International Journal Of Medical Informatics[JOURNAL]
Sun
200 papers
Drug prescribing in critically ill children with kidney impairment: Comparing clinical practice with clinical decision support recommendations.
Higi L, Wälti M, Gotta V … +3 more , Goischke A, Pfister M, Vonbach P
Int J Med Inform · 2026 Jul · PMID 41955912 · Publisher ↗
INTRODUCTION: The kidneys play a crucial role in eliminating drugs, and their functional impairment can lead to drug accumulation and adverse effects. To prevent this, drug dosages (doses or administration intervals) are...
INTRODUCTION: The kidneys play a crucial role in eliminating drugs, and their functional impairment can lead to drug accumulation and adverse effects. To prevent this, drug dosages (doses or administration intervals) are adjusted based on kidney function using the Glomerular Filtration Rate (GFR). Calculating dosages is an error prone task, particularly in paediatric patients where individual dosing strategies can be complex. AIM: This study assessed the potential application of a novel feature of the paediatric clinical decision support system (CDSS) PEDeDose that calculates adjusted drug dosages for children with impaired kidney function. Renal dosing practices in a paediatric intensive care patients with impaired kidney function were compared to recommendations provided by PEDeDose. METHODS: This retrospective observational feasibility study analysed drug prescriptions in patients with impaired kidney function from electronic health records of the paediatric intensive care unit (PICU) of the University Children's Hospital Basel in 2023. Extracted data included patients' age, weight, height, and length of PICU stay, drug administration records, serum creatinine and cystatin C values, and diagnoses. The primary outcome was the frequency of dosage adjustments adherent to CDSS' recommendations. RESULTS: Out of 436 patients, 20 (5%) patients were included. In this study of 20 patients, 7 (33%) were prescribed drugs that required dosage adjustments due to their impaired kidney function. Across these patients, dosage adjustment was indicated in 11 prescriptions, and 8 (73%) of these dosages were adjusted as recommended by the CDSS. CONCLUSION: The study observed good concordance between clinical practice and the CDSS' recommendations on dosage adjustment in paediatrics. This supports the potential of PEDeDose to facilitate dosing and maintain quality of drug prescribing in children with kidney impairment. Considering the complexity of drug prescribing in this patient population, the support provided by PEDeDose may benefit paediatric patients and their caregivers beyond a specialised university hospital setting.
"Can a chatbot be used in the full-text screening in a systematic review?".
Martins AM, Valero Juan LF, Oliveira A … +2 more , Martins JP, Santos M
Int J Med Inform · 2026 Jul · PMID 41950629 · Publisher ↗
INTRODUCTION: Large language model-based artificial intelligence tools are increasingly explored to support systematic reviews, yet evidence regarding their reliability in full-text screening remains limited. This study...
INTRODUCTION: Large language model-based artificial intelligence tools are increasingly explored to support systematic reviews, yet evidence regarding their reliability in full-text screening remains limited. This study evaluated the performance of two versions of ChatGPT (4.0 and 5.0) compared with human reviewers during article selection for a systematic review on influenza vaccine effectiveness. METHODS: A total of 170 full-text articles were independently assessed for eligibility using predefined inclusion and exclusion criteria. Human reviewers served as the gold standard. ChatGPT 4.0 and 5.0 were prompted using standardized instructions mirroring the review protocol. Agreement with human decisions was evaluated using accuracy, sensitivity, specificity, precision, F1-score, and Cohen's κ. Intra-model reproducibility was assessed for ChatGPT 5.0. RESULTS: ChatGPT 4.0 achieved an accuracy of 0.71 (95% CI: 0.64-0.78) and a Cohen's κ of 0.43, indicating moderate agreement with human reviewers. ChatGPT 5.0 demonstrated improved performance, with accuracy increasing 0.06 to 0.77 (95% CI: 0.70-0.83), sensitivity of 0.87, specificity of 0.70, and κ of 0.55, corresponding to moderate-to-substantial agreement. Intra-model reproducibility for ChatGPT 5.0 showed 80% agreement (κ = 0.60), indicating partial but imperfect consistency. CONCLUSIONS: ChatGPT 5.0 outperformed ChatGPT 4.0 in full-text screening accuracy and reproducibility, approaching but not matching human performance. These findings support the use of current LLMs as decision-support tools rather than autonomous reviewers in systematic reviews. Transparent reporting of model versions, prompts, and input quality is essential to ensure credible AI-assisted evidence synthesis.
Artificial intelligence in scholarly peer review: a scoping review of applications, risks, and governance challenges.
Nabavi A, Safari F, Shmoury AH … +3 more , Tabet S, Perdomo-Luna C, Celi LA
Int J Med Inform · 2026 Jul · PMID 41950628 · Publisher ↗
BACKGROUND: Artificial intelligence (AI) is increasingly integrated into scholarly publishing workflows, extending beyond manuscript preparation into editorial triage, reviewer assistance, and policy development. Peer re...
