Emergency Medicine

LLM-driven, near‑real‑time feedback increased SEP‑1 compliance in ED physicians

In a cluster randomized trial of 66 ED attendings (301 SEP-1 patients), LLM-delivered discharge feedback raised SEP-1 compliance from 70.1% to 82.9%. The absolute improvement was 13.0% (95% CI, 2.5%–23.4%; OR 2.10; P = .02). The largest change was fewer noncompletions of the documentation‑sensitive 30 mL/kg fluid bolus component. No differences were seen in 30‑day mortality or ICU admissions, and LLM–expert agreement was 92%. 1

Prehospital pediatric telemedicine in ambulances proved feasible and safe in Karachi pilot

In a cluster‑randomized pilot of 10 ambulances, 151 enrolled parent–patient encounters completed in 96% of intervention calls. Audio/video quality was rated acceptable in 87% and 91% of staff rated the system usable. PEWS completion was 100% in ambulances with telemedicine versus 86% in the ED. No interactions were judged unsafe, supporting feasibility and justification for larger effectiveness trials. 2

Generative AI assistance in Kenyan primary care was safe but did not reduce 14‑day treatment failures

Across 16 clinics and 9,691 patients, treatment failure within 14 days occurred in 2.2% (LLM) versus 2.0% (control). Adjusted odds ratio was 0.77 (95% CI, 0.55–1.08; P = 0.13), showing no significant reduction in the primary outcome. No serious adverse events were attributed to the LLM and independent review found no safety signal. The trial suggests safety but only modest, if any, clinical benefit from LLM decision support in this setting. 3