BMJ Health & Care Informatics最新文献

Objectives: Clinical trial enrolment is critical for the development and approval of novel cancer therapeutics, but patient identification and recruitment to clinical trials remains low and multiple trials accrue slowly or fail to meet accrual goals. Informatics solutions may facilitate clinical trial screening, ideally improving patient engagement and enrolment. Our objective is to develop and implement a system to efficiently screen queried patients for available clinical trials.

Methods: At Stanford, we designed and implemented a personalised clinical trial matching system, integrating our electronic medical record, clinical trials management system and a third-party software-based solution to directly connect providers with clinical research coordinators and appropriate trials.

Results: Over 3 years of a staged rollout, significant increases in clinical trial screening requests and subsequent enrolment have been observed. The total number of screening referrals increased from 20 in the first year to 236 in the third year. Enrolment related to screening referrals, the 'conversion rate', ranged from 16% to 26% of referred patients.

Conclusion: Clinical trial matching systems can increase awareness of available trials and provide a mechanism to increase clinical trial accrual, especially when implemented at the point of care for easy access at treatment decision points. Here, we describe the process of creating and implementing a bespoke clinical trial matching software integrated into the electronic medical record. Having validated the utility of the platform, we will focus on further efforts to drive utilisation through software features.

Clinical diagnostic tools can disadvantage subgroups due to poor model generalisability, which can be caused by unrepresentative training data. Practical deployment solutions to mitigate harm for subgroups from models with differential performance have yet to be established. This paper will build on existing work that considers a selective deployment approach where poorly performing subgroups are excluded from deployments. Alternatively, the proposed 'mitigated deployment' strategy requires safety nets to be built into clinical workflows to safeguard under-represented groups in a universal deployment. This approach relies on human-artificial intelligence collaboration and postmarket evaluation to continually improve model performance across subgroups with real-world data. Using a real-world case study, the benefits and limitations of mitigated deployment are explored. This will add to the tools available to healthcare organisations when considering how to safely deploy models with differential performance across subgroups.

Objectives: In the ever-evolving landscape of healthcare, the integration of digital systems and medical devices is increasingly important for modernising healthcare delivery. However, the acceptance and adoption of emerging technologies by healthcare staff present challenges. The purpose of this research was to apply relevant knowledge to inform and improve a conceptual framework (ARC): early exploration and agile adoption of emerging healthcare technology. We report on an expert-led Delphi study to evaluate consensus regarding the framework.

Method: The ARC conceptual framework, presented as four successive phases: imagine, educate, validate and score, was evaluated by 23 experts over two rounds. Experts first agreed/disagreed with 31 enabling statements relating to the early exploration and evaluation of new technology. The expert panel made recommendations (n=20), which were incorporated into round 2 with a checklist to evaluate the potential of a new technology.

Results: All participating experts completed round 1, and 13 completed round 2. Consensus (defined as >75% agreement) was achieved for 93.4% (n=57) of statements, with consensus without exception achieved for 34.4% (n=21) items and 16 new items added to the improved ARC framework, including on the appropriate use of simulation studies.

Discussion: The main findings highlight the importance of demonstration spaces, time in clinical environments with clinical teams, data-driven benefits and structured debriefs with staff.

Conclusion: A Delphi approach achieved expert consensus regarding the ARC framework for engaging with new technology and preparing the healthcare workforce for its use. Further advocacy is required to negotiate stakeholder involvement and interdisciplinary cooperation.

Objectives: To investigate whether machine learning (ML) algorithms can learn racial or ethnic information from the vital signs alone.

Methods: A retrospective cohort study of critically ill patients between 2014 and 2015 from the multicentre eICU-CRD critical care database involving 335 intensive care units in 208 US hospitals, containing 200 859 admissions. We extracted 10 763 critical care admissions of patients aged 18 and over, alive during the first 24 hours after admission, with recorded race or ethnicity as well as at least two measurements of heart rate, oxygen saturation, respiratory rate and blood pressure. Pairs of subgroups were matched based on age, gender, admission diagnosis and disease severity. XGBoost, Random Forest and Logistic Regression algorithms were used to predict recorded race or ethnicity based on the values of vital signs.

Results: Models derived from only four vital signs can predict patients' recorded race or ethnicity with an area under the curve (AUC) of 0.74 (±0.030) between White and Black patients, AUC of 0.74 (±0.030) between Hispanic and Black patients and AUC of 0.67 (±0.072) between Hispanic and White patients, even when controlling for known factors. There were very small, but statistically significant differences between heart rate, oxygen saturation and blood pressure, but not respiration rate and invasively measured oxygen saturation.

Discussion: ML algorithms can extract racial or ethnicity information from vital signs alone across diverse patient populations, even when controlling for known biases such as pulse oximetry variations and comorbidities. The model correctly classified the race or ethnicity in two out of three patients, indicating that this outcome is not random.

Conclusion: Vital signs embed racial information that can be learnt by ML algorithms, posing a significant risk to equitable clinical decision-making. Mitigating measures might be challenging, considering the fundamental role of vital signs in clinical decision-making.

Introduction: Multimorbidity (MM), defined as two or more chronic diseases in an individual, is linked to adverse outcomes. MM is increasing in sub-Saharan Africa due to rapidly advancing epidemiological and social transitions. The Multimorbidity in Africa: Digital Innovation, Visualisation and Application Research Hub (MADIVA) aims to address MM by developing data science solutions informed by stakeholder engagement.

Methods and analysis: MADIVA uses complex, individual-level datasets from research centres in rural Bushbuckridge, South Africa and urban Nairobi, Kenya. These datasets will be harmonised, linked and curated, and then used to develop MM risk prediction models, novel data science methods and interactive dashboards for research and clinical use. Pilot projects and mentorship programmes will support data science capacity development.

Ethics and dissemination: Ethics approval has been granted. Dissemination will occur through scientific meetings and publications. MADIVA is committed to making data FAIR: findable, accessible, interoperable and reusable.

Objective: To evaluate the effectiveness of an online Patient Decision Aid (PtDA) for patients with Generalised Anxiety Disorder (GAD).

Design: Randomised controlled trial comparing the PtDA to general information (fact sheet).

Setting: The study took place in 17 primary care centres in the Canary Islands (Spain).

Participants: Patients diagnosed with GAD and a score ≥8 in the GAD-7 questionnaire.

Intervention: Patients were randomly allocated to the PtDA group (n=58) or the control group (n=61).

Main outcome measure: The primary outcome was decisional conflict at postintervention, assessed with the Decisional Conflict Scale (DCS). Secondary outcomes include knowledge about GAD and its treatments, concordance between informed preference and 3 month actual choice, decision quality and GAD symptoms.

Results: There were no significant differences in decisional conflict at postintervention or 3 month follow-up in the intention-to-treat (ITT) or per-protocol sample (PPS). The PtDA significantly improved postintervention (MD=1.65, 95% CI: 0.84 to 2.46) and 3 month objective knowledge (MD=0.78, 95% CI: 0.02 to 1.55). In the PPS, anxiety symptoms at 3 months were significantly lower in the PtDA group (MD=-3.00, 95% CI: -5.69 to -0.30), but in the ITT sample, this difference did not reach significance (p=0.06). There were no significant differences in the rate of patients unsure about treatment preference at postintervention, nor on concordance or decision quality at follow-up.

Conclusion: The use of the PtDA led to improvements in knowledge at 3 months, but it did not result in a significant reduction of decisional conflict. These results must be interpreted with caution, given the methodological limitations of the study, mainly the high rate of dropouts. Further research is needed to confirm these results, the first published on the effectiveness of a PtDA for GAD patients.

Trial registration number: NCT04364958.

Objectives: To investigate the association between perioperative dexmedetomidine (DEX) use and oncological outcomes-including locoregional recurrence (LRR) and distant metastasis (DM)-in patients undergoing curative surgery for oral cavity squamous cell carcinoma (OCSCC).

Methods: This retrospective cohort study used data from the Taiwan Cancer Registry Database and included patients with stage I-IVB OCSCC who underwent curative surgery between 2007 and 2019. Patients were categorised by DEX exposure status and matched 1:1 using propensity score matching (PSM) based on key clinical and demographic variables. Cox proportional hazards models and competing risk analyses were used to estimate the association between DEX use and oncological outcomes.

Results: After PSM, 8024 patients (4012 per group) were included. Multivariable Cox regression showed that perioperative DEX use was significantly associated with increased risks of LRR (adjusted HR (aHR) 1.67; 95% CI 1.55 to 1.80; p<0.001) and DM (aHR 1.30; 95% CI 1.19 to 1.42; p<0.001).

Discussion: These findings suggest a potential oncological risk associated with perioperative DEX administration. Possible mechanisms include immune modulation and enhanced metastatic potential, as reported in preclinical studies. Further investigation is needed to clarify causal pathways and identify patient subgroups most affected.

Conclusions: Perioperative DEX use is independently associated with increased risks of LRR and DM in OCSCC patients. These results underscore the importance of cautious perioperative management and the need for prospective validation in randomised clinical trials.

The Integrated Disease Surveillance and Response (IDSR) framework, introduced by the WHO in 1998, aimed to unify disease surveillance across West Africa, replacing fragmented systems. However, challenges such as limited real-time reporting, inadequate data collection and workforce shortages continue to impede disease control and outbreak response. The resurgence of infectious diseases like Ebola, cholera, COVID-19 and monkeypox highlights the need to strengthen IDSR systems for effective public health management. This article reviews IDSR implementation in West Africa, identifying persistent gaps, including delayed outbreak detection, limited laboratory capacity and weak surveillance infrastructure. It emphasises the importance of policy development, capacity building and stakeholder engagement to secure political support and resources. Integrating technological innovations-such as mobile health (mHealth), geographic information systems (GIS), electronic health records and big data analytics-can enhance real-time data sharing and response coordination. Strengthening laboratories, workforce training and monitoring frameworks is essential to improve IDSR performance. Strategic investments are crucial to bolster public health capacities, accelerate response times and mitigate future epidemics in West Africa.

Objectives: To identify and characterise distinct subgroups of patients with asthma with severe acute exacerbations (AEs) by using a multistep clustering methodology that combines supervised and unsupervised machine learning.

Methods: This cohort study used anonymised, all-payer medical and prescription US claim data from October 2015 to May 2022. First, gradient-boosted decision trees were trained to predict AE in 4 132 973 patients with asthma, of whom 86 735 experienced AE. This model was applied to a holdout set of 86 434 patients with asthma with AE to derive SHapley Additive exPlanations (SHAP) values. SHAP values were then subjected to non-linear dimensionality reduction and density-based clustering to identify distinct subgroups among these patients. These subgroups were described using key clinical and demographic characteristics.

Results: Clustering identified five distinct subgroups of patients with asthma with AE, broadly differentiated by histories of acute care encounters, healthcare utilisation, AE treatments, coded asthma severity, specialist encounters, first-hand tobacco exposure, mood disorders and patient demographics. Notably, there was considerable between-cluster variability in the predicted likelihood of AE, with some subgroups comprised of patients who posed a challenge for the predictive model and would have been missed with predictive modelling alone.

Discussion: By identifying distinct subgroups among patients with asthma experiencing AE, this study highlights the heterogeneity within this population and emphasises the need for more personalised management of AE.

Conclusion: Applying predictive modelling and clustering to real-world data can help identify discrete phenotypes of patients and offer an important source of information for developing risk assessment and mitigation efforts.

Objectives: The objective of this study is to evaluate whether large language models (LLMs) can achieve performance comparable to expert-developed deep neural networks in detecting flow starvation (FS) asynchronies during mechanical ventilation.

Methods: Popular LLMs (GPT-4, Claude-3.5, Gemini-1.5, DeepSeek-R1) were tested on a dataset of 6500 airway pressure cycles from 28 patients, classifying breaths into three FS categories. They were also tasked with generating executable code for one-dimensional convolutional neural network (CNN-1D) and Long Short-Term Memory networks. Model performances were assessed using repeated holdout validation and compared with expert-developed models.

Results: LLMs performed poorly in direct FS classification (accuracy: GPT-4: 0.497; Claude-3.5: 0.627; Gemini-1.5: 0.544, DeepSeek-R1: 0.520). However, Claude-3.5-generated CNN-1D code achieved the highest accuracy (0.902 (0.899-0.906)), outperforming expert-developed models.

Discussion: LLMs demonstrated limited capability in direct classification but excelled in generating effective neural network models with minimal human intervention. This suggests LLMs' potential in accelerating model development for clinical applications, particularly for detecting patient-ventilator asynchronies, though their clinical implementation requires further validation and consideration of ethical factors.