Postural instability is one of the most disabling motor signs of Parkinson’s disease (PD) and often underlies an increased likelihood of falling and loss of independence. Current clinical assessments of PD-related postural instability are based on a retropulsion test, which introduces human error and only evaluates reactive balance. There is an unmet need for objective, multi-dimensional assessments of postural instability that directly reflect activities of daily living in which individuals may experience postural instability. In this study, we trained machine-learning models on insole plantar pressure data from 111 participants (44 with PD and 67 controls) as they performed simulated static and active postural tasks of activities that often occur during daily living. Models accurately classified PD from young controls (area under the curve (AUC) 0.99+/− 0.00), PD from age-matched controls (AUC 0.99+/− 0.01), and PD fallers from PD non-fallers (AUC 0.91+/− 0.08). Utilizing features from both static and active postural tasks significantly improved classification performances, and all tasks were useful for separating PD from controls; however, tasks with higher postural threats were preferred for separating PD fallers from PD non-fallers.

姿势不稳定是帕金森病 (PD) 最严重的运动障碍症状之一，并且常常导致跌倒和丧失独立性的可能性增加。目前对 PD 相关姿势不稳定性的临床评估基于后推力测试，该测试引入了人为误差并且仅评估反应性平衡。对姿势不稳定性的客观、多维评估的需求尚未得到满足，这些评估直接反映个人可能经历姿势不稳定性的日常生活活动。在这项研究中，我们根据 111 名参与者（44 名患有 PD 和 67 名对照者）的鞋垫足底压力数据训练了机器学习模型，因为他们执行了模拟日常生活中经常发生的活动的静态和主动姿势任务。模型准确地将 PD 分为年轻对照组（曲线下面积 (AUC) 0.99+/- 0.00）、年龄匹配对照组的 PD（AUC 0.99+/- 0.01）以及 PD 跌倒者和非 PD 者（AUC 0.91+/ − 0.08）。利用静态和主动姿势任务的特征显着提高了分类性能，并且所有任务都有助于将 PD 与控制区分开；然而，为了区分 PD 跌倒者和非 PD 者，首选具有较高姿势威胁的任务。