Analysis of pediatric absence epilepsy electroencephalograms using a mining-based association rule approach: A cross-sectional study based on realistic data.

Association rule mining (ARM) is an interpretable data mining method that can identify co-occurrence patterns among predefined clinical features. This study applied ARM to clinically labeled EEG data from children with absence epilepsy to identify association rules linking abnormal EEG discharge morphology, scalp distribution, and absence seizure duration. Key association rules included polyspike-and-wave discharge → spike-and-wave discharge (support = 0.517, confidence = 1.000, lift = 1.034), spikes → spike-and-wave discharge (support = 0.510, confidence = 0.975, lift = 1.012), mesial temporal involvement → central involvement (support = 0.601, confidence = 0.968, lift = 1.528), and frontal pole involvement → frontal involvement (support = 0.629, confidence = 0.741, lift = 1.536). Frontal involvement combined with the 10-20-seconds seizure-duration category also showed high co-occurrence with frontal pole involvement ({F, 15.0 seconds} → {Fp}: support = 0.289, confidence = 1.000, lift = 1.561). These findings suggest that ARM can provide a structured and clinically readable summary of EEG co-occurrence patterns in pediatric absence epilepsy. However, the study was limited by its relatively small sample size, single-center design, expert manual labeling, binarized EEG representation, and lack of independent external validation. Therefore, the findings should be interpreted as exploratory and hypothesis-generating rather than externally validated clinical rules. Future multicenter studies with standardized EEG labeling, additional clinical variables, comparison with alternative analytical approaches, and independent validation datasets are needed to confirm the reproducibility and clinical applicability of these association rules.