Researchers have found that high-frequency oscillations, or HFOs, in scalp EEG appeared far more often in recordings classified as hypsarrhythmia than in the study's IED comparison groups. The rate of these rapid wave patterns also separated recordings with active epileptic spasms from the other three categories. The authors present the measures as a possible extra layer of information, not as a clinical test ready to be used on its own.
The difference was stark at the recording level. At least one HFO was found in 100.0% of recordings with hypsarrhythmia and ES and 90.9% of recordings with hypsarrhythmia without ES. The corresponding figures were 36.0% for multifocal IED recordings and 29.4% for non-multifocal IED recordings. The overall difference across the four groups was statistically significant (Wald test, P < 0.001).
What the researchers measured
These were not four sets of independent children. The final analysis included 66 scalp EEG recordings from 22 pediatric patients, with 13 recordings in the hypsarrhythmia-with-ES group, 11 in hypsarrhythmia without ES, 25 in multifocal IED, and 17 in non-multifocal IED. Twelve of the patients were male.
To find HFOs, the researchers defined an event as a signal above 80 Hz with at least four oscillations. They filtered the EEG from 80 to 250 Hz, used a Hilbert-based detector with a five-standard-deviation envelope threshold, then visually checked the detections. In plain terms, the analysis looked for repeated fast waves that rose well above the set signal threshold.
Since recordings could repeat the same patient, the analysis used hierarchical statistical models that accounted for patient identity. A generalized estimating equation tested whether HFOs were present, a linear mixed-effects model compared detection rates, and a generalized linear mixed model estimated the odds of a high detection rate. This reduced the risk that one patient's multiple recordings would be treated as unrelated observations.
The strongest separation came from HFO rate
Rate differences were also substantial. The median was 1.8 HFO detections per minute in the hypsarrhythmia-with-ES recordings, with reported values from 0.2 to 6.4. It was 1.2 per minute in the hypsarrhythmia-without-ES group, compared with 0.0 per minute in both multifocal and non-multifocal IED recordings, whose reported upper values were 1.1 and 1.0. Comparisons with the ES group had P < 0.001 after age adjustment; the without-ES group also differed from multifocal IED (P = 0.006) and non-multifocal IED (P = 0.003).
After accounting for age, the odds of being in a high-detection category were 24.66 times as high for hypsarrhythmia with ES as for IED recordings, and 8.38 times as high for hypsarrhythmia without ES. These are odds ratios, not direct chances. The 95% confidence intervals were wide, 4.86 to 125.10 and 1.94 to 36.20, respectively, so the direction of the association was clear but its exact size remains uncertain.
The rate also performed well in a receiver operating characteristic analysis, a test of how well a measure separates two predefined groups. For hypsarrhythmia with ES versus the other three categories, HFO detection rate had an area under the curve of 0.92, with a 95% confidence interval of 0.82 to 1.00. Its sensitivity was 100.0% and specificity 77.4%. Spike detection rate had a lower area under the curve, 0.86, and lower specificity, 66.0%, although its sensitivity was also 100.0% and its confidence interval was 0.77 to 0.99.
A second clue came from frequency
The study found a second, more subtle difference in HFO morphology, meaning the features of individual events. Researchers compared frequency, amplitude, duration and cycle number with nested mixed-effects models that included patient identity and EEG recording date as random effects. Mean frequency was 117.6 ± 19.3 Hz in hypsarrhythmia with ES, 119.5 ± 16.2 Hz without ES, and 126.3 ± 20.4 Hz in the IED group. After age adjustment, both hypsarrhythmia groups had lower mean frequencies than IED (P = 0.008 and P = 0.018), while the two hypsarrhythmia groups did not differ significantly (P = 0.750).
Looking across the ordered categories from IED to hypsarrhythmia without ES to hypsarrhythmia with ES, HFO frequency showed a significant downward trend (P < 0.001). Amplitude, duration and cycle number did not show significant trends. In other words, frequency carried information about where a recording sat along this ordered set of study groups, while the other measured features did not show the same pattern.
A promising signal, not a finished test
The findings still come with a narrow base. Only 22 patients contributed the 66 recordings, and repeated recordings were allowed, even though the models accounted for patient-level clustering. The study therefore shows an association within this cohort and group discrimination in these recordings; it does not establish that HFOs cause hypsarrhythmia or epileptic spasms. The wide confidence intervals around the odds ratios also leave uncertainty about the size of the effect.
The authors say HFO detection rate and morphology could become complementary biomarkers for quantitative assessment of hypsarrhythmia and ictal states. But they explicitly stop short of recommending standalone clinical use and call for large-scale longitudinal validation. For now, the result is a promising measurement signal that needs further validation before its clinical utility is established.
Paper data and sources
Original title: Time-frequency characteristics of scalp high-frequency oscillations provide complementary biomarkers in infantile epileptic spasms syndrome.
Authors: Shunta Yamaguchi, Keisuke Maeda, Naoko Ishihara et al.
Journal/Repository: Experimental physiology
Status: Peer-reviewed
First online: 2026-08-21
DOI: 10.1113/ep094250
Original paper