Optimizing Clinical Assessments in Parkinson's Disease Through the Use of Wearable Sensors and Data Driven Modeling

Machine learning (ML) is the fastest growing field in computer science, and health informatics is among the greatest challenges. The goal of ML is to develop algorithms which can learn and improve over time and can be used for predictions. Most ML researchers concentrate on automatic machine learning (aML), where great advances have been made, for example, in speech recognition, recommender systems, or autonomous vehicles. Automatic approaches greatly benefit from big data with many training sets. However, in the health domain, sometimes we are confronted with a small number of data sets or rare events, where aML-approaches suffer of insufficient training samples. Here interactive machine learning (iML) may be of help, having its roots in reinforcement learning, preference learning, and active learning. The term iML is not yet well used, so we define it as “algorithms that can interact with agents and can optimize their learning behavior through these interactions, where the agents can also be human.” This “human-in-the-loop” can be beneficial in solving computationally hard problems, e.g., subspace clustering, protein folding, or k-anonymization of health data, where human expertise can help to reduce an exponential search space through heuristic selection of samples. Therefore, what would otherwise be an NP-hard problem, reduces greatly in complexity through the input and the assistance of a human agent involved in the learning phase.

Falls are a major health and injury problem for people with Parkinson disease (PD). Despite the severe consequences of falls, a major unresolved issue is the identification of factors that predict the risk of falls in individual patients with PD. The primary aim of this study was to prospectively determine an optimal combination of functional and disease-specific tests to predict falls in individuals with PD. A total of 101 people with early-stage PD undertook a battery of neurologic and functional tests in their optimally medicated state. The tests included Tinetti, Berg, Timed Up and Go, Functional Reach, and the Physiological Profile Assessment of Falls Risk; the latter assessment includes physiologic tests of visual function, proprioception, strength, cutaneous sensitivity, reaction time, and postural sway. Falls were recorded prospectively over 6 months. Forty-eight percent of participants reported a fall and 24% more than 1 fall. In the multivariate model, a combination of the Unified Parkinson's Disease Rating Scale (UPDRS) total score, total freezing of gait score, occurrence of symptomatic postural orthostasis, Tinetti total score, and extent of postural sway in the anterior-posterior direction produced the best sensitivity (78%) and specificity (84%) for predicting falls. From the UPDRS items, only the rapid alternating task category was an independent predictor of falls. Reduced peripheral sensation and knee extension strength in fallers contributed to increased postural instability. Falls are a significant problem in optimally medicated early-stage PD. A combination of both disease-specific and balance- and mobility-related measures can accurately predict falls in individuals with PD.

An ambulatory gait analysis method using body-attached gyroscopes to estimate spatio-temporal parameters of gait has been proposed and validated against a reference system for normal and pathologic gait. Later, ten Parkinson's disease (PD) patients with subthalamic nucleus deep brain stimulation (STN-DBS) implantation participated in gait measurements using our device. They walked one to three times on a 20-m walkway. Patients did the test twice: once STN-DBS was ON and once 180 min after turning it OFF. A group of ten age-matched normal subjects were also measured as controls. For each gait cycle, spatio-temporal parameters such as stride length (SL), stride velocity (SV), stance (ST), double support (DS), and gait cycle time (GC) were calculated. We found that PD patients had significantly different gait parameters comparing to controls. They had 52% less SV, 60% less SL, and 40% longer GC. Also they had significantly longer ST and DS (11% and 59% more, respectively) than controls. STN-DBS significantly improved gait parameters. During the stim ON period, PD patients had 31% faster SV, 26% longer SL, 6% shorter ST, and 26% shorter DS. GC, however, was not significantly different. Some of the gait parameters had high correlation with Unified Parkinson's Disease Rating Scale (UPDRS) subscores including SL with a significant correlation (r = -0.90) with UPDRS gait subscore. We concluded that our method provides a simple yet effective way of ambulatory gait analysis in PD patients with results confirming those obtained from much more complex and expensive methods used in gait labs.