BMJ digital health & AI最新文献

{"title":"Beyond protected characteristics: rethinking AI fairness through the lens of structural and social determinants of health.","authors":"Karol Basta, Uzma Zahid, Mahi Hardalupas, Dijana Vilic","doi":"10.1136/bmjdh-2026-000087","DOIUrl":"10.1136/bmjdh-2026-000087","url":null,"abstract":"","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000087"},"PeriodicalIF":0.0,"publicationDate":"2026-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528398/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915098","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":"How artificial intelligence is disrupting the field of medical knowledge and evaluation: a narrative review.","authors":"Kathrin Cresswell, Robin Williams","doi":"10.1136/bmjdh-2026-000104","DOIUrl":"10.1136/bmjdh-2026-000104","url":null,"abstract":"<p><p>Although artificial intelligence (AI) shows considerable promise in healthcare, its introduction and safe and effective exploitation involve several challenges. We argue that to maximise the benefits of AI and understand, track and mitigate potential harms there must be a shift in how AI is evaluated. Evaluation should be more closely linked to the local contexts in which AI is deployed and used and better integrated with clinical practice. To support this shift, we propose three interrelated levels of evaluation: evaluation of implementation, which examines transformations in contextually situated practices; evaluation of optimisation, which explores changes to clinical pathways over extended periods; and evaluation of scaling, which focuses on coordination through regional and national structures. This approach offers significant opportunities to advance both healthcare practice and the epistemology of medical knowledge.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000104"},"PeriodicalIF":0.0,"publicationDate":"2026-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528392/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915025","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":"The use of ambient voice technology (AVT) for clinician-patient consultations in healthcare practice: A rapid systematic scoping review.","authors":"Rachel Lawrence, Kevin Herbert, Theo Georghiou, Raj Mehta, Pei Li Ng, Lucina Rolewicz, Stephen Morris, Jenny Shand","doi":"10.1136/bmjdh-2026-000120","DOIUrl":"10.1136/bmjdh-2026-000120","url":null,"abstract":"<p><strong>Objective: </strong>Using ambient voice technology (AVT) to transcribe and summarise clinician-patient consultations in real time is increasing in healthcare practice, with the aim of reducing clinical documentation burden. However, the evidence base remains underdeveloped. This review synthesised current evidence on the use and impact of AVT for documenting clinician-patient consultations in healthcare. We also examined how impact has been evaluated and measured with the aim of informing future research and assessment in real-world practice.</p><p><strong>Methods: </strong>We conducted a rapid systematic scoping review (PROSPERO: CRD420251155579). Four databases and grey literature sources were searched in September 2025. Of 598 articles screened, 21 met the inclusion criteria. These studies evaluated user experiences with AVT (n<i>=</i>13), quantified impacts using objective numerical data (n<i>=</i>19) and perceived impacts based on self-reported experiences (n<i>=</i>17).</p><p><strong>Results: </strong>Evidence on the real-world impact of AVT remains limited. While some studies report reduced documentation burden and improvements in staff well-being and perceived care quality, these findings largely stem from short-term observational designs and self-reported measures. Evidence relating to wider, system-level impacts such as pathway productivity, patient safety and cost effectiveness is sparse. Approaches to evaluating AVT were heterogeneous and inconsistent, particularly when reporting on system-level impacts beyond the clinician-patient consultation. Different factors (e.g., workflow integration, output quality), the consultation context and clinical setting influenced experiences of using AVT.</p><p><strong>Conclusions: </strong>Current evidence does not yet support widespread adoption of AVT in clinical practice. Although early findings suggest potentially important benefits, available studies are heterogeneous, predominantly short-term and methodologically limited. It remains unclear whether the improvements reported to date translate into sustained gains in clinical capacity, care quality or system efficiency. Longer-term evaluations using standardised outcome measures complemented by qualitative methods, are needed to determine the impact and value of AVT in healthcare.</p><p><strong>Prospero registration number: </strong>CRD420251155579.