BACKGROUND: Artificial intelligence (AI) is increasingly integrated into scholarly publishing workflows, extending beyond manuscript preparation into editorial triage, reviewer assistance, and policy development. Peer review simultaneously faces long-standing problems including reviewer fatigue, bias, opacity, and publish-or-perish incentives. How AI interacts with these structural weaknesses remains unclear. OBJECTIVE: To map how AI is currently used in scholarly peer review, synthesize reported benefits and risks, and identify governance and research gaps relevant to health sciences. METHODS: A scoping review following Arksey and O'Malley was conducted and reported according to PRISMA-ScR. Scopus, Web of Science, PubMed/MEDLINE, and IEEE Xplore were searched (January 1, 2024-August 31, 2025) using terms combining artificial intelligence and peer review. Grey literature (publisher policies, professional guidelines, editorials) was identified through targeted searches of COPE, ICMJE, WAME, major publisher portals, and preprint servers. Duplicate screening/extraction with adjudication were done. Data were synthesized using inductive thematic analysis. RESULTS: Of 2,908 records, 189 met inclusion criteria. AI is used as AI assistive (triage, assistance) and autonomous (review generation, prediction).Reported benefits include improved workflow efficiency, standardized checks, and clearer feedback. However, current systems lack domain reasoning and ethical judgment for autonomous evaluation. Key risks are confidentiality breaches when manuscripts are submitted to third-party tools, algorithmic bias favoring elite institutions or male authors, and homogenization of scholarly voice. As of August 31, 2025, governance policies across publishers, journals, and professional societies remain fragmented. In many documented cases, reviewer use of generative AI is more restricted than author-side use; however, policies vary by publisher, journal, and society, and continue to evolve. CONCLUSIONS: AI can strengthen peer review when deployed as a transparent, auditable, privacy-preserving support tool under human oversight. Responsible integration in medical informatics requires coordinated governance, bias monitoring, secure infrastructures, and reforms to evaluation incentives.
Precision Grounding: augmenting large language models with evidence-based databases for trustworthy genetic variant summarization.
Du X, Nagy A, Oates MF … +6 more , Wang Y, Wang X, Plasek JM, Aronson SJ, Lebo MS, Zhou L
Int J Med Inform · 2026 Jul · PMID 41950627 · Full text
PURPOSE: To propose a novel method that augments LLMs with evidence-based, variant-specific information to improve summarization accuracy and support clinical decision making. METHODS: We proposed Precision Grounding whi...
PURPOSE: To propose a novel method that augments LLMs with evidence-based, variant-specific information to improve summarization accuracy and support clinical decision making. METHODS: We proposed Precision Grounding which uses a query tool that integrates domain expert-selected resources and allows users to query relevant factual information using identifiers in databases. In our case, which is genetic variant summarization, we developed CATT (ClinGen website: https://shorturl.at/pw81X; GitHub: https://github.com/mgbpm/clingen-ai-tools; Zenodo: https://doi.org/10.5281/zenodo.18896080), an open-source tool integrating publicly available ClinGen, ClinVar, and GenCC databases. Users can query and retrieve curated evidence via Variation IDs to ground LLM outputs. We compared our approach with baseline methods, including web-search grounding and retrieval-augmented generation (RAG; specifically, MedRAG) using 50 expert-selected variants. RESULTS: GPT-4o was selected due to its good performance on our task during a pilot test. Using GPT-4o, we found web-search grounding performed better than MedRAG since MedRAG failed to generate clinically useful summaries due to limitations of relevant information in its databases. Precision Grounding outperformed web-search grounding, achieving significantly higher accuracy and completeness scores, which were based on a 5-point Likert-Scale of 4.76 (+0.74) and 4.94 (+0.84), respectively. Error analysis revealed that Precision Grounding reduced clinically significant hallucinations, such as incorrect pathogenicity classification and summarizing the wrong variant. CONCLUSION: Precision Grounding outperformed existing grounding approaches for genetic variant summarization. Our open-source tool, CATT, enables integration of curated, domain-specific genetic variant knowledge and significantly reduces hallucinations in LLM outputs. By enhancing the accuracy and completeness of variant interpretation, this framework holds strong potential for real-world implementation in clinical decision-support systems. Its modular, interoperable design allows for seamless integration into clinical genomics workflows, supporting more efficient, trustworthy, and scalable variant review processes.
Community-based early-stage chronic kidney disease screening using explainable machine learning for low-resource settings.
Kabir MA, Munira S, Azad DT … +3 more , Ikram SM, Sarker MHR, Hanifi SMA
Int J Med Inform · 2026 Jul · PMID 41946122 · Publisher ↗
BACKGROUND: Early detection of chronic kidney disease (CKD) is essential for preventing progression to end-stage renal disease. However, existing screening tools-primarily developed using populations from high-income cou...
BACKGROUND: Early detection of chronic kidney disease (CKD) is essential for preventing progression to end-stage renal disease. However, existing screening tools-primarily developed using populations from high-income countries-often underperform in Bangladesh and South Asia, where risk profiles differ. Moreover, many machine learning-based CKD studies rely heavily on pathology-test variables and are developed using hospital-based datasets, limiting their applicability for community-level screening and for settings where laboratory testing is not readily accessible. OBJECTIVE: To develop and evaluate an explainable machine learning (ML) framework for community-based early-stage CKD screening that derives and evaluates optimized predictor subsets from both all available variables and variables excluding pathology tests, enabling accurate and practical risk assessment in low-resource settings. METHODS: A community-based CKD dataset from Bangladesh was used to develop predictive models. Variables were organized into clinically meaningful feature groups, and ten complementary feature selection methods were applied to identify robust predictor subsets. Twelve ML classifiers were evaluated using nested cross-validation. Model performance was benchmarked against established CKD screening tools and externally validated on three independent datasets from India, the UAE, and Bangladesh. SHAP (SHapley Additive exPlanations) was used to interpret model predictions. RESULTS: The best-performing model using the optimized feature subset derived from all variables achieved a balanced accuracy of 90.4%, while the subset derived from variables excluding pathology tests achieved comparable performance (89.23%), demonstrating the feasibility of accurate CKD risk prediction using minimal non-laboratory features. The proposed approach outperformed existing screening tools while requiring fewer and more accessible inputs. External validation demonstrated strong generalizability, with sensitivities ranging from 78% to 98%. SHAP analysis identified clinically meaningful predictors consistent with established CKD risk factors. CONCLUSIONS: Accurate, interpretable, and scalable early-stage CKD screening is achievable using only non-pathology-test features. This framework demonstrates the potential for community-level CKD screening in resource-constrained settings.
