Background: Large language models (LLMs) have demonstrated impressive performances in various medical domains, prompting an exploration of their potential utility within the high-demand setting of emergency department (ED) triage. This study evaluated the triage proficiency of different LLMs and ChatGPT, an LLM-based chatbot, compared to professionally trained ED staff and untrained personnel. We further explored whether LLM responses could guide untrained staff in effective triage. Objective: This study aimed to assess the efficacy of LLMs and the associated product ChatGPT in ED triage compared to personnel of varying training status and to investigate if the models’ responses can enhance the triage proficiency of untrained personnel. Methods: A total of 124 anonymized case vignettes were triaged by untrained doctors; different versions of currently available LLMs; ChatGPT; and professionally trained raters, who subsequently agreed on a consensus set according to the Manchester Triage System (MTS). The prototypical vignettes were adapted from cases at a tertiary ED in Germany. The main outcome was the level of agreement between raters’ MTS level assignments, measured via quadratic-weighted Cohen κ. The extent of over- and undertriage was also determined. Notably, instances of ChatGPT were prompted using zero-shot approaches without extensive background information on the MTS. The tested LLMs included raw GPT-4, Llama 3 70B, Gemini 1.5, and Mixtral 8x7b. Results: GPT-4–based ChatGPT and untrained doctors showed substantial agreement with the consensus triage of professional raters (κ=mean 0.67, SD 0.037 and κ=mean 0.68, SD 0.056, respectively), significantly exceeding the performance of GPT-3.5–based ChatGPT (κ=mean 0.54, SD 0.024; P<.001). When untrained doctors used this LLM for second-opinion triage, there was a slight but statistically insignificant performance increase (κ=mean 0.70, SD 0.047; P=.97). Other tested LLMs performed similar to or worse than GPT-4–based ChatGPT or showed odd triaging behavior with the used parameters. LLMs and ChatGPT models tended toward overtriage, whereas untrained doctors undertriaged. Conclusions: While LLMs and the LLM-based product ChatGPT do not yet match professionally trained raters, their best models’ triage proficiency equals that of untrained ED doctors. In its current form, LLMs or ChatGPT thus did not demonstrate gold-standard performance in ED triage and, in the setting of this study, failed to significantly improve untrained doctors’ triage when used as decision support. Notable performance enhancements in newer LLM versions over older ones hint at future improvements with further technological development and specific training.

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system (CNS) of still unclear etiology. In recent years, the search for biomarkers facilitating its diagnosis, prognosis, therapy response, and other parameters has gained increasing attention. In this regard, in a previous meta-analysis comprising 22 studies, we found that MS is associated with higher nitrite/nitrate (NOx) levels in the cerebrospinal fluid (CSF) compared to patients with non-inflammatory other neurological diseases (NIOND). However, many of the included studies did not distinguish between the different clinical subtypes of MS, included pre-treated patients, and inclusion criteria varied. As a follow-up to our meta-analysis, we therefore aimed to analyze the serum and CSF NOx levels in clinically well-defined cohorts of treatment-naïve MS patients compared to patients with somatic symptom disorder. To this end, we analyzed the serum and CSF levels of NOx in 117 patients (71 relapsing–remitting (RR) MS, 16 primary progressive (PP) MS, and 30 somatic symptom disorder). We found that RRMS and PPMS patients had higher serum NOx levels compared to somatic symptom disorder patients. This difference remained significant in the subgroup of MRZ-negative RRMS patients. In conclusion, the measurement of NOx in the serum might indeed be a valuable tool in supporting MS diagnosis.

The treatment of carotid artery stenosis (CAS) for stroke prevention is a matter of debate due to conflicting data, missing recent data, and advances in medical treatment options but also in interventional techniques and surgery. Therefore, the establishment of an easily available marker for brain damage might be a key tool in this patient group to guide treatment. A retrospective cross-sectional study was conducted leveraging a vascular surgery biobank of 95 patients aged 60 to 80 years. Serum neurofilament light chain (sNfL) and serum glial fibrillary acid protein (sGFAP) were evaluated using highly sensitive electrochemiluminescence immunoassays and z-scores. Discriminatory performance was assessed to differentiate between 19 symptomatic and 76 asymptomatic patients with CAS and their correlation with the degree of stenosis according to ultrasound-based North American Symptomatic Carotid Endarterectomy Trial (NASCET) criteria. SNfL levels were markedly elevated in symptomatic compared with asymptomatic patients (median 17.4 vs 3.8 pg/mL; P < .001). sNfL robustly discriminated between these patients (area under the curve = 0.83; 95% confidence interval, 0.72-0.94), as did the NfL z-score (area under the curve = 0.83; 95% confidence interval, 0.71-0.95). Interestingly, within the asymptomatic cohort, sNfL levels demonstrated a significant, positive correlation with the degree of stenosis (Spearman's ρ = 0.24; P = .036). Serum levels of SGFAP were also associated with symptomatic status, albeit with a P-value >.05 (0.1 vs 0.1 pg/mL; (P = .057). The study provides evidence of increased sNfL in symptomatic vs asymptomatic CAS and, of note, of ongoing neuronal or glial damage in some patients with clinically asymptomatic CAS, with a positive correlation between sNfL and the degree of CAS. sNfL is a promising and already accessible blood biomarker that may guide therapeutic decisions in this patient population. The role of sGFAP remains elusive and must be evaluated in larger studies. The CREST-2 trial has recently highlighted the high efficacy of intensive medical management in asymptomatic carotid artery stenosis (CAS), making the selection of patients for revascularization increasingly complex. Our study addresses this challenge by evaluating serum neurofilament light chain (sNfL) and serum glial fibrillary acid protein (sGFAP) as an objective biomarker for neuronal injury. Using individualized z-scores, we demonstrate that sNfL effectively differentiates symptomatic status (area under the curve = 0.828) and significantly correlates with the degree of stenosis in asymptomatic cohorts. These results suggest that sNfL can detect subclinical “silent” damage, providing a valuable biological tool to pinpoint high-risk patients who may require intervention beyond medical therapy alone. This represents a significant step toward personalized stroke prevention and refined risk stratification in the highly debated field of CAS management.

The Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS) and the Collum-Caput (Col-Cap) concept are tools for clinically assessing cervical dystonia severity. However, the accuracy of human ratings using these scales has not been systematically evaluated due to the lack of objective reference measurements. This study aims to assess and compare the accuracy of human TWSTRS and Col-Cap ratings to evaluate their robustness for clinical and research applications. One hundred pictures of 26 avatars mimicking cervical dystonia were created using the Rocketbox Avatar library. Forty-one movement disorder specialists rated the head and neck positioning of the avatars using either TWSTRS or Col-Cap. Movements were defined around two rotational levels (head, neck) in three rotational axes (pitch, yaw, roll). Ratings of angular deviations showed a mean absolute error of 5.8° (SD = 7.0). Rating accuracy was primarily influenced by the magnitude of angular deviation, with larger angles leading to greater estimation errors. Direct comparison of the rating scales revealed a higher accuracy through Col-Cap ratings (71 % vs. 63 % for TWSTRS). Years of clinical experience did not significantly affect rating accuracy. Both rating systems (TWSTRS and Col-Cap) show moderate accuracy in assessing head and neck positioning from computer-generated avatars, with Col-Cap showing slightly higher overall accuracy but struggling with precise differentiation between head and neck movements. These findings underscore the limitations of current clinical rating scales and highlight the need for more objective, reliable tools to effectively assess cervical dystonia.

BACKGROUND Large language models (LLMs) have demonstrated impressive performances in various medical domains, prompting an exploration of their potential utility within the high-demand setting of emergency department (ED) triage. This study evaluated the triage proficiency of different LLMs and ChatGPT, an LLM-based chatbot, compared to professionally trained ED staff and untrained personnel. We further explored whether LLM responses could guide untrained staff in effective triage. OBJECTIVE This study aimed to assess the efficacy of LLMs and the associated product ChatGPT in ED triage compared to personnel of varying training status and to investigate if the models’ responses can enhance the triage proficiency of untrained personnel. METHODS A total of 124 anonymized case vignettes were triaged by untrained doctors; different versions of currently available LLMs; ChatGPT; and professionally trained raters, who subsequently agreed on a consensus set according to the Manchester Triage System (MTS). The prototypical vignettes were adapted from cases at a tertiary ED in Germany. The main outcome was the level of agreement between raters’ MTS level assignments, measured via quadratic-weighted Cohen κ. The extent of over- and undertriage was also determined. Notably, instances of ChatGPT were prompted using zero-shot approaches without extensive background information on the MTS. The tested LLMs included raw GPT-4, Llama 3 70B, Gemini 1.5, and Mixtral 8x7b. RESULTS GPT-4–based ChatGPT and untrained doctors showed substantial agreement with the consensus triage of professional raters (κ=mean 0.67, SD 0.037 and κ=mean 0.68, SD 0.056, respectively), significantly exceeding the performance of GPT-3.5–based ChatGPT (κ=mean 0.54, SD 0.024; P<.001). When untrained doctors used this LLM for second-opinion triage, there was a slight but statistically insignificant performance increase (κ=mean 0.70, SD 0.047; P=.97). Other tested LLMs performed similar to or worse than GPT-4–based ChatGPT or showed odd triaging behavior with the used parameters. LLMs and ChatGPT models tended toward overtriage, whereas untrained doctors undertriaged. CONCLUSIONS While LLMs and the LLM-based product ChatGPT do not yet match professionally trained raters, their best models’ triage proficiency equals that of untrained ED doctors. In its current form, LLMs or ChatGPT thus did not demonstrate gold-standard performance in ED triage and, in the setting of this study, failed to significantly improve untrained doctors’ triage when used as decision support. Notable performance enhancements in newer LLM versions over older ones hint at future improvements with further technological development and specific training.