ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: FR-PO543

Development and Validation of Deep Learning Algorithm for Detecting Hyperkalemia Based on Electrocardiogram

Session Information

Category: Fluid‚ Electrolyte‚ and Acid-Base Disorders

  • 1002 Fluid‚ Electrolyte‚ and Acid-Base Disorders: Clinical

Authors

  • Lee, Dong Hee, Hallym University Sacred Heart Hospital, Anyang, Gyeonggi-do, Korea (the Republic of)
  • An, Jung Nam, Hallym University Sacred Heart Hospital, Anyang, Gyeonggi-do, Korea (the Republic of)
  • Kim, Sungmin, Hallym University Sacred Heart Hospital, Anyang, Gyeonggi-do, Korea (the Republic of)
  • Kim, Jwa-kyung, Hallym University Sacred Heart Hospital, Anyang, Gyeonggi-do, Korea (the Republic of)
  • Kim, Sung Gyun, Hallym University Sacred Heart Hospital, Anyang, Gyeonggi-do, Korea (the Republic of)
Background

Hyperkalemia is a common electrolyte abnormality, which in severe cases can cause fatal arrhythmia and cardiac arrest. However, it is often asymptomatic, and diagnosis is difficult without blood tests. The purpose of this study is to detect hyperkalemia quickly and easily through a deep learning-based model that has learned an electrocardiogram (ECG) that is non-invasive and can be quickly measured.

Methods

Among patients who underwent an ECG at least once from 2006 to 2020, patients with blood test results within 24 hours were included, and dialysis patients were excluded. All ECGs were acquired using a GE ECG machine and the raw data (XML datatype) were stored using the MUSE data management system. For model training and evaluation, the ECG-serum K pair was separated into train, validation, and test set. Hyperkalemia and severe hyperkalemia were defined as 5.5 mEq/L ≤ K+ and as 6.5 mEq/L ≤ K+. We trained a 4-class classification model using a Convolutional Neural Network. The model input was a standard 10-second, 12-lead ECG and the output being the likelihood of the ECG being from a patient with classes of serum potassium concentration.

Results

A total of 362,896 cases of ECG-serum K were analyzed in a total of 299431 patients, of which 330,137 cases were in the train set, 15,250 cases in the validation set, and 28,783 cases in the test set. In the validation set, AUROC for hyperkalemia was 0.97 and AUPRC was 0.79, and AUROC for severe hyperkalemia was 0.99 and AUPRC was 0.86. In the test set, for hyperkalemia, the sensitivity and specificity of deep learning model were 81.1% and 80.9%.

Conclusion

The deep learning model using the 12-lead ECG waveform detected hyperkalemia with high accuracy, and in severe hyperkalemia, the diagnostic predictive power was further increased. These results suggest the clinical applicability of AI software for diagnosing hyperkalemia using ECG.