Journal of Practical Emergency Medicine

Large language models (LLMs) are moving into clinical medicine faster than the evidence needed to define their safest role. Emergency medicine makes that gap particularly difficult to ignore. In the emergency department (ED), a convincing answer is not necessarily a safe answer, and a seemingly small error in prioritization may delay care for a patient with a time-sensitive condition. The relevant question, therefore, is no longer whether an LLM can produce a plausible diagnosis or assign an acuity category. It is whether the use of such a model can improve emergency care without introducing an unacceptable risk of under-triage, misplaced confidence, or loss of clinically important context.

Triage is an especially demanding test case. Decisions are made early, often with incomplete information, before the diagnostic picture has matured. A patient may be unable to provide a coherent history; laboratory and imaging results may not yet exist; and clinical deterioration can occur after the initial assessment. Moreover, triage errors are asymmetric. Over-triage consumes resources, but under-triage may postpone recognition of stroke, sepsis, acute coronary syndrome, internal bleeding, or another immediately threatening condition. Any technology positioned at this point in the care pathway should therefore be judged not only by overall accuracy but also by the clinical consequences of its errors.

Early evidence has nevertheless been impressive enough to justify serious attention. Williams et al. evaluated an LLM using data from 251,401 adult ED visits. In a balanced sample of 10,000 pairs of patients with different Emergency Severity Index (ESI) levels, the model correctly identified the patient with higher acuity in 8,940 pairs, corresponding to an accuracy of 89%. In a 500-pair subset reviewed by physicians, LLM accuracy was 88% compared with 86% for physicians (1).

These results are encouraging, but the task should not be mistaken for prospective triage. The model was asked to compare two written clinical histories and determine which represented the higher-acuity patient. It was not independently assessing patients arriving at a triage desk, nor was it operating prospectively within an ED workflow. The study therefore demonstrated that an LLM could extract useful acuity information from clinical text; it did not establish that autonomous LLM-based triage is safe.

Other evidence makes that distinction even more important. Zaboli et al. retrosp­ectively compared ChatGPT-4.0 with human triage decisions in 2,658 ED patients. Agreement between human and AI triage was low, with a Cohen κ of 0.125 (95% CI, 0.100–0.134). Human triage also performed better in predicting clinically important outcomes. For 30-day mortality, the area under the receiver operating characteristic curve was 0.88 for human triage and 0.70 for AI triage (P<0.001). For the need for life-saving interventions, the corresponding values were 0.98 and 0.87 (P=0.014). The authors specifically identified lower sensitivity in high-risk patients and consequent under-triage as a concern (2).

Yet another prospective study produced a different picture. Arslan et al. compared ChatGPT Plus, Copilot Pro, and triage nurses against ESI assessments assigned by an emergency physician. Overall accuracy was 66.5% for ChatGPT, 61.8% for Copilot, and 65.2% for nurses. Notably, identification of high-acuity patients was more frequent with ChatGPT and Copilot than with nurses in that study 87.8% and 85.7% versus 32.7%, respectively (3).

The apparent contradiction among these studies is itself informative. LLM performance cannot be treated as a fixed property of “AI.” Results depend on the model and version tested, the information supplied to it, the prompt, the reference standard, the triage system, the patient population, and the exact task being evaluated. A model that performs well when ranking two written cases may not perform equally well when assigning an absolute triage category. Likewise, good average accuracy can conceal clinically unacceptable errors in a small but high-risk subgroup.

How information is supplied to the model may be as important as the model itself. Yazaki et al. examined a retrieval-augmented generation (RAG) approach using 100 simulated triage scenarios derived from modified cases from the Japanese National Examination for Emergency Medical Technicians. GPT-3.5 combined with RAG achieved 70% correct triage, compared with 35% and 38% for two emergency medical technicians and 50% and 47% for two emergency physicians. Under-triage occurred in 8% of cases with GPT-3.5 plus RAG, compared with 33% for GPT-3.5 without RAG and 39% for GPT-4 without RAG (4). The findings suggest that connecting an LLM to task-specific knowledge may improve performance. At the same time, the study used simulated scenarios rather than live emergency encounters, which limits how far these results can be extended to routine practice.

Gaber et al. similarly evaluated several LLM configurations and a RAG-assisted workflow using 2,000 cases derived from the MIMIC-IV ED database. Their framework addressed triage, specialty referral, and diagnosis rather than treating the model as a stand-alone question-answering system (5). This is an important shift in emphasis. The relevant unit of evaluation may ultimately be the LLM-enabled clinical workflow, not the language model in isolation.

There are also reasons to remain cautious about extrapolating high performance on constrained tasks to autonomous clinical decision-making. Hager et al. evaluated selected LLMs using a curated dataset of 2,400 real patient cases involving four common abdominal pathologies. The tested models performed worse than physicians across the evaluated pathologies and showed limitations in following diagnostic and treatment guidelines, interpreting laboratory results, and responding consistently to changes in instructions (6). These findings should not be generalized automatically to every newer model—the technology evolves rapidly—but they demonstrate why conventional medical benchmarks are an incomplete proxy for bedside reliability.

The same caution applies to the concept of the AI “co-pilot.” Human supervision alone does not guarantee that an LLM will improve decisions. In a randomized clinical trial involving 50 physicians from internal medicine, family medicine, and emergency medicine, Goh et al. compared physicians using GPT-4 with physicians using conventional resources. Median diagnostic reasoning scores were 76% and 74%, respectively, an adjusted difference of 2 percentage points (95% CI, −4 to 8; P=0.60). Interestingly, the LLM operating alone scored 16 percentage points higher than the conventional-resources physician group (95% CI, 2–30; P=0.03) (7). Only five participants were emergency physicians, so the results are not an ED-specific effectiveness trial. Nevertheless, they expose an important problem: access to a capable model does not automatically translate into better human performance. Interface design, clinician training, appropriate reliance, and the way disagreement between clinician and model is handled may determine whether assistance is useful.

For this reason, the first successful applications of LLMs in emergency care may emerge from narrower tasks in which benefit can be measured without transferring final clinical authority to the model. Song et al. provide an instructive example. In a study involving six emergency physicians and 50 representative ED cases, LLM-assisted discharge documentation reduced median writing time from 69.5 seconds (95% CI, 65.5–78.0) to 32.0 seconds (95% CI, 29.5–36.0; P<0.001). LLM-assisted notes also received higher ratings than manually written notes for completeness, correctness, conciseness, and clinical utility (all P<0.001) (8). The study was performed in a single tertiary center and involved a limited validation set, but it illustrates a practical principle: the safest path to clinical adoption may begin with bounded tasks that reduce workload while leaving consequential decisions under direct clinician control.

Emergency triage requires an equally pragmatic standard. Before an LLM is used as more than supervised decision support, prospective multicenter studies should evaluate outcomes that matter to patients rather than relying predominantly on agreement with historical labels. Under-triage deserves particular attention, as do delays to time-sensitive treatment, unexpected intensive care unit admission, early ED return, and adverse clinical outcomes. Performance should also be examined across patient groups and clinical presentations rather than reported only as a single aggregate accuracy measure.

Model identity and version should be treated as part of the intervention. An evaluation performed with one version cannot be assumed to remain valid after a major model update. Similarly, changes in prompts, retrieval sources, or local workflow may alter performance. Clinical deployment will therefore require version control, post-implementation surveillance, and an auditable record of when AI-generated recomme­ndations are accepted or overridden.

Accountability must be designed before deployment rather than after an adverse event. If an LLM assigns a lower acuity level than the triage clinician, should its recommendation be ignored, reviewed, or trigger a mandatory second assessment? If the model recommends escalation and the clinician disagrees, what should be documented? These are not peripheral legal questions. They determine how the technology will influence actual behavior in the ED.

The current evidence does not support either extreme position. It would be premature to dismiss LLMs as unreliable text generators when they have demonstrated substantial ability to interpret clinical narratives and, in selected tasks, match or exceed human performance. It would be equally premature to treat those findings as evidence that autonomous AI triage is ready for routine deployment. What has been established is capability under specific experimental conditions. What remains to be established is dependable clinical benefit.

For emergency medicine, that distinction matters. The front door of the hospital is an unforgiving place to discover that benchmark performance does not translate into patient safety. For now, the most defensible role for LLMs in emergency triage is not that of an autonomous gatekeeper, but of a supervised clinical co-pilot whose recomme­ndations can be questioned, overridden, and continuously evaluated. The threshold for moving beyond that role should be set by prospective evidence of safer or better care not by the fluency of the model's answers.

Background: Acute abdomen is a common and clinically important emergency department presentation that requires timely evaluation and appropriate imaging selection. Ultrasonography (US) and computed tomography (CT) are widely used in this setting, but their real-world use may vary according to disease type, patient characteristics, admission time, and local resource availability. This study aimed to describe and compare the utilization frequency and acute-pathology detection rates of US and CT among emergency department patients hospitalized with acute abdomen.

Methods: This retrospective cross-sectional study included adult patients hospitalized with a final diagnosis of acute abdomen in the emergency department of a secondary-care hospital between January 1 and December 31, 2024. Demographic characteristics, admission profile, final diagnoses, imaging use, and imaging report results were extracted from the hospital electronic data system. Acute-pathology detection rate was defined as the proportion of imaged patients with an acute pathological finding reported on the corresponding modality. Paired US-CT findings were evaluated among patients who underwent both modalities. Because an independent reference standard was not available, the findings were interpreted as utilization and detection patterns rather than formal diagnostic accuracy estimates.

Results: Of 862 screened patients, 741 were included in the final analysis. The mean age was 50.43 +/- 19.85 years, and 413 patients (55.7%) were male. Acute appendicitis was the most common diagnosis (301 patients, 40.6%), followed by ileus (169 patients, 22.8%) and acute cholecystitis (124 patients, 16.7%). CT was performed in 689 patients (93.0%; 95% CI 90.9-94.6), whereas US was performed in 122 patients (16.5%; 95% CI 14.0-19.3). Both modalities were performed in 104 patients (14.0%; 95% CI 11.7-16.7). Acute pathological findings were reported in 69 of 122 US examinations (56.6%; 95% CI 47.7-65.0) and 629 of 689 CT examinations (91.3%; 95% CI 89.0-93.2). In paired cases, both modalities were acute-positive in 45 patients, US was acute-negative/CT acute-positive in 43 patients, US was acute-positive/CT acute-negative in 8 patients, and both were acute-negative in 8 patients.

