BMJ Health & Care Informatics最新文献

We present BODHI (Balanced, Open-minded, Diagnostic, Humble, and Inquisitive), an engineering framework for curiosity driven and humble clinical decision support artificial intelligence (AI) systems. Despite growing capabilities, large language models (LLMs) often express inappropriate confidence, conflating statistical pattern recognition with genuine medical understanding. BODHI addresses this through a dual reflective architecture that: (1) decomposes epistemic uncertainty into task specific dimensions, and (2) constrains model responses using virtue based stance rules derived from a Virtue Activation Matrix. We validate the framework through controlled evaluation on 200 clinical vignettes from HealthBench Hard, assessing GPT-4o-mini and GPT-4.1-mini across 5 random seeds (2000 total observations). Statistical analysis included bootstrap resampling, paired t tests, and effect size computation. BODHI improved overall clinical response quality (GPT-4.1-mini: +16.6 pp, p<0.0001, Cohen's d=11.56; GPT-4o-mini: +2.2 pp, p<0.0001, Cohen's d=1.56) and achieved very large effect sizes on curiosity (context seeking rate: Cohen's d=16.38 and 19.54) and humility (hedging: d=5.80 for GPT-4.1-mini) metrics. Crucially, 97.3% of GPT-4.1-mini responses and 73.5% of GPT-4o-mini responses included appropriate clarifying questions, compared with 7.8% and 0.0% at baseline, demonstrating the framework's effectiveness in eliciting information gathering behaviour. Findings suggest LLMs can be reliably constrained to operate within epistemic boundaries when provided with structured uncertainty decomposition and virtue aligned response rules, offering a pathway towards safer clinical AI deployment.

Objectives: The overwhelmed situation under the COVID-19 pandemic has worsened the quality of emergency medical care and the mortality rate due to out-of-hospital cardiac arrest (OHCA). However, there has been no research conducted for the validation of prognostic prediction models for OHCA patients using data collected during the pandemic. We sought to develop a pre-hospital prediction model for neurological outcome at 1 month in adult patients following OHCA using a machine-learning technique and validate the model for data collected during the pandemic.

Methods: The data of 1 740 212 adult OHCA patients from a nationwide registry in Japan between 2005 and 2019 were used for developing a prediction model. Neurological outcome at 1 month after OHCA was set as the prediction target. We validated the model using 96 525 patient data collected during the pandemic from March to December 2020.

Results: The optimal predictive factors were all ascertained at the emergency scene. Although the neurological outcome was less favourable during the pandemic compared with the corresponding pre-pandemic periods, the model yielded substantially high performance with precise calibration: the area under the receiver operating characteristics curve of 0.94 and 0.95 before and during the pandemic, respectively.

Discussion: The model will improve the quality of emergency care by enabling accurate triage and swift preparation for advanced life-saving care regardless of overwhelmed situations due to disastrous circumstances.

Conclusion: We developed a prediction model for neurological outcome in OHCA patients using machine learning techniques, which was adaptable to the medical situation during the COVID-19 pandemic.

Background: Mindfulness can positively impact physical and mental health, but face-to-face programmes are limited by poor accessibility, availability and cost. Virtual reality (VR) offers immersive audiovisual environments that could improve mindfulness practice.

Objectives: To evaluate commercially available VR apps related to mindfulness.

Methods: App stores and relevant online platforms were searched for VR apps related to mindfulness. Results were screened against eligibility criteria and relevant data extracted. Six raters used the Mobile App Rating Scale (MARS) to assess the quality of VR apps.

Results: Five VR apps related to mindfulness were included, that is, Headspace XR, Hoame, Innerworld, Maloka and TRIPP. These provided access to meditative and mindfulness sessions, guided by virtual instructors in some cases and situated in a range of virtual landscapes accompanied by sound or music. TRIPP received the highest average MARS score (4.1), followed by Hoame (3.8), Maloka (3.6), Headspace XR (3.4) and Innerworld (3.3). Most VR apps scored the highest on functionality (3.4-4.2), while the information category scored the lowest (3.1-3.7). The intraclass correlation was moderate.

Conclusion: This review provides important insights into VR apps related to mindfulness such as their availability and quality. Only five VR apps were identified related to mindfulness practice with an overall moderate MARS quality score (3.62/5.00). These may provide a convenient and immersive way to access and engage in regular mindfulness practice, particularly for novices. Rigorous scientific research should assess the effectiveness of these VR apps in improving physical and mental health through immersive digital mindfulness practice.

Objective: Digital health apps and patient portals are proposed as part of the drive from 'analogue to digital' care for the National Health Service (NHS) 10-Year Plan. Without mitigation strategies, digital inequalities could arise as a result, and more evidence is needed to understand how to mitigate this.

Methods: As part of an equality impact assessment, a retrospective cross-sectional analysis was conducted examining patient portal activation among patients invited to outpatient appointments at two large south-east London Hospital Trusts between 1 May and 1 November 2024.

