International Journal Of Medical Informatics[JOURNAL]
Sun
200 papers
Emergency department narratives for early risk stratification of serious disposition in older adults after falls: A survival modelling study with external validation.
Tang LYW, Gao ZH
Int J Med Inform · 2026 Sep · PMID 42214283 · Publisher ↗
BACKGROUND: Older adults presenting to the emergency department (ED) following a fall represent a clinically heterogeneous population, ranging from those with minor injuries to those requiring urgent intervention. Accura...
BACKGROUND: Older adults presenting to the emergency department (ED) following a fall represent a clinically heterogeneous population, ranging from those with minor injuries to those requiring urgent intervention. Accurate risk stratification at triage is critical, yet triage decisions are often made before physiological measurements are available. ED narratives documented in free-text fields may encode rich clinical signals collected at ED triage but remain largely underused for decision support. OBJECTIVE: To develop and externally validate a survival model that uses free-text ED narratives to rank older adult fall patients by risk of serious disposition at the point of ED triage, without requiring physiological measurements, to support early resource allocation in high-volume or resource-constrained emergency settings. METHODS: We conducted a retrospective cohort study using the National Electronic Injury Surveillance System (NEISS). The development cohort included ED visits from 2019-2022 with verified unintentional falls (n=7921). The external validation cohort comprised visits from 2013-2022 from hospitals not included in the development cohort. Text embeddings of processed narratives were used as inputs to an XGBoost-based survival model. The reported time from fall to ED arrival was extracted from narrative fields and used as the event time variable. Model performance was assessed using various metrics, including the Inverse Probability of Censoring Weighting Concordant Index (IPCW C-index). RESULTS: The model achieved an IPCW C-index of 0.776 (95% CI: 0.773-0.779) on the development cohort and 0.724 (95% CI: 0.719-0.728) on the external cohort, demonstrating discrimination above chance. Aggregated LIME analysis identified narrative features associated with high risk of serious disposition included mechanisms of injury and contextual factors. CONCLUSION: A survival model using only free-text ED narratives achieved good discrimination for serious disposition after geriatric falls, with external validation (IPCW C-index 0.724) supporting generalizability. This narrative-only approach, requiring no physiological measurements, may enable early risk stratification in resource-limited emergency settings. Prospective validation on real-time triage notes is warranted.
A multimodal EHR-based phenotyping framework integrating consensus clustering and transformer-based clinical NLP: application to autoimmune gastritis.
Pala D, Lenti MV, Santacroce G … +7 more , Bergomi L, Curgu C, Buonocore T, Sirtoli C, Parimbelli E, Lanzola G, Di Sabatino A
Int J Med Inform · 2026 Sep · PMID 42214282 · Publisher ↗
OBJECTIVE: To develop and evaluate a multimodal electronic health record (EHR)-based phenotyping pipeline integrating structured and unstructured clinical data to identify disease subgroups and characterize longitudinal...
OBJECTIVE: To develop and evaluate a multimodal electronic health record (EHR)-based phenotyping pipeline integrating structured and unstructured clinical data to identify disease subgroups and characterize longitudinal trajectories in a real-world setting. MATERIALS AND METHODS: We conducted a retrospective multicenter study including 1,598 patients with autoimmune gastritis. Structured demographic and clinical variables were combined with longitudinal endoscopic and histological data extracted from routine care. A consensus clustering strategy integrating partitioning (K-medoids) and hierarchical approaches was applied to identify robust patient subgroups. Free-text endoscopic reports were processed using a fine-tuned transformer-based natural language processing (NLP) model to automatically extract structured phenotypic features. To address irregular follow-up intervals, time-normalized progression indices were developed to capture both severity and temporal dynamics of disease evolution. RESULTS: After preprocessing, 607 patients were included in the analysis. The consensus clustering approach identified three clinically distinct subgroups. The NLP model demonstrated high performance in extracting endoscopic features (accuracy 90.2%, balanced accuracy 89.3%). Application of the proposed progression indices revealed significant differences in longitudinal patterns of mucosal damage across clusters (p < 0.01). CONCLUSION: This study demonstrates the feasibility of integrating clustering techniques and transformer-based clinical NLP within a unified EHR phenotyping pipeline. The proposed approach supports scalable secondary use of structured and narrative clinical data for subgroup discovery and trajectory modeling in chronic diseases.
Implementation of the VIVALDI Social Care data pipeline for monitoring and research of infections in care homes for older adults in the UK.
Stirrup O, Slator M, Monakhov I … +12 more , Adams N, Green M, Hall V, Hargreaves R, Knight L, Krutikov M, Kwiatkowska R, Meacock K, Copas A, Childe G, Fry Z, Shallcross L
Int J Med Inform · 2026 Sep · PMID 42214281 · Publisher ↗
OBJECTIVES: To create a new pipeline for routine health data of residents of care homes for older adults in England, enabling research and monitoring of infections. METHODS: Automated daily data exports containing person...
OBJECTIVES: To create a new pipeline for routine health data of residents of care homes for older adults in England, enabling research and monitoring of infections. METHODS: Automated daily data exports containing person- and care-home level identifiers for residents are transmitted from digital care record systems of participating providers to a secure platform hosted by National Health Service England, following application of project-specific and national opt-outs. Linkage to person-level routine healthcare data is performed before monthly export to a secure analysis platform. RESULTS: As of March 2026 we have a cohort of 188 care homes including 6762 current residents, with activation of the data pipeline ongoing for further enrolled homes. DISCUSSION: We have demonstrated feasibility of automating linkage of data from care homes to national routine health datasets. CONCLUSIONS: Sustained investment and collaboration between diverse stakeholders are required to realise the opportunities offered by digitisation of social care.
