Browse Prior Art Database

A Method for Extracting and Classifying Phenotypes from EHRs

IP.com Disclosure Number: IPCOM000247601D
Publication Date: 2016-Sep-19
Document File: 2 page(s) / 141K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a computational method for extracting clinically useful phenotypes from sparse, noisy, and high dimensional Electronic Health Records (EHRs), for classifying EHRs into cases/controls and an EHR event network, and capturing correlations among observed events.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 2

A Method for Extracting and Classifying Phenotypes from EHRs

Electronic Health Records (EHRs) contain diverse information about patients. Electronic phenotyping aims to extract concise, meaningful medical concepts from EHRs. EHRs are typically high dimensional, heterogeneous, sparse, recorded at irregular time stamps, and are often incomplete and biased.

EHRs present multiple problems. One problem is the occurrence of time-irregularities.

A record occurs when a patient enters the healthcare system. Most existing electronic

phenotyping methods group observations within a time window. Valuable time information could be ignored. Another problem is data noise and missing values in EHRs. Most existing methods use a deterministic framework ignoring biased noisy records. Additionally, a method is needed to scale a large corpus of data. As the volume of EHRs increases, computational methods capable of scaling to big data are required.

The novel solution is a computational method for extracting clinically useful phenotypes from sparse, noisy, and high dimensional EHRs, for classifying EHRs into cases/controls and an EHR event network, and capturing correlations among observed events.

The method comprises of the following:


• A framework for posing electronic phenotyping as a probabilistic inference problem

• Method for modeling missing data and noise in electronic health records • Method for dealing with temporal irregularity in EHRs • Method for incorporating prior knowledge about both disease progression and its clinical manifestations • Method for learning compact lower dimensional representations from high dimensional EHRs
• Algorithms for classifying EHRs and simulating new EHRs

• Outputs electronic phenotypes • Outputs an EHR event network, capturing correlations among observed eve...