By Dr. Walter Stewart
Heart failure remains nearly impossible to detect early.
Although doctors look for physical signs and symptoms, which are commonly known as the Framingham criteria, they can occur with illnesses other than heart failure. So, doctors usually diagnose heart failure after a patient is hospitalized, when the disease has progressed to a very serious stage and caused irreversible organ damage.
Sutter Health, IBM and Geisinger Health Systems have earned a $2 million grant from the National Institute of Health (NIH) to improve this diagnosis – to make it faster, more accurate, and more reliable with analytics. As part of this three year project, we will collect data on heart failure symptoms, test multiple approaches for quickly and accurately analyzing the data, and determine how we might structure a potential clinical trial.
The state of diagnosis today
Although recent research finds that patients hospitalized for acute heart failure are likely to survive longer compared to the prior decade, long-term survival rates still remain low. According to the American Heart Association, “only one in three patients hospitalized with acute decompensated heart failure (ADHF) in 2004 survived beyond five years.” In addition, patients with heart failure may experience reduced quality of life and decreased ability to function.
It’s almost impossible to conduct meaningful analysis of data collected in paper medical records. On the other hand, electronic health records (EHR) provide a rich, expansive view of a patient’s health history. Sophisticated analysis of existing EHR data could reveal important signals to help doctors recognize a patient’s symptom profile as indicative of an underlying problem with heart failure – characteristics distinct from other diseases – other than by alternative health problems.
Once patients are identified as high-risk for heart failure, physicians can better monitor their status, help motivate a patient to make potentially life-saving lifestyle changes and test clinical interventions to potentially slow or possibly reverse heart failure progression.
Improving diagnosis with EHR data
We’ve worked with IBM and Geisinger on how to use EHR data since 2009, publishing several, such as:
- Prediction modeling using EHR data: challenges, strategies, and a comparison of machine learning approaches
- Automatic identification of heart failure diagnostic criteria, using text analysis of clinical notes from electronic health records
- Combining knowledge and data driven insights for identifying risk factors using electronic health records
With this NIH grant, we expect to create a deeper understanding of how to use EHR data and advanced analytics to help detect heart failure up to three years sooner than is now possible. We’re also working to identify best practices that will help health systems nationwide integrate data analytics into primary care practices.
If we can help doctors predict heart failure years earlier, they can better partner with patients on a “smarter care” approach to treatment. Working together, patients and care teams can identify tailored treatment options and holistic approaches to disease management that are personalized for each individual.