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000120"},"PeriodicalIF":0.0,"publicationDate":"2026-07-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528325/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915139","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 AI for mental health: promise, risk and the path forward.","authors":"Matthew Brown, Naomi Herz","doi":"10.1136/bmjdh-2026-000158","DOIUrl":"10.1136/bmjdh-2026-000158","url":null,"abstract":"","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000158"},"PeriodicalIF":0.0,"publicationDate":"2026-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528334/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915062","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":"Benchmarking large language models and clinicians using locally generated primary healthcare vignettes in Kenya.","authors":"Paul Mwaniki, Wilkister Musau, Lynda Isaaka, Conrad Wanyama, Vaishnavi Menon, Alastair K Denniston, Xiaoxuan Liu, Mira Emmanuel-Fabula, Gwydion Williams, Bilal Akhter Mateen, Ambrose Agweyu","doi":"10.1136/bmjdh-2026-000081","DOIUrl":"10.1136/bmjdh-2026-000081","url":null,"abstract":"<p><strong>Objective: </strong>To determine the ability of large language models (LLMs) to answer open-ended healthcare questions (rather than multiple-choice questions) accurately in a low-resource setting.</p><p><strong>Methods and analysis: </strong>We benchmarked five LLMs (GPT-4.1, Gemini-2.5-Flash, DeepSeek-R1, MedGemma and o3) against Kenyan clinicians, using a randomly subsampled dataset of 507 vignettes (from a larger pool of 5107 clinical scenarios) spanning 12 nursing competency categories. Blinded physician panels rated responses using a 5-point Likert scale on an 11-domain rubric covering accuracy, safety, contextual appropriateness, and communication. We summarised mean scores and used Bayesian ordinal logistic regression to estimate probabilities of high-quality ratings (≥4) and to perform pairwise comparisons between LLMs and clinicians.</p><p><strong>Results: </strong>Clinician mean ratings were lower than those for LLMs in 9/11 domains: 2.86 vs 4.25-4.72 (guideline alignment), 2.76 vs 4.25-4.73 (expert knowledge), 2.96 vs 4.30-4.73 (logical coherence) and 2.58 vs 4.16-4.68 (low omission of critical information). On safety-related domains, LLMs received higher ratings: minimal extent of possible harm 3.16 vs 4.29-4.68; low likelihood of harm 3.68 vs 4.54-4.81. Performance was similar for low inclusion of irrelevant content (4.28 vs 4.25-4.35) and for avoidance of demographic bias (4.86 vs 4.91-4.94). In Bayesian models, LLMs had >90% probability of ratings ≥4 in most domains, whereas clinicians exceeded 90% only for contextual relevance and demographic/socioeconomic bias. Pairwise contrasts showed broadly overlapping credible intervals among LLMs, with o3 leading numerically most domains except contextual relevance, demographic/socio-economic bias and relevance to the question. Generating all LLM responses cost US$3.86-US$8.68 per model (US$0.008-US$0.017 per vignette), compared with US$3.35 per clinician-generated vignette.</p><p><strong>Conclusions: </strong>In controlled vignette-based tasks, LLMs produced responses that were more accurate, safer and more structured than clinicians, suggesting LLMs may have potential as supplementary knowledge and safety support tools. Findings support further evaluation in real patient encounters to determine effectiveness, safety and integration into clinical workflows, particularly in resource-constrained health systems.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000081"},"PeriodicalIF":0.0,"publicationDate":"2026-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528382/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915114","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":"Predicting cardiac index from coronary angiogram videos: a novel application of 3D convolutional neural networks.","authors":"Shih-Sheng Chang, Behrouz Rostami, Gerardo Lo Russo, Chia-Hao Liu, Eunjung Lee, Mohamad Alkhouli","doi":"10.1136/bmjdh-2026-000080","DOIUrl":"10.1136/bmjdh-2026-000080","url":null,"abstract":"<p><strong>Objective: </strong>Data on cardiac index (CI) in patients undergoing coronary angiography (CAG) can enhance decision-making and improve clinical outcomes. However, this necessitates an additional invasive procedure, and therefore, it is frequently omitted. This study explores the novel application of convolutional neural networks (CNNs) to predict CI from routine CAGs.</p><p><strong>Methods and analysis: </strong>Patients who underwent CAG and simultaneous right heart catheterisation at Mayo Clinic between 2002 and 2023 were included. A three-dimensional (3D)-CNN model (X3D architecture) was developed to predict binary CI categories (normal: CI≥2.2 vs abnormal: CI<2.2 L/min/m²) from CAG videos. CAGs were randomly split into training (70%), validation (15%) and test (15%) sets. All available videos from each study were entered into the model. The model's performance was evaluated using the area under the curve (AUC), sensitivity, specificity and F1 score.