BACKGROUND Artificial intelligence (AI) is increasingly being integrated into emergency department (ED) workflows to assist with time-sensitive decision-making, documentation, diagnostics, and education. AI encompasses multiple computational approaches, including traditional machine learning (ML), deep learning (DL), and large language models (LLMs). We aim to provide insight into the current integration of AI into the multiple clinical spheres of the emergency medicine.

METHODS This narrative review analyzes available literature on the implementation of ML-based predictive systems, DL-based image and signal interpretation tools, and LLM-driven documentation and clinical reasoning support within emergency medicine.

RESULTS A comprehensive literature search was conducted across PubMed and SCOPUS from its inception to October 30, 2025. Overall, 189 articles were found, among them 51 were included in the final review. ML and DL models demonstrated strong performance in electrocardiogram interpretation, radiographic triage, and sepsis prediction, in some cases outperforming traditional clinical scoring tools. LLMs showed promise in documentation support, discharge summary generation, triage assistance, and educational applications; however, concerns remain regarding generalizability and clinical reliability. AI-assisted triage systems improved prioritization and time-to-provider metrics in selected settings but require further validation.

CONCLUSION AI holds substantial potential to augment emergency care delivery. Nevertheless, issues of transparency, bias, accountability, and human oversight remain critical. Current evidence supports AI as a clinical support tool rather than a replacement for physician judgment.

BACKGROUND Acute Physiology and Chronic Health Evaluation (APACHE) scores are widely used in emergency medicine and critical care settings, yet their prognostic accuracy for sepsis-related mortality across different versions remains to be assessed thoroughly. This study aimed to compare the pooled prognostic performance of APACHE II, III, and IV in predicting in-hospital mortality in patients with sepsis.

METHODS We conducted a systematic review and meta-analysis by searching for articles in the PubMed, Embase, Web of Science, and Cochrane databases, with a publication date range from February 23, 2016, to June 8, 2024. Random effects models were employed to analyse pooled data, focusing on key metrics such as the area under the curve (AUC), diagnostic odds ratio (DOR), sensitivity, specificity, positive likelihood ratio (PLR), and negative likelihood ratio (NLR).

RESULTS A total of 33 articles were analyzed, 3 of which examined multiple APACHE scores. These articles collectively involved 60,041 patients: 25 articles with 9,619 patients assessed APACHE II, 4 articles with 32,862 patients assessed APACHE III, and 8 articles with 21,741 patients assessed APACHE IV. The AUC for in-hospital mortality improved progressively across versions: 0.75 (95% CI 0.66-0.83) for APACHE II, 0.81 (95% CI 0.73-0.89) for APACHE III, and 0.85 (95% CI: 0.78-0.92) for APACHE IV. APACHE IV showed numerically more favorable pooled prognostic estimates, with a DOR of 2.76 (95% CI 2.08-3.78), sensitivity of 0.78 (95% CI 0.66-0.86), specificity of 0.83 (95% CI 0.68-0.92), PLR of 4.29 (95% CI 2.33-7.89), and NLR of 0.28 (95% CI 0.17-0.46).

CONCLUSION APACHE IV demonstrated the superior prognostic accuracy to the other versions for sepsis-related mortality. With the increasing adoption of automated calculations through standardized software, APACHE IV might be incorporated into routine clinical workflows.

BACKGROUND Predicting mortality in elderly traumatic brain injury (TBI) patients is challenging. The Geriatric Trauma Outcome Score (GTOS) may underestimate risk in moderate to severe cases. We hypothesized that integrating the Glasgow Coma Scale (GCS) and the Rotterdam CT Score (RCS) into GTOS would enhance mortality prediction.

METHODS This retrospective cohort study included trauma patients aged ≥65 years with head Abbreviated Injury Scale (AIS) ≥3 and admission GCS ≤12, treated at a tertiary trauma center between January 2016 and December 2024. Three models were assessed: (1) GTOS only; (2) GTOS+GCS; and (3) GTOS+RCS, with GTOS logit as an offset for recalibration. Discrimination and calibration were evaluated using the area under the receiver operating characteristic curve (AUC), Brier score, calibration intercept, and slope. Internal validation of the models were performed using 200 bootstrap samples. Net clinical benefit was compared using decision curve analysis (DCA).

RESULTS A total of 171 patients were included, of whom 111 survived and 60 did not. The GTOS model showed poor performance (AUC=0.538, 95%CI 0.442-0.664; Brier score 0.292) of prediction of 30-day mortality. Adding GCS improved model performance (AUC=0.721, 95%CI 0.652-0.793; Brier score 0.209). The GTOS+RCS model showed superior results (AUC=0.883, 95%CI 0.843-0.934; Brier score 0.134), with the greatest net benefit in DCA.

CONCLUSION Integrating RCS into GTOS enhances mortality prediction accuracy and clinical utility for elderly TBI patients. The combined model demonstrates superior discrimination, calibration, and decision analytic benefit, suggesting the need for prospective validation in clinical settings

BACKGROUND The transfusion of uncross-matched O-type red blood cells (URBCs) is the global standard of care for emergency resuscitation in trauma. However, its application in China remains limited due to concerns over safety and resource availability. This study aimed to evaluate the clinical efficacy and safety of URBCs in patients with severe pelvic fractures.

METHODS In this multicenter, retrospective cohort study (2019-2023), patients with severe pelvic fractures (Abbreviated Injury Scale [AIS]≥3) transfused within 24 h of admission were stratified into the URBC and non-URBC groups. Propensity score overlap weighting (PSOW) was used to address confounding factors. The primary outcome was 30-day in-hospital mortality. Secondary outcomes included length of stay (LOS), 24-hour blood product consumption, and in-hospital complications.

RESULTS A total of 526 patients were included (URBC group, n=57; non-URBC group, n=469). After PSOW adjustment, there was no significant difference in 30-day mortality (21.4% vs. 21.3%, P=0.989) between the URBC and non-URBC groups. However, the URBC group received significantly lower volumes of plasma, platelets, and cryoprecipitate within the first 24 h (all P<0.05). No clinically significant acute hemolytic transfusion reactions were observed during the study period. Furthermore, subgroup analysis of non-O-type blood recipients (57.1%) showed no significant differences in mortality compared with their O-type counterparts, and no acute hemolytic transfusion reactions were observed.

CONCLUSION Early administration of URBCs for severe pelvic fractures is safe. While not associated with increased mortality or hemolytic transfusion reactions, early use of URBCs was linked to a significant reduction in coagulation-related blood product consumption. This study provides evidence supporting the implementation of URBC within resource-constrained trauma setting to improve resuscitation efficiency.

BACKGROUND Prehospital red blood cell concentrate (RBC) transfusion is clinically advantageous and has an impact on all-cause mortality, especially in cases of major traumatic hemorrhage. The most frequent mean of prehospital transportation for blood and blood products is the helicopter. However, data on how external conditions during the helicopter flight affect the RBC quality are currently insufficient. There are also no data on how long the latter can be transported by helicopter without qualitative changes. This study aims to investigate how helicopter flight conditions affect RBC quality and to determine the maximum safe transport duration.

METHODS Four pairs of RBCs from different donors were taken. Each RBC pair was first mixed and homogenised, and then divided into two parts. The first was transported by helicopter for 22 successive days and the second served as control and was stored at 2 °C-6 °C. On days 1, 8, 15, and 22, samples of each RBC were taken, and hemolysis rates were calculated. A final sample was taken on day 36, which also marked the expiry date of the RBCs.

RESULTS The study maintained stable storage temperatures within the ideal range throughout helicopter transport. Despite variable conditions including number of starts/landings, flight time, outside temperatures, wind speed, and the cabin temperature, the hemolysis rates of transported RBCs did not differ significantly from those of control RBCs (median 0.061% vs. 0.066%, P>0.4). Furthermore, after 14 days of post-transport storage (day 36), hemolysis rates in both groups remained well within acceptable limits (0.151% vs. 0.133%).

CONCLUSION External conditions during helicopter transport for 22 d do not affect the RBC quality in terms of hemoglobin, hematocrit, free hemoglobin and hemolysis rate, rendering RBC transport and administration safe.

BACKGROUND Small bowel obstruction (SBO) is a common emergency surgical disease for which identifying patients need emergency surgical resection due to nonviable small bowel tissue poses a significant challenge. This study aimed to develop and validate a predictive model to guide surgical decision-making using readily available clinical data.

