BMC Medical Informatics and Decision Making最新文献

{"title":"Performance of large language models in mitral valve surgery patient education: a comparative analysis.","authors":"Banu Bahriye Akdag, Mehmet Senel Bademci, Ihsan Peker, Guven Okay Karaca, Cagrı Kandemir, Barcin Özcem, Huseyin Durmaz, Meryem Cakır, Irem Ozcetin, Hidayet Onur Selcuk","doi":"10.1186/s12911-026-03662-3","DOIUrl":"10.1186/s12911-026-03662-3","url":null,"abstract":"<p><strong>Background: </strong>Large language models (LLMs), a form of artificial intelligence, are increasingly being utilized in healthcare to support patient education and information delivery. The aim of this study was to perform a comparative analysis of five different LLMs (i.e., ChatGPT-4o, Claude 3.7 Sonnet, Gemini 2.5 Pro Preview, DeepSeek-V3, and Microsoft Copilot) in terms of accuracy, completeness, and readability, based on their responses to frequently asked questions in preoperative patient education for mitral valve surgery (MVS).</p><p><strong>Methods: </strong>A standardized questionnaire comprising seven frequently asked questions by patients prior to MVS was developed. Prompting procedures and model parameters were fully reported to support reproducibility. These questions were presented to each LLM in an identical manner. The responses were evaluated by two academic experts in cardiac surgery using structured assessment criteria across three main dimensions: accuracy, completeness, and readability. For the readability analysis, the Simplified Measure of Gobbledygook (SMOG) Index and the Flesch-Kincaid Grade Level (FKGL) scale were utilized.</p><p><strong>Results: </strong>The ChatGPT-4o and Gemini 2.5 Pro Preview models received statistically significantly higher scores than Claude 3.7 Sonnet and Microsoft Copilot for both accuracy (median 5 for ChatGPT-4o and Gemini 2.5 Pro Preview vs. 4 for Claude 3.7 Sonnet and Microsoft Copilot, p < 0.001) and completeness (median 5 for Gemini 2.5 Pro Preview vs. 3 for Claude 3.7 Sonnet, p < 0.001). Claude 3.7 Sonnet achieved the highest readability scores, with significantly lower SMOG (10.90 for Claude 3.7 Sonnet vs. 12.24 for ChatGPT-4o, p = 0.006) and FKGL (8.0 for Claude 3.7 Sonnet vs. 9.04 for ChatGPT-4o, p = 0.004) scores, indicating simpler and more comprehensible sentence structures. Significant differences were observed among the evaluated models across all three assessment dimensions (p < 0.001 for all comparisons).</p><p><strong>Conclusions: </strong>The LLMs represent valuable supplementary tools in patient education processes. However, their implementation in clinical practice must be carefully evaluated, particularly with regard to accuracy and completeness. This study highlights the potential applicability of ChatGPT-4o and Claude 3.7 Sonnet models for preoperative patient education in MVS, while emphasizing that all LLMs should be used under the supervision and guidance of healthcare professionals. For LLMs to be reliably utilized in the medical field, improvement in medical accuracy and standardization are essential.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13579825/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148350725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Clinical subphenotypes in septic patients with new-onset atrial fibrillation: validation and parsimonious classifier model development.","authors":"Lin Zhang, Jiangjing Li, Xiaolin Shang, Yongfei Liu, Zhiyang Chen, Hongxia Xiang, Heming Zhang","doi":"10.1186/s12911-026-03652-5","DOIUrl":"10.1186/s12911-026-03652-5","url":null,"abstract":"<p><strong>Background: </strong>The substantial heterogeneity among septic patients with new‑onset atrial fibrillation (NOAF) complicates intensive care unit (ICU) management, and the absence of a parsimonious subphenotype classifier has impeded the implementation of personalized therapeutic strategies. In this study, we aim to identify the subphenotype of septic patients with NOAF, develop a concise classifier model, and reveal the differential efficacy of heart-rate control therapies across these subphenotypes.</p><p><strong>Methods: </strong>This retrospective study utilized data from the Medical Information Mart in Intensive Care (MIMIC)-IV and MIMIC-III database. Consensus clustering based on hierarchical clustering was employed for subphenotype derivation. Nine supervised classifiers were employed to construct the subphenotype classifier, including random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), partial least squares (PLS), neural network (NN), naïve Bayes (NB), linear discriminant analysis (LDA), least absolute shrinkage and selection operator (LASSO), and adaptive boosting (AdaBoost.M1). Subphenotype-specific propensity score-derived stabilized inverse probability of treatment weights was used to weight comparisons of intensive care unit and hospital lengths of stay and discharge dispositions across subphenotypes.</p><p><strong>Results: </strong>Among 1535 septic patients with NOAF, three distinct subphenotypes emerged: Phenotype A was the mildest subgroup; Phenotype B exhibited pronounced metabolic acidosis, highest anion gap, severe renal impairment, and elevated severity scores; Phenotype C showed intermediate laboratory values and score profiles with notable hyperchloremia and inflammatory markers. A parsimonious model established by SVM showed the best effectiveness. HR-control therapy was associated with subphenotype-specific reductions in hospital and ICU lengths of stay and more favorable discharge dispositions in Phenotype A, while more favorable survival outcomes were observed across all three subphenotypes in IPTW-weighted analyses.</p><p><strong>Conclusion: </strong>Cluster analysis revealed three clinically relevant subphenotypes among septic patients with NOAF, each demonstrating distinct clinical outcomes and heterogeneous responses to heart rate control therapy.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13576258/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148350835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and external validation of an interpretable machine learning model for predicting prolonged postoperative ICU length of stay in coronary artery bypass grafting patients using MIMIC-IV 3.1 and eICU-CRD 2.0.","authors":"Dayan Liu, Pengyu Lu, Yulan Meng, Xianglong Liu, Wei Huang","doi":"10.1186/s12911-026-03659-y","DOIUrl":"10.1186/s12911-026-03659-y","url":null,"abstract":"<p><strong>Background: </strong>Prolonged postoperative intensive care unit (ICU) length of stay (LOS) after coronary artery bypass grafting (CABG) drives resource use yet remains difficult to anticipate at the 24-hour ICU evaluation moment when extended monitoring beyond 72 h is first considered. We developed and externally validated an interpretable machine-learning decision-support calculator for this deployment moment.</p><p><strong>Methods: </strong>Adult CABG patients were identified from MIMIC-IV 3.1 (n = 6,919; 7:3 stratified split for development) and eICU-CRD 2.0 (n = 5,972; external validation). The outcome was prolonged ICU LOS (> 3 days). Elastic Net plus Boruta selected eight bedside features collected within the first 24 h of ICU admission: 24-hour fluid intake, Charlson Comorbidity Index (CCI), Sequential Organ Failure Assessment (SOFA) score, Simplified Acute Physiology Score II (SAPS-II), Glasgow Coma Scale (GCS), vasopressor use, congestive heart failure, and atrial fibrillation. Nine machine-learning algorithms were compared by 10-fold cross-validation paired t-tests with Bonferroni and Benjamini-Hochberg correction. Calibration metrics included Hosmer-Lemeshow test, Integrated Calibration Index (ICI), expected-to-observed ratio, intercept, and slope (full Methods). SHapley Additive exPlanations (SHAP) provided per-patient feature attribution.</p><p><strong>Results: </strong>CatBoost was selected as the deployed model. On the MIMIC-IV internal test set (n = 2,076), the area under the receiver-operating-characteristic curve (AUC) was 0.7739 (95% confidence interval 0.7379-0.8099), Hosmer-Lemeshow p = 0.224, calibration slope 0.973. On the eICU-CRD external cohort, AUC was 0.6452 (95% CI 0.6311-0.6602), calibration slope 0.998, ICI 0.023. At the prevalence-anchored threshold of 0.30, sensitivity was 0.55, specificity 0.65, positive predictive value 0.40, and negative predictive value 0.77. Decision Curve Analysis showed positive net benefit over treat-all and treat-none across t = 0.20-0.40. SHAP top-3 features were 24-hour fluid intake, CCI, and atrial fibrillation.</p><p><strong>Conclusions: </strong>The model provides decision support at the post-CABG 24-hour ICU evaluation moment with modest discrimination and well-calibrated probabilities at deployment. The deployed online calculator renders per-patient feature contributions transparent at the bedside via SHAP; prospective validation is required before clinical deployment.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13579721/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148337778","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Data analysis tool for identifying multidimensional health profiles associated with frailty in older adults.","authors":"Alondra Sánchez-Salinas, Anilu Franco-Arcega, Manuel A Ojeda-Misses, Lilian E Bosques-Brugada, Christian J Flores-Pérez, Aldo Márquez-Grajales","doi":"10.1186/s12911-026-03645-4","DOIUrl":"10.1186/s12911-026-03645-4","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Background: &lt;/strong&gt;Frailty in older adults is a multidimensional condition associated with physical, psychological, and sociodemographic factors that may affect functionality and quality of life during aging. The analysis of these factors often requires the management of heterogeneous information, which can represent a challenge for specialists during evaluation and follow-up processes. In this context, computational tools and data-driven approaches may support the organization and exploratory analysis of multidimensional health information. This study presents the development of a web-based computational system designed to support specialists in the collection, management, and analysis of health-related data from older adults in the State of Hidalgo, México.