Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Traceable Drug Recommendation over Medical Knowledge Graphs

Yu Lin, Zhen Jia, Philipp Christmann, Xu Zhang, Shengdong Du, Tianrui Li

TL;DR

TraceDR tackles the problem of non‑explainable drug recommendations by integrating a medical knowledge graph with a novel evidence graph and a patient-aware graph neural network. It retrieves candidate drugs with BM25, constructs an evidence graph from MKG facts, and uses a patient-attention GNN to jointly predict drugs and the corresponding evidence, enabling traceability. The DrugRec benchmark, built from an MKG and an automated LLM-driven data-construction pipeline, provides 21,000 synthetic EHRs across 14,023 diseases and 100,186 drugs for large-scale evaluation. Empirical results show TraceDR outperforms LLMs, traditional DR methods, retrieval baselines, and standard GNNs while maintaining a low DDI rate and offering interpretable evidence for each recommendation. This work advances practical, explainable DR and supplies a scalable dataset to spur future research in clinical decision support.

Abstract

Drug recommendation (DR) systems aim to support healthcare professionals in selecting appropriate medications based on patients' medical conditions. State-of-the-art approaches utilize deep learning techniques for improving DR, but fall short in providing any insights on the derivation process of recommendations -- a critical limitation in such high-stake applications. We propose TraceDR, a novel DR system operating over a medical knowledge graph (MKG), which ensures access to large-scale and high-quality information. TraceDR simultaneously predicts drug recommendations and related evidence within a multi-task learning framework, enabling traceability of medication recommendations. For covering a more diverse set of diseases and drugs than existing works, we devise a framework for automatically constructing patient health records and release DrugRec, a new large-scale testbed for DR.

Traceable Drug Recommendation over Medical Knowledge Graphs

TL;DR

TraceDR tackles the problem of non‑explainable drug recommendations by integrating a medical knowledge graph with a novel evidence graph and a patient-aware graph neural network. It retrieves candidate drugs with BM25, constructs an evidence graph from MKG facts, and uses a patient-attention GNN to jointly predict drugs and the corresponding evidence, enabling traceability. The DrugRec benchmark, built from an MKG and an automated LLM-driven data-construction pipeline, provides 21,000 synthetic EHRs across 14,023 diseases and 100,186 drugs for large-scale evaluation. Empirical results show TraceDR outperforms LLMs, traditional DR methods, retrieval baselines, and standard GNNs while maintaining a low DDI rate and offering interpretable evidence for each recommendation. This work advances practical, explainable DR and supplies a scalable dataset to spur future research in clinical decision support.

Abstract

Drug recommendation (DR) systems aim to support healthcare professionals in selecting appropriate medications based on patients' medical conditions. State-of-the-art approaches utilize deep learning techniques for improving DR, but fall short in providing any insights on the derivation process of recommendations -- a critical limitation in such high-stake applications. We propose TraceDR, a novel DR system operating over a medical knowledge graph (MKG), which ensures access to large-scale and high-quality information. TraceDR simultaneously predicts drug recommendations and related evidence within a multi-task learning framework, enabling traceability of medication recommendations. For covering a more diverse set of diseases and drugs than existing works, we devise a framework for automatically constructing patient health records and release DrugRec, a new large-scale testbed for DR.

Paper Structure

This paper contains 12 sections, 2 equations, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. The TraceDR Methodology
  3. Relevant Drugs Retrieval
  4. Evidence Graph Construction
  5. Recommendation Prediction
  6. The DrugRec Benchmark
  7. Limitations of Existing Benchmarks
  8. Data Construction Framework
  9. Experimental Setup
  10. Experimental Results
  11. Related Work
  12. Conclusion

Figures (1)

  • Figure 1: Overview of TraceDR's three stages: TraceDR first retrieves candidate drugs via BM25, for which an evidence graph with relevant medical information is constructed. Our GNN, with dedicated patient-attention, scores drugs in the graph.