Postural Complexity during Listening in Young and Middle-Aged Adults

Postural behavior has traditionally been studied using linear assessments of stability (e.g., center of pressure ellipse area). While these assessments may provide valuable information, they neglect the nonlinear nature of the postural system and often lead to the conflation of variability with pathology. Moreover, assessing postural behavior in isolation or under otherwise unrealistic conditions may obscure the natural dynamics of the postural system. Alternatively, assessing postural complexity during ecologically valid tasks (e.g., conversing with others) may provide unique insight into the natural dynamics of the postural system across a wide array of temporal scales. Here, we assess postural complexity using Multiscale Sample Entropy in young and middle-aged adults during a listening task of varying degrees of difficulty. It was found that middle-aged adults exhibited greater postural complexity than did young adults, and that this age-related difference in postural complexity increased as a function of task difficulty. These results are inconsistent with the notion that aging is universally associated with a loss of complexity, and instead support the notion that age-related differences in complexity are task dependent.


Introduction
How might aging manifest in the dynamics of biological systems? One hypothesis that has received much attention in recent years is the loss of complexity hypothesis, which claims that the loss of function associated with aging is the result of a loss of complexity in the underlying physiological systems [1]. As aging is associated with an increase in the incidence and severity of falls, the complexity of the postural system is of particular relevance [2]. Moreover, as individuals rarely perform postural tasks in the absence of an additional task (e.g., conversing with others or manipulating nearby objects), this postural complexity should be measured under ecologically valid, multi-task conditions.
Traditionally, however, research on postural behavior has not had as its focus the complexity of the underlying physiological systems. Rather, research has focused primarily on quantifying the stability of the postural system, often during laboratory-specific paradigms. In assessing stability, linear assessments of the center of pressure (e.g., ellipse area) and relations of the center of mass to the base of support (e.g., proximity) have been widely studied [3]. While it is true that these measures may provide valuable information in certain contexts, inherent assumptions of linearity restrict the breadth and depth of information that might be gleaned. When measuring the ellipse area of the center of pressure, for example, the traditionally dominant perspective is that an increase in variability is alone indicative of a pathological system. Such a perspective ignores the nonlinear nature of the postural system and conflates variability with pathology [4].
A more modern approach is to study the response of the postural system to mechanical perturbation [5,6]. This approach is particularly attractive to those who wish to study the is able to adapt to increased perceptual demands might be gained. Moreover, assessing the complexity of postural behavior across physiologically relevant timescales (e.g., those associated with physiological tremor) might serve to further elucidate age-related changes in complexity. It is expected that the complexity exhibited by the young adults will be greater than the complexity exhibited by the middle-aged adults, and that this difference will increase as a function of task difficulty.

Participants
As part of a larger study, 16 young adults (18 to 28 years of age; mean 22 years) and 16 middle-aged adults (48 to 64 years of age; mean 58 years) were recruited, with both groups being comprised of 11 females and 5 males. Note that data from one young participant and two middle-aged participants were corrupted during collection, resulting in groups of 15 and 14 being used for the present study. The young participants all had clinically normal hearing (pure-tone thresholds <20 dB HL from 250 to 8000 Hz), and all but two of the middle-aged participants had clinically normal hearing (pure-tone thresholds <30 dB HL) from 250 to 4000 Hz. Participants were recruited from the general population and were screened for visual, vestibular, and motor impairments prior to collection. For a full description of participant characteristics, see Helfer et al. [11]. All procedures were approved by the University of Massachusetts Amherst Institutional Review Board (IRB).

