BMJ digital health & AI最新文献

{"title":"Ambient voice technology, regulatory clarity and the importance of clinical governance.","authors":"Ari Ercole, Jessica D Workum","doi":"10.1136/bmjdh-2026-000328","DOIUrl":"10.1136/bmjdh-2026-000328","url":null,"abstract":"","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000328"},"PeriodicalIF":0.0,"publicationDate":"2026-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646919/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149140393","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":"When are large language models medical devices? Navigating the MDR and AI Act for healthcare applications.","authors":"Jessica D Workum, Sade J Krijgsman, Ildiko Vajda, R M Van Stokkum, Jan van de Brand, Diederik Gommers, Ari Ercole, Michel van Genderen","doi":"10.1136/bmjdh-2026-000003","DOIUrl":"10.1136/bmjdh-2026-000003","url":null,"abstract":"<p><p>The flexibility of generative artificial intelligence (AI), and specifically large language models (LLMs), holds promise to transform healthcare by offering new tools to improve patient care and enhance clinician efficiency. However, this flexibility introduces new challenges, as LLM applications might inadvertently be used in ways that do not align with the intended purpose of the manufacturer, raising patient safety concerns and liability issues. We have identified a regulatory gap whereby LLM applications that could be repurposed, malfunction or be inadvertently misused for medical purposes fall outside the scope of the Medical Device Regulation (MDR) because they lack a medical intended purpose, and outside meaningful AI Act safety obligations because they do not qualify as high-risk systems, yet still present patient safety risks without adequate regulatory guidance. In this manuscript, we aim to address this regulatory gap for LLM applications in healthcare by identifying these ambiguous grey-area applications and providing four recommendations-validation, mitigation, compliance and alignment-focused on prevention and risk mitigation rather than reclassification as medical devices. These recommendations are intended to advance the responsible implementation of LLM applications in healthcare while ensuring patient safety.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000003"},"PeriodicalIF":0.0,"publicationDate":"2026-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646968/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149140376","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":"Public perspectives on artificial intelligence in healthcare in the USA: a cross-sectional survey.","authors":"Vishnu Prakas Nair, Rebecca Handler, Megha D Tandel, Yingjie Weng, Laura M Holdsworth, Ethan Goh, Roger W Sands, Natasha Zr Steele, Laura T Derry, Robert D Boutin, Matthew DeBell, Tamara Scott, David Svec, Neera Ahuja, Fatima Rodriguez, Roxana Daneshjou, Akshay Chaudhari, Sigall K Bell, Jonathan H Chen, Andre Kumar, Adam Rodman, Jason Hom","doi":"10.1136/bmjdh-2026-000183","DOIUrl":"10.1136/bmjdh-2026-000183","url":null,"abstract":"<p><strong>Objective: </strong>To evaluate public perspectives on artificial intelligence (AI) in healthcare in the USA and assess variation in trust, comfort and perceived fairness across demographic groups.</p><p><strong>Methods and analysis: </strong>This cross-sectional survey of US adults was conducted between 21 November and 4 December 2025, via YouGov, a private public opinion research firm, using a non-probability-based online panel. The sample included 500 general population respondents and an oversample of 750 adults aged 60+. Survey weights were applied to approximate the US adult population. Responses were measured using Likert-type or categorical scales and analysed using survey-weighted statistical tests with false discovery rate (FDR) correction for multiple comparisons. Primary outcomes included comfort with healthcare providers using AI, attitudes towards AI-generated second opinions, prior use of AI for healthcare, relative trust in AI versus clinicians for health information, preference for AI chatbots versus providers in explaining care plans and perceived effects of AI on healthcare fairness.</p><p><strong>Results: </strong>Among weighted respondents (n=500), 34.0% reported prior healthcare AI use, which was more common among college graduates than those without a high school diploma (46.5% vs 22.9%). Only 25.4% reported being comfortable or very comfortable with their provider using AI, while 43.0% were uncomfortable and 31.5% were neutral. Regarding information trust, 44.5% trusted AI much less than clinicians; 88.4% reported a provider would better help them understand their care plan than a chatbot. Prior AI use was associated with more favourable attitudes towards laboratory/imaging interpretation and AI-assisted referrals. Differences by political affiliation, insurance, health status and age were not statistically significant after FDR correction.</p><p><strong>Conclusions: </strong>AI implementation appears to be outpacing public trust. Higher educational attainment was the most consistent predictor of favourable attitudes. Notably, one-third of respondents were neutral across outcomes, suggesting public attitudes are potentially modifiable. These findings underscore the need for transparent deployment and targeted trust-building strategies to support equitable AI adoption in healthcare.