Using foundation models to learn how to represent electronic health records

Placeholder Show Content


The invention and use of electronic health records has enabled new applications of machine learning in the field of healthcare. One particularly important application is the use of machine learning to predict the risk of a variety of clinical outcomes given electronic health records. However, training machine learning models on electronic health records is often challenging due to limited data set sizes and the intrinsic complexity of electronic health record data. In this dissertation, I propose and validate the use of foundation models as a way to learn representations for electronic health records, taking advantage of the longitudinal structure of medical record data to learn transferable representations. These representations can then be used to improve our ability to predict patient risk, achieving superior ranking performance, increased robustness across both time and space, and better label efficiency.


Type of resource text
Form electronic resource; remote; computer; online resource
Extent 1 online resource.
Place California
Place [Stanford, California]
Publisher [Stanford University]
Copyright date 2024; ©2024
Publication date 2024; 2024
Issuance monographic
Language English


Author Steinberg, Ethan Hannan
Degree supervisor Leskovec, Jurij
Degree supervisor Shah, Nigam
Thesis advisor Leskovec, Jurij
Thesis advisor Shah, Nigam
Thesis advisor Ré, Christopher
Degree committee member Ré, Christopher
Associated with Stanford University, School of Engineering
Associated with Stanford University, Computer Science Department


Genre Theses
Genre Text

Bibliographic information

Statement of responsibility Ethan Steinberg.
Note Submitted to the Computer Science Department.
Thesis Thesis Ph.D. Stanford University 2024.

Access conditions

© 2024 by Ethan Hannan Steinberg

Also listed in

Loading usage metrics...