Unintentional home and leisure injuries (HLIs) are a leading cause of morbidity and mortality among older adults, with falls representing the main source of harm. In France, as in many countries, the recurrence of such injuries has received limited attention, despite their major health and social consequences. We conducted a prospective cohort study using data from the French MAVIE cohort, which continuously collects information on household characteristics, socio-demographics, health conditions, and injuries among older adults. We applied duration models with selection correction to estimate both the probability of sustaining an HLI and the time interval until injury recurrence, accounting for previous injuries and unobserved heterogeneity. Among 2,179 participants aged 65 and older (mean age = 69.9 years; 57.2

OBJECTIVE:This study aims to assess whether the emotions experienced during an urgent health problem represent risk factors for developing chronic pain. METHOD:A pain study was carried out as part of a randomized multicentre study on the prevention of post-concussion syndrome and post-traumatic stress syndrome (SOFTER) following emergency hospitalisation. Nine hundred and fourteen patients not suffering from chronic pain at admission provided information on the presence and intensity of eight emotions (anger, fear, regret, sadness, relief, contentment, joy, and interest) during their stay at the emergency department. Four months later, they were called to assess if chronic pain had developed. The predisposition to experience these emotions prior to the emergency room was questioned at this occasion. RESULTS:Four months after their admission to the emergency department (ED), 30 % of the patients described experiencing chronic pain. Adjusted for patient's perception of own health, intensity of acute pain, reason for visit and level of education at the time of admission, sadness (OR = 1.5 95 % CI = [1.1-2.2]) and anger (OR = 1.6 95 % CI = [1.1-2.5]) declared in the ED were predictive of pain status at four months. By taking into account patient pre-disposition to feel each of these two emotions, we observed a significantly higher risk of chronic pain at four months among patients pre-disposed to anger who declared a relatively strong anger during their time spent in the ED (OR = 2.90, IC = 1.12-7.52). CONCLUSION:This study shows that the emotional state of patients admitted to the ED must be taken into consideration in order to detect people likely to experience chronic pain and therefore offer personalised preventive treatment. Further studies will be necessary to better comprehend the role played by emotions in the development of chronic pain.

Background The digitization of health care, facilitated by the adoption of electronic health records systems, has revolutionized data-driven medical research and patient care. While this digital transformation offers substantial benefits in health care efficiency and accessibility, it concurrently raises significant concerns over privacy and data security. Initially, the journey toward protecting patient data deidentification saw the transition from rule-based systems to more mixed approaches including machine learning for deidentifying patient data. Subsequently, the emergence of large language models has represented a further opportunity in this domain, offering unparalleled potential for enhancing the accuracy of context-sensitive deidentification. However, despite large language models offering significant potential, the deployment of the most advanced models in hospital environments is frequently hindered by data security issues and the extensive hardware resources required. Objective The objective of our study is to design, implement, and evaluate deidentification algorithms using fine-tuned moderate-sized open-source language models, ensuring their suitability for production inference tasks on personal computers. Methods We aimed to replace personal identifying information (PII) with generic placeholders or labeling non-PII texts as “ANONYMOUS,” ensuring privacy while preserving textual integrity. Our dataset, derived from over 425,000 clinical notes from the adult emergency department of the Bordeaux University Hospital in France, underwent independent double annotation by 2 experts to create a reference for model validation with 3000 clinical notes randomly selected. Three open-source language models of manageable size were selected for their feasibility in hospital settings: Llama 2 (Meta) 7B, Mistral 7B, and Mixtral 8×7B (Mistral AI). Fine-tuning used the quantized low-rank adaptation technique. Evaluation focused on PII-level (recall, precision, and F1-score) and clinical note-level metrics (recall and BLEU [bilingual evaluation understudy] metric), assessing deidentification effectiveness and content preservation. Results The generative model Mistral 7B performed the highest with an overall F1-score of 0.9673 (vs 0.8750 for Llama 2 and 0.8686 for Mixtral 8×7B). At the clinical notes level, the model’s overall recall was 0.9326 (vs 0.6888 for Llama 2 and 0.6417 for Mixtral 8×7B). This rate increased to 0.9915 when Mistral 7B only deleted names. Four notes of 3000 failed to be fully pseudonymized for names: in 1 case, the nondeleted name belonged to a patient, while in the others, it belonged to medical staff. Beyond the fifth epoch, the BLEU score consistently exceeded 0.9864, indicating no significant text alteration. Conclusions Our research underscores the significant capabilities of generative natural language processing models, with Mistral 7B standing out for its superior ability to deidentify clinical texts efficiently. Achieving notable performance metrics, Mistral 7B operates effectively without requiring high-end computational resources. These methods pave the way for a broader availability of pseudonymized clinical texts, enabling their use for research purposes and the optimization of the health care system.

With pedestrian crossings implicated in a significant proportion of vehicle-pedestrian accidents and the French government's initiatives to improve pedestrian safety, there is a pressing need for efficient, large-scale evaluation of pedestrian crossings. This study proposes the deployment of advanced deep learning neural networks to automate the assessment of pedestrian crossings and roundabouts, leveraging aerial and street-level imagery sourced from Google Maps and Google Street View. Utilizing ConvNextV2, ResNet50, and ResNext50 models, we conducted a comprehensive analysis of pedestrian crossings across various urban and rural settings in France, focusing on nine identified risk factors.Our methodology incorporates Mask R-CNN for precise segmentation and detection of zebra crossings and roundabouts, overcoming traditional data annotation challenges and extending coverage to underrepresented areas. The analysis reveals that the ConvNextV2 model, in particular, demonstrates superior performance across most tasks, despite challenges such as data imbalance and the complex nature of variables like visibility and parking proximity.The findings highlight the potential of convolutional neural networks in improving pedestrian safety by enabling scalable and objective evaluations of crossings. The study underscores the necessity for continued dataset augmentation and methodological advancements to tackle identified challenges. Our research contributes to the broader field of road safety by demonstrating the feasibility and effectiveness of automated, image-based pedestrian crossing audits, paving the way for more informed and effective safety interventions.

OBJECTIVE:To determine the impact of pain, stress, and negative emotions like anger, sadness, fear and regret in the persistence or development of chronic pain four months after admission to the emergency department (ED). METHODS:Data from 641 ED patients in the SOFTER IV clinical trial were analyzed. Pain, stress, and negative emotions were assessed at discharge, then dichotomized as non-severe or severe. Patients with chronic pain history were included in the analysis. Chronic pain at four months was evaluated using a binary yes/no question, and its predictors were identified using multivariable logistic regression with variable selection based on statistical significance. RESULTS:At four months post-ED admission, 35.1 % of patients reported chronic pain. As expected, a prior history of chronic pain was a strong predictor. Among patients with no history of chronic pain, those who reported severe anger at discharge were at nearly three times the risk of developing chronic pain (OR = 2.8, 95 % CI: 1.4-5.6). In addition, patients admitted for traumatic injuries and female patients also showed elevated risk, with odds ratios of 1.7 (95 % CI: 1.2-2.4) and 1.4 (95 % CI: 1.0-2.0), respectively. CONCLUSION:Anger may affect the development or persistence of chronic pain after an emergency department admission.

Faced with the challenges of patient confidentiality and scientific reproducibility, research on machine learning for health is turning towards the conception of synthetic medical databases. This article presents a brief overview of state-of-the-art machine learning methods for generating synthetic tabular and textual data, focusing their application to the automatic classification of trauma mechanisms, followed by our proposed methodology for generating high-quality, synthetic medical records combining tabular and unstructured text data.

Emergency Rooms (ERs) are at the center of various optimization research due to the growing number of visits in recent decades. The accurate logging of patient movement and time spent in ERs is essential for studying patient pathways and improving operations. Despite the innovative digital tracking system employed at the Bordeaux University Hospital, over 90% of these logs suffer from missing or incoherent values, primarily due to manual entry errors and software changes. This study explores the application of Transformer neural networks for cleaning ER logs at the Bordeaux University Hospital, addressing the challenge of inaccuracies in patient movement records accumulated over ten years. By leveraging a T5 Transformer model, we aim to demonstrate the model's ability to correct these errors by transforming unreliable datasets into valuable resources for optimizing ER operations. Our findings reveal that the Transformer model can achieve an overall accuracy of 95.79% in identifying and correcting faulty entries, showcasing the potential of advanced machine learning techniques in enhancing data integrity for healthcare applications. This research highlights the effectiveness of Transformer networks in data cleaning tasks and opens up opportunities for their application in various fields that are burdened with corrupt sequential data.

Background/Objectives: Very few studies describe the various feelings experienced in the emergency department (ED). Our study describes the pain, stress, and negative and positive emotions experienced by patients admitted to the ED in relation to age, gender, and reason for ED admission. Methods: Patients admitted to the ED of seven French hospitals were surveyed as part of the randomised multicentre study SOFTER IV (n = 2846). They reported the intensity of their pain on a numerical rating scale of 0 to 10, the intensity of their stress on an equivalent scale, and their emotions on a five-point rating scale using an adapted version of the Geneva Emotion Wheel proposed by Scherer, based on eight core emotions: fear, anger, regret, sadness, relief, interest, joy, and satisfaction. Results: Patients reported an average pain rating of 4.5 (SD = 3.0) and an average stress rating of 3.4 (SD = 3.1). Forty-six percent reported at least one strong negative emotion, and the two most frequently reported were fear and sadness. Forty-seven percent of patients described feeling at least one strong positive emotion, and the two most frequently reported were interest and relief. Pain was significantly higher among female patients under 60 admitted for injury. Stress was significantly higher among female patients under 60 admitted for illness. Emotions of negative valency were significantly higher among women admitted for injury. Emotions of positive valency were significantly higher among men over 60 admitted for illness. Conclusions: Experiences of pain, stress, and emotions have a strong presence in the ED. The reporting of these feelings varies depending on age, gender, and reason for ED admission.

[This corrects the article DOI: 10.1016/j.heliyon.2025.e42428.].

We present a domain-agnostic paired-comparison approach that uses Large Language Models (LLMs) to quantify sex/gender-related asymmetries in documented clinical decision-making. The method trains an LLM to emulate observed decisions, then evaluates sex-swapped pairs in which only sex is flipped, holding documented clinical content constant. We apply it to emergency triage, analyzing more than 140,000 Bordeaux University Hospital (France) admissions and testing methodological portability on MIMIC-IV, spanning a different language, population, and healthcare system. Fine-tuning Mistral NeMo 12B for triage prediction and using Mistral Small 24B for pair generation, we find otherwise identical presentations were more likely to receive a lower-severity predicted score as female than male: 1.1

The surge in AI-based research for emergency healthcare presents challenges such as data protection compliance and the risk of exacerbating health inequalities. Human biases in demographic data used to train AI systems may indeed be replicated. Yet, AI also offers a chance for a paradigm shift, acting as a tool to counteract human biases.Our study focuses on emergency triage, swiftly categorizing patients by severity upon arrival. Objectives include conducting a literature review to identify potential human biases in triage and presenting a preliminary study. This involves a qualitative survey to complement the review on factors influencing triage scores. Additionally, we analyze triage data descriptively and pilot AI-driven triage using a Large Language Model model with data from University Hospital of Bordeaux. Finally, assembling these pieces, we outline an experimental plan to assess AI effectiveness in detecting biases in triage data.

BACKGROUND: Road accidents are the leading type of work-related fatalities, but the impact of work-related travel on overall traffic safety has been scarcely studied. OBJECTIVE: The main objective of the present study was to assess drivers’ relative road accident risk between work-related and personal journeys. METHODS: A responsible/non-responsible case-control study was performed on a sample of 7,051 road accidents in France from the VOIESUR project. Logistic regression determined odds-ratios according to work-related versus personal travel, and identified risk factors for responsibility, specific to each of the two sub-groups. RESULTS: Drivers traveling on duty or commuting home were significantly less often responsible for accidents than drivers on personal journeys: OR = 0.75 [0.63; 0.89] and 0.65 [0.53; 0.80] respectively. Responsibility was significantly more frequent in commuting to versus from work: OR = 1.38 [1.06; 1.78]. Among on-duty drivers, professional passenger-transport drivers had the lowest risk of responsibility (OR = 0.25 [0.11; 0.58]), while those on temporary or work/study contracts and professional light goods vehicle drivers had the highest risk (OR = 11.64 [2.15; 62.94] and OR = 29.83 [5.19; 171.38] respectively). When driving under the influence of alcohol, risk of responsibility was higher in commuting home than in personal journeys. CONCLUSION: On-duty drivers showed lower risk of responsibility for an accident than other drivers. However, on-duty drivers on temporary or work/study contracts, who are usually not subject to specific regulations, showed higher risk, and should be the subject of particular attention regarding occupational risk prevention.

The digitization of healthcare, facilitated by the adoption of electronic health record (EHR) systems, has revolutionized data-driven medical research and patient care. While this digital transformation offers substantial benefits in healthcare efficiency and accessibility, it concurrently raises significant concerns over privacy and data security. Initially, the journey towards protecting patient data de-identification saw the transition from rule-based systems to more mixed approaches including machine learning for de-identifying patient data. Subsequently, the emergence of Large Language Models (LLMs) has represented a further opportunity in this domain, offering unparalleled potential for enhancing the accuracy of context-sensitive de-identification. However, despite LLMs offering significant potential, the deployment of the most advanced models in hospital environments is frequently hindered by data security issues and the extensive hardware resources required. The objective of our study is to design, implement, and evaluate de-identification algorithms by employing fine-tuning of moderate-sized open-source language models, ensuring their suitability for production inference tasks on personal computers. We aimed at replacing personal identifying information (PII) with generic placeholders or labeling non-PII texts as 'ANONYMOUS', ensuring privacy while preserving textual integrity. Our dataset, derived from over 425,000 clinical notes from the adult emergency department of the Bordeaux University Hospital in France, underwent independent double annotation by two experts to create a reference for model validation with 3,000 clinical notes randomly selected. Three open-source language models of manageable size were selected for their feasibility in hospital settings: Llama 2 7B, Mistral 7B, and Mixtral 8x7B. Fine-tuning utilized the quantized Low-Rank Adaptation (qLoRA) technique. Evaluation focused on PII-level (Recall, Precision and F1-Score) and clinical note-level metrics (Recall and BLEU metric), assessing de-identification effectiveness and content preservation. The generative model Mistral 7B demonstrated the highest performance with an overall F1-score of 0.9673 (vs. 0.8750 for Llama 2 and 0.8686 for Mistral 8x7B). At the clinical notes level, the same model achieved an overall recall of 0.9326 (vs. 0.6888 for Llama 2 and 0.6417 for Mistral 8x7B).This rate increased to 0.9915 for the anonymization of names with Mistral 7B. Four notes out of the total 3000 failed to be fully anonymized for names: in one case, the non-anonymized name belonged to a patient, while in the other cases, it belonged to medical staff. Beyond the fifth epoch, the BLEU score consistently exceeded 0.9864, indicating no significant text alteration due to the process. Our research underscores the significant capabilities of generative NLP models, with Mistral 7B standing out for its superior ability to de-identify clinical texts efficiently. Achieving notable performance metrics, Mistral 7B operates effectively without requiring high-end computational resources. These methods pave the way for a broader availability of anonymized clinical texts, enabling their use for research purposes and the optimization of the healthcare system.

With pedestrian crossings implicated in a significant proportion of vehicle-pedestrian accidents and the French government's initiatives to improve pedestrian safety, there is a pressing need for efficient, large-scale evaluation of pedestrian crossings. This study proposes the deployment of advanced deep learning neural networks to automate the assessment of pedestrian crossings and roundabouts, leveraging aerial and street-level imagery sourced from Google Maps and Google Street View. Utilizing ConvNextV2, ResNet50, and ResNext50 models, we conducted a comprehensive analysis of pedestrian crossings across various urban and rural settings in France, focusing on nine identified risk factors.Our methodology incorporates Mask R-CNN for precise segmentation and detection of zebra crossings and roundabouts, overcoming traditional data annotation challenges and extending coverage to underrepresented areas. The analysis reveals that the ConvNextV2 model, in particular, demonstrates superior performance across most tasks, despite challenges such as data imbalance and the complex nature of variables like visibility and parking proximity.The findings highlight the potential of convolutional neural networks in improving pedestrian safety by enabling scalable and objective evaluations of crossings. The study underscores the necessity for continued dataset augmentation and methodological advancements to tackle identified challenges. Our research contributes to the broader field of road safety by demonstrating the feasibility and effectiveness of automated, image-based pedestrian crossing audits, paving the way for more informed and effective safety interventions.

To examine the risk factors for severe pain upon discharge from the emergency department, assuming appropriate pharmacological treatment of pain, in order to improve pain relief in emergency departments and reduce the risk of potential chronic pain. An analytic study was conducted utilizing data from a multicenter randomized controlled trial to evaluate patients’ experiences upon admission and discharge from the emergency department (ED). Severe pain was defined by a score of six on a numerical rating scale of zero to ten. Stress and negative emotions (including anger, fear, sadness, and regret) were evaluated using numerical rating scales, respectively ranging from 0 to 10 and 1 to 5. The risk factors of severe pain at discharge (SPD) from ED were calculated using logistic regression considering patient characteristics evaluated at their admission to the ED. From the 1240 patients analyzed, 22.2

There is a burgeoning interest in harnessing artificial intelligence (AI) to enhance patient flow within emergency departments (EDs). However, this advancement is accompanied by a significant risk: by relying on historical healthcare data, these AI tools may perpetuate existing systemic biases associated with gender, age, ethnicity, and socioeconomic status. This paper surveys studies identifying biases in ED data, offering context for concern about these biases. These insights are valuable for researchers developing AI to optimize ED workflows while accounting for ethical considerations.

The emergence of artificial intelligence in healthcare is probably leading to another two-speed world. On one hand, widely accessible AI applications such as language models are becoming ubiquitous, while on the other, resource-intensive technologies like robotic surgery and personalized medicine will be reserved for a privileged few. This development signifies a growing disparity in access to AI advancements. The paper also discusses the inevitability of widespread automated medical consultation, and the need for a quality assurance system to oversee the burgeoning use of AI in healthcare.

Judgment biases in emergency triage can adversely affect patient outcomes. This study examines sex/gender biases using four advanced language models fine-tuned on real-world emergency department data. We introduce a novel approach based on the testing method, commonly used in hiring bias detection, by automatically altering triage notes to change patient sex references. Results indicate a significant bias: female patients are assigned lower severity ratings than male patients with identical clinical conditions. This bias is more pronounced with female nurses or when patients report higher pain levels but diminishes with increased nurse experience. Identifying these biases can inform interventions such as enhanced training, protocol updates, and machine learning tools to support clinical decision-making.