Reliability of assessing postural control during seated balancing using a physical human-robot interaction

Abstract

This study evaluated the within- and between-visit reliability of a seated balance test for quantifying trunk motor control using input–output data. Thirty healthy subjects performed a seated balance test under three conditions: eyes open (EO), eyes closed (EC), and eyes closed with vibration to the lumbar muscles (VIB). Each subject performed three trials of each condition on three different visits. The seated balance test utilized a torque-controlled robotic seat, which together with a sitting subject resulted in a physical human-robot interaction (pHRI) (two degrees-of-freedom with upper and lower body rotations). Subjects balanced the pHRI by controlling trunk rotation in response to pseudorandom torque perturbations applied to the seat in the coronal plane. Performance error was expressed as the root mean square (RMSE) of deviations from the upright position in the time domain and as the mean bandpass signal energy (E_mb) in the frequency domain. Intra-class correlation coefficients (ICC) quantified the between-visit reliability of both RMSE and E_mb. The empirical transfer function estimates (ETFE) from the perturbation input to each of the two rotational outputs were calculated. Coefficients of multiple correlation (CMC) quantified the within- and between-visit reliability of the averaged ETFE. ICCs of RMSE and E_mb for all conditions were ≥0.84. The mean within- and between-visit CMCs were all ≥0.96 for the lower body rotation and ≥0.89 for the upper body rotation. Therefore, our seated balance test consisting of pHRI to assess coronal plane trunk motor control is reliable.

Introduction

Several sensory systems are involved in maintaining upright postural control and balance: visual, vestibular, and proprioception (Goodworth and Peterka, 2009, Maurer et al., 2006, Peterka, 2002). To determine the contribution of different sensory feedback pathways to trunk postural control, experiments are performed with sensory information either removed (e.g., closed eyes to remove visual information) (Goodworth and Peterka, 2009, Radebold et al., 2001) or corrupted (e.g., added vibration to reduce the quality of proprioceptive information) (Brumagne et al., 2004, Brumagne et al., 1999, Slota et al., 2008). Identifying the contributions of selected feedback pathways to trunk postural control is important for understanding the mechanisms of control and possible impairments. To that end, several studies have investigated the role of the trunk muscles in maintaining postural control using sitting balancing tests (Larivière et al., 2015, Oomen et al., 2015, Priess et al., 2015, Reeves et al., 2009, Sung et al., 2015, Van Daele et al., 2009, van Dieën et al., 2010a; van Drunen et al., 2013, Willigenburg et al., 2013). Sitting posture has been selected, because the lower body is fixed so that ankle (Brumagne et al., 2008), knee, or hip (Mok et al., 2004) stabilization strategies are not possible; the movement is limited to the trunk to maintain balance.

For the seated balance test to be useful, it must reliably assess trunk control. Three studies addressed the reliability of seated balance tests, and the reported results range from poor to excellent (Cholewicki et al., 2000, Larivière et al., 2013, van Dieën et al., 2010b). These inconsistencies could stem from the fact that unperturbed postural sway was used as a measure of performance. In such a case, the postural sway (output signal) is transient behavior contaminated with neuromuscular noise, which is not repeatable. Thus, data reproducibility and test reliability may vary from one study to another. To overcome this problem, we developed a physical human-robot interaction (pHRI) system (Priess et al., 2015) that consists of a compliant unstable seat and, at the same time, provides torque perturbation as a known input to the system (Cruise et al., 2017, van Drunen et al., 2016, van Drunen et al., 2015, van Drunen et al., 2013). This pHRI also allows for the precise regulation of test difficulty by controlling rotational seat stiffness.

The purpose of this study was to assess the within- and between-visit reliability of a seated balance test under three conditions: eyes open (EO), eyes closed (EC), and eyes closed with lumbar muscle vibration (VIB). The reliability was evaluated using both time and frequency domain analysis, which capture the control attributes such as accuracy and speed. This study uses nonparametric analysis of input-output data, therefore, the contributions of feedback pathways (e.g., reflexive, and intrinsic) cannot be distinguished in this framework. Because the pHRI involved with this test was designed with a known input perturbation, we hypothesized that the performance measures in our seated balance test would be reliable.