Harnessing machine learning to decode dietary Impacts on cardiometabolic multimorbidity.
Liu R, Tang L, Zhang F … +15 more , Li Y, Tang Q, Zhang X, Li W, Wang S, Zhang Y, Chen L, Ma J, Zou X, Yao T, Tang R, Cai Z, Yi Y, Zeng Y, Zhang L
Int J Med Inform · 2026 Sep · PMID 42208085 · Publisher ↗
OBJECTIVE: The objective of this research is to develop and validate the efficacy of a machine learning model that integrates 13 nutrients with baseline characteristics to predict Cardiometabolic Multimorbidity (CMM). Fu...
OBJECTIVE: The objective of this research is to develop and validate the efficacy of a machine learning model that integrates 13 nutrients with baseline characteristics to predict Cardiometabolic Multimorbidity (CMM). Furthermore, the study aims to examine the role of key nutrients in assessing disease risk and to elucidate the underlying mechanisms involved. METHODS: Data were synthesized from two population-based databases: the NHANES and the CHNS. The analysis included 13 nutrients and seven demographic and health variables. Binary and gradient logistic regressions were used to assess associations. Six machine learning models were trained and validated for generalizability on both NHANES and CHNS datasets. SHAP values were used to interpret feature contributions and understand variable relationships with prediction outcomes. RESULTS: The SVM model demonstrated the best performance, achieving an external validation AUC of 0.874, indicating good predictive ability across populations. SHAP analysis identified age, magnesium, BMI, total fat, vitamin B1, and dietary fiber as important contributors to model predictions. Magnesium, vitamin B1, and dietary fiber showed inverse associations with CMM risk within the modeling framework. While total fat exhibited an inverse association in logistic regression, it played a significant role in model prediction, suggesting that its relationship with CMM may be complex and context-dependent. CONCLUSIONS: A machine learning model integrating nutritional and baseline characteristics may provide a useful approach for predicting CMM risk, with the SVM model showing the best performance. The findings highlight the relevance of multiple dietary factors in risk prediction; however, these associations should be interpreted with caution. Further longitudinal and interventional studies are needed to clarify potential causal relationships.
Mapping, displaying, and facilitating access to autism services: an Italian eHealth solution.
Caruso A, Micai M, Gila L … +4 more , Sabbioni M, Galati G, Fulceri F, Scattoni ML
Int J Med Inform · 2026 Sep · PMID 42202664 · Publisher ↗
BACKGROUND: Autism Spectrum Disorder (ASD) presents a significant global challenge, necessitating robust healthcare services and strategic policy planning. This study describes the development and implementation of a nat...
BACKGROUND: Autism Spectrum Disorder (ASD) presents a significant global challenge, necessitating robust healthcare services and strategic policy planning. This study describes the development and implementation of a national eHealth solution to systematically map ASD healthcare centers across Italy. METHODS: The Italian National Institute of Health, under the mandate of the Ministry of Health, established a multidisciplinary working group to lead a co-design process for developing and implementing an eHealth platform that maps healthcare services for individuals with ASD across all Italian Regions and Autonomous Provinces. RESULTS: The dynamic eHealth platform was developed and refined through a two‑step evaluation process to ensure technical reliability, standardized data collection, and alignment with operational needs. It is publicly accessible through the National Observatory for Autism website and systematically collects geographical and operational characteristics of healthcare services, the composition of multidisciplinary workforce across services, access procedures and public availability of information. As of 2024-2025, the platform has mapped 1,228 services for individuals with ASD and/or intellectual disability, covering both child and adult populations. More than 30,000 professionals are involved in delivering diagnostic and rehabilitative interventions. Analysis highlights regional disparities in service distribution and workforce composition. CONCLUSION: The national eHealth platform provides a standardized, accessible solution that enhances transparency and coordination of ASD services across Italy. By consolidating verified service data, it supports navigation for individuals with ASD and their family and enables policymakers to monitor disparities and inform evidence‑based planning.
Evaluating large language models for structuring cardiology reports: a real-world clinical study on patient subtyping and trial recruitment.
Pescol F, Buonocore TM, Tibollo V … +6 more , Failla G, Traversi E, La Rovere MT, Sacchi L, Ricciardi W, Bellazzi R
Int J Med Inform · 2026 Sep · PMID 42202663 · Publisher ↗
BACKGROUND: Artificial Intelligence (AI) methods have emerged as useful tools for supporting patient recruitment in clinical trials (CTs). Despite several studies having recently proposed promising applications of Large...
