Answers are grounded in this paper and cite their sources.
Patient portal messaging carries a substantial share of outpatient communication, and rising message volume has slowed clinical response to urgent concerns embedded in routine traffic. Whether large language models (LLMs) can triage these messages, and whether locally hosted open-weight models can match cloud-hosted ones, remains unclear. We evaluated six LLMs (GPT-4o, GPT-5, GPT-5 Mini, Llama 4 Scout, GPT-OSS 20B, Gemma 3 27B) against a physician reference standard in three cohorts: 500 primary care messages enriched for urgent cases, 503 epilepsy clinic messages from the same health system, and 300 patient-authored messages from public platforms outside it. Under Majority adjudication in primary care, GPT-5 missed no urgent message, and four of six models, including one deployable on-premises, reached sensitivity 0.97 or higher and negative predictive value 0.98 or higher. Rank ordering held across cohorts. Both cloud-hosted and on-premise LLMs can rule out urgent messages with high sensitivity, though specificity remained limited.
No takes yet. Share an insight, caveat, or question.
Junior et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: