BMJ Health & Care Informatics最新文献

Objectives: To evaluate provider-level variability across the full perioperative workflow using a computer vision-based artificial intelligence (AI) system that automatically detects and timestamps operating room events.

Methods: A cross-sectional study of total knee arthroplasty cases performed between September 2022 and March 2025 at a regional health system was conducted. An ambient surgical platform equipped with wall-mounted cameras continuously captured perioperative activity. A YOLO-based model identified patients, staff and equipment, and a transformer-based event detector predicted key perioperative events in real time. Detected events were used to segment cases into eight workflow phases: anaesthesia induction, patient preparation, final preparation, active procedure, postoperation, patient exit, room cleanup and room setup. Provider-level variability in segment durations was evaluated after adjusting for case characteristics, daily surgical volume and team composition.

Results: The computer vision event detection system achieved high agreement with ground truth annotations. Across 2502 surgical cases, significant provider-level variability was observed in all workflow segments except for room exit. Active procedure showed the greatest variation among surgeons (F=28.4, p<0.001; β IQR=-20.9 to 8.8 min) followed by room setup among circulating nurses (F=1.3, p<0.001; β IQR=-5.2 to 4.4 min) and room setup among scrub nurses (F=1.4, p<0.001; β IQR=-3.7 to 3.2 min).

Conclusions: Automated workflow segmentation using computer vision provides a scalable method to evaluate perioperative efficiency with greater granularity. Broader case segmentation may support more targeted and effective surgical quality improvement initiatives.

Objective: This study aims to develop an algorithm to detect steatotic liver disease (SLD) risk in low-resource settings without requiring imaging.

Methods: This retrospective cohort study included 826 measurements from 444 participants aged 45-60 years who participated in the MAUCO+ study. Data included ultrasound, vibration-controlled transient elastography (VCTE), anthropometrics and biomarkers. Logistic multivariable regression was used to develop two predictive models for SLD risk, with and without ultrasound, using VCTE as gold standard. Missing data were minimal and retained in the analysis, as their proportion was not statistically relevant. Predictive performance (sensitivity, specificity, positive predictive value and negative predictive value) was compared with the clinically used Fatty Liver Index (FLI).

Results: The algorithm without ultrasound achieved a sensitivity of 81.1% (95% CI 71.7% to 88.4%) and specificity of 71.4% (95% CI 57.9% to 80.4%). The model with ultrasound demonstrated a sensitivity of 91.5% (95% CI 84.1% to 95.6%) and specificity of 70% (95% CI 59.9% to 80.7%). FLI showed an area under the curve (AUC) of 0.762, while our models achieved higher AUCs: 0.878 (with ultrasound) and 0.794 (without ultrasound).

Discussion: Our models offer screening tools for SLD in low-resource primary care. The model without ultrasound outperformed FLI, making it a feasible alternative where imaging is unavailable. The ultrasound-based model demonstrated higher performance, underscoring the value of ultrasound when it is accessible. Integrating these algorithms into preventive programmes could improve early diagnosis, especially in populations with a high burden of obesity and diabetes.

Conclusions: We developed two predictive models for SLD screening in a Chilean cohort. Both showed strong performance and potential for implementation in primary care to support early detection and better disease management.

Objectives: Little research focuses on mechanisms underlying the well-recognised relationship between mental and physical health, or its potential to influence adherence and response to treatments. This short report summarises results of the National Institute for Health and Care Research-funded 'Core Mental Health Data Set (CMHDS)' study to embed a digital tool for routine collection of mental health data in physical health studies.

Methods: Four chief investigators of physical health trials were approached to embed the CMHDS into their study. Two trials, one for people receiving specialist cystic fibrosis (CF) care, and the established Salford Kidney Study (SKS) successfully managed to embed CMHDS.

Results: A combined 478 participants from both studies were invited to complete the CMHDS. Of those approached, 88% agreed to complete CMHDS; 44% completed it. In the SKS, people who completed CMHDS were significantly younger and had higher estimated glomerular filtration rates and were from least deprived areas. In the CF study, there was no significant difference in characteristics of participants who did or did not complete the tool.