</p><p><strong>Results: </strong>A total of 15 297 CAG studies were considered. After applying a quality-check algorithm to exclude low-quality or irrelevant CAGs, 11 475 studies from 9103 patients were included. CI was abnormal in 25.2% of studies. The model achieved an AUC of 0.83 (95% CI 0.81 to 0.85), sensitivity of 0.72 (95% CI 0.68 to 0.77), specificity of 0.78 (95% CI 0.75 to 0.80) and F1 score of 0.62 (95% CI 0.58 to 0.65) on the test dataset. Limiting the input to 4, 3 or 2 prespecified CAG videos representing standard views did not significantly affect the model's performance, supporting its applicability in varied clinical workflows.</p><p><strong>Conclusions: </strong>This study demonstrates the feasibility of leveraging deep learning models to predict CI from CAG videos. The proposed 3D-CNN model shows promise as a tool for providing physicians with real-time data to aid decision-making during CAG procedures.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000080"},"PeriodicalIF":0.0,"publicationDate":"2026-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528320/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915117","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":"Academia and industry in the age of artificial intelligence: a narrative review.","authors":"Elmer Victor Bernstam, Todd R Johnson, Fares Alahdab","doi":"10.1136/bmjdh-2026-000047","DOIUrl":"10.1136/bmjdh-2026-000047","url":null,"abstract":"<p><p>Recent advances in artificial intelligence (AI) have fuelled a boom phase and promise to revolutionise many domains, including the traditional activities of academic medical centres (AMCs): biomedical research, education and healthcare services. Progress in AI is driven by three factors: (1) faster and cheaper processors (compute), (2) increasingly efficient training algorithms and neural network architectures and (3) an increase in electronically readable data. Outside of healthcare, industry investment dwarfs AI investment by governments. The largest computer clusters and datasets are now managed by industry. Similarly, most PhDs specialising in AI research are employed by industry. Therefore, it is not surprising that most of the largest and best-performing models are now developed by industry. Despite the industry lead in the general domain, we argue that AMCs have a key advantage in healthcare; they steward high-value biomedical data, a key driver of AI progress. Data are a strategic asset for AMCs. However, to capitalise on that asset, AMCs must invest in data ecosystems that improve data quality and utility, including the processes that produce data as well as data management. AMCs must also implement policies and procedures that lower the barriers to data use and reuse. With burgeoning interest in AI, AMCs should take on the role of careful, yet efficient, stewards of healthcare data to maximise its potential, while acting in the public's best interests.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000047"},"PeriodicalIF":0.0,"publicationDate":"2026-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528323/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915094","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":"Computer-aided polyp detection and characterisation systems to support colonoscopy: a systematic review with results stratified by each individual artificial intelligence system.","authors":"Nicole Downes, Clare Dadswell, Sophie Ip, Isaac Mackenzie, Victoria Wakefield, Steven J Edwards","doi":"10.1136/bmjdh-2026-000046","DOIUrl":"10.1136/bmjdh-2026-000046","url":null,"abstract":"<p><strong>Objective: </strong>To assess the clinical effectiveness and diagnostic accuracy of specific artificial intelligence (AI) systems with computer-aided polyp detection (CADe) and/or computer-aided polyp characterisation (CADx) functionalities.</p><p><strong>Methods and analysis: </strong>Studies using CADe/CADx systems alongside endoscopist judgement during real-time colonoscopies for any indication were included from searches performed in June 2025. Adenoma detection rate (ADR) and diagnostic accuracy (eg, sensitivity and specificity) were key outcomes, with other polyp detection outcomes, procedural duration and adverse events (AEs) also analysed. For CADe, meta-analyses were stratified by each individual AI system. Meta-analyses were not performed for CADx due to heterogeneity concerns. Subgroup analyses for colonoscopy indication and endoscopist experience were explored.</p><p><strong>Results: </strong>Overall, 52 studies (43 for CADe and 9 for CADx) covering eight CADe/CADx systems were included. ADR increased with each individual CADe system versus colonoscopy without CADe; results for two systems lacked statistical significance. While more uncertain, some systems also increased the detection of advanced and non-advanced adenomas, different-sized adenomas, sessile serrated lesions (SSLs) and/or hyperplastic/non-neoplastic polyps. Adjunct CADx diagnostic accuracy results were inconsistent and had limitations; comparisons against endoscopist optical diagnoses were often not available, SSLs were often analysed as non-neoplastic and low-confidence diagnoses were often excluded. CADe may increase procedural duration slightly, and no concerns about AEs were identified. No data for the impact of CADe/CADx on long-term outcomes, such as interval colorectal cancer (CRC) or mortality, were available. No robust conclusions about effects across different colonoscopy indications or endoscopist experience subgroups could be made.