Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-Explaining Hypergraph Neural Networks for Diagnosis Prediction

Leisheng Yu, Yanxiao Cai, Minxing Zhang, Xia Hu

TL;DR

SHy introduces a self-explaining diagnosis predictor that models each patient as a unique hypergraph to capture higher-order disease interactions. It learns personalized hierarchical disease embeddings, augments histories with plausible false negatives, and extracts $K$ temporal phenotypes via Gumbel-Softmax-based masking, all feeding a GRU-based predictor with a multi-objective loss that enforces fidelity, distinctness, and interpretable attention. The approach yields competitive predictive performance while providing concise, faithful, and editable explanations that clinicians can intervene on, addressing robustness to false negatives in real-world EHR data. Overall, SHy demonstrates a practical, interpretable framework for diagnosis prediction with strong potential for clinical adoption and interactive model refinement.

Abstract

The burgeoning volume of electronic health records (EHRs) has enabled deep learning models to excel in predictive healthcare. However, for high-stakes applications such as diagnosis prediction, model interpretability remains paramount. Existing deep learning diagnosis prediction models with intrinsic interpretability often assign attention weights to every past diagnosis or hospital visit, providing explanations lacking flexibility and succinctness. In this paper, we introduce SHy, a self-explaining hypergraph neural network model, designed to offer personalized, concise and faithful explanations that allow for interventions from clinical experts. By modeling each patient as a unique hypergraph and employing a message-passing mechanism, SHy captures higher-order disease interactions and extracts distinct temporal phenotypes as personalized explanations. It also addresses the incompleteness of the EHR data by accounting for essential false negatives in the original diagnosis record. A qualitative case study and extensive quantitative evaluations on two real-world EHR datasets demonstrate the superior predictive performance and interpretability of SHy over existing state-of-the-art models.

Self-Explaining Hypergraph Neural Networks for Diagnosis Prediction

TL;DR

SHy introduces a self-explaining diagnosis predictor that models each patient as a unique hypergraph to capture higher-order disease interactions. It learns personalized hierarchical disease embeddings, augments histories with plausible false negatives, and extracts temporal phenotypes via Gumbel-Softmax-based masking, all feeding a GRU-based predictor with a multi-objective loss that enforces fidelity, distinctness, and interpretable attention. The approach yields competitive predictive performance while providing concise, faithful, and editable explanations that clinicians can intervene on, addressing robustness to false negatives in real-world EHR data. Overall, SHy demonstrates a practical, interpretable framework for diagnosis prediction with strong potential for clinical adoption and interactive model refinement.

Abstract

The burgeoning volume of electronic health records (EHRs) has enabled deep learning models to excel in predictive healthcare. However, for high-stakes applications such as diagnosis prediction, model interpretability remains paramount. Existing deep learning diagnosis prediction models with intrinsic interpretability often assign attention weights to every past diagnosis or hospital visit, providing explanations lacking flexibility and succinctness. In this paper, we introduce SHy, a self-explaining hypergraph neural network model, designed to offer personalized, concise and faithful explanations that allow for interventions from clinical experts. By modeling each patient as a unique hypergraph and employing a message-passing mechanism, SHy captures higher-order disease interactions and extracts distinct temporal phenotypes as personalized explanations. It also addresses the incompleteness of the EHR data by accounting for essential false negatives in the original diagnosis record. A qualitative case study and extensive quantitative evaluations on two real-world EHR datasets demonstrate the superior predictive performance and interpretability of SHy over existing state-of-the-art models.

Paper Structure

This paper contains 27 sections, 18 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Learning on Longitudinal EHRs
  4. Intrinsically Interpretable Models
  5. Methodology
  6. Problem Formulation
  7. Personalized Disease Representation Learning
  8. Learning Hierarchical Disease Embeddings
  9. Constructing Patient Hypergraphs
  10. Modeling Individual Comorbidities
  11. Temporal Phenotype Extraction & Modeling
  12. Prediction and Objectives
  13. Experiments
  14. Data Description & Experimental Setup
  15. Predictive Performance Comparison
  16. ...and 12 more sections

Figures (3)

  • Figure 1: An illustration of diagnosis prediction using the longitudinal EHR of a patient. Diagnoses are denoted by ICD-9 codes.
  • Figure 2: Overview of SHy.
  • Figure 3: An illustration of how SHy extracts five temporal phenotypes from the EHR of a 53-year-old female patient and how a clinician can refine the prediction by directly adjusting the generated phenotypes. Underlined text highlights human modifications, while check marks indicate correct predictions. The red numbers indicate the importance weights.