In 31 HUP SEEG patients, first-10-s recruitment extent did not predict seizure duration within patients

We tested if the spatial extent of early seizure recruitment predicts seizure duration within a patient. This is important for closed-loop neurostimulation, because a device could increase stimulation early only for long seizures. We used per-second SEEG features of 133 seizures (10-600 s) from 36 patients in OpenNeuro ds004100 (HUP, one center). The features are copied from sci publication 7fbada2e. For each seizure, E is the fraction of bipolar pairs with mean 20-100 Hz log power z >= 2 in [0,10) s after the clinical onset mark. The z values use a [-40,-25) s baseline. A plan written before the analysis specified all tests. In the 31 patients with 3 or more seizures, the within-patient Spearman rho of E vs log duration was near zero: mean -0.09, bootstrap 95% CI -0.29 to 0.11, 12 positive and 15 negative patients, one-sided sign-flip p = 0.80. A sham pre-onset window gave a similar result (mean rho -0.15). Line length, mean z, and cumulative windows from [0,1) s to [0,20) s also did not predict duration (all mean rho between -0.18 and 0.11). The within-patient pair AUC was 0.47 and the between-patient rho was -0.18 (p = 0.29). Only one secondary test had p < 0.05: line-length extent in [10,20) s, pooled within-patient r = 0.22, p = 0.02, with no correction for multiple tests. Thus the extent of early recruitment does not tell which seizures of a patient will be long. The CI excludes a mean within-patient rho larger than about 0.11. Limits: one center, 3-5 seizures per patient, clinical offset marks of unknown accuracy, a fixed z >= 2 threshold, and 1-s bipolar features. The package contains the plan, code, input features with S3 provenance, results, a figure, and reproduction commands.

seizure duration · seizure recruitment · stereoelectroencephalography · high-frequency power · within-patient analysis · responsive neurostimulation