Frontiers in digital health最新文献

{"title":"Machine learning based multimodel prediction of poor wellbeing, generalized anxiety disorder, and insomnia among adults in Indian urban informal settlements: toward explainable digital mental health screening.","authors":"Rupeshkumar Deshmukh, Sudhanshu Mahajan, Reva Wani","doi":"10.3389/fdgth.2026.1935205","DOIUrl":"10.3389/fdgth.2026.1935205","url":null,"abstract":"<p><strong>Introduction: </strong>Mental health morbidity among urban informal settlements-dwelling populations is shaped by a complex interplay of behavioural exposures including screen time, working hours, and sleep duration and sociodemographic vulnerability.</p><p><strong>Methods: </strong>We analyzed cross-sectional data from 1,600 adults recruited from 16 clusters within the urban informal settlement field practice area of a medical college's Urban Health Training Centre in Pune, India, to develop and compare ten supervised machine learning algorithms predicting poor wellbeing (WHO-5), moderate-to-severe generalized anxiety disorder (GAD-7), and moderate-to-severe insomnia (ISI); internal consistency of all three instruments was confirmed from item-level responses. Because participants were nested within 16 clusters, our primary analysis used leave-one-cluster-out (LOCO) cross-validation, in which each cluster was held out in turn as an independent test set; a conventional participant-level 80/20 split with 10-fold cross-validation was retained as a secondary analysis. Model discrimination, calibration, and clinical utility were assessed using AUC, sensitivity, specificity, predictive values, Brier score, decision curve analysis, and DeLong's test, each with cluster-level bootstrap 95% confidence intervals. SHAP was used to interpret the best-performing model, and feature-attribution concordance was additionally assessed across the Rashomon set of near-equivalent top models.</p><p><strong>Results: </strong>The WHO-5, GAD-7, and ISI showed acceptable-to-excellent internal consistency in this sample (Cronbach's alpha 0.730-0.938; McDonald's omega 0.815-0.953). LightGBM achieved the highest primary (LOCO-CV) AUC for all three outcomes: 0.977 (95% CI 0.970-0.983) for poor wellbeing, 0.987 (95% CI 0.984-0.992) for GAD, and 0.988 (95% CI 0.982-0.993) for insomnia. Several sociodemographic predictors showed unusually strong univariate associations with the outcomes, traced to a repeating between-cluster pattern in the sampling frame; this structure means that even cluster-aware validation may remain optimistic relative to a genuinely external population, a limitation we discuss in detail. SHAP-based rankings were highly concordant (Spearman <i>ρ</i> = 0.84-0.95) across the three top-performing, statistically near-equivalent models. A class-weighting sensitivity analysis improved positive predictive value for poor wellbeing from 0.478 to 0.600 with minimal AUC cost.</p><p><strong>Conclusion: </strong>Gradient-boosted tree ensembles, particularly LightGBM, showed strong and internally consistent discrimination for these three outcomes, assessed using reliable instruments; however, given the between-cluster homogeneity identified in this sample, external validation in an independent population is essential before any claim of generalizable clinical utility.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1935205"},"PeriodicalIF":5.2,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13661956/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186175","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TransmiRisk-AI: an agentic AI-based risk assessment framework for zoonotic diseases using integrated surveillance data.","authors":"Ramya Jeyachandran, Rama Parvathy Lakshmanan","doi":"10.3389/fdgth.2026.1915858","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1915858","url":null,"abstract":"<p><strong>Introduction: </strong>Zoonotic disease outbreaks are increasing in frequency and complexity, highlighting the need for intelligent surveillance systems capable of real-time monitoring and risk assessment. Existing surveillance approaches are often reactive, fragmented, and dependent on manual analysis, limiting timely public health responses.</p><p><strong>Methods: </strong>This study presents TransmiRisk-AI, an agentic AI-driven surveillance framework designed to support automated zoonotic disease risk monitoring and decision-making. The framework integrates a hybrid risk scoring model with a multi-agent architecture comprising data ingestion, risk assessment, alert generation, execution, logging, and orchestration agents. These agents collaboratively automate data collection, preprocessing, risk evaluation, outbreak monitoring, and alert dissemination. To enhance situational awareness, the system incorporates outbreak intelligence from WHO Disease Outbreak News and dynamically updates disease risk profiles based on outbreak recency, frequency, and severity. An interactive Streamlit dashboard provides transparent visualization of risk categories, outbreak trends, and system activities.</p><p><strong>Results: </strong>Case studies involving the Nipah virus and Crimean-Congo hemorrhagic fever demonstrate the framework's ability to adapt to evolving epidemiological conditions.</p><p><strong>Discussion: </strong>TransmiRisk-AI offers a scalable and interpretable digital health solution that supports proactive surveillance and timely decision-making within a One Health context.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1915858"},"PeriodicalIF":5.2,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13661919/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149198360","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI-based clinical decision support systems for antibiotic therapy in common adult infections: a systematic review from development to clinical implementation.","authors":"Miguel Hueda-Zavaleta, Rodrigo Flores-Palacios, Fabrizzio Bardales-Silva, Eric Cotacallapa-Pino, Ari Frenkel, Diana Minchon-Vizconde, Juan Carlos Gomez de la Torre Pretell","doi":"10.3389/fdgth.2026.1944745","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1944745","url":null,"abstract":"<p><strong>Background: </strong>Antimicrobial resistance is among the leading causes of global mortality, and artificial intelligence (AI)-based clinical decision support systems (CDSS) have been proposed to guide antibiotic prescribing. Several AI models for antibiotic decision-making have been developed and validated, but their actual deployment in clinical practice and effect on patient outcomes have not been systematically characterized.</p><p><strong>Objective: </strong>To identify, classify, and appraise studies evaluating AI-based CDSS for antibiotic therapy in hospitalized adults with common infections, stratified by level of clinical implementation.</p><p><strong>Methods: </strong>Search-strategy development began on June 7, 2026; definitive searches of PubMed/MEDLINE, Scopus, Web of Science Core Collection, and Embase were executed on June 28, 2026, after submission of the PROSPERO record. Four reviewers independently screened records and assessed reports. Studies were classified as deployed in clinical practice (Level 1), evaluated against physicians' decisions without deployment (Level 2), or tested through retrospective simulation (Level 3). Risk of bias was assessed with RoB 2, ROBINS-I, or PROBAST, as appropriate. Owing to clinical and methodological heterogeneity, findings were synthesized narratively, with an outcome-specific GRADE assessment.</p><p><strong>Results: </strong>Of 1,157 records identified, 10 studies met the inclusion criteria: three Level 1 (30%), four Level 2 (40%), and three Level 3 (30%). The three deployed studies evaluated pneumonia, predominantly pulmonary <i>Stenotrophomonas maltophilia</i> infection, and <i>Staphylococcus aureus</i> bacteremia and reported favorable signals for mortality, hospital processes, prescribing, or medication safety. Kanjilal et al., a published conference abstract, was included as a Level 3 retrospective evaluation of 29,508 adults with community-onset sepsis. Among the seven Level 2/3 studies assessed with PROBAST, three were classified as low, three as moderate under the prespecified descriptive adaptation, and one as high overall risk of bias. Certainty was low for mortality/survival and very low for the remaining clinically relevant outcomes.</p><p><strong>Conclusions: </strong>Three of 10 included studies reached clinical deployment, whereas seven remained non-deployed. Deployed systems showed favorable outcome and process signals, but the evidence was heterogeneous and of low to very-low certainty, depending on the outcome. Prospective multicenter evaluations of deployed systems are required before computational performance can be interpreted as clinical effectiveness.</p><p><strong>Systematic review registration: </strong>https://www.crd.york.ac.uk/PROSPERO/view/CRD420261435593, identifier CRD420261435593.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1944745"},"PeriodicalIF":5.2,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13662083/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149198317","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"What and when: a Delphi study identifying AI competencies and learning objectives for medical students.","authors":"Jennifer Simoni, José Luis Pereira, Sofía Crespo Sánchez, Diego Egaña-Yin, Carlota García-Gargallo, Oliver L A Hertog, Andrés Idoate, Lourdes López-Ortiz, Ian Morales-Gutiérrez, Alexandra Romero, Paula Santiago-Martínez, Gabriela Sawczyn, Adrián Vahamaki, Judith Urtubia-Fernandez, Elisa Mengual","doi":"10.3389/fdgth.2026.1918847","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1918847","url":null,"abstract":"<p><strong>Introduction: </strong>As Artificial Intelligence (AI) becomes embedded in clinical practice, medical graduates must be prepared to use it competently and critically. AI competencies and learning objectives (LOs) for medical students are needed. This study identifies these competencies and LOs for undergraduate medical education (UGME), particularly following the public release of large language models in 2022.</p><p><strong>Methods: </strong>A four-round modified Delphi process was conducted between May 2024 and January 2025. The panel included medical students (<i>n</i> = 14) and medical professors (<i>n</i> = 2) in the first three rounds, and international experts in medical education and AI in the final round (<i>n</i> = 10). In Round 1, using an open-ended question, 77 potential LOs were identified. In Rounds 2 and 3, these items were rated using a 7-point Likert scale (1 = \"Low Priority\", 7 = \"High Priority\"), and participants were provided an area for comments. Median scores and interquartile ranges were used to assess the strength of agreement, and the LOs were modified accordingly. In Round 4, validation input was sought from 10 international experts in medical education and AI. Learning objectives were organized into categories representing broader competencies.</p><p><strong>Results: </strong>The process identified 7 competencies and 37 supporting LOs essential for medical graduates. Timelines for incorporating these across UGME were also identified as guidelines.</p><p><strong>Discussion: </strong>This study offers a framework for integrating AI into UGME, combining theoretical understanding with ethical use to prepare future physicians for clinical practice. Developed with input from students and multinational AI expert stakeholders, these competencies and LOs constitute a student-partnered, internationally informed framework that may guide curricular development in other settings.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1918847"},"PeriodicalIF":5.2,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13662149/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149198300","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Radiomics-Based prediction of renal prognosis in ANCA-associated vasculitis.","authors":"Zuoxian Hou, Shuo Zhang, Peng Xia, Shilu Cao, Linkun Bai, Wenling Ye, Feng Feng, Limeng Chen","doi":"10.3389/fdgth.2026.1915391","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1915391","url":null,"abstract":"<p><strong>Objective: </strong>To develop and validate a non-invasive prediction model combining renal CT radiomics features and clinical markers for assessing long-term renal prognosis at 1, 3, and 5 years in patients with antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV).</p><p><strong>Methods: </strong>A retrospective cohort of 262 AAV patients from Peking Union Medical College Hospital (2015-2020) was analyzed. Radiomics features (morphological, first-order, and texture) were extracted from segmented renal parenchyma on abdominopelvic CT. Clinical markers included demographics, serum creatinine, complement component 3 (C3), and other laboratory indicators. Renal volume, serum creatinine, and complement component 3 (C3) were pre-specified as predictors based on <i>a priori</i> biological rationale, and XGBoost was employed for model construction. Performance was evaluated via ROC-AUC (receiver operating characteristic/area under the curve), F1 score, decision curve analysis, and calibration curves. SHAP (SHapley Additive exPlanations) analysis assessed feature contributions, and Kaplan-Meier curves with Bonferroni-corrected log-rank tests were used for survival stratification.</p><p><strong>Results: </strong>The mean follow-up was 40.8 months; 59 patients (23%) developed ESRD (56 within 3 years) and 29 (11%) died. The combined model (renal volume + sCr + C3) achieved an AUC of 0.93 for 3-year ESRD prediction (clinical-only model: 0.91), with renal volume independently associated with ESRD (Wald <i>p</i> = 0.004). For the 3-year composite outcome, the combined model achieved an AUC of 0.84 (clinical-only: 0.81) with significantly improved reclassification (IDI <i>p</i> = 0.018; category-free NRI <i>p</i> = 0.024). In pre-specified median-based Kaplan-Meier analyses, reduced renal volume (≤ 130.6 mL), elevated serum creatinine (> 177.5 μmol/L) and lower C3 (≤ 0.97 g/L) were associated with poorer 3-year ESRD-free survival (log-rank <i>p</i> ≤ 0.002).</p><p><strong>Conclusion: </strong>The integrated radiomics-clinical model provides accurate, non-invasive risk stratification for long-term renal outcomes in AAV, facilitating personalized treatment.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1915391"},"PeriodicalIF":5.2,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13662048/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149198308","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid multi-model framework with knowledge distillation for precise and efficient sickle cell detection.","authors":"Metta Manikanta Karthikeya Reddy, Cherukuri Krishna Lathvik, Arumuga Arun R, Karthik K","doi":"10.3389/fdgth.2026.1938284","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1938284","url":null,"abstract":"<p><p>Accurate and timely detection of sickle red blood cells has a significant impact on the patients' treatment. Developing a computer vision system that is both lightweight and resource-efficient is important for enabling real-time edge-based applications in resource-constrained healthcare environments. This study proposes a hybrid sickle cell detection framework that uses Bi-lightweight Knowledge Distillation (KD)-guided CNNs, combining two lightweight student models from different teacher networks with Machine Learning (ML) classifiers to identify sickle cells with accuracy and efficiency. Using depthwise separable convolution layers, lightweight models are created that drastically lower model complexity. These lightweight models are improved by extracting knowledge from top-performing teacher networks, such as MobileNetV2 and DenseNet121. Using this Knowledge Distillation learning technique, lightweight models can enhance their representational capabilities without adding to their computational complexity. To improve decision-making performance, the Bi-distilled student models are combined with conventional machine learning KNN classifiers to create a hybrid ensemble that combines the advantages of ML-based classification with dual KD-guided CNN-based feature extraction. The Sickle Cell Disease Dataset is used to train and validate the proposed system. The learned student models are then converted into TensorFlow Lite (TFLite) format using Float32 quantization to further reduce computational complexity. In combination with dual KD-guided CNN models and a KNN classifier, the proposed hybrid model achieves a high classification accuracy of 95.45% and an F1-score of 95.02% while being small at only 3.96 MB. Deploying and evaluating the framework on a resource-constrained Raspberry Pi 4 Model B demonstrate its potential for real-time, edge-based, and point-of-care applications.