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000183"},"PeriodicalIF":0.0,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13646972/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149140415","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":"Generative artificial intelligence adoption and the future of the primary care profession: an online survey of UK general practitioners' experiences and opinions.","authors":"Charlotte Blease, Carolina Garcia Sanchez, Jens Gaab, Joseph E Alderman, Luke Allen, Stephanie Kumpunen","doi":"10.1136/bmjdh-2026-000253","DOIUrl":"https://doi.org/10.1136/bmjdh-2026-000253","url":null,"abstract":"<p><strong>Objective: </strong>Use of large language model-based systems in healthcare has expanded rapidly since late 2022. Building on previous national surveys, we examined current generative AI (GenAI) use among UK general practitioners (GPs) and how GPs conceptualise artificial intelligence (AI), its clinical role and implications for professional practice.</p><p><strong>Methods and analysis: </strong>We analysed data from the 2026 wave of the Generative AI in UK General Practice (GPAI-UK) survey is an annual, convenience sample online survey of UK GPs administered via Doctors.net.uk within a scheduled monthly omnibus targeting approximately 1000 responses. The questionnaire retained core items on GenAI use and added items on conceptualisations of AI, anticipated effects on skills and meaningfulness of work, perceived patient responses, career advice and long-term professional expectations. Descriptive statistics were used to analyse responses.</p><p><strong>Results: </strong>A total of 1004 GPs completed the online survey. Just over half reported using GenAI in clinical practice (52%, 522/1004), most commonly ChatGPT, Heidi Health and Microsoft Copilot. Among users, 70% (363/522) reported that GenAI had reduced their work burden.Nearly half conceptualised GenAI as a collaborative technology working alongside clinicians (46%, 462/1004), while around one-fifth viewed it as a simple tool (21%, 209/1004) and 1 in 10 as a potential competitor (10%, 103/1004). Most expected GenAI to play at least a moderate role in routine practice within 5 years (69%, 693/1004).Looking ahead 10 years, one-third expected GenAI to change which GP skills matter most (34%, 337/1004), and 16% (163/1004) anticipated replacement of some core skills, most often documentation (811/1004), knowledge recall (509/1004) and diagnostic reasoning (442/1004). Many GPs said they would continue to encourage a career in general practice in the context of developments in AI tools; almost half had reservations (48%, 478/1004), nearly, one-third were not able to advise either way (32%, 317/1004). Most expected GenAI to substantially reshape the GP role while leaving GPs central to care (59%, 590/1004).</p><p><strong>Conclusion: </strong>In this survey, over half of UK GP respondents used GenAI and primarily viewed it as a collaborative technology. GPs anticipated reconfiguration of skills and roles rather than widespread professional displacement. Understanding clinicians' conceptualisations alongside adoption patterns is important for workforce planning, training and regulation.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000253"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13663762/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149198847","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":"High-risk without safeguards? The EU AI Act and the push for deregulation of medical AI.","authors":"Hannah Van Kolfschooten, Barry Solaiman, Daria Onitiu","doi":"10.1136/bmjdh-2026-000184","DOIUrl":"https://doi.org/10.1136/bmjdh-2026-000184","url":null,"abstract":"<p><p>Artificial intelligence (AI) is increasingly embedded in clinical decision-making, yet recent proposals to simplify the European Union (EU) Artificial Intelligence Act (AI Act) have reopened the question of whether AI-enabled medical devices should remain subject to the AI Act's full high-risk regime. Under the current framework, AI-enabled medical devices are regulated through a combination of sectoral medical device legislation (the Medical Devices Regulation (MDR) and In Vitro Diagnostic Medical Devices Regulation (IVDR)) and additional AI-specific safeguards contained in the AI Act. This narrative review analyses the Health Innovation Package and related legislative developments under the EU simplification agenda and examines their impact on the relationship between the AI Act and medical device law. It shows that the proposed reform would retain the high-risk classification of AI-enabled medical devices while removing most associated AI Act obligations. In effect, this decouples risk classification from the safeguards that give it regulatory meaning, including requirements on data governance, risk management, human oversight and postmarket monitoring. Rather than resolving regulatory overlap, the amendments shift the centre of gravity back to product-focused medical device law without ensuring equivalent AI-specific safeguards. These changes may narrow attention to fundamental rights, weaken oversight in clinical use and increase legal uncertainty regarding accountability and responsibility. In a domain where AI systems directly shape clinical decisions and patient outcomes, these changes risk undermining the conditions for safe, equitable and accountable deployment of medical AI.