BACKGROUND: Artificial Intelligence (AI) methods have emerged as useful tools for supporting patient recruitment in clinical trials (CTs). Despite several studies having recently proposed promising applications of Large Language Model (LLMs) for patient recruitment in CTs, their implementation in routine clinical practice remains limited. METHODS: In this study, we present a comprehensive pipeline, developed and tested in a real-world clinical setting, to obtain highly detailed patient subtyping and eligibility assessment for specific CTs. Our solution leverages cardiological discharge letters, a rich yet underutilized source of patient data, to extract detailed structured clinical information through LLMs. Patient subtyping and eligibility assessment are performed through a rule-based approach, based on the extracted information, to maximize deterministic and interpretable outputs. We employed OpenAI's GPT-4.1 within the cloud-based service Microsoft Azure Machine Learning Studio, deployed in the hospital infrastructure. Validation was conducted on a sample of 100 discharge letters through exact-match comparison between the model's output and a ground-truth template, pre-populated by expert clinicians. RESULTS: Our results confirm the feasibility and effectiveness of the proposed approach in real-world clinical scenarios. GPT-4.1 achieved high values of information extraction accuracy for most clinical variables (0.94 ± 0.08), resulting in a limited number of false negatives (FN) and false positives (FP) in both patient subtyping (0.12 and 0.13, respectively) and eligibility assessment. At the criterion-level, the proportion of FNs and FPs was below 3% for most criteria (13 and 11 of the 14 criteria examined, respectively). CONCLUSION: Overall, our study presents a notable step towards the integration of AI-driven approaches into real-world clinical practice for patient recruitment in CTs, highlighting both its practicality and effectiveness in meeting the stringent demands of healthcare settings.
Enabling real-world evidence for medication safety in pregnancy: Determinants of research participation and linked EHR data use.
Hu YJ, Chan EY, Sipio MD … +2 more , Wang J, Yuan J
Int J Med Inform · 2026 Sep · PMID 42202662 · Publisher ↗
OBJECTIVE: Real-world evidence for medication safety in pregnancy increasingly relies on linked electronic health record (EHR) data. However, the success of such systems depends on pregnant women's willingness to partici...
OBJECTIVE: Real-world evidence for medication safety in pregnancy increasingly relies on linked electronic health record (EHR) data. However, the success of such systems depends on pregnant women's willingness to participate in research and to permit secondary use of their clinical data. We synthesised evidence on determinants shaping research participation and acceptability of linked EHR data use in pregnancy. METHODS: We conducted a mixed-methods systematic review in accordance with PRISMA 2020 and ENTREQ guidelines. MEDLINE, PubMed, CINAHL, and Google Scholar were searched (January 2008-August 2025). Eligible studies examined perspectives of pregnant women, women with children, healthcare professionals, researchers, or the public regarding research participation or EHR data sharing. Thematic synthesis was undertaken, and methodological quality was assessed using CASP checklists. RESULTS: Sixty-one studies (44,722 participants; 22 countries) were included. Participation and EHR data-sharing decisions reflected a balance between perceived benefit and perceived burden, mediated by trust and transparency, governance, regulatory, and health system structures, and infant-centred considerations. Facilitators included perceived maternal or societal benefit, altruistic motivation, trust and transparency, collaborative decision making, quality patient-provider communication, governance transparency, and electronic health literacy. Barriers included recruitment logistics, privacy and confidentiality concerns, biological specimen collection, culture and socioeconomic factors. Willingness to share EHR data varied widely (22-100%), with governance safeguards, trust in institutions and participant control strongly influencing acceptability. CONCLUSION: Acceptability of linked EHR use in pregnancy is shaped by socio-technical factors extending beyond technical safeguards. Real-world evidence initiatives aiming to strengthen medication safety should incorporate transparent governance, proportionate consent models, clinician-mediated engagement, and burden-reducing study designs to support equitable inclusion in learning health systems.
Cognitive adaptability predicts digital health technology use in older adults.
Shakya S, Paudel A
Int J Med Inform · 2026 Sep · PMID 42190430 · Publisher ↗
PURPOSE: Use of digital health technology varies among older adults. The psychological factors influencing the acceptance and use of digital health technology among older adults are limitedly examined. The purpose of thi...
PURPOSE: Use of digital health technology varies among older adults. The psychological factors influencing the acceptance and use of digital health technology among older adults are limitedly examined. The purpose of this study was to examine the association of psychological factors with digital health technology use in this population. METHODS: Cross-sectional data from the National Health Aging Trends Study (NHATS) were used to investigate four types of digital health technology use: seeking health information online, managing health insurance data, refilling prescriptions/scheduling appointments online, and participating in telehealth visits. Psychological factors included internal and external loci of control and cognitive adaptability. We used multivariable logistic regressions to examine associations between each psychological factor and each type of digital health technology use in older adults, adjusting for sociodemographic and clinical factors. RESULTS: There was variation among participants (N = 4651) in the use of digital health technology: making telehealth visits (51.9%), managing health insurance data (42.4%), filling prescriptions/scheduling health appointments (30.3%), and obtaining health information (25.8%). Cognitive adaptability was statistically significantly associated with greater use of seeking online health information, managing health insurance data, refilling prescriptions/scheduling appointments online, and participating in telehealth visits. Internal and external loci of control were not statistically significantly associated with any type of digital health technology use among participants. CONCLUSIONS: This study is among the first to demonstrate that cognitive adaptability is a significant predictor of digital health technology use among older adults. These findings highlight the need to implement user-centered design and to develop and test digital interventions. Furthermore, interventions that capitalize on the cognitive adaptability of older adults using a guided or problem-solving approach may improve their digital health technology usage. These measures can mitigate disparities in digital health technology use in this population.
Assessing the generalisability of foundation models to ultra-wide field retinal imaging for diabetic retinopathy screening in Denmark and Greenland.
Li LY, Thambawita V, Byberg S … +1 more , Hulman A
Int J Med Inform · 2026 Sep · PMID 42184619 · Publisher ↗
BACKGROUND: Foundation models have shown promising performance in ophthalmology image analysis, but their ability to generalize to unseen imaging types and populations remains unknown. We evaluated the generalizability o...
