International Journal Of Medical Informatics[JOURNAL]
Sun
200 papers
Large language models exhibit greater diagnostic anchoring than physicians in a forced-choice vignette study.
Sheppert A, Shen C, Geissal E … +2 more , Riley C, Riley S
Int J Med Inform · 2026 Jun · PMID 42335861 · Publisher ↗
BACKGROUND: Anchoring bias is a well-described contributor to diagnostic error in clinical reasoning. However, the susceptibility of large language models (LLMs) to diagnostic anchoring remains poorly characterized. Usin...
BACKGROUND: Anchoring bias is a well-described contributor to diagnostic error in clinical reasoning. However, the susceptibility of large language models (LLMs) to diagnostic anchoring remains poorly characterized. Using a forced-choice ranking paradigm, we compared the susceptibility of LLMs and human physicians to anchoring in a controlled vignette task. METHODS: We conducted a forced-choice ranking study using nine pairs of internal medicine outpatient-focused clinical vignettes. Each pair consisted of an anchored vignette containing a patient statement suggesting a plausible but unlikely diagnosis and a matched control without the suggestion. Postgraduate year 1-3 (PGY1-3) internal medicine residents (N = 20) and attending physicians (N = 5) each selected and ranked a top-5 differential diagnosis from a fixed list for each vignette (225 total vignette responses). Eight LLMs were evaluated on identical forced-choice ranking tasks (72 anchored and 72 control outputs). The primary endpoint was the proportion of rankings where the anchor diagnosis was ranked first in anchored vignettes. Comparisons used Cochran-Mantel-Haenszel (CMH) tests stratified by vignette. Secondary endpoints included anchor inclusion in the top-5 diagnoses and anchoring magnitude quantified by Borda count score shifts between anchored and control conditions. RESULTS: In anchored vignettes, residents ranked the anchor first in 17/80 responses (21.2%) and attendings in 2/20 (10.0%), compared with LLMs in 40/72 responses (55.6%). LLMs had higher odds of ranking the anchor first versus residents (OR 3.94, 95% CI 2.32-7.33; P < 0.001) and attendings (OR 4.07, 95% CI 2.47-7.21; P < 0.001). LLMs included the anchor in the top-5 in 97.2% of outputs versus 67.5% (residents) and 65.0% (attendings). Mean anchoring-induced Borda score shifts were 2.29 (LLMs), 0.88 (residents), and 0.87 (attendings); the LLM-resident difference was 1.41 (P = 0.008). CONCLUSIONS: In this forced-choice vignette paradigm, LLMs exhibited significantly greater diagnostic anchoring than physicians. These experimental findings motivate evaluation of anchoring susceptibility before integrating LLMs into clinical diagnostic workflows and underscore the need for studies in open-ended, real-world clinical settings.
Patient experience in an EMR-enabled outpatient clinic: a cross-sectional convergent mixed-methods study.
Wang W, Samadbeik M, Puri G … +5 more , McLeod DSA, Lobo E, Duong T, Kirwa T, Sullivan C
Int J Med Inform · 2026 Jun · PMID 42335860 · Publisher ↗
BACKGROUND: Electronic medical records (EMRs) are widely implemented across health settings and function as sociotechnical systems that shape clinical workflows, information use, and patient-clinician interaction. While...
BACKGROUND: Electronic medical records (EMRs) are widely implemented across health settings and function as sociotechnical systems that shape clinical workflows, information use, and patient-clinician interaction. While EMR impacts on clinician experience have been extensively studied, patient experience of EMR-enabled care remains underexplored. This study aims to examine patient experience in an EMR-enabled outpatient clinic and identify actionable recommendations to optimise clinic outcomes. METHODS: A cross-sectional, convergent mixed-methods survey was conducted in a fully digital public diabetes outpatient clinic in Queensland, Australia. Quantitative data, collected using the Patient Experience Monitor (PEM) Adult Outpatient short-form aligned with Picker principles, assessed patient experience across multiple outpatient care domains. Qualitative data, collected through two open-ended items, explored how patients experienced care in the context of clinician-mediated EMR use during consultations and identified opportunities for improvement. Data were collected concurrently and analysed separately. Integration occurred at the reporting stage, where qualitative findings were used to explain and contextualise the quantitative results and to inform practical recommendations. RESULTS: One hundred patients participated. Quantitative findings showed highly favourable but ceiling-affected patient experience ratings across PEM domains. Qualitative analysis identified four themes: perceived facilitation of informational continuity and coordination of care; perceived reduction in personal interaction; limited patient and GP access beyond the public hospital EMR environment; and background trust and neutral perceptions of EMR use. Integration of findings informed a set of actionable recommendations to optimise EMR-supported workflows, preserve interpersonal engagement, strengthen information continuity across care settings, and enable more participatory models of outpatient care. CONCLUSIONS: Patients perceived aspects of EMR-enabled outpatient care as supporting patient-centred care, particularly when clinicians used integrated information effectively during consultations. Findings highlight the importance of implementing EMRs as sociotechnical systems that not only align with consultation workflows but also preserve interpersonal connection and support participatory care. Achieving this requires meaningful information access and sharing across patients, clinicians, and care settings, providing practical guidance for designing digitally enabled outpatient services.
Fast healthcare interoperability resources (FHIR) implementation guide creation process: scoping review.