Beyond block time: a head-to-head comparison of reinforcement learning, genetic algorithms, and predict-then-optimize scheduling for operating room workflow using discrete-event simulation.
Çalışkan YK, Başak F, Erdem O … +1 more , Kudaş İ
Int J Med Inform · 2026 Jul · PMID 41946121 · Publisher ↗
BACKGROUND: Operating room (OR) inefficiency persists despite decades of process improvement, largely due to stochastic case durations, emergency disruptions, and resource coupling across pre-, intra-, and postoperative...
BACKGROUND: Operating room (OR) inefficiency persists despite decades of process improvement, largely due to stochastic case durations, emergency disruptions, and resource coupling across pre-, intra-, and postoperative steps. While artificial intelligence (AI) methods are increasingly proposed for OR scheduling and allocation, most evaluations are single-method, single-site, or non-comparative, limiting actionable adoption decisions. The specific unresolved gap is not whether AI can improve isolated OR subproblems, but whether distinct algorithmic paradigms can be benchmarked head-to-head under identical stochastic conditions, shared-resource constraints, and downstream bottlenecks to support implementation choices. METHODS: We conducted a simulation-based comparative study using a discrete-event simulation (DES) model representing a general surgery OR suite with elective and emergency arrivals, shared anesthesia and nursing resources, turnover processes, and downstream constraints. The model was implemented in Python 3.10 using SimPy and compared four strategies: a fully specified rule-based baseline, predict-then-optimize (PTO), genetic algorithm (GA), and reinforcement learning (RL). Across 500 replications with common random numbers, strategies were compared using repeated-measures inference with multiplicity control. Calibration targeted operational plausibility rather than institution-specific emulation: input distributions were anchored to general-surgery workflow patterns and literature-consistent throughput assumptions, and the scope of inference was therefore bounded to comparative performance within a stylized but resource-coupled OR environment. RESULTS: Compared with baseline, all AI strategies reduced delays and overtime and improved utilization. RL achieved the greatest gains, reducing total delay from 182.4 ± 45.1 to 104.2 ± 31.0 min (42.9% relative reduction), overtime from 76.3 ± 28.4 to 41.2 ± 18.1 min (46.0%), increasing utilization from 73.2 ± 4.1% to 80.1 ± 3.4% (+6.9 percentage points), and decreasing same-day cancellations from 0.42 ± 0.19 to 0.19 ± 0.12 (54.8%) (all overall p < 0.001). RL also showed the narrowest delay dispersion (IQR 22.5 min versus 48.2 min for baseline and 35.6 min for PTO). CONCLUSION: In a controlled DES environment, AI-based scheduling meaningfully improved OR workflow metrics versus conventional rules, with RL offering the most robust performance. However, RL's advantage should be interpreted as conditional on the modeled reward structure, resequencing latitude, and disruption profile; accordingly, the present findings are best viewed as decision-support evidence for staged local validation rather than as proof of universal real-world superiority.
Socio-technical risks of clinical speech-to-text systems: Transparency, privacy, and reliability challenges in AI-driven documentation.
Elsayed N
Int J Med Inform · 2026 Jul · PMID 41936257 · Publisher ↗
BACKGROUND: AI-driven speech-to-text (STT) documentation systems are increasingly adopted in clinical settings to reduce documentation burden and improve workflow efficiency. However, adoption has outpaced the systematic...
BACKGROUND: AI-driven speech-to-text (STT) documentation systems are increasingly adopted in clinical settings to reduce documentation burden and improve workflow efficiency. However, adoption has outpaced the systematic evaluation of socio-technical risks related to transparency, reliability, patient autonomy, and organizational accountability. OBJECTIVE: To develop a socio-technical framework for identifying and governing risks associated with the implementation of clinical speech-to-text systems. METHODS: This study synthesizes interdisciplinary evidence from technical automatic speech recognition research, clinical workflow and human factors studies, ethical guidance on consent and patient autonomy, and regulatory and organizational governance sources. Using a structured narrative synthesis approach, relevant literature was iteratively reviewed and thematically analyzed to identify recurring socio-technical risk mechanisms. The synthesis was used to develop a layered conceptual framework for evaluating and governing clinical speech-to-text systems. RESULTS: Findings show that clinical STT systems operate within tightly coupled socio-technical environments where model performance, audio capture conditions, clinician oversight, patient understanding, workflow design, and institutional governance are interdependent. Key risks include inconsistent disclosure and consent practices, performance disparities for accented speech and speech/voice disorders, accuracy degradation under real clinical acoustics, automation complacency and variable clinician review, unclear accountability across vendors and healthcare organizations. These risk domains informed a six-layer socio-technical governance model spanning technical, human/workflow, ethical, organizational, regulatory, and sociocultural dimensions. CONCLUSION: The study proposes a socio-technical governance framework and implementation roadmap to support the responsible deployment of clinical STT systems. The framework emphasizes transparency, patient autonomy, documentation integrity, and accountable governance to enable safe and equitable adoption of speech-based documentation technologies.
Benchmarking text encoding strategies in multimodal clinical data for surgical case duration prediction.