Discussion: It was feasible, and researchers and participants considered it acceptable, to embed the CMHDS in physical health studies as part of routine data collection.

Conclusion: Future studies should embed the CMHDS routinely and encourage completion to minimise bias and optimise the added value of having mental health covariates or predictor variables in physical health studies.

Introduction: Emergency departments are facing increasing strain due to overcrowding and resource shortages, leading to the suspension of some services. Stratifying the clinical risk-defined as the severity and likelihood of harm-is crucial for anticipating care needs and supporting decision-making. Implementing predictive models for clinical risk management offers a technological solution to this challenge. This systematic review will evaluate the performance and usefulness of a predictive model for managing the clinical risk of people who visit the emergency department.

Methods and analysis: Eight electronic databases will be searched (CINAHL Plus, Health Technology Assessment Database, MedicLatina, MEDLINE, PubMed, Scopus, Cochrane Plus Collection, Web of Science). Risk of bias will be assessed using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies and Prediction Model Risk of Bias Assessment Tool.

Ethics and dissemination: Ethical approval is not required. Results will be disseminated through peer-reviewed publications.

Prospero registration number: CRD42024556926.

Background: Monitoring early childhood growth is vital, as growth faltering could indicate nutritional or health issues requiring prompt intervention. Our study's aim was to assess the performance of a length-weight artificial intelligence (LWAI) tool for predicting children's length and weight from smartphone images.

Methods: This observational, single-centre study recruited children aged 0-18 months. Investigators measured length and weight in clinic using WHO standard recommendations and captured six images per child in a supine position, while parents took six similar images at home. Within each image, LWAI identifies specific body landmarks and a reference object, then extracts and uses image features to predict the child's length and weight. The LWAI's performance was assessed by comparing length/weight prediction versus actual measurements. User experience was collected through questionnaires.

Results: A total of 215 participants (mean age 6.1 months) were included, and length/weight predictions were generated for 98% (2184/2224) of the images. The mean absolute error (MAE) and mean absolute percentage error (MAPE) for length were 2.47 cm (4.04%) for individual images and 1.89 cm (3.18%) for grouped images (participants with ≥9 images). The corresponding MAE/MAPE for weight were 0.69 kg (11.68%) and 0.56 kg (9.02%), respectively. Regarding usability, 97% of parents who reported not routinely measuring their child's growth indicated that they would start doing so regularly if a digital tool was available to them.

Conclusions: The LWAI tool can predict length and weight in children ≤18 months, offering a practical, convenient, artificial intelligence-powered alternative for growth monitoring in home and clinical settings.

Trial registration number: NCT05079776.

Background: Advance directives (AD) are crucial for aligning healthcare with end-of-life preferences, yet documentation rates remain low, often only 5-15%. Leveraging electronic health records (EHRs) for automated outreach may offer a promising strategy to enhance AD completion without placing additional burdens on already busy clinicians.

Methods: We evaluated the feasibility and effectiveness of EHR-based AD outreach within the Yale New Haven Health System (YNHHS). In April 2024, a targeted message was sent via Epic's MyChart over seven business days, coinciding with National Healthcare Decisions Day. Patients aged ≥65 years with an active MyChart, no existing AD documentation and at least one primary care visit within 2 years were eligible; those hospitalised or in hospice were excluded. The message provided education about advance care planning, encouraged completion of a Healthcare Representative Form and/or Living Will Form and offered instructions for uploading documents directly to the EHR or returning them to a primary care provider's office. A reminder was sent 90 days later.

Results: Outreach reached 25 571 patients, with 61% viewing the MyChart message. Six months after intervention, AD completion across YNHHS rose from 39.9% (28 324/70 911) to 42.8% (30 230/70 583), translating to a 7.5% conversion rate in the targeted cohort. There was no observed increase in patient messaging or clinical staff workload.

Conclusion: These findings suggest that EHR-integrated campaigns can effectively increase AD documentation among older adults without straining providers. By prompting patients to complete forms at their convenience, this scalable and sustainable intervention may be adapted for wider populations and other preventive or chronic care needs.

