BMJ Health & Care Informatics最新文献

Electronic health records (EHRs) provide comprehensive patient data, which could be better used to enhance informed decision-making, resource allocation and coordinated care, thereby optimising healthcare delivery. However, in mental healthcare, critical information, such as on risk factors, precipitants and treatment responses, is often embedded in unstructured text, limiting the ability to automate at scale measures to identify and prioritise local populations and patients, which potentially hinders timely prevention and intervention. We describe the development and proof-of-concept implementation of Visual & Interactive Engagement With Electronic Records, a clinical informatics platform designed to enhance direct patient care and population health management by improving the accessibility and usability of EHR data. We further outline strategies that were employed in this work to foster informatics innovation through interdisciplinary and cross-organisational collaboration to support integrated, personalised care and detail how these advancements were piloted and implemented within a large UK mental health National Health Service Foundation Trust to improve patient outcomes at an individual patient, clinician, clinical team and organisational level.

Objectives: Tailoring medication dosing to an individual's traits is complex, but artificial intelligence (AI) advancements enable greater precision. Our study objectives were to gauge healthcare providers' perspectives on AI-guided precision dosing and to identify barriers and enablers for adopting AI-guided precision dosing into clinical practice.

Methods: We conducted a qualitative study using purposive sampling to select a diverse group of healthcare providers, thereby broadening the viewpoints. We explored their receptiveness to AI-enabled dosing and sought to uncover implementation challenges. During the interviews, we introduced CURATE.AI as an example of an AI dosing tool. We analysed the data using deductive methods, coding the data according to the Unified Theory of Acceptance and Use of Technology framework.

Results: We interviewed 16 participants (9 doctors, 4 nurses and 3 pharmacists). Interviews revealed diverse perspectives, from hopeful anticipation to recognised challenges. While acknowledging AI's potential to enhance decision-making and patient safety, concerns about AI's suitability for complex cases, erosion of critical thinking, liability protection, and trust arose. Moreover, transparency, understandability of AI output and human oversight were seen as essential to mitigate risks and promote acceptance.

Discussion: AI-enabled dosing tools have the potential to optimise dosing and improve patient safety, but adoption barriers remain. Successful implementation will require technically robust tools and careful alignment with clinical workflows and user expectations.

Conclusion: Our study highlights the hopeful anticipation and complex challenges of introducing AI-enabled dosing into clinical practice. As AI inevitably becomes a part of healthcare, ongoing evaluation is essential to demonstrate value and promote adoption.

Objectives: Electronic prescriptions (e-prescriptions) introduce drug product selection mismatches during pharmacy data entry. System Approach to Verifying Electronic Prescriptions (SAV E-Rx) detects and alerts pharmacy staff to clinically significant occurrences. This study evaluates outcomes of the identified mismatches.

Methods: A retrospective analysis was conducted using 1 year of e-prescriptions and dispensing data from 14 community pharmacies across 9 US states. SAV E-Rx screened the data, and flagged mismatches were reviewed by pharmacists using the Common Formats for Event Reporting. Data were analysed using descriptive statistics, the Mann-Whitney U test and χ² tests.

Results: Of 1 250 804 records processed, 699 662 included sufficient data for comparison. Pharmacists classified 587 (88.7%) flagged records as intended mismatches and 75 (11.3%) as unintended. Intended mismatches involved ingredients (26.2%), strengths (53.7%) and dosage forms (47.4%), mainly due to prescriber-approved substitutions (62.4%). Unintended mismatches stemmed from ingredients (42.7%), strengths (36.0%) and dosage forms (54.7%) discrepancies, primarily reported as human error (82.7%) and labelling issues (76.0%). Future alerts were favoured for unintended mismatches (96.0%) compared with intended mismatches (56.7%) (p<0.001).

Discussion: While routine substitutions are a normal part of quality and timely care, unintended mismatches may pose clinical risks. These errors can arise from human factors and workflow challenges, including high prescription volumes and manual overrides. SAV E-Rx serves as an independent, automated safety net that flags mismatches, catching postdispensing errors that would otherwise go unnoticed.

Conclusions: E-prescription errors remain a safety concern. Routine implementation of SAV E-Rx could enhance error detection and enable timely interventions.

Objectives: Large language model (LLM)-based tools offer potential for clinical practice but raise concerns regarding output accuracy, patient safety and data security. We aimed to assess Swiss clinicians' use, knowledge and perception of LLMs and identify associated factors.

Methods: An anonymous online survey was distributed via 34 medical societies in Switzerland. The primary outcome was frequent use of LLMs (at least weekly use). The secondary outcome was higher knowledge regarding LLMs (score above the median in an 11-item test). Qualitative analysis explored clinicians' perceptions of LLM-related opportunities and risks.

Results: Among 685 participants (response rate 29.0%), 225 (32.8%) reported frequent use of LLMs, 25 (3.6%) reported having used a specific medical LLM and 42 (6%) reported the availability of workplace LLM guidelines. The median knowledge test score was 6 points (IQR 4-8 points). Multivariable analysis showed that younger age, male sex and research activity were significantly associated with frequent use and higher knowledge. Qualitative analysis identified administrative support, analytical assistance and access to information as key opportunities. The main risks identified were declining clinical skills, poor output quality and legal or ethical concerns.

Discussion: The study highlights a notable adoption of LLMs among Swiss clinicians, particularly among younger, male and research-active individuals. However, the limited availability of workplace guidelines raises concerns about safe and effective use.

Conclusion: The gap between widespread LLM use and the scarcity of workplace guidelines underscores the need for accessible educational resources and clinical guidelines to mitigate potential risks and promote informed use.

Objectives: The 'tumour, node, metastasis' (TNM) classification of colorectal cancer (CRC) predicts prognosis and so is vital to consider in analyses of patterns and outcomes of care when using electronic health records. Unfortunately, it is often only available in free-text reports. This study aimed to develop regex-based text-processing algorithms that identify the reports describing CRC and extract the TNM staging at a low computational cost.

Methods: The CRC and TNM extraction algorithms were iteratively developed using 58 634 imaging and pathology reports of patients with CRC from the Oxford University Hospitals (OUH) and Royal Marsden (RMH) NHS Foundation Trusts (FT), with additional input from Imperial College Healthcare and Christie NHS FTs. The algorithms were evaluated on a stratified random sample of 400 OUH development data reports and 400 newer 'unseen' OUH reports. The reports were annotated with the help of two clinicians.

Results: The CRC algorithm achieved at least 93.0% positive predictive value (PPV), 72.1% sensitivity, 64.0% negative predictive value (NPV) and 90.1% specificity for primary CRC on pathology reports. On imaging reports, it demonstrated at least 78.0% PPV, 91.8% sensitivity, 93.0% NPV and 80.9% specificity. For the main T/N/M categories, the TNM algorithm achieved PPVs of at least 93.9% (T), 97.7% (N) and 97.2% (M), and sensitivities of 63.6% (T), 89.6% (N) and 64.8% (M). NPVs were at least 45.0% (T), 91.1% (N), 88.4% (M), and specificities 95.7% (T), 98.1% (N), 99.3% (M). Reductions in performance were mostly due to implicit staging. For extracting explicit TNM stages, current or historical, the algorithm made no errors on 400 pathology reports and six errors on 400 imaging reports.

Conclusion: The TNM algorithm accurately extracts explicit TNM staging, but other methods are needed for retrieving implicit stages. The CRC algorithm is accurate on non-supplementary reports, but outputs need additional review if higher precision is required.

Objective: This study aims to develop a transformer-based deep learning model for real-time activity recognition and fall detection, addressing the limitations of existing methods in terms of accuracy and real-time applicability.

Methods: The proposed system uses sliding window segmentation technique to process wearable sensor data, including accelerometer, gyroscope and orientation signals. The transformer encoder models temporal dependencies through a self-attention mechanism, enabling the extraction of global and local temporal patterns. The performance of the model is evaluated on an updated version of the MobiAct data set, which includes over 14 million sensor records collected from 66 participants and 16 activities, including four types of falls and multiple scenario-based activities of daily living.

Result: The transformer model achieved an accuracy of over 98% and demonstrated excellent precision and recall for difficult fall categories such as forward-lying and sideward-lying. Comparative analysis shows that transformers outperform convolutional neural networks long short-term memory (CNN-LSTM) and temporal convolutional networks in terms of classification metrics, confusion matrix results and training stability.

Discussion: The results highlight the effectiveness of the transformer model in capturing complex temporal dependencies, addressing key challenges such as misclassification and false positives. Compared with traditional models, its parallel processing capabilities improve real-time deployment efficiency.

Conclusion: This research establishes transformer-based models as powerful solutions for activity recognition and fall detection, providing reliable applications for elderly care and fall prevention. Future work will focus on optimising edge devices and validating on real-world data sets.

Objective: Many hospitals struggle to transform electronic health record (EHR) data to support performance, continuous improvement and patient safety. Our study aimed to explore the feasibility of semiautomated surveillance for healthcare-associated infections (HAIs) in Australian hospitals, focussing on Staphylococcus aureus bloodstream infection (SABSI) surveillance.

Method: National surveillance case definitions were reviewed with an inventory list of data elements created to identify high-probability healthcare-associated SABSI events. An interview schedule was developed to assess the availability, characteristics and quality of EHR data for data elements. Interviews were conducted with hospital infection prevention and control (IPC) staff.

Results: 12 IPC staff representing 12 hospitals and 11 healthcare organisations were interviewed. EHRs were in place at nine (75%) sites, supplied by six different vendors. Heterogeneity was observed in EHR functionalities, data capture methods for routine care and local approaches to use electronic systems to reduce HAI surveillance workload. None reported using automated surveillance systems. Most core data elements for the SABSI algorithm were present in EHRs, suggesting only minor modification to the SABSI definitions may be needed for automation, but issues with data quality were also described.

Discussion: We propose that modification of the national SABSI definitions is needed for automation. While many Victorian hospitals have adopted EHRs, data quality and interoperability issues limit the leveraging of EHR data for secondary purposes.

Conclusions: We have taken the initial steps of evaluating the feasibility of semiautomated HAI surveillance in Victorian hospitals. With further development, this offers the promise of enhanced efficiency and reduced human resources required for HAI surveillance.

Objectives: Surgical pressure injuries (SPIs) are a significant patient safety risk due to prolonged immobility and tissue hypoperfusion under general anaesthesia. Existing risk assessment tools lack real-time predictive capabilities. This study developed and validated a machine-learning model for SPI prediction and clinical integration.

Method: This retrospective cohort study analysed electronic health records from 931 surgical inpatients under general anaesthesia between January 2016 and December 2021. SPI cases were identified using ICD-10 codes with 1:1 matching by medical specialty. Data preprocessing included imputation, normalisation and outlier removal. Logistic regression (LR), multilayer perceptron (MLP) and decision tree (DT) models were developed and validated via cross-validation. Model performance was assessed using area under the curve (AUC), accuracy, precision, recall and F1 score.

Results: Significant SPI predictors included the Charlson Comorbidity Index (p<0.001), number of medication types (p=0.001) and body mass index (p<0.001). The MLP outperformed LR (AUC=0.707) and DT (AUC=0.717), achieving the highest AUC (0.836), accuracy (0.773), precision (0.812), recall (0.688) and F1 score (0.745).

Discussion: The MLP model effectively identified key SPI risk factors, outperforming LR and DT by capturing non-linear relationships. Its integration into clinical workflows may enhance perioperative risk management through early detection and targeted interventions.

Conclusion: Machine learning integration can improve early SPI detection and personalised prevention. The MLP model demonstrated the highest potential for real-time SPI risk stratification. Future research should validate this model across diverse surgical populations and develop scalable strategies for clinical implementation.

Background: The significance of patient safety has been acknowledged in healthcare systems, prompting the need for effective patient safety monitoring systems (PSMSs). These systems' endeavour is to manage patient safety data and improve overall safety within healthcare organisations. This study aims to characterise the implementation of and outputs of such systems across hospital settings.

Methods: A systematic literature review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The review included a comprehensive search of databases such as PubMed, EMBASE, Scopus, Web of Science and Google Scholar for studies published in English up to 30 July 2024. The focus was on monitoring systems that manage patient safety in medical care, with inclusion criteria that required studies to examine the application of PSMSs and report their implementation outputs.

Results: The literature search yielded 23 relevant studies published between 2009 and 2023. PSMSs were used in various clinical contexts, including emergency departments, radiology wards, intensive care units and operating rooms, addressing various issues such as medication safety, healthcare-associated infections, blood transfusion errors, surgical site infections, laboratory and radiology adverse events. The findings indicated positive outputs from the implementation of PSMSs. Furthermore, these systems provide valuable information and timely alerts and contribute to a culture of safety in healthcare facilities.

Conclusions: PSMSs can be used for enhancing safety practices, reducing adverse events and promoting a culture of patient safety. Further research and continued implementation of PSMSs are essential to further augment patient safety standards in healthcare settings.

Objective: Artificial intelligence (AI) holds promise for predicting sepsis. However, challenges remain in integrating AI, natural language processing (NLP) and free text data to enhance sepsis diagnosis at emergency department (ED) triage. This study aimed to evaluate the effectiveness of AI in improving sepsis diagnosis.

Methods: This retrospective cohort study analysed data from 134 266 patients admitted to the ED and subsequently hospitalised between 1 January 2016 and 31 December 2021. The data set comprised 10 variables and free-text triage comments, which underwent tokenisation and processing using a bag-of-words model. We evaluated four traditional NLP classifier models, including logistic regression, LightGBM, random forest and neural network. We also evaluated the performance of the BERT classifier. We used area under precision-recall curve (AUPRC) and area under the curve (AUC) as performance metrics.

Results: Random forest exhibited superior predictive performance with an AUPRC of 0.789 (95% CI: 0.7668 to 0.8018) and an AUC of 0.80 (95% CI: 0.7842 to 0.8173). Using raw text, the BERT model achieved an AUPRC of 0.7542 (95% CI: 0.7418 to 0.7741) and AUC of 0.7735 (95% CI: 0.7628 to 0.8017) for sepsis prediction. Key variables included ED treatment time, patient age, arrival-to-treatment time, Australasian Triage Scale and visit type.

Discussion: This study demonstrates AI, particularly random forest and BERT classifiers, for early sepsis detection in EDs using free-text patient concerns.

Conclusion: Incorporating free text into machine learning improved diagnosis and identified missed cases, enhancing sepsis prediction in the ED with an AI-powered clinical decision support system. Large, prospective studies are needed to validate these findings.