Noorchenarboo M, Kwong M, Elnahas A … +3 more , Hawel J, Schlachta CM, Grolinger K
Int J Med Inform · 2026 Jul · PMID 41936256 · Publisher ↗
BACKGROUND: Operating rooms (ORs) are highly resource-intensive, yet surgical case duration is often estimated using heuristics that are prone to errors. While machine learning models based on structured perioperative da...
BACKGROUND: Operating rooms (ORs) are highly resource-intensive, yet surgical case duration is often estimated using heuristics that are prone to errors. While machine learning models based on structured perioperative data improve accuracy, unstructured clinical text remains underutilized despite containing valuable contextual details. OBJECTIVE: To benchmark classical and contextual text encoding strategies, combined with structured perioperative data, for predicting surgical case durations. METHODS: We retrospectively analyzed 180,370 elective surgical cases from three tertiary care hospitals (2015-2020). Structured variables such as age, sex, BMI, ASA score, and case service were combined with unstructured text features (procedure descriptions), which were encoded using five different methods (label encoding, count vectorization, TF-IDF, ClinicalBERT, Sentence-BERT). We trained a diverse set of machine learning models including linear regression, tree-based ensembles, and neural networks and evaluated predictive accuracy using standard error metrics with cross-validation. RESULTS: Adding unstructured clinical text to structured perioperative variables improved prediction accuracy across all models. Contextual embeddings consistently outperformed structured-only and traditional text encodings. Sentence-BERT and ClinicalBERT achieved comparable best performance, reducing MAE to approximately 26.4 minutes and SMAPE to 21.6%, with R of 0.86; neither encoder was statistically superior to the other (p>0.82). Improvements over structured-only baselines were statistically significant (p<0.01), corresponding to up to 16% reduction in prediction error. Traditional encodings (label, count, TF-IDF) provided limited benefit. CONCLUSION: Integrating semantically rich clinical text with structured perioperative data substantially improves surgical duration prediction. Our multimodal approach which combines structured and unstructured data with contextual embeddings, directly improves prediction accuracy, which in turn supports more reliable OR scheduling, better resource utilization, and improved patient care. Future work should incorporate additional narrative sources and interpretability techniques to support clinical adoption.
Evaluating the quality of online health information: A comparative study of COPD patient information in Facebook forums versus ChatGPT.
Diers CS, Høj S, Meteran H … +2 more , Backer V, Meteran H
Int J Med Inform · 2026 Jul · PMID 41934770 · Publisher ↗
BACKGROUND: Patients with chronic obstructive pulmonary disease (COPD) increasingly turn to online platforms for health information. However, the quality of information shared in social media forums remains uncertain. Th...
BACKGROUND: Patients with chronic obstructive pulmonary disease (COPD) increasingly turn to online platforms for health information. However, the quality of information shared in social media forums remains uncertain. This study compared the accuracy of user-generated Facebook responses to AI-generated replies from ChatGPT. METHODS: Posts and comments were extracted from two COPD-related Facebook groups in October 2024. A total of 48 posts from each group were selected, yielding 2,761 comments. Each post was also submitted to ChatGPT. Responses were categorized as 'useful', 'misleading', or 'neither' and rated on a 5-point Likert scale by three independent reviewers. Interrater reliability was assessed using Fleiss' kappa. Differences in quality scores were analyzed using the Wilcoxon signed-rank test and Hodges-Lehmann 95% CI. RESULTS: In the Danish group, 47% (667/1,433) of comments were relevant, with 34% deemed useful and 9% misleading. Results were similar in the English group (734/1,315 relevant; 38% useful, 9% misleading). Critically, 16% and 15% of comments encouraging behavioral changes contained misleading information in the Danish and English group, respectively. Likert ratings showed significantly higher accuracy for ChatGPT compared to Facebook. For the Danish group, the median score for Facebook was 4.0 (IQR 3.0-4.5) versus 5.0 (IQR 4.0-5.0) for ChatGPT (p < 0.001; Hodges-Lehmann difference: -0.75, 95% CI -1.0 to -0.5). For the English group, Facebook scored a median of 4.0 (IQR 3.0-4.0) versus 5.0 (IQR 5.0-5.0) for ChatGPT (p < 0.001; Hodges-Lehmann difference: -1.0, 95% CI -1.25 to -0.75). No significant performance differences were found between the Danish and English groups for either platform. CONCLUSION: ChatGPT delivered more accurate responses than Facebook users, highlighting its potential as a reliable educational tool. However, Facebook remains valuable for peer support, suggesting that combining AI and social platforms may enhance digital care strategies for COPD.
Barriers and facilitators to the implementation and use of computerized clinical decision support systems for predicting and managing in-patient clinical deterioration: a systematic review of qualitative research.
Saikali M, Baysari M, Lichtner V … +3 more , Pelayo S, Carland JE, Marcilly R
Int J Med Inform · 2026 Jul · PMID 41932265 · Publisher ↗
Abstract loading — click title to view on PubMed.
Development and validation of an eye-tracking-based cognitive impairment screening system for older adults in China: a cross-sectional study.
Si W, Ma X, Lin J … +4 more , Xu T, Wang R, Zhu A, Cao W
Int J Med Inform · 2026 Jul · PMID 41921372 · Publisher ↗
BACKGROUND: Global population ageing has rendered cognitive impairment a critical public health issue. Early screening is essential for timely intervention; however, traditional tools are limited by their reliance on pro...
BACKGROUND: Global population ageing has rendered cognitive impairment a critical public health issue. Early screening is essential for timely intervention; however, traditional tools are limited by their reliance on professionals, cultural and educational biases, and high cost. PURPOSE: This study aimed to develop and validate an unobtrusive, eye-tracking-based cognitive screening system (CIS-ET) for older adults in China, evaluating its efficacy in distinguishing between healthy controls (HCs), mild cognitive impairment (MCI), and Alzheimer's disease (AD). METHODS: In this cross-sectional study, 113 participants (41 HCs, 41 with MCI, and 31 with AD) were recruited in Shanghai. All participants completed the CIS-ET (assessing 6 cognitive domains via 43 items), the Mini-Mental State Examination (MMSE), and the Montreal Cognitive Assessment (MoCA). Statistical analyses included receiver operating characteristic (ROC) curve analysis, partial correlation, and Cronbach's α for reliability. RESULTS: The CIS-ET demonstrated excellent discriminative validity. The area under the curve (AUC) for distinguishing HCs from participants with MCI was 0.878 (95% CI: 0.796 to 0.959), with a sensitivity of 85.37% and specificity of 87.80%. For differentiating MCI from AD, the AUC was 0.893 (95% CI: 0.821 to 0.965; sensitivity 77.42%, specificity 87.80%). When distinguishing HCs from the combined cognitive impairment group (MCI + AD), the AUC reached 0.927 (95% CI: 0.876 to 0.978; sensitivity 91.67%, specificity 87.80%). After adjusting for age and education, the CIS-ET total score showed strong positive correlations with MMSE (r = 0.870, p < 0.001) and MoCA (r = 0.891, p < 0.001). Internal consistency reliability was acceptable (Cronbach's α = 0.685). CONCLUSIONS: The CIS-ET is a valid, reliable, and user-friendly tool for the early screening of cognitive impairment in older adults in China. Its design supports potential for large-scale use in community healthcare settings. (Trial registration: Chinese Clinical Trial Registry, ChiCTR2400085172.).
Unlocking the potential of clinical decision support in cardiovascular care: A mixed-methods systematic review of implementation barriers and enablers.
Sallam S, Cebulla A, Brommeyer M … +2 more , Aljudaibi S, Balasubramanian M
Int J Med Inform · 2026 Jul · PMID 41921371 · Publisher ↗
BACKGROUND: Clinical decision support significantly enhances effective healthcare delivery by empowering healthcare providers with the necessary tools and resources to make evidence-based decisions. To date, only limited...
BACKGROUND: Clinical decision support significantly enhances effective healthcare delivery by empowering healthcare providers with the necessary tools and resources to make evidence-based decisions. To date, only limited research studies have discussed the effective implementation of clinical decision support systems (CDSS) in managing cardiovascular care. OBJECTIVE: To identify the barriers and facilitators of integrated CDSS implementation through electronic health records in cardiovascular care. METHODS: A convergent integrated design aligned with Joanna Briggs Institute (JBI) methodology for mixed methods review was followed. Four databases were included from October 2023 to July 2025. Eligible peer-reviewed studies (quantitative, qualitative, or mixed-methods) reported on barriers or facilitators to implementing CDSS in cardiovascular care. RESULT: A total of 718 articles were screened; 12 studies were included in this review (2008-2025). Studies spanned diverse settings across five countries and included qualitative (n = 4), quantitative (n = 2), and mixed-methods (n = 6) designs. All studies evaluated knowledge-based CDSS applied in cardiovascular care, in primary (n = 9) and tertiary care settings. Barriers and facilitators were synthesised into a three-level framework: macro (organisational and regulatory), meso (technical and clinical), and micro (patient and training). Common barriers included workflow disruption, alert fatigue, limited interoperability, regulatory burden, and lack of staff training. Key facilitators included leadership commitment, workflow integration, stakeholder engagement, iterative tool refinement, and targeted clinician training. CONCLUSION: Our review underscored the critical need for contextual, multi-level strategies to enable the effective adoption and sustainability of CDSS in cardiovascular care. The establishment of standardised guidelines to systematically address implementation barriers was also considered essential. A coordinated governance framework encompassing organisational commitment, technical integration, clinician involvement, patient engagement, and regulatory alignment must be supported by ongoing training and capacity building. Recognising and addressing these interdependent factors can equip healthcare systems to scale CDSS adoption, optimise clinical workflows, and improve patient outcomes through enhanced decision-making.
Artificial intelligence in gastroenterology clinical practice: Scoping review of large language model applications.
Yazarkan Y, Sonmez G, Simsek C
Int J Med Inform · 2026 Jul · PMID 41921370 · Publisher ↗
BACKGROUND: Advances in artificial intelligence have brought renewed attention to tools that can work with the large amount of written information generated in clinical practice. Among these, large language models (LLMs)...
BACKGROUND: Advances in artificial intelligence have brought renewed attention to tools that can work with the large amount of written information generated in clinical practice. Among these, large language models (LLMs) stand out for their ability to interpret and generate medical text in a flexible, context-aware way. Gastroenterology, like many specialties, produces a wide range of narrative and semi-structured data, and this has encouraged researchers to explore how LLMs might help clinicians manage everyday tasks. Recent studies have examined their potential contributions to patient education, communication between care teams and patients, decision support, and routine documentation, reflecting a growing interest in how these systems might fit into real-world clinical workflows. OBJECTIVE: This scoping review aimed to map current applications of LLMs in gastroenterology clinical practice, including subspecialty focus, study designs, model types, and reported outcomes. METHOD: Following PRISMA guidelines, a systematic search was conducted in PubMed, Scopus, and Web of Science for studies published between January 2022 and August 2025. Eligible studies included original research assessing LLM applications in gastroenterology clinical practice. Data were extracted on subspecialty, application domain, LLM type, data source, and outcomes. We employed thematic analysis to address our primary research question. RESULTS: 73 out of 2895 studies identified in the initial search met the inclusion criteria. Six subspecialties and six application domains emerged from our review. Hepatology (20/73 studies, 27.3%) and endoscopy (17/73 studies, 23.2%) were the most represented subspecialties. The most frequently investigated application domains were patient education and communication (38 studies) and decision support and clinical guidance (24 studies). Most studies were simulation-based or literature-based cases, although an increasing number have used real-world clinical data, particularly in recent years. The majority evaluated general-purpose models such as GPT-3.5 and GPT-4, with some incorporating retrieval augmentation or fine-tuning. Reported outcomes varied by application domain and included measures of accuracy, concordance, completeness, relevance, safety, reliability, usability, user satisfaction, efficiency, time savings, and educational value. Commonly described limitations included variable reliability, incomplete responses, and challenges in generalizing from simulated to clinical settings. CONCLUSIONS: Research on LLMs in gastroenterology has expanded across multiple subspecialties and application domains. Current evidence is primarily based on simulation studies, with limited but growing evaluation using real-world clinical data. Further work is needed to assess performance in prospective and applied clinical contexts.
Evolution of artificial intelligence at Hospital Italiano de Buenos Aires: A retrospective review of experience and lessons learned.
Luna DR, Otero CM, Cancio AH … +14 more , Mazzuoccolo LD, Kowalczuk VR, Rabellino JM, Aineseder M, Frutos EL, Esposito MI, Rusconi Lagarrigue AB, Saguier A, Marin FE, Segalini A, Campos FA, Castaño JM, Rubin L, Benitez SE
Int J Med Inform · 2026 Jul · PMID 41916095 · Publisher ↗
BACKGROUND: Hospital Italiano de Buenos Aires (HIBA) has progressively integrated artificial intelligence (AI) technologies into clinical practice over more than two decades. Describing this process may provide useful in...
BACKGROUND: Hospital Italiano de Buenos Aires (HIBA) has progressively integrated artificial intelligence (AI) technologies into clinical practice over more than two decades. Describing this process may provide useful insights for healthcare institutions aiming to adopt AI in a structured and sustainable manner. OBJECTIVES: To describe the evolution of AI implementation at HIBA, identifying key stages, technologies, and lessons learned throughout this institutional journey. METHODS: A retrospective institutional review was conducted, describing four sequential stages of AI adoption at HIBA. Each stage was characterized according to the predominant AI technologies, their clinical applications, and their degree of integration into routine healthcare workflows. RESULTS: The first stage (1997-2009) focused on establishing digital foundations and clinical decision support systems, improving patient safety through pharmacological alerts and structured clinical recommendations. The second stage (2010-2017) involved natural language processing and speech technologies, enabling the extraction of structured information from unstructured clinical text and the development of automatic speech recognition systems. The third stage (2018-2022) encompassed computer vision applications in medical imaging, including convolutional neural networks for breast density assessment and triage systems for chest radiographs, with emphasis on iterative validation and integration into clinical workflows. The fourth stage (2023-present) explores generative AI and large language models, exemplified by the internally developed chatbot TANA, supporting clinical decision-making, digital triage, and patient engagement. Across all stages, key lessons emerged related to data quality, interdisciplinary collaboration, model validation, user training, ethical safeguards, and responsible AI implementation. CONCLUSIONS: This overview highlights the institutional strategies and challenges associated with long-term AI adoption in healthcare. The experience at HIBA may offer relevant guidance for other hospitals seeking to integrate AI into clinical practice.
HL7 FHIR consent for healthcare data sharing: challenges, opportunities and integrity implications.
Phuyal S, Bhandari M, Bista R … +1 more , Ferreira JC
Int J Med Inform · 2026 Jul · PMID 41905015 · Publisher ↗
OBJECTIVE: To assess whether HL7 FHIR Consent, as currently specified and deployed, is sufficient to support verifiable, regulation-aligned consent governance in distributed and cross-organisational health data sharing....
OBJECTIVE: To assess whether HL7 FHIR Consent, as currently specified and deployed, is sufficient to support verifiable, regulation-aligned consent governance in distributed and cross-organisational health data sharing. METHODS: We conducted a qualitative critical analysis of FHIR Consent informed by (i) peer-reviewed implementation literature, (ii) national-scale consent exchange initiatives, and (iii) accountability requirements under GDPR and the European Health Data Space (EHDS). The analysis is organized into four dimensions: semantic interpretability, consent lifecycle management, runtime enforcement, and cross-organisational trust/auditability. RESULTS: FHIR Consent provides an interoperable representation of authorisation intent, but large-scale deployments remain limited by (1) non-canonical semantics across implementations, (2) lack of standardized lifecycle versioning and cross-organisational revocation propagation, (3) heterogeneous translation of declarative consent into enforceable access control, and (4) limited capability for independent verification of consent provenance and historical integrity across institutional boundaries. CONCLUSION: We derive an architecture pattern that separates (a) standards-based consent representation (FHIR Consent), (b) local policy interpretation/enforcement, and (c) cross-organisational integrity verification. Cryptographic integrity anchoring is discussed as a complementary mechanism for tamper-evident verification of off-chain consent artifacts and lifecycle events, without externalizing consent semantics or personal data.
Assessing the representativeness of single-center EMR data on ten cancer types: A comparative analysis with national statistics from South Korea (2011-2021).
