Otolaryngology (ENT)

Protocol: RCT of AI-assisted personalized education vs SMS in older postoperative HNC patients

This single-center randomized trial will enroll 100 postoperative head and neck cancer patients aged ≥60 to test an AI assistant versus standardized SMS education. The intervention delivers 12 months of personalized, phased multimedia education and real-time AI Q&A; controls receive matched-frequency SMS. Primary outcomes are patient-reported mental health, social support, and quality-of-life measures assessed at five timepoints to 12 months. Generalisability may be limited by single-center recruitment and reliance on self-reported outcomes. 1

Systematic map: LLM clinical interventions show mixed benefits and reporting gaps

This evidence map of 55 studies (2022–2025) found most were human–AI collaborative designs (65.5%) focusing on decision support and symptom management. LLM-assisted interventions commonly improved psychological health endpoints, while LLM-only studies emphasized functional and process outcomes. Diagnostic accuracy in randomized trials ranged 0.65–0.88 and was lower and more variable than non-randomized reports. Reporting quality was suboptimal with mean 78.8% CONSORT-AI adherence and frequent omissions about data quality and error handling. 2

Exploratory triangulation identifies 8 early ICU features linked to mortality in grade 4 irAEs

In 26 ICU patients with grade 4 immune‑related adverse events, triangulation of univariate and machine learning methods highlighted eight robust mortality-associated features. Features linked to higher mortality included vasopressor use, oxygen therapy, day‑1 and day‑2 lactate, pneumonia, and higher neutrophil percentage. Higher BMI and concurrent VEGF‑inhibitor exposure were inversely associated with mortality in this cohort. Findings are hypothesis-generating because of the small sample and exploratory design and need external validation. 3