Biomarkers to Predict Outcome from Responsive Brain Stimulation for Epilepsy

Abstract ________________________________________________________________________________________ The current FDA-approved responsive neurostimulation (RNS) device offers a promising alternative to surgery for more than 600,000 Americans with intractable epilepsy who are not candidates for resective surgery.

Unfortunately, there are no validated biomarkers to predict seizure outcomes before these devices are placed, and approximately 1/3 of patients do not benefit from RNS long-term. There is a critical need to develop biomarkers based upon clinical and electrophysiological data to determine the most effective therapy for patients

with medication-resistant seizures, and to bring quantitative rigor to clinical decision making. The long-term goal of this proposal is to discover and validate a predictive biomarker signature for RNS response that can be used in epilepsy surgery decision making and broadly adopted. To achieve this goal, our overall objective is to

develop this prognostic biomarker signature using machine learning applied to a carefully selected set of features and models calculated from intracranial EEG (IEEG) obtained during presurgical evaluation that incorporates qualitative clinical features. We will collaborate across centers and with industry partners via a novel federated

approach, whereby each clinical site will post data in a common format to their own, private, cloud-based data store, which will be accessible to analysis pipelines run centrally from our cloud-based platform. Our central hypothesis is that biomarker signatures derived from multimodal data collected during evaluation prior to device

implant can be used to predict patient response to RNS therapy. Our preliminary data, analyzing 10 RNS patients each from UCSF, NYU and UPenn, demonstrates our ability to perform the proposed research. In the R61 Phase, we will test this hypothesis retrospectively in 125 patients who underwent IEEG prior

to RNS device placement at the UPenn, UCSF and NYU epilepsy centers. Our specific aims for this phase are: 1) To build a federated processing pipeline for biomarker discovery using presurgical evaluation neuroimaging, IEEG and clinical metadata, 2) To identify a predictive biomarker signature from this data. Our federated analysis

framework will enable us to: (a) accelerate biomarker discovery across multiple sites and industry partners, (b) satisfy patient and industry limitations on sharing proprietary data, (c) provide a practical framework for rapid adoption across clinical centers worldwide. In the R33 phase, the biomarker signature will be validated in 100

additional patients followed longitudinally at 9 clinical sites. The proposed research is innovative because it represents a substantive departure from the status quo by rigorously analyzing multimodal patient data to predict response to RNS and guide decisions on device implantation. The proposed research is significant because it

has the potential to dramatically improve the success rate of RNS for epilepsy through better patient selection. This study also puts into place a novel, versatile, federated data mining infrastructure for multi-center and industry collaboration in translational neuroscience.