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000184"},"PeriodicalIF":0.0,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13629938/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149123002","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":"State of clinical AI in 2026.","authors":"John Emmett Worth, Anastasia Perez, David Jh Wu, Peter Brodeur, Emily Tat, Arjun Manrai, David Wu, Chase Walton, Liam G McCoy, Priyank Jain, Rebecca Handler, Jason Hom, Laura Zwaan, Vishnu Ravi, Brian Han, Kevin Schulman, Kathleen Lacar, Kameron Black, Adrian Haimovich, Adam Rodman, Ethan Goh, Jonathan H Chen","doi":"10.1136/bmjdh-2026-000113","DOIUrl":"https://doi.org/10.1136/bmjdh-2026-000113","url":null,"abstract":"<p><p>Clinical artificial intelligence (AI) has advanced rapidly, with frontier large language models now matching or exceeding physician performance on simulated diagnostic reasoning and clinical decision-support tasks. Yet adoption has outpaced the evidence base: fewer than 5% of cleared U.S. Food and Drug Administration (FDA) AI/machine learning (ML) devices have undergone peer-reviewed evaluation and prospective randomised trials remain scarce. This narrative review synthesises evidence from ARISE's State of Clinical AI Report 2026 across six domains: model performance and human benchmarking, evaluation methodology and benchmark validity, foundational methods including multimodal and multiagent systems, clinical workflow integration, patient-facing applications and applied domain-specific AI. We conducted a targeted literature search of PubMed, medRxiv and arXiv supplemented by expert nomination, including empirical studies of AI in clinical contexts. Across domains, frontier models demonstrate strong clinical reasoning capabilities but exhibit persistent failures in uncertainty calibration, metacognition and robustness to distributional shift. Standard benchmarks have saturated, prompting development of multidimensional evaluation frameworks that assess safety, real-world workflows and agentic capabilities. New trends in foundational methods include converting medical data into tokens and developing multiagent, multimodal systems. In clinical workflows, early prospective trials in clinical decision support and diagnostic imaging show promising results, though human-AI collaboration remains suboptimal and risks of automation bias and clinician deskilling are emerging. In tandem, patient-facing AI is progressing with more personalised health assistance; with this comes the risk of patient overtrust, raising the bar for guardrails. The most credible translational progress occurs on narrowly defined tasks with clear endpoints, while broader clinical autonomy awaits advances in evaluation and human-AI interaction.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000113"},"PeriodicalIF":0.0,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13629876/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149122980","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":"Contested futures of AI ethics for global population health.","authors":"James Shaw, Núria Vallès-Peris, Miquel Domènech","doi":"10.1136/bmjdh-2026-000098","DOIUrl":"https://doi.org/10.1136/bmjdh-2026-000098","url":null,"abstract":"<p><p>Ethical debate about artificial intelligence (AI) for health is fundamental to the good governance and long-term success of this group of technologies in healthcare and public health. Building on foundational contributions on this topic from the field of bioethics, we propose that interdisciplinary work on AI ethics should deepen the already strong engagements that exist with Science and Technology Studies (STS) to uphold core commitments of global population health. STS is an interdisciplinary field that specifically explores the meaning, development and governance of science and technology and offers enormous conceptual resources for AI ethics. We present a path towards deeper integration of STS and bioethics for an AI ethics that can more meaningfully address the most important issues presented by AI for global population health.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000098"},"PeriodicalIF":0.0,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13629987/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149123010","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":"From principles to practice: an actionable framework for AI governance in healthcare organisations.","authors":"Sam Freeman, Amy Wang, Sudeep Saraf, Erica Potts, Amy Mckimm, Enrico Coiera, Farah Magrabi","doi":"10.1136/bmjdh-2026-000018","DOIUrl":"https://doi.org/10.1136/bmjdh-2026-000018","url":null,"abstract":"<p><strong>Objective: </strong>Artificial intelligence (AI) technologies are being rapidly adopted in healthcare, yet organisational governance often lacks the processes needed to oversee their safe and responsible use. Previous AI governance frameworks largely focus on high-level AI ethics principles, leaving healthcare organisations struggling to translate these principles into practice, assess risk and embed AI oversight into existing processes. This study aimed to develop and validate a practice-oriented AI governance framework for healthcare organisations.</p><p><strong>Methods and analysis: </strong>We developed an AI governance framework using an exploratory multimethod design that drew on a scoping review, document analysis and semistructured interviews. The framework was validated through stakeholder workshops and application to three case studies of AI systems to identify patients for home-based care, radiology decision support and infection surveillance within a large tertiary healthcare organisation. The case studies represented real-world governance decisions faced by healthcare organisations and reflected the major pathways for acquiring AI capabilities: in-house development, commercial procurement and codevelopment with external partners.