Procedures
Participants stood with their feet shoulder-width apart on a 40 × 60 cm piezoelectric force platform (Kistler Instruments Corporation, Amherst, NY, USA) while listening to and repeating pre-recorded sentences. These sentences were designed to have low predictability while being grammatically feasible (e.g., Theo found the pink menu and the true item here). During the listening task, a loudspeaker was placed 1.2 m in front of the participants and was adjusted to the ear height of each participant. The loudspeaker simultaneously played both a target sentence (i.e., a sentence which the participants were instructed to repeat) as well as two masking sentences (i.e., sentences similar to the target sentence which the participants were instructed to ignore). These masking sentences were played from random starting points as to minimize any grammatical alignment of the masking and target sentences.
For the present study, three conditions were analyzed. The first condition consisted of participants standing on the force platform without any listening task. This condition, referred to as the baseline condition, served to provide the experimenters with the baseline postural dynamics of each participant. The remaining two conditions consisted of participants standing on the force platform while performing the listening task described above. The difficulty of the listening task was modified for these two conditions by adjusting the signal-to-noise ratio (SNR) of the stimuli. In the first of these remaining conditions, the 0 dB condition, the combined energy of the masking sentences was equal to that of the target sentence. In the final condition, the −6 dB condition, the combined energy of the masking sentences was 6 dB greater than that of the target sentence. The order of conditions was randomized for each participant, with each condition lasting approximately 80 s.
Multiscale Sample Entropy was computed for the CoP data in the following manner ( Figure 1). First, the original CoP time series was coarse-grained according to the following equation: where y j is a sample of the coarse-grained signal, τ is the scale, and x i is a sample of the original signal. This procedure was repeated for a total of 40 scales, with each scale greater than one yielding a new, coarse-grained signal. Sample Entropy was then computed for each of the 40 scales by the following equation: where m is the number of samples being compared across segments of the signal, r is the radius of similarity, and N is the length of the time series. Here, Φ m (r) is the probability that the segments will be sufficiently similar when comprised of m samples, and Φ m+1 (r) is the probability that the segments will be sufficiently similar when comprised of m + 1 samples. Note that in the present study, r was set to 0.15 times the standard deviation of the signal and m was set to 2 [12]. the original signal is coarse-grained into a given number of signals by Equation (1). Then, Sample Entropy is computed for each of these coarse-grained signals by Equation (2). Complexity Indices are then computed by taking the sum of the Sample Entropy values within a given range of scales. Here, the high-frequency Complexity Index corresponds to 8-12 Hz (scales 8 to 12) and the low-frequency Complexity Index corresponds to 2.5-6 Hz (scales 17 to 40). Note that the area under the entire curve represents the overall Complexity Index, corresponding to 2.5-100 Hz.
After computing the Sample Entropy for each scale, Complexity Indices were computed by taking the sum of the Sample Entropy values within a given range of scales, as defined by the following equation: where S E is the Sample Entropy for scale τ and the Complexity Index is comprised of scales b through n. Here, three Complexity Indices were selected based on their corresponding timescales: an overall Complexity Index comprised of scales 1 through 40 (corresponding to 2.5-100 Hz), a low-frequency Complexity Index comprised of scales 17 to 40 (corresponding to 2.5-6 Hz), and a high-frequency Complexity Index comprised of scales 8 to 12 (corresponding to [8][9][10][11][12]. Note that frequencies of 2.5-6 Hz are typically associated with voluntary movement, while frequencies of 8-12 Hz are typically associated with involuntary movement (e.g., physiological tremor).

Statistical Analysis
Two-way repeated measures analyses of variance (ANOVAs) with condition and age as within-and between-subjects factors, respectively, were performed for the three Complexity Indices (overall, low-frequency, and high-frequency) and for the two CoP directions (anteroposterior (A/P) and mediolateral (M/L)). Greenhouse-Geisser corrections were applied in any instance where the assumption of sphericity was violated (Mauchly's W; p < 0.05). Post hoc pairwise comparisons were performed using t-tests with Bonferroni corrections. Effect sizes were reported in partial eta-squared (η 2 p ) (0.01 = small; 0.06 = medium; 0.14 = large) for the ANOVAs and in Cohen's d (0.20 = small; 0.50 = moderate; 0.80 = large) for the post hoc comparisons. A significance threshold of α = 0.05 was used for all inferential statistics. All statistical analyses were performed in JASP (JASP, Amsterdam, The Netherlands).

