Skip to main content
Journal of Neurophysiology logoLink to Journal of Neurophysiology
. 2018 Feb 28;120(1):37–52. doi: 10.1152/jn.00330.2017

Sensorimotor control of the trunk in sitting sway referencing

Adam D Goodworth 1,, Kimberly Tetreault 1, Jeffrey Lanman 1, Tate Klidonas 1, Seyoung Kim 2, Sandra Saavedra 1
PMCID: PMC6093949  PMID: 29488840

Abstract

We developed a sway-referenced system for sitting to highlight the role of vestibular and visual contributions to trunk control. Motor control was investigated by measuring trunk kinematics in the frontal plane while manipulating visual availability and introducing a concurrent cognitive task. We examined motor learning on three timescales (within the same trial, minutes), within the same test session (1 h), and between sessions (1 wk). Posture sway was analyzed through time-based measures [root mean square (RMS) sway and RMS velocity], frequency-based measures (amplitude spectra), and parameterized feedback modeling. We found that posture differed in both magnitude and frequency distribution during sway referencing compared with quiet sitting. Modeling indicated that sway referencing caused greater uncertainty/noise in sensory feedback and motor outputs. Sway referencing was also associated with lower active stiffness and damping model parameters. The influence of vision and a cognitive task was more apparent during sway referencing compared with quiet sitting. Short-term learning was reflected by reduced RMS velocity in quiet sitting immediately following sway referencing. Longer term learning was evident from one week to the next, with a 23% decrease in RMS sway and 9% decrease in RMS velocity. These changes occurred predominantly during cognitive tests at lower frequencies and were associated with lower sensory noise and higher stiffness and integral gains in the model. With the findings taken together, the sitting sway-referenced test elicited neural changes consistent with optimal integration and sensory reweighting, similar to standing, and should be a valuable tool to closely examine sensorimotor control of the trunk.

NEW & NOTEWORTHY We developed the first sway-referenced system for sitting to highlight the role of vestibular and visual contributions to trunk control. A parametric feedback model explained sensorimotor control and motor learning in the task with and between two test sessions. The sitting sway-referenced test elicited neural changes consistent with optimal integration and sensory reweighting, similar to standing, and should be a valuable tool to closely examine sensorimotor control of the trunk.

Keywords: modeling, posture, sitting, sway referencing, trunk, vestibular, visual

INTRODUCTION

The human trunk (head, arms, torso, and pelvis) comprises over half of the body’s mass (de Leva 1996), and posture control of the trunk is a foundational motor skill because of its importance in nearly all voluntary activities (Edwards et al. 2017; Karthikbabu et al. 2012; Rachwani et al. 2015; Teng and Powers 2014). Trunk posture requires sensory feedback from visual, vestibular, and somatosensory systems (Andreopoulou et al. 2015; Goodworth and Peterka 2010b; Maaswinkel et al. 2015; Wu et al. 2016) and can be influenced by reflexes and intrinsic biomechanical properties of the trunk (Brown and McGill 2009; Goodworth and Peterka 2009; van Drunen et al. 2015). Impairments in trunk control can have a major debilitating effect on daily activities and socialization. Trunk impairments are common in a variety of orthopedic conditions, such low back pain (Mokhtarinia et al. 2016; Sung et al. 2015), and neurological conditions, such as hemiparetic stroke (de Oliveira et al. 2008; Karthikbabu et al. 2012), Parkinson’s disease (Maetzler et al. 2012), spinal cord injury (Harel et al. 2013; Seelen et al. 1997), and cerebral palsy (Saavedra and Woollacott 2015). Impairments in sensory feedback, perceptual decision, and muscular control are also common in these populations (de Oliveira et al. 2008; Hwang et al. 2016; Perugini et al. 2016; Willigenburg et al. 2013). However, the impact of specific sensory and motor impairments on trunk control is still poorly understood. More sensory-specific tools that characterize sensorimotor integration could delineate the various contributions to trunk posture.

In standing posture, the most common tool to evoke a large shift toward reliance on visual and/or vestibular systems is the surface sway-referenced paradigm (Alahmari et al. 2014; Clark and Riley 2007; Peterka 2002). In surface sway-referenced tests, spontaneous body motion is most often measured from changes in center of pressure on a force plate. Center of pressure is low-pass filtered to approximate the body’s center of mass and then used as a reference signal to tilt the surface to the same angle as the body sway. A slightly more refined technique uses a single-degree-of-freedom backboard, eliminating intersegmental motion, to directly measure the body sway tilt angle as the reference signal (Cenciarini and Peterka 2006; Peterka 2002). In either case, when the surface tilt angle approximates the body sway tilt angle, the modulation of proprioceptive signals in the lower leg is diminished. To maintain balance, the brain must shift reliance away from the proprioceptive feedback and shift toward vestibular (in eyes closed, EC) and visual (in eyes open, EO) conditions. Impairments in the vestibular system can greatly increase body sway and result in falling (Nashner et al. 1982; Peterka 2002). To our knowledge, surface sway referencing has yet to be developed in sitting posture. A sitting sway referencing test would have advantages over existing techniques to assess trunk control (Maaswinkel et al. 2016) because sway referencing could isolate the contributions of visual and vestibular feedback, providing a unique probe into sensorimotor control of the trunk (as opposed to the ankle muscle action that is the primary end effector in standing posture studies). In addition, sitting sway referencing could offer a window into posture mechanisms used by people with more severe disability who cannot stand independently.

The goal of the current study is to better understand sensorimotor processes of trunk control by developing a sitting sway-referenced system and to describe characteristics of motor control. First, we complete a detailed comparison of body dynamics in quiet sitting to sway referencing. A sensorimotor integration model is used to interpret the neural processes that change from quiet sitting to sway referencing. Second, we quantify changes in trunk sway with visual availability (EO vs. EC) and test the hypothesis that subjects exhibit similar sensory reweighting mechanisms between visual conditions to those observed in standing. Third, we introduce a concurrent cognitive task and test the hypothesis that the cognitive task disrupts posture control. Fourth, we examine motor learning during sitting sway referencing. Learning is assessed at three time intervals: within each trial (minutes), within each test session (~1 h), and between two test sessions (~1 wk). Finally, we determine the relation between subjects’ perception of difficulty and their performance on quantitative measures of trunk control.

Addressing the above questions provides new insight into sensory mechanisms of trunk control and provides the first baseline characteristics of sitting sway reference in healthy adults. In populations with poor trunk control, sitting sway referencing has the potential to identify the particular sensory system degrading trunk control and could be a tool to examine the physiological components of improved trunk control or effectiveness of interventions aimed at improving trunk control.

METHODS

Participants.

Twelve healthy subjects were recruited (7 women), provided written, informed consent, and were tested according to a protocol approved by the Human Subjects Committee at the University of Hartford. Subjects’ mean age was 23 ± 1.9 yr, height was 170 ± 9.7 cm, weight was 70.7  ± 15 kg, and upper body height (distance from greater trochanter to glenohumeral joint) was 49 ± 5.5 cm.

Backboard configuration and bench motion.

For each balance test, subjects were seated on a bench with legs separated at a comfortable distance and with feet on a footrest that moved with the bench (Fig. 1). Throughout each trial, the bench was either stationary during quiet sitting or could tilt up and down in the frontal plane during sway referencing (Fig. 1B). Subjects were strapped against a lightweight backboard. The backboard had comfortable straps and foam wedges around the waist and head, similar to other studies (Cenciarini and Peterka 2006; Peterka 2002). The backboard could only rotate in the frontal plane. All subjects reported the backboard to be comfortable. The backboard provided an unambiguous kinematic signal of body tilt to control bench motion during sway referencing (compared with multisegmental motion where the head and trunk segments may be moving differently; Wu et al. 2016). The backboard facilitated interpretation of results, because in a single-link upper body and head model, center of mass motion is directly related to visual and vestibular cues. Finally, a backboard configuration may be necessary in future tests with people who lack independent sitting to limit the degrees of freedom (Goodworth et al. 2017). In the current study, the backboard weighed 6.76 kg, with about half of the mass (3.4 kg) located around the axis of rotation, resulting in a small inertia (0.7 kg·m2). By comparison, the average upper body mass and inertia about the axis of rotation were 47.9 kg and 7.3 kg·m2, respectively. The weight of the backboard was supported by a rigid structure so that in a completely upright position, subjects did not need to generate any additional force.

Fig. 1.

Fig. 1.

Schematic of sitting sway-referenced system. A: surface tilt motion creates tilts of a bench that the subject sits on. B: horizontal translation of the bench is eliminated with rigid connectors on both sides of the vertical legs. A photograph of the bench and surface is included in Goodworth et al. (2017).

The backboard rotated about a shaft (i.e., axis of rotation) that was located midline of the body at the same height as the bench so that the pivot point was at the base of support (Fig. 1). Attached to the shaft was a frictionless potentiometer (CP-2UTX series; Midori America, Fullerton, CA) that measured upper body tilt angle (Goodworth and Peterka 2010a). The upper body tilt angle was sampled at 200 Hz with a real-time operating system (cRIO; National Instruments, Austin, TX).

During sway-referenced conditions, the system was designed to minimize somatosensory feedback through motion of an articulating bench and footrest. Specifically, when the body tilted away from upright, the bench tilted to the same angle, keeping the upper body and pelvis in an approximately neutral position, and the footrest moved with the bench so that legs remained in a neutral position. The body’s tilt position was used as the driving signal within a proportional-derivative (PD) controller. The output from the PD controller was a voltage signal that controlled a servomotor. The servomotor tilted a surface and the articulating bench (Goodworth et al. 2017) to the same angle as the body’s tilt. The time delay between backboard and bench tilt was estimated at 33 ms by calculating the cross-correlation between the backboard and bench waveform during sway referencing. This electromechanical delay suggests that modulation of proprioceptive cues was not fully eliminated in sway referencing. For safety, bench tilt was limited to ±4.5°. Subjects’ body tilt rarely exceeded this tilt angle.

