BACKGROUND: Paroxysmal slow wave events (PSWEs), defined as electroencephalography (EEG) segments where the median power frequency falls below 6 Hz for ≥5 s, have been shown to predict epilepsy in patients with a first seizure. We evaluated the prevalence and localisation of PSWEs in large independent EEG datasets, exploring their potential as biomarkers for drug-resistant epilepsy (DRE). METHODS: An exploratory analysis used 1064 participants from the Temple University EEG corpus, comparing patients with epilepsy (N = 903) with participants with seizure-mimics and normal EEGs (N = 161). Validation analysis used an independent cohort from Bonn University, comprising drug-responsive (N = 51) and patients with DRE (N = 44). FINDINGS: In the exploratory analysis, the proportion of time PSWEs were detected was longer in epilepsy compared with participants without epilepsy (P < 0.0001). Analyses of aetiology and EEG localisation revealed that PSWEs were most prolonged in patients with reported focal epilepsy (P = 0.004), particularly with temporal lobe involvement (P = 0.005). Patients with DRE had prolonged time in PSWEs (P = 0.005), corresponding with an increased risk of refractoriness (OR = 1.9; 95% CI 1.2-2.9). Validation analysis confirmed these findings, with prolonged PSWEs in DRE vs. drug-responsive patients (P < 0.0001, AUC = 0.829). Based on the cutoff established in the exploratory cohort, prolonged time in PSWEs in the validation cohort was associated with increased DRE risk (OR = 5.14, 95% CI 2.1-12.3). In patients with poor surgical outcomes (Engel IB-IV, N = 13), pre-surgical EEGs showed prolonged time in PSWEs compared with Engel IA (N = 24, P = 0.038). INTERPRETATION: Analysis of 1159 EEGs from two independent cohorts demonstrated that PSWEs are more prevalent and prolonged in patients with focal epilepsy and may indicate a lack of therapeutic response. FUNDING: The Canadian Institutes of Health Research (168164, 180636).