Won JH, Lee H
Int J Med Inform · 2026 Jul · PMID 41895025 · Publisher ↗
BACKGROUND: Real-world data (RWD) from electronic medical records (EMRs) is increasingly utilized in oncology to complement evidence from clinical trials by reflecting routine clinical practice and diverse patient popula...
BACKGROUND: Real-world data (RWD) from electronic medical records (EMRs) is increasingly utilized in oncology to complement evidence from clinical trials by reflecting routine clinical practice and diverse patient populations. However, many EMR-based studies rely on single-center data, limiting the generalizability of their findings. We aimed to evaluate the representativeness of single-center EMR data from Seoul National University Hospital (SNUH) by comparing it with national cancer data from the Korean Statistical Information Service (KOSIS). METHODS: We compared annual cancer statistics from SNUH EMR and KOSIS (2011-2021) for ten cancer types: breast, gallbladder/biliary tract, gastric, kidney, liver, lung, pancreatic, prostate, thyroid cancers, and leukemia. We calculated the coverage proportion of cancer cases in the SNUH EMR relative to KOSIS. Differences in age and gender distributions between the two databases were analyzed. Annual trends in cancer cases were compared between two databases. RESULTS: From 2011 to 2021, SNUH data included 8.2% of national incident and 10.7% of prevalent cases, with high coverage for liver (20.4%) and pancreatic (20.3%) cancers. No significant differences in age and gender distribution were found across all cancer types (p > 0.05), with high cosine similarity (>0.8). Strong correlations in annual trends were observed for breast, lung, and pancreatic cancers (r > 0.9), while negative correlations were found for thyroid cancer prevalence (r = - 0.62) and liver cancer incidence (r = - 0.59). CONCLUSION: Single-center EMR data can be a valuable resource for oncology research in South Korea. However, external factors including changes in clinical guidelines should be considered when generalizing findings from such data to broader populations.
Can ensemble methods improve predictive performance of existing models estimating chronic kidney disease among patients with diabetes?
Black JE, Campbell DJT, Ronksley PE … +2 more , McBrien KA, Williamson TS
Int J Med Inform · 2026 Jul · PMID 41895024 · Publisher ↗
BACKGROUND: Clinical prediction models often suffer from poor model transportability and/or subgroup performance resulting from using a single data source. We aimed to determine whether ensemble methods can combine multi...
BACKGROUND: Clinical prediction models often suffer from poor model transportability and/or subgroup performance resulting from using a single data source. We aimed to determine whether ensemble methods can combine multiple existing models to improve predictive performance when compared to component models. METHODS: As a case study, we used electronic medical records from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) to test ensemble methods for models estimating the risk of developing chronic kidney disease (CKD) among people with diabetes in a cohort of 37,604 individuals. We considered 13 models identified from prior systematic reviews and combined their unique risk estimates using many strategies (e.g., averaging or mixture-of-experts). We assessed discrimination, precision, recall, calibration, net reclassification index, and integrated discrimination improvement. RESULTS: Ensemble methods performed well, but no better than the best performing component model. Among ensemble methods, the averaging or selection process with the best performance weighted the predictions from all component models by their development cohort size (AUROC: 0.827 [95% CI: 0.821 to 0.833]). However, this did not exceed the best performing component model (AUROC: 0.826 [95% CI: 0.820 to 0.832]). Similarly, based on the NRI, estimated risks based on the ensemble methods were often worse than the best performing component model. CONCLUSIONS: This study suggests ensemble methods may not improve predictive performance, though further research should confirm these findings. SUMMARY TABLE: Many clinical prediction models exist that predict the same outcome, but commonly suffer poor performance when applied in new settings. Ensemble methods provide a method of combining multiple models developed across diverse settings to potentially improve predictive performance. When applied in primary care electronic medical records, we found that ensemble models based on existing clinical prediction models could match, but did not surpass the performance of the best performing component model. Ensemble methods may not be necessary to combine existing models; rather, the best performing component model can be used.
Predictive value of early blood glucose trajectory for poor prognosis in ARDS patients.
Fang Z, Chen J, Lin J … +2 more , Lin Z, Zheng G
Int J Med Inform · 2026 Jul · PMID 41895023 · Publisher ↗
BACKGROUND: Patients with acute respiratory distress syndrome (ARDS) often develop stress-induced hyperglycemia, and glycemic variability is associated with adverse outcomes. Traditional static glycemic indicators cannot...
BACKGROUND: Patients with acute respiratory distress syndrome (ARDS) often develop stress-induced hyperglycemia, and glycemic variability is associated with adverse outcomes. Traditional static glycemic indicators cannot capture dynamic glucose changes, and current ARDS prognostic models lack integration of dynamic glycemic trajectories, leading to insufficient precision in early risk stratification. This study aimed to investigate the association between early glycemic trajectory and 30-day mortality in ARDS patients, while constructing and validating a prognostic prediction model integrating dynamic glycemic features. METHODS: A total of 8,103 ARDS patients from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database were enrolled as training set, and 158 patients from a single center served as external validation set. Group-based trajectory modeling (GBTM) was used to cluster the blood glucose trajectory within 48 h of admission. Independent prognostic factors were identified using Cox proportional hazards models, and a nomogram model was constructed. And the nomogram was validate using receiver operating characteristic curve, calibration curve and decision curve analysis. RESULTS: Four blood glucose trajectories were identified: G1 (20.01%), G2 (37.43%), G3 (34.64%) and G4 (7.91%). Among them, G2 group with stable and low blood glucose levels had the highest 30-day survival probability, while G4 group with an initial high blood glucose level that decreased sharply and then slightly increased had a significantly higher 30-day death (log-rank P < 0.0001). The nomogram integrating 16 predictors including glucose trajectory, age, respiratory rate, and anion gap achieved good discrimination (AUC = 0.72 in training set, 0.75 in validation set), favorable calibration, and high net clinical benefit. CONCLUSIONS: Early blood glucose trajectory is an independent predictor of 30-day mortality in patients with ARDS. The nomogram constructed based on dynamic blood glucose evolution has good predictive efficacy and clinical applicability, and can provide a quantitative tool for the early intervention of high-risk patients.
