Parkinson’s disease (PD) is a progressive neurodegenerative disorder that manifests with the cardinal motor signs of bradykinesia, rigidity, tremor and postural instability (PI). Postural instability is a major cause of falls and as the disease process progresses, the most common patient complaints center on gait and balance difficulties leading to falls.1 Falls are the most common reason for hospitalization in PD patients2, impose a significant economic burden to the US healthcare system3 and are a major cause of diminished quality of life4, reduced mobility, disability5 and death. Both clinical6,7 and laboratory-based18,9 assessments of balance provide only a brief window of time into patients’ function. Prior studies have demonstrated poor correlation between capacity- based measurements in the clinic or lab (i.e., what can a patient do when asked) and performance-based measurements in the real world.10,11 We have developed methods to use wearable sensors in the ambulatory setting that can accurately detect a variety of activities and have created a number of quantitative metrics that are specific to PI. In this manner, we can monitor and analyze PI in the real world ambulatory setting in Veterans with PD. At present, there are no effective long term treatments for PI. Major impediments to progress in this field are an understanding of how patients actually experience PI at home11 and categorizing PI into meaningful phenotypic subtypes in order to understand its underlying pathophysiology10,12,13 and evaluate new treatments. The goal of this project is to better understand the underlying kinematic and electrophysiological components of postural instability in Veterans with PD. Aim 1 sends PD patients home for one week with five wearable sensors and a neck-worn video camera to create a massive video-validated quantitative dataset of a variety of events that are relevant to analyzing PI at home (walking, turning, sit to stand transitions, near falls/stumbles). For each video-validated event, we use deep learning algorithms to predict which activity occurred and create ROC curves to examine the algorithms’ predictive accuracy. Aim 2 will use kinematic data obtained from the wearable sensors to develop “deep clinical phenotypes” of postural instability using principal component analysis (PCA) and unsupervised clustering machine learning methods. Using these deep clinical phenotypes, we will then test specific hypotheses related to patients’ future fall risk, their experience of PI at home and the relationship of these phenotypes to clinical data such as the presence of co-morbidities like peripheral neuropathy. In Aim 3, a subset of Veterans with PD from the first two aims will undergo subthalamic nucleus (STN) DBS. We will use local field potential recordings from their leads to understand the physiological signature(s) that occur just prior to, during and after a perturbation evoking a reactive postural response. By recording from contacts in motor and associative regions while undergoing simultaneous kinematic recordings and associative STN stimulation, we can investigate the physiological basis of postural instability in these patients.

Balance problems and falls are among the most common complaints in Veterans with Parkinson’s Disease (PD), but we have no effective treatments and our ability to measure balance and falls remains quite poor. This study uses wearable sensors to measure balance and uses deep brain stimulation electrodes to measure electric signals from the brain in Veterans with PD. We hope to use this data to better understand the brain pathways underlying balance problems in PD so that we can design new treatments to improve balance and reduce falls in Veterans with PD.