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Method: &lt;/strong&gt;A web platform was developed to register and organize sociodemographic, psychological, and physical information from older adults. The system integrated questionnaires, physical measurements, statistical analysis tools, and computational intelligence techniques within a single environment. Data from 210 participants were collected and analyzed using exploratory statistical methods, feature selection techniques, and K-means clustering to identify patterns associated with functional and health-related characteristics. Internal validation metrics, including the Silhouette coefficient, Davies-Bouldin Index, and Calinski-Harabasz Index, were used to support cluster evaluation.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;The proposed system facilitated the structured storage, visualization, and exploratory analysis of multidimensional health information. The clustering analysis identified two differentiated groups characterized by distinct physical and psychological profiles. Differences were observed in variables such as muscular strength, calf circumference, body mass index, gratitude, and functional self-perception. The interpretation of the obtained profiles considered the demographic composition of the groups, particularly the predominance of women in one cluster, which may influence physical measurements associated with aging. Additionally, the usability assessment conducted with specialists using the SUS instrument indicated an acceptable level of usability for exploratory and research support purposes.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;The results suggest that integrating computational intelligence and statistical analysis techniques into a unified platform may support the exploratory assessment of multidimensional factors related to aging and frailty in older adults. The proposed system does not aim to establish clinical diagnoses, but rather to provide specialists with an additional resource for organizing and interpreting complex health-related information. This approach may contribute to the development of data-driven tools to support gerontological research and comprehensive evaluation processes in older populations.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Trial registration: ","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13573356/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148337723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and validation of machine learning models for early diagnosis of hemophagocytic lymphohistiocytosis in pediatric Epstein-Barr virus infection.","authors":"Yingying Ye, Yuqin Li, Yuewen Su, Meng Cao, Jiaxue Liu, Jiaying Ding, Shaoyan Hu, Jianmei Tian, Weifang Zhou","doi":"10.1186/s12911-026-03658-z","DOIUrl":"10.1186/s12911-026-03658-z","url":null,"abstract":"<p><strong>Objective: </strong>To establish a machine learning (ML) model for the early diagnosis of Epstein-Barr virus-associated hemophagocytic lymphohistiocytosis (EBV-HLH) based on clinical features and to utilize the Shapley Additive Explanations (SHAP) method to interpret the ML model, thereby providing reliable factors for diagnosing EBV-HLH.</p><p><strong>Methods: </strong>We collected clinical data from 1,026 children with Epstein-Barr virus (EBV) infection who were hospitalized at Children's Hospital of Soochow University from October 2017 to September 2024. First, we compared the clinical data of Epstein-Barr virus-associated infectious mononucleosis (EBV-IM) and EBV-HLH through univariate analysis and least absolute shrinkage and selection operator (LASSO) regression to select key features for machine learning training. Subsequently, we applied six machine learning algorithms and logistic regression to build diagnostic models, and the optimal model was selected based on multiple evaluation metrics. Finally, we utilized the Shapley Additive Explanations (SHAP) algorithm to clarify the importance of variables in the model to facilitate its application in clinical settings.</p><p><strong>Results: </strong>Among the six machine learning models and logistic regression evaluated, the Extreme Gradient Boosting (XGBoost) model demonstrated the strongest discrimination ability, with an area under the receiver operating characteristic curve (AUC) of 0.9775, sensitivity of 0.9461 and specificity of 0.9784. The SHAP analysis indicated that the most important predictors for EBV-HLH were D-dimer, cervical lymphadenopathy (CLA), gamma-glutamyl transferase (GGT), lactate dehydrogenase (LDH), and CD3 + CD4+ T cells.</p><p><strong>Conclusions: </strong>The XGBoost model demonstrated excellent predictive performance for the early identification of EBV-HLH in children. Compared with other models, it achieved higher sensitivity and may serve as a promising decision-support tool pending external validation.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528047/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148350771","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Construction and validation of a predictive nomogram for 28-day mortality in critically ill patients with toxic encephalopathy.","authors":"Mingzhu Wang, Chengchao Peng, Zeguang Ye, Min Lu, Lehao Ren","doi":"10.1186/s12911-026-03655-2","DOIUrl":"10.1186/s12911-026-03655-2","url":null,"abstract":"<p><strong>Background: </strong>Toxic encephalopathy (TE) is a serious neurological complication among critically ill patients and is associated with considerable disability and mortality. Early risk prediction is essential for timely intervention and improved outcomes. However, evidence on prognostic modeling in TE remains limited. The primary objective of this study was to develop and validate a prognostic nomogram for predicting 28-day mortality in critically ill patients diagnosed with TE and to evaluate its clinical utility.</p><p><strong>Methods: </strong>Adult patients diagnosed with TE were retrospectively selected from the MIMIC-IV database based on relevant International Classification of Diseases (ICD) codes. Baseline demographic, clinical, and laboratory variables were extracted, and candidate predictors were selected through univariate logistic regression analyses based on clinical relevance and statistical significance. Subsequently, a multivariable logistic regression model was developed and presented as a nomogram. The value of the predictive model was evaluated using the area under the ROC curve (AUC), calibration curve, and decision curve analysis (DCA). Its potential clinical applicability was further assessed with a clinical impact curve (CIC).</p><p><strong>Results: </strong>A total of 3039 patients with TE were included, among whom 526 (17.3%) died within 28 days. The cohort was randomly divided into training and validation cohorts at a 7:3 ratio. The final model incorporated age (OR = 1.022, 95% CI: 1.013-1.032), heart rate (OR = 1.007, 95% CI: 1.001-1.013), hemoglobin (OR = 0.904, 95% CI: 0.852-0.958), albumin (OR = 0.677, 95% CI: 0.546-0.836), PaO₂ (OR = 0.998, 95% CI: 0.996-0.999), SOFA score (OR = 1.094, 95% CI: 1.043-1.148), APACHE II score (OR = 1.038, 95% CI: 1.011-1.066), and acute kidney injury (OR = 1.520, 95% CI: 1.132-2.501). The nomogram exhibited acceptable discriminative ability, with AUC values of 0.750 (95% CI: 0.725-0.776) and 0.757 (95% CI: 0.715-0.799) in the training and validation cohorts, respectively, along with satisfactory calibration. Furthermore, decision curve analysis (DCA) and clinical impact curve (CIC) results supported its clinical utility for individualized risk stratification.</p><p><strong>Conclusion: </strong>Utilizing the MIMIC-IV database, we constructed and validated a prognostic nomogram to predict 28-day mortality in critically ill patients with TE. This tool may help clinicians identify high-risk individuals and optimize monitoring and supportive care.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13563740/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148337566","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards an information-centric architecture framework for health information logistics: a design science research study.","authors":"Øivind Skeidsvoll Solvang, Francis Odeh, Conceição Granja, Terje Solvoll","doi":"10.1186/s12911-026-03642-7","DOIUrl":"10.1186/s12911-026-03642-7","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Background: &lt;/strong&gt;Integrated care depends on timely awareness of parallel treatment activities and safe access to distributed patient records. However, multimorbid patients often navigate complex, non-linear pathways across care levels, resulting in information fragmentation that compromises decision-making, safety, and coordination. Existing electronic health record systems and interoperability initiatives offer limited support for pathway-oriented information logistics, particularly with respect to governance, provenance, and the integration of the patient's voice. This study aims to design a conceptual information-centric architecture framework to address these systemic gaps in awareness and access.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methods: &lt;/strong&gt;Applying Design Science Research (DSR), the study synthesized rigor from a scoping review with relevance derived from a longitudinal analysis of medical records from 14 multimorbid patients (1954-2024) in Northern Norway. A representative 19-year user scenario, validated by clinical expertise, together with stakeholder and root-cause analyses, informed the identification of recurrent failures in information logistics. High-level non-functional and functional requirements were iteratively derived, foregrounding Information Governance (IG), Data Management (DM), Security and Privacy, and Health Information Quality (HIQ) as essential socio-technical constraints. The resulting architecture was examined through a scenario-based demonstration and an early-stage ex-ante evaluation.