A visualized nurse-led ePRO system for chemotherapy toxicity: Content design and validation.
Kuan CC, Guo JC, Shen SH … +4 more , Hsu C, Chu CM, Lin CK, Pan HH
Int J Med Inform · 2026 Jul · PMID 41880919 · Publisher ↗
BACKGROUND: Chemotherapy is a cornerstone treatment for patients with cancer but frequently induces toxicities. Standard text-heavy symptom monitoring tools often induce visual fatigue, compromising treatment adherence,...
BACKGROUND: Chemotherapy is a cornerstone treatment for patients with cancer but frequently induces toxicities. Standard text-heavy symptom monitoring tools often induce visual fatigue, compromising treatment adherence, quality of life, and survival. OBJECTIVES: This study aimed to design and validate the content of a visualized chemotherapy toxicity self-assessment system. The goal was to integrate patient-reported outcomes (PROs) with standardized severity grading and nurse-developed graphic interventions for systematic post-discharge monitoring. METHODS: The Chemotherapy Toxicity Self-Assessment Questionnaire (CTSAQ) evaluates 12 chemotherapy-related symptoms using a four-grade severity scale. Symptoms and interventions were identified and refined via a Delphi-based consensus with 11 multidisciplinary experts. Reliability testing included internal consistency and test-retest reliability. Construct validity was examined using an exploratory structure analysis with varimax rotation. Feasibility was assessed in patients with cancer completing weekly CTSAQ reports after discharge. RESULTS: The CTSAQ demonstrated excellent content validity (S-CVI 0.98), acceptable internal consistency (Cronbach's α = 0.744), and high test-retest reliability (Spearman's ρ = 0.788-1.000) in a clinical cohort (n = 56). Five symptom clusters were identified, psychological, functional, digestive, systemic, and peripheral clusters, explaining 74.13% of the total variance. Weekly completion rates exceeded 90% (mean completion time, 10-12 min). CONCLUSIONS: The CTSAQ is a valid and feasible nurse-led ePRO instrument for post-discharge monitoring. Its visualized guidance is designed to facilitate symptom self-assessment; preliminary feedback suggests it may enhance patients' sense of security and nursing support during home recovery. Future comparative studies are required to formally evaluate its clinical efficacy and user burden.
Digital twin technologies for supporting self-care in adults with diet-related chronic conditions: a systematic review.
Saeedian Y, Wright C, Jansons P … +3 more , Shen Y, Zhang Y, Maddison R
Int J Med Inform · 2026 Jul · PMID 41880918 · Publisher ↗
BACKGROUND: Diet-related chronic conditions are major contributors to global morbidity and mortality. Effective management of these conditions requires consistent engagement in self-care behaviours such as healthy eating...
BACKGROUND: Diet-related chronic conditions are major contributors to global morbidity and mortality. Effective management of these conditions requires consistent engagement in self-care behaviours such as healthy eating, physical activity, and medication adherence. However, behavioural interventions often lack personalisation, limiting their impact, whereas digital twin (DT) systems, which use digital technologies to generate real-time representations of individuals, offer the potential to support people through adaptive and patient-centred approaches by integrating health data to personalise and optimise self-care strategies. OBJECTIVE: A systematic review was conducted to synthesise and evaluate the use of DT for supporting self-care behaviours among adults with diet-related chronic conditions METHODS: Five electronic databases (PubMed, Web of Science, Embase, CINAHL, and IEEE Xplore) were searched from inception to 31 July 2025. Studies including adults with diet-related chronic conditions, using DT interventions, and reporting changes in self-care maintenance, monitoring, or management were eligible. RESULTS: Of 3,685 records identified, four studies (N = 2,662) met the inclusion criteria, all focusing on type 2 diabetes. All four studies used continuous glucose monitoring (CGM), and three additionally used wearable devices and dietary logs integrated with AI-driven DT platforms to provide personalised feedback and recommendations, often alongside human coaching. Two retrospective studies found substantial reductions in antidiabetic medication use, with one reporting a 74% reduction over one year including discontinuation of insulin in 94% of baseline insulin users, and large class‑specific declines, such as sulfonylureas (-99%) and DPP‑4 inhibitors (-88%), with many participants maintaining HbA1c < 7% on no therapy or metformin monotherapy, and the other showing that 42.8% of participants eliminated all diabetes medications within 90 days while maintaining glycaemic control (via a staged medication‑reduction framework). One randomised controlled trial found that 94% of participants discontinued all T2D medications and 72.7% achieved diabetes remission after one year, accompanied by increased daily steps, improved sleep, and reduced sedentary time (with only 4.7% remaining on metformin alone at one year). Another study found that personalised insulin infusion guided by a DT over 14 days reduced insulin doses by 14-29% and increased time in the target glucose range to 86-97% (personal insulin pump data over a 14‑day collection period) CONCLUSIONS: DT-based interventions demonstrated potential to enhance multi-behavioural self-care and clinical outcomes in adults with type 2 diabetes. Evidence is currently limited to diabetes, highlighting the need for studies in other diet-related chronic conditions and standardised assessment of self-care behaviours.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe