English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

MedGraph: Graph Memory for Maintaining Patient State in Long-Term Medical Dialogue

Forum topic · 小凯 · 2026-11-07

Summary

MedGraph (arXiv:2611.08433) is a graph-based memory system for LLM medical assistants that track patients across visits. The authors analyze errors of five existing memory systems on longitudinal medical dialogue and identify three recurring failure modes: retrieved evidence reflects an outdated clinical state, includes another person's health facts, or lacks safety facts the answer depends on. MedGraph addresses these by extracting dated facts from each session, attributing each fact to the person it describes, and linking medication, diagnosis, and measurement records into chains that preserve earlier states. Retrieval ranks a patient's chains as units, reads each chain at the time point the question refers to, appends original dialogue text, and includes all recorded safety facts in a patient profile. On MedMemoryBench, MedGraph achieves the highest overall accuracy with both GPT and Qwen backbones, exceeding the strongest of five baselines by 10.5 and 10.4 percentage points respectively, with efficient memory construction and answering in the GPT setting.

Paper Overview

Research Area: NLP Authors: Donghua Zhang, Zizheng Wang, Ze Dong, Yiding Feng, Siyu Qian Published: 2026-11-05 arXiv: 2611.08433

Abstract

LLM-based medical assistants that follow a patient across visits need a memory that keeps the patient's clinical state correct as it changes. The authors examine the errors of five memory systems on longitudinal medical dialogue and find three recurring causes:

1. The retrieved evidence gives an outdated state. 2. It includes another person's health facts. 3. It lacks a safety fact the answer depends on.

They present MedGraph, a graph memory built for these three cases:

  • Construction extracts dated facts from each session, attributes each fact to the person it describes, and links medication, diagnosis, and measurement records into chains that keep earlier states.
  • Retrieval ranks the patient's chains as units, reads each chain at the time the question asks about, and adds the original dialogue text; a patient profile lists every recorded safety fact with every question.

Results

On MedMemoryBench, MedGraph achieves the highest overall accuracy with both a GPT and a Qwen backbone — 10.5 and 10.4 points above the strongest of five baselines — with efficient memory construction and answering in the GPT setting.

--- *Auto-collected on 2026-11-08*

Tags

#medgraph#llm#medical-ai#graph-memory#longitudinal-dialogue#nlp#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178636946