In data science, Missing Data (MD) are handled by approaches from basic imputation (e.g., mean) to more complex (e.g., Machine Learning (ML)). Assessing these methods consists in amputating data then imputating back amputated values to compare with ground truth. The existing amputation methods are most often Random Amputation (RA) or parameterized approaches using Missing Data Mechanisms (MDM) involved. MD from real-world datasets can follow complex patterns and the specific MDM involved can be impossible to determine with certainty. Thus, existing data amputation methods are often difficult to properly apply on real-world datasets.In this article, we propose Missing Mechanisms Respectful Data Amputation (MMRDA) as a new method to generate synthetic MD and we study the impact of data amputation methods on the results of imputation techniques assessment.MMRDA has been compared to the RA method and assessed on open datasets. MMRDA significantly outperforms RA. Four well-known imputation techniques were used to compare imputation performance post-MMRDA with imputation performance post-RA. Differences were significant and the best imputation performance is not always achieved by the same imputer depending on the amputation technique used for a given dataset. This supports that MDM involved affect imputation and that the attribution of complex mechanisms is still to be explored. Therefore, we recommend all researchers to pay close attention to the amputation method used during imputation techniques assessment or comparison, especially on real-world datasets.

INTRODUCTION:Conditions requiring emergency gastrointestinal surgery pose substantial challenges to healthcare systems and patient outcomes. For emergency gastrointestinal surgery, the mortality rate is higher than after other types of surgery, and 30-day readmission rates can exceed 30%. Unlike elective surgery (for which the application of enhanced recovery after surgery (ERAS) procedures has led to demonstrably better recovery), emergency surgery patients are still managed in an ad hoc manner. The Réhabilitation Augmentée pour les Urgences Chirurgicales (RAUC) multi-faceted research program has been designed to transform the care of patients undergoing emergency gastrointestinal surgery [7]. The ancillary RAUCisable study will develop a classification model that can automatically flag up RAUC-eligible patients early in their visit to the emergency department (ED). The study's secondary objectives include the identification of key features and the prediction of time to surgery. METHOD:RAUCisable is single-centre, retrospective, observational cohort study of electronic health records in the ED and digestive surgery department at Amiens-Picardie University Hospital (Amiens, France). All adult patients having attended the ED between January 1st, 2021, and December 31st, 2024, will be considered for inclusion. The primary classification outcome is eligibility for the RAUC pathway. EXPECTED RESULTS:We expect to identify ~2,400 RAUC-eligible patients from among ~250,000-300,000 ED visits over a 4-year period. These patients are likely to be significantly older than non-surgical ED patients (e.g., more over-65s), with a higher proportion of acute abdominal conditions (e.g., ~ 24% with appendicitis, ~ 13% with bowel obstruction, ~ 10% with peritonitis, etc.), and greater disease acuity on triage. DISCUSSION:The RAUCisable study's findings will directly guide a concomitant, prospective, controlled study (RAUC-AMIENS) of the augmented recovery pathway, including ERAS elements and remote monitoring. CONCLUSION:The RAUCisable study is a pivotal step toward digitally enhanced emergency surgical care. By learning from past data, we are seeking to improve the future management of emergency surgery patients through timely identification and targeted care pathways. This protocol article details our methodological approach for ensuring rigor and reproducibility. TRIAL REGISTRATION:NCT07037719.

Medical emergency calls are complex interactions in which critical information must be identified rapidly from fragmented, emotional, and evolving conversations. In French emergency medical services, the decision to dispatch a mobile intensive care unit is one of the most consequential triage decisions, yet little is known about how severity-related information is represented across call transcripts and clinical notes. This retrospective study analyzed 8 904 first calls received by a French emergency medical service, using automatic call transcripts, dispatcher notes, and physician notes. We compared lexical and contextual language models to determine whether these textual sources could predict mobile intensive care unit deployment. Calls leading to deployment showed distinct lexical patterns, while clinical notes contained more condensed medical terminology and abbreviations. The first quarter of the transcript already provided meaningful predictive information, and performance improved when the full transcript was used. Clinical notes, especially physician notes, carried strong predictive signal, and combining transcripts with notes achieved the highest performance. Among calls in which deployment occurred more than 15 minutes after call start, the contextual model identified a substantial proportion as potentially requiring earlier attention. These findings suggest that language-based decision-support tools could help flag complex or discordant calls for secondary medical review while preserving human oversight.