</p><p><strong>Conclusion: </strong>The included CADe systems may each increase ADR, with increased detection of other polyp types possible and no major concerns about procedure duration or AEs. Further research on CADx diagnostic accuracy, CADe/CADx impact on long-term outcomes such as CRC incidence and effects across colonoscopy indication and endoscopist experience subgroups is required.</p><p><strong>Prospero registration number: </strong>CRD42024586541.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000046"},"PeriodicalIF":0.0,"publicationDate":"2026-06-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528337/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915104","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":"Machine learning-based prediction of bacterial infection foci: model development, calibration and uncertainty quantification using conformal prediction methods.","authors":"Jacob Bahnsen Schmidt, Karen Leth Nielsen, Dmytro Strunin, Nikolai S Kirkby, Jesper Q Thomassen, Steen C Rasmussen, Ruth Frikke-Schmidt, Frederik Boetius Hertz, Allan Peter Engsig-Karup","doi":"10.1136/bmjdh-2026-000068","DOIUrl":"10.1136/bmjdh-2026-000068","url":null,"abstract":"<p><strong>Objective: </strong>To develop and evaluate machine learning (ML) models capable of reliably predicting the focus of bacterial infections in hospitalised patients, addressing limitations of conventional microbiological diagnostics.</p><p><strong>Methods and analysis: </strong>We conducted a retrospective study using data from 10 153 patients admitted to Rigshospitalet, Denmark, between 1 November 2019 and 3 June 2023. The dataset included microbiological findings, biochemical measurements and vital parameters. ML models were trained and evaluated, with outputs calibrated using Venn-ABERS calibration. Model uncertainty was quantified through conformal risk control to provide statistically robust uncertainty estimates.</p><p><strong>Results: </strong>This study shows that the XGBoost model demonstrated the best performance, achieving a log loss of 0.209±0.006 (mean±SD) and an area under the receiver operating characteristic curve of 0.93±0.007. Incorporating conformal prediction techniques enhanced predictive reliability by combining high accuracy with calibrated probabilistic predictions and uncertainty estimates.</p><p><strong>Conclusion: </strong>We find that ML models, particularly XGBoost combined with conformal prediction frameworks, can accurately predict the focus of bacterial infections while providing robust uncertainty quantification. This approach offers a promising addition to conventional diagnostic methods, potentially improving clinical decision-making.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000068"},"PeriodicalIF":0.0,"publicationDate":"2026-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528319/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915125","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":"Analysis of the challenges in ensuring robust and effective clinical evaluation of AI-based digital medical devices.","authors":"Sandrine Boulet, Vincent Thévenet, Judith Abecassis, Corinne Collignon, Bruno Giraudeau, Edouard Lhomme, François Petit, Raphaël Porcher, Laura Richert, Florence Saillour, Moreno Ursino, Gaël Varoquaux, Vincent Vercamer, Sarah Zohar, Rodolphe Thiébaut","doi":"10.1136/bmjdh-2026-000043","DOIUrl":"10.1136/bmjdh-2026-000043","url":null,"abstract":"<p><p>The rapid evolution of algorithms over time, the validity of the digital criteria used for evaluation and the organisational impact are examples of issues that are both diverse and more specific to digital medical devices (DMDs) than drugs. Furthermore, some DMDs that use artificial intelligence (AI) have the feature of being developed using continuous learning systems, in which the algorithms learn from all the data they use. Despite these unique features, their clinical evaluation often follows existing frameworks designed for drugs. Nevertheless, the specific features of AI-based DMDs call for an adapted clinical evaluation approach and raise new methodological questions. We propose a narrative review of these characteristics and the challenges they could imply for the evaluation of DMDs. Hence, the definition of users and intended use, the choice of datasets for analytical and clinical validation, the specification of outcomes of interest, the use of real-world data studies in addition to randomised clinical trials, and the choice of an appropriate comparator are key elements that can influence the design, analysis and generalisability of the clinical assessment.</p>","PeriodicalId":524116,"journal":{"name":"BMJ digital health & AI","volume":"2 1","pages":"e000043"},"PeriodicalIF":0.0,"publicationDate":"2026-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13528383/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148915136","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}