Deep Learning for EHR-Based Diagnosis Prediction: A General Recipe Open Access

Yu, Leisheng (Spring 2022)

Permanent URL:


With the rapid accumulation of Electronic Health Records (EHRs) and the recent advance in data-driven algorithms, deep learning models have been increasingly applied to tasks in EHR-based predictive healthcare. This paper, motivated by the hierarchical structure of EHR data and the identified challenges in predictive healthcare, mathematically formulates a general architecture for learning patient representations for diagnosis prediction. With the guidance of the proposed general architecture, this work further discusses how existing works have incorporated various model designs to overcome certain challenges from four levels: diagnosis, visit, sequence, and framework-level. Through these discussions, this paper serves as a summary of existing works in modeling sequential EHR data, a cookbook for novices interested in EHR-based predictive healthcare, and a foundation for a future work, the idea of which are also introduced in this paper.

Table of Contents

Introduction 1

General Recipe 6

Diagnosis-Level Representation Learning 9

Visit-Level Representation Learning 13

Sequence-Level Representation Learning 16

Framework-Level Design 20

Conclusion 22

Future Work 23

About this Honors Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
  • English
Research Field
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files