Patient-Conditioned Dual Hypergraph Reasoning for Auditable Traditional Chinese Medicine Prescription Support

Jul 16, 2026

Upload by Eliana

Traditional Chinese medicine (TCM) prescription support demands patient-specific reasoning from clinical narratives to syndromes, treatment principles, herbs, and doses, yet free-form language-model generation is hard to audit against explicit evidence. Static TCM knowledge resources offer useful priors but do not decide which diagnostic and prescription relations to emphasize per patient. A patient-conditioned dual-hypergraph framework is proposed for auditable prescription construction. Dynamic patient-weighting activates individualized diagnostic and herb-dose paths while retaining case-level auditability, evaluated on TCM-SD and TCM-BEST4SDT.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

Introduction

Motivation and problem setting

Related resources and limitations

Proposed dual-hypergraph framework

Key contribution

Patient-Conditioned Dual Hypergraph Reasoning for Auditable Traditional Chinese Medicine Prescription Support Weizhi Nie, Shaojin Bai, Weijie Wang, and Yuting Su arXiv :2607 .04025v 1 [ cs .IR] 4 Jul 2026 Abstract—Traditional Chinese medicine (TCM) prescription support requires patient-specific reasoning from clinical narratives to syndromes, treatment principles, herbs, and doses. Direct language-model generation can produce fluent prescriptions, but its decisions are difficult to audit against explicit clinical evidence. Static TCM knowledge resources provide useful priors, but they cannot determine which diagnostic and prescription relations should be emphasized for an individual patient. We propose a patient-conditioned dual hypergraph framework for auditable TCM prescription support. The first hypergraph organizes symptom, tongue, pulse, and other clinical evidence around syndrome and treatment-principle reasoning. The second hypergraph organizes syndrome, treatment, disease-context, herb, retrieval, and dose-prior evidence for prescription construction. Unlike static knowledge graphs or fixed hypergraphs, both hypergraphs are dynamically weighted by the patient representation. This design enables individualized activation of diagnostic and prescription paths, supporting personalized syndrome differentiation and herb-dose recommendation while preserving case-level auditability. Experiments on TCM-SD show that dynamic weighting in the first hypergraph improves MacBERT syndrome differentiation to 0.8297 accuracy and 0.3288 macro-F1. On TCM-BEST4SDT, the second hypergraph achieves the best mean Herb-F1 of 0.3111 across three seeds, and the full connected pipeline reaches 0.3074 Herb-F1, close to the oracle setting. A 50-case real-world CAP audit further suggests practical review potential, while highlighting the need for prospective dose-safety validation. Index Terms—Biomedical informatics, clinical decision support, hypergraph reasoning, interpretable artificial intelligence, traditional Chinese medicine, large language models. I. INTRODUCTION TRADITIONAL Chinese medicine (TCM) prescription is a structured clinical reasoning process rather than a single text-generation task. A clinician interprets chief complaints, symptom evolution, tongue and pulse signs, disease context, and treatment response. The clinician then maps this evidence to syndrome differentiation, therapeutic principles, formula organization, herb selection, and dose adjustment. This reasoning chain is relational and many-to-many. Multiple symptoms may jointly support one syndrome, one syndrome may imply several therapeutic principles, and one principle may activate a coordinated group of formula structures and herb roles. These properties make TCM prescription support a useful but difficult setting for interpretable biomedical artificial intelligence. This work is a computational decision-support study. It is not intended for autonomous clinical prescription. Weizhi Nie, Shaojin Bai, Weijie Wang, and Yuting Su are with Tianjin University, Tianjin, China. Recent language models provide useful interfaces for clinical text processing [1]–[5] . They can summarize narratives, extract signs, and produce fluent explanations. However, direct generation is poorly aligned with the requirements of medical decision support. A generated prescription may be linguistically plausible while remaining disconnected from explicit syndrome evidence or domain constraints. In biomedical settings, model outputs should therefore be auditable and constrained by prior knowledge rather than treated as free-form natural language decisions [6], [7] . Knowledge resources for TCM provide a complementary foundation. Databases such as SymMap, ETCM, TCMSP, TCMID, BATMAN-TCM, and HERB encode relations among symptoms, diseases, syndromes, formulas, herbs, ingredients, targets, and pharmacological evidence [8]–[14] . Yet these resources are typically static. They specify

FAQs

Why is direct language-model generation insufficient for TCM prescription support?

Generated prescriptions can be fluent but may not remain connected to explicit syndrome evidence or medical constraints, making outputs difficult to audit reliably.

What does the dual-hypergraph framework model?

H1 links symptoms and clinical evidence to syndrome and treatment-principle reasoning, while H2 organizes syndrome, disease context, retrieval evidence, herb roles, and dose priors for prescription construction.

How does patient conditioning improve interpretability and personalization?

Both hypergraphs are dynamically weighted by the patient representation, enabling individualized activation of diagnostic and prescription paths while preserving case-level auditability.

Jul 16, 2026

Upload by Eliana