BMJ Health & Care Informatics最新文献

{"title":"Omission and hallucination prevalence of clinical guidelines in diagnostic large language model outputs.","authors":"Robin van Kessel, Michael Anderson, Brian McMillan, Marc R Matthews, Paul Rust, Pauline Pearcy, Khurram Nasir, Elias Mossialos","doi":"10.1136/bmjhci-2025-101959","DOIUrl":"10.1136/bmjhci-2025-101959","url":null,"abstract":"<p><strong>Objective: </strong>Meaningful assessments of how large language models (LLMs) incorporate clinical guidelines require large-scale testing over many queries. Here, we evaluate the prevalence of clinical guideline omissions and hallucinations in a large sample of diagnostic LLM outputs.</p><p><strong>Methods: </strong>We used simulated case vignettes and zero-shot prompting to generate diagnostic outputs and rationales from GPT-4.1 and DeepSeek-V3. English case vignettes were created for hypercholesterolaemia and type-2 diabetes mellitus. Each vignette contained identical medical information, while sociodemographic characteristics varied in terms of sex, ethnicity and location. We calculated the prevalence of existing and hallucinated clinical guidelines in LLM outputs across disease, LLM and sociodemographic characteristics.</p><p><strong>Results: </strong>We analysed a total of 12 197 LLM outputs, which quantifies three hazard areas: omissions (up to 97% for DeepSeek-V3 and 46% for GPT-4.1), hallucinations (up to 9%) and inconsistencies (guideline citation rate ranging from 0% to 78.39% across sociodemographic vignettes). Omission and hallucination rates were generally similar across vignettes with different sex or ethnicity data, yet were particularly sensitive to patient location.</p><p><strong>Discussion: </strong>This study highlights significant variability in clinical guideline prediction across two different diseases, three different sociodemographic variables and two LLMs, even when the LLMs were instructed by identical prompts, establishing clinical guideline prediction in LLM outputs as a stochastic event.</p><p><strong>Conclusion: </strong>The stochastic nature of LLMs creates a unique challenge for evidence generation and clinical deployment. Being able to measure and capture this stochasticity within high-quality research designs will be a prerequisite to advancing the responsible deployment of LLMs in healthcare.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13110572/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147761688","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":"Does the accuracy of medication administration documentation improve with electronic medication systems? A stepped-wedge cluster randomised trial.","authors":"Tim Badgery-Parker, Ling Li, Amanda Woods, Magdalena Raban, Johanna Westbrook","doi":"10.1136/bmjhci-2025-101507","DOIUrl":"10.1136/bmjhci-2025-101507","url":null,"abstract":"<p><strong>Objectives: </strong>Accurate documentation of medication administration is crucial for patient safety. This study compares the accuracy of documentation between paper-based and electronic medication management (eMM) systems.</p><p><strong>Methods: </strong>A secondary analysis of data from a stepped-wedge cluster randomised trial of an eMM implementation was conducted. The eMM module allowed electronic prescribing, recording of drug dispensing, drug administration and medication reconciliation and monitoring. Medication administration processes were observed in a tertiary paediatric hospital in Sydney, Australia. Trained nurse observers recorded details of medication administrations which were then compared with the information documented by nurses about the administrations.</p><p><strong>Results: </strong>5095 administrations were analysed. With eMM compared with paper, significantly fewer administrations had no time documented (adjusted OR (aOR) 0.15; 95% CI 0.07 to 0.34; p<0.001), but documented details were more often inaccurate (aOR 4.6; 95% CI 3.3 to 6.6). Fewer doses were administered from incorrect prescriptions with eMM (aOR 0.25; 95% CI 0.18 to 0.36).</p><p><strong>Conclusion: </strong>eMM improves certain aspects of documentation but introduces new challenges, particularly in accurately recording administration details. Further investigation is needed to optimise these systems.</p><p><strong>Trial registration number: </strong>Australian New Zealand Clinical Trials Registry 370325.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13110603/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147761710","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":"i-MoMCARE: AI-enabled mobile app for maternal and child health care in Cambodia - a pilot implementation and evaluation study.","authors":"Hendra Goh, Dyna Khuon, Mengieng Ung, Sreymom Oy, Chhavarath Dary, Chhorlika Khim, Sveng Chea Ath Chhay, Yan Fang Lee, Rattana Kim, Siyan Yi, Vonthanak Saphonn","doi":"10.1136/bmjhci-2025-101691","DOIUrl":"10.1136/bmjhci-2025-101691","url":null,"abstract":"<p><strong>Objectives: </strong>In Cambodia, village health support groups (VHSGs) are central to maternal and child health (MCH) service delivery, yet maternal mortality remains high with regional disparities. Although digital health solutions show promise, their real-world use is underexplored. This study therefore aims to evaluate the i-MoMCARE app in supporting VHSGs and health centre staff for delivering MCH services.</p><p><strong>Methods: </strong>A mixed-methods approach guided by the Reach, Effectiveness, Adoption, Implementation and Maintenance (RE-AIM) framework was used. Reach was assessed using programme records (number of users trained and MCH cases managed). Effectiveness and adoption were evaluated using the mHealth App Usability Questionnaire (MAUQ) and qualitative data on user experience and intention to use. Implementation was examined through system usage monitoring and feedback on workflow integration. Maintenance was explored using qualitative insights on long-term use, barriers and recommendations.</p><p><strong>Results: </strong>44 VHSGs and 10 health centre staff were trained, and 96 MCH cases were managed with the app. Usefulness was rated highly (MAUQ 5.58, SD 1.04), with users appreciating the user-centric design that improved usability and integration into daily workflows. Effectiveness was similarly high (MAUQ 5.50, SD 0.95); the clinical decision support system and point-of-care alerts were valued as reliable decision-support tools, and the electronic medical record (EMR) was seen as a secure data repository. However, adoption was hindered by hardware and software limitations (MAUQ 5.79, SD 0.75), including device availability and low digital literacy.</p><p><strong>Discussion: </strong>Our findings highlight the value of the i-MoMCARE app in supporting VHSGs, particularly those with limited training, by enhancing their decision-making. The study also lays the groundwork for a unified EMR in primary care. However, concerns around hardware and software limitations hinder adoption and sustained use.</p><p><strong>Conclusion: </strong>To enable scale-up, we recommend reducing access barriers and supporting sustained use through diversifying the platform's design and implementing phased incentivisation strategies.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13110616/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147761732","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":"Barriers associated with the implementation, adoption, scale-up and sustainability of mHealth in Sub-Saharan Africa: a systematic review guided by the NASSS Framework.","authors":"Jordan Murray, Thomas Connolly, Zaeem Haq, Bernard David Naughton","doi":"10.1136/bmjhci-2025-101485","DOIUrl":"10.1136/bmjhci-2025-101485","url":null,"abstract":"<p><strong>Objectives: </strong>The objectives of this study were to highlight and evaluate the factors affecting the non-adoption, abandonment of and challenges to scale-up, spread and sustainability of mHealth in Sub-Saharan Africa (SSA) and to assess the appropriateness of the Non-adoption, Abandonment, Scale-up, Spread, and Sustainability (NASSS) Framework in the SSA context.</p><p><strong>Methods: </strong>A systematic search of published research was conducted using PubMed and Embase research databases from 2012 to 2025. Peer-reviewed studies from SSA countries, evaluating mHealth interventions, were included in this study. Papers were analysed to identify barriers according to the seven domains of the NASSS Framework, which are (1) condition, (2) technology, (3) value proposition, (4) adopters, (5) organisation(s), (6) wider system, (7) embedding and adaptation over time RESULTS: This study identified 409 barriers across 91 papers. 309 barriers were covered by the seven NASSS Framework domains, while 100 were unaddressed by the framework. The main barriers were associated with domain 2, the technical aspects of the technology (27%). Other prominent barriers related to the wider system (18%), the adopter (11%), the organisation (9%) or condition (8%), with the value proposition (1%) and embedding and adaptation over time (1%), least hindering implementation. Challenges which were unaddressed by the framework (24%) included socioeconomic and infrastructural issues, distinct to lower and middle-income countries (LMIC).</p><p><strong>Discussion: </strong>This paper highlights the importance of evaluating rudimentary aspects of mHealth and preparing infrastructure before addressing more complex issues during mHealth implementation in SSA.</p><p><strong>Conclusion: </strong>Although the NASSS Framework is applicable to the SSA region, it would benefit from some further development to improve its applicability in LMICs.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13110535/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147761647","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":"Using a large language model artificial intelligence agent to improve the efficiency of clinical quality measure evidence evaluation: a case study.","authors":"Jeffrey Geppert, Barbara Jones, Jeremy Bellay, Gerrit Bryan","doi":"10.1136/bmjhci-2025-101621","DOIUrl":"10.1136/bmjhci-2025-101621","url":null,"abstract":"<p><strong>Objectives: </strong>To evaluate the feasibility and performance of a large language model (LLM)-based artificial intelligence (AI) agent, implemented within a structured Claim-Argument-Evidence System (CAES), for supporting the review of clinical quality measure (CQM) evidence in the Centers for Medicare & Medicaid Services Consensus-Based Entity (CBE) endorsement process.</p><p><strong>Methods: </strong>The CBE conducted a pilot study using a previously endorsed measure. CAES extracted claims and citations from a submitted diagnostic performance measure for pneumonia, automatically retrieved additional relevant evidence from PubMed abstracts and assessed the quality, confidence and agreement of evidence supporting each claim. The system's assessments were compared with the judgement of a subject matter expert (SME).</p><p><strong>Results: </strong>CAES completed the assessment in approximately 5 hours. The SME agreed with the CAES-assigned claim statuses for 69% of claims, was neutral for 11% and disagreed for 14%. Disagreements primarily stemmed from the need for contextual interpretation beyond abstracts.</p><p><strong>Discussion: </strong>Manual evaluation of CQM evidence requires significant time and resources, estimated at over 2400 labour hours per review cycle, limiting efficiency and transparency.The AI agent evaluated 64 claims and 355 claim-evidence pairs related to the pneumonia diagnosis measure. It assigned claim statuses based on evidence strength and generated justifications.</p><p><strong>Conclusion: </strong>This pilot demonstrated the feasibility and potential of LLM-based AI agents to improve the efficiency and transparency of evidence review for CQMs. Further development is needed to incorporate additional data sources and extend applicability across the measure development lifecycle.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13110563/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147761662","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":"Supporting street medicine through medical informatics: a manifesto on research needs.","authors":"Alfred Franz Winter, Andrea Alverà, Elske Ammenwerth, Alina M Breaz, Catherine Chronaki, Mihaela Crişan-Vida, Jaime Delgado, Parisis Gallos, Franziska Jahn, Philipp Plicka, Birgit Schneider, Sven Speerforck, Lăcrămioara Stoicu-Tivadar","doi":"10.1136/bmjhci-2025-102005","DOIUrl":"10.1136/bmjhci-2025-102005","url":null,"abstract":"<p><p>Homeless individuals face major barriers in accessing regular healthcare, leading to the development of street medicine as a distinct humanitarian field. To enhance continuity and quality of care, consistent documentation is crucial. However, electronic health records are rarely used in European street medicine. An interdisciplinary European workshop identified four key research fields for medical informatics in street medicine: (1) Research ethics: The creation of ethical guidelines that promote co-creation and active communication with patients. (2) System development: Designing a secure, anonymous and European Health Data Space-compliant personal Electronic Health and Social Record (pEHSR), establishing standardised health and social care datasets, and implementing dynamic consent management methods. (3) Education and training: Developing targeted programmes to improve digital and health literacy among street medicine clients and training professionals in the effective use of the pEHSR system. (4) Management and evaluation: Creating management structures and evaluation frameworks suited to the unique challenges of street medicine.This manifesto calls for internationally coordinated scientific efforts to build effective, integrated digital health solutions for vulnerable populations. It encourages researchers and practitioners from medical informatics and related fields to engage with these priorities and support innovation that advances equitable healthcare access.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13084957/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147670330","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":"Acceptable accuracy for medical AI: a survey of physicians and the general population in Sweden.","authors":"Rasmus Arvidsson, Jonathan Widén, Lina Al-Naasan, Ronny Kent K Gunnarsson, Peter Nymberg, Charlotte R Blease, Anna Moberg, Pär-Daniel Sundvall, Carl Wikberg, David Sundemo","doi":"10.1136/bmjhci-2025-101899","DOIUrl":"10.1136/bmjhci-2025-101899","url":null,"abstract":"<p><strong>Objectives: </strong>To identify the lowest sensitivity and specificity that physicians and the general population consider acceptable for medical artificial intelligence (AI), relative to current human performance.</p><p><strong>Methods: </strong>In a nationwide, cross-sectional survey in Sweden, 2025, random samples of 500 physicians and 500 adults from the general population were mailed a questionnaire presenting three vignettes (chest pain triage, sore throat triage, ECG myocardial infarction detection) with the corresponding human performance. Participants reported the maximum number of cases an AI should be allowed to miss or over-refer.</p><p><strong>Results: </strong>Response rates were 45% among physicians and 31% in the general population. Both groups demanded higher AI accuracy than the human benchmark for all cases. In the chest pain triage vignette, the nurse correctly referred 84 of 100 true emergencies; physicians required the AI to correctly refer 11 additional patients (95% sensitivity) and the general population demanded referral of 16 additional patients (100% sensitivity) (p<0.001 for both groups). Among 100 patients not requiring referral, the nurse would mistakenly refer 66. Both groups required the AI to reduce unnecessary referrals by 16 (50% specificity) (p<0.001). A similar pattern was observed in the other vignettes.</p><p><strong>Discussion: </strong>The accuracy thresholds required by the respondents exceed the performance of many existing systems, although emerging AI research shows promise in narrowing the gap.</p><p><strong>Conclusion: </strong>Physicians and the general population require medical AI systems to outperform human clinicians. When implementing AI in healthcare settings, early engagement with both groups may be necessary to align expectations with real-world system performance.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13052720/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147608087","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":"Robotic process automation for identifying missing codes on insurance claims.","authors":"Jiyun Lee, Jun Hwan Cho, Won Joo Lee, Dong Hoon Kim, Ye Lim Gong, Won Tae Kim, Chan Woong Kim","doi":"10.1136/bmjhci-2025-101821","DOIUrl":"10.1136/bmjhci-2025-101821","url":null,"abstract":"<p><strong>Objectives: </strong>This study aimed to develop and implement robotic process automation (RPA) for identifying missing codes during insurance claim post-review at a tertiary hospital and to evaluate its feasibility and effectiveness METHODS: As a single-centre, operational implementation, an RPA system integrated with optical character recognition (OCR) and electronic medical record (EMR) platforms was developed using Blue Prism. The system compared 532 surgical procedure codes with 21 cutting device codes, automatically flagging discrepancies. Accuracy and efficiency were compared with manual review.</p><p><strong>Results: </strong>Between 1 and 31 May 2025, the RPA system analysed 61 claim statements and performed 199 OCR processes. The Google Cloud Vision API (application programming interface) achieved 100% detection accuracy without false positives, while Tesseract yielded lower accuracy. The RPA reduced average processing time from 120 (manual review) to 54 min, representing a 55% efficiency gain.</p><p><strong>Discussion: </strong>RPA reliably automated repetitive, rule-based administrative tasks, improving accuracy and standardisation of insurance claim audits. Secure system architecture ensured compliance with healthcare data protection standards. User-centred development and integration with EMR demonstrated feasibility in complex healthcare workflows.</p><p><strong>Conclusion: </strong>Implementing RPA for insurance claim post-review significantly enhanced efficiency and accuracy, reduced administrative workload and provided a scalable model for digital transformation in healthcare administration.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13052645/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147608051","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":"Novel two-stage deep learning framework for automated pressure injury classification.","authors":"Ting-Yu Lai, Yi-Jiun Chou, Chun-You Liu, Chien-Wei Chen, Ching-Ting Lin, Wei-Chun Wang, Yimin Hsu, Ming-Li Hsieh, Shih-Sheng Chang","doi":"10.1136/bmjhci-2025-101636","DOIUrl":"10.1136/bmjhci-2025-101636","url":null,"abstract":"<p><strong>Objective: </strong>The study aims to develop an artificial intelligence (AI) framework for automatic pressure injury (PI) staging directly from raw clinical images, without requiring manual lesion localisation. By integrating a two-stage deep learning approach-combining object detection and image classification-the proposed system seeks to improve diagnostic accuracy, reduce subjectivity and support consistent staging in real-world clinical workflows.</p><p><strong>Methods: </strong>We conducted a retrospective study using 1807 PI images from China Medical University Hospital collected between 2020 and 2024. Lesions were annotated by five senior nurses with consensus. YOLOv9 was used for lesion detection, and DenseNet161 for staging. Performance metrics included accuracy, sensitivity, specificity, F1 score, area under the curve (AUC) and mean average precision (mAP)@0.5.</p><p><strong>Results: </strong>The object detection model (YOLOv9) achieved an mAP@0.5 of 0.796. On the independent test set (n=365), the staging model demonstrated an overall accuracy of 0.775, sensitivity of 0.775, specificity of 0.955, precision of 0.779 and F1 score of 0.775.</p><p><strong>Discussion: </strong>The proposed two-stage AI framework-using YOLOv9 for lesion localisation and DenseNet161 for staging-demonstrated performance comparable to or exceeding previously reported approaches, while offering improved clinical interpretability when applied to real-world clinical images. By separating detection and classification, the framework reduced background noise and improved the ability to distinguish visually similar PI stages. However, challenges remain in handling intra-wound heterogeneity, subtle early-stage boundaries and variability in image quality.</p><p><strong>Conclusions: </strong>This two-stage AI system may assist clinical staff by providing standardised, reproducible PI staging from clinical images, with potential for integration into nursing workflows.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13034204/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147526989","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":"Introduction to secure data sharing in primary care using the federated causal learning models.","authors":"Miaoshuang Chen, Zongqi Chang, Peng Gong, Zihuan Tang, Lin Hu, Xingyu Zhang, Shiyang Ma, Jiaqiang Liao, Xia Jiang, Jiayuan Li, Tao Zhang","doi":"10.1136/bmjhci-2025-101543","DOIUrl":"10.1136/bmjhci-2025-101543","url":null,"abstract":"<p><strong>Objectives: </strong>In primary healthcare research, there are core challenges such as data silos and missing data. Furthermore, the current high technical barriers severely limit effective cross-regional data analysis.</p><p><strong>Methods: </strong>This work was the first to apply the federated causal learning framework to primary healthcare. Through two case studies, we demonstrated how to estimate cross-regional causal effects without sharing raw data, guided by a detailed step-by-step protocol. Furthermore, we designed a systematic simulation study tailored to the characteristics of primary healthcare data to evaluate the performance of this framework under various missingness mechanisms and proportion settings.</p><p><strong>Results: </strong>This framework was effectively applied to both chronic non-communicable disease and infectious disease, two major issues that remain public health priorities requiring sustained attention. In the cardiovascular disease case, the estimated average treatment effect (ATE) from the federated model (ATE=0.017) was very close to the result of the centralised model (ATE=0.018). Under all missing data scenarios, the stable model consistently achieved perfect or near-perfect coverage rates, maintaining performance even under missingness rates as high as 20%. In addition, the coverage of the unstable model remained robustly above 96.10% even when model assumptions were violated.</p><p><strong>Discussion: </strong>This work demonstrated the effectiveness and practicality of federated causal learning in primary healthcare data, which was characterised by decentralisation and susceptibility to missing data.</p><p><strong>Conclusion: </strong>This framework provided a feasible solution for primary healthcare workers to safely conduct federated causal inference. It held promise for advancing data-driven precision decision-making in primary care.</p>","PeriodicalId":9050,"journal":{"name":"BMJ Health & Care Informatics","volume":"33 1","pages":""},"PeriodicalIF":6.2,"publicationDate":"2026-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13064139/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147527051","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}