Background: Overcrowded emergency departments (EDs) create significant challenges for patient management and hospital efficiency. In response, Amiens Picardy University Hospital (APUH) developed the “Prediction of the Patient Pathway in the Emergency Department” (3P-U) model to enhance patient flow management. Objectives: To develop and clinically validate an explainable artificial intelligence (XAI) model for hospital admission predictions, using structured triage data, and demonstrate its real-world applicability in the ED setting. Methods: Our retrospective, single-center study involved 351,019 patients consulting in APUH’s EDs between 2015 and 2018. Various models (including a cross-validation artificial neural network (ANN), a k-nearest neighbors (KNN) model, a logistic regression (LR) model, and a random forest (RF) model) were trained and assessed for performance with regard to the area under the receiver operating characteristic curve (AUROC). The best model was validated internally with a test set, and the F1 score was used to determine the best threshold for recall, precision, and accuracy. XAI techniques, such as Shapley additive explanations (SHAP) and partial dependence plots (PDP) were employed, and the clinical explanations were evaluated by emergency physicians. Results: The ANN gave the best performance during the training stage, with an AUROC of 83.1% (SD: 0.2%) for the test set; it surpassed the RF (AUROC: 71.6%, SD: 0.1%), KNN (AUROC: 67.2%, SD: 0.2%), and LR (AUROC: 71.5%, SD: 0.2%) models. In an internal validation, the ANN’s AUROC was 83.2%. The best F1 score (0.67) determined that 0.35 was the optimal threshold; the corresponding recall, precision, and accuracy were 75.7%, 59.7%, and 75.3%, respectively. The SHAP and PDP XAI techniques (as assessed by emergency physicians) highlighted patient age, heart rate, and presentation with multiple injuries as the features that most specifically influenced the admission from the ED to a hospital ward. These insights are being used in bed allocation and patient prioritization, directly improving ED operations. Conclusions: The 3P-U model demonstrates practical utility by reducing ED crowding and enhancing decision-making processes at APUH. Its transparency and physician validation foster trust, facilitating its adoption in clinical practice and offering a replicable framework for other hospitals to optimize patient flow.

Chronic Obstructive Pulmonary Disease (COPD) is a major global health issue. Early detection and identification of risk factors are critical for effective management. This study applies explainable machine learning techniques using LIME and SHAP to analyze a large US dataset and provide interpretable insights into COPD risk factors. While established factors such as smoking and age remain central, the analysis also highlights the underappreciated influence of mental health, physical inactivity, and comorbid conditions. These insights enable a more holistic understanding of COPD risk profiles and support the development of tailored prevention strategies. While limited to US data, the findings pave the way for future studies incorporating diverse populations

Despite the significant advances made by Artificial Intelligence (AI) models in enhancing medical diagnostics and prognostics, their opacity poses a hurdle to widespread clinical adoption. In this regard, Explainable AI (XAI) aims to demystify these complex models, such as neural networks, by revealing the reasoning behind predictions. However, a notable gap exists in enabling non-experts to verify these explanations, necessitating human-in-the-loop evaluation. This paper introduces a systematic protocol, including a novel “consistency” metric, to evaluate the SHAP-based explanations of XAI, comparing them against the clinical knowledge of expert clinicians. We demonstrate how this metric could facilitate both global and feature-specific analyses, operating at the level of individual instances, and thus enhancing AI transparency. It is conceived that the implications of this work may extend beyond the medical context, offering a standardized methodology that could potentially improve the interpretability and acceptance of AI systems in diverse domains.

Large Language Models (LLMs) hold promise for advancing Emergency Medicine by enhancing operational efficiency and supporting decision-making. This scoping review explores the ethical, legal, and global considerations influencing LLM deployment in emergency care. Key ethical concerns, including patient safety, data privacy, and transparency, emphasise the need for explainable AI (XAI) to build trust and prevent biased outputs. Legal challenges highlight the importance of regulatory compliance, especially regarding data protection laws like the GDPR. Significant international variability in LLM adoption further underscores the need for harmonised guidelines to ensure safe and equitable AI integration across diverse healthcare systems. To advance the responsible use of LLMs, future research should prioritise model transparency, consider resource-limited settings, and focus on establishing robust regulatory frameworks.

In the field of emergency medicine (EM), the use of decision support tools based on artificial intelligence has increased markedly in recent years. In some cases, data are omitted deliberately and thus constitute “data not purposely collected” (DNPC). This accepted information bias can be managed in various ways: dropping patients with missing data, imputing with the mean, or using automatic techniques (e.g., machine learning) to handle or impute the data. Here, we systematically reviewed the methods used to handle missing data in EM research. A systematic review was performed after searching PubMed with the query “(emergency medicine OR emergency service) AND (artificial intelligence OR machine learning)”. Seventy-two studies were included in the review. The trained models variously predicted diagnosis in 25 (35%) publications, mortality in 21 (29%) publications, and probability of admission in 21 (29%) publications. Eight publications (11%) predicted two outcomes. Only 15 (21%) publications described their missing data. DNPC constitute the “missing data” in EM machine learning studies. Although DNPC have been described more rigorously since 2020, the descriptions in the literature are not exhaustive, systematic or homogeneous. Imputation appears to be the best strategy but requires more time and computational resources. To increase the quality and the comparability of studies, we recommend inclusion of the TRIPOD checklist in each new publication, summarizing the machine learning process in an explicit methodological diagram, and always publishing the area under the receiver operating characteristics curve—even when it is not the primary outcome.

The research on artificial intelligence (AI) applied to emergency medicine and its daily use in emergency departments has significantly increased in recent years. AI should be considered as a tool to assist in the diagnosis and therapeutic management of patients and to improve the orga-nization of the emergency departments, particularly conside-ring contextual constraints from several aspects. AI has not only advantages (reproducibility, speed) but also risks (error, loss of critical thinking). Like the General Data Protection Regulation, especially for health, the European Commission has published a draft regulation called "AI Act" for the design, development, and use of AI algorithms. It wishes to impose, among other things, a human guarantee, in other words, human supervision to ensure the safety of patients, caregivers, and institutions. The establishment of a multipro-fessional human guarantee college aiming to guarantee the supervision of AI tools from design, development, deploy-ment, and daily use will thus ensure the long-term safety of patients.

Explainable Artificial Intelligence (XAI) has the potential to revolutionize healthcare by providing more transparent, trustworthy, and understandable predictions made by AI models. To this end, the present study aims to develop an explainable NLP model for predicting patient admissions to the emergency department based on triage notes. We utilize transformer models to leverage the extensive textual data captured in triage notes, while also delivering interpretable results by using the LIME approach. The results show that the proposed model provides satisfactory accuracy along with an interpretable understanding of the factors contributing to patient admission. In general, this work highlights the potential of NLP in improving patient care and decision-making in emergency medicine.

La recherche sur l’intelligence artificielle (IA) appliquée à la médecine d’urgence et son utilisation au quotidien dans les structures d’urgences (SU) ont augmenté significativement ces dernières années. L’IA doit être considérée comme un outil d’aide à la prise en charge diagnostique et thérapeutique des patients et d’amélioration de l’organisation des SU, notamment par la prise en compte de contraintes « métiers », contextuelles, relatives aux patients et plus généralement structurelles. L’IA comporte des avantages (reproductibilité, rapidité) mais aussi des risques (erreur, perte d’esprit critique). À l’image du Règlement général sur la protection des données et notamment de santé, la Commission européenne a publié un projet de règlement nommé « AI Act » pour la conception, le développement et l’utilisation des algorithmes d’IA. Elle souhaite imposer, entre autres, une garantie humaine, autrement dit une supervision humaine pour assurer la sécurité des patients, des soignants et des institutions. La mise en place d’un collège de garantie humaine pluriprofessionnel visant à garantir la supervision des outils d’IA de la conception au développement, au déploiement et à l’utilisation quotidienne permettra ainsi d’assurer durablement la sécurité des patients.

BACKGROUND Recently, validated clinical decision rules have been developed that avoid unnecessary use of computed tomographic pulmonary angiography (CTPA) in patients with suspected pulmonary embolism (PE) in the emergency department (ED). OBJECTIVE To measure any resulting change in CTPA use for suspected PE. DESIGN Retrospective analysis. SETTING 26 European EDs in 6 countries. PATIENTS Patients with CTPA performed for suspected PE in the ED during the first 7 days of each odd month between January 2015 and December 2019. MEASUREMENTS The primary end points were the CTPAs done for suspected PE in the ED and the number of PEs diagnosed in the ED each year adjusted to an annual census of 100 000 ED visits. Temporal trends were estimated using generalized linear mixed regression models. RESULTS 8970 CTPAs were included (median age, 63 years; 56% female). Statistically significant temporal trends for more frequent use of CTPA (836 per 100 000 ED visits in 2015 vs. 1112 in 2019; P < 0.001), more diagnosed PEs (138 per 100 000 in 2015 vs. 164 in 2019; P = 0.028), a higher proportion of low-risk PEs (annual percent change [APC], 13.8% [95% CI, 2.6% to 30.1%]) with more ambulatory management (APC, 19.3% [CI, 4.1% to 45.1%]), and a lower proportion of intensive care unit admissions (APC, -8.9% [CI, -17.1% to -0.3%]) were observed. LIMITATION Data were limited to 7 days every 2 months. CONCLUSION Despite the recent validation of clinical decision rules to limit the use of CTPA, an increase in the CTPA rate along with more diagnosed PEs and especially low-risk PEs were instead observed. PRIMARY FUNDING SOURCE None specific for this study.

Severe forms of coronavirus disease 2019 （COVID-19）can lead to hypoxemic pneumonia and the need for treatment with high-flow nasal cannula （HFNC） oxygen therapy. [1] This treatment decreases the need for subsequent invasive ventilation in patients with hypoxemic pneumonia.The oxygen flow in the nasal cannula is driven pneumatically or using a turbine.Oxygen and air are mixed （giving the fraction of inspired oxygen[FiO 2 ]） and propelled at up to 80L·min -1 .The propelled gases must be heated and humidified.HFNC system is not intended for use outside a hospital environment.

Background: During the coronavirus disease 2019 (COVID-19) pandemic, calculation of the number of emergency department (ED) beds required for patients with vs. without suspected COVID-19 represented a real public health problem. In France, Amiens Picardy University Hospital (APUH) developed an Artificial Intelligence (AI) project called “Prediction of the Patient Pathway in the Emergency Department” (3P-U) to predict patient outcomes. Materials: Using the 3P-U model, we performed a prospective, single-center study of patients attending APUH’s ED in 2020 and 2021. The objective was to determine the minimum and maximum numbers of beds required in real-time, according to the 3P-U model. Results A total of 105,457 patients were included. The area under the receiver operating characteristic curve (AUROC) for the 3P-U was 0.82 for all of the patients and 0.90 for the unambiguous cases. Specifically, 38,353 (36.4%) patients were flagged as “likely to be discharged”, 18,815 (17.8%) were flagged as “likely to be admitted”, and 48,297 (45.8%) patients could not be flagged. Based on the predicted minimum number of beds (for unambiguous cases only) and the maximum number of beds (all patients), the hospital management coordinated the conversion of wards into dedicated COVID-19 units. Discussion and conclusions: The 3P-U model’s AUROC is in the middle of range reported in the literature for similar classifiers. By considering the range of required bed numbers, the waste of resources (e.g., time and beds) could be reduced. The study concludes that the application of AI could help considerably improve the management of hospital resources during global pandemics, such as COVID-19.

The advent of transformer models has allowed for tremendous progress in the Natural Language Processing (NLP) domain. Pretrained transformers could successfully deliver the state-of-the-art performance in a myriad of NLP tasks. This study presents an application of transformers to learn contextual embeddings from freetext triage notes, widely recorded at the emergency department. A large-scale retrospective cohort of triage notes of more than 260K records was provided by the University Hospital of Amiens-Picardy in France. We utilize a set of Bidirectional Encoder Representations from Transformers (BERT) for the French language. The quality of embeddings is empirically examined based on a set of clustering models. In this regard, we provide a comparative analysis of popular models including Came mBERT, FlauBERT, and mBART. The study could be generally regarded as an addition to the ongoing contributions of applying the BERT approach in the healthcare context.

The implementation of Data Analytics has achieved a significant momentum across a very wide range of domains. Part of that progress is directly linked to the implementation of Text Analytics solutions. Organisations increasingly seek to harness the power of Text Analytics to automate the process of gleaning insights from unstructured textual data. In this respect, this study aims to provide a meeting point for discussing the state-of-the-art applications of Text Analytics in the healthcare domain in particular. It is aimed to explore how healthcare providers could make use of Text Analytics for different purposes and contexts. To this end, the study reviews key studies published over the past 6 years in two major digital libraries including IEEE Xplore, and ScienceDirect. In general, the study provides a selective review that spans a broad spectrum of applications and use cases in healthcare. Further aspects are also discussed, which could help reinforce the utilisation of Text Analytics in the healthcare arena.

L'enseignement est un des lieux dans lesquels l'histoire de la Revolution francaise peut aujourd'hui se renouveler. Depuis la fin du XIXe siecle, l'ecole mais aussi l'universite ont eu pour fonction sociale de defendre et promouvoir la Revolution autant que de la comprendre et de l'expliquer. Les difficultes a enseigner la Revolution francaise sont en partie dues a l'etonnante longevite de cette tradition missionnaire et patrimoniale. Face a celle-ci, la pedagogie du jeu et de l'incarnation peut constituer un recours. Dans cette optique, nous avons organise un atelier d'histoire avec des etudiantes et des etudiants des universites Paris 1 et Paris 8. Fonde sur un travail documentaire et des methodes d'improvisation theâtrale, cet atelier, croisant des debats sur l'abolition de la traite (1790) et sur l’adoption du Civil Rights Act (1964) a partiellement dejoue les attendus d'une histoire trop souvent jouee d'avance. Utilises en complement des methodes de transmission plus classiques, ces ateliers contribuent, le temps d'une experience partagee, a l'appropriation nouvelle d'une histoire parfois fatiguee.

Data Analytics is rapidly expanding within the healthcare domain to help develop strategies for improving the quality of care and curbing costs as well. Natural Language Processing (NLP) solutions have received particular attention whereas a large part of clinical data is stockpiled into unstructured physician or nursing notes. In this respect, we attempt to employ NLP to provide an early prediction of the medical specialties at hospital admission. The study uses a large-scale dataset including more than 260K ED records provided by the Amiens-Picardy University Hospital in France. Our approach aims to integrate structured data with unstructured textual notes recorded at the triage stage. On one hand, a standard MLP model is used against the typical set of features. On the other hand, a Convolutional Neural Network is used to operate over the textual data. While both learning components are conducted independently in parallel. The empirical results demonstrated a promising accuracy in general. It is conceived that the study could be an additional contribution to the mounting efforts of applying NLP methods in the healthcare domain.

Overcrowding in Emergency Departments (ED) is considered as an international issue, which could have adverse impacts on multiple care outcomes such as the length of stay for example. Part of the solution could lie in the early prediction of the patient outcome as discharge or hospitalization. This study applies Deep Learning to this end. A large-scale dataset of about 260K ED records was provided by the Amiens-Picardy University Hospital in France. In general, our approach is based on integrating structured data with unstructured textual notes recorded at the triage stage. The key idea is to apply a multi-input of mixed data for training a classification model to predict hospitalization. In a simultaneous manner, the model training utilizes the numeric features along with textual data. On one hand, a standard Multi-Layer Perceptron (MLP) model is used with the standard set of features (i.e. numeric and categorical). On the other hand, a Convolutional Neural Network (CNN) is used to operate over the textual data. The two components of learning are conducted independently in parallel. The empirical results demonstrated that the classifier could achieve a very good accuracy with ROC-AUC≈0.83. The study is conceived to contribute to the mounting efforts of applying Natural Language Processing in the healthcare domain.