From Basic to Extra Features: Hypergraph Transformer Pretrain-then-Finetuning for Balanced Clinical Predictions on EHR

TL;DR

This paper introduces HTP-Star, a hypergraph transformer model with a pretrain-then-finetune approach, designed to improve clinical predictions from EHR data by effectively integrating basic and extra patient features.

Contribution

The paper presents a novel hypergraph transformer framework with regularization techniques for balanced and robust EHR-based clinical predictions.

Findings

HTP-Star outperforms baseline models on real EHR datasets.

The model effectively balances predictions for patients with varying feature availability.

Regularization enhances robustness during fine-tuning.

Abstract

Electronic Health Records (EHRs) contain rich patient information and are crucial for clinical research and practice. In recent years, deep learning models have been applied to EHRs, but they often rely on massive features, which may not be readily available for all patients. We propose HTP-Star, which leverages hypergraph structures with a pretrain-then-finetune framework for modeling EHR data, enabling seamless integration of additional features. Additionally, we design two techniques, namely (1) Smoothness-inducing Regularization and (2) Group-balanced Reweighting, to enhance the model's robustness during fine-tuning. Through experiments conducted on two real EHR datasets, we demonstrate that HTP-Star consistently outperforms various baselines while striking a balance between patients with basic and extra features.