This Arkangel AI whitepaper is a targeted narrative review of 28 sources on selected uses of artificial intelligence in emergency medicine and critical care: triage and prioritization, sepsis prediction, and surveillance of clinical deterioration. It classifies the evidence by stage (retrospective development, external validation, prospective evaluation, impact on processes or outcomes, and routine implementation) and adds research questions on clinical information and documentation support.
Maturity is uneven. Retrospective models show strong discrimination, such as an internal-test AUC of 0.94 for a sepsis model combining nursing notes and clinical variables, yet the external validation of the widely deployed Epic Sepsis Model reported an AUC of 0.63 (95% CI 0.62–0.64). Prospective evidence is limited to a single-center triage evaluation and a Thai cluster-randomized trial of a sepsis alert that improved timely antibiotics. The corpus contains no primary evaluation of these AI tasks in Latin America; regional sources, such as one describing 257 heterogeneous ICUs, cover burden and infrastructure. The authors recommend silent prospective evaluation only after the clinical response, capacity and stopping criteria are defined. It is a whitepaper, not peer reviewed.
