Electronic Health Records

Use & Optimization News

EHR Data, Machine Learning Predict Risk of Cardiovascular Disease

Researchers developed a health management tool that leverages EHR data and machine learning to test for cardiovascular disease.

Integrating EHR data into machine learning algorithms can help to predict risk of cardiovascular disease.

Source: Thinkstock

By Kate Monica

- A team of researchers from Stanford University created a personal health management tool that combines EHR data with machine learning to accurately diagnose patients with abdominal aortic aneurysm, also called AAA.

A new report recently published in Cell describes how researchers integrated genome and EHR data into a new machine learning framework to predict patient diagnoses of the heart condition. 

The cardiovascular disease is asymptomatic as it grows. As a result, healthcare providers often diagnose the condition at a late stage. By analyzing genome and EHR data using machine learning, researchers attempted to diagnose the condition earlier for more timely treatment.

Researchers also performed whole genome sequencing on patients with the form of aneurysm and modeled personal genomes with EHR data to assess the effectiveness of adjusting personal lifestyles to manage patient health. 

The study serves as a proof-of-principle for using available clinical information to diagnose the heart condition earlier and promote healthy lifestyle changes for improved health outcomes.

READ MORE: Breaking Down How the Apple Health Records EHR Data Viewer Works

Researchers used the hierarchical estimate from agnostic learning (HEAL) health management tool to test 474 patients from VA Palo Alto Healthcare System, Stanford University, and Kaiser Permanent.

“Integration of participant EHR data was derived from the clinical records as well as patient interviews, including individuals’ various physiological measurements and lifestyles upon their last clinical visit before the initiation of the project,” explained researchers in the report.

HEAL identified AAA-associated genetic components in patients using an algorithm that examines genomes for single nucleotide variants (SNVs) and assesses the potential clinical relevance of each individual mutation. The tool’s machine learning framework agnostically identified a subset of genes showing distinct mutational patterns in cases compared to controls and used this pattern to predict health outcomes.

HEAL also mapped identified genes onto biological networks to offer a complete picture of disease-associated pathways.

Researchers combined genome analysis with assessments of patient EHRs, personal lifestyle surveys, and visit summaries about patient’s physiological measurements from their most recent doctors’ visits.

READ MORE: EHR Data Helps to Predict Short-Term Mortality for Cancer Patients

Ultimately, the tool analyzed genome sequences and patient EHRs to accurately identify 313 patients with abdominal aortic aneurysm and 161 patients without the cardiovascular disease.

“For each individual, HEAL accurately predicted his/her AAA risk using personal genome and EHR data,” stated researchers. “On the other hand, for the same individual with newly adopted lifestyles resulting in physiological changes (e.g., from a high cholesterol to a low cholesterol diet), HEAL can immediately update his/her AAA risk upon corresponding changes conditioned on the person’s genome baseline.”

“This allows us to further investigate the interplay between personal genomes and lifestyles underlying disease predisposition,” researchers continued.

The tool may be helpful to improve diagnosis, treatment, and prevention efforts for this heart condition. The tool’s integrative model is capable of identifying genome baselines for certain diseases and offering patients and providers actionable guidelines for disease management.

“For example, we know that smoking has a tremendous influence on AAA development,” Stanford University School of Medicine Professor of Medicine Philip S. Tsao, PhD, said in a press release. “If you knew you had a genetic predilection for AAA, you would strongly be advised not to pick up a smoking habit.”

READ MORE: FDA Provides Guidance for EHR Data Use in Clinical Investigations

HEAL could also be used to detect neurological diseases including autism and schizophrenia, Tsao said.

“Our study presents a new framework for disease genome analysis,” maintained Tsao.

Furthermore, HEAL could give providers a risk score for patients who may have the potential to develop certain diseases and deliver relevant recommendations about treatment and lifestyle.

“It is now possible to determine a person's risk for AAA right from their genome sequence,” said Stanford University School of Medicine Director of Genomics and Personalized Medicine Michael Snyder, PhD. “This is important because the disease is irreversible, and most people discover they have AAA when their aorta bursts, which is usually lethal.”

Grants from the National Institute of Health (NIH), the University of California, and the VA Office of Research and Development helped to support research about HEAL’s effectiveness.

While genome profiles can help to predict and manage diseases earlier for more effective treatment, Snyder emphasized that patients own their genome data and are entitled to decide whether researchers, providers, or other individuals are allowed to access and analyze their information.

“The genome data belongs to the person and it is up to them to decide how to share it,” concluded Snyder. “If placed in a medical database, it should be secure.”

Stanford University also participated in a study earlier this year about the effectiveness of using de-identified EHR data for more accurate predictive analytics.

Stanford researchers joined others from Google, University of California San Francisco, and University of Chicago Medicine and found representations of a comprehensive patient EHR using Fast Healthcare Interoperability Resources (FHIR) can contribute to more accurate predictive analytics.

Researchers found deep learning — a type of machine learning that mimics the decision-making structure of the human brain — can produce valid predictions across a variety of clinical problems and health outcomes ranging from mortality rates to readmissions.

“A deep learning approach that incorporated the entire electronic health record, including free-text notes, produced predictions for a wide range of clinical problems and outcomes that outperformed state-of-the-art traditional predictive models,” wrote researchers.

Combining EHR data with machine learning holds promise as a way to anticipate and prevent health problems and improve health outcomes.



Sign up to continue reading and gain Free Access to all our resources.

Sign up for our free newsletter and join 60,000 of your peers to stay up to date with tips and advice on:

EHR Optimization
EHR Interoperability

White Papers, Webcasts, Featured Articles and Exclusive Interviews

Our privacy policy

no, thanks

Continue to site...