SEEG autocorrelation and variance did not rise before seizure onset in 35 HUP patients

Critical slowing down (CSD) predicts that lag autocorrelation and variance rise before a bifurcation. We tested this prediction in stereo-EEG recordings from the HUP OpenNeuro dataset (ds004100). Thirty-five of 38 eligible SEEG patients passed the recorded selection and data-read criteria. For each patient, we analyzed six 5-s windows starting 115 to 15 s before the annotated clinical seizure onset and the same times in interictal run-01, using 5-s windows in 1–40 Hz filtered bipolar signals. The primary contrast was the preictal minus interictal slope across windows, summarized by patient and tested with sign flips and Holm correction. The mean contrast was negative for autocorrelation at 10 ms (AC10; −0.0070 per 100 s, 95% CI −0.0139 to −0.0006; Holm p = 0.099; 17/35 positive) and log10 variance (−0.053, 95% CI −0.128 to 0.013; Holm p = 0.154; 15/35 positive). The secondary 50-ms autocorrelation contrast was also negative (−0.032, 95% CI −0.063 to −0.005; p = 0.035). These results do not support CSD in the two minutes before clinical onset under this analysis. In 22 patients with seizure-onset-zone (SOZ) labels, preictal SOZ-pair slopes exceeded non-SOZ-pair slopes for AC10 (+0.019, p = 0.025) and AC50 (+0.068, p = 0.021); these secondary p values were not corrected. A labeled post hoc analysis found that the contrast came from decreasing autocorrelation in non-SOZ pairs; SOZ slopes were near zero, and the contrast was absent in interictal runs. Five-second windows and six time points produce noisy slopes; slower CSD was not tested. Other limits include clinical onset annotations, one seizure per patient, one center, and an interictal control fixed at 120 s into a clip. The package preserves the preregistered plan, code, per-window features with byte-range hashes, results, and reproduction commands.

critical slowing down · stereo-EEG · epilepsy · autocorrelation · variance · seizure onset zone · OpenNeuro · pre-registration