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1938284"},"PeriodicalIF":5.2,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13658618/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186213","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Strengthening mental health care in Somalia through artificial intelligence: opportunities, challenges, and a roadmap for implementation.","authors":"Nur Adam Mohamed","doi":"10.3389/fdgth.2026.1965858","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1965858","url":null,"abstract":"<p><p>Mental health needs in Somalia are substantial, yet access to timely, high-quality care remains constrained by workforce shortages, limited infrastructure and financing, fragmented services, weak health information systems, and persistent sociocultural barriers. Artificial intelligence (AI) may complement mental health system strengthening in this fragile and resource-constrained setting. This Perspective examines potential applications of AI across Somalia's mental health system and considers the associated implementation, ethical, governance, and equity challenges. Potential applications include AI-assisted screening, clinical decision support, telepsychiatry, digital interventions, predictive analytics, surveillance, health-system planning, and research. Responsible adoption will require locally representative data, appropriate infrastructure and governance, workforce capacity, sociocultural adaptation, equitable access, and sustainable financing. This Perspective proposes a phased roadmap progressing from infrastructure and governance to pilot testing, capacity building, implementation research, and conditional national scale-up. AI should augment rather than replace mental health professionals and remain integrated within broader health-system investment. With local adaptation, community engagement, and continuous evaluation, AI could contribute to more accessible, equitable, resilient, and human-centered mental health care in Somalia.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1965858"},"PeriodicalIF":5.2,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13658170/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186345","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Facilitators and barriers to implementing AI-assisted diagnostic systems in primary healthcare services in low-resource settings: a mixed-methods study protocol.","authors":"Junhan Mu, Can Li, Ting Kang, Chengcheng Dong, Cong Chen, Bijie Ou, Huijuan Liang, Nan Zhang","doi":"10.3389/fdgth.2026.1967645","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1967645","url":null,"abstract":"<p><strong>Introduction: </strong>Although AI-assisted diagnostic systems are increasingly deployed in primary care globally, evidence from multiple countries indicates that deployment does not necessarily ensure clinical uptake. Resource-constrained and dispersed regions face implementation challenges, yet the mechanisms remain unclear. This mixed-methods protocol employs the Consolidated Framework for Implementation Research (CFIR) to systematically identify barriers and facilitators to AI implementation in primary care in such settings.</p><p><strong>Methods and analysis: </strong>This study adopts a sequential mixed-methods design. Phase I involves semi-structured in-depth interviews with multi-level stakeholders (defined by the 7P principle: Patients, Providers, Purchasers, Payers, Public, Policymakers, and Product suppliers), including primary care clinicians, facility managers, external administrators, and vendor staff. Purposive and maximum-variation sampling will cover diverse geographic areas and usage levels (high, low, and non-use). The interview guide will be developed based on the CFIR 2.0 domains. Interviews will be audio-recorded, transcribed verbatim, thematically analysed, and mapped to CFIR 2.0 constructs. Phase I will recruit professional stakeholders and patients until information saturation is reached. Phase II comprises a cross-sectional survey of primary care patients targeting a minimum of 482 valid responses, to identify and quantify patient-level acceptance of AI-assisted diagnosis and its influencing factors, informed by patients' perspectives identified in Phase I. Phase III integrates findings from the first two phases and applies the Behaviour Change Wheel (BCW) and Implementation Mapping to formulate context-sensitive optimisation strategies.</p><p><strong>Ethics and dissemination: </strong>This study has been approved by the Medical Ethics Committee of Inner Mongolia Medical University (approval number: YKD202602099). All participants will provide written informed consent prior to data collection. Findings will be disseminated through peer-reviewed journal publications, national and international conference presentations, and policy briefs submitted to local health authorities. Beyond the pastoral regions of China, this protocol aims to generate an evidence base and a replicable methodological template (integrating CFIR 2.0, COM-B, and Implementation Mapping) that may inform the scaled implementation of AI in comparable resource-constrained and geographically dispersed health systems. The transferability of the findings to settings such as those in Central Asia, sub-Saharan Africa, and Small Island Developing States will be explicitly assessed as a research question rather than assumed.