Protocol.

Subjects were tested in two sessions. Each session was separated by 6 or 7 days. In each session, subjects completed 10 separate trials. Throughout each trial, subjects listened to white noise through earphones and were instructed to “stay as upright as possible,” because instructions can impact trunk control (van Drunen et al. 2015). Different trials were designed to manipulate sensory availability and cognitive processes while providing repetition for averaging kinematics and identifying any motor learning. Eight of the 10 trials were presented in random order: 2 eyes-closed trials where subjects only listened to white noise (ECwhite), 2 eyes-closed trials where subjects performed a cognitive task (ECcog), 2 eyes-open trials with white noise (EOwhite), and 2 eyes-open trials with a cognitive task (EOcog). The remaining two trials (ECwhite pretest and ECwhite posttest) were not randomized and were included at the start and end of each test session. During each cognitive task, subjects continued to listen to white noise but were also instructed to either verbally count backward by 3 or 7 from 500 or to verbally go through the alphabet and name animals or foods that started with each letter. The different counting and alphabet tasks were presented in random order for each cognitive trial. The different cognitive tasks provided variety for subjects and did not bias the cognitive tasks toward those with good math or alphabetic recall skills.

Each 3.5-min-long trial consisted of 30 s of quiet sitting while the bench was stationary, followed by 150 s of sway referencing while the bench was moving with the spontaneous body sway, followed by 30 s of quiet sitting while the bench was stationary (see Fig. 2A). A minimum break of 60 s occurred after each trial.

Fig. 2.

Fig. 2.

A: experimental design of each trial included 30 s of quiet sitting with no bench motion (25 s included in analyses), followed by 150 s of sway referencing where the bench moved with spontaneous body tilt angles, followed by 30 s of quiet sitting (25 s included in analyses). Sample experimental body sway for 2 trials, where the subject listened to white noise in either eyes open (EOwhite) or eyes closed (ECwhite) conditions. B: sample amplitude spectra for quiet sitting and sway referencing, averaged between repeated trials in the same session. ECcog and EOcog refers to trials with concurrent cognitive tasks. C: correlations in root mean square (RMS) sway and RMS velocity between sway referencing and quiet sitting.

Analysis.

Each trial was broken into three periods of analysis (Fig. 2A): the first period of quiet sitting, sway referencing, and the last period of quiet sitting. Dependent variables included the zero-mean root mean square of the body sway angle with respect to vertical (referred to as RMS sway) and the zero-mean root mean square of the body sway velocity (referred to as RMS velocity). RMS sway and RMS velocity were calculated across 25 s of quiet sitting for the first and last periods of quiet sitting in each trial. In sway referencing, the 150-s time period was divided into six 25-s intervals. To isolate steady-state behavior, RMS measures were calculated for each of the intervals 2–5. The average across these intervals was taken as the steady-state sway-referenced RMS sway and RMS velocity.

We also used amplitude spectrum as a frequency-domain-dependent variable. An amplitude spectrum decomposes body sway across frequencies and therefore provides more detail into sway dynamics than RMS measures. Amplitude spectra were calculated as the magnitude of discrete Fourier transform (Pintelon and Schoukens 2001) of the body sway time series. The body sway time series used to calculate amplitude spectra were the same 25-s intervals described above for RMS measures. In sway referencing, a separate amplitude spectrum was calculated for each 25-s interval, and then the average amplitude spectrum across intervals was considered the steady-state sway-referenced amplitude spectrum. The 25-s interval was selected so that the fundamental frequency (1/25 s = 0.04 Hz) was the same for quiet sitting and sway referencing. Amplitude spectra were calculated from 0.04 to 2.7 Hz, with additional averaging across adjacent frequencies at higher frequencies so that 9 frequency points are represented with approximately equal distance on a logarithmic scale (Goodworth and Peterka 2009; Peterka 2002).

Statistics on experimental data.

Dependent variables (RMS sway and RMS velocity) were analyzed in repeated-measures ANOVA statistical models using SAS JMP software. Periods of quiet sitting and sway referencing were analyzed separately because sway characteristics in quiet sitting and sway referencing were dramatically different (Figs. 24). To determine the influence of “condition,” vision (EO vs. EC) and cognition (white noise vs. cognitive task) were included as model effects, and post hoc comparisons were made with Tukey’s honestly significant difference test. To investigate motor learning, model effects included test session, repeated trials, first vs. last 25 s of quiet sitting (only used in quiet sitting models), and pretest vs. posttest. Amplitude spectra were compared by first calculating the average amplitude of body sway across low frequencies (0.04–0.6 Hz) and high frequencies (>0.6 Hz) and then using paired t-tests for EC vs. EO trials, white noise vs. cognitive trials, and session 1 vs. session 2. Correlations were calculated with the Pearson’s correlation coefficient between 1) quiet sitting and sway referencing for RMS sway and RMS velocity and 2) RMS sway and answers to perception questions. In all statistical analyses, significance was considered at P < 0.05.

Fig. 4.

Fig. 4.

A: summary of root mean square (RMS) sway and RMS velocity for quiet sitting and sway referencing averaged across subjects and between repeated trials and sessions. Error bars are ±SE across subjects. B: interaction effects showing the influence of cognitive tasks differed in session 2 compared with session 1 for RMS sway (top). C: average subject amplitude spectra averaged across repeated trials. Error bars are ±SE across subjects (error bars are similar in size to symbols).

Sensorimotor integration modeling.

We used a feedback control model of posture to understand the underlying changes in posture control across test conditions (Fig. 3A and appendix a). The model was based on previous studies that represented the spontaneous sway measures in healthy adults and in Parkinson’s disease during quiet stance (Maurer and Peterka 2005) and perturbed stance (van der Kooij and Peterka 2011). The overall approach to parameter estimation was similar to Qu et al. (2009). The model represented the body as a single-link inverted pendulum with mass (m), center of mass height (h), and inertia (J) equal to the upper body (i.e., HAT segment in Winter 2005) plus the backboard. Body motion was stabilized via corrective torque (Tc) from sensory feedback generated from a time-delayed position (Kp), derivative (Kd), and integral (Ki) controller. Kp and Kd represent the net effect of the corrective torque that moves the body toward upright in proportion to body sway position and velocity, respectively. Ki represents corrective torque generated in proportion to body sway away from upright summed across time and generally affects lower frequencies (Goodworth and Peterka 2012). The time delay (τ) represents the inherent delay from receptor stimulation to torque generation from receptor activation, neural transmission, sensory processing, and muscle activation.

Fig. 3.

Fig. 3.

A: feedback control model of sitting posture where active stiffness (Kp), damping (Kd), and integral (Ki) gains generate corrective torque (Tc) in proportion to body sway, velocity, and time integration of body sway, respectively. Sensory noise and torque noise represent the inherent uncertainty and noise in sensory and motor systems. Sensory noise was a low-pass-filtered pink noise, and torque noise was low-pass-filtered white noise. J, m, and h are upper body inertia, mass, and center of mass height, respectively, and g is acceleration due to gravity. B: sample simulated body sway in various eyes open (EO) or eyes closed (EC) conditions in subjects listening to white noise. C: amplitude spectra of sensory and motor noise averaged across the repeated 25-s cycles. Gray lines represent spectra from different noise seeds, and the black line is the average.

Two noise sources (torque and sensory) initiated body movement in the model (van der Kooij and Peterka 2011). Figure 3B shows an example of simulated body sway evoked in the model. Torque noise, based on Maurer and Peterka (2005), was band-limited white noise with a low-pass filter. Sensory noise, similar to Boonstra et al. (2013), was pink noise but included an additional low-pass filter to better approximate sway dynamics. Similar to Maurer and Peterka (2005), we found that filter time constants could be fixed to a single value across all trials without compromising model accuracy. Our time constant values were set to the average of 20 model fits (fitting procedure is described below) where time constants were free to vary in both ECwhite quiet sitting and ECwhite sway referencing. Our time constant for torque noise was 0.0017 Hz (Maurer and Peterka 2005 used 0.0016 Hz) and for sensory noise was 0.46 Hz.

Model parameters m, h, and J were estimated from body and backboard measurements and were 54.7 kg, 0.307 m, and 8.04 kg·m2, respectively. The remaining neural control parameters (Kp, Kd, Ki, τ, and noise amplitudes) were obtained through the “fmincon” optimization function in MATLAB (The MathWorks, Natick, MA). The fmincon function iterated through parameter combinations to determine the set of model parameters that minimized the normalized mean square error (MSE) between the model amplitude spectra and the mean subject experimental amplitude spectra across frequencies 0.04–2.7 Hz (Goodworth and Peterka 2009; Peterka 2002). Model amplitude spectra were obtained from 200 s of simulated time series in MATLAB Simulink using the discrete Fourier transform across 25-s intervals, similar to the experimental time series. For each simulation, the sampling rate was 200 Hz and the eighth-order Dormand-Prince solver was used. Note that the same 12 noise seeds were used for each test condition so that there was no bias in comparisons across test conditions. Also, noise input waveforms were independent of the fitting routine and were not included in the fit. However, the amplitude of noise input waveforms was included in the fit, which can be understood as the scaling or sensitivity of the system to the input noise and reflects an internal change in the system.

Parameter sensitivity.

To determine the sensitivity of amplitude spectra to each parameter, we used model parameters obtained from an ECwhite sway referencing and then increased and decreased each parameter by 25%. For each increase and decrease of a parameter, a separate simulation was run in Simulink and the resulting amplitude spectrum was calculated.

To represent sensitivity of each parameter to different noise realizations (i.e., different noise seeds in the model), parameters were estimated for 12 different noise seeds for each trial and the standard deviations across different noise seeds are displayed. Theoretically, if a near-infinite duration noise input were included in the model simulation, then there would be no variability in parameter values across different noise seeds. The 200-s duration for simulations was heuristically selected 1) to satisfy the practical consideration that long-duration simulations were not feasible within the optimization routine, which required a separate simulation and amplitude spectra calculations for each of the thousands of iterations of parameter combinations, and 2) to maintain averaging of amplitude spectra similar to that of the experimental data (i.e., a similar number of 25-s intervals). When exploring longer and shorter simulation durations, we found very similar mean parameter values across different noise seeds and very similar relative standard deviations across parameters compared with results presented in the current study.

Perception questions.

At the end of each session, subjects were asked questions to aid our understanding of their comfort and the relation between subjects’ perception of difficulty and their quantitative performance during sway referencing. Questions included, “How much difficulty did you have maintaining balance in eyes open/eyes closed?” (Q1/Q2), “How much did the cognitive task affect your balance?” (Q3), “How much do you feel fatigue affected your performance?” (Q4), and “Did you experience any motion sickness?” (Q5). Questions were answered on a 5-point rating scale with 1 indicating small and 5 indicating large. These questions are important to document to understand a subject’s experience in this novel experimental setup and how perception may be related to motor control and motor learning.

RESULTS

Quiet sitting and sway referencing.

Kinematic results demonstrated unique behavior in time periods that included sway referencing compared with quiet sitting. Figure 2A shows a sample time series and amplitude spectra from one subject in an ECwhite and EOwhite trial. Body sway significantly increased at the onset of sway referencing compared with quiet sitting (on average, 4–6 times higher RMS sway). There was a visually noticeable oscillation in body sway during sway referencing that was not present in quiet sitting. This oscillation was evident in amplitude spectra around 0.1–0.3 Hz in sway referencing, whereas the amplitude spectra of quiet sitting smoothly decreased with increasing frequency on the log scale in Fig. 2B. Averaging across subjects and test sessions made the difference in sway dynamics between sway referencing and quiet sitting even more evident (Fig. 4).

Across all trials, there were generally weak correlations between time periods of sway referencing and quiet sitting (Fig. 2C): RMS sway in the first 25 s of quiet sitting to sway referencing (R = 0.32, P < 0.001), RMS sway in the last 25 s to sway referencing (R = 0.36, P < 0.001), RMS velocity in the first 25 s of quiet sitting to sway referencing (R = 0.06, P = 0.37), and RMS velocity in the last 25 s to sway referencing (R = 0.22, P < 0.001). These correlations were influenced by subjects and trials with particularly large body sway. When data outside of 2SD were excluded, correlations dropped to R = 0.12 for RMS sway in the 1st 25 s, R = 0.25 for RMS sway in the last 25 s, R = 0.02 for RMS velocity in the 1st 25 s, and R = 0.17 for RMS velocity in the last 25 s, but they remained significant between sway referencing and the last 25 s for both RMS sway and RMS velocity. Thus, although there was not a strong relationship between the quiet sitting and sway referencing in the majority of subjects and trials, we found that trials with particularly large body sway and velocity in sway referencing tended to have large sway and velocity in quiet sitting, especially during the last 25 s of quiet sitting.

Influence of vision and cognition.

In periods of quiet sitting, varying the test condition (vision and cognition) had a significant effect on RMS sway (P < 0.001; Fig. 4A, top left). Post hoc analyses revealed that sway during ECcog was significantly higher than during both EOwhite (P = 0.001) and EOcog (P = 0.004). In quiet sitting, RMS velocity was not significantly influenced by test condition (Fig. 4A, bottom left).

In periods of sway referencing, varying test condition had a larger influence on behavior, with a significant effect on both RMS sway (P < 0.001; Fig. 4A, top right) and RMS velocity (P < 0.001; Fig. 4A, bottom right). RMS sway during both EC conditions was significantly higher than during both EO conditions (P < 0.001), and sway in EOcog was significantly higher than in EOwhite (P < 0.001; Fig. 4A, top right). The highest to lowest RMS velocity conditions were ECcog > ECwhite > EOcog > EOwhite. All pairwise RMS velocity comparisons were significantly different at P < 0.001, except ECcog vs. ECwhite (P = 0.008) and ECwhite vs. EOcog (P = 0.013; Fig. 4A, bottom right).

In the frequency domain, visual availability influenced a wide bandwidth of frequencies of body sway (EO vs. EC in both quiet sitting and sway referencing; Fig. 4C). In sway referencing, EO conditions had significantly lower amplitude spectra than EC conditions when averaged across both low (0.04–0.6 Hz) and high frequencies (above 0.6Hz), with P < 0.0001 for all comparisons. In quiet sitting, EOcog was significantly lower than ECcog across low (P = 0.03) and high frequencies (P = 0.013), but EOwhite and ECwhite were not significantly different. In sway referencing, the EOcog condition resulted in significantly larger amplitude spectra compared with the EOwhite condition across low (P = 0.012) and high frequencies (P = 0.017) of body sway, but no statistical differences were found between ECwhite and ECcog. In quiet sitting, there were no significant differences in amplitude spectra between cognitive and white noise conditions.

Motor learning.

We investigated motor learning on three different timescales. First, because quiet sitting with a stationary bench was performed at the beginning and end of each trial, we investigated short-term learning within each trial by comparing the first to last 25 s of quiet sitting. Second, because each test condition (EOwhite, ECwhite, ECcog, and EOcog) was repeated two times during each session, we investigated motor learning within each session by comparing these repeated trials. We also investigated motor learning within one session by comparing ECwhite pretest with ECwhite posttest. Third, we investigated more long-term learning across sessions by comparing sway behavior between sessions (1 wk).

Figure 5A shows that within each trial there was a significant decrease in RMS velocity (but not RMS sway) in the time domain between the first and last 25-s period of quiet sitting (P < 0.001 for all conditions). Within each session, there were no significant differences between repeated trials in the eight randomized conditions in either RMS sway or RMS velocity in either quiet sitting or sway referencing. When ECwhite posttest was compared with ECwhite pretest, there were no significant differences in RMS sway or RMS velocity in quiet sitting periods. However, in sway referencing, there was a significant decrease of 33% in ECwhite posttest compared with pretest in both RMS sway (P = 0.035) and RMS velocity (P < 0.001) in session 1 (Fig. 5B), but no significant differences were found between pre- and posttest in session 2. Between sessions, there was a significant 23% decrease in RMS sway (P < 0.006), but not RMS velocity, in the quiet sitting periods (data not shown). In sway referencing, session 2 was associated with 23% lower RMS sway (P < 0.001) and also an 8.7% lower RMS velocity (P = 0.007) across all conditions (data not shown). Figure 5C, left, shows the amplitude spectra averaged across all conditions for sessions 1 and 2. There was a significant decrease in body sway in session 2 vs. session 1 across lower frequencies (<0.6 Hz; P = 0.005), but not higher frequencies (Fig. 5C, middle). In addition, subjects improved most in conditions that involved the concurrent cognitive task (Fig. 5C, right). There was a significant reduction in sway amplitude across lower frequencies in session 2 vs. session 1 in ECcog (P = 0.004), EOcog (P = 0.029), and ECwhite (P = 0.033).

Fig. 5.

Fig. 5.

A: short-term motor learning was evident through changes in root mean square (RMS) velocity from the 1st 25 s of quiet sitting (preceding sway referencing) to the last 25 s of quiet sitting (following sway referencing) in each trial. Bar plots show the mean subject data averaged across sessions. Error bars are ±SE across subjects. ECwhite and EOwhite, white noise trials with eyes closed and eyes open, respectively; ECcog and EOcog, cognitive trials with eyes closed and eyes open, respectively. B: learning within and between sessions for ECwhite test conditions. Note that only ECwhite conditions had pre- and posttests. There was a significant decrease in RMS sway and RMS velocity from pre- to posttest in session 1 but not in session 2. *P < 0.05. C: longer term learning was evident in sway referencing amplitude spectra at lower frequencies of body sway (left). In plots at middle and right, a ratio of amplitude spectra equal to 1 means that body sway was identical in both sessions. A ratio <1 means that body sway was lower in session 2. Lower body sway in session 2 was more evident for cognitive compared with white noise trials.

Interaction effects.

There was a significant interaction effect in sway referencing (P = 0.02) between condition and session in RMS sway. In session 1, subjects’ body sway was more affected by changing test conditions compared with session 2. Figure 4B, top, shows that in session 2, the influence of the cognitive task had a much smaller effect on sway compared with session 1. This interaction was not seen in RMS velocity. This finding is consistent with Fig. 5C, which shows how differences in sway dynamics between sessions were not uniform across all test conditions. Specifically, sway dynamics in the white noise conditions were more similar between sessions, whereas sway dynamics differed more between sessions for cognitive conditions.

Modeling results.

We used a feedback model to interpret how the neuromotor system changed between test conditions (Fig. 2). Model results were obtained for ECwhite and EOwhite in quiet sitting and for ECwhite, EOwhite, and EOcog in sway referencing, because these conditions exhibited large differences in RMS sway and amplitude spectra. Other comparisons were possible (EOwhite vs. EOcog in quiet sitting and ECwhite vs. ECcog in quiet sitting and sway referencing) but considered redundant because differences in body sway across these conditions were small. Below is a description of model accuracy, parameter changes across test conditions, the sensitivity analysis, and model results for exploring changes between test sessions.

Figure 6A shows the model results alongside experimental amplitude spectra. In general, the model was able to accurately account for sway dynamics across a range of frequencies (0.04–2.7 Hz). The average (SD) MSE across the different noise seeds in sway referencing was 0.0052 (0.0046) for ECwhite, 0.0033 (0.0025) for EOcog, and 0.0024 (0.0015) for EOwhite, and that in quiet sitting was 0.0046 (0.0041) for ECwhite and 0.011 (0087) for EOwhite, which was over 2 times higher than that for the other conditions, suggesting some limitation in the model to fully account for EOwhite quiet sitting data.

Fig. 6.

Fig. 6.

A: model (lines) and experimental (symbols) amplitude spectra. The model presented is the average model fit across 12 different random noise seeds. The experimental data presented are the mean subject data averaged between repeated trials and sessions. B: means ± SD of parameter values across 12 different random noise seeds for various conditions. The torque and sensory noise vertical axis are scaled by the root mean square (RMS) of their waveforms, so a value of 1 indicates that the torque or sensory noise input waveform had an RMS value of 1 N·m or 1 rad, respectively.

Model parameters are shown in Fig. 6B, with test conditions on the horizontal axis ordered according to amplitude spectra magnitude (lowest gains on the left, EOwhite quiet sitting; highest gains on the right, ECwhite sway referencing). From quiet sitting to sway referencing, the most notable parameter changes were the increases in noise amplitudes (torque and sensory) and integral gain (Ki). The stiffness (Kp) and damping (Kd) neural control parameters showed modest decreases from quiet sitting to sway referencing. The sensorimotor time delay (τ) was fairly similar between quiet sitting and sway referencing.

Across the three sway referencing conditions, the most notable changes in parameters were in sensory and motor noise. There was a pattern of increasing sensory noise associated with larger magnitude of amplitude spectra. Lack of visual availability (ECwhite vs. EOwhite) resulted in larger noise (sensory and torque). The cognitive task (EOcog vs. EOwhite) resulted in larger sensory noise and Ki. In quiet sitting, lack of visual availability (ECwhite vs. EOwhite) was associated with a slightly larger torque noise, lower Kp, and higher τ.

The sensitivity analysis (Fig. 7) shows that each parameter impacted sway dynamics differently. The nominal model (circles) represent the amplitude spectra derived from a model fit to an ECwhite condition in sway referencing. The lines represent the amplitude spectra after increasing (solid) or decreasing (dashed) each parameter by 25%. Changes in sensory noise uniformly increased or decreased body sway across all frequencies. Other parameters impacted sway more across specific frequency ranges. Sensitivity results were consistent with parameter changes observed across sway referenced conditions and between quiet sitting and sway referencing. In sway referencing, experimental EOwhite body sway was lower than ECwhite across a wide range of frequencies, with slightly larger differences at low frequencies. This result is consistent with a lower EO sensory noise (impacting all frequencies) and lower EO torque noise (impacting frequencies below ~0.6 Hz). Experimental sway in EOcog was higher than EOwhite across all frequencies, with the most similar sway amplitudes at the lowest two frequencies (0.04 and 0.1 Hz). This result is consistent with a higher sensory and torque noise in EOcog compared with EOwhite, along with an increase in Ki in EOcog that reduced body sway at the lowest two frequencies. Finally, for sway referencing compared with quiet sitting, it is clear that the increase in sensory and motor noise contributed to the overall increase in the body sway amplitude. The increase in Ki contributed to the slight drop in sway amplitude at the lowest frequencies in sway referencing vs. quiet sitting conditions, whereas decreases in Kp and Kd contributed to the amplitude spectra shape between 0.04 and 3Hz.

Fig. 7.

Fig. 7.

Sensitivity of amplitude spectra to each model parameter. The nominal model was derived from a fit to an eyes-closed white noise (ECwhite) sway referencing condition. Parameters are shown in Fig. 6. The lines show the amplitude spectra after each parameter value was increased (solid) or decreased (dashed) by 25%. Sensory noise uniformly increased or decreased body sway across frequencies, whereas other parameters impacted sway more across specific frequencies.

We explored alternative hypotheses about the changes between quiet sitting and sway referencing. In particular, we used the model to test if all changes between ECwhite in quiet sitting vs. sway referencing could be explained by only changes in neural control parameters (Kp, Kd, Ki) and the time delay (τ). In this scenario, model errors were over 30 times larger (average MSE = 0.189) than the model with both neural control and noise parameters varying. We also explored varying only torque and/or sensory noise. Although an increase in sensory noise alone could account for the main increase in trunk sway (higher amplitude spectra) in sway referenced vs. quiet sitting, further coordinated changes in Kp, Kd, and Ki were needed to fully account for sway dynamics. We found similar results exploring the differences across test conditions in sway referencing. That is, changes in noise (sensory and torque) could account for much of the differences across test conditions, but coordinated changes in Kp, Kd, and Ki were needed to more accurately describe body sway across all frequencies. However, the time delay had the lowest impact. If τ was held constant across different test conditions and noise seeds, model accuracy was nearly the same.

Finally, we used the model to better understand the neural changes between sessions in sway referencing. The model was fit separately to amplitude spectra (averaged across all conditions) in sessions 1 and 2 (Fig. 8). The main difference in parameters between sessions was an increased Kp, increased Ki, and lower sensory noise.

Fig. 8.

Fig. 8.

A: model and experimental sway-referencing amplitude spectra and ratio of amplitude spectra in sessions 1 and 2 (averaged across trials). B: means ± SD of parameter values across 12 different random noise seeds for sway referencing in sessions 1 and 2.

Subject perception.

The relation between subjects’ perception of their performance and quantitative measures depended on the question asked. All subjects reported the test comfortable. As expected, subjects reported eyes closed as being more difficult than eyes open, consistent with RMS measures. There was a modest positive relation between the perception of difficulty in eyes closed vs. the ECwhite–EOwhite RMS sway in sway referencing (R = 0.34). In sway referencing, there was a moderately strong positive correlation between perception of how cognition negatively impacted sway and the average RMS sway across cognitive minus white noise trials (R = 0.57), with the strongest relation being perception and EOcog–EOwhite RMS sway (R = 0.68; Fig. 9). One person reported motion sickness. This person had notably large sway velocity during quiet sitting compared with the other 11 subjects in most trials during session 1, with 53% and 112% larger RMS velocity (averaged across trials) in the first and last 25 s of quiet sitting, respectively, compared with the other 11 subjects. This subject was similar to others in RMS sway and in RMS sway and RMS velocity during sway referencing conditions. Self-reporting of fatigue varied considerably among subjects, ranging from 1 to 5 with an average of 3 (1.2). Although there was no correlation between fatigue and performance from pre- to posttests in sway referencing, there was a significant correlation between perception of how fatigue impacted performance and the average RMS sway across sway referencing conditions (R = 0.65, P = 0.021). A summary of correlations between perception and sway referencing is presented in Table 1.

Fig. 9.

Fig. 9.

An example of correlations between perception of performance and quantitative measures for question 3 (Q3). The average sway difference is based on root mean square (RMS) sway during the period of sway referencing, averaged across repeated trials. EOcog − EOwhite, difference in average sway between cognitive and white noise trials with eyes open.

Table 1.

Perception vs. performance (RMS sway) in sway reference

EO Conditions
EC Conditions
Cognitive Conditions
EC − EO
Cognitive − White Noise
ECwhite − EOwhite
ECcog − EOcog
EOcog − EOwhite
ECcog − ECwhite
Perception Sway Sway velocity Sway Sway velocity Sway Sway velocity Sway diff. Velocity diff. Sway diff. Velocity diff. Sway diff. Velocity diff. Sway diff. Velocity diff. Sway diff. Velocity diff. Sway diff. Velocity diff.
Q1 R 0.24 0.42 0.51 0.32 0.56 0.20 0.40 0.42
P 0.45 0.18 0.087 0.30 0.057 0.53 0.19 0.18
Q2 R 0.37 −0.11 0.13 −0.35 0.34 −0.26 −0.11 −0.39
P 0.24 0.73 0.70 0.27 0.27 0.41 0.74 0.21
Q3 R 0.27 0.22 0.57 0.32 0.68 0.38 0.37 0.22
P 0.39 0.50 0.05 0.31 <0.05* 0.23 0.24 0.50

Values are correlation coefficients (R) and P values indicating perception (answers to questions) vs. performance [root mean square (RMS) body sway] during and between (differences, diff.) conditions with subjects’ eyes open (EO) or eyes closed (EC) while listening to white noise (EOwhite, ECwhite) or performing cognitive tasks (EOcog, ECcog). The following questions were asked: Q1, How much difficulty did you have maintaining balance in eyes open? Q2, How much difficulty did you have maintaining balance in eyes open? Q3, How much did the cognitive task affect your balance? Q4, How much do you feel fatigue affected your performance? and Q5, Did you experience any motion sickness? Answers (perception) ranged as follows: not at all ← (1, 2, 3, 4, 5) → very much. Q4 and Q5 had fewer analyses because only 1 subject reported motion sickness, and it did not seem necessary to explore extensive correlations with Q4. In Q4, there was significant correlation between perception of fatigue and RMS body sway averaged across all sway referencing tests, but there was no correlation between perception of fatigue and pre- minus posttests in sway referencing.

*

Statistically significant.

DISCUSSION

The overarching goal of our study was to better understand sensorimotor control of the trunk by developing and testing a sitting sway-referenced system. Because this is the first study to describe the sway reference paradigm in sitting, we investigated various aspects of motor control, summarized in four questions below.

How does sensorimotor control change between quiet sitting and sway referencing?

We found clear differences in kinematic behavior between quiet sitting and sway referencing, suggesting that sitting sway referencing offers a unique window into trunk control. The differences in sway referencing compared with quiet sitting included 1) a larger sway magnitude with minimal correlation in RMS measures for the majority of trials, 2) unique amplitude spectra, and 3) larger influence of vision and cognitive processes. These differences were interpreted through a feedback model that showed large increases in sensory and motor noise with modest changes in neural control parameters in sway referencing compared with quiet sitting.

