Researchers use deep learning, ECGs to detect hyperkalemia

A deep learning model trained on more than 1.5 million electrocardiograms can reliably detect hyperkalemia—or abnormally high potassium levels in the blood—among patients with chronic kidney disease (CKD), Mayo Clinic researchers reported April 3 in JAMA Cardiology.

The training set was derived from ECGs of nearly 450,000 patients seen at the Mayo Clinic in Rochester, Minnesota, from 1994 through 2017.

When tested on patients with stage 3 or worse CKD at three Mayo Clinic sites from other states, the model achieved an area under the curve (AUC) ranging from 0.853 to 0.883 for detecting hyperkalemia, defined as a serum potassium level above 5.5 mmol/L. Sensitivities spanned from 88.9% to 91.3%, while specificities ranged from 55% to 63.2%.

Senior author Paul A. Friedman, MD, and colleagues noted this accuracy far outpaced what physicians achieved with ECG analysis in previous reports, when they demonstrated a sensitivity of just 34% to 43%.

“The model was robust across diverse patients, geography and year,” the researchers wrote. “At a high-sensitivity operating point, the deep learning model performed well as a potential screening tool to rule out hyperkalemia, with a negative predictive value greater than 99%.”

Finding a better, noninvasive way to screen patients for this condition is a worthy goal, according to Friedman et al., because hyperkalemia is often asymptomatic but is associated with cardiac arrhythmias and death.

Read more from AI in Healthcare below: