Northwestern Medicine Builds Machine-Learning Model To Identify Patients With Heart Failure

Scientists and physicians create a custom tool within the Epic electronic health record to preemptively find patients with heart failure, expediting assessment and treatment.

One of the greatest strengths of healthcare software as ubiquitous as Epic’s electronic health record (EHR) is that it can amass so much data. In fact, the company estimates that hospitals using its software hold the medical records of more than half of all patients in the U.S. But with such a high volume of information held in one system, a common question facing today’s healthcare experts is how that data can be more efficiently leveraged to advance patient care.

For a team of scientists and physicians at Northwestern Medicine Bluhm Cardiovascular Institute, the answer to this question involved a form of artificial intelligence. In the fall of 2020, they began creating a custom machine-learning model within the Epic EHR that could sort through millions of patient records to make predictions about whether a patient could have heart failure, which currently impacts more than 6 million people in the United States.*

Prioritizing Patients

Built using Epic’s cloud-computing platform, the tool integrates three existing peer-reviewed models into an ensemble model, combining algorithms to analyze data. Working within an existing patient population and a clear set of biomarkers, such as lab results, diagnostics and medications, the tool preemptively identifies and classifies patients in different stages of heart failure. Manually sorting through such an enormous dataset to accomplish the same task would have been impossible for clinicians.

The model is being used in the Cardiology Department’s existing clinical workflow, requiring no additional training or resources. Batches of predictions are evaluated and processed into a list for nurse coordinators to review along with other elements from patient charts. Once the machine’s prediction about heart disease is validated, clinicians can reach out to patients so they can be seen and treated by specialty physicians much faster.

Enhancing Workflows

One of the most integral elements of this model is its built-in transparency, which allows Northwestern Medicine to use a computer-driven technology without overreach. The clinicians using the tool know what data elements are being pulled in to support the machine’s predictions, and the feedback loop incorporates human review, removing any potential for machine error while honing the model.

Supporting clinician intake, facilitating patient outreach and allowing for faster treatment are among the myriad benefits this machine-learning tool offers. In the future, it has the potential to identify patients at risk of developing more severe forms of heart failure, a huge step in advancing the care and treatment of patients.

* Centers for Disease Control and Prevention