BMJ Health & Care Informatics最新文献

Objectives: To describe the implementation of a multidisciplinary, ethically grounded hackathon as a model to develop and evaluate generative AI (GenAI) solutions for real-world clinical challenges within a hospital setting.

Methods: The GenAI Health Hackathon (GAHH) organised at Hospital Clínic de Barcelona included 13 challenges were selected via an internal call based on clinical impact, feasibility and data availability. Participants accessed anonymised real-world data through a secure cloud environment. Teams employed large language models and retrieval-augmented generation to build prototypes addressing tasks such as clinical text structuring, decision support and workflow automation. Human-in-the-loop validation, explainability and regulatory safeguards were emphasised.

Results: The hackathon yielded multiple AI prototypes tested on real data. Results varied: entity recognition reached 90.5% accuracy, summarisation >90% clinician concordance and nutritional models achieved F1 scores of 0.75-0.93. Lower scores (F1<0.52, Jaccard Index <0.4) were seen in complex reasoning or multilingual tasks. Bias was explored in 10 projects, with mitigations such as stratified sampling, prompt tuning, disclaimers and expert oversight. A transferable framework was proposed to replicate responsible GenAI hackathons in clinical contexts.

Discussion: Interdisciplinary collaboration and real-world testing proved essential for aligning GenAI with clinical needs. The hackathon revealed challenges in bias, evaluation and integration but offered a transferable framework for responsible innovation under General Data Protection Regulation and the European Union Artificial Intelligence Act.

Conclusions: The GAHH demonstrated that GenAI can be safely and effectively applied in healthcare with rigorous governance and interdisciplinary collaboration, offering a scalable model for responsible AI innovation.

Objectives: Antimicrobial resistance is a critical public health threat. Large language models (LLMs) show great capability for providing health information. This study evaluates the effectiveness of LLMs in providing information on antibiotic use and infection management.

Methods: Using a mixed-method approach, responses to healthcare expert-designed scenarios from ChatGPT 3.5, ChatGPT 4.0, Claude 2.0 and Gemini 1.0, in both Italian and English, were analysed. Computational text analysis assessed readability, lexical diversity and sentiment, while content quality was assessed by three experts via DISCERN tool.

Results: 16 scenarios were developed. A total of 101 outputs and 5454 Likert-scale (1-5) scores were obtained for the analysis. A general positive performance gradient was found from ChatGPT 3.5 and 4.0 to Claude to Gemini. Gemini, although producing only five outputs before self-inhibition, consistently outperformed the other models across almost all metrics, producing more detailed, accessible, varied content and a positive overtone. ChatGPT 4.0 demonstrated the highest lexical diversity. A difference in performance by language was observed. All models showed a median score of 1 (IQR=2) regarding the domain addressing antimicrobial resistance.

Discussion: The study highlights a positive performance gradient towards Gemini, which showed superior content quality, accessibility and contextual awareness, although acknowledging its smaller dataset. Generating appropriate content to address antimicrobial resistance proved challenging.

Conclusions: LLMs offer great promise to provide appropriate medical information. However, they should play a supporting role rather than representing a replacement option for medical professionals, confirming the need for expert oversight and improved artificial intelligence design.

Objectives: To evaluate the impact of implementing a multidisciplinary integrated telehealth platform in central Taiwan on healthcare accessibility, emergency response and chronic disease management.

Methods: We analysed data from 26 institutions within a central Taiwan telehealth network between 2022 and 2024. The study evaluated the use and benefits of teleconsultation, artificial intelligence-assisted ECG monitoring during prehospital ambulance transfers and outcomes in patients with cryptogenic stroke following the platform integration. Satisfaction surveys were performed.

Results: By 2024, more than 300 teleconsultations were performed across 26 partner facilities. Non-emergent referral rates fell from 30% in 2022 to 10% in 2024 following teleconsultations. Emergent stroke teleconsultations allowed thrombolytic therapy within the golden hour in 83% of cases. At-home ECG monitoring helped detect atrial fibrillation in 25% of cryptogenic stroke patients within 2 weeks, ensuring timely recall and initiation of appropriate antiarrhythmic therapy to prevent recurrent stroke. Surveys indicated that 83% of healthcare providers and patients were satisfied with telehealth services.

