We've updated our Privacy Policy to make it clearer how we use your personal data. We use cookies to provide you with a better experience. You can read our Cookie Policy here.


Novel Technique Analyzes 53 Million Points of Clinical Data

An image reflecting genomic data.
Credit: iStock.
Listen with
Register for free to listen to this article
Thank you. Listen to this article using the player above.

Want to listen to this article for FREE?

Complete the form below to unlock access to ALL audio articles.

Read time: 2 minutes

Researchers from Children’s Hospital of Philadelphia (CHOP) and Drexel University were able to analyze 53 million patient notes from more than 1.5 million individual patients to identify similarities in their medical histories that can help pinpoint potential risks for developing future diseases and the trajectory of those conditions.

This method of identifying phenotypic similarities exceeds the capacity any other current computational models. The findings were recently published in the journal Artificial Intelligence in Medicine.

While many technological advances have made it possible to analyze genetic data on a large scale, there are challenges with applying the same techniques to clinical data because it is often not standardized, meaning researchers may describe or record the same symptoms differently. The Human Phenotype Ontology (HPO) was established to serve as a dictionary of more than 15,000 clinical terms to help standardize how clinical information for patients can be analyzed, accelerating the integration of precision medicine into clinical practice.

Want more breaking news?

Subscribe to Technology Networks’ daily newsletter, delivering breaking science news straight to your inbox every day.

Subscribe for FREE

Prior studies from the Epilepsy Neurogenetics Initiative (ENGIN) at CHOP have analyzed clinical data from tens of thousands of patient notes that have helped reveal novel genetic targets in patients with different genetic childhood epilepsies. To expand on this success, researchers from CHOP’s Department of Biomedical Health Informatics (DBHi), Drexel’s College of Computing & Informatics, and ENGIN utilized Arcus, a suite of tools and services developed at CHOP that links biological, clinical, research and environmental data for the purpose of conducting innovative, data-driven research. The Arcus platform is designed to help reveal how sets of data overlap at a massive scale.

“This work follows trends in the artificial intelligence and machine learning field where complex, disparate information is transformed into something we can represent in a standardized way, thereby allowing machines to classify medical conditions and predict future disease risk,” said senior study author Scott Haag, PhD, Assistant Research Professor in the Department of Computer Science at Drexel’s College of Computing & Informatics and the Supervisor of the Arcus Data Science Team at CHOP. 

In this study, researchers analyzed data from 1,504,582 patients with a variety of diagnoses and syndromes with 53,955,360 electronic notes in the Arcus data repository. As a result, they identified 9,477 distinct phenotypes. This technique demonstrated a high degree of agreement with the judgment of experts in the various clinical fields represented in this data.

“This study demonstrated that utilizing an array made up of the clinical terms we identified could exceed the capacity of other much more computationally complex methods of analyzing phenotypes,” said first author Maryam Daniali, a PhD candidate in Computer Science at Drexel University. “This allowed us to map similarities between phenotypes using millions of points of data, significantly surpassing previous methods that relied on thousands of data points.”

Reference: Daniali M, Galer PD, Lewis-Smith D, et al. Enriching representation learning using 53 million patient notes through human phenotype ontology embedding. Artif Intell Med.2023;139:102523. doi: 10.1016/j.artmed.2023.102523.

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source.