Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-supervised Hierarchical Representation for Medication Recommendation

Yuliang Liang, Yuting Liu, Yizhou Dang, Enneng Yang, Guibing Guo, Wei Cai, Jianzhe Zhao, Xingwei Wang

TL;DR

A novel hierarchical encoder named HIER is proposed to hierarchically represent diagnoses and procedures, which is based on standard medical codes and compatible with any existing methods, and develops the position encoding to explicitly introduce global hierarchical position.

Abstract

Medication recommender is to suggest appropriate medication combinations based on a patient's health history, e.g., diagnoses and procedures. Existing works represent different diagnoses/procedures well separated by one-hot encodings. However, they ignore the latent hierarchical structures of these medical terms, undermining the generalization performance of the model. For example, "Respiratory Diseases", "Chronic Respiratory Diseases" and "Chronic Bronchiti" have a hierarchical relationship, progressing from general to specific. To address this issue, we propose a novel hierarchical encoder named HIER to hierarchically represent diagnoses and procedures, which is based on standard medical codes and compatible with any existing methods. Specifically, the proposed method learns relation embedding with a self-supervised objective for incorporating the neighbor hierarchical structure. Additionally, we develop the position encoding to explicitly introduce global hierarchical position. Extensive experiments demonstrate significant and consistent improvements in recommendation accuracy across four baselines and two real-world clinical datasets.

Self-supervised Hierarchical Representation for Medication Recommendation

TL;DR

A novel hierarchical encoder named HIER is proposed to hierarchically represent diagnoses and procedures, which is based on standard medical codes and compatible with any existing methods, and develops the position encoding to explicitly introduce global hierarchical position.

Abstract

Medication recommender is to suggest appropriate medication combinations based on a patient's health history, e.g., diagnoses and procedures. Existing works represent different diagnoses/procedures well separated by one-hot encodings. However, they ignore the latent hierarchical structures of these medical terms, undermining the generalization performance of the model. For example, "Respiratory Diseases", "Chronic Respiratory Diseases" and "Chronic Bronchiti" have a hierarchical relationship, progressing from general to specific. To address this issue, we propose a novel hierarchical encoder named HIER to hierarchically represent diagnoses and procedures, which is based on standard medical codes and compatible with any existing methods. Specifically, the proposed method learns relation embedding with a self-supervised objective for incorporating the neighbor hierarchical structure. Additionally, we develop the position encoding to explicitly introduce global hierarchical position. Extensive experiments demonstrate significant and consistent improvements in recommendation accuracy across four baselines and two real-world clinical datasets.

Paper Structure

This paper contains 26 sections, 15 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Medication Recommendation
  4. The Representation of Medical Codes
  5. Preliminary
  6. Electrical Health Records (EHR)
  7. Patient Representation
  8. International Classification of Diseases (ICD)
  9. Proposed Method
  10. Hierarchical Encoder
  11. Relation Embedding
  12. Position Encoding
  13. Implementation
  14. Theoretical Insights
  15. What knowledge do existing models derive from the data?
  16. ...and 11 more sections

Figures (3)

  • Figure 1: An illustration of medication recommendation models, where (a) is the paradigm of existing methods; (b) is our method with the hierarchical encoder.
  • Figure 2: Convergence efficiency curve in the test set during training stage: the first row for GAMENet and the second row for SafeDrug. Our method significantly improves the training convergence efficiency in both baselines.
  • Figure 3: The visualization of diagnoses and procedures in the 2-d space using t-SNE, where (a) and (c) for embeddings of diagnoses and procedures; (b) and (d) for our relation embeddings of diagnoses and procedures.