BACKGROUND: Foundation models have shown promising performance in ophthalmology image analysis, but their ability to generalize to unseen imaging types and populations remains unknown. We evaluated the generalizability of ophthalmology foundation models to ultra-wide field (UWF) retinal images for diabetic retinopathy (DR) screening in a Danish and a Greenlandic population. METHODS: Three ophthalmology foundation models (RETFound DINOv2, VisionFM, and EyeCLIP) were fine-tuned and evaluated using 6,374 UWF retinal images from 1,760 participants in Denmark and 6,558 images from 1,146 participants in Greenland. Binary DR classification (normal vs. any retinopathy) was performed under four experimental settings: fine-tuning on the Danish dataset, fine-tuning on the Greenlandic dataset, external validation of Danish-fine-tuned models on the Greenlandic dataset, and sequential fine-tuning from the Danish to the Greenlandic dataset. Model discrimination and calibration were assessed. RESULTS: DR prevalence differs between the Danish and Greenlandic datasets, with 45% and 14% of all images having DR, respectively. When fine-tuned and evaluated within the same population, discrimination was similar in Denmark and Greenland, with RETFound DINOv2 achieving the highest AUROC (0.76 [95% CI: 0.73, 0.78] and 0.76 [0.73, 0.80], respectively). External validation on the Greenlandic dataset showed worse performance across models (AUROC 0.59-0.62). Sequential fine-tuning improved discrimination (AUROC 0.70-0.78). However, calibration remained poor across all settings, with calibration intercepts ranging from -1.69 to 0.37 and slopes from 0.25 to 0.78. CONCLUSION: Foundation models showed limited generalizability when applied to unseen imaging contexts and populations, with disparities in model performance observed across Danish and Greenlandic populations. Local fine-tuning improved discrimination, but did not resolve calibration issues, underscoring the importance of careful calibration evaluation to ensure clinical relevance.
Improving physical activity in men with prostate cancer through wearable devices and online education: Randomized controlled trial.
O'Neill M, Sabiston CM, Tomlinson G … +3 more , Santa Mina D, Sibley D, Alibhai SMH
Int J Med Inform · 2026 Sep · PMID 42176601 · Publisher ↗
OBJECTIVES: Cancer-related fatigue (CRF) is common in men with prostate cancer (PC). Physical activity is helpful in reducing CRF; however, it is unclear if reducing sedentary behaviour (SB) improves CRF. Technology-base...
OBJECTIVES: Cancer-related fatigue (CRF) is common in men with prostate cancer (PC). Physical activity is helpful in reducing CRF; however, it is unclear if reducing sedentary behaviour (SB) improves CRF. Technology-based interventions, wearable technology (WEAR) and online educational workshops (EDU), have shown to increase physical activity and affect health behaviour; however, their impact on CRF are less clear. The aim of this pilot study was to determine the feasibility of a pilot randomized controlled trial of technology-based interventions to reduce SB and improve CRF. METHODS: Participants were adult males with PC, currently sedentary (<150 min per week of activity), fluent in English, and had a computer with internet access. Participants were randomized in a 2x2 factorial design (WEAR only, EDU only, WEAR + EDU, and control). Feasibility metrics included consent rate, retention rate, adherence, and study acceptability. Assessments consisted of Functional Assessment of Cancer Therapy - Fatigue (FACT-F) (Minimal Clinically Important Difference (MCID) = 3), Sit-Q-7d (MCID = 1.9), Accelerometry, Functional Assessment of Cancer Therapy - General (FACT-G) and Patient Health Questionnaire (PHQ-9). Descriptive statistics were used for feasibility outcomes and a constrained longitudinal data analysis was used for efficacy outcomes. RESULTS: Twenty-one participants, mean age of 67 (SD 8.7) years and mainly retired (52.4%), were recruited. Neither consent rate (28/231, 12.1%) nor retention rate (11/21, 52.4%) met feasibility targets; however, there was moderate intervention adherence (3/5, 60%). Trends of reduced SB and CRF were seen in participants who used a WEAR device. Average effects across times for fatigue (FACT-F) were 3.8 (95% CI -5.4, 12.9) for WEAR and -1.1 (-10.4, 8.3) for EDU. Effects on SB (Sit-Q-7d) were 2.3 (-2.8, 7.4) for WEAR and -1.0 (-5.7, 3.7) for EDU. CONCLUSION: Although consent and retention rates were low, participants found the study acceptable and were adherent. Unfortunately, the study was not large enough to precisely estimate treatment effects. PRACTICAL IMPLICATIONS: Future studies should continue to evaluate WEAR and EDU, using strategies to increase recruitment and retention. Clinicians should encourage cancer survivors to reduce SB, increase physical activity through education and wearable devices.
An HL7 FHIR® IG for lifestyle medicine in learning health systems: Multi-vendor wearable interoperability with documented terminology gaps.
Lourenço Santos R, Cruz-Correia RJ
Int J Med Inform · 2026 Sep · PMID 42176600 · Publisher ↗
BACKGROUND: Consumer wearable devices collect clinically useful physiological data from patients, yet less than 5% of this patient-generated health data (PGHD) reaches clinical systems. The European Health Data Space (EH...