Corrêa Rampinelli VP, Dos Santos RA, Celuppi IC … +5 more , Santos GN, Hammes JF, Cunha CL, Wazlawick RS, Dalmarco EM
Int J Med Inform · 2026 Jun · PMID 42322886 · Publisher ↗
BACKGROUND: The Fast Healthcare Interoperability Resources (FHIR) standard is a global benchmark for digital health data exchange. Despite its widespread adoption, the scientific literature on the methodological processe...
BACKGROUND: The Fast Healthcare Interoperability Resources (FHIR) standard is a global benchmark for digital health data exchange. Despite its widespread adoption, the scientific literature on the methodological processes for creating FHIR Implementation Guides (IGs) remains fragmented and lacks systematization. OBJECTIVE: This scoping review aims to synthesize the scientific literature on the process of developing FHIR IGs for electronic health records, identifying methodological steps, toolchains, governance patterns, and critical gaps that limit clinical adoption. METHODS: Following JBI and PRISMA-ScR guidelines, a comprehensive search was conducted across nine databases in August 2025. From an initial 5,552 records, eleven studies published between 2021 and 2025 were selected for analysis. Data extraction focused on development stages, team composition, authoring tools, and validation workflows. RESULTS: The study identified a synthesized seven-step implementation lifecycle: requirements, modeling, terminology, narrative documentation, technical validation, clinical validation, and publication. A significant methodological shift toward "Infrastructure as Code" was observed, with frequent use of FHIR Shorthand (FSH) and Continuous Integration/Continuous Deployment (CI/CD) pipelines, particularly in European national initiatives. While technical validation was nearly universal (10 out of 11 studies), clinical validation was inconsistently addressed (7 out of 11 studies), often relegated to future work, resulting in IGs that are syntactically correct but insufficiently aligned with real-world clinical workflows. Narrative documentation, essential for non-technical stakeholders, was reported as comprehensive in only four studies, limiting broader clinical adoption. All studies reported multidisciplinary team involvement, confirming that IG development is fundamentally an exercise in clinical governance and consensus. CONCLUSION: The creation of FHIR IGs has evolved into a complex discipline requiring a convergence of software engineering, clinical semantics, and institutional governance. This review advances beyond existing FHIR literature by providing the first systematic synthesis focused specifically on the IG creation process, presenting a replicable seven-step cycle that can guide implementers, researchers, and policy makers. Closing the gap between technical readiness and clinical applicability remains a critical challenge. Future research should prioritize ongoing clinical validation, the establishment of standardized reporting frameworks for IG development, and the integration of generative AI tools to enhance narrative documentation and terminology binding.
From bedside observations to clinical decision support system (CDSS) rules: using real-world adverse drug events (ADEs) data to identify high-risk iatrogenic situations.
Dintilhac A, Lohan L, Laureau M … +4 more , Perier D, Cestac P, Juillard-Condat B, Breuker C
Int J Med Inform · 2026 Jun · PMID 42322885 · Publisher ↗
INTRODUCTION: Adverse drug events (ADEs) constitute a major clinical and economic burden in Europe. While hospital pharmacy activities improve prescription safety, pharmacists cannot review all orders in time and must pr...
INTRODUCTION: Adverse drug events (ADEs) constitute a major clinical and economic burden in Europe. While hospital pharmacy activities improve prescription safety, pharmacists cannot review all orders in time and must prioritize high-risk patients. Rule-based clinical decision support systems (CDSS) offer an additional preventive strategy but often generate excessive, low-relevance alerts. OBJECTIVE: To develop rules for detecting iatrogenic risk in accordance with methodological standards reported in the literature, using ADEs identified in a prospective cohort of adult patients admitted to the emergency department of a French healthcare institution (2,600-bed tertiary care center). METHODS: ADEs were identified through a structured medication history interview conducted by a trained clinical pharmacist upon the patient's admission to the emergency department. To focus on the most critical situations, drug classes defined at the fourth level of the Anatomical Therapeutic Chemical (ATC) classification system (ATC4) and associated with the highest risk were identified by considering prescription frequency, ADE occurrence, and ADE severity. For each selected ATC level 4 class, logistic regression models were used to assess the association between ADE probability and specific explanatory factors. These factors were then operationalized into rules designed to detect iatrogenic risk. RESULTS: A total of 245 ATC4 classes were involved in at least one ADE. Among these, 22 classes were identified as high iatrogenic risk, accounting for approximately 50% of prescriptions leading to an ADE, with vitamin K antagonists and heparins showing the highest risk. Regression analyses resulted in 58 distinct rules: 31 (53.4%) combined prescription data with at least one laboratory parameter, 8 (13.8%) incorporated demographic variables (age or sex), and 19 (32.8%) were based solely on medication prescription data. CONCLUSION: The clinical and pharmaceutical relevance of the proposed rules must be further evaluated to reduce excessive alert generation, which may lead to disengagement from both pharmacists and prescribers. The institutional health data warehouse could provide an appropriate environment for this evaluation.
Telehealth utilization among adults with cognitive difficulty: Evidence from the National Health Interview Survey, 2020-2023.
Gonzalez Villarreal E, Hong YR, Breithaupt A … +3 more , Kripalani S, Kiselica A, Kulshreshtha A
Int J Med Inform · 2026 Jun · PMID 42320084 · Publisher ↗
BACKGROUND: Telehealth may expand care access by reducing transportation barriers, yet little is known the extent to which cognitive difficulties impact telehealth usage. METHODS: We analyzed cross-sectional data from th...
