Computer-assisted image processing 12 lead ECG model to diagnose hyperkalemia

J Electrocardiol. 2017 Jan-Feb;50(1):131-138. doi: 10.1016/j.jelectrocard.2016.09.001. Epub 2016 Sep 9.

Abstract

Background: We sought to develop an improved 12 lead ECG model to diagnose hyperkalemia by use of traditional and novel parameters.

Methods: We retrospectively analyzed ECGs in consecutive hyperkalemic patients (serum potassium (K)>5.3mEq/L) by blinded investigators with normokalemic ECGs as internal controls. Potassium levels were modeled using general linear mixed models followed by refit with standardized variables. Optimum sensitivity and specificity were determined using cut point analysis of ROC-AUC.

Results: The training set included 236 ECGs (84 patients) and validation set 97 ECGs (23 patients). Predicted K=(5.2354)+(0.03434*descending T slope)+(-0.2329*T width)+(-0.9652*reciprocal of new QRS width>100msec). ROC-AUC in the validation set was 0.78 (95% CI 0.69-0.88). Maximum specificity of the model was 84% for K>5.91 with sensitivity of 63%.

Conclusion: ECG model incorporating T-wave width, descending T-wave slope and new QRS prolongation improved hyperkalemia diagnosis over traditional ECG analysis.

Keywords: ECG; Hyperkalemia; QRS prolongation; T wave slope; T wave width.

MeSH terms

  • Aged
  • Algorithms*
  • Diagnosis, Computer-Assisted / methods*
  • Electrocardiography / methods*
  • Electroencephalography / methods*
  • Female
  • Humans
  • Hyperkalemia / blood*
  • Hyperkalemia / diagnosis*
  • Machine Learning
  • Male
  • Middle Aged
  • Pattern Recognition, Automated / methods
  • Potassium / blood*
  • Reproducibility of Results
  • Sensitivity and Specificity

Substances

  • Potassium