METHODS A retrospective cohort study was conducted using data from Wuhan Union Hospital between 2017 and 2023 to establish the prediction model, with an external validation cohort from two other medical centers. Six machine learning algorithms (Logistic Regression [LR], Random Forest [RF], K Nearest-Neighbours [KNN], Multilayer Perceptron [MLP], Adaptive Boosting [AdaBoost] and eXtreme Gradient Boosting [XGBoost]) were employed, and the final model was based on LR classifier. Performance was evaluated with various metrics including AUC-ROC, accuracy, and decision curve analysis. SHapley Additive exPlanations (SHAP) method was used for model interpretation.

RESULTS High-risk and low-risk SBO groups differed significantly in clinical, laboratory, imaging, treatment, and outcome variables. Ten predictors were retained: white blood cell count, history of abdominal operation, platelet, albumin, neutrophil, spiral sign, acute bellyache, ascites, lymphocyte count, and rebound tenderness. After comparing the six algorithms, LR was selected because it showed the most consistent generalization and calibration. The final LR model achieved an AUC of 0.889 (95% CI 0.855-0.923) in the training data, a mean cross-validation AUC of 0.876 (95% CI 0.773-0.978), and an AUC of 0.873 (95% CI 0.818-0.929) in the independent test set. In the external validation cohort from the two other medical centers, the model achieved an AUC of 0.762 (95% CI 0.693-0.831).

CONCLUSION An interpretable machine learning model was developed and validated for prediction of high-risk SBO patients upon admission. The model may offer decision support for clinicians, aiding in risk stratification and guiding treatment strategies.

BACKGROUND This study aims to investigate the prognostic value of computed tomography (CT)-quantified visceral fat area (VFA) and subcutaneous fat area (SFA) in sepsis.

METHODS This study comprised a retrospective cohort (n=347) between January 2022 and July 2024 and a prospective internal validation cohort (n=160) between August 2024 and February 2025, all enrolling sepsis patients defined by the Sepsis-3 criteria. Abdominal CT at the third lumbar vertebra (L3) was used for VFA and SFA quantification. Prognostic factors for 28-day mortality were identified through least absolute shrinkage and selection operator (LASSO) regression and multivariable analysis. A nomogram was developed and validated using discrimination, calibration, and decision curve analysis (DCA).

RESULTS The retrospective cohort demonstrated area under the curve (AUC) values of 0.74 (95% CI 0.66-0.82) (VFA) and 0.76 (95% CI 0.69-0.84) (SFA) for 28-day mortality prediction. The nomogram, incorporating VFA, SFA, lactate (LAC), total cholesterol (TC), age, septic shock, and procalcitonin (PCT), was found to outperform Sequential Organ Failure Assessment (SOFA) and Acute Physiology And Chronic Health Evaluation (APACHE) II scores with an AUC of 0.93 (95% CI 0.89-0.96) compared to 0.84 and 0.82.

CONCLUSION The CT-based nomogram integrating adiposity metrics provides a valuable prognostic tool for sepsis patients, though optimal adiposity ranges require further investigation.

BACKGROUND Aortic dissection (AD) is still a challenging diagnosis with poor prognosis, and its misdiagnosis remains an unresolved problem. Postmortem examination has long been deemed important for understanding the diseases and determining the cause of death. We aim to investigate the concordance between clinical diagnosis and autopsy findings and to identify the most common clinical “red flags” and the cognitive errors involved in the management of AD.

METHODS This retrospective, observational, single-centre study analysed a series of 30 forensic autopsies focusing on the 7 cases of AD. Clinical-autopsy discrepancies were categorised into six predefined classes, and cognitive biases were classified into four types. All records were independently reviewed by two emergency physicians and two forensic pathologists, with consultation from a cognitive psychologist and consensus reached through collegial discussion.

RESULTS These seven patients had a mean age of 59 years (range 48-71 years). Loss of consciousness (4/7 patients), lumbar pain (4/7), psychomotor agitation (4/7) and two or more ER visits were considered red flags for AD. The most common malperfusion syndromes were coronary and femoral-iliac. Among the types of diagnostic errors, faulty verification bias was the most common, followed by biases related to faulty information processing. In three patients (43%), the failure to diagnose AD directly contributed to death, as an earlier diagnosis would likely have altered clinical management and improved survival outcomes.

CONCLUSION There remains a significant unmet need to increase awareness and clinical suspicion of AD in patients presenting with atypical symptoms, non-traditional risk factors, or ambiguous physical findings. Furthermore, identifying and sharing the cognitive underpinnings of errors may aid in reducing misdiagnosis and optimizing both the timing and modality of clinical management.