Results: Of the 503 688 patients invited to attend outpatient clinics during the study period, 52.7% had activated the patient portal. Availability of email contact details was the strongest determinant of onboarding likelihood (OR 10.86). Multivariate logistic regression models showed that the following groups were less likely to activate the patient portal: men (OR 0.84), individuals at the extremes of age (71-80 or 11-20 years), those of mixed or undefined ethnicity (OR 0.58), those of black ethnicity (OR 0.62) and those with the highest degree of socioeconomic deprivation (Index of Multiple Deprivation group 1; OR 0.68).

Conclusion: This large-scale roll-out of a digital health portal provides empirical evidence of factors that drive digital inequalities for patients of two major London NHS Trusts. The observed disparities across demographic and socioeconomic dimensions and simple reliable digital contact mechanisms highlight the risk that digital healthcare initiatives may inadvertently produce new types of inequalities.

Objectives: To evaluate the National Health Service (NHS) Federated Data Platform (FDP) Inpatient (IP) Care Coordination Solution (CCS) digital scheduling tool on elective theatre utilisation.

Methods: An interrupted time series assessed changes in theatre utilisation and cancellations following tool adoption (January 2022). Weekly data spanned 90 weeks (April 2021-December 2023). Outcomes included weekly median theatre utilisation (actual, booked and bookings per session) and the percentage of cancelled bookings. Models incorporated a 5-week lag and estimated level (step-change) and trend (slope) effects.

Results: Postintervention level and trend increases were observed for booked (β=4.40, p=0.045; β=0.26, p=0.002) and actual (β=3.98, p=0.064; β=0.23, p=0.006) utilisation. Bookings per session showed a significant level increase (β=0.34, p=0.002) with no trend change (β=0.00, p=0.790). Across the postintervention period, compared with counterfactual estimates, booked and actual utilisation were 15.0% (95% CI 13.4% to 16.5%, p<0.0001) and 12.2% (95% CI 10.8% to 13.5%, p<0.0001) higher, while bookings per session were 10.9% (95% CI 9.5% to 12.4%, p<0.0001) higher. Significant positive effects were observed for urology, general surgery, gynaecology, plastic surgery and ophthalmology. A significant upward trend in cancellation rates was associated with the introduction of the tool (β=2.1, p=0.001).

Discussion: Findings suggest that centralised digital scheduling tools can improve theatre capacity by enabling more efficient use of existing capacity through improved scheduling visibility. Future research should explore differences in specialty-level usage and long-term sustainability of gains.

Conclusion: The introduction of the NHS FDP IP CCS product was associated with improved elective theatre utilisation.

Objectives: To evaluate the ability of large language models (LLMs) to simulate multidisciplinary team (MDT) decision-making in colorectal cancer, a malignancy that often requires complex treatment planning.

Methods: We retrospectively analysed 1423 colorectal cancer cases discussed at MDT meetings at Peking University Cancer Hospital between January 2023 and December 2024. Three LLMs-OpenAI o3-mini-2025-01-31, DeepSeek-R1 671b and Qwen qwq-plus-2025-03-05-were tested for their ability to replicate MDT recommendations using a standardised treatment categorisation framework. Each case was processed three times per model; only cases with consistent outputs across all three runs were included. Concordance between AI-generated decisions and expert MDT consensus was assessed using agreement percentages and Cohen's kappa.

Results: O3 demonstrated the highest intramodel stability, with an agreement rate of 81.0% (Fleiss' kappa=0.794), yielding 1153 cases with consistent outputs. Concordance with MDT consensus was comparable across the three models, ranging from 62.5% to 65.4%. Multivariable analysis of O3 outputs identified treatment-naïve status, non-metastatic disease and colon tumour location as independent predictors of higher concordance with experts.

Discussion: LLMs showed fair overall agreement with expert MDT decisions, with stronger performance in standardised and less complex clinical scenarios. Areas of higher concordance included treatment-naïve non-metastatic colon cancer, treated non-metastatic rectal cancer and treated non-metastatic colon cancer.

Conclusion: LLMs can partially replicate expert MDT recommendations in colorectal cancer. Their integration into clinical workflows should aim to complement, rather than replace, human expertise.

Objectives: To explore the effects of bidirectional interoperability between electronic health records (EHR) and smart infusion pumps on medication errors (MEs), system compliance and workflow efficiency and economic aspects.

Methods: This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 criteria. The literature search on Scopus, MEDLINE (Ovid), Web of Science, Cumulated Index in Nursing and Allied Health Literature, and Evidence-Based Medicine Reviews was conducted on 3 October 2024. Peer-reviewed studies considering bidirectional interoperability between EHR and smart pumps in hospitals were included. Study selection according to a predetermined Population, Intervention, Comparison(s) and Outcome tool, data extraction and evidence quality assessment (Joanna Briggs Institute critical appraisal tool assessment and the Grading of Recommendations Assessment, Development and Evaluation approach) were carried out by two individual reviewers.

Results: Seven studies from the USA, published between 2011 and 2024, were included. The studies used variable designs to compare the effects of bidirectional interoperability between smart infusion pumps and EHR on system compliance and workflow efficiency (n=6 studies), MEs (n=3) and economic outcomes (n=2) before and after implementation. The observed effects were mainly positive; however, evidence quality was low because of the observational nature of studies.

Discussion: The interoperability between EHR systems and smart infusion pumps remains a relatively novel research topic. Evidence is geographically concentrated, limiting its generalisability to different healthcare systems, regulatory environments and technology adoption patterns.

Conclusion: While bidirectional interoperability may reduce MEs, improve system compliance and workflow efficiency and enhance hospitals' charging accuracy of provided care, future studies should prioritise controlled designs, robust data and economic outcomes to justify the investment.

Prospero registration number: CRD42024538518.

Objectives: To develop recommendations to inform development and integration of predictive digital health and artificial intelligence tools in primary care.

Methods: Recommendation development involved two stages. The initial scoping phase comprised an umbrella review to identify barriers to implementation for risk prediction tools in primary care. The consensus phase involved a stakeholder workshop with 22 stakeholders. The draft recommendations were then refined via a stakeholder survey completed by 13 participants and three online meetings attended by 14 individuals to generate the final output.

Results: The umbrella review included 12 reviews and identified 15 barriers to implementation of risk prediction models, including lack of integration with electronic health records and poor interoperability across them. The final recommendations include 14 core features of risk prediction models and tools, including the need for codesign with clinicians and the public and integration with digital infrastructure and workflows.

Discussion: These findings particularly emphasise the value of early engagement with key stakeholders and health record system providers, and a need for shared understanding of the needs of end-users.

Conclusions: We have developed recommendations detailing 14 key characteristics for a digital risk prediction model to be successfully used in primary care settings. This profile should be used to guide development of new risk prediction tools and is also applicable more widely to other digital health innovations within primary care. Future research should work to resolve the identified system-level barriers to implementation.

Objectives: To develop and evaluate a machine learning (ML) model that predicts Crohn's disease (CD) patients responsible for the top quartile of healthcare spending.

Methods: De-identified commercial claims (2016-2018) from ~267 000 continuously enrolled members in a Midwestern state were analysed, including 994 CD cases. Monthly data for each patient was aggregated into data points that included healthcare spending amounts, encounter interactions, demographics and binary flags for diagnoses, procedures and drug codes. Seven algorithm families were tuned using five-fold cross-validation (January 2016 to September 2017) and tested prospectively (November 2017 to February 2018). Monthly performance evaluations assessed the accuracy of predicting high-cost healthcare spending, using 4-month and 1-month historical cost analyses for comparison.

Results: ML models predicted an average of 80% of the dollars spent by top-quartile members during the 4-month evaluation period, compared with 67% for the 4-month baseline and 62% for the prior-month benchmark. The models identified an average of 51 new members entering the high-cost group each month, nearly double the yield of the 4-month historical method. These ML models more accurately anticipated inpatient encounters that drove excess spending.

Discussion: Claims-based ML offers actionable lead time for payers and clinicians to enhance monitoring, adjust biological therapy or schedule elective care before emergency admissions occur. Because this framework relies exclusively on standard claim fields, it can be quickly extended to other episodic, high-variance conditions.

Conclusion: Prospectively tested, claims-only ML models enhance short-term risk stratification in CD by identifying future high-cost patients. Future studies should confirm the clinical impact, cost savings and ensure equitable performance across diverse populations.

Objectives: Interoperability, the seamless exchange and use of data across digital health systems, is essential for integrated, efficient healthcare delivery. However, evidence on its adoption in Africa remains limited and fragmented. This scoping review aimed to map existing evidence, identify key barriers and highlight emerging opportunities for strengthening interoperability across all levels on the continent.

Methods: We conducted the review in line with the Joanna Briggs Institute (JBI) methodology and the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) framework. Searches were carried out across PubMed/MEDLINE, IEEE Xplore, African Index Medicus and Google Scholar, focusing on English-language publications from January 2010 to March 2025. Eligible sources included peer-reviewed articles, conference papers and relevant policy documents.

Results: Sixteen studies met the inclusion criteria. The findings revealed wide disparities in the adoption of interoperability standards, with countries such as Uganda, South Africa and Kenya showing greater momentum due to national digital strategies and health information exchange initiatives. Common challenges included limited technical capacity, fragmented infrastructure and inadequate regulatory support. However, there were encouraging developments around the use of open-source platforms like OpenHIE, regional policy alignment through the African Union Digital Health Strategy and growing public-private partnerships.

Discussion: Progress remains uneven, shaped by each country's digital maturity, workforce capabilities and policy landscape. Capacity-building and better alignment with global standards could bridge current gaps.

Conclusion: To build resilient digital health systems, African countries must strengthen governance, invest in infrastructure and develop technical expertise. Future work should assess how interoperability influences clinical care and explore regional readiness for cross-border data exchange.