BACKGROUND: Consumer wearable devices collect clinically useful physiological data from patients, yet less than 5% of this patient-generated health data (PGHD) reaches clinical systems. The European Health Data Space (EHDS) Regulation 2025/327 mandates standardized health data exchange, creating an urgent need for validated transformation frameworks. OBJECTIVE: To develop and validate a FHIR® R4 Implementation Guide enabling multi-vendor wearable data transformation for lifestyle medicine, while documenting terminology gaps across seven vendor ecosystems. METHODS: We developed FHIR® profiles, extensions, and ConceptMaps using a multi-layer architecture covering eleven domains (six lifestyle medicine pillars and five wearable-specific domains). Technical validation included HL7 IG Publisher compilation, FHIR® Validator conformance testing, and systematic terminology verification. RESULTS: The IG comprises 74 profiles, 50 extensions, 14 CodeSystems, 174 ValueSets, and 28 ConceptMaps (546 artifacts). A two-phase terminology audit refined 1173 custom codes to 718 without standard direct equivalents, including genuine gaps (RMSSD, pNN50), wearable-specific metrics, and granularity not yet addressed by LOINC or SNOMED CT. Build validation produces 23 errors (all IPS upstream) and 73 warnings (97% non-actionable). Of six core HRV metrics, only SDNN has a LOINC code (80404-7). The multi-layer architecture achieved 75% weighted reuse across seven vendor ecosystems. CONCLUSIONS: This work extends the HL7 Physical Activity IG's "Temporary Codes" pattern into a semantic convergence buffer-a vendor-neutral intermediary CodeSystem with outbound ConceptMap coverage to LOINC, SNOMED CT, and OMOP enabling multi-vendor semantic convergence. The 75% weighted reuse and iterative terminology audits progressively reduce custom codes as standards mature. All profiles are vendor-agnostic and EHDS-compliant.
Patient Perceptions and preferences for the disclosure of artificial intelligence generated draft replies to electronic messages - A qualitative study.
Barrison PD, Platt J, Ackerman MS … +2 more , Friedman CP, Vinson AH
Int J Med Inform · 2026 Sep · PMID 42176599 · Publisher ↗
BACKGROUND: Since their public release, artificial intelligence-generated draft replies (GDRs) to patient portal messages have been rapidly adopted across healthcare systems. Concurrently, debate has arisen regarding the...
BACKGROUND: Since their public release, artificial intelligence-generated draft replies (GDRs) to patient portal messages have been rapidly adopted across healthcare systems. Concurrently, debate has arisen regarding the extent to which GDRs should be disclosed to patients. This qualitative interview study, using vignettes, examines adult patients' reactions to and preferences regarding written AI disclosure statements. METHODS: Semi-structured interviews were conducted with 30 adult portal users with no prior experience with GDRs. Participants were recruited from a patient research registry at a large academic health system. Eligible participants were stratified by age and educational attainment and randomly sampled. Participants shared their initial interpretations, reactions, and preferences regarding GDRs with written disclosure statements, using four case-based vignettes. Interview data were analyzed through an interpretivist epistemology and a systematic process of inductive code and category generation and synthesis. RESULTS: Patient interpretations of disclosure statements suggested that written disclosure statements did not provide sufficient information to clarify how AI was used, with nearly half of participants unable to identify AI as the draft's writer. GDR disclosure statements elicited participant concerns about error, authorship, and depersonalization. Despite these concerns, most participants expressed a preference for disclosure practices, citing a perceived right to information, agency, and autonomy in their healthcare. Those participants who expressed the most concerns when presented with written disclosure statements argued for preemptive disclosure methods that facilitate education and trust building. CONCLUSION: As healthcare systems establish best practices for transparency and disclosure of GDRs, accounting for evolving patient perspectives on AI will be crucial to maintaining patient trust and confidence. The reactions and preferences for disclosure methods outlined by participants in this study suggest that careful deliberation should be given not to whether to disclose, but how to disclose, as healthcare systems balance tensions between AI transparency and patient trust.
From documentation to discovery: clinicians' perspectives on the generation, usability and standardization of real-world data.
Ross E, Bouissou O, Helland Å … +1 more , Faxvaag A
Int J Med Inform · 2026 Sep · PMID 42176598 · Publisher ↗
BACKGROUND: Real-world data (RWD) are increasingly recognized as essential for understanding patient populations underrepresented in clinical trials and for supporting data-driven learning in healthcare. For smaller subg...
BACKGROUND: Real-world data (RWD) are increasingly recognized as essential for understanding patient populations underrepresented in clinical trials and for supporting data-driven learning in healthcare. For smaller subgroups, the value of RWD depends on standardization and interoperability that enable meaningful reuse across institutions. This study examines how clinicians perceive the reuse and standardization of RWD within a federated Learning Health System (LHS), with emphasis on data quality, clinical relevance, and implications for continuous learning. METHODS: A qualitative case study was conducted at Oslo University Hospital, informed by Learning Health System theory. Ten oncologists representing seven cancer subspecialties participated in focused, semi structured interviews. Data were analyzed using the stepwise deductive inductive (SDI) method to support empirically grounded conceptual development. The study was situated in the hospital's implementation of the OMOP Common Data Model (CDM) for oncology data. RESULTS: Clinicians highlighted the value of RWD for capturing patient groups often excluded from clinical trials. They described substantial variation in documentation practices, noting that clinically relevant information is frequently unstructured or inconsistently recorded. Time constraints and uncertainty about documentation requirements were identified as barriers to data quality. When reviewing data transformed into the OMOP CDM, participants generally found the mappings accurate but expressed concerns about loss of nuance and semantic drift. Across interviews, there was strong support for involving domain experts in validation and for using standardized data to enable collaboration and learning across institutions. CONCLUSIONS: RWD can strengthen oncology practice by supporting insights into diverse patient populations and enabling continuous learning. Standardization through models such as OMOP CDM facilitates reuse and cross institutional collaboration, but success depends on structured documentation, semantic fidelity, clinician engagement, and robust technical infrastructure. These findings underscore the sociotechnical conditions required to realize the potential of RWD within emerging frameworks such as the European Health Data Space.