Conclusion: In this hospitalized acute abdomen cohort, CT was used substantially more often than US and had a higher acute-pathology detection rate. These findings reflect real-world imaging practice in a secondary-care emergency setting and should not be interpreted as definitive diagnostic superiority because imaging selection was not standardized and no independent reference standard was available.

Background: Paranasal sinus computed tomography (CT) is often examined in urgent sinonasal conditions, especially when complicated infection or possible intervention is being considered. The aim is to highlight how these factors are important in emergency radiology, especially in urgent sinonasal situations.

Methods: A retrospective cross-sectional study was performed at Sulaimani Teaching Hospital between November 2023 and December 2024 on rhinosinusitis patients. The depth of the olfactory fossa on both sides was recorded and classified according to the Keros classification using patients' files.

Results: The study included 501 patients (57.1% male; mean age 38.37 ± 13.82 years). Most of the patients were classified as Keros type II, found in 360 of the right fossae (71.9%) and 361 of the left fossae (72.1%). In addition, 297 patients (59.3%) exhibited symmetrical Keros type II configurations, while 128 patients (25.5%) showed olfactory fossa asymmetry. The most common asymmetry seen was subtype IVc (right type II, left type I). We observed a moderate level of right–left agreement (Cohen’s kappa = 0.421). Overall, 156 patients (31.1%) had skull base anatomy that was clinically significant, highlighting the importance of surgical awareness.

Conclusion: Our findings indicate that Keros type II is the most prevalent configuration, and nearly a quarter of the patients presented with olfactory fossa asymmetry. Additionally, about one-third of the patients had clinically significant skull base anatomy. These results emphasize the necessity of a thorough evaluation to improve communication and operative planning in urgent sinonasal assessments.

Background: Nutritional risk is relevant to early inpatient and acute-care pathways because early recognition of malnutrition may support timely referral. This study estimated the prevalence of malnutrition risk among adult inpatients in Sulaimani using the Malnutrition Universal Screening Tool (MUST) and Subjective Global Assessment (SGA), and examined its association with age, sex, length of hospital stay, and primary cause of admission.

Methods: This observational cross-sectional study was conducted in four public hospitals in Sulaimani from March 1 to March 31, 2024. Individual-level data from 169 adult inpatients aged 18 years or older who had remained hospitalized for more than three days were analyzed. Nutritional status was assessed between days 3 and 7 of hospitalization using MUST and SGA. Therefore, the findings reflect patients who remained hospitalized beyond three days rather than nutritional status at the exact time of emergency department arrival or hospital admission.

Results: The mean age was 53.6 ± 18.08 years, and 87 participants (51.5%) were female. MUST identified 131 patients (77.5%; 95% CI: 70.6-83.2) as being at medium or high risk of malnutrition, while SGA classified 140 patients (82.8%; 95% CI: 76.4-87.8) as moderately or severely malnourished. Hospital stay of 8 days or longer was associated with SGA-defined moderate/severe malnutrition after adjustment for available covariates (adjusted OR: 12.30; 95% CI: 1.60-94.78; P = 0.016), although the temporal direction of this association cannot be determined. Severe malnutrition was present in 69 patients (40.8%; 95% CI: 33.7-48.4) and was associated with age >65 years, hospital stay ≥8 days, and oncological admission after adjustment for available covariates. MUST and SGA showed high observed agreement after dichotomization (92.3%), with Cohen's kappa = 0.759.

Conclusion: Malnutrition risk was very common among adult inpatients who remained hospitalized for more than three days in Sulaimani. The findings may be relevant to early acute-care screening, particularly for patients who remain hospitalized after urgent admission pathways. However, this study was not designed to estimate malnutrition at emergency department triage, and causal conclusions about malnutrition and length of stay should be avoided.

Background: Abnormal uterine bleeding (AUB) is a frequent gynecologic problem that may present to acute-care settings when bleeding is heavy, persistent, or associated with symptoms suggestive of anemia. Initial emergency evaluation prioritizes exclusion of pregnancy-related bleeding and assessment for hypovolemia or hemodynamic instability; however, background gynecologic and metabolic characteristics may also help clinicians recognize patients likely to have clinically significant AUB. This study examined factors associated with AUB among reproductive-aged women attending a tertiary gynecology outpatient clinic and interpreted the findings for acute-care practice.

Methods: This retrospective case-control study included 700 women aged 15-55 years: 350 with AUB and 350 controls without AUB. Demographic, reproductive, metabolic, and gynecologic variables were extracted from medical records. The primary outcome was AUB status. Bleeding patterns were summarized among cases. Group comparisons, crude odds ratios (ORs), and a clinically selected multivariable logistic regression model were used. Findings were interpreted as associations, not prevalence, incidence, or causal effects.

Results: The mean age was 41.7 ± 9.7 years among cases and 40.7± 9.4 years among controls. Women with AUB had higher BMI than controls (27.5± 4.0 vs 25.8± 3.6 kg/m2; p<0.001), and obesity was more common among cases (21.7% vs 10.9%). Menorrhagia was the most common recorded bleeding pattern (53.1%), followed by intermenstrual bleeding (18.9%). After adjustment, uterine fibroid had the strongest association with AUB (adjusted OR 62.90; 95% CI 22.37-176.89), followed by PCOS (adjusted OR 23.02; 95% CI 8.18-64.76), history of anemia (adjusted OR 4.17; 95% CI 2.37-7.32), hypertension (adjusted OR 3.46; 95% CI 2.00-6.01), number of pregnancies (adjusted OR 1.19 per pregnancy; 95% CI 1.05-1.36), BMI (adjusted OR 1.13 per kg/m2; 95% CI 1.08-1.20), and family history of AUB (adjusted OR 1.66; 95% CI 1.01-2.71). IUCD use was associated with AUB in crude analysis but not after adjustment.

Conclusion: In this clinic-based case-control study, fibroid and PCOS were the dominant gynecologic factors associated with AUB. BMI, anemia history, hypertension, number of pregnancies, and family history also remained associated after adjustment. These findings are relevant to acute-care clinicians as background risk-recognition information, but the study was not emergency-department-based and did not include pregnancy test results, vital signs, hemoglobin values, transfusion, admission, treatment, or short-term outcomes. Prospective ED-based studies are needed to develop severity and disposition models for women presenting with AUB.

Background: Methanol poisonings (MP) are responsible for a considerable number of fatal intoxications on an annual basis. It has been demonstrated that even relatively small ingestions of these alcohols can produce significant toxicity. It is imperative to recognize the signs of methanol poisoning promptly in order to initiate treatment in a timely manner. The objective of this study was to evaluate the clinical and laboratory data of patients with the aim of predicting the prognosis of MP cases brought to the Emergency Department (ED).

Methods: This single-center, retrospective, observational study was conducted with MP cases that were brought to the ED of Health Sciences University Antalya Training and Research Hospital between January 1, 2018 and December 31, 2022. The demographic, clinical, laboratory, and outcome data were systematically recorded and analyzed, and survivors were compared with non-survivors.

Results: The study's sample size was 40 patients. The mean age of the patients was 48.62 years (±12.8 years), 17.5% of whom were female and 82.5% of whom were male. The most prevalent symptoms were altered consciousness, observed in 80% of cases, and visual symptoms, noted in 50% of cases. The mean Glasgow Coma Scale (GCS) value was 8.55 ± 5.33, and 25% of the patients were comatose. Renal replacement therapy (RRT) was administered to 90% of the patients, ethanol therapy to 82.5%, and mechanical ventilation to 45% of the patients. Non-surviving patients exhibited a larger anion gap, higher partial pressure of carbon dioxide (PCO2) levels, and higher lactate levels, while non-surviving patients demonstrated lower pH levels and lower bicarbonate levels. The area under the curve (AUC) analysis revealed that pH (AUC= 0.911 [95% confidence interval (CI): 0.820, 0.970]) and base excess (AUC=0.906 [95% CI: 0.819, 0.970]) were the blood gas parameters with the highest AUC values in relation to in-hospital mortality.

Conclusion: The in-hospital mortality rate was 47.5%. In MP cases, coma on arrival at the emergency department, more severe metabolic abnormalities, and the need for mechanical ventilation were associated with in-hospital mortality.

Background: Carpal tunnel syndrome (CTS) is a common median nerve entrapment neuropathy that may present in emergency or urgent-care settings with hand pain, numbness, paresthesia, or weakness. Because clinicians must distinguish chronic CTS from acute median nerve compression and other neurologic or vascular mimics, this study evaluated serum vitamin D and vitamin B12 levels across electrodiagnostic CTS severity grades.

Methods: This cross-sectional study included 100 adults with clinically suspected and electrodiagnostically confirmed CTS. Participants were classified by Bland electrodiagnostic severity as mild, moderate, or severe CTS. Fasting serum 25-hydroxyvitamin D and vitamin B12 levels were measured on admission (the same day as electrodiagnostic testing). Continuous variables were compared across severity groups using one-way analysis of variance, categorical variables using chi-square testing, and correlations using Pearson correlation coefficients.

Results: Among 100 patients, 32 had mild CTS, 41 moderate CTS, and 27 severe CTS. Mean serum vitamin D levels declined progressively across mild, moderate, and severe groups (28.6 ± 7.2, 19.4 ± 6.8, and 12.1 ± 5.3 ng/mL, respectively; p<0.001). Vitamin D deficiency increased from 12.5% in mild CTS to 51.2% in moderate CTS and 81.5% in severe CTS (p<0.001). Mean serum vitamin B12 levels also declined with increasing severity (342.5 ± 88.4, 241.3 ± 76.2, and 163.7 ± 54.9 pg/mL, respectively; p<0.001), while B12 deficiency increased from 9.4% to 34.1% and 70.4% across the same groups (p<0.001). Vitamin D and vitamin B12 showed inverse correlations with the nerve-conduction-derived severity score (r= -0.71 and r= -0.67, respectively; both p<0.001).