BACKGROUND: Telehealth may expand care access by reducing transportation barriers, yet little is known the extent to which cognitive difficulties impact telehealth usage. METHODS: We analyzed cross-sectional data from the National Health Interview Survey (NHIS) from 2020 to 2023. Multivariate logistic regression assessed the association between cognitive difficulty and telehealth utilization, controlling for age, sex, race/ethnicity, education, insurance, and year. RESULTS: Among 72,611 adults (mean age 54.6 years ± 18.5 years, 57.0 % female, 11.1 % African American), telehealth use was higher among individuals with significant cognitive difficulty (51.1 %) than those without (32.7 %) (OR = 2.41, 95 % CI: 2.16-2.69). Age modified this association (interaction p < 0.001), with a weaker effect among adults ≥ 50 (OR = 1.33, 95 % CI: 1.26-1.40) than among younger adults (OR = 1.71, 95 % CI: 1.63-1.80). CONCLUSIONS: Participants with greater cognitive difficulty were more likely to use telehealth, though the association was attenuated in adults > 50 years.
Current progress and obstacles for automated classification of causes of death based on death certificates: A systematic review.
Li C, Meng Q, Wells S … +3 more , Barbieri S, Jackson R, Poppe K
Int J Med Inform · 2026 Jun · PMID 42320083 · Publisher ↗
OBJECTIVE: While manual coding or rule-based software are approaches used by most countries for cause of death classification, the application of advanced deep learning tools is likely to enhance the efficiency and accur...
OBJECTIVE: While manual coding or rule-based software are approaches used by most countries for cause of death classification, the application of advanced deep learning tools is likely to enhance the efficiency and accuracy of national mortality statistics. To systematically review the current implementation of automated coding or categorising tools for cause of death classification, summarising the methodologies applied, performance achieved, and the progress and obstacles for application. METHODS: PubMed and Scopus were systematically searched from 2018 to 2024 to identify studies that used automated tools to code or categorise the causes of death. Two researchers independently selected the papers with disagreement adjudicated by a third supervisor. For each study, the general profile, detailed methodology, and performance of the tools were extracted with progress and potential obstacles for implementation assessed qualitatively. An adapted version of QUADAS-2 was used to assess the risk of bias. RESULTS: Among the 46 included studies, the training sample size ranged from 165 to 10,519,268 people. The most popular approaches used were deep learning (n = 22, of which 7 were recurrent neural network and 6 were transformer) and rule-based (n = 15) automation. Large disparities existed in the performance, with recall (sensitivity) ranging from 0.253 to 1.000 and precision (positive predictive value) ranging from 0.396 to 1.000. Precision was often higher than recall and could vary substantially for different settings within the same study. Quality of text was the major obstacle to implementation of older automated tools, while for deep learning models, target task and materials were required for pre-training. The performance of deep learning was unsatisfactory for infrequent causes of death and head-to-head comparisons of performance with rule-based tools were limited. CONCLUSION: Despite deep learning applications gaining popularity over rule-based tools, their performance is inconsistent and evidence of head-to-head comparisons is insufficient. All approaches are influenced by the quality of the training data.
Exploring risk factors for long-term sickness absence during emerging adulthood: Continuous and discrete time models using Young-HUNT data on psychological distress and chronic pain.
Gorosito MA, Yazidi A, Hermansen Å … +5 more , Berg B, Øiestad BE, Grotle M, Richardsen KR, Haugerud H
Int J Med Inform · 2026 Jun · PMID 42320082 · Publisher ↗
INTRODUCTION: Long-term sickness absence (LTSA) in young adults has important consequences for labour market participation and future work disability. Chronic pain and psychological distress are key risk factors and freq...
INTRODUCTION: Long-term sickness absence (LTSA) in young adults has important consequences for labour market participation and future work disability. Chronic pain and psychological distress are key risk factors and frequently co-occur, yet their combined impact during adolescence on later LTSA remains insufficiently understood. This study aims to explore factors that influence adolescents' and young people's risk of receiving LTSA benefits during emerging adulthood. METHODS: This longitudinal study used data from the Young-HUNT1 (1995-1997; n = 8736) and Young-HUNT3 (2006-2008; n = 7935) cohorts linked to Norwegian registry data and followed into early adulthood. The outcome was time to LTSA (≥90 or ≥180 days). Associations were examined using Cox proportional hazards models and Kaplan-Meier analyses. Continuous- and discrete-time models were developed and evaluated using the concordance index, time-dependent AUC, and integrated Brier score. Risk factors were analysed using SurvSHAP, SHAP, and regression-based methods. RESULTS: Chronic pain and co-occurring pain and psychological distress were consistently associated with increased LTSA risk (adjusted HRs between 1.3 and 1.5 for pain and between 1.6 and 1.7 for co-occurrence). In contrast, psychological distress alone showed no consistent association. Model performance was moderate and similar across approaches (C-index between 0.63 and 0.67). Key predictors included female sex, low parental education, chronic pain, poor perceived health, and indicators of early health problems. CONCLUSION: Adolescent chronic pain, particularly when co-occurring with psychological distress, is an important predictor of LTSA in early adulthood. While absolute LTSA levels may vary across cohorts, underlying risk patterns remain stable. More complex models did not outperform traditional approaches. These findings highlight the importance of early-life conditions and support early identification and intervention to reduce later work absence.
Development and external validation of an interpretable machine learning model for early prediction of stroke-associated pneumonia: a multicenter study.
