Automated spinal cord stimulation programming using EEG

Chronic pain affects more people than heart disease, diabetes and cancer combined and will affect 1 in 4 Americans over their lifetime1. Spinal cord stimulation (SCS), involving electrical stimulation of the spinal cord from the epidural space, is a promising option to treat chronic neuropathic pain in the low back, legs and feet from multiple etiologies. Presently, SCS uses one-size-fits-all electrical stimulation without regard to ongoing pain state or unique pain related changes in the brain2. SCS has three key drawbacks: 1) After implant, effective stimulation parameters must be chosen through a painstaking, trial-and-error process 3. 2) SCS loses benefit after 1 year in up to 20% of patients4. 3) Before implant, patients must have an expensive one-week trial period because we lack an objective method to predict which patients will receive pain relief. Developing data-driven methods to optimize the programming of electrical stimulation and predict trial success without an invasive implant has the potential to improve therapy efficacy, make SCS accessible to more patients and uncover fundamental mechanisms of chronic pain in the brain.

Aim 1: To validate personalized biomarkers of chronic back and leg pain and guide the optimal tuning of SCS parameters in the ambulatory setting, we will record EEG during varying natural pain states and device settings from 15 patients with SCS implants. Aim 1a. Conditions will include 1) baseline with SCS off for 12 hours; 2) ineffective SCS programs; 3) and established, most comfortable SCS settings. Aim 1b. We will then develop machine learning models to predict optimal settings for each patient from neural data. 10 additional patients will then undergo automated parameter prediction using our method and their time to achieve optimal relief will be compared to 10 patients using conventional programming methods.

Aim 2: To predict which patients will receive benefit from SCS and avert a costly and invasive trial period, we will also record EEG during the above states in 25 patients undergoing standard, temporarily implanted one-week trial periods. We will compare brain activity patterns with machine learning models between patients who have good relief from the trial (and at the 6 months timepoint) vs those that fail to achieve at least 50% pain relief. These analyses will form the basis of future trial prediction algorithms and reveal key insights about brain circuit connectivity related to SCS responsiveness.