Abstract Background Etomidate, a well‐established intravenous induction agent valued for haemodynamic stability, has emerged as a recreational drug via e‐cigarette vaping (“space oil”), particularly in Asia, including Hong Kong, prompting regulatory controls amid rising emergency presentations. In Hong Kong, etomidate misuse via e‐cigarettes was first identified in late 2023, with rapid escalation leading to its classification as a dangerous drug in 2025. Methods Six electronic databases (PubMed/MEDLINE, Embase, Web of Science Core Collection, Scopus, SciFinder and Google Scholar) were searched from inception to March 2026 using keywords for etomidate, vaping (“space oil,” “Kpod”), toxicity and pharmacology. Relevant papers, case reports, preclinical studies and reviews were narratively synthesised. Google searches identified additional authority announcements and policy documents. Results Acute etomidate vaping produces rapid‐onset hypnotic effects, including confusion, drowsiness, incoordination and myoclonus, leading to the colloquial term “zombie cigarettes”. Chronic exposure inhibits adrenal 11β‐hydroxylase, mimicking a rare form of congenital adrenal hyperplasia, with accumulation of steroid precursors that cause hyperandrogenism and aldosteronism (hypokalemia and hypertension) in the context of cortisol deficiency. Case of withdrawal with irritability and visual hallucination after cessation of etomidate vaping has been reported. Heterogeneity in e‐cigarette devices and formulations limits characterisation of the pharmacokinetics of etomidate delivered by inhalation. Conclusions Etomidate vaping poses acute and chronic neurological and endocrine risks, necessitating clinician vigilance for intoxication recognition and timely supportive care. Multidisciplinary surveillance, e‐liquid composition analysis and research on the pharmacology of aerosolised are urgently needed.
Abstract Background Out‐of‐hospital cardiac arrest (OHCA) is a leading cause of death worldwide. Telephone‐assisted cardiopulmonary resuscitation (TCPR) and automated external defibrillators (AEDs) improve survival, but traditional emergency systems face delays due to traffic and geography. Objectives This narrative review examined the current application, technical challenges, and future potential of combining TCPR with unmanned aerial vehicle (UAV)‐mounted AED systems in OHCA. Methods We searched PubMed, Embase, and Scopus up to March 2025 using terms including “telephone CPR,” “TCPR,” “unmanned aerial vehicle,” “drone,” “AED,” “out‐of‐hospital cardiac arrest,” and “emergency response.” Inclusion criteria: English or Chinese peer‐reviewed studies on TCPR, UAV‐AED, or their integration in OHCA. Exclusion criteria: conference abstracts, editorials, and studies without original data or clear methodology. Results Evidence from simulation and feasibility studies suggests that UAV‐AED systems reduce AED delivery time compared with emergency medical services. TCPR increases bystander CPR rates and quality. However, direct clinical outcome data on survival improvement from integrated systems remain limited. Key challenges include UAV endurance, regulatory barriers, public acceptance, and integration with existing emergency systems. Conclusions TCPR combined with UAV‐AED has potential to improve OHCA response, but evidence is largely from simulation and feasibility studies. Future research should focus on real‐world outcome evaluation, cost‐effectiveness, and system integration.
Abstract Introduction We compared the Kiwi‐grip technique with the traditional pencil grip technique for bougie‐assisted videolaryngoscopic endotracheal intubation in terms of intubation time on a manikin model with normal and simulated cervical spine injury airways. Methods In this prospective randomised crossover manikin trial, emergency department doctors performed videolaryngoscopic intubation using both the Kiwi‐grip and traditional pencil grip techniques. Each participant completed scenarios with a normal airway and a simulated cervical spine injury. Participants were randomly assigned to the order of techniques. Following standardised demonstration and familiarisation, intubation time, success rate and dental injury in each scenario were recorded. Upon completion of all scenarios, participants rated the two techniques using a questionnaire. Results Thirty‐six participants completed the study. In the normal airway scenario, mean intubation time was significantly shorter with the Kiwi‐grip technique (25.14 ± 7.23 s) compared with the traditional pencil grip (30.64 ± 8.55 s; p < 0.001). In the cervical spine injury scenario, mean intubation time was also significantly shorter with Kiwi‐grip (22.94 ± 4.66 s) versus traditional pencil grip (28.57 ± 8.75 s; p < 0.001). Dental injury rates were lower with Kiwi‐grip in both scenarios (normal: 22.2% vs. 36.1%; cervical: 47.2% vs. 58.3%). Twenty‐nine participants (80.6%) would recommend the Kiwi‐grip technique to colleagues, and 23 (63.9%) felt it improved their confidence in intubation. Conclusion In this manikin trial, the Kiwi‐grip technique, with the tube preloaded, was associated with a shorter intubation time than the traditional pencil grip in both a normal airway and a simulated cervical spine injury, but the effect of the grip cannot be isolated.
Abstract Introduction High‐volume urban emergency departments (EDs) must allocate experienced physicians without creating another front‐end bottleneck. In July 2025, our ED introduced a bundled redesign in which senior physicians assumed continuous primary responsibility for an Emergency Severity Index (ESI) level 3 consultation‐room pathway; residents were removed from consultation‐room care and total on‐site physician staffing increased. Methods This single‐center retrospective before‐and‐after study compared random samples of potentially eligible encounters from January 2025 (preimplementation) to December 2025 (postimplementation). The primary outcome was time from triage to the first physician order. Primary regression models adjusted for age, sex, weekday arrival, night‐shift arrival, and fever/respiratory chief complaint; sensitivity analyses additionally adjusted for final diagnosis category. Results The analysis included 443 preimplementation and 404 postimplementation encounters. Median time to first physician order was 37.3 versus 17.1 min (median difference, −20.2 min; bootstrap 95% CI, −23.8 to −16.0). The adjusted postimplementation‐to‐preimplementation geometric mean ratio was 0.52 (95% CI, 0.46–0.59; p < 0.001). Adjusted resource‐use estimates were inconclusive. Exploratory adjusted ORs were 0.62 (95% CI, 0.41–0.96) for any 72‐h revisit and 0.63 (95% CI, 0.41–0.97) for related revisit; these secondary analyses were not adjusted for multiple comparisons. The exploratory unadjusted risk ratio for the composite acute‐care admission outcome was 1.02 (95% CI, 0.50–2.09). Conclusions A continuous senior physician‐led consultation‐room responsibility redesign was associated with earlier documented physician ordering in a high‐volume ESI level 3 medical pathway. However, the evaluated intervention was a bundled operational change, and the study cannot determine the independent contribution of senior physician responsibility, resident removal, or increased staffing.
Abstract Background The growing demand for emergency services highlights the necessity for innovative approaches to expedite patient management. The introduction of artificial intelligence tools has created the possibility of improving emergency triage efficiency. Objective This study examined the accuracy of ChatGPT, a baseline untrained large language model (LLM), in predicting emergency triage categories compared with frontline emergency department (ED) nurses, using local ED triage guidelines in Hong Kong. Methods This study is a single‐centre, retrospective analysis of 235 triage cases sampled over a year. We evaluated the concordance of triage decisions between the triage team and the audit group, and between GPT‐4 and the audit group, using quadratic weighted kappa (QWK). The overall accuracy between groups was assessed using the McNemar test. Results The level of agreement in triage decisions between the triage team and the audit group was QWK 0.922 (95% CI., 0.897–0.947), which was significantly superior to that of GPT‐4 (QWK 0.847; 95% CI., 0.808–0.886). The rate of appropriate triage in GPT‐4 was 78.72% (95% CI: 73.05%–83.47%), which was significantly lower than that of nurses' triage ( p < 0.001). Conclusion This study found that untrained GPT‐4 achieved good agreement with the gold standard but underperformed frontline nurses in emergency triage of ED patients. GPT4 triage had a higher over‐triage rate but a similar under‐triage rate to the triage team. Prospective studies comparing trained LLM models with frontline nurses should be considered in ED settings.