Advancing stroke prevention in atrial fibrillation: a systematic review of machine learning-based risk prediction models.
Islam MM, Nkemdirim Okere A
Int J Med Inform · 2026 Sep · PMID 42176597 · Publisher ↗
BACKGROUND: Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and confers a four to fivefold increase in ischemic stroke risk, accounting for approximately 15 - 20% of all stroke events globally. D...
BACKGROUND: Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and confers a four to fivefold increase in ischemic stroke risk, accounting for approximately 15 - 20% of all stroke events globally. Despite this burden, the predominant risk stratification tool, the CHADS-VASc score, achieves only modest discrimination, constrained by its static, additive architecture that cannot capture the nonlinear, high-dimensional interactions inherent in real-world electronic health record (EHR) data. This evidence gap creates a dual clinical hazard: under-anticoagulation in high-risk patients and unnecessary bleeding exposure in those whose risk is overestimated. This study aimed to systematically evaluate the predictive performance, methodological rigor, and clinical readiness of machine learning (ML) models derived from EHR data for the prediction of ischemic stroke in patients with AF. METHODS: A systematic search of PubMed, Embase, Scopus, and Web of Science was conducted from inception through September 2025, following PRISMA 2020 guidelines. Studies were eligible if they developed or validated ML models for ischemic stroke prediction using EHR data in adults with AF and reported at least one quantitative performance metric. Methodological quality was assessed using the PROBAST and TRIPOD-AI frameworks. RESULTS: Eight studies (2017 to 2024) encompassing 809,523 patients across seven countries were included. Supervised ensemble methods consistently outperformed CHADS-VASc, with AUROCs ranging from 0.66 to 0.91 versus 0.54 to 0.68 for the traditional score. However, performance varied substantially: several models achieved only marginal gains (AUROC 0.63 - 0.69), and the AUROC range reflects pronounced heterogeneity rather than uniform superiority. Critical barriers persist - only one study performed external validation; fewer than half applied explainable AI techniques; class imbalance was rarely addressed; and 88% of studies received a high risk of bias rating in the analysis domain under PROBAST, a finding that substantially limits confidence in the reported performance estimates. CONCLUSION: In light of the pervasive methodological limitations identified, including high analytic risk of bias, absence of external validation, and lack of model interpretability, claims of ML superiority over CHADS-VASc must be interpreted with caution. While ML models demonstrate potential discriminative improvements, current evidence is insufficient to support clinical adoption. Translating algorithmic promise into bedside impact requires dynamic longitudinal modeling, rigorous multisite external validation, transparent risk attribution, and prospective evaluation within real-world EHR workflows.
Development and validation of an interpretable machine learning model for early risk prediction of acute myocardial infarction.
Cui S, Gao L, Zhang N … +2 more , Zhang H, Gong N
Int J Med Inform · 2026 Sep · PMID 42172727 · Publisher ↗
BACKGROUND: Acute myocardial infarction (AMI) remains a leading cause of global morbidity and mortality, with early prediction critical for timely intervention. Traditional risk assessment tools are limited because of th...
BACKGROUND: Acute myocardial infarction (AMI) remains a leading cause of global morbidity and mortality, with early prediction critical for timely intervention. Traditional risk assessment tools are limited because of their reliance on limited variables and static thresholds. This study aims to develop an interpretable machine learning (ML) model using multidimensional clinical data for early AMI risk prediction. METHODS: We performed a retrospective cohort study of 7939 patients enrolled from the second hospital of Shandong University from January 2020 to January 2024. A total of 108-dimensional clinical features composed of epidemiological data and biochemical data were collected, followed by data preprocessing. ML models were constructed via various algorithms with GridSearchCV hyperparameter tuning, and model performance was evaluated via 5-fold cross-validation. The SHapley Additive exPlanations (SHAP) method was employed to interpret the model, and the top 10 features were selected to simplify the model and maximize predictive performance. The final model was externally validated using an independent cohort of 532 patients collected from January 2025 to April 2025. RESULTS: The weighted model with the XGBoost algorithm achieved the best performance, with accuracy of 0.864, F1-score value of 0.797, and prediction uncertainty lower than 0.01 on the test set. SHAP analysis revealed nonlinear interactions among metabolic profile, coagulation status, and demographic factor, identifying Hs-cTnI as the primary AMI predictor alongside NT-proBNP, LDL-C, CG, D-dimer, AST, PLT, GLU, female sex, and BMI. The optimal model showed wide applicability and strong robustness, confirmed by the accuracy of 0.932 on the independent validation dataset (n = 488 in applicability domain). An interactive webserver embedded with the optimal model was developed to enhance practicability (https://www.mips.net.cn). CONCLUSIONS: An explainable ML model effectively predicted AMI risk integrating multimodal clinical data, offering a publicly accessible webserver generated for the optimal model facilitated its utility in clinical settings.
Developing and validating a sequence-aware deep learning model for infection risk prediction in home care.
Xu Z, Zhou S, Song J … +4 more , Russell D, Zolnoori M, Shang J, Topaz M
Int J Med Inform · 2026 Sep · PMID 42172726 · Publisher ↗
BACKGROUND: Infections are a leading cause of hospitalizations and emergency department (ED) visits in home care. Existing prediction tools often underutilize longitudinal patterns of home care electronic health record (...