Discussion: The single-centre study showcases a multidisciplinary integrated telehealth model. However, confounders existed, including changes in the healthcare system, selection bias and technology disparities. Satisfaction data may be biased. The short timeframe precludes long-term analysis, underscoring the need for broader, controlled studies to assess the sustained impact of telehealth.

Conclusion: The integrated telehealth centre model provides a scalable and replicable approach for healthcare delivery. Studies for long-term benefits and outcomes will help improve telehealth models.

Study objectives: Chronic wounds represent a significant economic and personal burden. For their successful treatment, the causes must be known and treated. Wounds caused by pyoderma gangrenosum (PG), a rare inflammatory skin disease, are often misdiagnosed. This study, therefore, aims to develop a machine learning model capable of differentiating PG from other wound types, focusing on chronic leg wounds to address this diagnostic challenge.

Methods: We used 3674 wound photographs from three specialised wound centres with the four most common types of foot and leg ulcers and the rare inflammatory differential diagnosis PG. The convolutional neural network classifier ConvNeXt 'B' was pretrained on LAION2B, ImageNet12k and ImageNet 1k before being trained and fine-tuned on an 85:15 train, validation split.

Results: The model achieved an overall high accuracy in multiclass classification of the chronic wounds (unbalanced accuracy 90%, balanced accuracy 87%). The sensitivity for identifying PG was 94%, while the sensitivity forother chronic wound types was 97% for diabetic foot ulcers (DFU), 92% for venous leg ulcers (VLU), 78% for mixed leg ulcers and 74% for arterial leg ulcers.

Discussion: The machine learning model effectively differentiates PG from the most common leg and foot ulcers and was very accurate for classifying DFU and VLU. A higher rate of misclassifications occurred for the other vascular ulcers, that is, mixed and arterial leg ulcers. This aligns with the challenges in clinical practice.

Conclusion: Despite the limited number of wound images, this novel multiclass wound classification model accurately identified PG and differentiated leg and foot ulcer subtypes, providing a foundation for a diagnostic support system.

Objective: The increasing burden of hypertension in Africa underscores the need to embrace digital health innovations to improve delivery and access to quality hypertension care. This review aimed at (1) identifying barriers and facilitators to the implementation and uptake of digital health tools and (2) examining the scope and use of digital health tools based on the Practical Reviews in Self-Management Support (PRISMS) taxonomy.

Materials and methods: We searched PubMed, CINAHL (Medline) and HINARI from inception to June 2024. The Joanna Briggs Institute (Population, Concept, and Context (PCC)) framework guided the formulation of research questions, and the PRISMS taxonomy was used to analyse the functions of digital tools.

Results: Sixteen studies (k=16) across three African regions were included. Common digital health tools were mobile Health (mHealth) or electronic Health applications and short message service (SMS)-based interventions. Supported self-management functions included medication adherence (k=10), lifestyle counselling (k=12) and home blood pressure monitoring (k=9). Implementation strategies included prior training (k=10), continuous digital support (k=10) and provision of resources including BP devices and data credit (k=11). Targeted users were healthcare workers (k=8), patients (k=11) or both (k=3). Barriers included limited digital literacy, poor communication among healthcare workers, privacy concerns and weak internet infrastructure. Facilitators included competency-based training, contextual adaptations, continuous technical support and enhanced user experience through effective feedback systems between users.

Conclusion: Digital health tools, particularly mHealth apps and SMS, support key hypertension self-management tasks in African settings. Addressing technological and contextual barriers while reinforcing training and support systems is critical to successful implementation and scale-up.

Objectives: This study developed clinical evidence-based pathways (CEBPWs) to standardise treatment protocols, align diagnosis-reimbursement criteria, detect upcoding and enable early overtreatment warnings.

Methods: The CEBPWs were developed based on hospitalised patient-level data from 1 January 2022 to 31 June 2024 in 166 public hospitals in 16 administrative districts of Shanghai. It includes a total of 5 319 336 cases, involving 3 688 108 groups of 'diagnosis+therapy'. 2.61 billion records of hospitalisation charges and 876.45 million records of outpatient charges were collected. GROWTH algorithm was used to find the combination of frequently charged items for examination, treatment, drugs and devices in 'diagnosis+therapy' group.

Results: CEBPWs comprise five key elements: objective evidence identification, accurate classification, value weighting, frequency weighting and temporal sequencing of evidence. We applied CEBPWs to 224 diseases, detecting issues including upcoding, overtreatment and fragmented care episodes to enhance healthcare quality. CEBPWs achieve 100% coverage in diagnostics, therapy and consumables, with 81.81% drug coverage. The pilot evaluation showed that there were violations in 433 cases, with a frequency deviation of 8.64% and cost deviation of 10.82%, with 8.95% for diagnosis, 9.44% for therapy, 14.81% for drugs and 8.98% for consumables.

Discussion: We were developed CEBPWs, breaking the limitations of the clinical pathways is that the experience of clinical experts rather than objective criterion based on the characteristics of big data and lack of diagnostic and therapy standards integrated with payment standards.

Conclusion: The results indicate that CEBPW is critical tool for hospital management and regulation, address many drawbacks of clinical pathways.

Objectives: Due to the big disease burden of undiagnosed chronic obstructive pulmonary disease (COPD), we aimed to investigate the differences in the characteristics and risk factors of patients with undiagnosed COPD in China.

Methods: We used data from the 'Happy Breathing' Programme through April 2023. Current study is a cohort design. Participants were divided into high risk, undiagnosed and diagnosed COPD. Univariate logistic regression, lasso regression, decision tree, random forest and gradient boosting machine were used to screen the variables. Comparisons were conducted between undiagnosed and patients with diagnosed COPD.

Results: A total of 1603 high-risk, 4688 undiagnosed and 1634 patients with diagnosed COPD were identified. Patients with undiagnosed COPD had the lowest level of education, the poorest COPD-related knowledge and most biofuel users compared with high-risk populations and diagnosed patients (p<0.001). After multivariable logistic regression analysis, COPD-related knowledge score (OR=0.96, 95% CI 0.95 to 0.97), COPD Assessment Test Score (OR=1.01, 95% CI 1.00 to 1.02) and modified Medical Research Council Dyspnea Scale (OR=1.26, 95% CI 1.14 to 1.39) remained significant. Analysis of follow-up data showed that patients with undiagnosed COPD had lighter symptoms and experienced less acute exacerbations than diagnosed patients (p<0.001).

Discussion: Most patients with COPD remain undiagnosed until they feel dyspnoea or hospitalisation due to acute exacerbation. Undiagnosed COPD contributes significantly to the disease burden.

Conclusion: In China, patients with undiagnosed COPD were poorly educated, consumed more biofuels, smoked more and had limited COPD-related knowledge. Patients with undiagnosed COPD are also at risk of acute exacerbation.

Trial registration number: NCT04318912.

Objectives: This study aimed to evaluate the feasibility of using a large language model (LLM) to generate patient information leaflets (PILs) with improved readability based on PILs in the field of interventional radiology.

Methods: PILs were acquired from the Cardiovascular and Interventional Radiology Society of Europe website, reformatted, and uploaded to the GPT-4 user interface with a prompt aimed to simplify the language. Automated readability metrics were used to evaluate the readability of original and LLM-modified PILs. Factual accuracy was assessed by human evaluation from three consultant interventional radiologists using an agreed marking scheme.

Results: LLM-modified PILs had significantly lower mean reading grade (9.5±0.5) compared with original PILs (11.1±0.1) (p<0.01). However, the recommended reading grade of 6 (expected to be understood by 11- to 12-year-old children) was not achieved. Human evaluation revealed that most LLM-modified PILs had minor concerns regarding factual accuracy, but no errors that could result in serious patient harm were detected.

Discussion: LLMs appear to be a powerful tool in improving the readability of PILs within the field of interventional radiology. However, clinical experts are still required in PIL development to ensure the factual accuracy of these augmented documents is not compromised.

Conclusion: LLMs should be considered as a useful tool to assist with the development and revision of PILs in the field of interventional radiology.