Past HFO count pooling gave mixed forecast gains on 13 later dates from four sleep SEEG patients

We tested whether past HFO counts improved forecasts of a later channel map. We used supplied FRandR marks from public OpenNeuro sleep SEEG data. Four depth electrode patients had at least three recording dates. We fixed channel sets from the first two dates. We compared pooled past counts with the most recent date on 13 later dates. Pooling had positive log score gains on seven dates and in two patients. The equal patient mean gain was 0.001291 natural log units per target mark. Its 95 percent t interval was [-0.039460, +0.042042]. This interval assumes independent, approximately normal patient gains from a common distribution. The test did not establish a positive mean gain. A fixed stronger smoothing check had four positive patient gains. Both forecast scores were lower than those from the primary rule. We report all checks and their limits. The package preserves source marks, hashes, the plan, code, forecasts, figures, and repeat instructions. Separate code confirmed counts, scores, and uncertainty summaries. A fresh source download and process reproduced all 24 analysis files byte for byte. The study does not test clinical localization or treatment outcome.

high frequency oscillations · stereoelectroencephalography · sleep · channel maps · forecasting · log score · epilepsy