Playing the Devil’s Advocate A recent JAMA Commentary provides an analysis of ED triage research and the application of AI and Large Language Models (LLM) to ED Triage on-site. Friedmans’s analysis suggests it may be too early to know AI’s effect on ED triage. He suggests a human-machine hybrid — AI as a clinical-co-pilot to clinical decision makers. Williams asks “Is just being able to do something the bar for using AI, or is it being able to do something well, for all types of patients?”

One wonders: Are EDs the appropriate venue for which to develop an AI/triage tool? Increasingly crowded EDs are a major systemic problem that may need to be solved first.  Given ESI triage mismatch (specific to ED resource use), can the ESI tool meaningfully apply to self-triaged patients, some of whom are inappropriately now on-site?

Since the 1980s, ED overcrowding has continued to increase. Are patients (perhaps low-income groups, lacking adequate access) increasingly self-triaging to the ED, compounding overcrowding? Recent studies point out negative consequences: reduced patient safety, increased staff stress level, increased error, and delays in care.  One wonders if ED overcrowding will be solved by enlisting AI to support ED staff to triage more rapidly?

In my opinion, ED overcrowding is a symptom of inadequate patient access to more appropriate, less emergent venues – open for expanded hours, not just office hours. Inadequate access also can also sabotage the effectiveness of telephone triage, its original purpose — to reduce inappropriate ED (Urgent Care, Office/Clinic visits) and reduce Hospital Readmissions. It was also intended to make health systems overall more cost-effective. Without expanded access and improved pre-hospital triage systems, ED overcrowding will continue to worsen.

Is pre-hospital triage still viable? Will the future of triage be AI-Enhanced Symptom Checkers to help patients self-triage to the ED where AI-Supported ED Triage eliminates the need for, (and expense of)  pre-hospital triage altogether? Does the healthcare industry have the luxury of time and financial resources required to develop two AI tools - one for patients, the other for ED settings?