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1967645"},"PeriodicalIF":5.2,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13658501/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186716","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Operationalizing clinical artificial intelligence through secure FHIR pipelines: a zero-trust cloud reference architecture for EHR interoperability.","authors":"Femi Oke","doi":"10.3389/fdgth.2026.1953969","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1953969","url":null,"abstract":"<p><p>Fast Healthcare Interoperability Resources (FHIR) standardizes clinical exchange but does not establish that a downstream artificial intelligence (AI) use remains authorized, semantically consistent, or accountable. This Hypothesis and Theory article proposes evidence-continuous boundary control: a material data, model, or workflow transition is governed by an applicable policy decision and linked to versioned evidence of its prerequisites, execution, and disposition. The framework arose from continuing healthcare-interoperability research and was formalized through standards analysis, targeted evidence synthesis, and explicit design inference. Documented revision searches distinguish external evidence from architectural decisions. Eight requirements are organized into five functional layers: clinical sources, an interoperability edge, a zero-trust control plane, a segregated cloud data and AI plane, and governance and monitoring. Applications are consumers, not a sixth layer. An illustrative adult readmission-risk workflow connects identity, policy, source versions, features, model release, scoring, presentation, and human action. Its symbolic FHIR R4 evidence linkage includes explicit R5 translation cautions and policy-aware authorization, including consent where applicable. Six propositions test incremental value against a well-run contemporary platform without removing its native safeguards; simulated ablations isolate particular links. Independent reference standards, denominators, support, counterevidence, and inconclusive conditions expose operational and human-factors trade-offs. A bounded synthetic-data companion implementation makes selected exchange controls inspectable; it does not implement the complete clinical-AI architecture or validate its propositions. No empirical architecture evaluation, patient dataset, clinical effectiveness, or independent expert validation is reported. Staged future studies must establish incremental benefit without disproportionate burden or inappropriate reliance.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1953969"},"PeriodicalIF":5.2,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13658595/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186332","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A modular risk management framework for integrating cybersecurity and patient safety in medical device software development.","authors":"Pekka Pohjosaho, Tommi Mikkonen, Kyriakos Stefanidis, Vlad Stirbu, Tuomas Granlund","doi":"10.3389/fdgth.2026.1903787","DOIUrl":"https://doi.org/10.3389/fdgth.2026.1903787","url":null,"abstract":"<p><p>Medical device software operates in increasingly connected environments where cybersecurity incidents can directly compromise patient safety. While established safety risk management standards such as ISO 14971 provide mature frameworks for managing unintended hazards, they implicitly assume non-malicious causes and that risks are amenable to probability estimation. Cybersecurity risk management, by contrast, addresses adversarial threats, exploitability, and inherent uncertainty, requiring fundamentally different vocabularies, risk models, and acceptance criteria. In practice, organisations often force these domains into a single monolithic risk structure, producing terminological confusion, incommensurable risk values, and siloed assessments with no mechanism to escalate findings between safety and security processes. This paper presents the Modular Risk Schema Framework, a concrete, repeatable approach to integrating safety and cybersecurity risk management in medical device software development. Rather than forcing divergent domains into a unified risk model, the framework supports multiple domain-specific risk schemas in parallel, each with its own information model, risk model, acceptance criteria, and level of rigor, connected through explicit traceability interfaces. These interfaces ensure systematic escalation of safety-relevant cybersecurity risks and identification of safety controls with security implications. The framework is embedded within a single quality management system, enforces security by design through phase-gated deliverables aligned with the software lifecycle, and is designed for implementation in modern application lifecycle management environments without dependence on specific tools. The approach has been developed and applied in industrial contexts and aligns with the requirements of ISO 14971, IEC 62304, IEC 81001-5-1, and current regulatory guidance.</p>","PeriodicalId":73078,"journal":{"name":"Frontiers in digital health","volume":"8 ","pages":"1903787"},"PeriodicalIF":5.2,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13658581/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}