Ramy Azzouz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ramy Azzouz has a total of 58 co-authors. Here are the top co-authors by number of shared publications:
- Amélie Vromant
Emergency Department, Hôpital Pitié-Salpêtrière, Assistance Publique des Hôpitaux de Paris, Paris, France
3 shared publications - Edouard Lansiaux
Emergency Department, Lille University Hospital, Lille, France
3 shared publications - Emmanuel Chazard
University of Lille, Lille University Hospital, ULR 2694 METRICS, CERIM, Department of Public Health , Lille
3 shared publications - Eric Wiel
Emergency Department, Lille University Hospital, Lille, France
3 shared publications - Camille Paradis
Centre Antipoison de Bordeaux, GH PELLEGRIN
2 shared publications - Christine Tournoud
East Centre antipoison
2 shared publications
Latest Publications
- 19 March 2026
- Published by BMJ in Emergency Medicine Journal
- Vol. 43 (8) , 493-499
Spatial clustering of out-of-hospital cardiac arrest in northern France and its association with social deprivation: a population-based registry study
Background Out-of-hospital cardiac arrest (OHCA) shows marked geographic variability. Socio-economic deprivation may contribute to this variability, but evidence from mixed urban–rural regions remains limited. We aimed to identify spatial clusters of OHCA incidence in northern France and to assess whether these clusters were associated with deprivation, care processes and outcomes. Methods We conducted a retrospective, population-based cohort study using data from the French National Out-of-Hospital Cardiac Arrest Registry for all OHCAs managed by mobile intensive care units between 1 July 2015 and 30 June 2016 in Nord-Pas-de-Calais. Age-adjusted and sex-adjusted standardised incidence ratios were smoothed with a Besag-York-Mollié Bayesian model. Spatial scan statistics identified incidence clusters. Cluster-level socioeconomic indicators, care processes and outcomes were compared Results Incidence was mapped across 1541 municipalities. Among 2867 OHCAs, mean annual OHCA incidence was 68.8 per 100 000 inhabitants (range 1.5 across municipalities). Two high-incidence clusters (relative risk between 1.61 and 1.68) and three low-incidence clusters (relative risk between 0.48 and 0.66) were detected. High-incidence clusters displayed greater deprivation and lower median income. Bystander basic life support was less frequent (26.7%) and 12-month survival lowest (1.4%) in the most deprived cluster. Conclusions Pronounced spatial inequality in OHCA burden exists within northern France. Municipalities with higher deprivation experience disproportionately higher incidence and poorer long-term survival. Targeted cardiopulmonary resuscitation training and strategic automated external defibrillator deployment should prioritise these vulnerable communities.
- 19 March 2026
- Published by BMJ in Emergency Medicine Journal
- Vol. 43 (8) , 493-499
Spatial clustering of out-of-hospital cardiac arrest in northern France and its association with social deprivation: a population-based registry study
Background Out-of-hospital cardiac arrest (OHCA) shows marked geographic variability. Socio-economic deprivation may contribute to this variability, but evidence from mixed urban–rural regions remains limited. We aimed to identify spatial clusters of OHCA incidence in northern France and to assess whether these clusters were associated with deprivation, care processes and outcomes. Methods We conducted a retrospective, population-based cohort study using data from the French National Out-of-Hospital Cardiac Arrest Registry for all OHCAs managed by mobile intensive care units between 1 July 2015 and 30 June 2016 in Nord-Pas-de-Calais. Age-adjusted and sex-adjusted standardised incidence ratios were smoothed with a Besag-York-Mollié Bayesian model. Spatial scan statistics identified incidence clusters. Cluster-level socioeconomic indicators, care processes and outcomes were compared Results Incidence was mapped across 1541 municipalities. Among 2867 OHCAs, mean annual OHCA incidence was 68.8 per 100 000 inhabitants (range 1.5 across municipalities). Two high-incidence clusters (relative risk between 1.61 and 1.68) and three low-incidence clusters (relative risk between 0.48 and 0.66) were detected. High-incidence clusters displayed greater deprivation and lower median income. Bystander basic life support was less frequent (26.7%) and 12-month survival lowest (1.4%) in the most deprived cluster. Conclusions Pronounced spatial inequality in OHCA burden exists within northern France. Municipalities with higher deprivation experience disproportionately higher incidence and poorer long-term survival. Targeted cardiopulmonary resuscitation training and strategic automated external defibrillator deployment should prioritise these vulnerable communities.
- Top Cited PapersOpen Access
- 10 March 2026
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 14, e83318
Artificial Intelligence Models for Predicting Triage in Emergency Departments: Seven-Month Retrospective Comparative Study of Natural Language Processing, Large Language Model, and Joint Embedding Predictive Architectures
Background: Triage errors in emergency departments (EDs), including undertriage and overtriage, pose significant risks to patient safety and resource allocation. With increasing patient volumes and staffing challenges, artificial intelligence (AI) integration into triage protocols has gained attention as a potential solution. Objective: This study aims to develop and compare 3 AI models—natural language processing (NLP), large language model (LLM), and Joint Embedding Predictive Architecture (JEPA)—for predicting triage outcomes according to the French Emergency Nurses Classification in Hospital (FRENCH) scale and to assess their performance relative to nurse triage and clinical expert consensus. Methods: We conducted a retrospective analysis of prospectively collected data from adult patients triaged at Roger Salengro Hospital ED (Lille, France) over 7 months (June-December 2024). Three AI models were developed: TRIAGEMASTER (NLP with Doc2Vec + MLP), URGENTIAPARSE (LLM with FlauBERT + Extreme Gradient Boosting [XGBoost]), and EMERGINET (JEPA with variance-invariance-covariance regularization). Of 73,236 ED visits, 657 (0.90%) had complete audio recordings and structured data. Data were split 80:20 into training and validation sets with stratification. Gold-standard labels were established by senior clinician consensus (minimum 5 years of ED experience). The primary outcome was concordance with the gold-standard FRENCH triage level, assessed using weighted κ, Spearman correlation, F1-score, area under the receiver operating characteristic (AUC-ROC) curve, mean absolute error (MAE), and root mean square error (RMSE). Secondary analyses evaluated Groupes d’Etude Multicentrique des Services d’Accueil (GEMSA) prediction and performance by input data type. Results: URGENTIAPARSE demonstrated superior performance, with a composite z score of 2.514 compared with EMERGINET (0.438), TRIAGEMASTER (–3.511), and nurse triage (–4.343). URGENTIAPARSE achieved an F1-score of 0.900 (95% CI 0.876-0.924), an AUC-ROC of 0.879 (95% CI 0.851-0.907), a weighted κ of 0.800 (P<.001), a Spearman correlation of 0.802 (P<.001), an MAE of 0.228, and an RMSE of 0.790. Exact agreement was 90.0%, with near-agreement (+1 or –1 level) of 92.8%. However, training showed perfect accuracy (1.0) with poor validation performance (~0.5), indicating overfitting. EMERGINET achieved moderate performance (F1-score=0.731, AUC 0.686), while TRIAGEMASTER and nurse triage performed poorly (F1-score=0.618 and 0.303, respectively). For GEMSA prediction, URGENTIAPARSE maintained superiority (κ=0.863, Spearman=0.864, P<.001). Class 1 (highest acuity) was underrepresented (4/657, 0.61%), limiting undertriage risk assessment. Conclusions: The LLM-based architecture (URGENTIAPARSE) demonstrated the highest accuracy for ED triage prediction among the tested models, outperforming traditional NLP, JEPA, and current nurse triage practices. However, severe overfitting, extreme selection bias (657/73,236, 0.90%, inclusion), a monocentric design, and sparse high-acuity representation limit clinical applicability. Before deployment, the model requires regularization, external validation across diverse EDs, prospective testing, and comprehensive safety evaluation, particularly for undertriage detection. Integration of AI triage support systems shows promise but demands rigorous validation, bias mitigation, and transparent uncertainty quantification to ensure patient safety.
- 10 March 2026
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 14, e83318
Top Cited PapersOpen AccessArtificial Intelligence Models for Predicting Triage in Emergency Departments: Seven-Month Retrospective Comparative Study of Natural Language Processing, Large Language Model, and Joint Embedding Predictive Architectures
Background: Triage errors in emergency departments (EDs), including undertriage and overtriage, pose significant risks to patient safety and resource allocation. With increasing patient volumes and staffing challenges, artificial intelligence (AI) integration into triage protocols has gained attention as a potential solution. Objective: This study aims to develop and compare 3 AI models—natural language processing (NLP), large language model (LLM), and Joint Embedding Predictive Architecture (JEPA)—for predicting triage outcomes according to the French Emergency Nurses Classification in Hospital (FRENCH) scale and to assess their performance relative to nurse triage and clinical expert consensus. Methods: We conducted a retrospective analysis of prospectively collected data from adult patients triaged at Roger Salengro Hospital ED (Lille, France) over 7 months (June-December 2024). Three AI models were developed: TRIAGEMASTER (NLP with Doc2Vec + MLP), URGENTIAPARSE (LLM with FlauBERT + Extreme Gradient Boosting [XGBoost]), and EMERGINET (JEPA with variance-invariance-covariance regularization). Of 73,236 ED visits, 657 (0.90%) had complete audio recordings and structured data. Data were split 80:20 into training and validation sets with stratification. Gold-standard labels were established by senior clinician consensus (minimum 5 years of ED experience). The primary outcome was concordance with the gold-standard FRENCH triage level, assessed using weighted κ, Spearman correlation, F1-score, area under the receiver operating characteristic (AUC-ROC) curve, mean absolute error (MAE), and root mean square error (RMSE). Secondary analyses evaluated Groupes d’Etude Multicentrique des Services d’Accueil (GEMSA) prediction and performance by input data type. Results: URGENTIAPARSE demonstrated superior performance, with a composite z score of 2.514 compared with EMERGINET (0.438), TRIAGEMASTER (–3.511), and nurse triage (–4.343). URGENTIAPARSE achieved an F1-score of 0.900 (95% CI 0.876-0.924), an AUC-ROC of 0.879 (95% CI 0.851-0.907), a weighted κ of 0.800 (P<.001), a Spearman correlation of 0.802 (P<.001), an MAE of 0.228, and an RMSE of 0.790. Exact agreement was 90.0%, with near-agreement (+1 or –1 level) of 92.8%. However, training showed perfect accuracy (1.0) with poor validation performance (~0.5), indicating overfitting. EMERGINET achieved moderate performance (F1-score=0.731, AUC 0.686), while TRIAGEMASTER and nurse triage performed poorly (F1-score=0.618 and 0.303, respectively). For GEMSA prediction, URGENTIAPARSE maintained superiority (κ=0.863, Spearman=0.864, P<.001). Class 1 (highest acuity) was underrepresented (4/657, 0.61%), limiting undertriage risk assessment. Conclusions: The LLM-based architecture (URGENTIAPARSE) demonstrated the highest accuracy for ED triage prediction among the tested models, outperforming traditional NLP, JEPA, and current nurse triage practices. However, severe overfitting, extreme selection bias (657/73,236, 0.90%, inclusion), a monocentric design, and sparse high-acuity representation limit clinical applicability. Before deployment, the model requires regularization, external validation across diverse EDs, prospective testing, and comprehensive safety evaluation, particularly for undertriage detection. Integration of AI triage support systems shows promise but demands rigorous validation, bias mitigation, and transparent uncertainty quantification to ensure patient safety.
- Open Access
- 11 September 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Artificial Intelligence Models for Predicting Triage in Emergency Departments: Seven-Month Retrospective Comparative Study of Natural Language Processing, Large Language Model, and Joint Embedding Predictive Architectures (Preprint)
BACKGROUND Triage errors in emergency departments (EDs), including undertriage and overtriage, pose significant risks to patient safety and resource allocation. With increasing patient volumes and staffing challenges, artificial intelligence (AI) integration into triage protocols has gained attention as a potential solution. OBJECTIVE This study aims to develop and compare 3 AI models—natural language processing (NLP), large language model (LLM), and Joint Embedding Predictive Architecture (JEPA)—for predicting triage outcomes according to the French Emergency Nurses Classification in Hospital (FRENCH) scale and to assess their performance relative to nurse triage and clinical expert consensus. METHODS We conducted a retrospective analysis of prospectively collected data from adult patients triaged at Roger Salengro Hospital ED (Lille, France) over 7 months (June-December 2024). Three AI models were developed: TRIAGEMASTER (NLP with Doc2Vec + MLP), URGENTIAPARSE (LLM with FlauBERT + Extreme Gradient Boosting [XGBoost]), and EMERGINET (JEPA with variance-invariance-covariance regularization). Of 73,236 ED visits, 657 (0.90%) had complete audio recordings and structured data. Data were split 80:20 into training and validation sets with stratification. Gold-standard labels were established by senior clinician consensus (minimum 5 years of ED experience). The primary outcome was concordance with the gold-standard FRENCH triage level, assessed using weighted κ, Spearman correlation, F1-score, area under the receiver operating characteristic (AUC-ROC) curve, mean absolute error (MAE), and root mean square error (RMSE). Secondary analyses evaluated Groupes d’Etude Multicentrique des Services d’Accueil (GEMSA) prediction and performance by input data type. RESULTS URGENTIAPARSE demonstrated superior performance, with a composite z score of 2.514 compared with EMERGINET (0.438), TRIAGEMASTER (–3.511), and nurse triage (–4.343). URGENTIAPARSE achieved an F1-score of 0.900 (95% CI 0.876-0.924), an AUC-ROC of 0.879 (95% CI 0.851-0.907), a weighted κ of 0.800 (P<.001), a Spearman correlation of 0.802 (P<.001), an MAE of 0.228, and an RMSE of 0.790. Exact agreement was 90.0%, with near-agreement (+1 or –1 level) of 92.8%. However, training showed perfect accuracy (1.0) with poor validation performance (~0.5), indicating overfitting. EMERGINET achieved moderate performance (F1-score=0.731, AUC 0.686), while TRIAGEMASTER and nurse triage performed poorly (F1-score=0.618 and 0.303, respectively). For GEMSA prediction, URGENTIAPARSE maintained superiority (κ=0.863, Spearman=0.864, P<.001). Class 1 (highest acuity) was underrepresented (4/657, 0.61%), limiting undertriage risk assessment. CONCLUSIONS The LLM-based architecture (URGENTIAPARSE) demonstrated the highest accuracy for ED triage prediction among the tested models, outperforming traditional NLP, JEPA, and current nurse triage practices. However, severe overfitting, extreme selection bias (657/73,236, 0.90%, inclusion), a monocentric design, and sparse high-acuity representation limit clinical applicability. Before deployment, the model requires regularization, external validation across diverse EDs, prospective testing, and comprehensive safety evaluation, particularly for undertriage detection. Integration of AI triage support systems shows promise but demands rigorous validation, bias mitigation, and transparent uncertainty quantification to ensure patient safety.
- 11 September 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessArtificial Intelligence Models for Predicting Triage in Emergency Departments: Seven-Month Retrospective Comparative Study of Natural Language Processing, Large Language Model, and Joint Embedding Predictive Architectures (Preprint)
BACKGROUND Triage errors in emergency departments (EDs), including undertriage and overtriage, pose significant risks to patient safety and resource allocation. With increasing patient volumes and staffing challenges, artificial intelligence (AI) integration into triage protocols has gained attention as a potential solution. OBJECTIVE This study aims to develop and compare 3 AI models—natural language processing (NLP), large language model (LLM), and Joint Embedding Predictive Architecture (JEPA)—for predicting triage outcomes according to the French Emergency Nurses Classification in Hospital (FRENCH) scale and to assess their performance relative to nurse triage and clinical expert consensus. METHODS We conducted a retrospective analysis of prospectively collected data from adult patients triaged at Roger Salengro Hospital ED (Lille, France) over 7 months (June-December 2024). Three AI models were developed: TRIAGEMASTER (NLP with Doc2Vec + MLP), URGENTIAPARSE (LLM with FlauBERT + Extreme Gradient Boosting [XGBoost]), and EMERGINET (JEPA with variance-invariance-covariance regularization). Of 73,236 ED visits, 657 (0.90%) had complete audio recordings and structured data. Data were split 80:20 into training and validation sets with stratification. Gold-standard labels were established by senior clinician consensus (minimum 5 years of ED experience). The primary outcome was concordance with the gold-standard FRENCH triage level, assessed using weighted κ, Spearman correlation, F1-score, area under the receiver operating characteristic (AUC-ROC) curve, mean absolute error (MAE), and root mean square error (RMSE). Secondary analyses evaluated Groupes d’Etude Multicentrique des Services d’Accueil (GEMSA) prediction and performance by input data type. RESULTS URGENTIAPARSE demonstrated superior performance, with a composite z score of 2.514 compared with EMERGINET (0.438), TRIAGEMASTER (–3.511), and nurse triage (–4.343). URGENTIAPARSE achieved an F1-score of 0.900 (95% CI 0.876-0.924), an AUC-ROC of 0.879 (95% CI 0.851-0.907), a weighted κ of 0.800 (P<.001), a Spearman correlation of 0.802 (P<.001), an MAE of 0.228, and an RMSE of 0.790. Exact agreement was 90.0%, with near-agreement (+1 or –1 level) of 92.8%. However, training showed perfect accuracy (1.0) with poor validation performance (~0.5), indicating overfitting. EMERGINET achieved moderate performance (F1-score=0.731, AUC 0.686), while TRIAGEMASTER and nurse triage performed poorly (F1-score=0.618 and 0.303, respectively). For GEMSA prediction, URGENTIAPARSE maintained superiority (κ=0.863, Spearman=0.864, P<.001). Class 1 (highest acuity) was underrepresented (4/657, 0.61%), limiting undertriage risk assessment. CONCLUSIONS The LLM-based architecture (URGENTIAPARSE) demonstrated the highest accuracy for ED triage prediction among the tested models, outperforming traditional NLP, JEPA, and current nurse triage practices. However, severe overfitting, extreme selection bias (657/73,236, 0.90%, inclusion), a monocentric design, and sparse high-acuity representation limit clinical applicability. Before deployment, the model requires regularization, external validation across diverse EDs, prospective testing, and comprehensive safety evaluation, particularly for undertriage detection. Integration of AI triage support systems shows promise but demands rigorous validation, bias mitigation, and transparent uncertainty quantification to ensure patient safety.
- 28 February 2024
- Published by Taylor & Francis in Clinical Toxicology
- Vol. 62 (2) , 107-111
Exposure to ammonia solution due to substance use: a retrospective study from the French poison centres database (2009–2018)
Ammonia solution (ammonium hydroxide) is used to convert cocaine hydrochloride to freebase cocaine. Due to its causticity, unintentional exposure to ammonia in a substance use context can result in injury. The objective of this study was to describe the characteristics of unintentional oral and buccal ammonia solution exposure in a substance use context. A retrospective study was conducted using the French poison centres database over a 10-year period (2009-2018). A total of 1,546 files were extracted, and 263 substance users were included. There was a significant increase in the number of these exposures between 2009 and 2018. Unintentional ingestion of ammonia solution was linked to product decanting in 89 per cent of cases. Substance use prior to the exposure and a festive context, such as free parties or teknivals, were identified in 25 per cent and 21 per cent of cases, respectively. Patients received a hospital examination in 87 per cent of cases. The severity of intoxication in substance users was graded as moderate (33 per cent) or severe (15 per cent) using the Poisoning Severity Score. The increased number of ammonia solution cases reported was consistent with an increase in the number of crack users in Europe in the same period. Ammonia solution exposure can suggest the possibility of substance use disorders. In such cases, patients can be referred to receive appropriate treatment and support. This study had some limitations, such as the lack of available information due to the retrospective nature of the study and the non-standardized questions asked by the poison centre during the medical phone interviews. Oral and buccal ammonia solution exposure in known substance users in France increased between 2009 and 2018. These users were mostly young men. A festive context and decanting were frequent. Patients were mainly referred to emergency departments to receive clinical examination and care. The potential severity of oral or buccal ammonia solution exposure in substance users requires increased vigilance among all healthcare professionals involved in the management of these intoxication cases.
- 28 February 2024
- Published by Taylor & Francis in Clinical Toxicology
- Vol. 62 (2) , 107-111
Exposure to ammonia solution due to substance use: a retrospective study from the French poison centres database (2009–2018)
Ammonia solution (ammonium hydroxide) is used to convert cocaine hydrochloride to freebase cocaine. Due to its causticity, unintentional exposure to ammonia in a substance use context can result in injury. The objective of this study was to describe the characteristics of unintentional oral and buccal ammonia solution exposure in a substance use context. A retrospective study was conducted using the French poison centres database over a 10-year period (2009-2018). A total of 1,546 files were extracted, and 263 substance users were included. There was a significant increase in the number of these exposures between 2009 and 2018. Unintentional ingestion of ammonia solution was linked to product decanting in 89 per cent of cases. Substance use prior to the exposure and a festive context, such as free parties or teknivals, were identified in 25 per cent and 21 per cent of cases, respectively. Patients received a hospital examination in 87 per cent of cases. The severity of intoxication in substance users was graded as moderate (33 per cent) or severe (15 per cent) using the Poisoning Severity Score. The increased number of ammonia solution cases reported was consistent with an increase in the number of crack users in Europe in the same period. Ammonia solution exposure can suggest the possibility of substance use disorders. In such cases, patients can be referred to receive appropriate treatment and support. This study had some limitations, such as the lack of available information due to the retrospective nature of the study and the non-standardized questions asked by the poison centre during the medical phone interviews. Oral and buccal ammonia solution exposure in known substance users in France increased between 2009 and 2018. These users were mostly young men. A festive context and decanting were frequent. Patients were mainly referred to emergency departments to receive clinical examination and care. The potential severity of oral or buccal ammonia solution exposure in substance users requires increased vigilance among all healthcare professionals involved in the management of these intoxication cases.
- 2 December 2023
- Published by Taylor & Francis in Clinical Toxicology
- Vol. 61 (12) , 1047-1054
Pre-hospital triage of children at risk of oesophageal button battery impaction: the button battery impaction score
Button battery ingestion in children can be fatal if oesophageal perforation occurs. Such children require chest radiography in the emergency department to determine the button battery position and number. Current guidelines recommend that a button battery impacted in the oesophagus should be removed within two hours. We developed a clinical tool (the button battery impaction score) to estimate the risk of oesophageal impaction and help determine the most appropriate healthcare facility for initial assessment, either a local medical centre or a medical centre with the infrastructure for endoscopic retrieval. A multi-centre retrospective study was conducted over seven years in eight French poison centres. We included patients aged less than 12 years with radiography showing the button battery position and a symptom description before radiography. Button battery impaction scores were calculated using backward stepwise selection. A total of 1,430 patients were included, of whom 86, 461, and 375 had a button battery in their oesophagus, stomach, and post-pyloric position, respectively. No button batteries were identified by radiography in 508 patients. Sixteen of thirty-five factors independently predicted oesophageal impaction before chest radiography (P < 0.05). After the backward stepwise selection, the following seven factors contributed to the button battery impaction score: cough, drooling, dysphagia/food refusal, fever, pain (unspecified location), vomiting, and button battery ≥ 15 mm. The button battery impaction score showed an area under the curve value of 0.87, a negative predictive value of 0.98, and a sensitivity of 0.86. No cases of death, stricture, or haemorrhage were observed in patients with negative scores, including those with oesophageal impaction. A button battery impaction score used readily available data to predict the risk of oesophageal impaction after button battery ingestion and before chest radiography. When further validated, this rapid tool may be widely applicable in determining an appropriate facility for patient transfer to either a local medical centre or a medical centre with the infrastructure for endoscopic retrieval.
- 2 December 2023
- Published by Taylor & Francis in Clinical Toxicology
- Vol. 61 (12) , 1047-1054
Pre-hospital triage of children at risk of oesophageal button battery impaction: the button battery impaction score
Button battery ingestion in children can be fatal if oesophageal perforation occurs. Such children require chest radiography in the emergency department to determine the button battery position and number. Current guidelines recommend that a button battery impacted in the oesophagus should be removed within two hours. We developed a clinical tool (the button battery impaction score) to estimate the risk of oesophageal impaction and help determine the most appropriate healthcare facility for initial assessment, either a local medical centre or a medical centre with the infrastructure for endoscopic retrieval. A multi-centre retrospective study was conducted over seven years in eight French poison centres. We included patients aged less than 12 years with radiography showing the button battery position and a symptom description before radiography. Button battery impaction scores were calculated using backward stepwise selection. A total of 1,430 patients were included, of whom 86, 461, and 375 had a button battery in their oesophagus, stomach, and post-pyloric position, respectively. No button batteries were identified by radiography in 508 patients. Sixteen of thirty-five factors independently predicted oesophageal impaction before chest radiography (P < 0.05). After the backward stepwise selection, the following seven factors contributed to the button battery impaction score: cough, drooling, dysphagia/food refusal, fever, pain (unspecified location), vomiting, and button battery ≥ 15 mm. The button battery impaction score showed an area under the curve value of 0.87, a negative predictive value of 0.98, and a sensitivity of 0.86. No cases of death, stricture, or haemorrhage were observed in patients with negative scores, including those with oesophageal impaction. A button battery impaction score used readily available data to predict the risk of oesophageal impaction after button battery ingestion and before chest radiography. When further validated, this rapid tool may be widely applicable in determining an appropriate facility for patient transfer to either a local medical centre or a medical centre with the infrastructure for endoscopic retrieval.
- 1 December 2022
- Published by Elsevier in Journal d'imagerie diagnostique et interventionnelle
- Vol. 6 (1) , 7-11
Comment je fais une formation par la simulation pour les internes du DES de radiologie ? Exemple de la prise en charge d’une alerte AVC
- 1 December 2022
- Published by Elsevier in Journal d'imagerie diagnostique et interventionnelle
- Vol. 6 (1) , 7-11
Comment je fais une formation par la simulation pour les internes du DES de radiologie ? Exemple de la prise en charge d’une alerte AVC