Discovery of clinical disease patterns from patient data

Paper Title:

Unsupervised Discovery of Clinical Disease Signatures Using Probabilistic Independence

Published on:

8 February 2024

Primary Category:

Machine Learning

Paper Authors:

Thomas A. Lasko,

John M. Still,

Thomas Z. Li,

Marco Barbero Mota,

William W. Stead,

Eric V. Strobl,

Bennett A. Landman,

Fabien Maldonado

Bullets

Key Details

•

Presents unsupervised method to discover precise disease patterns in EHR data

•

Uses probabilistic independence to separate multiple overlapping disease signals

•

Applied to 269K records and 9.2K variables, infers 2K interpretable disease patterns

•

Patterns include recognizable pictures of rare conditions

•

Patterns gave better lung cancer prediction and explanation than original variables

Explore the topics in this paper

clinical prediction

data mining

disease pattern identification

electronic health records

unsupervised learning

AI generated summary

Discovery of clinical disease patterns from patient data

This paper presents a method to identify precise patterns of clinical disease from electronic health records, without supervision. It uses the principle of probabilistic independence to separate multiple overlapping disease signals present in patient data. Applying the method to 269,099 records with 9,195 variables, it produced 2,000 interpretable patterns of disease, including recognizable rare conditions. The patterns gave better prediction and explanation of lung cancer status for 13,252 patients than using original variables.