</p><p><strong>Results: </strong>The framework unifies ethics and governance principles, incorporates tiered oversight aligned with organisational digital maturity and includes a structured review checklist to support consistent decision-making about the implementation and operation of AI systems across clinical and operational contexts. Stakeholders found the framework relevant, usable and feasible to implement. Across the diverse AI use cases, the review checklist identified governance, safety and implementation risks and supported structured assessment and decision-making. The framework also enabled AI oversight to be integrated into existing organisational governance processes while supporting a proportionate approach to risk management across the AI life cycle.</p><p><strong>Conclusion: </strong>This study addresses an important gap between high-level AI governance principles and organisational implementation. The framework provides healthcare organisations with a structured approach to assessing, governing and monitoring AI systems throughout their life cycle and may support the safe and responsible adoption of AI in practice. Further research is needed to evaluate the framework across diverse healthcare settings and emerging AI technologies, including generative and agentic AI.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000018"},"PeriodicalIF":0.0,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13629979/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149122971","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":"Shifting bottleneck: AI capability and healthcare adaptation.","authors":"James Siddle, Alistair Connell","doi":"10.1136/bmjdh-2026-000264","DOIUrl":"https://doi.org/10.1136/bmjdh-2026-000264","url":null,"abstract":"","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000264"},"PeriodicalIF":0.0,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13630049/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149123013","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":"Me, my doctor and AI: a qualitative interview study on patients' requirements for trustworthy use of AI-CDSS in ophthalmology.","authors":"Insa Schaffernak, Julia Cecil, Anjesa Mahmuti, Selina Angerer, Tobias Blaschta, Eva Lermer","doi":"10.1136/bmjdh-2026-000185","DOIUrl":"10.1136/bmjdh-2026-000185","url":null,"abstract":"<p><strong>Objective: </strong>Artificial intelligence-based clinical decision support systems (AI-CDSS) are increasingly entering care and redefining how medical decisions are made. As patients are directly affected by hybrid physician-AI decisions without interacting with the system itself, trust, transparency and involvement in decision-making become important concerns. This study explored patients' requirements for trustworthy use of AI-CDSS in ophthalmology, with particular attention to expectations regarding disclosure of AI use and informational needs.</p><p><strong>Methods and analysis: </strong>We conducted semi-structured interviews with 23 ophthalmology patients and caregivers (aged 22-84 years) in Germany between January and March 2026. Participants were recruited via patient organisations, ophthalmology practices, clinics, social media, or purposively, with the aim of obtaining a sample diverse in age, gender and eye conditions. We used inductive qualitative content analysis to examine their perceptions of AI-assisted decision-making, factors shaping trust in AI-supported medical decisions and expectations regarding communication and AI disclosure.</p><p><strong>Results: </strong>Participants were generally open to the use of AI-CDSS in ophthalmology but had several requirements for trustworthy use of AI-CDSS. These related to three domains: the physician (eg, AI competence, critical evaluation of outputs), the AI system (eg, performance, explanations) and governance of its use (eg, transparency, consent). Participants considered it essential that physicians be able to identify incorrect AI output. Relatedly, the physician emerged as an important trust mediator: some indicated that pre-existing trust in the physician would extend to trust in the use of AI-CDSS, while for others, trust appears to depend on the perceived physician-AI interaction. Transparency about AI use was particularly important in this context. Almost all participants expected to be informed when AI-CDSS were used, although preferences regarding the amount and format of information varied.</p><p><strong>Conclusion: </strong>The findings suggest that patients' comfort with AI-assisted clinical decision-making depends strongly on the perceived physician-AI interaction and physician-patient communication. Specifically, disclosure appears to be important for informational purposes and for enabling patients to assess the trustworthiness of physician-AI decisions and remain involved in decision-making. As implementation and governance decisions likely shape how patients experience AI-assisted decision-making, they should be made with explicit attention to patient concerns.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000185"},"PeriodicalIF":0.0,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13599939/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149043415","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}