Building mood state classifiers to inform deep brain stimulation (DBS) of treatment-resistant bipolar depression - bipolar disorder (BD) is a recurrent neuropsychiatric disorder characterized by…

Building mood state classifiers to inform deep brain stimulation (DBS) of treatment-resistant bipolar depression - bipolar disorder (BD) is a recurrent neuropsychiatric disorder characterized by wide fluctuations in mood, energy, and activity. Although the hallmark of BD is mania in BD-i (hypomania in BD-II), depressive episodes dominate the longitudinal course and account for a disproportionate degree of the morbidity and mortality. About 25% of patients with BD meet criteria for treatment-resistant bipolar depression (trbd). Given the gravity of the illness and paucity of available therapies, there exists a large unmet need for more effective interventions. We propose an early feasibility study to advance the therapy and neurobehavioral monitoring of trbd with deep brain stimulation (DBS) targeting the ventral capsule/ventral striatum (VC/VS) using a sensing capable device. We will leverage a smart phone-based software platform paired with the device's chronic sensing capability to record local field potentials (lfps) time-locked with remote, high-density measures of natural behavior (e.g., from wearables). Our goal is to detect relevant behavioral states including both depression and mania and transitions between these states, including high risk mixed features. Real-time, remote detection of affective states via both behavioral and neural data monitoring will enable new interventional strategies to shift behavior away from pathological behavioral states towards healthier ones, contributing to the long-term stability of patients with BD. Aim 1: efficacy and safety of VC/VS DBS for trbd will be examined in a 9-month open-label early feasibility study of 10 subjects with BD-i, followed by a 3-month stabilization period, and then blinded discontinuation at month 12 to confirm active VS. Sham response. Efficacy outcome will be based on rating scale measures of initial response at 9 months, absence of recurrence of depression during follow-up at 18 months, and management of emergent (hypo)mania. Aim 2: mania/mixed feature detection. 2A. Given the risks of inducing mania or mixed states, a clinician-facing dashboard will be created and updated every 24 hours displaying raw data for: affect/mood, energy/activity, sleep, speech, and anxiety. Behavioral and physiological data will be derived from multimodal methods (e.g., computer vision, voice recordings, apple watch, oura ring, etc.), in/outside the clinic, to objectively assess changes in key behavioral domains. 2B. Composite measures of mood and energy/activity will be developed using machine learning (and other methods) based upon data from the trial subjects. The clinician will be notified when a state change (e.g., emerging mania) is identified that warrants adjustment of stimulation. Aim 3: identify neural markers to be used as classifiers for state changes from depression into either manic or mixed states. Development of neural classifiers for (hypo)mania builds upon research from an NIH funded study of VC/VS DBS in subjects with OCD. The outcome of this aim, coupled with aim 2, has implications for advancing DBS programming with the clinician-in-the-loop and possible future development of an adaptive DBS system in BD. Meeting all milestones could also inform studies of DBS in depression with respect to neurobehavioral feedback to drive DBS programming decisions.