracticality of traditional blood biomarker- or imaging-based screening for large populations. This study proposes a graph-based deep learning approach that leverages electronic health records (EHR) to enhance the predictive assessment of PTE risk.</div></div><div><h3>Methods</h3><div>We utilized Oracle Real-World Data (ORWD) to construct a Heterogeneous Graph Attention Network (HeteroGAT) that contains patient and diagnosis nodes, with temporal information represented using patients-to-diagnosis edges, and comorbidity connectivity embedded using diagnosis-to-diagnosis edges. The HeteroGAT was trained on a cohort of 1,598,998 TBI-only patients and 102,687 individuals who developed epilepsy after TBI. Model performance was evaluated using sensitivity, specificity, macro F1-score, and area under the receiver operating characteristic curve (AUC-ROC), benchmarked against traditional machine learning models. Attention scores of nodes were used to evaluate node importance. The capabilities of the HeteroGATs trained to differentiate early vs late PTE patients following TBI were also assessed.</div></div><div><h3>Results</h3><div>HeteroGAT significantly outperformed conventional models in PTE prediction by effectively integrating demographic data and comorbidity profiles spanning from 20 to 500 distinct conditions. The model’s multi-head attention mechanisms, in combination with learned comorbidity connectivity, enhanced its ability to capture complex dependencies within EHR data. HeteroGAT achieved an AUC-ROC of 0.80, outperforming the best-performing traditional model, random forest (AUC-ROC = 0.77). HeteroGAT also demonstrated capabilities in differentiating early and late PTEs. Ranking of nodes based on attention scores also identified predictors of PTE that are clinically relevant.</div></div><div><h3>Conclusion</h3><div>By modeling sparse EHR data through patient encounter embeddings, HeteroGAT effectively captures temporal and relational patterns in comorbidities critical for PTE prediction. Our findings highlight the potential of graph-based deep learning models, synergized with large-scale EHR data, in advancing personalized risk assessment, ultimately addressing the urgent need for more precise and proactive management of PTE in TBI patients.</div></div>","PeriodicalId":54950,"journal":{"name":"International Journal of Medical Informatics","volume":"208 ","pages":"Article 106239"},"PeriodicalIF":4.1,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号