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;The study produced a prescriptive information-centric architecture framework (Meta-artifact) organized into layered, modular components that separate user interaction, pathway logic, and secure storage. Information tokens, consisting of metadata, clinical codes, and location pointers, enable governed, read-only virtual access to source records, reducing duplication and avoiding traditional replication burdens. HIQ operationalizes DM/IG by enforcing requirements for accuracy, currency, consistency, completeness, and contextual relevance across inter-organizational access. In a hospitalization-discharge scenario, the architecture provided timely notifications, governed access to distributed records, and transparent consent and audit controls, demonstrating conceptual feasibility and alignment with the identified problem and requirements.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;This study contributes early-stage prescriptive design knowledge to improve inter-organizational information logistics. By embedding IG, DM, HIQ, and Security and Privacy as structural pillars and decoupling information from application constraints, the architecture provides a governance-aligned foundation for pathway-oriented coordination, proactive awareness, and safe, on-demand virtual access. This information-centric, token-based architecture, with a modular, standards-neutral design, supports adaptability and provides a foundatio","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13573431/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148337768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An integrated evaluation framework for synthetic clinical data in severely imbalanced settings: fidelity, privacy-risk profiling, and diagnostic utility.","authors":"Youngtae Kim, Jungwoo Lee, Sang Baek Koh, Kyu Hee Lee","doi":"10.1186/s12911-026-03648-1","DOIUrl":"10.1186/s12911-026-03648-1","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Background: &lt;/strong&gt;The development of clinical artificial intelligence models is constrained by limited access to high-quality electronic health record data, a challenge that is particularly pronounced in rare and highly imbalanced clinical cohorts. Synthetic data generation has been proposed as a strategy to mitigate data-sharing barriers. However, an integrated evaluation framework that jointly examines distributional fidelity, diagnostic behavior, and privacy risk under such conditions remains lacking.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methods: &lt;/strong&gt;We developed an integrated evaluation framework to assess variational autoencoders (VAE) and conditional generative adversarial networks (CTGAN). The framework jointly characterizes distributional fidelity, privacy risk, and diagnostic behavior using structured electronic health record data from the KoGES cohort, with disease prevalence ranging from 0.8% to 7.5%, adopting a prevalence-aware approach in which evaluation metrics are stratified by disease-specific class frequency. To address the sensitivity of p-value-based tests in large samples, distance-based metrics, including Jensen-Shannon divergence and Wasserstein distance, were employed. Diagnostic behavior was evaluated using XGBoost, random forest, and logistic regression classifiers, with emphasis on minority-class-sensitive metrics such as recall and Macro-F1.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;In multivariate structural analyses, the correlation similarity between empirical and synthetic data was 0.794 for VAE-generated data and 0.667 for CTGAN-generated data. Across diseases with moderate outcome prevalence, multivariate and stratified distributions exhibited numerical overlap between VAE-generated and empirical data. In diagnostic evaluations, classifiers trained on empirical data alone yielded zero recall for the rarest outcome (0.8% prevalence), whereas CTGAN-trained classifiers produced non-zero recall values at the cost of reduced overall accuracy. Across evaluated threat models, membership inference attack performance remained near the random-guessing reference (AUROC ≈ 0.500).&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;This study presents an integrated, prevalence-aware evaluation framework for synthetic clinical data that systematically identified failure modes undetectable by conventional single-metric approaches, including minority-class metric instability, heterogeneous tail-end privacy exposure, and qualitative divergence in membership score calibration. The evaluated generative models exhibited distinct trade-off profiles: VAE preserved multivariate structure while exhibiting lower proximity-based exposure under the evaluated threat model, whereas CTGAN achieved higher minority-class detection at the cost of structural fidelity. Supplementary augmentation experiments confirmed that increased synthetic data volume does not uniformly improve minority-class detection under extreme prevalence constraints. These findings demonstrate that