BACKGROUND: Infections are a leading cause of hospitalizations and emergency department (ED) visits in home care. Existing prediction tools often underutilize longitudinal patterns of home care electronic health record (EHR) and narrative clinical notes. OBJECTIVES: To evaluate a sequence-aware deep learning approach that integrates longitudinal structured EHR data with Natural Language Processing (NLP)-derived features from clinical notes to predict infection-related hospitalizations or ED visits, derive a three-tier risk stratification tool from model outputs, and assess performance, interpretability, and fairness. METHODS: This retrospective cohort study analyzed 23,321 home care episodes (2015-2017), of which 1,528 (6.5%) involved an infection-related hospitalization or ED visit within 30 days of admission. We compared sequence-aware models against a non-sequential baseline. Models used 28 admission-level and 30 visit-level features (6 structured measures and 24 NLP-derived indicators) to represent longitudinal home visit data. Models estimated the risk of first infection-related hospitalization or ED visit over 2-, 3-, and 4-day windows. Evaluation metrics included the Area Under the Receiver Operating Characteristic (AUROC) curve, Area Under the Precision-Recall Curve (AUPRC), and F1-score. RESULTS: Sequence-aware models consistently outperformed the non-sequential baseline, with larger gains observed when NLP-derived features were included. The best configuration (bidirectional long short-term memory at 4-day lead time) achieved AUROC 0.991, AUPRC 0.885, and F1-score 0.774. The three-tier stratification concentrated 77.7% of infection events within the highest-risk 5% of episodes. Interpretability analyses highlighted the importance of recent visit information, and fairness analyses showed equitable performance across most demographic and socioeconomic subgroups. CONCLUSION: A sequence-aware deep learning model integrating longitudinal structured EHR data with NLP-derived features can accurately and equitably predict near-term risks of infection-related hospitalizations or ED visits among home care patients. The three-tier tool may support targeted triage and early intervention for patients at risk of infection-related acute care utilization.
Psychometric evaluation of the Spanish-language version of the health information technology usability evaluation scale (Health-ITUES).
Gerchow L, Schnall R, Liu J
Int J Med Inform · 2026 Sep · PMID 42172725 · Full text
INTRODUCTION: Rigorous development of mobile health technologies must include assessments to ensure interventions are effective and sustainable across diverse populations. Spanish-speaking individuals represent the large...
INTRODUCTION: Rigorous development of mobile health technologies must include assessments to ensure interventions are effective and sustainable across diverse populations. Spanish-speaking individuals represent the largest linguistic minority in the United States and globally. Several general and modality-specific usability instruments exist in Spanish; however, a validated Spanish version of Health-ITUES-a health IT-specific, customizable instrument designed for broad HIT contexts-has been lacking. The purpose of this study was to evaluate the psychometric properties of the Spanish-language version of the Health Information Technology Usability Evaluation Scale (Health-ITUES). METHODS: We analyzed survey data from 125 Spanish-speaking participants (aged 18-64 years), including 18 health professionals from the Dominican Republic and 107 persons living with HIV in New York City or the Dominican Republic. Participants interacted with a mobile health technology and completed the Health-ITUES after use. Psychometric evaluation included exploratory factor analysis, internal consistency reliability, construct validity, and criterion validity assessed through correlations with the Post-Study System Usability Questionnaire (PSSUQ). RESULTS: Exploratory factor analysis supported a three-factor structure (Impact and Perceived Usefulness, Perceived Ease of Use, and User Control) with strong model fit (RMSR = 0.04) and internal consistency (Cronbach's α = 0.79-0.96). Within-factor correlations were higher than between-factor correlations, supporting construct validity. Criterion validity was established through significant correlations with PSSUQ subscales (r = 0.44-0.86, p < 0.001). CONCLUSION: The Spanish Health-ITUES demonstrated strong reliability and validity in both patient and provider samples. Findings will contribute to the rigor of research by developing a linguistically and culturally relevant usability instrument to improve inclusion of Spanish-speaking populations in digital health research.
Early diagnosis and risk stratification of aortic stenosis using artificial intelligence applied to echocardiography: scoping review.
Malagón Tarqui GE, García JV, Hernández Rincón EH
Int J Med Inform · 2026 Sep · PMID 42161114 · Publisher ↗
INTRODUCTION: Aortic stenosis (AS) is the most common acquired valvular heart disease worldwide, accounting for 43% of valvular diseases. It is estimated that 40-50% of patients with severe symptomatic AS do not receive...
INTRODUCTION: Aortic stenosis (AS) is the most common acquired valvular heart disease worldwide, accounting for 43% of valvular diseases. It is estimated that 40-50% of patients with severe symptomatic AS do not receive intervention, resulting in a mortality rate of over 90%. Transthoracic echocardiography remains the gold standard for diagnosis, making it critical for early detection of the disease. However, it is operator-dependent and varies according to the patient's clinical presentation. In this context, artificial intelligence algorithms, especially deep learning algorithms applied to echocardiography, are emerging as tools with the potential to automate and improve the detection of aortic stenosis. OBJECTIVE: To evaluate the available evidence on the usefulness of artificial intelligence tools applied to echocardiography for the early diagnosis of aortic stenosis, identifying their performance, clinical applicability, and methodological limitations. METHODS: A scoping review was conducted in four databases (PubMed, Scopus, Web of Science, and BIREME) in accordance with the PRISMA-ScR guideline, which included 25 studies between January 2020 and December 2025 that used AI systems applied to echocardiography for the early diagnosis and risk stratification of aortic stenosis. RESULTS: Twenty-five studies met the inclusion criteria for this review. Artificial intelligence (AI) algorithms, especially convolutional neural networks, achieved heterogeneous performance. The AUC ranged from 0.82 to 0.99; sensitivity was 82.2-90% and specificity was 88-99%. Multivision models performed better than single-vision models. CONCLUSIONS: Artificial intelligence algorithms perform well in detecting and classifying the severity of AS. Their performance shows high diagnostic potential in retrospective datasets, reaching metrics that emulate expert accuracy. Critical barriers remain, such as lack of external validation, interpretability, and clinical integration. Prospective multicenter studies with harmonized regulatory frameworks are needed for global validation.
Continuous learning and improvement cycles to improve first contact provider assignments at a large academic health system.
Will J, Kothari U, Blecker SB … +4 more , Roncoli T, Moeller B, Testa P, Feldman J
Int J Med Inform · 2026 Sep · PMID 42161113 · Publisher ↗
BACKGROUND: Communication failures are a leading cause of sentinel events in U.S. healthcare, often due to unclear provider contact identification. The electronic health record (EHR) system offers a solution by enabling...
BACKGROUND: Communication failures are a leading cause of sentinel events in U.S. healthcare, often due to unclear provider contact identification. The electronic health record (EHR) system offers a solution by enabling the discrete assignment of a first contact provider (FCP), who oversees and coordinates patient care. However, adoption of this practice is inconsistent across many hospital settings. This study describes the impact of continuous learning and improvement cycles to address this challenge. METHODS: Following the Plan-Do-Study-Act (PDSA) lifecycle, we completed five quality improvement cycles. Each PDSA cycle included a technological intervention accompanied by evolving operational expectations for clinical staff. We evaluated improvement after each PDSA by measuring the percent of a hospitalized patient's time with an assigned FCP. RESULTS: FCP coverage significantly improved from a baseline average of 5.1% to 59.0% after PDSA Cycle 1 (p < 0.001), 67.4% after Cycle 2 (p < 0.001), 79.7% after Cycle 3 (p < 0.001), 87.5% after Cycle 4 (p < 0.001), and 99.4% after Cycle 5 (p < 0.001). CONCLUSION: Having a reliable FCP at any point during a patient's hospital admission is an important safety practice. Continuous learning and improvement cycles, driven by a strong partnership between technology and operations, led to significant and sustained improvements in FCP assignments.
Explainable TabNet for gestational diabetes prediction with physician-in-the-loop and multi-site clinical validation.
Narayanan A, Sankaran P, Sankar UV … +2 more , Kurian S, John T
Int J Med Inform · 2026 Aug · PMID 42155534 · Publisher ↗
BACKGROUND: Gestational diabetes mellitus (GDM) affects 15-25% of pregnancies worldwide and poses serious risks of macrosomia, preeclampsia, neonatal hypoglycaemia, and long-term type 2 diabetes. Existing machine learnin...
BACKGROUND: Gestational diabetes mellitus (GDM) affects 15-25% of pregnancies worldwide and poses serious risks of macrosomia, preeclampsia, neonatal hypoglycaemia, and long-term type 2 diabetes. Existing machine learning models lack prospective multi-site external validation and formal physician trust evaluation, limiting real-world applicability. OBJECTIVES: To develop a clinically validated, explainable deep learning framework for GDM prediction using routinely available first-antenatal-visit clinical features, and to evaluate clinical readiness through dual-stage physician-in-the-loop (PITL) validation. METHODS: A TabNet binary classifier was developed on 3,525 clinical records using a three-stage feature-tailored hybrid imputation strategy (GAIN for HDL and OGTT; MissForest for Systolic BP; Mean for BMI). To prevent data leakage, SMOTE-based class balancing was applied exclusively within the training folds of a 5-fold stratified cross-validation pipeline, with validation folds remaining untouched. Explainability was delivered through TabNet intrinsic feature masks, SHAP, and LIME. Two-stage clinical validation comprised: (1) blinded PITL review by four certified obstetricians evaluating 30 patient cases with XAI explanations; and (2) prospective external validation across three independent Kerala hospitals totaling 80 patients. RESULTS: The proposed TabNet model achieved 97.13% accuracy, 94.05% precision, 98.91% recall, and 96.22% F1-score, outperforming ten baseline classifiers including Random Forest, XGBoost, and SVM under identical preprocessing conditions. Compared to recent state-of-the-art GDM prediction studies, the proposed model consistently outperformed comparable methods-under a rigorous 5-fold cross-validation strategy with confidence intervals, while most existing studies rely on single train-test splits without cross-validation. PITL validation yielded 96.7% concordance, an average Cohen's kappa of 0.909, and Fleiss' kappa of 0.963, with no prior GDM study reporting such formal physician endorsement. External multi-site F1 scores ranged from 83.70% to 87.00% across all three hospitals, reflecting an expected performance reduction in prospective real-world data, partly attributed to inter-site variability in feature availability and clinical data recording protocols. SHAP analysis identified a strong model-level interaction between PCOS and prediabetes as the dominant combined GDM risk signals, independently corroborated by all four obstetricians. CONCLUSION: The proposed framework integrates explainable deep learning with prospective dual-stage clinical validation, demonstrating promising performance as a clinically oriented proof-of-concept for the assessment of risk of GDM using routine clinical variables. TRIAL REGISTRATION: Clinical Trials Registry India, CTRI/2024/08/073158.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe