Background: Accurate burn size and depth assessment at first contact guides fluid resuscitation, referral, and operative planning, yet both tasks show meaningful inter-clinician variability. General-purpose multimodal large language models may offer scalable, image-based decision support in emergency care, but prospective benchmarking against clinicians and a robust reference standard remains limited. Methods: We conducted a prospective, single-centre diagnostic accuracy and agreement study in a tertiary emergency department (22 July–8 September 2025). Consecutive acute burn presentations (< 24 h) were screened; protocol-conformant cases contributed standardized three-view photographs per anatomically distinct burn region. A multimodal large language model generated region-level estimates of total body surface area (TBSA) contribution and burn depth class. Eighteen emergency physicians independently rated the same images and minimal metadata, blinded to model and reference outputs. A three-member expert panel served as the reference standard by consensus. The primary endpoint was non-inferiority of the model versus the physician median for region-level absolute TBSA error relative to the panel, with a pre-specified margin of 3 percentage points, using patient-level cluster bootstrap for inference. Secondary endpoints included TBSA agreement and depth agreement (quadratic-weighted kappa). Results: Of 413 screened presentations, 52 patients were enrolled, yielding 64 analyzable burn region-cases (35 pediatric, 29 adult). The model’s mean absolute TBSA error versus the panel was 1.40 percentage points (median 1.00); 87.5% of cases were within ± 3 percentage points and 98.4% within ± 5. The physician median had a mean absolute error of 0.89 percentage points (median 0.75). The paired non-inferiority analysis met the pre-specified criterion (Hodges–Lehmann median Δ = 0.25; one-sided 95% upper bound = 0.50), indicating the model was non-inferior to physicians for TBSA estimation. In contrast, depth agreement versus the panel was slight for the model (quadratic-weighted kappa 0.14), with systematic underestimation of deeper burns, while physician consensus showed substantially higher agreement (quadratic-weighted kappa 0.65). Conclusions: In this prospective emergency department evaluation, a general-purpose multimodal model achieved non-inferior performance to emergency physicians for region-level TBSA estimation but performed substantially worse for burn depth classification. These findings support a narrowly defined adjunct role for TBSA estimation, while depth-dependent decisions should remain clinician-led and require further method development and external validation.

Background: Large Language Models (LLMs), such as GPT-4o, are increasingly investigated for clinical decision support in emergency medicine. However, their real-world performance in disposition prediction remains insufficiently studied. This study evaluated the diagnostic accuracy of GPT-4o in predicting ED disposition—discharge, ward admission, or ICU admission—in complex emergency respiratory cases requiring pulmonology consultation and chest CT, representing a selective high-acuity subgroup of ED patients. Methods: We conducted a retrospective observational study in a tertiary ED between November 2024 and February 2025. We retrospectively included ED patients with complex respiratory presentations who underwent pulmonology consultation and chest CT, representing a selective high-acuity subgroup rather than the general ED respiratory population. GPT-4o was prompted to predict the most appropriate ED disposition using three progressively enriched input models: Model 1 (age, sex, oxygen saturation, home oxygen therapy, and venous blood gas parameters); Model 2 (Model 1 plus laboratory data); and Model 3 (Model 2 plus chest CT findings). Model performance was assessed using accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score. Results: Among the 221 patients included, 69.2% were admitted to the ward, 9.0% to the intensive care unit (ICU), and 21.7% were discharged. For hospital admission prediction, Model 3 demonstrated the highest sensitivity (91.9%) and overall accuracy (76.5%), but the lowest specificity (20.8%). In contrast, for discharge prediction, Model 3 achieved the highest specificity (91.9%) but the lowest sensitivity (20.8%). Numerical improvements were observed across models, but none reached statistical significance (all p > 0.22). Model 1 therefore performed comparably to Models 2–3 while being less complex. Among patients who were discharged despite GPT-4o predicting admission, the 14-day ED re-presentation rates were 23.8% (5/21) for Model 1, 30.0% (9/30) for Model 2, and 28.9% (11/38) for Model 3. Conclusion: GPT-4o demonstrated high sensitivity in identifying ED patients requiring hospital admission, particularly those needing intensive care, when provided with progressively enriched clinical input. However, its low sensitivity for discharge prediction resulted in frequent overtriage, limiting its utility for autonomous decision-making. This proof-of-concept study demonstrates GPT-4o’s capacity to stratify disposition decisions in complex respiratory cases under varying levels of limited input data. However, these findings should be interpreted in light of key limitations, including the selective high-acuity cohort and the absence of vital signs, and require prospective validation before clinical implementation.

Early risk stratification may support emergency department (ED) decision-making for patients hospitalized with heart failure (HF), yet commonly used tools may perform differently across clinically relevant outcomes. We conducted a retrospective observational cohort study using MIMIC-IV (v3.1) linked to MIMIC-IV-ED (2011–2019). Adult ED encounters with HF International Classification of Diseases-9/10 codes from the ED diagnosis table that were linkable to an inpatient admission were included. The primary analytic cohort was restricted to complete-case encounters with sufficient data to compute both Acute Decompensated Heart Failure National Registry (ADHERE) CART and Get With The Guidelines–Heart Failure (GWTG-HF). The primary outcome was 30-day all-cause mortality; secondary outcomes were intensive care unit (ICU) admission within 24 hours after ED disposition, any ICU admission, acute kidney injury (AKI; Kidney Disease Improving Global Outcomes creatinine criteria), and hospital/ICU length of stay. Associations were evaluated using univariable logistic regression. Discrimination was assessed by area under the receiver operating characteristic curve (bootstrap 95% confidence intervals [CI]s) and compared using DeLong test. Calibration was assessed with patient-level 10-fold cross-validated calibration intercept/slope and calibration curves. Among 5508 eligible encounters, 4812 had complete data for both scores and constituted the analytic cohort. Thirty-day mortality occurred in 317/4812 (6.6%). ICU admission within 24 hours after ED disposition occurred in 822/4812 (17.1%), any ICU admission in 1098/4812 (22.8%), and AKI in 1373/4810 (28.5%). GWTG-HF was associated with higher 30-day mortality (odds ratio 2.3 per 10 points, 95% CI 2.1–2.6); ADHERE CART also showed higher odds across strata (group 4 vs 1 odds ratio 5.2, 95% CI 3.1–8.8). Mortality discrimination was higher for GWTG-HF than ADHERE (area under the receiver operating characteristic curve 0.75 vs 0.64; difference 0.10; P < .001), with near-ideal calibration for both. In ED patients hospitalized with HF, GWTG-HF more reliably stratified 30-day mortality risk than ADHERE CART, while both scores showed limited utility for ICU utilization and AKI, supporting outcome-specific and dynamic risk assessment.

Background: Emergency Medical Services (EMS) ambulance transport is a common route of emergency department (ED) arrival for adults aged ≥ 80 years, but factors associated with future EMS-transported ED presentation remain uncertain. We assessed whether baseline comprehensive geriatric assessment (CGA) measures were associated with EMS-transported ED presentation within 12 months. Methods: We performed a retrospective cohort study in Türkiye linking the Healthy Aging Center (YAŞAM) CGA registry, which records routine multidomain geriatric assessments in community-dwelling older adults, to ED administrative records from a tertiary hospital. Consecutive adults aged ≥ 80 years who underwent baseline CGA between October 5, 2023 and December 10, 2024 were included ( n = 587). The primary outcome was ≥ 1 EMS-transported ED presentation within 12 months; repeat use (≥ 2 presentations) was a prespecified secondary outcome. Separate age- and sex-adjusted domain-specific association logistic regression models were fitted for each CGA measure; exploratory age- and sex-adjusted logistic models were also used for repeat use. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC). Cause-specific Cox models were used as sensitivity analyses for time to first EMS-transported ED presentation, censoring at death before any EMS event. Results: During follow-up, 159/587 participants (27.1%) had ≥ 1 EMS-transported ED presentation, 55/587 (9.4%) had repeat use, and 51/587 (8.7%) died. Higher frailty was associated with the primary outcome (odds ratio [OR] per 1-point increase, 1.30; 95% confidence interval [CI], 1.16–1.47), whereas higher Katz Activities of Daily Living scores (OR, 0.77; 95% CI, 0.70–0.85), Mini Nutritional Assessment-Short Form scores (OR, 0.85; 95% CI, 0.80–0.92), and Timed Up and Go indicator = 1 (OR, 0.34; 95% CI, 0.23–0.50) were associated with lower odds. Discrimination was modest across models (AUC, 0.575–0.643). Cause-specific Cox sensitivity analyses showed directionally similar associations. Conclusions: In adults aged ≥ 80 years, multiple routinely collected CGA domains were associated with subsequent EMS-transported ED use. These measures should be interpreted as indicators of geriatric vulnerability rather than stand-alone predictive tools; any predictive application would require externally validated multivariable models.

Point-of-care ultrasound (POCUS) is increasingly used in emergency and critical care for rapid, radiation-free bedside imaging. However, its integration in

Background/Objective: Spontaneous pneumothorax (SP) commonly presents to the emergency department (ED), and clinicians must rapidly decide between conservative care and tube thoracostomy. Most artificial intelligence tools focus on pneumothorax detection rather than quantitative size estimation and actionable management recommendations. We aimed to evaluate whether a vision-enabled generative pre-trained transformer (GPT) model could estimate apical SP depth on chest radiographs (CXRs) and predict initial ED management. Methods: We conducted a single-center retrospective observational study of adult patients (≥ 18 years) with confirmed SP on pre-intervention posteroanterior CXR (January 1, 2023–December 31, 2025). GPT received only the CXR plus age and sex and returned laterality, estimated apical depth (Depth_cm), and a binary management recommendation (tube thoracostomy vs conservative). The reference outcome was tube thoracostomy within 24 h. We calculated diagnostic performance metrics and area under the receiver operating characteristic curve (AUC). We assessed agreement between GPT depth estimates and blinded radiologist measurements using intraclass correlation coefficient (ICC), Bland–Altman analysis, and mean absolute error (MAE). Results: Among 101 patients, mean age was 33.1 ± 14.6 years and 89 (88.1%) were male; 53 (52.5%) underwent tube thoracostomy within 24 h. GPT showed sensitivity 86.8% (95% CI, 75.2%–93.5%), specificity 93.8% (95% CI, 83.2%–97.9%), and accuracy 90.1% (95% CI, 82.7%–94.5%), with Cohen’s κ = 0.80 and AUC 0.90 (95% CI, 0.84–0.96). Depth agreement was strong (ICC, 0.893), with MAE 0.69 cm and mean bias −0.51 cm (95% limits of agreement, − 2.30 to 1.27 cm). Conclusions: In confirmed SP, a vision-enabled GPT model produced apical depth estimates that closely agreed with radiologist measurements and generated management recommendations that substantially matched real-world ED decisions, supporting its potential role as adjunct imaging decision support.