Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OntoMedRec: Logically-Pretrained Model-Agnostic Ontology Encoders for Medication Recommendation

Weicong Tan, Weiqing Wang, Xin Zhou, Wray Buntine, Gordon Bingham, Hongzhi Yin

TL;DR

OntoMedRec tackles data sparsity in medication recommendation by leveraging medical ontologies through Logic Tensor Networks to create logically-grounded, domain-agnostic ontology encoders. The three encoders (for diagnoses, procedures, and medications) are pre-trained with DAG-aware axioms and an axiom-oriented sampling strategy, then aligned with indication data from MEDI. The pre-trained embeddings are model-agnostic and can initialize both instance-based and longitudinal downstream models, improving performance especially in few-shot, sparse scenarios. This provides a scalable, interpretable pretraining paradigm for ontology-aware clinical decision support with demonstrated improvements on the MIMIC-III benchmark.

Abstract

Most existing medication recommendation models learn representations for medical concepts based on electronic health records (EHRs) and make recommendations with learnt representations. However, most medications appear in the dataset for limited times, resulting in insufficient learning of their representations. Medical ontologies are the hierarchical classification systems for medical terms where similar terms are in the same class on a certain level. In this paper, we propose OntoMedRec, the logically-pretrained and model-agnostic medical Ontology Encoders for Medication Recommendation that addresses data sparsity problem with medical ontologies. We conduct comprehensive experiments on benchmark datasets to evaluate the effectiveness of OntoMedRec, and the result shows the integration of OntoMedRec improves the performance of various models in both the entire EHR datasets and the admissions with few-shot medications. We provide the GitHub repository for the source code on https://anonymous.4open.science/r/OntoMedRec-D123

OntoMedRec: Logically-Pretrained Model-Agnostic Ontology Encoders for Medication Recommendation

TL;DR

OntoMedRec tackles data sparsity in medication recommendation by leveraging medical ontologies through Logic Tensor Networks to create logically-grounded, domain-agnostic ontology encoders. The three encoders (for diagnoses, procedures, and medications) are pre-trained with DAG-aware axioms and an axiom-oriented sampling strategy, then aligned with indication data from MEDI. The pre-trained embeddings are model-agnostic and can initialize both instance-based and longitudinal downstream models, improving performance especially in few-shot, sparse scenarios. This provides a scalable, interpretable pretraining paradigm for ontology-aware clinical decision support with demonstrated improvements on the MIMIC-III benchmark.

Abstract

Most existing medication recommendation models learn representations for medical concepts based on electronic health records (EHRs) and make recommendations with learnt representations. However, most medications appear in the dataset for limited times, resulting in insufficient learning of their representations. Medical ontologies are the hierarchical classification systems for medical terms where similar terms are in the same class on a certain level. In this paper, we propose OntoMedRec, the logically-pretrained and model-agnostic medical Ontology Encoders for Medication Recommendation that addresses data sparsity problem with medical ontologies. We conduct comprehensive experiments on benchmark datasets to evaluate the effectiveness of OntoMedRec, and the result shows the integration of OntoMedRec improves the performance of various models in both the entire EHR datasets and the admissions with few-shot medications. We provide the GitHub repository for the source code on https://anonymous.4open.science/r/OntoMedRec-D123
Paper Structure (31 sections, 2 equations, 3 figures, 2 tables)

This paper contains 31 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Instance-Based Medication Recommendation
  4. Longitudinal Medication Recommendation
  5. Existing Solutions to Data Sparsity Issue
  6. Preliminaries
  7. Electronic Medical Record
  8. Medical Concept Ontologies
  9. Medication Recommendation
  10. Logic Tensor Networks
  11. The Indication Relationship between Medications and Diagnoses
  12. The OntoMedRec Model
  13. Pre-training Ontology Encoders
  14. The Knowledge Formulation of Ontology Data
  15. The Alignment of Diagnosis and Medication Representations
  16. ...and 16 more sections

Figures (3)

  • Figure 1: Frequency distribution of diagnoses and medications in MIMIC-III dataset. The last bin is the cropped diagnoses/medications with a frequency higher than 200/40000.
  • Figure 2: An excerpt of the ATC ontology. Some nodes are omitted.
  • Figure 3: Performance of randomly initialised embedding table and OntoMedRec embedding table in different downstream models and tail percentages