AI home triage tool for TCM gynecology scored 100% on a safety test, study finds

In testing, the model beat every baseline on RAGAS automatic scoring and on expert human evaluation. On a safety test set it reached 100% sensitivity and specificity in separating conditions suited to self-care from those needing urgent hospital care.

The work targets a gap in home access to TCM gynecology care, where hospital visit costs, privacy worries and lack of immediate guidance are barriers. The results come from evaluation on test sets, not from use by patients at home.

The researchers say home access to TCM gynecology care is hampered by the cost of frequent hospital visits, privacy concerns and the lack of immediate professional guidance. General-purpose AI is convenient, but the gap between colloquial symptom descriptions and formal TCM diagnostic logic, plus hallucination risk, keeps it from giving safe advice.

To address this, the team built a domain knowledge graph under expert supervision, drawing 10,231 entities and 32,051 relationships from 741 high-quality clinical case records. A framework of four specialized agents uses it to simulate the expert diagnostic process.

Evaluation used RAGAS automatic scoring and expert human ratings, where MAGR-TCM outperformed all baselines, plus a safety test set on which it scored 100% sensitivity and specificity. The authors say this suggests it can mitigate hallucinations and support women's daily self-care and preliminary triage.