Conclusions: Lower serum vitamin D and vitamin B12 levels were significantly associated with greater electrodiagnostic severity of CTS in this cohort. For stable subacute or chronic CTS discharged from emergency or urgent care, identifying vitamin D or B12 deficiency may be reasonable as part of outpatient follow-up, especially in moderate-to-severe, bilateral, recurrent, or metabolically complex cases.

Early-career researchers (ECRs) occupy a heterogeneous set of career stages, from doctoral training and postdoctoral appointments to the transition toward independent research leadership. Across countries and disciplines, the evidence repeatedly identifies challenges related to insecure career structures, access to research funding, mentorship, professional networks, research and communication skills, publishing and research-evaluation pressures, and work-life and mental-health concerns. This narrative review synthesizes peer-reviewed evidence and current reports from research-funding and scientific organizations to examine these barriers and the support strategies most consistently proposed or evaluated. The literature indicates that no single definition of an ECR is universally applicable and that experiences differ substantially by career stage, discipline, national research system, gender, and family circumstances. Mentoring is generally associated with career development, but the evidence favors structured and network-based approaches over reliance on a single informal mentor. Capacity-building programs can address gaps in scientific writing, communication, grant development, and broader academic competencies, particularly where institutional training is limited. Recent evidence also documents a substantial burden of psychological distress among doctoral and early postdoctoral researchers, while qualitative studies link poor well-being to job insecurity, high expectations, and limited institutional support. Policies for ECRs should therefore combine career-stage-appropriate funding, transparent career development, high-quality mentoring, skills training, responsible research assessment, publishing support, and measures that address work-family and equity barriers. The evidence base remains heterogeneous, and many interventions require stronger longitudinal evaluation.

Hantavirus disease is an uncommon zoonotic infection that can progress rapidly from a nonspecific febrile illness to life-threatening cardiopulmonary or renal dysfunction. In the Americas, hantavirus cardiopulmonary syndrome (HCPS), also termed hantavirus pulmonary syndrome (HPS), is characterized by increased vascular permeability, pulmonary edema, hypoxemia, and circulatory failure, whereas hemorrhagic fever with renal syndrome (HFRS), encountered predominantly in Europe and Asia, is distinguished by acute kidney injury, thrombocytopenia, and variable hemorrhagic manifestations (1-3). The emergency department is an important point of recognition because patients may initially present with fever, myalgia, gastrointestinal symptoms, or headache before respiratory or hemodynamic deterioration becomes apparent. Emergency nurses participate in exposure assessment, serial physiologic observation, implementation of supportive-care protocols, fluid-balance surveillance, recognition of respiratory and circulatory deterioration, and infection-prevention measures. However, nursing-specific outcome evidence identified by the targeted search used for this mini-review is limited; the nursing role should therefore be derived cautiously from established clinical-management principles rather than presented as independently proven to improve survival. This mini-review examines evidence relevant to emergency nursing practice, including 2026 publications on critical-care management, prognosis, outbreak response, and healthcare-worker exposure. It emphasizes early recognition of HCPS, monitoring during the transition to cardiopulmonary failure, careful fluid and hemodynamic management, escalation to critical care, renal surveillance in HFRS, and infection prevention in suspected Andes virus infection. The multinational Andes virus outbreak linked to expedition cruise travel in 2026 was declared contained by WHO on 2 July 2026; nevertheless, the event illustrates the need for emergency departments to maintain operational readiness for an uncommon but potentially severe infection in which limited person-to-person transmission is operationally relevant (7,21).

Hyponatremia is a frequent electrolyte disorder in emergency and inpatient care, but its clinical meaning varies widely according to acuity, severity, symptoms, and underlying mechanism. Beer potomania is an important low-solute form of hypotonic hyponatremia that develops when heavy beer intake is combined with markedly reduced dietary protein and salt intake. The condition is clinically distinctive because the kidney is not primarily failing to dilute urine; rather, it lacks enough urinary osmoles to eliminate the patient's free-water load safely. As a result, even when antidiuretic hormone (ADH) is appropriately suppressed, maximal water excretion may be capped at a relatively low volume. Emergency clinicians must recognize this mechanism because treatment can be paradoxically hazardous: solute reintroduction through food, isotonic saline, hypertonic saline, potassium chloride, or saline-containing vitamin infusions may abruptly restore water excretion and produce rapid autocorrection of serum sodium. This narrative review synthesizes guideline recommendations, mechanistic literature, expert discussions, recent systematic reviews of case reports, and representative case reports on beer potomania-associated hyponatremia. It emphasizes a practical bedside approach: confirm hypotonicity, assess neurologic severity, evaluate volume status and solute intake, interpret urine studies in clinical context, avoid reflexive isotonic fluid administration in stable patients, and monitor closely for brisk aquaresis and overcorrection. Severe symptomatic hyponatremia should be treated promptly with hypertonic saline boluses to achieve an initial limited rise, whereas stable patients usually require careful fluid restriction, gradual nutritional repletion, and frequent reassessment. Prevention of osmotic demyelination syndrome depends on controlled correction, recognition of high-risk features such as alcohol use disorder and malnutrition, accounting for the sodium-raising effect of potassium replacement, and individualized use of 5% dextrose in water (D5W), desmopressin, or both when sodium rises faster than intended.

To the Editor,

Accurate, reproducible assessment of consciousness is fundamental to decision-making in acute and critical care settings, guiding prioritization of imaging, airway protection and intubation strategies, escalation pathways, and reliability and interpretability of serial handoffs. Despite its ubiquity, contemporary evidence suggests that the Glasgow Coma Scale (GCS), particularly when reported as a single total score, can obscure clinically meaningful between-patient variability and compress distinct neurologic states into the same overall score (1). A particular challenge in applying the three GCS components (eye, verbal, and motor) in critical care is that verbal response is not reliably measurable in patients with airway compromise or intubation, facial trauma, aphasia, intoxication, or sedative exposure (2). This represents a major limitation and supports the need for alternative assessment tools that remain fully scorable under such conditions.

In analyses from TRACK-TBI, patients with similar total GCS scores exhibited discrepancies in consciousness levels, underscoring that an overall GCS score alone may not reliably represent the underlying neurologic states. Notably, a GCS score of 8 was associated with minimally conscious state without language function (MSC-) in 78% of patients, MSC with language function (MSC+) in 17% and vegetative state in 5% (1). Moreover, any sum score can arise from different combinations of GCS components, which may not carry equivalent prognostic value. Among TBI patients with a total GCS of 8, the reported discharge mortality rate ranges from 21% for E1V2M5 (eye 1, verbal 2, motor 5) to 43% for E4V1M3 (3). The practical consequence is that identical total scores can map to varying neurologic profiles, rendering threshold-based pathways inherently ambiguous. The GCS was originally developed as a bedside clinical communication tool, and the entrenched use of an “initial GCS ≤8” threshold, widely applied in TBI patients, to dichotomize “severe” from mild to moderate injury, has been challenged as an outdated heuristic that has migrated from research convenience into clinical care and triage and rigid reliance on this threshold may delay care for patients requiring urgent care (4). More broadly, expert commentary at the 50-year milestone of the GCS has emphasized that modern TBI assessment is shifting toward multidimensional assessment frameworks that incorporate modifiers beyond a single consciousness score, reflecting the need for more comprehensive evaluation (5). In non-TBI contexts, a multicenter trial on comatose patients with acute poisoning and a GCS <9 reported clinical benefit for composite in-hospital endpoints with a conservative strategy of withholding intubation, illustrating that a low GCS alone is an insufficient basis for airway decisions (6). The key question is not whether to completely discard the GCS, but how to preserve its familiar structure while mitigating recognized limitations, particularly in patients for whom more precise neurologic discrimination is most consequential.

The Full Outline of UnResponsiveness (FOUR) score offers a practical alternative approach by retaining eye and motor assessment while replacing the verbal domain with brainstem reflexes and respiratory pattern, enabling structured scoring in intubated patients and improving distinction among deeply impaired states. Brainstem reflexes and respiratory drive could reflect the severity of coma and evolving deterioration, representing clinically valuable features that the GCS does not explicitly assess (7, 8). Across a range of medical, neurological and neurosurgical intensive care unit (ICU) cohorts, encompassing patients with traumatic and non-traumatic brain injury, stroke, sepsis and cardiac arrest, the FOUR score has generally achieved higher predictive accuracy for hospital and short-term mortality than GCS and has shown greater responsiveness when patients are clustered at very low GCS scores (9). In TBI, our systematic review and meta-analysis comparing the FOUR score and GCS for the prediction of in-hospital mortality yielded broadly comparable results (10), supporting clinical interchangeability in contexts where an untestable verbal response constrains GCS. Studies on non-traumatic altered mental status, one of the most challenging emergency department (ED) presentations, although limited, further support the feasibility and clinical comparability of the two scores. In a prospective ED study of adults with non-traumatic altered mental status, the FOUR score demonstrated comparable prognostic performance to GCS for predicting mortality and poor functional outcome. The study further proposed FOUR score strata aligned with familiar GCS severity bands (11).

There may be concerns that the FOUR score is less reproducible and more difficult to teach. Nevertheless, a quantitative systematic review spanning ICU and ED professionals has reported high reliability and validity for both tools, with the analyses slightly favoring the FOUR score (12). Considering that neurologic assessment is inherently serial and team-based, even modest gains in interrater reliability can reduce inconsistencies and improve the quality of decisions based on change over time. Additionally, it has been reported that emergency medicine residents find the FOUR score easy to learn and interpret (11), which supports the notion that the FOUR score can be implemented without an undue training burden.

The clinical implication is subtle but important. When two tools perform similarly for key outcomes, the one that preserves more useful bedside information is often the more rational choice for routine documentation. What should change in practice, therefore, is a refinement of how neurologic status is captured and communicated. First, the FOUR score should be routinely added, or preferentially used, when the verbal component is unavailable or unreliable, and when brainstem reflexes and respiratory pattern are expected to inform decision-making meaningfully (9, 12). Second, when GCS is used, given the demonstrated variability within patients with similar total scores, documenting each component should be prioritized over a sum score alone (1). Third, care and decision pathways should abandon rigid dichotomies, such as “GCS ≤8,” in favor of a structured, multidomain assessment and trajectory-based decision-making (4, 5).

In conclusion, we propose that GCS can be replaced with the FOUR score to reduce information loss and enhance specificity, particularly in intubated and deeply obtunded patients. Future multicenter implementation studies should standardize scoring strata and examination time points, and evaluate whether the FOUR score-guided documentation improves communication, earlier detection of deterioration, and escalation accuracy in acute care practice.

Dear Editor;

The recently contained 2026 multi-country cluster of Andes virus (ANDV) infection, aboard the M/V Hondius cruise ship, showed that hantavirus disease can present at the emergency department door far beyond the rural settings clinicians expect, exposing a persistent diagnostic blind spot. As of 2 July 2026, the World Health Organization (WHO) reported 13 linked cases, including three deaths (case fatality ratio, 23%) (1). Although uncommon, hantavirus carries disprop­ortionate emergency-medicine relevance: a nonspecific febrile illness can evolve rapidly into hemorrhagic fever with renal syndrome (HFRS) or hantavirus cardiop­ulmonary syndrome (HCPS)/hantavirus pulmonary syndrome (HPS), with acute kidney injury, shock, pulmonary edema, and respiratory failure (1-3).

Hantaviruses are zoonotic RNA viruses maintained in rodent reservoirs. Infection usually follows inhalation of aerosolized particles contaminated with rodent excreta, typically during farming, forestry, camping, or cleaning poorly ventilated rodent-infested spaces (3-5). Human-to-human transmission is not a general feature; the exception is ANDV, documented in Argentina and Chile, reaffirmed during the 2026 outbreak, which showed onboard spread (1, 2, 6, 7).

Geographic patterns are heterogeneous. HFRS predominates in Asia and Europe: Hantaan and Seoul viruses are major Asian causes, while Puumala and Dobrava-Belgrade viruses account for most European cases (3, 5). China recorded 166,975 HFRS cases and 1,689 deaths during 2004-2016 (8). The European Union/European Economic Area (EU/EEA) reported 1,885 cases in 2023 (0.4/100,000), mostly from Finland and Germany (9). These figures are not directly comparable given differing surveillance (5, 9).

HCPS occurs predominantly in the Americas, caused by New World hantaviruses including Sin Nombre virus (North America) and ANDV (southern South America); it is less frequent than HFRS but more fatal (2, 3). In 2025, eight American countries reported 229 confirmed HPS cases and 59 deaths, concentrated in the Southern Cone (10). In the United States, 890 laboratory-confirmed hantavirus cases, 859 meeting the HPS definition, were reported from 1993-2023; local epidemiology is more informative than continental estimates (11, 12).

Incidence reflects reservoir ecology and human behavior, though effects vary by virus and region (3, 13). A key diagnostic blind spot lies in reducing exposure assessment to "Have you seen a rodent?" More useful questions address cleaning dusty spaces, agricultural work, sleeping in cabins, camping, travel to endemic areas, and contact with a person with suspected ANDV infection (4, 5).

Early recognition is difficult because the prodrome, fever, headache, myalgia, gastroi­ntestinal symptoms, resembles influenza, coronavirus disease 2019 (COVID-19), sepsis, leptospirosis, or dengue (2, 4, 5). This nonspecificity is the central diagnostic blind spot; subsequent organ involvement is more informative. HFRS features increased vascular permeability, hypotension, thrombo­cytopenia, proteinuria/hematuria, occasionally disseminated intravascular coagulation (DIC), and acute kidney injury (5, 14, 15). HCPS may progress within 5-10 days to bilateral pulmonary edema, hypoxemic respiratory failure, and shock; thrombo­cytopenia, hemocon­centration, leukocytosis with left shift, and circulating immunoblasts are prodromal clues, though none replaces epidemiologic assessment and confirmatory testing (2, 4, 16).

The practical task is not universal testing but risk-stratified early recognition. Thrombo­cytopenia with worsening respiratory failure after plausible rodent exposure warrants greater suspicion than uncomplicated viral illness. Fever with thrombo­cytopenia, proteinuria/hematuria, oliguria, rising creatinine, or unexplained hypotension should raise concern for HFRS when exposure history is compatible (4, 5, 14). Because pulmonary deterioration in HCPS may be abrupt, patients with evolving respiratory or hemodynamic abnormalities warrant close monitoring and early transfer if needed (2, 4, 17).

Confirmation relies on detection of immunoglobulin M (IgM) and immunoglobulin G (IgG) antibodies, typically by enzyme-linked immunosorbent assay (ELISA), with reverse transcription-quantitative polymerase chain reaction (RT-qPCR) offering earlier detection and species differe­ntiation (2). Because turnaround time varies, escalation of care often must precede confirmation, a delay clinicians cannot afford. Early consultation with infectious-disease, critical-care, nephrology, and public-health teams should follow the dominant syndrome (2, 4, 5). When ANDV is suspected, infection-prevention measures and contact tracing require prompt public-health involvement (1, 6, 7).

Hantavirus should not become a reflex diagnosis for every febrile patient, but it deserves a defined place in the differential when exposure, geography, thrombo­cytopenia, renal dysfunction, vascular leakage, pulmonary edema, or rapid deterioration form a coherent picture. Closing this blind spot means recognizing hantavirus not by frequency, but by common illnesses, how quickly it progresses, and how little time remains to act.

Background: Waterhouse-Friderichsen syndrome is a rare, rapidly fatal condition characterized by bilateral adrenal hemorrhage and acute adrenal insufficiency, most commonly associated with fulminant meningococcemia. Early manifestations may be nonspecific, and the diagnosis is often delayed or established postmortem.

Case presentation: A 65-year-old woman with a history of migraine was brought to the emergency department after awakening with aphasia and right-sided weakness. Initial neurological assessment showed severe focal deficits, with a National Institutes of Health Stroke Scale (NIHSS) score of 22, and an acute stroke pathway was activated. Her admission temperature was 36.6 °C, and no cutaneous lesions were initially documented. Brain magnetic resonance imaging showed no acute ischemic lesion, and computed tomography angiography of the head and neck showed no extracranial or intracranial vascular occlusion. During the diagnostic workup, she developed profound hemodynamic deterioration. Initial laboratory tests showed leukopenia, mild thrombo­cytopenia, acute kidney injury, markedly elevated D-dimer, and hyperlactatemia. She rapidly progressed to refractory shock, metabolic acidosis, fever, mottling, and multiorgan failure despite fluid resuscitation, vasopressors, empiric antibiotic therapy, hydrocortisone, airway management, and rescue therapy. Lumbar puncture was performed because of persistent neurological symptoms and negative neuroimaging for acute stroke. The patient died approximately 12 hours after arrival. Blood cultures were positive for Neisseria meningitidis, and clinical autopsy demonstrated bilateral adrenal hemorrhage, disseminated intravascular coagulation, and findings consistent with Waterhouse-Friderichsen syndrome.

Conclusion: Fulminant meningo­coccemia and Waterhouse-Friderichsen syndrome may initially mimic acute stroke, particularly when focal neurological deficits precede overt septic features. In patients with apparent stroke symptoms and early hemodynamic instability, hyperla­ctatemia, cytopenias, or coagulopathy, sepsis and meningo­coccemia should remain high-priority differential diagnoses.

A 43-year-old man with hypertension and insulin resistance presented with an 8-month history of pulsating left hemicranial headache that worsened over the preceding 4 days and culminated in an acute thunderclap exacerbation that reached maximal intensity in less than one minute. The headache was severe (10/10), non-radiating, and accompanied by photophobia, phonophobia, a retro-orbital burning sensation, and right temporal visual-field loss. Neuroimaging revealed a 38 x 27 x 25 mm sellar/suprasellar pituitary neuroendocrine tumor with predominantly subacute hemorrhagic components and marked optic pathway compression, consistent with pituitary apoplexy. Vascular imaging (CTA, MRA, or MRV) and lumbar puncture were not performed because non-contrast head CT and sellar MRI demonstrated a hemorrhagic sellar/suprasellar lesion explaining the presentation, without clinical features mandating additional vascular or cerebrospinal fluid evaluation. The patient underwent urgent subtotal endoscopic transsphenoidal resection/decompression and received stress-dose hydrocortisone for suspected central adrenal insufficiency. Histopathology confirmed a corticotroph PitNET/adenoma (WHO 2022), ACTH/synaptophysin/CAM5.2 positive, prolactin-negative, with Ki-67 1%. Prolactin decreased from 555.09 ng/mL at presentation to 55.09 ng/mL on February 16, 2024, supporting stalk-effect hyperprolactinemia rather than prolactinoma. Follow-up evaluations showed headache resolution, stable visual function, secondary panhypopituitarism requiring hormonal replacement, and a small residual sellar lesion under surveillance.

A middle-aged woman with no known past medical history presented to the emergency department with acute visual loss that had begun approximately 4 hours before admission. She had experienced flu-like symptoms during the preceding days. Before presentation to our emergency department, she had sought care at a local clinic because of a thunderclap headache. Intravenous ketorolac was administered, after which she developed an anaphylactic reaction. After treatment of the anaphylactic reaction, she reported severe visual impairment and was transferred to our emergency department by emergency medical services.

Ophthalmologic examination showed visual acuity limited to light perception. No relative afferent pupillary defect was reported (negative Marcus Gunn sign). The remainder of the physical examination was unremarkable. Cerebrospinal fluid analysis was normal.

Axial brain computed tomography (CT) was reported to show bilateral parieto-occipital subcortical hemorrhage (Figure 1). Brain magnetic resonance imaging (MRI) showed bilateral posterior parieto-occipital abnormalities on T2-weighted and fluid-attenuated inversion recovery (FLAIR) sequences (Figure 2). Diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) sequences are shown in Figure 3. Magnetic resonance angiography (MRA), which is not included in the available figure set, was reported to show multifocal segmental narrowing and dilatation of the cerebral arteries with a string-of-beads appearance.