AI trong phân loại điện thoại y tá

Telephone triage nurses assess symptoms without seeing the patient and decide how urgently care is needed. Most U.S. nurse advice lines and after-hours services use written decision-support protocols. The best known are the Schmitt-Thompson protocols, developed by pediatrician Barton Schmitt and internist David Thompson. Each protocol covers a symptom such as abdominal pain or fever. Questions are ordered from most to least serious, and the first positive answer points to a disposition, from calling emergency services now, to going to the emergency department, to being seen within 24 hours, to home care advice. The American Academy of Ambulatory Care Nursing publishes practice standards for telehealth nursing that expect nurses to use protocols together with their own judgment. AI fits around this structure. Intake bots collect basic information before a nurse calls back. Speech recognition transcribes calls and drafts documentation. Language models map a caller's free-text description to the most likely protocol. Prioritization models rank the callback queue. Post-call analytics review recordings for missed red flags. A common misconception is that consumer symptom checkers do the same job as nurse triage. A 2015 BMJ study by Semigran and colleagues tested many online symptom checkers and found their triage advice was often wrong, though generally cautious. Nurses catch things apps miss: a breathless voice, a confused answer, a parent who sounds panicked, or a caller who says 'no' to chest pain but describes pressure. The central safety concern is undertriage, meaning sending someone to a lower level of care than they need. Safe designs follow a few rules. Any red-flag phrase triggers immediate escalation. Ambiguous or incomplete input escalates to a nurse rather than being guessed. AI can recommend raising urgency but cannot lower a disposition without a nurse. Every automated step is logged for audit.