Seventy patients with syncope and 105 patients with epilepsy were included in the analysis. The duration of loss of consciousness (LOC), neuron-specific enolase (NSE), prolactin, NH3, CPK, and myoglobin level were measured within 6 hours after the symptom onset. Univariate analysis was carried out to identify the value of each demographic and the measured parameter for differentiation of epileptic seizure from syncope using a logistic regression. Multivariate analyses controlling for age and sex were conducted to identify the possible independent biomarkers. Machine learning algorithms, including support vector machine, decision tree, K-nearest neighbor, and random forest classifiers were applied to develop an algorithm for differential diagnosis.

Compared to syncope patients, epilepsy patients exhibited longer LOC duration, increased NSE, prolactin, NH3, and myoglobin levels. The level of NH3 at cut-off value of 45.5 µmol/L was found to be the most powerful variable to discriminate the conditions with 71.23% specificity and 68.8% sensitivity. Using the LOC duration, NH3, and myoglobin levels as input features, support vector machine algorithm could classify epileptic seizure and syncope with 85.1% accuracy.

The LOC duration, levels of NSE, prolactin, NH3, and myoglobin could be independent biomarkers for differential diagnosis between epileptic seizure and syncope. Moreover, differential diagnosis algorithm made by combining potential biomarkers might be useful as a screening tool.