Low-Frequency Complexity Index (M/L)
Main effects of condition (p = 0.017; η 2 p = 0.140) and age (p = 0.002; η 2 p = 0.297) were found for M/L postural complexity in the 2.5-6 Hz range, with postural complexity being greater in the middle-aged participants than in the young participants (Tables A13 and A14; Figure 3). Post hoc analysis revealed that postural complexity was greater in the baseline condition than in the −6 dB condition (p = 0.015; d = 0.557), with no significant difference existing between the baseline and 0 dB conditions (p = 0.921; d = 0.197) or between the 0 dB and −6 dB conditions (p = 0.192; d = 0.361) (Table A15; Figure 3). While no significant condition × age interaction was found (p = 0.068; η 2 p = 0.095), post hoc analysis revealed significant age-related differences in postural complexity in the −6 dB condition (p = 0.002; d = 1.503), but not in the 0 dB (p = 1.000; d = 0.607) or baseline (p = 0.239; d = 0.924) conditions (Tables A13 and A16; Figure 3).

Discussion
The present study assessed the complexity of postural behavior exhibited by young and middle-aged participants during a listening task of varying degrees of difficulty. Consistent with the loss of complexity hypothesis, it was expected that the young participants would exhibit greater postural complexity than the middle-aged participants, and that this difference would increase as a function of task difficulty. In comparing the postural complexity exhibited by the two groups of participants, however, the opposite result was observed; the middle-aged participants exhibited greater postural complexity across all Complexity Indices and across both CoP directions. Moreover, this difference appeared to increase as a function of task difficulty, with significant age-related differences in postural complexity emerging consistently in the −6 dB condition while not being present in the baseline or 0 dB conditions. While these results were not expected, they are not entirely without precedent. In 2007, Costa et al. [13] found that postural complexity did not differ significantly between young and older individuals during quiet standing, provided the individuals in the latter group were of sufficient health (i.e., had no history of falling). Rather, it was found that healthy young and healthy older individuals both exhibited greater postural complexity than did older individuals with a history of falling. As the middle-aged participants in the present study were screened for visual, vestibular, and motor impairments, it is reasonable that no significant age-related differences in postural complexity were observed in the baseline or 0 dB conditions. Indeed, these participants were first recruited by Helfer et al. [11] in an effort to investigate age-related changes in postural control and speech perception that may appear despite a lack of associated impairments.
Additionally, Duarte and Sternad [14] found that older individuals exhibited greater postural complexity than did young adults during prolonged standing. Importantly, however, these age-related differences were found to depend upon the radius of similarity, r, expressed in Equation (2). As prolonged standing is accompanied by natural postural adjustments (e.g., shifting one's weight), the resulting signal often contains a substantial number of outliers. If one group performs more frequent postural adjustments, these outliers may lead to differences in signal complexity, as r is typically multiplied by the standard deviation of the signal. When Duarte and Sternad corrected for these outliers by using a fixed r value, these age-related differences in postural complexity were no longer found.
Does such a finding by Duarte and Sternad suggest that the age-related differences in the present study are similarly dependent upon the radius of similarity (r)? First, one must consider the nature of the postural tasks in question. While the prolonged standing task implemented by Duarte and Sternad consisted of unconstrained standing lasting thirty minutes, the postural task in the present study lasted only eighty seconds. It is therefore unlikely that the postural data analyzed in the present study contained a similar number of outliers. Moreover, the standard deviations in the present study were, on average, greater in the middle-aged participants [11]. As a greater standard deviation increases the radius of similarity, using a fixed r value in the present study would serve only to increase the relative degree of postural complexity observed in the middle-aged participants.
How, then, might one reconcile the nature of these differences with the existing literature on postural complexity and aging? As significant age-related differences in postural complexity emerged consistently in the −6 dB condition, it is possible that the two groups of participants adopted two distinct strategies in response to the listening task and that these strategies became distinguishable only under more difficult acoustic conditions. If, for example, maintaining upright posture while engaging with the listening task proved more difficult for the middle-aged participants, these participants may have been unable to incur any reduction in the degrees of freedom being used for postural control [15]. In contrast, the young participants may have prioritized the perception of auditory information, even if this prioritization resulted in such a reduction in the degrees of freedom. This explanation is further supported by the finding that the postural complexity exhibited by the middle-aged participants did not differ between conditions, while the postural complexity exhibited by the young participants appeared to decrease as a function of task difficulty (Figures 2-4).
Returning to the first study conducted by Helfer et al. [11], a moderate reduction in listening task performance was observed between the young and middle-aged participants. While this reduction may have been the result of early age-related changes in hearing, it is consistent with the notion that individuals in the latter group may have been unable to prioritize the perception of auditory information. Helfer et al. also reported a moderate increase in the 95% confidence ellipse of the CoP between the young and middle-aged participants during the listening task, further suggesting that maintaining upright posture while engaging with the listening task was more difficult for the middle-aged participants. These results, when combined with those of the present study, suggest that early aging may be accompanied by a decrease in the ability of individuals to regulate their CoP while conversing with others.
One additional finding is that the age-related differences in postural complexity appeared consistent across the three Complexity Indices. Within each Complexity Index, middle-aged participants exhibited significantly greater postural complexity in both the A/P and M/L directions than did young adults in the −6 dB condition (Figures 2-4). Likewise, moderate-to-large effect sizes were found within each Complexity Index in the 0 dB condition, suggesting that the middle-aged participants exhibited greater A/P postural complexity than did the young participants, though these differences did not reach statistical significance. Note that direct comparison of postural complexity between Complexity Indices is not possible given the different number of scales within each Complexity Index (e.g., the overall Complexity Index is comprised of 40 scales while the high-frequency Complexity Index is comprised of only five scales).
While the results presented here provide unique insight into the dynamics of the postural system during listening, there are several limitations that should be noted. Firstly, participants in the present study were restricted to quiet standing, which may not be reflective of the multitude of postures individuals adopt while conversing. Secondly, the physical activity levels of the participants were not assessed despite these levels being potentially relevant to the interpretation of our results. Lastly, as noted by Helfer et al. [11], the baseline condition was measured in relative silence, which may have altered postural behavior by rendering participants unable to orient themselves towards a reference signal in the surrounding environment [16].
Overall, the results of the present study do not support the loss of complexity hypothesis put forth by Lipsitz and Goldberger [1] but rather the notion that age-related differences in complexity are task dependent [15]. Specifically, it was found that the postural complexity exhibited by young and middle-aged adults may not differ significantly during quiet standing if the individuals in the latter group are of sufficient health. Moreover, it was found that middle-aged adults may exhibit greater postural complexity than young adults when quiet standing is coupled with a perceptual task. These results should serve to caution against universally associating the degree of complexity exhibited by a physiological system with the functional capacity of that system, as the degree of complexity may differ between tasks and ecological conditions.

Appendix B
A sensitivity analysis was performed to examine the effects of filter selection on Multiscale Sample Entropy (MSE). First, MSE was computed for the CoP data using six filtering approaches: raw (i.e., no filter), a lowpass filter at 15 Hz, a bandpass filter at 0.5-15 Hz, a bandpass filter at 1-15 Hz, a bandpass filter at 2-15 Hz, and a bandpass filter at 1-20 Hz. These results are shown below for the CoP data in the A/P and M/L directions in Figures A1 and A2, respectively.  Once the MSE was computed, Complexity Indices were computed for each of the six filtering techniques across a range of scale factors. Group mean differences were then computed and plotted as a function of maximum scale ( Figure A3). Note that a maximum scale of 40 was used in the present study based on recommendations by Yentes et al. [12] to compute Sample Entropy on no fewer than 200 samples when using the recommended m criterion of 2. As each CoP signal analyzed in the present study contained approximately 8000 samples, the coarse-graining procedure would yield approximately 200 samples per signal at scale 40. Figure A3. The effects of filter selection and maximum scale factor on group mean Complexity Index differences.
In comparing filtering approaches, we selected the 1-15 Hz bandpass filter in order to remove low-frequency nonstationarities as well as to retain only those frequencies which are physiologically feasible [17]. This filtering approach also appeared to maximize age-related differences while yielding consistent trends across CoP directions ( Figure A3).