Tao LLM (care transitions)
2026
An LLM Chatbot to Facilitate Primary-to-Specialist Care Transitions: A Randomized Controlled Trial
In 2,069 patients attending specialist clinics, an LLM chatbot taking the history and drafting a referral before the visit reduced specialist consultation duration compared with no chatbot before the specialist visit.
Result
Primary endpoint: Physician consultation duration, physician-perceived care coordination and patient-reported communication ease (co-primary)
Consultation 3.14 ± 2.25 vs 4.41 ± 2.77 min (28.7% reduction; P<0.001); physician-perceived care coordination 3.69 ± 0.90 vs 1.73 ± 0.95 (P<0.001); patient-reported communication ease 3.99 ± 0.62 vs 3.44 ± 0.97 (P<0.001). PreA-only and PreA-human arms were equivalent.
- Subtopic
- Patient-facing use
- Design
- Three-arm RCT (PreA alone, PreA with staff support, no PreA), two health centres, China
- Relevance
- Important: one of several pillars.
- Source
- Tao X, et al. Nat Med. 2026;32(3):934-942.doi 10.1038/s41591-025-04176-7
Discussion & critique
One of the first patient-randomised trials of a patient-facing LLM in routine care, showing an autonomously operating chatbot can shorten specialist visits and that co-design with local stakeholders beat fine-tuning on local dialogue data. Endpoints are process and perception measures rather than clinical outcomes, and the trial ran at two centres.