BMJ Health & Care Informatics最新文献

Objectives: Predicting mortality is vital for tailoring treatments, improving care and reducing costs. Machine learning (ML) has shown strong potential, often outperforming traditional severity-of-illness scoring systems in intensive care units (ICUs). However, the black-box nature of ML limits adoption. This study evaluates the accuracy of several ML algorithms on the Medical Information Mart for Intensive Care III (MIMIC-III) dataset and applies explainable artificial intelligence to identify variables influencing ICU mortality.

Methods: A retrospective cohort of 600 MIMIC-III patient records was analysed. ML algorithms tested included support vector machine (SVM), K-nearest neighbour (KNN), decision trees (DT), gradient boosting (GB), random forests (RF), Naive Bayes (NB), logistic regression (LR) and extra trees (ET). Models were assessed with threefold cross-validation using F1 Score, sensitivity, specificity, confusion matrix and accuracy. SHapley Additive exPlanations (SHAP) was applied to explain key factors influencing predictions.

Results: Among the eight algorithms evaluated, ET and GB achieved the highest accuracy (98.33% and 98.23%, respectively) with F1-scores above 96%. SVM also performed strongly (97.50% accuracy, F1=94.34%). RF and DT yielded accuracies of 95.00% and 94.17%, respectively. NB reached 96.67% accuracy with 100% recall, while KNN and LR showed lower discriminative performance (76.67% and 75.83% accuracy, respectively).SHAP, LIME and Shapley values consistently identified hypertension, tumours and endocrine and digestive disease as the leading predictors of mortality.

Discussion: Findings highlight ML's potential to optimise ICU decision-making and support clinicians. However, reliance on retrospective data remains a limitation, and clinical validation is required.

Conclusion: ML algorithms are highly effective for mortality prediction, and explainability is key for trust and adoption. When combined, accuracy and interpretability enable ML to safely support informed ICU decisions and improve patient outcomes.

Objectives: Lung cancer is the leading cause of cancer-related mortality worldwide, with poor prognosis largely due to late-stage diagnosis. Current screening methods such as low-dose CT face accessibility and cost barriers in resource-limited settings. This study develops a lightweight multichannel convolutional neural network for lung cancer screening support through longitudinal risk stratification using routine pre-diagnostic healthcare data.

Methods: We conducted a retrospective cohort study using Taiwan's National Health Insurance Research Database, comprising 99 615 individuals (575 lung cancer cases; 99 040 non-cancer controls). Diagnostic codes, medication records and medical orders within a 36-month observation window were extracted. Log-likelihood ratio feature selection was implemented to reduce dimensionality, achieving 99.8% reduction in computational requirements while retaining clinical relevance. A multichannel Convolutional Neural Network (CNN) architecture was designed to process these heterogeneous data modalities simultaneously.

Results: The proposed method achieved an F₁-score of 0.5738, precision of 0.7149, Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.8316 and Area Under the Precision-Recall Curve (AUPRC) of 0.1617, outperforming baseline methods in precision and F₁-score. Ablation studies confirmed that medical orders provide primary predictive value, while medication features contribute limited discriminative signal in the pre-diagnostic phase. SHapley Additive exPlanations analysis revealed that routine healthcare utilisation patterns, rather than cancer-specific features, drive risk stratification.

Discussion: The lightweight architecture enables deployment in resource-constrained clinical environments while maintaining robust performance, offering potential as a preliminary screening tool to identify high-risk individuals for further diagnostic examination.

Conclusion: Efficient deep learning models using routine clinical data can facilitate lung cancer risk stratification and screening, providing a scalable solution for clinical implementation.

Objectives: Evaluate the technical integration and usability of an intraoperative predictive machine learning model for colorectal anastomotic leakage within the Epic electronic health records (EHRs) at a single academic centre, with outputs blinded.

Methods: The system used 28 data elements from patient records, intraoperative monitoring equipment and structured operating room (OR) observations. Data were collected every 15 min and processed in the cloud with encrypted, pseudonymised transfer. Usability was assessed using the System Usability Scale (SUS) and additional questions addressing access, clarity, responsiveness, workflow impact, safety and training needs. Convenience sampling was used, with all available OR staff involved in eligible procedures invited to participate.

Results: 15 procedures (9 October 2024 to 6 March 2025) were included. Nine unique users responded (≈75% of 12 exposed; four surgeons, five OR assistants). The interface was accessed in all cases, and predictions were generated each time from all three sources. Mean SUS was 79.2 (SD 10.4; 95% CI 71.2 to 87.2). Diagnostic items favoured access speed and clarity; prediction responsiveness scored lower.

Discussion: The system could be implemented and operated reliably within a single centre and EHR environment and was perceived as easy to use, despite the small sample size. However, findings may not generalise to other hospitals or EHRs without adaptation and further multi-site evaluation.

Conclusion: The system functioned reliably and was positively received, supporting readiness for activation in real-time clinical use and prospective evaluation. Future deployment should incorporate regulatory planning and a quality-management framework to monitor performance, safety and changes in model behaviour over time.

The deployment of artificial intelligence (AI) translation tools in healthcare is accelerating rapidly, yet regulatory frameworks lag dangerously behind clinical practice. Recent data reveal that 57% of US physicians are already using or planning to adopt AI translation services within the next year. This creates a critical policy vacuum where clinicians deploy tools with variable performance across languages, risking patient safety and deepening health inequities. We examine the fractured regulatory landscape, document performance disparities between well-resourced and digitally under-represented languages, and argue for an urgent, evidence-informed policy framework centred on patient comprehension rather than linguistic accuracy.We delineate a risk-stratified validation approach comprising two distinct tracks: a 'Streamlined Pathway' for tool-language combinations with robust existing evidence (eg, Spanish) and a 'Standard Pathway' requiring independent, prospective validation for digitally under-represented languages (eg, Haitian Creole). To ensure accountability, we propose establishing oversight bodies within the U.S. Department of Health and Human Services (HHS) or the Food and Drug Administration (FDA) to mandate pre-deployment validation and post-market monitoring. Without such action, AI translation risks creating a two-tier system where the 25.7 million Americans with non-English language preferences receive dramatically different care quality based solely on the language they speak.

Objectives: Hybrid hospital-at-home (HaH) models, combining remote patient monitoring (RPM) with home visits, offer an alternative to inpatient care. Yet, evidence on how RPM is delivered is limited. This paper reports a substudy embedded within a feasibility study, examining RPM quality in a hybrid HaH programme for patients with lower respiratory tract infections before a large trial.

Methods: We analysed 19 patient trajectories in a multimethod feasibility study (April 2022-May 2023). We hypothesised that effective RPM implementation enables the timely detection of deterioration, prompting patients' return to the hospital and preventing harm. Quality was assessed via implementation fidelity and patient safety. Fidelity referred to adherence to RPM protocols for virtual ward rounds and alert management. Data sources include telemedicine logs, electronic health records, clinical observations and clinician feedback.

Results: No severe adverse events occurred. Four patients (21%) returned to the hospital. A total of 48 virtual ward rounds were scheduled, and 46 (96%) were provided, 37 of which (77%) were conducted via video. Most RPM alerts were logged outside protocol-defined timeframes, as clinicians prioritised clinical action over documentation.

Discussion: We demonstrate a gap between protocol assumptions and real-world clinical workflows. Ultimately, rigid performance metrics may overlook essential adaptive practices, underestimating true quality.

Conclusions: Despite protocol deviations, our findings suggest that RPM practices may have supported timely detection of deterioration during early-phase testing. This study underscores the need to balance protocol adherence with clinical flexibility, emphasising core functions over administrative compliance in complex interventions like hybrid HaH.

Trial registration number: NCT05087082.

Objectives: Female-specific cancers, including breast, ovarian, cervical and uterine malignancies, lack comprehensive early detection approaches, particularly for ovarian and endometrial cancers where effective population-level screening remains limited. This study aimed to develop and validate a computational method for early detection of female-specific cancers using longitudinal healthcare records.

Methods: We developed a multichannel convolutional neural network (MCNN) to analyse 36-month pre-diagnostic healthcare records from Taiwan's National Health Insurance Research Database. The study included 19 954 female patients (596 cancer cases, 19 358 controls) from 1999 to 2013. Log-likelihood ratio feature selection identified top 10 features across three data modalities (diagnostic codes, medications, medical orders). The six-channel architecture processed temporal patterns through stratified 10-fold cross-validation, with performance compared against nine baseline algorithms.

Results: MCNN achieved superior balanced performance with Macro-F₁ score of 0.8443, precision of 0.9135 and recall of 0.7978, outperforming traditional machine learning and deep learning approaches. Feature analysis revealed clinically relevant patterns including tamoxifen therapy, immunohistochemical procedures and cancer-specific diagnostic codes. SHapley Additive exPlanations (SHAP) interpretability analysis demonstrated the model's ability to identify pre-diagnostic phases through temporal healthcare utilisation patterns. Systematic feature selection reduced computational requirements by over 99%, enabling validation on Taiwan's population-scale National Health Insurance Research Database (NHIRD).

Discussion: The multichannel deep learning approach enables unified early detection across four female cancer types using routine administrative data, addressing detection gaps for ovarian and endometrial cancers while providing complementary risk stratification for existing screening programmes.

Conclusion: Clinical implementation through electronic health record (EHR) integration offers practical pathways for accessible cancer risk assessment during routine healthcare encounters.

Objectives: To evaluate the usability, usefulness and impact of a novel point of care natural language processing (NLP) system, Medical information AI Data Extractor (MiADE), to assist structured diagnosis recording in electronic health records.

Methods: Mixed methods evaluation of the implementation of MiADE in a major National Health Service hospital, with surveys, interviews and observed outpatient consultations. The number of structured diagnoses recorded per outpatient encounter was compared before and after MiADE, and completeness of inpatient problem lists was evaluated using billing diagnoses as a gold standard.

Results: 85 clinicians consented to the study and were provided access to MiADE and 24 used MiADE to receive structured data suggestions during the study period. Baseline survey data and observations showed wide variation in structured data recording despite clinicians considering it to be important. Half of postimplementation survey respondents considered MiADE to be 'very' or 'moderately' useful. Multilevel quasi-Poisson regression of 12 309 outpatient encounters (accounting for time and clustering by clinician) estimated that the post-MiADE period was associated with 23.7% more diagnoses recorded per encounter. No improvement was seen in the inpatient setting.

Discussion: Structured recording of key information such as diagnoses using a clinical terminology is essential for safe, efficient patient care, but is currently done incompletely because it is time-consuming for clinicians. MiADE was associated with an increase in outpatient structured diagnosis recording despite low uptake of the tool.

Conclusion: Point of care NLP using MiADE can potentially improve structured data recording, but further development and better clinician engagement are needed to maximise its impact.

Trial registration number: ISRCTN58300671.

Objective: To expose reasoning pathways of a reinforcement learning policy for Medicaid care coordination, develop an error taxonomy and implement fairness-aware guardrails.

Design: Retrospective interpretability audit using attention analysis, Shapley explanations, sparse autoencoder feature discovery and blinded clinician adjudication.

Setting: Medicaid care coordination programmes in Washington, Virginia and Ohio (July 2023-June 2025).

Participants: 250 000 intervention decisions; 200 divergent cases reviewed by five clinicians.

Main outcome measures: Calibrated harm prediction; algorithmic clearance and residual harm rates; error taxonomy frequencies; subgroup fairness metrics.

Results: The conformal model achieved area under the receiver operating characteristic curve of 0.80 (95% CI 0.78 to 0.82), clearing 89.5% (95% CI 88.9% to 90.1%) of decisions with 1.22% (95% CI 1.14% to 1.30%) residual harm versus 6.67% (95% CI 6.02% to 7.32%) for flagged decisions. Sparse autoencoders identified seven reasoning motifs linking social determinants to clinical cascades. The error taxonomy revealed premise errors (48%, 95% CI 41% to 55%), calibration failures (27%, 95% CI 21% to 33%) and contextual blind spots (25%, 95% CI 19% to 31%). Divergence was higher for telehealth visits (11.2%) and behavioural health patients (10.7% vs 6.9%, p<0.001). Fairness optimisation reduced race-group disparity by 37% (95% CI 22% to 48%) and sex-group disparity by 28% (95% CI 14% to 39%). Reviewers rated 23% (95% CI 17% to 29%) of overridden recommendations as well-matched, confirming appropriate human oversight.

Conclusions: Mechanistic interpretability transforms opaque algorithmic assistance into auditable decision support, providing a governance scaffold for clinical artificial intelligence deployment.

Objectives: To develop and validate a tool for standardised quality assessment of data-driven algorithms in healthcare, focusing on the underlying data pipeline.

Methods: Data Assessment Tool for Algorithm Critical Appraisal and Robust Evidence (DATA-CARE) was iteratively developed from the established Quality In Prognosis Studies framework, selected after reviewing 10 existing quality assessment tools for observational and artificial intelligence studies. DATA-CARE evaluates five quality domains of the data pipeline: study population, data, algorithm, outcome and report transparency. Each domain comprises three to five quality criteria. With a total score of 75 points, study quality is categorised as low (<45), moderate (45-59) or high (≥60). DATA-CARE was validated during a systematic review on data-driven algorithms using continuous physiological monitoring data within the paediatric intensive care unit. Two independent reviewers performed quality assessment using DATA-CARE of included studies. Tool validation was evaluated using inter-rater agreement and intraclass correlation coefficient (ICC).

Results: DATA-CARE demonstrated robust inter-rater agreement (93.5%) with ICC 0.98 (95% CI 0.96 to 0.99). Of 3858 screened studies, 31 were reviewed in the use case, describing diverse algorithms. Studies were predominantly low (32.3%) to moderate (41.9%) and sporadically (25.8%) high quality.

Discussion: Predominance of low-to-moderate quality studies reveals critical barriers to clinical implementation of data-driven algorithms, including low quality data capture and processing, lacking validation strategies and non-transparent reporting of findings.

Conclusions: DATA-CARE allows standardised and reliable critical appraisal for a wide variety of algorithms, addressing current gaps in standardised and reproducible algorithm development.

Objectives: This study aims to assess electronic health record (EHR) use in physiotherapy, identify factors influencing its adoption and evaluate physiotherapists' perceptions of its relevance.

Methods: A cross-sectional study was conducted with 138 licensed physiotherapists recruited through digital platforms. EHR utilisation was evaluated using the RSEFisio scale, a validated instrument designed to capture multiple dimensions of EHR use in physiotherapy. Descriptive and inferential statistical analyses were applied to examine usage patterns and contextual factors. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Results: The EHR utilisation rate was 78.3%. Higher utilisation was significantly associated with adequate time allocated for documentation (p=0.001), systematic recording for all patients (p=0.013) and multi-professional access to records (p=0.043). The frequency of documentation was closely linked to the perceived clinical relevance of recorded items.

Discussion: Despite the high level of EHR utilisation, physiotherapy documentation remains incomplete and driven by perceived clinical relevance. Utilisation improves with adequate time, standardised recording and interprofessional access. Inconsistent data quality undermines continuity of care and limits secondary uses, including artificial intelligence integration. Strengthening documentation is essential to improve clinical workflows and support data-driven decision-making in physiotherapy.

Conclusion: Physiotherapists recognise the value of comprehensive documentation, but report limited time and incomplete records. The disconnect between awareness and practice highlights the need for practical, system-level strategies to support more consistent and effective EHR use in physiotherapy.