Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluating the Predictive Features of Person-Centric Knowledge Graph Embeddings: Unfolding Ablation Studies

Christos Theodoropoulos, Natasha Mulligan, Joao Bettencourt-Silva

TL;DR

This paper addresses ICU readmission prediction by leveraging person-centric knowledge graphs (PKGs) that integrate demographic, clinical, and social-context information within a graph neural network framework. It introduces a repeatable, ablation-driven methodology to identify which features across demographic, clinical, and social facets most strongly influence predictive performance, using the MIMIC-III dataset and the Health & Social Patient-centric Ontology (HSPO). Key findings show that while clinical factors (diseases and medications) are strong predictors, social determinants (notably household context) and certain demographics (marital status, race) substantially shape model performance, underscoring the importance of comprehensive data collection and holistic modeling. The proposed framework offers a robust, interpretable approach to feature discovery in PKG-based predictions and is adaptable to other datasets and healthcare tasks, with future work pointing to broader scope and integration of generative methods for improved interpretability and discovery.

Abstract

Developing novel predictive models with complex biomedical information is challenging due to various idiosyncrasies related to heterogeneity, standardization or sparseness of the data. We previously introduced a person-centric ontology to organize information about individual patients, and a representation learning framework to extract person-centric knowledge graphs (PKGs) and to train Graph Neural Networks (GNNs). In this paper, we propose a systematic approach to examine the results of GNN models trained with both structured and unstructured information from the MIMIC-III dataset. Through ablation studies on different clinical, demographic, and social data, we show the robustness of this approach in identifying predictive features in PKGs for the task of readmission prediction.

Evaluating the Predictive Features of Person-Centric Knowledge Graph Embeddings: Unfolding Ablation Studies

TL;DR

This paper addresses ICU readmission prediction by leveraging person-centric knowledge graphs (PKGs) that integrate demographic, clinical, and social-context information within a graph neural network framework. It introduces a repeatable, ablation-driven methodology to identify which features across demographic, clinical, and social facets most strongly influence predictive performance, using the MIMIC-III dataset and the Health & Social Patient-centric Ontology (HSPO). Key findings show that while clinical factors (diseases and medications) are strong predictors, social determinants (notably household context) and certain demographics (marital status, race) substantially shape model performance, underscoring the importance of comprehensive data collection and holistic modeling. The proposed framework offers a robust, interpretable approach to feature discovery in PKG-based predictions and is adaptable to other datasets and healthcare tasks, with future work pointing to broader scope and integration of generative methods for improved interpretability and discovery.

Abstract

Developing novel predictive models with complex biomedical information is challenging due to various idiosyncrasies related to heterogeneity, standardization or sparseness of the data. We previously introduced a person-centric ontology to organize information about individual patients, and a representation learning framework to extract person-centric knowledge graphs (PKGs) and to train Graph Neural Networks (GNNs). In this paper, we propose a systematic approach to examine the results of GNN models trained with both structured and unstructured information from the MIMIC-III dataset. Through ablation studies on different clinical, demographic, and social data, we show the robustness of this approach in identifying predictive features in PKGs for the task of readmission prediction.
Paper Structure (6 sections, 2 figures)

This paper contains 6 sections, 2 figures.

Table of Contents

  1. Introduction and Background
  2. Methods
  3. Pre-processing and Graph Neural Network Model Training
  4. Ablation Study
  5. Results and Discussion
  6. Conclusions

Figures (2)

  • Figure 1: Overall approach (A), Knowledge Graphs and GNN (B).
  • Figure 2: Ablation study results (A, B).