Objectives: Critical incident reporting systems (CIRS) collect narrative reports on medical errors, but emotional signals within these reports, potential indicators of perceived risk and systemic weakness, are rarely examined. This cross-sectional study applied large language model-based sentiment analysis to explore how emotional expression in CIRS data may support artificial intelligence-enhanced patient safety monitoring.

Methods: We analysed 11 056 anonymised German incident reports submitted between 2005 and 2024 using GPT-4 (Generative Pre-trained Transformer 4, gpt-4-turbo-2024-04-09, zero shot) to assign sentiment labels and quantify five emotions (fear, frustration, anger, sadness, guilt; scale 0-1). Emotional profiles were clustered (k-means) and thematic patterns extracted via Latent Dirichlet allocation. Associations were examined using non-parametric tests.

Results: Negative sentiment dominated (95.6%, 95% CI 94.9% to 96.2%). Fear (mean=0.63, SD=0.21) and frustration (mean=0.59, SD=0.19) were most prevalent. Emergency care settings showed higher fear (p<0.05) and guilt (p<0.001). Reports with strong emotional expression, especially fear, guilt and sadness, were less likely to receive formal feedback (43.1% (95% CI 41.7% to 44.5%) vs 48.1% (95% CI 46.5% to 49.7%); absolute difference=5.0 percentage points (95% CI 2.7 to 7.3); p=0.001).

Discussion: Emotion intensity did not consistently correlate with harm severity but was linked to care context and systemic complexity. Emotion clusters reflected distinct clinical and organisational patterns, from acute emergencies to procedural failures.

Conclusion: Emotion-based analysis of incident reports provides insight into perceived burden and care context. Sentiment profiling may improve system interpretability and support emotion-sensitive safety culture and feedback. Leveraging large language models can reduce reviewer workload and enable more targeted triage of emotionally complex reports.

IntroductionRelapse remains a major challenge in the treatment of substance use disorders (SUDs), particularly during follow-up. Digital tools are emerging as supportive resources, but few deliver real-time interventions. This study will examine the effectiveness of a digital relapse prevention plan (DRPP) integrated into a certified mobile application to detect early risk signs and provide immediate, personalised responses.Methods and analysisA multicentre randomised controlled trial will recruit adults with SUD. Participants will be randomised to standard treatment plus a restricted app version (control) or the same treatment with the full app, including automated alerts and DRPP access (experimental). The plan can be activated manually or automatically through smartphone sensors detecting risk patterns. The primary outcome will be time to first clinical relapse, while secondary outcomes will include patient satisfaction with the DRPP, adherence and perceived emotional self-regulation. Findings are expected to provide robust evidence on the feasibility, acceptability and clinical utility of digital relapse prevention strategies.Ethics and disseminationThis study obtained ethical approval (code 25/327) from Committee of Hospital Universitario 12 de Octubre.Trial registration number:NCT07052175.

Objectives: Postoperative complications (PCs) require substantial resources to manage and are cumbersome to monitor. Artificial intelligence (AI), particularly natural language processing (NLP), offers a potential solution by automating and streamlining these processes, but perceived PC rates may differ depending on model optimisation strategies. This study aimed to develop a mock-up AI-driven automated registry for PCs. We hypothesised that using NLP to obtain longitudinal overviews of key quality metrics is feasible, but that optimisation strategies impacted the observed rate of PCs.

Methods: We analysed 100 505 surgical cases from 12 Danish hospitals between 2017 and 2021. Previously validated NLP models were applied to detect seven types of PCs, using two different threshold settings: a set of thresholds optimised for positive predictive value (precision), referred to as F-score of 0.5, and a set of thresholds optimised for sensitivity, referred to as F-score of 2. Trends in PC rates over time were assessed, and hospital-level variations were examined using logistic regression models.

Results: The NLP models detected 8512 or 15 892 PCs, depending on threshold selection, corresponding to total PC rates of 9.14% and 17.1%, respectively. Most PCs showed stable or increasing trends over time, regardless of threshold setting. Regression analyses demonstrated that threshold selection significantly influenced findings, impacting hospital comparisons.

Conclusion: We demonstrate that NLP can be used for automated PC detection. However, threshold selection and additional performance metrics must be carefully considered.