InfEHR: Clinical phenotype resolution through deep geometric learning on electronic health records

Justin Kauffman, Emma Holmes, Akhil Vaid, Alexander W. Charney, Patricia Kovatch, Joshua Lampert, Ankit Sakhuja, Marinka Zitnik, Benjamin S. Glicksberg, Ira Hofer and Girish N. Nadkarni ()
Additional contact information
Justin Kauffman: Icahn School of Medicine at Mount Sinai
Emma Holmes: Icahn School of Medicine at Mount Sinai
Akhil Vaid: Icahn School of Medicine at Mount Sinai
Alexander W. Charney: Icahn School of Medicine at Mount Sinai
Patricia Kovatch: Icahn School of Medicine at Mount Sinai
Joshua Lampert: Icahn School of Medicine at Mount Sinai
Ankit Sakhuja: Icahn School of Medicine at Mount Sinai
Marinka Zitnik: Broad Institute of MIT and Harvard
Benjamin S. Glicksberg: Icahn School of Medicine at Mount Sinai
Ira Hofer: Icahn School of Medicine at Mount Sinai
Girish N. Nadkarni: Icahn School of Medicine at Mount Sinai

Nature Communications, 2025, vol. 16, issue 1, 1-21

Abstract: Abstract Electronic health records contain multimodal data that can inform clinical decisions but are often unsuited for advanced machine learning analyses due to lack of labeled data. Here, we present InfEHR, a framework to automatically compute clinical likelihoods from whole electronic health records without requiring large volumes of labeled training data. InfEHR applies deep geometric learning through a procedure that converts whole electronic health records to temporal graphs that naturally capture phenotypic dynamics, leading to unbiased representations. Using only few labeled examples, InfEHR computes and automatically revises probabilities achieving highly performant inferences, especially in low-prevalence diseases. We test InfEHR using electronic health records from Mount Sinai Health System and UC Irvine Medical Center against physician-provided heuristics on neonatal culture-negative sepsis (3% prevalence) and postoperative acute kidney injury (21% prevalence). InfEHR demonstrated superior performance: for culture-negative sepsis (sensitivity: 0.60 vs. 0.04, specificity: 0.98 vs. 0.99) and post-operative acute kidney injury (sensitivity: 0.71 vs. 0.20, specificity: 0.93 vs. 0.98). Our study demonstrates the application of geometric deep learning in electronic health records for probabilistic inference in real-world clinical settings at scale.

Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
https://www.nature.com/articles/s41467-025-63366-6 Abstract (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:nat:natcom:v:16:y:2025:i:1:d:10.1038_s41467-025-63366-6

Ordering information: This journal article can be ordered from
https://www.nature.com/ncomms/

DOI: 10.1038/s41467-025-63366-6

Access Statistics for this article

Nature Communications is currently edited by Nathalie Le Bot, Enda Bergin and Fiona Gillespie

More articles in Nature Communications from Nature
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().