Zhao M, Zhou Q, Zhang Q … +7 more , Wang L, Wang P, Qin Y, Chen J, Liu M, Wanyan T, Sun C
Int J Med Inform · 2026 Jun · PMID 42320081 · Publisher ↗
OBJECTIVE: To develop and externally validate an interpretable machine learning model for predicting 7-day stroke-associated pneumonia (SAP) using routine clinical data collected within 24 h of admission. METHODS: This m...
OBJECTIVE: To develop and externally validate an interpretable machine learning model for predicting 7-day stroke-associated pneumonia (SAP) using routine clinical data collected within 24 h of admission. METHODS: This multicenter study utilized a development cohort from the Henan Stroke Cohort and an independent external validation cohort from three Zhengzhou hospitals. Adult patients with imaging-confirmed ischemic or hemorrhagic stroke were eligible. Patients with infection at admission or who developed SAP within 24 h of admission were excluded. We evaluated 26 candidate predictors obtained within 24 h of admission. Nine machine learning algorithms were trained following recursive feature elimination. Model performance was evaluated based on discrimination, calibration, and clinical utility. The optimal model was interpreted using SHapley Additive exPlanations (SHAP) and deployed as an online calculator. RESULTS: The development and external validation cohorts comprised 1201 and 645 patients, with 7-day SAP incidences of 20.6 % (n = 247) and 24.8 % (n = 160), respectively. Among the nine algorithms, stochastic gradient boosting (SGBT) demonstrated the most balanced overall performance. In internal validation, SGBT achieved an area under the receiver operating characteristic curve (AUC) of 0.947 in the training set and 0.905 in the test set. In external validation, the model yielded an AUC of 0.906, alongside an accuracy of 0.864, sensitivity of 0.712, specificity of 0.918, positive predictive value (PPV) of 0.756, negative predictive value (NPV) of 0.899, and F1-score of 0.733. The final model retained 10 predictors: stroke subtype, fibrinogen, D-dimer, C-reactive protein, uric acid, triglycerides, homocysteine, and clinical scores (ADL, GCS, NIHSS). SHAP analysis identified early neurological impairment and inflammatory burden as the primary contributors to SAP prediction. CONCLUSION: An interpretable SGBT model utilizing routine admission data accurately predicted 7-day SAP and remained robust during external validation. The accompanying online calculator facilitates individualized risk estimation to support early preventive decision-making in hospitalized patients with stroke.
From development to clinical practice: deployment of an interoperable and secure ML-based CDSS to aid in the early detection of sepsis.
Serrano García A, López D, Macias-Fassio E … +7 more , Salas-Sosa S, Pascual I, Pruenza C, Borges-Sa M, Giglio A, Cruz-Rojo J, Pacheco-Puig R
Int J Med Inform · 2026 Jun · PMID 42308952 · Publisher ↗
Recent research has increasingly focused on machine learning (ML) models for early disease prediction, yet practical frameworks for integrating these models into clinical workflows remain limited. BIAlert is a microservi...
Recent research has increasingly focused on machine learning (ML) models for early disease prediction, yet practical frameworks for integrating these models into clinical workflows remain limited. BIAlert is a microservices-based framework designed to operate as a real-time early-warning system for ML-driven disease prediction in hospitalised patients. It can be deployed remotely on physical or virtual servers and is composed of coupled microservices that communicate through Apache Kafka queues, using HL7 FHIR resources as the message format. The system comprises four core components: (1) the Connector, which ingests raw hospital data and converts it into standardised healthcare formats; (2) the Writer, which stores FHIR-formatted data in an internal database and triggers the prediction pipeline; (3) the Predictor, which hosts ML models and generates patient-specific alerts; and (4) the Model Evaluator, which supports prospective monitoring of model performance. Alerts are displayed through the BIAlert user interface and can also be integrated directly into the electronic health record (EHR). BIAlert is currently deployed and operating in real-time clinical settings in two hospitals, demonstrating its feasibility as a scalable and interoperable solution for ML-based clinical decision support.
Why almost all ML models for medicine are wrong-and what we need for evidence-based medical AI.
Cabitza F, Jurman G, Molinari F … +1 more , Bellazzi R
Int J Med Inform · 2026 Jun · PMID 42308951 · Publisher ↗
Machine learning (ML) models are increasingly proposed to support clinical decision-making, yet their evidentiary basis remains weaker than their publication volume suggests. This editorial argues that the problem is not...
Machine learning (ML) models are increasingly proposed to support clinical decision-making, yet their evidentiary basis remains weaker than their publication volume suggests. This editorial argues that the problem is not only translational, regulatory, or infrastructural, but methodological. Many medical ML pipelines rely on uncertain ground truths, optimize performance around clinically irrelevant thresholds, report unstable or prevalence-dependent metrics, neglect calibration and uncertainty, and lack rigorous external and temporal validation. These weaknesses produce optimistic estimates that do not reliably anticipate performance in heterogeneous clinical settings. We call for an evidence-based medical AI grounded in more reliable annotation practices, explicit modeling of uncertainty, clinically meaningful threshold selection, calibration and decision-utility analyses, robustness testing, external validation on independent datasets, and post-deployment monitoring. The editorial also invites authors, reviewers, users, and vendors to adopt stricter standards so that predictive models can become credible, accountable, and clinically useful tools in everyday practice, rather than merely publishable artifacts.
Technological solutions for multiple sclerosis: a scoping review.
Lima B, Guimaraes J, Fernandes CS … +1 more , Ferreira MC
Int J Med Inform · 2026 Jun · PMID 42308950 · Publisher ↗
BACKGROUND: Multiple sclerosis (MS) is a chronic neurodegenerative disease of the central nervous system (CNS) that affects nearly 3 million people worldwide. It can lead to cognitive impairment, physical disability, and...
BACKGROUND: Multiple sclerosis (MS) is a chronic neurodegenerative disease of the central nervous system (CNS) that affects nearly 3 million people worldwide. It can lead to cognitive impairment, physical disability, and a reduced quality of life. Technological innovations have demonstrated significant potential in supporting individuals living with chronic conditions, including MS. PURPOSE: This study aims to synthesize existing evidence on technological solutions designed to support people with MS across various aspects of disease management and daily living. METHODS: A literature search was conducted in PubMed, Scopus, Web of Science, and the Cumulative Index of Nursing and Allied Health Literature (CINAHL), following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. RESULTS: Forty-nine studies were included. These studies investigated a wide range of technologies, such as mobile applications, websites, video games, wearables, and virtual reality, used to support individuals with MS in several domains, including fatigue, cognition, mental health, motor function, physical activity, medication and treatment adherence, and communication and decision-making. CONCLUSION: This study highlights the growing role of technological solutions in supporting assessment, self- management, health literacy, and telerehabilitation for people with MS. Continued research is essential to enhance the development, adoption, and long-term effectiveness of these technologies in promoting sustainable self-management of the condition.
Patient safety concerns in the go-live phase of an electronic health record implementation.
Jämsä J, Tissari P, Kovanen J … +1 more , Lehtonen L
Int J Med Inform · 2026 Sep · PMID 42296876 · Publisher ↗
BACKGROUND: Transitions to new electronic health record systems can disrupt workflows and patient safety. At Helsinki University Hospital, the Epic Radiant imaging module went live in April 2021 as part of the Apotti ele...
BACKGROUND: Transitions to new electronic health record systems can disrupt workflows and patient safety. At Helsinki University Hospital, the Epic Radiant imaging module went live in April 2021 as part of the Apotti electronic health record rollout. OBJECTIVE: To characterise patient safety incidents associated with the Radiant go-live and identify contributing factors for safer future implementations. METHODS: We analysed three sources: radiology-targeted patient safety incident reports from six months before to six months after go-live; a one-time free-text screen of reports made by radiology units one month after go-live, used to identify Radiant-related reports targeted outside radiology; and root cause analysis minutes for serious incidents in IT Management during the six months after go-live. Incidents were classified by type, risk level, seriousness, and Radiant-relatedness. RESULTS: Radiology-targeted reports increased from 161 before to 296 after go-live. Of 284 non-serious post-go-live reports, 100 were Radiant related. The most common categories were information flow and data management and imaging processes. Delayed or missing examinations or reports predominated, usually at risk levels II-III with no, low, or unspecified recorded harm. The free-text screen identified 72 additional Radiant-related reports targeted outside radiology. Serious radiology-targeted incidents increased from 1 before to 12 after go-live; 11 were Radiant related and mainly involved prolonged MRI or CT reporting delays affecting high-risk patients. Root cause analyses from IT management highlighted increased workload, changed work procedures, insufficient end-to-end testing, and interoperability failures. CONCLUSIONS: Radiant implementation coincided with a cluster of incidents, notably serious delays in imaging reporting, linked to integration gaps and workflow strain. End-to-end interface testing, code-set harmonisation, clear surge procedures and active post-go-live monitoring may mitigate risks.
Embedding LLMs in the patient portal to summarize acute minor illness information: a three-arm experimental study.
Esmaeilzadeh P
Int J Med Inform · 2026 Jun · PMID 42296704 · Publisher ↗
BACKGROUND: Patient portals provide direct access to clinical encounter data; however, comprehension barriers, especially for patients with limited health literacy, frequently prevent meaningful use, even for common acut...
BACKGROUND: Patient portals provide direct access to clinical encounter data; however, comprehension barriers, especially for patients with limited health literacy, frequently prevent meaningful use, even for common acute conditions. Large language models (LLMs) embedded in the portal interface may convert clinical documentation into accessible plain-language summaries, yet no experimental study has evaluated this approach for acute minor illness. OBJECTIVE: This study evaluated the impact of embedding two commercially available frontier LLMs (Claude Sonnet 4.5 [Anthropic] and GPT-5.1 Thinking [OpenAI]) into the patient portal of a small primary care clinic on patients' comprehension of information about acute minor illnesses, portal engagement, self-management adherence, and unnecessary healthcare utilization. METHODS: We conducted a three-arm experimental study at a five-physician primary care practice. Adults (≥18 years) presenting with acute upper respiratory infection, influenza-like illness, or similar acute minor ailment were assigned 1:1:1 to: (A) Claude Sonnet 4.5 portal summaries, (B) GPT-5.1 Thinking portal summaries, or (C) standard portal access (control). The primary outcome was health information comprehension at Day 14, assessed via a study-specific, pilot-tested 10-item quiz (0-100 scale). A clinician audit of 20% of LLM summaries assessed accuracy and safety. RESULTS: Of 186 enrolled participants, 174 completed the 4-week follow-up (93.5% retention). At Day 14, comprehension scores were significantly higher in LLM arms versus control (Claude: M = 81.4, SD = 9.3; GPT-5.1: M = 79.6, SD = 10.1; control: M = 63.2, SD = 12.8; F(2,169) = 47.34, p < 0.001, partial η = 0.36). Both LLM arms showed significantly higher total portal login frequency (Claude: p = 0.001; GPT: p = 0.007), longer session durations, and greater self-management adherence (p = 0.007) compared with the control. Unnecessary return visits were approximately 61% lower in combined LLM arms (OR = 0.34, 95% CI [0.14-0.81], p = 0.015). Minor factual imprecision rates in audited summaries were 5.7% (Claude) and 11.8% (GPT-5.1), with zero clinically significant hallucinations across all audited summaries. No significant difference was observed between the two LLMs. CONCLUSION: LLM-embedded patient portal summaries significantly improved comprehension of acute minor illnesses and reduced unnecessary healthcare utilization in a primary care setting. Because the intervention combined an LLM-generated plain-language summary with its prominent presentation in a dedicated portal panel, the observed effects should be interpreted as the result of this combined plain-language-plus-interface intervention rather than as a test of LLM capability alone. Both models performed comparably and safely, supporting the feasibility of further evaluation in diverse clinical settings. Further multi-site replication and implementation science studies are needed.
Functional near-infrared spectroscopy-based machine learning techniques for autism spectrum disorder diagnosis: a systematic review and meta-analysis.
Li X, Yan F, Wu Y … +3 more , Zhou Y, He L, Gao J
Int J Med Inform · 2026 Jun · PMID 42296703 · Publisher ↗
BACKGROUND: Diagnosis of autism spectrum disorder (ASD) primarily relies on subjective behavioral assessments, posing risks of delayed and missed diagnoses. The combination of functional near-infrared spectroscopy (fNIRS...
BACKGROUND: Diagnosis of autism spectrum disorder (ASD) primarily relies on subjective behavioral assessments, posing risks of delayed and missed diagnoses. The combination of functional near-infrared spectroscopy (fNIRS) technology and machine learning (ML) offers a novel approach for objective ASD identification; however, its overall diagnostic efficacy and key influencing factors require systematic evaluation. METHODS: Following the PRISMA guidelines, we systematically searched PubMed, EMBASE, Web of Science, The Cochrane Library, and Wiley Online Library databases to comprehensively collect English-language literature on fNIRS-based machine learning techniques for ASD diagnosis published from the inception of our database to December 2025. Data were extracted to calculate pooled sensitivity (sen), specificity (spe), positive likelihood ratio (LR + ), negative likelihood ratio (LR-), diagnostic odds ratio (DOR), and their 95% confidence intervals (CI). Pooled receiver operating characteristic (ROC) curves were plotted, and area under the curve (AUC) was calculated to assess diagnostic value. The I test assessed study heterogeneity, which was explored through meta-regression and subgroup analyses. Publication bias was evaluated using Deeks funnel plot asymmetry tests. The review protocol was registered in PROSPERO (CRD420251250866). RESULTS: A total of 17 studies were included in the systematic review, with 15 studies included in the meta-analysis. The pooled analysis showed that the sensitivity of fNIRS-based ML techniques for diagnosing ASD was 0.92 (95% CI 0.87-0.95),with a specificity of 0.94 (95% CI 0.90-0.97) and a composite area under the receiver operating characteristic curve (AUC) of 0.98 (95% CI 0.96-0.99). Based on these findings, fNIRS-based ML techniques demonstrate good diagnostic value for ASD. CONCLUSION: fNIRS-based ML techniques demonstrate excellent diagnostic accuracy for ASD, presenting a highly promising objective diagnostic tool. Future research should focus on establishing standardized, multicenter datasets and analytical workflows, while exploring multimodal data fusion and model interpretability to advance this technology toward stable, reliable clinical diagnostic support.
Organizational factors influencing electronic health record adoption in Philippine public hospitals: A systematic review.
Villarino RT, Zheng L
Int J Med Inform · 2026 Sep · PMID 42287959 · Publisher ↗
CONTEXT: Electronic Health Record (EHR) systems are vital in realizing universal health coverage in Low- and Middle-Income Countries. The Philippines' Universal Health Care Act mandates nationwide EHR adoption; however,...
CONTEXT: Electronic Health Record (EHR) systems are vital in realizing universal health coverage in Low- and Middle-Income Countries. The Philippines' Universal Health Care Act mandates nationwide EHR adoption; however, implementation remains inconsistent across the devolved health system. OBJECTIVE: To systematically review the role of organizations in implementing EHRs in Philippine public hospitals, using the Consolidated Framework for Implementation Research (CFIR). METHODS: A systematic review protocol was developed according to the PRISMA 2020 guidelines and was registered in PROSPERO (CRD420251247527). Eight databases were searched: PubMed, Scopus, Web of Science, CINAHL, IEEE Xplore, ACM Digital Library, Google Scholar, and HERDIN from January 2010 to November 2025. Quality was assessed using Joanna Briggs Institute checklists; evidence certainty was evaluated using GRADE and GRADE-CERQual frameworks. RESULTS: A total of seven studies (2016-2024) qualified for inclusion. The quality assessment of the included studies showed that 57 % (4/7) had high quality (≥80 %), 43 % (3/7) had moderate quality (60-79 %), and the average quality was 80 %. The CFIR-informed thematic synthesis identified twenty-three contextual organizational variables underpinning five domains. Self-efficacy emerged as the sole significant predictor of EHR acceptance, with acceptance rates of 96 %-98 % across urban, rural, and remote areas. The most thoroughly investigated domains among inner-setting factors were leadership engagement (100 %), implementation climate (86 %), organizational readiness (86 %), and infrastructure constraints (100 %). Notably, the domains of infrastructure (100 %), information technology support (100 %), training (86 %), and software-workflow fit, including system complexity (86 %), emerged as critical sociotechnical challenges rather than user-side deficits. Key facilitators included leadership (86 %), high-quality training support (86 %), user involvement (86 %), and phased implementation (57 %). CONCLUSION: Organizational change management factors and inner setting constructs have been identified as the critical determinants of EHR adoption success in resource-limited hospitals. The findings also challenge technology-deterministic approaches and emphasize leadership commitment, readiness assessment, authentic user engagement, and sustainability planning.
Effectiveness of telerehabilitation in managing chronic low back pain: a pragmatic randomized controlled non-inferiority trial.
Alahmri F, Alkhawajah HA, Nuhmani S … +1 more , Muaidi Q
Int J Med Inform · 2026 Sep · PMID 42284831 · Publisher ↗
BACKGROUND: For patients with chronic low back pain (CLBP), telerehabilitation (TR) has been suggested as an alternative to conventional rehabilitation. However, evidence from pragmatic trials comparing TR with routine c...
BACKGROUND: For patients with chronic low back pain (CLBP), telerehabilitation (TR) has been suggested as an alternative to conventional rehabilitation. However, evidence from pragmatic trials comparing TR with routine clinical rehabilitation is still limited. OBJECTIVES: To investigate whether TR is non-inferior to conventional in-person rehabilitation for pain intensity in individuals with CLBP. Secondary objectives included disability, pain catastrophizing, quality of life, and kinesiophobia. METHODS: A pragmatic non-inferiority randomized controlled trial was conducted with 50 adults with CLBP who were randomly assigned to TR or in-person rehabilitation. The TR program included individualized exercise, education, and self-management delivered remotely, while the control group received usual clinic-based care. The primary outcome was pain intensity assessed using the Numeric Rating Scale (NRS), with disability and other patient-reported outcomes assessed as secondary measures. Between-group comparisons were performed using analysis of covariance, with adjustments made for baseline values, and non-inferiority was evaluated using a pre-specified margin. RESULTS: Both groups showed comparable improvements across outcomes, with no meaningful between-group differences. Non-inferiority was not demonstrated for pain intensity within the pre-specified margin. However, TR demonstrated non-inferiority to conventional rehabilitation for disability measured by the Oswestry Disability Index (adjusted mean difference -0.82 points; 95% CI -5.13 to 3.49). For other outcomes, no clear between-group differences were observed. CONCLUSIONS: These findings indicate that telerehabilitation can be considered a clinically acceptable alternative option to in-person rehabilitation for individuals with CLBP, particularly when the primary goal is functional improvement. Careful patient selection may be required when pain reduction is the main treatment objective.
Predicting recurrence within 5 years in Early-Stage lung adenocarcinoma with micropapillary and solid patterns.
Wang Z, Chen J, Xu Y … +6 more , Lin T, Chen C, Shen Y, Huang J, Xu S, Chen S
Int J Med Inform · 2026 Sep · PMID 42284830 · Publisher ↗
BACKGROUND: Micropapillary (MP) and solid (S) pathological patterns are recognized as high-risk patterns. Even early-stage invasive adenocarcinoma (IAC) with MP/S pathological patterns carry a high risk of postoperative...
BACKGROUND: Micropapillary (MP) and solid (S) pathological patterns are recognized as high-risk patterns. Even early-stage invasive adenocarcinoma (IAC) with MP/S pathological patterns carry a high risk of postoperative recurrence. This study aimed to construct and validate machine learning (ML) models to predict recurrence within 5 years in patients with IAC with MP/S pathological components and tumor diameter ≤ 3 cm without lymph node metastasis and distant metastasis. METHODS: We retrospectively analyzed 974 patients from two centers. Univariate and multivariate logistic regression analyses were used to select independent predictive variables. Eight ML models were evaluated based on the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). The contribution of variables was interpreted using SHapley Additive exPlanations (SHAP) method. RESULTS: Univariate and multivariate analyses identified five independent predictors for the model: surgical procedure, consolidation-to-tumor ratio (CTR), visceral pleural invasion, epidermal growth factor receptor mutation status, and smoking history. The Neural Network model demonstrated the best performance, with an AUC of 0.794 in the training set, 0.764 in the test set and 0.775 in the validation set. Calibration curves showed good predictive accuracy, and DCA indicated substantial clinical utility across most risk thresholds. SHAP analysis identified CTR and surgical procedure as the most important predictive factors, highlighting their significant roles in postoperative recurrence. CONCLUSION: We developed and validated a machine learning model integrating multidimensional clinicopathological features and imaging features to assess the 5-year recurrence risk in pN0M0 invasive adenocarcinoma (≤3 cm) patients with MP/S pathological components.
European cancer data sharing: Analyses of an international survey.
Hogstrom L, Laurinavicius A, Alvarez MG … +10 more , Nakken S, Guldvik IJ, Sanchez CM, Papadodima O, Voutetakis K, Mæhle PM, Zerbe N, Šekerija M, Fullaondo A, Hovig E
Int J Med Inform · 2026 Sep · PMID 42284829 · Publisher ↗
BACKGROUND: Clinical and biomarker data sharing across institutions accelerates cancer research, yet European data sharing practices are not well characterized. METHODS: We surveyed 75 cancer institutions across 23 Europ...
BACKGROUND: Clinical and biomarker data sharing across institutions accelerates cancer research, yet European data sharing practices are not well characterized. METHODS: We surveyed 75 cancer institutions across 23 European countries on data sharing practices, data standards adoption, and perceived challenges. RESULTS: Most institutions (>80 %) actively share data for research. However, adoption of clinical data standards remains fragmented: only 33 % use Fast Healthcare Interoperability Resources (FHIR) and 21 % use the Observational Medical Outcomes Partnership (OMOP) data model. The top non-technical hurdles were General Data Protection Regulation (GDPR) compliance (31 %) and legal challenges (24 %). The leading technical hurdle was interoperability (35 %). Despite challenges, 77 % of institutions aim to expand sharing capabilities. CONCLUSIONS: Among those surveyed, GDPR compliance and legal complexity are the dominant barriers to European cancer data sharing. Inconsistent data standards adoption poses risks to European Health Data Space (EHDS) implementation. Investments in legal support, interoperability, and personnel are needed to advance cancer research networking.
Multi-site health research integrating complementary data sources: A scoping review of statistical inference methods for vertically partitioned data.
Domingue MP, Lévesque S, Burgun A … +2 more , Ethier JF, Camirand Lemyre F
Int J Med Inform · 2026 Sep · PMID 42275911 · Publisher ↗
BACKGROUND AND OBJECTIVES: To address the multidimensional nature of health-related questions, advances in health research often require integrating information from various data sources within statistical analyses. When...
BACKGROUND AND OBJECTIVES: To address the multidimensional nature of health-related questions, advances in health research often require integrating information from various data sources within statistical analyses. When complementary information pertaining to the same set of individuals are distributed across different institutions, vertical methods make it possible to obtain analysis results without sharing or pooling individual-level data. To guide stakeholders toward a transparent and rigorous use of vertical methods with sensitive health data, this study aims to (1) Identify existing vertical methods enabling statistical inference (confidence interval estimation and hypothesis testing); and (2) Characterize the methodological properties of these methods and the current extent of their use with health data. METHODS: We conducted a scoping review following PRISMA-ScR using four interdisciplinary databases. We then systematically extracted the characteristics of identified vertical methods with respect to comparability with the pooled analysis, efficiency of communication schemes and confidentiality. We additionally screened studies that cited included articles to identify applications on vertically partitioned real-world health data. RESULTS: Among 2887 articles initially screened, 30 were included in the review, of which a majority mentioned health analytics. Inference for the linear and the logistic regression framework were the most frequent statistical inference tasks undertaken in proposed methods. Equivalence with the pooled analyses was not systematically addressed and most methods required multiple communications between participating parties. Almost all articles described their approach as privacy-preserving, although a minority provided privacy assessments. Very few published health studies were found to report the use these methods. CONCLUSION: The scope of existing approaches enabling statistical inference for vertically partitioned data is still relatively limited. Most existing methods do not concurrently achieve results equivalent to centralized analyses, high communication efficiency, and guaranteed protection of individual-level data.
Machine learning-based prediction of early invasive mechanical ventilation in ICU patients with pneumonia: Development and external validation.
Lu Y, Chen H, Li M … +2 more , Liang H, Lv H
Int J Med Inform · 2026 Sep · PMID 42275910 · Publisher ↗
BACKGROUND: Pneumonia is a common critical illness in the intensive care unit (ICU), and a subset of patients rapidly progresses to respiratory failure requiring invasive mechanical ventilation (IMV). Current decisions o...
BACKGROUND: Pneumonia is a common critical illness in the intensive care unit (ICU), and a subset of patients rapidly progresses to respiratory failure requiring invasive mechanical ventilation (IMV). Current decisions often rely on fragmented clinical indicators, highlighting the need for integrated, data-driven tools for early risk assessment. METHODS: We conducted a retrospective cohort study using the Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) database as the development cohort. The cohort was randomly split into training (80%) and test (20%) sets. Candidate predictors at ICU admission were selected through multi-step feature selection. Eight machine learning models were trained with 5-fold cross-validation and hyperparameter tuning. Model performance was evaluated in terms of discrimination, calibration, and clinical utility. The best-performing model was externally validated in an independent cohort from Maoming People's Hospital. A web-based calculator was further developed to facilitate potential clinical application and individualized risk assessment. RESULTS: The development cohort included 5,608 patients, of whom 856 (15.3%) required IMV within 24 h. The final model retained seven predictors: age, oxygen flow, FiO, pH, PaO, PaCO, and platelet count. LightGBM showed the best performance in the internal test set (AUC = 0.799) and achieved an AUC of 0.702 in the external validation cohort (n = 155). Calibration showed acceptable agreement, and decision curve analysis demonstrated net clinical benefit. Risk stratification further enabled identification of clinically distinct patient groups with progressively increasing IMV incidence. SHAP analysis further enhanced model interpretability by identifying key predictors associated with IMV risk. CONCLUSIONS: This study developed and externally validated a machine learning model and online calculator for predicting early IMV in ICU pneumonia patients. The model provides an interpretable, data-driven approach for early risk stratification and may serve as an adjunctive tool to assist clinical decision-making. Multicenter prospective studies are warranted to confirm clinical utility.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe