Fostering psychological safety in classroom assessments: moderated-mediation effects of bullying, gender and counselling.

The integration of artificial intelligence [] into clinical decision support systems exemplifies one of the most revolutionary advancements in contemporary medicine. Large Language Models (LLMs), in particular, have exhibited exceptional capacities in analyzing unstructured clinical data, responding to medical inquiries, and assisting in diagnostic reasoning []. Their application encompasses various medical specialties, providing potential tools for education, documentation, and initial analysis. Nevertheless, considerable challenges remain in implementing these models within complex, high-stakes clinical settings, particularly in specialized domains such as hepatology. Conventional large language models frequently produce fluent and contextually appropriate responses; however, these outputs are not consistently underpinned by explicit clinical reasoning, systematic application of established medical guidelines, or dependable execution of multi-step diagnostic decision-making processes [, ]. This limitation manifests in inconsistencies, factual hallucinations, and an inability to reliably apply structured clinical algorithms to nuanced patient presentations []. These limitations are particularly apparent in the management of acute cholangitis. Acute cholangitis is a potentially life-threatening biliary tract infection that requires prompt diagnosis, severity assessment, and timely intervention []. The Tokyo Guidelines 2018 (TG18) provide a standardized framework for the diagnosis and severity grading of acute cholangitis, incorporating clinical signs, laboratory findings, and imaging features []. Optimal management requires precise integration of these parameters within the TG18 framework to determine appropriate treatment strategies, including antibiotic therapy, biliary drainage timing, and intervention modality selection [].Current LLMs, when tasked with such challenges, frequently exhibit guideline misapplication, difficulty in synthesizing multimodal data, and failure in complex differential diagnosis, leading to potentially unsafe recommendations []. Consequently, their utility as standalone clinical tools remains limited without mechanisms for verification, reasoning traceability, and adherence to medical knowledge structures. To address these limitations, neuro-symbolic AI has emerged as a promising paradigm. This approach combines the pattern recognition capabilities of neural networks with the logical reasoning and explicit knowledge representation of symbolic AI []. Neural networks excel at processing unstructured data, while symbolic systems operate on defined rules and enable deductive reasoning. A neuro-symbolic system can parse clinical vignettes using its neural components, extract relevant features, and map them onto a symbolic knowledge graph of medical guidelines []. This process enables explicit and auditable clinical reasoning. Recent studies have demonstrated the feasibility of such architectures for specific medical tasks, with improvements in accuracy, reliability, and explainability compared to conventional LLMs [, ]. However, the application of neuro-symbolic AI to the comprehensive management of complex disease groups such as cholangitis, which involves multiple subtypes and management phases, remains largely unexplored [].