Symmetric channel selection preserves a smaller, fragile SEEG adaptation contrast in 13 patients

Alternating-polarity SEEG results reported a larger decrease in the polarity-invariant component S than in the polarity-flipping component D, using channels selected from the first second of each train. Because S_first also sets the denominator of S_last/S_first, this rule can create regression-to-the-mean bias. In archived per-channel RMS values from OpenNeuro ds005169, a preregistered symmetric rule based on sqrt(S_first × S_last) > 3 × pre-train noise gave a mean contrast C = −0.187 (ratio 0.83; 95% CI −0.326 to −0.029; 12/13 patients negative; exact two-sided sign-flip p = 0.038), about half the first-selected contrast (C = −0.334; ratio 0.72; p = 0.006). The estimated selection bias was +0.142 log units (p = 0.001). A no-stimulation placebo produced a first-selected decrease of −0.416 (p = 0.0005); because it had half as many pseudo-pulses and more noise, it is an upper bound, not an exact bias estimate. The primary result was fragile to patient removal (leave-one-out p = 0.0005–0.075). In the independently verified alternation subset, C was −0.042 (5 patients, 13 trains; p = 0.81). In the descriptive ds006519 sample, the symmetric contrast had the same sign (C = −0.212; 4/5 patients; p = 0.125). Values came from one upstream pipeline; the two same-site datasets may share patients and were not pooled. RMS includes noise, and the test does not identify which part of S is neural. No new raw data were read.

stereo-electroencephalography · alternating-polarity stimulation · channel selection · regression to the mean · adaptation · intracranial EEG · reproducibility