Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Early Risk Prediction of Pediatric Cardiac Arrest from Electronic Health Records via Multimodal Fused Transformer

Jiaying Lu, Stephanie R. Brown, Songyuan Liu, Shifan Zhao, Kejun Dong, Del Bold, Michael Fundora, Alaa Aljiffry, Alex Fedorov, Jocelyn Grunwell, Xiao Hu

TL;DR

PedCA-FT presents a multimodal fused transformer framework that jointly leverages tabular and textualized EHR views to predict pediatric cardiac arrest within the first 24 hours of admission. By employing dedicated tabular and textual transformers followed by a fusion transformer and focal loss, the approach outperforms ten baselines across key metrics in a CHOA-CICU cohort with low CA incidence. The study highlights the value of modality-specific representations and late fusion for capturing complex interactions among high-dimensional risk factors, with feature-permutation analysis confirming clinically meaningful predictors such as CRT, bilirubin, and hemodynamic markers. These findings suggest substantial potential for timely CA detection and improved patient care in pediatric ICUs, with planned extensions to additional data sources and real-time monitoring.

Abstract

Early prediction of pediatric cardiac arrest (CA) is critical for timely intervention in high-risk intensive care settings. We introduce PedCA-FT, a novel transformer-based framework that fuses tabular view of EHR with the derived textual view of EHR to fully unleash the interactions of high-dimensional risk factors and their dynamics. By employing dedicated transformer modules for each modality view, PedCA-FT captures complex temporal and contextual patterns to produce robust CA risk estimates. Evaluated on a curated pediatric cohort from the CHOA-CICU database, our approach outperforms ten other artificial intelligence models across five key performance metrics and identifies clinically meaningful risk factors. These findings underscore the potential of multimodal fusion techniques to enhance early CA detection and improve patient care.

Early Risk Prediction of Pediatric Cardiac Arrest from Electronic Health Records via Multimodal Fused Transformer

TL;DR

PedCA-FT presents a multimodal fused transformer framework that jointly leverages tabular and textualized EHR views to predict pediatric cardiac arrest within the first 24 hours of admission. By employing dedicated tabular and textual transformers followed by a fusion transformer and focal loss, the approach outperforms ten baselines across key metrics in a CHOA-CICU cohort with low CA incidence. The study highlights the value of modality-specific representations and late fusion for capturing complex interactions among high-dimensional risk factors, with feature-permutation analysis confirming clinically meaningful predictors such as CRT, bilirubin, and hemodynamic markers. These findings suggest substantial potential for timely CA detection and improved patient care in pediatric ICUs, with planned extensions to additional data sources and real-time monitoring.

Abstract

Early prediction of pediatric cardiac arrest (CA) is critical for timely intervention in high-risk intensive care settings. We introduce PedCA-FT, a novel transformer-based framework that fuses tabular view of EHR with the derived textual view of EHR to fully unleash the interactions of high-dimensional risk factors and their dynamics. By employing dedicated transformer modules for each modality view, PedCA-FT captures complex temporal and contextual patterns to produce robust CA risk estimates. Evaluated on a curated pediatric cohort from the CHOA-CICU database, our approach outperforms ten other artificial intelligence models across five key performance metrics and identifies clinically meaningful risk factors. These findings underscore the potential of multimodal fusion techniques to enhance early CA detection and improve patient care.

Paper Structure

This paper contains 17 sections, 14 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Study Design
  3. Problem Definition
  4. Data Source and Cohort Selection
  5. Methods
  6. Tabular Transformer
  7. Textual Transformer
  8. Fusion Transformer
  9. Experimental Results
  10. Experimental Settings
  11. Compared Baseline Models
  12. Clinician Derived Risk Scores
  13. Last Observed Risk Factors
  14. Numerical Time-Series Risk Factors
  15. Experimental Results & Discussion
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Model Architecture of PedCA-FT.
  • Figure 2: Bar plots with error bars illustrating PPV, NPV, Sensitivity, and Specificity.
  • Figure 3: Feature importance analysis using feature permutation. Error bar indicates p95 high and p95 low.

Theorems & Definitions (2)

  • Definition 1: Patient EHR Data
  • Definition 2: Early Risk Prediction of Cardiac Arrest