The changes in noise can be understood through the concept of optimal integration and sensory reweighting in sway referencing (Kuo 2005; Nashner et al. 1982; Peterka 2002; van der et al. Kooij 2001). Sensory and motor systems are noisy (Faisal et al. 2008). In quiet sitting, somatosensory, visual, and vestibular cues all provide congruent information about body orientation relative to upright. By combining multiple noisy sensory cues (Ernst and Banks 2002; van Beers et al. 1999), and by relying more on signals with lower noise, the brain can minimize uncertainties in body orientation estimation and generate highly tuned muscle activations to minimize body sway (Goodworth and Peterka 2010b; Mahboobin et al. 2009). In contrast, in sway referencing, somatosensory feedback is minimized, and the brain has fewer signals providing information about body orientation relative to vertical. Moreover, surface sway referencing is associated with sensory reweighting toward the use of vestibular feedback (Nashner et al. 1982; Peterka 2002), and previous studies suggest vestibular noise associated with posture control is larger than proprioception noise (Fitzpatrick and McCloskey 1994; Mahboobin et al. 2009; Peterka and Loughlin 2004; van der Kooij and Peterka 2011). Vestibular noise has been modeled with a magnitude of six times larger than proprioception to describe standing posture in surface sway referencing (Peterka and Loughlin 2004). The present study found a similar increase in magnitude of sensory noise in sway referencing compared with quiet sitting. This similarity is noteworthy: although both studies investigated posture control, the biomechanics (standing vs. sitting) and planes of motion (sagittal vs. frontal) differed. Taken together, the increase in noise in the sway referencing model represents the increase in uncertainty in estimating body orientation with fewer sensory systems, an increased reliance on vestibular feedback (and vision in EO conditions), and an increase in motor output variability. These sensorimotor changes are the same as those described for standing sway referencing.

However, increased noise alone could not explain all the changes observed in sway referencing compared with quiet sitting. To fully account for sway dynamics, additional changes in neural controller parameters (Kp, Kd, and Ki) occurred. The increase in Ki indicates that subjects generated more torque as body sway deviated away from upright for a longer time. An increase in Ki was noted as a compensation for reduced sensory feedback in a previous study investigating trunk control in participants with bilateral vestibular loss (Goodworth and Peterka 2010b). The decrease in Kp and Kd indicates lower overall stiffness and damping to orient the body vertical. The sensitivity analysis showed that this decrease contributed to the increase in body sway between 0.1 and 0.3 Hz. A very similar experimental finding was shown during standing sway referencing, where body sway was maximal across similar frequencies, and these sway dynamics were attributed to a reduction in torque generated from sensory feedback (see Figs. 3 and 5 in Peterka and Loughlin 2004), similar to the reduction in stiffness and damping observed in the present study.

Changes in control strategy and the influence of sensory and motor noise between quiet sitting and sway referencing can also be interpreted in light of the differences in biomechanical task and difficulty. In sway referencing, corrective torques are generated about an approximately neutral pelvis-to-trunk position regardless of the trunk tilt angle, whereas in quiet sitting, muscle stretch, angle of pull, and lever arms modulate with trunk tilt angle. Task difficulty also differs. In quiet sitting, the pelvis and hips provide a static and relatively large base of support for most trunk displacements, whereas sway referencing has only one position (perfectly upright) where the base of support is not in motion. Some studies have suggested an increase in coactivation with unstable sitting (Oomen et al. 2015). In sway referencing, coactivation could be modeled as additional “stiffness” and “damping” that tends to orient the trunk toward the pelvis and surface (Goodworth and Peterka 2009; Peterka 2002; van Drunen et al. 2015) and with increased torque noise (Reeves et al. 2006). This surface-orienting stiffness would effectively counter the vertical-orienting stiffness and damping modeled with parameters Kp and Kd in the current model. If coactivation occurred during sway referencing, the net effect would be a lowering of Kp and Kd and an increase in torque noise in sway referencing, similar to what we found. We also speculate that co-contraction could have contributed to the increase in motor noise. Future studies using surface electromyography could help clarify this topic.

Unstable sitting has also been associated with changes in parameters obtained from nonlinear stabilogram diffusion analyses (Cholewicki et al. 2000). In particular, parameters related to stochastic activity and the correlation between time step increments increased in unstable sitting compared with quiet sitting. Results from a previous standing study (Maurer and Peterka 2005) help to merge our findings with the stabilogram diffusion results. Maurer and Peterka (2005) compared parameters in a feedback model with those in the stabilogram diffusion and found a positive correlation between sensorimotor noise and the same stabilogram diffusion parameters that were reported to increase in unstable sitting, which is consistent with our findings in sitting that sensorimotor noise increased. This result helps understand the relation of stabilogram diffusion to feedback modeling and provides further evidence that many of the neural processes (and methods of inquiry) in standing can be applied to sitting.

Are subjects more sensitive to vision and cognitive processes during sway referencing?

Body sway was much more influenced by visual availability and the cognition task in sway referencing compared with quiet sitting. During sway referencing, visual availability was associated with a 50% decrease in RMS sway and reduced sway across all frequencies (ECwhite vs. EOwhite). This decrease in EO vs. EC conditions is comparable in magnitude to decreases in posture sway found in standing-surface sway referencing (Horak et al. 2002). In the current study, the main model parameter accounting for these changes was a 43% decrease in sensory and motor noise from ECwhite to EOwhite. The decrease in sensorimotor noise in EO compared with EC is consistent with a better estimation of body orientation and lower motor output variability in EO compared with EC conditions. A previous study found that visual and vestibular inputs had much larger impact during unstable sitting than during quiet sitting (Andreopoulou et al. 2015). These results were attributed to the notion of sensory reweighting, where the brain uses visual and vestibular cues to a greater extent in unstable sitting. Clinically, unstable sitting and sitting sway referencing can be used to probe sensory integrity of visual and vestibular sensorimotor pathways for posture control, similar to standing balance (Cohen et al. 1996). In general, assessing balance in more dynamic environments has been shown to highlight differences in sensorimotor processes (Goodworth et al. 2015; Goodworth and Peterka 2010a; Hafström et al. 2002; Horak et al. 1997; McAndrew et al. 2010).

We also found that the concurrent cognitive task had a significant effect during sway referencing but not quiet sitting. Previous studies investigated the influence of cognitive tasks on posture show mixed results. Some studies report a decrease in posture sway with a cognitive task (e.g., Rankin et al. 2000; Resch et al. 2011), others report an increase (Rankin et al. 2000; Redfern et al. 2004; Pellecchia 2003), and others report negligible effects (Resch et al. 2011; Yardley et al. 2001). In the present study, we found negligible effects in quiet sitting, consistent with the notion that during any “easy” posture task, the cognitive effect is small and other movements, such as utterances and respiration, play a larger role (Yardley et al. 2001). In EO quiet sitting, body sway was so small that respiration could have played a notable role in the observed body sway. However, because respiration was not explicitly modeled, we speculate that respiration contributed to the lower model accuracy in the EO quiet sitting compared with other conditions.

During sway referencing, the cognitive task increased posture sway in EO conditions but had little impact in EC. These results are similar to existing standing posture studies during surface sway referencing in EO (Mujdeci et al. 2016) and EC (Mujdeci et al. 2016; Resch et al. 2011). The minimal impact of the cognitive task in EC has been explained as “posture first.” That is, posture is prioritized over performance on the concurrent cognitive task when posture is significantly challenged. We did not measure performance on the cognitive task, but it is possible that performance on the cognitive task was lower in EC vs. EO conditions. It was beyond the scope of the present study to investigate the various factors that contribute to changes with cognition, such as cognitive difficulty, spatial vs. verbal task, etc. (Barra et al. 2006; Mujdeci et al. 2016). Our goal was to understand if a cognitive task impacted posture sway in sitting sway referencing, and, if so, to use a model to help understand why this occurred. Our model suggested that the main difference between EOcog and EOwhite is increased noise. Thus an interference with body orientation estimation and motor outputs accompanied the cognitive task in EO conditions. Other studies have shown longer reaction times with dual tasking (Resch et al. 2011; Yardley et al. 2001). However, these longer reaction times were related to voluntary tasks and are likely minimally related to the time delay associated with posture control, because our model showed minimal change in the time delay with the cognitive task.

Do subjects exhibit motor learning?

We found evidence for motor learning in some, but not all, measures and timescales. One striking learning effect was a decrease in sway velocity in quiet sitting following the 2.5 min of sway referencing (Fig. 5A). This learning was temporary and dissipated before the start of the subsequent trial, which typically started after about 2 min of rest. The cause of this short-term adaptation is unknown. It is likely that neural changes in Kp, Kd, Ki, or possibly co-contraction during sway referencing did not fully return to normal sitting state.

We also found evidence for longer term learning, with a 23% reduction in RMS sway and 9% reduction in RMS velocity during sway referencing from session 1 to session 2 (separated by 1 wk). Thus the longer timescale (1 wk) was mostly associated with changes in body position, whereas the shorter timescale (minutes) was associated with changes in velocity, supporting the idea that these two types of learning are from different neural processes (Tjernström et al. 2002; Wrisley et al. 2007).

The reduction in body sway in session 2 occurred predominantly at lower frequencies and included lower sensory noise, higher stiffness (Kp), and higher integral (Ki) gains. The minimal difference in damping (Kd) may reflect the smaller change in RMS velocity compared with RMS. The reduction in sensory noise does not imply that the inherent variability in sensory receptor and transduction decrease, but rather that subjects improved their ability to interpret and use visual and vestibular cues to balance (because the sway referencing test was designed to focus on these sensory systems). Interestingly, in session 2, subjects improved more on cognitive vs. white noise tests (Fig. 5C), suggesting that the way subjects were influenced by the dual tasking changed between sessions. Others have used dual tasking (cognitive and balance/mobility) during training and found better outcomes on tests that required cognitive and mobility compared with participants in the study who did not have dual task training (Silsupadol et al. 2009). Finally, given that we also found a 21% decrease in quiet sitting in RMS sway from session 1 to session 2, the long-term learning was not restricted to the sway referencing task. Although improvements in sensory utilization are expected to improve quiet sitting also, it is possible that comfort/confidence contributed to the reduction in sway position. Comfort/confidence may have been the explanation for the consistently higher sway observed on the first attempt at sway referencing (pretest compared with posttest) in session 1, whereas there were no pretest-to-posttest differences in session 2.

Previous studies have also described motor learning in standing (Pasma et al. 2016; Van Ooteghem et al. 2009; Wrisley et al. 2007) and unstable sitting (Barbado et al. 2017). In these previous studies, notably higher body sway was observed during the very first posture trial, similarly to our study. Two of the standing studies showed continued improvement with repeated exposure to a translating platform (Van Ooteghem et al. 2009) and sway referencing tests (Wrisley et al. 2007). Also, retention was noted after 24 h (Van Ooteghem et al. 2009) and after 1–4 wk (Wrisley et al. 2007). Similarly, Barbado et al. (2017) examined adaptation in trunk control in both anterior-posterior and medial-lateral directions while subjects were sitting on a hemispherical seat. This configuration has similarity to sway referencing. Subjects were tested in 3 sessions across 3 wk. The authors generally showed reductions in center of pressure and trunk kinematic motion across repeated trials and between 1 and 2 wk, with retention maintained in the third week. In contrast, other studies found minimal changes in trunk responses to perturbations between repeated sessions 1–3 days apart (Griffioen et al. 2016; Pasma et al. 2016; Reeves et al. 2014). The different findings may be due to the test structure. In the studies showing minimal differences across sessions, posture sway was elicited through external perturbations such that linear input-output processes dominated. Also, the perturbations were not designed to elicit sensory reweighting. In sway referencing, there is no external perturbation, and central changes in sensory reweighting should impact posture sway.

To what extent do subjects’ perceptions relate to their performance?

On the basis of subjects’ responses to questions, we found a relation between perception and measured performance, similar to previous studies (Goodworth et al. 2015; McAuley et al. 1985). Because questions were asked after each session, subjects likely had the entire experience in mind when answering each question and may have rated difficulty on a relative scale across test conditions. One interesting correlation related to how the cognitive task influenced posture sway. The strongest correlation was between answers to Q3 (“How much did the cognitive task affect your balance?”) and RMS sway difference between EOcog and EOwhite in sway referencing (Table 1). This correlation was much higher than the same comparison in EC conditions (ECcog minus ECwhite), consistent with the finding that the cognitive task significantly influenced EO sway, but not EC. Because sensory noise was the main parameter that differed in these conditions, one can speculate that perception of difficulty was related to uncertainty in the estimation of body orientation (Lim et al. 2017), which the cognitive task interfered with. Finally, the strong correlation between how self-reported fatigue affected performance and RMS sway across sway referencing conditions suggests that subjects who sway the most were the same subjects who perceived that fatigue influenced them the most. However, it is unclear how subjects determined fatigue (physical, cognitive, or emotional).

Limitations.

The sensitivity analysis showed that some model parameters had redundancy in their influence on amplitude spectra. A wider variety of experimental conditions and analyses, such as the use of external perturbations, electromyography, and impulse response functions, could help better quantify parameters (Goodworth and Peterka 2009, 2012). In addition, it would be important to consider anterior-posterior posture control in future studies, although we anticipate similar results because a previous study generally showed similar changes in trunk control across repeated tests between anatomical planes in unstable sitting (Barbado et al. 2017). Also, we did not tabulate performance on cognitive tasks. Thus it is possible that performance on cognitive tasks was worse in EC trials compared with EO trials and that subjects’ posture sway was similar between ECwhite and ECcog because they simply did not prioritize the cognitive task in EC conditions, where posture sway was largest. Finally, although we found it interesting that the one subject who felt motion sickness had consistently large sway velocity, we are not able to generalize this finding in any way.

Conclusion.

We developed a sway-referenced system in sitting to emphasize the contribution of vestibular and visual contributions to trunk control. Posture sway during sway referencing differed from quiet sitting in both magnitude and frequency distribution. Modeling suggested that sway referencing produced greater uncertainty/noise in sensory feedback and motor outputs and resulted in a reduction in active stiffness and damping gains on sensory cues that orient the body upright. The heightened noise is consistent with optimal integration principles and sensory reweighting toward “noisier” vestibular feedback. The influence of vision and a concurrent cognitive task was much more apparent during sway referencing compared with quiet sitting, supporting the notion that more challenging posture tasks highlight different neuromotor processes. Modeling suggested that sensory and motor noise accounted for most of the visual and cognitive effects within each test session. With this novel sitting task, short-term motor learning was evident in body sway velocity immediately following sway referencing. Longer term motor learning was evident from one week to the next, with 23% decrease in body sway and 9% decrease in sway velocity. These changes in body sway in session 2 occurred predominantly during cognitive tests at lower frequencies and were associated with a lower influence of sensory noise and increased stiffness and integral gains in the model. Perception of difficulty with the cognitive task was consistent with quantitative measures. Taken together, the sitting sway reference test elicited similar neural processes to standing sway referencing and should be a valuable tool to more closely examine sensorimotor control of the trunk.

GRANTS

This work was supported by National Institute of Deafness and Other Communications Disorders Grant R03 DC013858 and a University of Hartford Institute for Translational Research Grant.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.D.G. and S.S. conceived and designed research; A.D.G., K.T., J.L., and T.K. performed experiments; A.D.G., K.T., J.L., T.K., and S.K. analyzed data; A.D.G., S.K., and S.S. interpreted results of experiments; A.D.G. prepared figures; A.D.G. drafted manuscript; A.D.G., K.T., J.L., T.K., S.K., and S.S. edited and revised manuscript; A.D.G., K.T., J.L., T.K., S.K., and S.S. approved final version of manuscript.

APPENDIX: FEEDBACK MODEL SIMULATION AND PARAMETER FITTING DETAILS

The feedback model (Fig. 3) was generated in MATLAB Simulink (The MathWorks). The body was modeled as a single-link pendulum rotating with respect to upright above the axis of rotation of the bench. Two noise sources (torque and sensory) initiated movement of the pendulum (van der Kooij and Peterka 2015). The pendulum was stabilized by corrective torque in the feedback system from time-delayed stiffness, damping, and integral gains (collectively referred to as “neural control parameters”; see methods for more detail).

Neural control parameters and the amplitudes of noise inputs were estimated by minimizing the normalized mean square error (MSE) between the model-derived amplitude spectra and the mean subject experimental amplitude spectra across frequencies 0.04–2.7 Hz (Goodworth and Peterka 2009; Peterka 2002). The noise waveform itself was not part of the fit; only the amplitude (or sensitivity of the system to the input) was included in the fit. To fit the model to experimental data, the minimization was performed with the “fmincon” function in MATLAB. The fmincon function iterated through hundreds to thousands of different combinations of parameters and noise amplitude values until a minimum was reached. For each iteration, a 200-s time series output was simulated in MATLAB Simulink and then divided into 8 separate 25-s cycles. To focus on steady-state behavior, the first cycle was not included. Each subsequent 25-s cycle was converted to an amplitude spectrum using the discrete Fourier time series averaged across adjacent frequencies, identical to experimental analyses. The average amplitude spectrum across the cycles was then defined as the model-derived amplitude spectrum for that particular iteration.

The fmincon function requires the user to define upper and lower limits on parameters and to provide an initial guess for each parameter. Poorly chosen bounds or initial guesses can cause the optimization routine to converge to a local minimum that does not accurately describe the true system, which was recently demonstrated in a two-segment modeling study (Goodworth and Peterka 2018). In the current study, reasonable initial guesses and bounds were identified through the sensitivity analysis, previous studies (e.g., defining a reasonable range of possible time delays associated with trunk posture control), biomechanical constraints (e.g., minimum stiffness of mass times gravity times center of mass height), and preliminary modeling of data where extensive initial guesses and upper and lower bounds were explored and examined for consistency, quality of fits, and global minima. Table A1 shows the approximate first round of upper and lower bounds and initial guesses used initially in all fits to experimental data. Parameters and MSEs were evaluated after every fit. If MSEs were large or if parameters converged to upper or lower bounds, then the fit was repeated with multiple initial guesses and/or different bounds.

We also considered the use of identical upper and lower bounds for quiet sitting and sway referencing. This resulted in more local minimums and poorer fits, but for parameters associated with good fits (low MSEs) that did not converge to upper or lower bounds, results were similar to those presented in results. When poorly fit parameters were refit with multiple initial guesses and different lower and upper bounds, the parameters also converged to values similar to those reported in results.

To examine robustness in model fitting, we also determined how well the fitting routine would converge to a known set of parameter values. First, a nominal set of parameters was selected for sway referencing and quiet sitting. We then simulated an amplitude spectrum using one of the sensory and motor noise waveforms used in the present study. Next, using fitting procedures identical to those in the current study, we estimated parameters in a model fit to the simulated amplitude spectrum. With a reasonable initial guess (equal to the mean values shown in Fig. 6), parameters converged nearly perfectly to the known system in both quiet sitting and sway referencing. When random (sometimes unreasonable) initial guesses were used, we found that parameters converged to the known system when MSEs were lowest and that higher MSEs were associated with incorrect parameters estimates that often converged to upper and lower bounds. Thus, as noted above, in fitting experimental data, both MSEs and parameter values were closely examined to ensure that global minima were approximated in the fitting routine and that optimal parameters were estimated.

Table A1.

Approximate parameter bounds in fitting routine and initial guesses for quiet sitting and sway referencing

Parameter Lower Bound Upper Bound Initial Guess
Kp, N·m·rad−1 180, 180 350, 500 300, 400
Kd, N·m·rad−1·s 40, 40 130, 130 80, 80
Ki, N·m·rad−1·s−1 0, 0 70, 180 1, 70
τ, s 0.04, 0.04 0.12, 0.12 0.07, 0.07
Sensory noise amplitude 0 0.002, 0.03 0.001, 0.015
Motor noise amplitude 0 300, 1,000 150, 400

Data are approximate parameter bounds in fitting routine and initial guesses for quiet sitting (1st value) and sway referencing (2nd value). Note that sensory and motor noise amplitudes in this table differ in magnitude from the converged values in Figs. 6 and 8, which show parameters normalized by the root mean square of the noise waveform. Kp, stiffness; Kd, damping; Ki, integral gain; τ, time delay.

REFERENCES

  1. Alahmari KA, Marchetti GF, Sparto PJ, Furman JM, Whitney SL. Estimating postural control with the balance rehabilitation unit: measurement consistency, accuracy, validity, and comparison with dynamic posturography. Arch Phys Med Rehabil 95: 65–73, 2014. doi: 10.1016/j.apmr.2013.09.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Andreopoulou G, Maaswinkel E, Cofré Lizama LE, van Dieën JH. Effects of support surface stability on feedback control of trunk posture. Exp Brain Res 233: 1079–1087, 2015. doi: 10.1007/s00221-014-4185-5. [DOI] [PubMed] [Google Scholar]
  3. Barbado D, Moreside J, Vera-Garcia FJ. Reliability and repetition effect of the center of pressure and kinematics parameters that characterize trunk postural control during unstable sitting test. PM R 9: 219–230, 2017. doi: 10.1016/j.pmrj.2016.08.029. [DOI] [PubMed] [Google Scholar]
  4. Barra J, Bray A, Sahni V, Golding JF, Gresty MA. Increasing cognitive load with increasing balance challenge: recipe for catastrophe. Exp Brain Res 174: 734–745, 2006. doi: 10.1007/s00221-006-0519-2. [DOI] [PubMed] [Google Scholar]
  5. Boonstra TA, Schouten AC, van der Kooij H. Identification of the contribution of the ankle and hip joints to multi-segmental balance control. J Neuroeng Rehabil 10: 23, 2013. doi: 10.1186/1743-0003-10-23. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Brown SH, McGill SM. The intrinsic stiffness of the in vivo lumbar spine in response to quick releases: implications for reflexive requirements. J Electromyogr Kinesiol 19: 727–736, 2009. doi: 10.1016/j.jelekin.2008.04.009. [DOI] [PubMed] [Google Scholar]
  7. Cenciarini M, Peterka RJ. Stimulus-dependent changes in the vestibular contribution to human postural control. J Neurophysiol 95: 2733–2750, 2006. doi: 10.1152/jn.00856.2004. [DOI] [PubMed] [Google Scholar]
  8. Cholewicki J, Polzhofer GK, Radebold A. Postural control of trunk during unstable sitting. J Biomech 33: 1733–1737, 2000. doi: 10.1016/S0021-9290(00)00126-3. [DOI] [PubMed] [Google Scholar]
  9. Clark S, Riley MA. Multisensory information for postural control: sway-referencing gain shapes center of pressure variability and temporal dynamics. Exp Brain Res 176: 299–310, 2007. doi: 10.1007/s00221-006-0620-6. [DOI] [PubMed] [Google Scholar]
  10. Cohen H, Heaton LG, Congdon SL, Jenkins HA. Changes in sensory organization test scores with age. Age Ageing 25: 39–44, 1996. doi: 10.1093/ageing/25.1.39. [DOI] [PubMed] [Google Scholar]
  11. de Leva P. Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters. J Biomech 29: 1223–1230, 1996. doi: 10.1016/0021-9290(95)00178-6. [DOI] [PubMed] [Google Scholar]
  12. de Oliveira CB, de Medeiros IR, Frota NA, Greters ME, Conforto AB. Balance control in hemiparetic stroke patients: main tools for evaluation. J Rehabil Res Dev 45: 1215–1226, 2008. doi: 10.1682/JRRD.2007.09.0150. [DOI] [PubMed] [Google Scholar]
  13. Edwards S, Austin AP, Bird SP. The role of the trunk control in athletic performance of a reactive change-of-direction task. J Strength Cond Res 31: 126–139, 2017. doi: 10.1519/JSC.0000000000001488. [DOI] [PubMed] [Google Scholar]
  14. Ernst MO, Banks MS. Humans integrate visual and haptic information in a statistically optimal fashion. Nature 415: 429–433, 2002. doi: 10.1038/415429a. [DOI] [PubMed] [Google Scholar]
  15. Faisal AA, Selen LPJ, Wolpert DM. Noise in the nervous system. Nat Rev Neurosci 9: 292–303, 2008. doi: 10.1038/nrn2258. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Fitzpatrick R, McCloskey DI. Proprioceptive, visual and vestibular thresholds for the perception of sway during standing in humans. J Physiol 478: 173–186, 1994. doi: 10.1113/jphysiol.1994.sp020240. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Goodworth A, Perrone K, Pillsbury M, Yargeau M. Effects of visual focus and gait speed on walking balance in the frontal plane. Hum Mov Sci 42: 15–26, 2015. doi: 10.1016/j.humov.2015.04.004. [DOI] [PubMed] [Google Scholar]
  18. Goodworth AD, Peterka RJ. Contribution of sensorimotor integration to spinal stabilization in humans. J Neurophysiol 102: 496–512, 2009. doi: 10.1152/jn.00118.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Goodworth AD, Peterka RJ. Influence of frontal plane stance width on sensory reweighting and coordination in human balance control. J Neurophysiol 104: 1103–1118, 2010a. doi: 10.1152/jn.00916.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Goodworth AD, Peterka RJ. Influence of bilateral vestibular loss on spinal stabilization in humans. J Neurophysiol 103: 1978–1987, 2010b. doi: 10.1152/jn.01064.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Goodworth AD, Peterka RJ. Sensorimotor integration for multisegmental frontal plane balance control in humans. J Neurophysiol 107: 12–28, 2012. doi: 10.1152/jn.00670.2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Goodworth AD, Peterka RJ. Identifying mechanisms of stance control: a single stimulus multiple output model-fit approach. J Neurosci Methods 296: 44–56, 2018. doi: 10.1016/j.jneumeth.2017.12.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Goodworth AD, Wu YH, Felmlee D, Dunklebarger E, Saavedra S. A trunk support system to identify posture control mechanisms in populations lacking independent sitting. IEEE Trans Neural Syst Rehabil Eng 25: 22–30, 2017. doi: 10.1109/TNSRE.2016.2541021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Griffioen M, Maaswinkel E, Zuurmond WW, van Dieën JH, Perez RS. Trunk stabilization estimated using pseudorandom force perturbations, a reliability study. J Biomech 49: 244–251, 2016. doi: 10.1016/j.jbiomech.2015.12.007. [DOI] [PubMed] [Google Scholar]
  25. Hafström A, Fransson PA, Karlberg M, Ledin T, Magnusson M. Visual influence on postural control, with and without visual motion feedback. Acta Otolaryngol 122: 392–397, 2002. doi: 10.1080/00016480260000076. [DOI] [PubMed] [Google Scholar]
  26. Harel NY, Asselin PK, Fineberg DB, Pisano TJ, Bauman WA, Spungen AM. Adaptation of computerized posturography to assess seated balance in persons with spinal cord injury. J Spinal Cord Med 36: 127–133, 2013. doi: 10.1179/2045772312Y.0000000053. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Horak FB, Dickstein R, Peterka RJ. Diabetic neuropathy and surface sway-referencing disrupt somatosensory information for postural stability in stance. Somatosens Mot Res 19: 316–326, 2002. doi: 10.1080/0899022021000037782. [DOI] [PubMed] [Google Scholar]
  28. Horak FB, Henry SM, Shumway-Cook A. Postural perturbations: new insights for treatment of balance disorders. Phys Ther 77: 517–533, 1997. doi: 10.1093/ptj/77.5.517. [DOI] [PubMed] [Google Scholar]
  29. Hwang S, Agada P, Grill S, Kiemel T, Jeka JJ. A central processing sensory deficit with Parkinson’s disease. Exp Brain Res 234: 2369–2379, 2016. doi: 10.1007/s00221-016-4642-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Karthikbabu S, Chakrapani M, Ganeshan S, Rakshith KC, Nafeez S, Prem V. A review on assessment and treatment of the trunk in stroke: a need or luxury. Neural Regen Res 7: 1974–1977, 2012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Kuo AD. An optimal state estimation model of sensory integration in human postural balance. J Neural Eng 2: S235–S249, 2005. doi: 10.1088/1741-2560/2/3/S07. [DOI] [PubMed] [Google Scholar]
  32. Lim K, Karmali F, Nicoucar K, Merfeld DM. Perceptual precision of passive body tilt is consistent with statistically optimal cue integration. J Neurophysiol 117: 2037–2052, 2017. doi: 10.1152/jn.00073.2016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Maaswinkel E, Griffioen M, Perez RS, van Dieën JH. Methods for assessment of trunk stabilization, a systematic review. J Electromyogr Kinesiol 26: 18–35, 2016. doi: 10.1016/j.jelekin.2015.12.010. [DOI] [PubMed] [Google Scholar]
  34. Maaswinkel E, van Drunen P, Veeger DJ, van Dieën JH. Effects of vision and lumbar posture on trunk neuromuscular control. J Biomech 48: 298–303, 2015. doi: 10.1016/j.jbiomech.2014.11.030. [DOI] [PubMed] [Google Scholar]
  35. Maetzler W, Mancini M, Liepelt-Scarfone I, Müller K, Becker C, van Lummel RC, Ainsworth E, Hobert M, Streffer J, Berg D, Chiari L. Impaired trunk stability in individuals at high risk for Parkinson’s disease. PLoS One 7: e32240, 2012. doi: 10.1371/journal.pone.0032240. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Mahboobin A, Loughlin P, Atkeson C, Redfern M. A mechanism for sensory re-weighting in postural control. Med Biol Eng Comput 47: 921–929, 2009. doi: 10.1007/s11517-009-0477-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Maurer C, Peterka RJ. A new interpretation of spontaneous sway measures based on a simple model of human postural control. J Neurophysiol 93: 189–200, 2005. doi: 10.1152/jn.00221.2004. [DOI] [PubMed] [Google Scholar]
  38. McAndrew PM, Dingwell JB, Wilken JM. Walking variability during continuous pseudo-random oscillations of the support surface and visual field. J Biomech 43: 1470–1475, 2010. doi: 10.1016/j.jbiomech.2010.02.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. McAuley E, Duda J, Fujita A, Gauvin L, Halliwell W, Hanin YL, Hatfield BD, Horn T, Qi WM, Spink K, Weiss M, Yukelson D. Success and causality in sport: the influence of perception. J Sport Psychol 7: 13–22, 1985. doi: 10.1123/jsp.7.1.13. [DOI] [Google Scholar]
  40. Mokhtarinia HR, Sanjari MA, Chehrehrazi M, Kahrizi S, Parnianpour M. Trunk coordination in healthy and chronic nonspecific low back pain subjects during repetitive flexion-extension tasks: effects of movement asymmetry, velocity and load. Hum Mov Sci 45: 182–192, 2016. doi: 10.1016/j.humov.2015.11.007. [DOI] [PubMed] [Google Scholar]
  41. Mujdeci B, Turkyilmaz D, Yagcioglu S, Aksoy S. The effects of concurrent cognitive tasks on postural sway in healthy subjects. Rev Bras Otorrinolaringol (Engl Ed) 82: 3–10, 2016. doi: 10.1016/j.bjorl.2015.10.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Nashner LM, Black FO, Wall C 3rd. Adaptation to altered support and visual conditions during stance: patients with vestibular deficits. J Neurosci 2: 536–544, 1982. doi: 10.1523/JNEUROSCI.02-05-00536.1982. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Oomen NM, Reeves NP, Priess MC, van Dieën JH. Trunk muscle coactivation is tuned to changes in task dynamics to improve responsiveness in a seated balance task. J Electromyogr Kinesiol 25: 765–772, 2015. doi: 10.1016/j.jelekin.2015.07.001. [DOI] [PubMed] [Google Scholar]
  44. Pasma JH, Engelhart D, Maier AB, Aarts RG, van Gerven JM, Arendzen JH, Schouten AC, Meskers CG, van der Kooij H. Reliability of system identification techniques to assess standing balance healthy elderly. PLoS One 11: e0151012, 2016. doi: 10.1371/journal.pone.0151012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Pellecchia GL. Postural sway increases with attentional demands of concurrent cognitive task. Gait Posture 18: 29–34, 2003. doi: 10.1016/S0966-6362(02)00138-8. [DOI] [PubMed] [Google Scholar]
  46. Perugini A, Ditterich J, Basso MA. Patients with Parkinson’s disease show impaired use of priors in conditions of sensory uncertainty. Curr Biol 26: 1902–1910, 2016. doi: 10.1016/j.cub.2016.05.039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Peterka RJ. Sensorimotor integration in human postural control. J Neurophysiol 88: 1097–1118, 2002. doi: 10.1152/jn.2002.88.3.1097. [DOI] [PubMed] [Google Scholar]
  48. Peterka RJ, Loughlin PJ. Dynamic regulation of sensorimotor integration in human postural control. J Neurophysiol 91: 410–423, 2004. doi: 10.1152/jn.00516.2003. [DOI] [PubMed] [Google Scholar]
  49. Pintelon R, Schoukens J. System Identification: A Frequency Domain Approach. Piscataway, NJ: IEEE, 2001. doi: 10.1002/0471723134. [DOI] [Google Scholar]
  50. Qu X, Nussbaum MA, Madigan ML. Model-based assessments of the effects of age and ankle fatigue on the control of upright posture in humans. Gait Posture 30: 518–522, 2009. doi: 10.1016/j.gaitpost.2009.07.127. [DOI] [PubMed] [Google Scholar]
  51. Rachwani J, Santamaria V, Saavedra SL, Woollacott MH. The development of trunk control and its relation to reaching in infancy: a longitudinal study. Front Hum Neurosci 9: 94, 2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Rankin JK, Woollacott MH, Shumway-Cook A, Brown LA. Cognitive influence on postural stability: a neuromuscular analysis in young and older adults. J Gerontol A Biol Sci Med Sci 55: M112–M119, 2000. doi: 10.1093/gerona/55.3.M112. [DOI] [PubMed] [Google Scholar]
  53. Redfern MS, Talkowski ME, Jennings JR, Furman JM. Cognitive influences in postural control of patients with unilateral vestibular loss. Gait Posture 19: 105–114, 2004. doi: 10.1016/S0966-6362(03)00032-8. [DOI] [PubMed] [Google Scholar]
  54. Reeves NP, Everding VQ, Cholewicki J, Morrisette DC. The effects of trunk stiffness on postural control during unstable seated balance. Exp Brain Res 174: 694–700, 2006. doi: 10.1007/s00221-006-0516-5. [DOI] [PubMed] [Google Scholar]
  55. Reeves NP, Popovich JM Jr, Priess MC, Cholewicki J, Choi J, Radcliffe CJ. Reliability of assessing trunk motor control using position and force tracking and stabilization tasks. J Biomech 47: 44–49, 2014. doi: 10.1016/j.jbiomech.2013.10.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Resch JE, May B, Tomporowski PD, Ferrara MS. Balance performance with a cognitive task: a continuation of the dual-task testing paradigm. J Athl Train 46: 170–175, 2011. doi: 10.4085/1062-6050-46.2.170. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Saavedra SL, Woollacott MH. Segmental contributions to trunk control in children with moderate-to-severe cerebral palsy. Arch Phys Med Rehabil 96: 1088–1097, 2015. doi: 10.1016/j.apmr.2015.01.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Seelen HA, Potten YJ, Huson A, Spaans F, Reulen JP. Impaired balance control in paraplegic subjects. J Electromyogr Kinesiol 7: 149–160, 1997. doi: 10.1016/S1050-6411(97)88884-0. [DOI] [PubMed] [Google Scholar]
  59. Silsupadol P, Shumway-Cook A, Lugade V, van Donkelaar P, Chou LS, Mayr U, Woollacott MH. Effects of single-task versus dual-task training on balance performance in older adults: a double-blind, randomized controlled trial. Arch Phys Med Rehabil 90: 381–387, 2009. doi: 10.1016/j.apmr.2008.09.559. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Sung W, Abraham M, Plastaras C, Silfies SP. Trunk motor control deficits in acute and subacute low back pain are not associated with pain or fear of movement. Spine J 15: 1772–1782, 2015. doi: 10.1016/j.spinee.2015.04.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Teng HL, Powers CM. Sagittal plane trunk posture influences patellofemoral joint stress during running. J Orthop Sports Phys Ther 44: 785–792, 2014. doi: 10.2519/jospt.2014.5249. [DOI] [PubMed] [Google Scholar]
  62. Tjernström F, Fransson PA, Hafström A, Magnusson M. Adaptation of postural control to perturbations–a process that initiates long-term motor memory. Gait Posture 15: 75–82, 2002. doi: 10.1016/S0966-6362(01)00175-8. [DOI] [PubMed] [Google Scholar]
  63. van Beers RJ, Sittig AC, Gon JJ. Integration of proprioceptive and visual position-information: An experimentally supported model. J Neurophysiol 81: 1355–1364, 1999. doi: 10.1152/jn.1999.81.3.1355. [DOI] [PubMed] [Google Scholar]
  64. van der Kooij H, Jacobs R, Koopman B, van der Helm F. An adaptive model of sensory integration in a dynamic environment applied to human stance control. Biol Cybern 84: 103–115, 2001. doi: 10.1007/s004220000196. [DOI] [PubMed] [Google Scholar]
  65. van der Kooij H, Peterka RJ. Non-linear stimulus-response behavior of the human stance control system is predicted by optimization of a system with sensory and motor noise. J Comput Neurosci 30: 759–778, 2011. doi: 10.1007/s10827-010-0291-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. van Drunen P, Koumans Y, van der Helm FC, van Dieën JH, Happee R. Modulation of intrinsic and reflexive contributions to low-back stabilization due to vision, task instruction, and perturbation bandwidth. Exp Brain Res 233: 735–749, 2015. doi: 10.1007/s00221-014-4151-2. [DOI] [PubMed] [Google Scholar]
  67. Van Ooteghem K, Frank JS, Horak FB. Practice-related improvements in posture control differ between young and older adults exposed to continuous, variable amplitude oscillations of the support surface. Exp Brain Res 199: 185–193, 2009. doi: 10.1007/s00221-009-1995-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Willigenburg NW, Kingma I, Hoozemans MJ, van Dieën JH. Precision control of trunk movement in low back pain patients. Hum Mov Sci 32: 228–239, 2013. doi: 10.1016/j.humov.2012.12.007. [DOI] [PubMed] [Google Scholar]
  69. Winter DA. Biomechanics and Motor Control of Human Movement. Hoboken, NJ: Wiley, 2005. [Google Scholar]
  70. Wrisley DM, Stephens MJ, Mosley S, Wojnowski A, Duffy J, Burkard R. Learning effects of repetitive administrations of the sensory organization test in healthy young adults. Arch Phys Med Rehabil 88: 1049–1054, 2007. doi: 10.1016/j.apmr.2007.05.003. [DOI] [PubMed] [Google Scholar]
  71. Wu YH, Duncan K, Saavedra S, Goodworth A. Segmental trunk and head dynamics during frontal plane tilt stimuli in healthy sitting adults. J Biomech 49: 2831–2837, 2016. doi: 10.1016/j.jbiomech.2016.06.023. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Yardley L, Gardner M, Bronstein A, Davies R, Buckwell D, Luxon L. Interference between postural control and mental task performance in patients with vestibular disorder and healthy controls. J Neurol Neurosurg Psychiatry 71: 48–52, 2001. doi: 10.1136/jnnp.71.1.48. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Journal of Neurophysiology are provided here courtesy of American Physiological Society

RESOURCES