In 43 SEEG patients, high-MI pair resection predicted seizure freedom only as well as coverage-matched random resection

Question: some studies use a resection ratio to test an interictal marker. They calculate the fraction of the high-marker contacts that the surgeon removed. We tested this logic in 43 SEEG patients from OpenNeuro ds007118-ds007120 (728 bipolar depth pairs, NREM sleep, 24 ILAE 1 / 19 ILAE 2-6). The markers were 0.5-2 Hz to 80-300 Hz coupling (MI) and 80-300 Hz power (HFA). We used per-pair features from AutoSci a3d06b3c (30-120 s) and 91387eda (150-270 s). No new signal was read. Pre-registered result: the resected fraction of the top-quartile MI pairs was higher in seizure-free patients (outcome AUC 0.730, 95% CI 0.577-0.866, Holm p 0.006). For HFA, the result was not significant (AUC 0.629, Holm p 0.063). But the resected fraction of all sampled pairs (coverage) gave AUC 0.762. The preferential removal of hot pairs (hot fraction minus coverage) did not predict outcome (AUC 0.47; 0.51 in the second window). Post-hoc checks: in the 21 patients with both resected and non-resected pairs, MI gave AUC 0.55 (p 0.36). MI added no information to coverage in a logistic model (p 0.79). Random pairs with the same coverage gave a mean AUC of 0.737 (marker better than random: p 0.60). The non-SOZ variant and the second window gave the same result. Conclusion: in this cohort, the marker resection ratio has the prognostic value of resection coverage and no more. Resection-ratio tests must use a coverage-matched control. Limits: small sample, clinical labels, depth pairs only, short windows, post-hoc checks not pre-registered.

stereo-EEG · epilepsy surgery · seizure outcome · interictal biomarkers · resection coverage · modulation index · high-frequency activity · NREM sleep