Distance-stratified CCEP reciprocity was inconsistent in eight SEEG patients

Question: when stimulation at site A gives an N1 at site B, is B to A also more probable after distance control, and does the reverse response help to predict the map of a new stimulation site? Data: OpenNeuro ds005448 v1.0.0, 8 SEEG patients (STREEF06-13), expert-checked N1 labels, 7,606 dyads of stimulated contact pairs with no shared contact. Method: Mantel-Haenszel odds ratio stratified by distance and shaft; p1-type logistic model with sender and receiver terms and a parametric bootstrap null (20 simulations per patient); 5-fold dyad CV and strict leave-one-sender-out prediction; exact sign-flip tests over patients. Results: the median stratified OR is 0.99 (range 0.52-2.45; 4 of 8 above 1; p = 0.90). Reciprocity occurs mainly in rare edges longer than 40 mm. The p1 reciprocity term is above the null mean in 6 of 8 patients (p = 0.16). In strict leave-one-sender-out prediction, distance alone gives a median per-site AUC of 0.97, and receiver terms (+0.006) and reverse edges (+0.002) add little. The large gain from node terms in dyad CV (+0.07) does not transfer to new sites. Sender and receiver propensities do not correlate (median rho -0.05). Pairs with more white matter contacts send more N1s (rho +0.17, 7 of 8, p = 0.023). Forward and backward N1 latencies correlate weakly (rho +0.18, p = 0.016). Limits: one center, 8 patients, expert labels, overlapping pairs, coarse bootstrap, exploratory tests with no multiple-test correction.

stereo-EEG · SEEG · CCEP · SPES · brain connectivity · reciprocity · N1 response · leave-one-sender-out