num","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13560368/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148327157","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Machine learning for predicting institutionalization and mortality risks among older home care recipients with routinely collected need assessment data: explainable AI for long-term care.","authors":"Hongsoo Kim, Jimin Yoon, Hyerim Noh, Sunghun Yun, Seungyeon Chun, Woojoo Lee","doi":"10.1186/s12911-026-03641-8","DOIUrl":"10.1186/s12911-026-03641-8","url":null,"abstract":"<p><strong>Background: </strong>Identifying older home care recipients at risk of institutionalization in advance is crucial for providing preventive services. Supporting the decision-making of care workers in prioritizing their limited resources would be useful for promoting an aging-in-place policy.</p><p><strong>Objective: </strong>This study aims to develop and interpret explainable machine learning (ML) methods for predicting institutionalization and mortality as competing events among older adults newly entering the Long-Term Care Insurance (LTCI) home care system.</p><p><strong>Materials and methods: </strong>We explored and evaluated ML algorithms using routinely collected standardized functional assessment data to examine the care needs of 36,526 older beneficiaries enrolled in a nationwide public LTCI program. Individuals were classified according to whether institutionalization or death occurred first during follow-up. Accordingly, we employed multinomial logistic regression (MLR), least absolute shrinkage and selection operator (LASSO), extreme gradient boosting (XGBoost), and random forest (RF) algorithms to predict institutionalization and death in a multinomial approach. The performance of the model was assessed via the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and expected calibration error (ECE). Youden's J statistic was used to determine the cutoff point.</p><p><strong>Results: </strong>Among newly enrolled home care beneficiaries, 10.4% were institutionalized in long-term care facilities within six months, increasing to 18.8% within 12 months after the initiation of long-term care services. The LASSO model achieved the best performance, with the most predictive features of institutionalization including eating nonfood items which marked the advanced dementia, need for suctioning (aspiration care) and dementia diagnosis. The most predictive features of mortality included need for oxygen therapy, cancer pain management and diabetic foot care.</p><p><strong>Conclusion: </strong>The newly developed algorithm may be implemented to examine the care needs of older home care users without additional risk assessment and for early intervention to reduce institutionalization risks.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13560349/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148327136","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards Electronic Medical Record (EMR) utilization in developing countries: a conceptual model based on systematic literature review.","authors":"Abolfazl Rezaei, Maryam Jahanbakhsh, Hamid Reza Dehghan","doi":"10.1186/s12911-026-03488-z","DOIUrl":"10.1186/s12911-026-03488-z","url":null,"abstract":"<p><strong>Background: </strong>Paper-based medical records remain prevalent in many developing countries, limiting the accuracy and efficiency of healthcare delivery. Electronic Medical Record (EMR) offer significant potential to improve care quality, yet their adoption faces infrastructural, financial, and contextual challenges in resource-limited settings.</p><p><strong>Methods: </strong>A systematic review of five databases was conducted to identify studies on EMR design, development, and implementation. Eligible articles underwent qualitative analysis to identify key components and challenges. These findings were synthesized into a conceptual model, using the System Development Life Cycle (SDLC) as a comprehensive Conceptual map to cover all system phases rather than a linear execution plan.</p><p><strong>Result: </strong>From 80 eligible studies, qualitative analysis yielded 204 refined codes, synthesized into a three-layered framework (Foundation, Core Cycle, and Optimization) comprising seven conceptual categories: strategic foundations, system architecture, user experience, interoperability standards, implementation processes, ethical considerations, and emerging technologies. Together, these categories outline the key dimensions of EMR system development in resource-limited settings.</p><p><strong>Conclusions: </strong>The proposed model offers a constructive and iterative approach for EMR implementation in resource-limited settings. By distinguishing between core necessities and optimization features, the framework allows for a flexible, scalable adoption strategy that addresses local infrastructural constraints better than rigid, linear models.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":" ","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13551989/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148315739","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}