Abstracts

Rationale: Long-term open label studies have demonstrated safety and efficacy of the RNS system with median rates of seizure reduction of 75% at 9 years post-implant compared to pre-implant baseline (Nair et al., 2018). Despite the clinical success of responsive stimulation, its mechanism of action remains largely unknown. Understanding how brain activity changes in the few seconds following stimulation, i.e., the acute effects (AE) of responsive neurostimulation, may provide insights into how the treatment works. Further, finding correlations between the AE of stimulation and clinical outcomes may help with iterating through a large number of stimulation settings and finding the optimal settings faster than currently possible. Methods: Two analyses were performed on ECoG and clinical outcomes data captured from patients who participated in the NeuroPace RNS clinical trials (Bergey et al., 2015; M. Morrell, 2011). In the first analysis, spectral power (SP) features and AE of stimulation on ECoG records from the sham stimulation period (i.e., responsive stimulation OFF) of the pivotal trial was compared to SP and AE of ECoG records from the stimulation period (i.e., responsive stimulation ON) within the same patient. In the second analysis, SP and AE of stimulation on ECoG records were compared before and after patients had a substantial (5x or greater) decrease in clinical seizure rate.For measuring AE, ECoG activity in 2 second windows before detection (pre event window, -2 to 0 seconds w.r.t. detection), and ECoG activity in 2 second windows after detection (post-event window, +2 to +4 seconds) were featurized and compared using equation 1. Acute Effect (AE)(n,k)=((postSP(n,k)-preSP(n,k)))/preSP(n,k) (equation 1)n is the ECoG event numberk is the frequency bin numberpreSP is the spectral power in the pre event windowpostSP is the spectral power in the post event window Results: ECoG data from 20 RNS System clinical trial patients who received sham stimulation followed by active stimulation were retrospectively analyzed. In 19 out of these 20 patients, there was a greater reduction in total spectral power following detection during the active stimulation period compared to sham. Across the 20 patients, active stimulation caused an acute decrease of 22.03% in spectral power on maximum channels (i.e., the channel with maximum change in spectral power following stimulation within a patient) in the 2 second window following active stimulation compared to the 2 second window preceding stimulation. This compared to a 1.36% spectral power decrease on maximum channels during the sham period. The average difference in power decrease between stimulation and sham events was 20.66 % and statistically significant (p = 0.00014, Wilcoxon signed rank test). Additionally, the spectral power difference was more pronounced at spectral frequencies < 40 Hz. Modest associations were found between the acute effect (AE) of stimulation and clinical outcomes within individual detection channels. Even though no strong trends were seen across 47 detection channels from 35 patients included in this analysis, AE features could predict clinical outcomes with an average classification accuracy of 70% within individual detection channels. Conclusions: Responsive neurostimulation results in an acute reduction in the ECoG power in epileptic patients. The AE of stimulation is associated with clinical outcomes, but controlled prospective studies are needed to further investigate the association in detail. Funding: Lundbeck Foundation Clinical Research Fellowship Program