EHR data can identify patients at risk of heart failure

Photo credit: Getty/pandpstock001

Algorithms based on electronic health record data can help flag hospitalized patients at risk of heart failure, with the most complicated algorithms yielding the most accurate results, according to a recent study in JAMA Cardiology.

Real-time, accurate identification of patients with heart failure can catch the problem earlier and results in better outcomes. However, heart failure is a condition not easily identified. The researchers, from the New York University School of Medicine and elsewhere, hypothesized that data analysis using algorithms may help.

The authors developed five algorithms of increasing complexity to determine how well they would work to identify acute and chronic heart failure. They conducted a retrospective study of 47,119 hospitalizations at New York University Langone Medical Center using EHR data for adult patients admitted from Jan. 1, 2013 through Feb. 28, 2015. The potential structured data elements used for heart failure classification were demographics, lab results, vital signs, problem list diagnoses and medications used in the treatment of heart failure. 

The first algorithm used only the problem list. In contrast, the third algorithm used logistic regression and 30 structured data elements. The fourth algorithm used a machine-learning approach on free text and included 1,118 elements in the model. The fifth algorithm used a machine-learning approach to identify 947 structured and unstructured data elements.

The problem list algorithm identified only half of hospitalized patients with heart failure. The other algorithms boasted improved accuracy for the identification of heart failure. However, while the fourth and fifth algorithms had the best predictive accuracy because they relied on free text notes and reports, they would be more difficult to implement because they relied on unstructured data. The authors suggested that the best approach may depend on a provider’s clinical and operational needs.

“The implementation of complex algorithms into an EHR for real-time identification of heart failure may require special expertise and resources," the researchers said. "There may be a tradeoff of cost of implementation and benefit of improvement."