Prediction of Hyperkalemia in Dogs from Electrocardiographic Parameters Using an Artificial Neural Network

Abstract

Abstract. Objective: To predict severe hyperkalemia from single electrocardiogram (ECG) tracings. Methods: Ten conditioned dogs each underwent this protocol three times: Under isoflurane anesthesia, 2 mEq/kg/hr of potassium chloride was given intravenously until P-waves were absent from the ECG and ventricular rates decreased Ն20% in Յ5 minutes. Serum potassium levels (K ϩ ) were measured at regular intervals with concurrent digital storage of lead II of the surface ECG. A three-layer artificial neural network with four hidden nodes was trained to predict K ϩ from 15 separate elements of corresponding ECG data. Data were divided into a training set and a test set. Sensitivity, specificity, and diagnostic accuracy for recognizing hyperkalemia were calculated for the test set based on a prospectively defined K ϩ = 7.5. Results: The model produced data for 189 events; 139 were placed in the training set and 50 in the test set. The test set had 37 potassium levels at or above 7.5 mmol/L. The neural network had a sensitivity of 89% (95% CI = 75% to 97%) and a specificity of 77% (95% CI = 46% to 95%) in recognizing these. The positive likelihood ratio was 3.87. Overall accuracy of this model was 86% (95% CI = 73% to 94%). Mean (ϮSD) difference between predicted and actual K ϩ values was 0.4 Ϯ 2.0 (95% CI = Ϫ0.2 to 1.0). Conclusions: An artificial neural network can accurately diagnose experimental hyperkalemia using ECG parameters.Further work could potentially demonstrate its usefulness in bedside diagnosis of human subjects. Key words: hyperkalemia; neural network; diagnosis; ECG. ACADEMIC EMERGENCY MEDICINE 2001; 8:599-603 H YPERKALEMIA is commonly seen in emergency department (ED) patients and may be associated with lethal dysrhythmias. While hyperkalemia is easily diagnosed by laboratory testing, the result may be unavailable for 45 to 120 minutes or more, unless the ED has the capability of bedside measurement of electrolytes, a practice that is still not widespread. During this time, the undiagnosed hyperkalemic patient may decompensate suddenly and without warning. Unanticipated decompensation of the clinically occult hyperka- Because of its key role in cardiac electrophysiology, hyperkalemia often results in surface electrocardiogram (ECG) changes such as peaked Twaves, low-amplitude P-waves, and increases in PR or QRS intervals. Since the ECG may be obtained at the bedside much more quickly than the results of blood testing, ECG findings have been used for rapid bedside diagnosis of hyperkalemia, 1 but there are limitations. Since hyperkalemic ECG abnormalities are nonspecific, diagnostic accuracy may be decreased by many false-positive results. Additionally, sensitivity acceptable for screening for a life-threatening condition has not been demonstrated. Wrenn et al.2 tested two experienced ED clinicians in evaluating 220 ECGs for the diagnosis of hyperkalemia. In this group there were 87 subjects (40%) with hyperkalemia, defined as K ϩ > 5.0 mmol/L. Twenty-nine of those 87 subjects had K ϩ > 6.5 mmol/L....

Retrospective Review of the Frequency of ECG Changes in Hyperkalemia

“…Given the lack of validation of this method, it is unclear whether the lack of documented correlation with serum potassium concentration reflects the inadequacy of the measurement or a physiologic difference in patients with ESRD. Porter et al (18) used a combination of electrocardiographic markers incorporated into a neural network algorithm to diagnose hyperkalemia in dogs with high sensitivity (89%) and specificity (77%). Although compelling, these results have not been reproduced in human clinical settings, and the use of a neural network makes their application to patient care logistically more complicated.…”

Predicting hyperkalemia by a two-staged artificial neural network

“…However, the accuracy is very low (about 50%) even by experienced cardiologist [1,2]. Because the ECG features in response to hyperkalemia are not obvious.…”

“…Based on clinical reports [1,2], hyperkalemia was associated with the change of ECG by some characteristics. For examples, peaked T waves and widen T waves could be observed in some chest leads during mild hyperkalemia, and flatten P waves in some limb and chest leads.…”

“…The applications of neural networks on the detection of cardiovascular diseases had big progress such as myocardial infarctions and atrial fibrillation [5]. Recent Poter etc [1] used successfully traditional back-propagation neural networks to detect severe hyperkalemia with a sensitivity 89% by single lead II ECG signals. However, the hyperkalemic data were collected by controlled conditions.…”

Predicting hyperkalemia by the use of a 12-lead temporal-spatial electrocardiograph: clinical evaluations and model simulations

“…The peaked T waves could be observed in some chest leads or limb lead II in early hyperkalemia, and flatten P waves in some limb and chest leads. Other variables such as the increased PR interval, the QRS interval, and the RR interval could be seen in mild to severe hyperkalemia [5]. Nevertheless, the choice of the exact features in response to hyperkalemia is difficult to make because of the variability of the features.…”