New review looks at early, real‑world tests of artificial intelligence in crowded emergency departments.

A new systematic review of seven prospective studies suggests that artificial intelligence (AI) can help emergency teams sort patients more quickly and, in many cases, just as accurately as traditional methods—but it’s not ready to replace human judgment. The studies, conducted in hospitals across Asia and Europe, evaluated AI tools that assist triage by predicting how urgent a patient’s condition is when they arrive at the emergency department (ED).  

Why this matters:

ED crowding is linked to worse outcomes for the sickest patients. Triage—the rapid process of deciding who needs care first—has to be both fast and reliable. The review found that AI systems, used as decision support for triage nurses, could reduce errors caused by fatigue, time pressure, or inconsistent data entry, potentially smoothing patient flow.  

What the researchers did

Following PRISMA methods, the authors screened 1,633 records and ultimately included seven real‑world, prospective studies (not just algorithm testing on old datasets). Most sites used 5‑level national triage scales (such as ESI, KTAS, TTAS, MTS) and implemented AI models ranging from machine learning and deep learning to fuzzy‑logic expert systems. Figure 1 in the review shows the flow from 1,633 records down to seven included studies.  

What they found (in plain language)

• Accuracy was generally high: Across studies, triage prediction accuracy ranged from 80.5% to 99.1%. A fuzzy‑logic system reported 99% accuracy with 99% sensitivity and specificity in one hospital; deep‑learning models performed well overall but were less consistent at the very highest urgency levels.  
• Small but real time savings: In one ED, an AI‑assisted, voice‑enabled intake tool cut triage documentation time by about 27 seconds per patient compared with manual entry—small per case, but potentially meaningful at scale.  
• Fewer dangerous misses: One large study reported a lower “miss‑triage” rate for life‑threatening cases when an AI protocol was used (0.9% vs. 1.2%).  
• Not ideal for the very sickest patients (yet): Several projects either struggled with, or intentionally excluded, the top two acuity levels (patients needing immediate resuscitation) because early AI versions can’t capture the vital “first‑impression” cues nurses assess face‑to‑face in seconds.  
• What drives triage mistakes: Factors linked with mis‑triage included arrival by ambulance, off‑hours presentation, older age, sex, and vital‑sign patterns (e.g., heart rate, blood pressure, oxygen saturation).  

What this means for patients and clinicians

Expect to see more hospitals pilot AI as a support tool, especially for the large middle of the triage spectrum (levels 3–5) that contributes most to ED crowding. In those cases, AI can highlight subtle risk patterns and standardize documentation. But the triage nurse remains in charge, and clinicians will continue to make the final call—particularly for the most critical patients where hands‑on assessment is irreplaceable.  

Caveats

These were prospective observational studies, often with modest sample sizes, and many used clinicians’ ratings as the “gold standard,” which introduces subjectivity. The authors urge larger, more rigorous trials that track objective outcomes—like ED length of stay, ICU admission, and in‑hospital mortality—before hospitals scale up AI triage widely.  

By the numbers (at a glance)

• 1,633 records screened → 7 prospective studies included.  
• Accuracy range: 80.5%–99.1% across models.  
• One fuzzy‑logic system: 99% sensitivity & specificity.  
• Time saved per triage with AI input aid: ~27 seconds.  

Bottom line for readers of The Nano Post

Early real‑world evidence says AI can be a helpful extra set of eyes at triage—speeding up data capture, flagging overlooked risks, and standardizing assessments—while nurses and physicians keep control. The next step is high‑quality clinical trials to prove that AI‑assisted triage not only predicts urgency but improves outcomes for patients who need help the most.  

Source: Nayeon Yi, Dain Baik, and Gumhee Baek. “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.