Joseph Nano Uncategorized

AI is starting to help nurses triage in crowded ERs — but it’s not ready to replace them

A new review of real‑world studies finds artificial intelligence can speed up and sometimes sharpen emergency‑department (ED) triage, while highlighting big gaps that need fixing before hospitals lean on it widely.  

What’s new

Researchers from Ewha Womans University reviewed seven prospective (real‑world) studies that tested AI tools alongside standard triage in university or tertiary‑care EDs across Korea, Germany, Iran, Taiwan, China and Greece. The flow diagram on page 4 shows the team screened 1,633 papers and included seven that actually trialed AI in practice.  

What the studies tried

Most tools used machine‑learning models to sort patients into urgency levels; one used fuzzy logic. Nearly all worked with common five‑level triage systems such as ESI, MTS, KTAS or TTAS (one used a four‑level scale). Table 1 (pages 5–6) summarizes sample sizes (from 146 to 17,072 patients), settings and outcomes.  

Key takeaways (in plain language)

  • Faster documentation: A voice‑AI system that helped capture vital signs and chief complaint cut triage time by ~27 seconds (median 204 s vs. 231 s) but left some fields incomplete (≈82% completion for chief concern). Page 10.  
  • Accuracy ranged from good to excellent: Across studies, triage prediction accuracy ran ~80.5%–99.1%. A deep‑learning model’s overall accuracy hit 84.6%, while a fuzzy‑logic tool reported ~99% sensitivity and specificity in its setting. Pages 8–9.  
  • Safety signals were mixed: In one large Chinese study, an AI triage protocol reduced life‑threatening mis‑triage from 1.2% to 0.9% versus usual care. Page 6 and page 8 (Table 4).  
  • Not all apps get urgency right: A symptom‑checker used on walk‑in patients over‑triaged 57% and under‑triaged 9%, though 94.7% of cases were considered clinically “safe.” Page 10.  
  • Weak spots at the sickest end: Performance often dipped for the most critical levels (1–2); one neural‑network model’s precision for level‑1 cases was 33%. Pages 8–9.  

Why it matters

ED crowding is tied to worse outcomes. AI that standardizes decisions, flags high‑risk patients sooner, or speeds routine data entry could help nurses and physicians move the right patients to care faster and safely park the rest. The review suggests AI is most promising for the large middle group (levels 3–5) that drives ED bottlenecks. Discussion, pages 10–12.  

Important caveats

  • Evidence is still early and uneven: different AI types, different triage scales, and small samples in some studies make comparisons hard. Pages 7–11.  
  • Human judgment remains crucial—especially for the sickest patients, where “first‑impression” assessment in the first seconds can’t be captured by data alone. Pages 10–11.  
  • Most studies judged “high quality” on reporting checklists, but several lacked details on sample‑size justification and handling of missing data. Table 2 on page 7; Table 3 on page 8.  

Bottom line

AI triage tools are promising teammates for ED nurses—not replacements. They can save time, reduce certain errors, and aid risk sorting, but performance varies, and the most critical cases still rely on expert eyes and hands. The authors call for larger, rigorous trials using hard outcomes like ED length of stay, ICU admission and mortality before broad rollout. Conclusion, pages 12–13.  

Source: Yi N, Baik D, Baek G. “The effects of applying artificial intelligence to triage in the emergency department: A systematic review of prospective studies.” Journal of Nursing Scholarship (2025).  

Editor’s note: This article is for information only and is not a substitute for professional medical advice.