Do postural responses to transient and continuous perturbations show similar vision and amplitude dependence?

J. Bzomechanics Vol. 26. NO. IO. pp. 1181-1190. Printed in Great Britam

1993.

0021-9290,93 fcJ 1993 Pergamon

$6.00+.00 Press Ltd

DO POSTURAL RESPONSES TO TRANSIENT AND CONTINUOUS PERTURBATIONS SHOW SIMILAR VISION AND AMPLITUDE DEPENDENCE?* B. E. MAKI and G. OSTROVSKI Centre for Studies in Aging, Sunnybrook Health Science Centre, University of Toronto, Toronto, Canada Abstract-This study compares the influence of vision and perturbation amplitude on postural center-ofpressure responses to transient (acceleration pulse) versus continuous (pseudorandom acceleration) platform-translation perturbations. To control possible confounding effects, efforts were directed at making the perturbations unpredictable and matching the transient and continuous waveforms in terms of frequency content and level of challenge to stability. It was hypothesized that, under these conditions, the same control mechanisms would be activated; hence, responses to transient and continuous perturbations would show similar vision and amplitude dependence. A repeated-measures experimental design was used to test 12 healthy males. The transient and continuous perturbations were each administered at three acceleration amplitude levels, in random order, and the tests were repeated under eyes-open and blindfolded conditions. When order-of-testing effects were taken into account, the results failed to demonstrate any differences between transient and continuous postural control with regard to the influence of vision, as both types of response were unaffected by vision deprivation. This finding is consistent with the experimental hypothesis. Responses to the two types of perturbation did appear to differ, however, with regard to the influence of perturbation amplitude, as the transient-perturbation responses showed greater evidence of nonlinearity. This finding would seem to reject the hypothesis that the responses to transient and continuous perturbations would show similar amplitude dependence. However, it is suggested that the amplitude distribution of the particular continuous waveform that was used, rather than the continuous nature of the perturbation per se, may have led to the greater linearity in the continuous-perturbation responses. Although the present results fail to provide unequivocal support for the hypothesis that the same control mechanisms are utilized in transient and continuous balance recovery, they do demonstrate that distinctions between these two types ofpostural task are not nearly so clearcut as has been suggested. Nonetheless, from a practical perspective, the findings of substantial nonlinearity indicate that attempts to predict responses to transient perturbations from continuous-perturbation tests should be performed with caution.

INTRODUCTION Our postural balance is continually challenged by destabilizing perturbations. The posture control mechanisms must respond to sudden transient perturbations, such as slips, trips and pushes, and must also continuously compensate for the intrinsic instability of the musculoskeletal linkage due to gravita-

tional forces. Other forms of continuous perturbations can also arise, e.g. while standing in a moving vehicle. The balance mechanisms that act to maintain an upright posture under these various conditions are often modelled as a feedback control system, in which neural centers utilize visual, vestibular and/or somatosensory feedback about the body orientation and motion to generate stabilizing motor commands (e.g. Werness. 1982; Lee, 1989). It has been suggested, however, that the posture control system may respond differently to transient and continuous perturbations, using feedback control for continuous regulation but Received in$nal form 13 January 1993.

Address correspondence to: Brian Maki, Centre for Studies in Aging, 2075 Bayview Avenue, Toronto, Ontario, M4N 3M5 Canada. *Portions of the work described in this paper were presented in preliminary form at the Tenth International Symposium on Disorders of Posture and Gait, Munich, 1990.

switching, for transient recovery, to a feedforward (preprogrammed) mode where ‘reflex-like responses are . . stereotypically released without feedback control’ (Diener and Dichgans, 1988a). Diener et al. (1986) and Diener and Dichgans (1988a) reported differences in the postural responses to transient and continuous perturbations with regard to the influence of vision, as well as vestibular and somatosensory inputs. However, these findings may have been confounded by other factors, such as (1) perturbation predictability, which may promote anticipatory adaptations to continuous perturbations (Hocherman et al., 1988) and/or central preprogramming of responses to transient perturbations (Horak er al., 1989), (2) differences in the frequency content of the transient and continuous waveforms, which may affect the relative contributions of the visual, vestibular and somatosensory inputs (Nashner, 1981) and (3) differences in the degree to which functional stability (i.e. the ability to maintain balance without moving the feet) was challenged, which could affect postural strategy selection (Horak and Nashner, 1986). The feedbacklfeedforward dichotomy suggested by Diener et al. (1986) and Diener and Dichgans (1988a) predicts that sensory inputs, such as vision, modulate responses to continuous perturbations, but have little effect on the responses to transient perturbations. It

1181

B. E. MAKI and G. OSTROVSKI

1182

can be further predicted, if this dichotomy exists, that scaling of postural responses to perturbations of differing magnitude is likely to differ for transient and continuous perturbations, particularly where unpredictable conditions might preclude accurate preprogramming of responses to transient perturbations. The purpose of the present study was to test these predictions by comparing postural responses to transient versus continuous perturbations with regard to the influence of changes in visual conditions and changes in perturbation amplitude. Conceptually, a continuous perturbation is no more than a sequence of multiple transient perturbations; therefore, we suggest that the same posture control mechanisms (likely to be a hybrid of feedforward and feedback control) will be utilized in dealing with both types of perturbation, provided that the perturbations are truly unpredictable and that other confounding factors are controlled. Specifically, we hypothesize that, under conditions where the two types of perturbation are unpredictable, have similar frequency content and represent similar challenge to stability, responses to transient and continuous perturbations will show similar behavior in terms of vision and amplitude dependence. Delineation of possible differences between transient balance recovery and continuous postural regulation is of importance in understanding the neurophysiology of postural control, but this issue also has clinical implications. If there is, in fact, a feedforward/ feedback dichotomy, then clinical balance testing programs should assess responses to both transient and continuous perturbations. Conversely, the ability to predict transient balance recovery from continuousperturbation tests could have practical advantages, e.g. use of a ‘gentle’ continuous perturbation to test the frail elderly, for whom transient tests can be a safety hazard and may also elicit anxiety that could affect the performance (Maki et al., 1991; Maki and Whitelaw, 1993). From a practical perspective, we anticipated that the findings of this study would help to assess the feasibility of using continuous-perturbation tests to predict transient balance recovery.

METHODS

Twelve healthy males, aged 25 to 45, were tested. A questionnaire was used to screen out subjects with deficits or disorders that might affect postural balance, in particular: (1) history of dizziness or balance/coordination problems; (2) history of neurological, musculoskeletal, visual or vestibular disorders; (3) recent illnesses or injuries; or (4) use of drugs and medications. The perturbations were applied via horizontal anterior-posterior acceleration of a platform on which the subject stood. Mounted on the platform were two custom-built force plates which recorded the location of the center of pressure on the feet to within +_1 mm (error standard deviation). A servo force = balance

accelerometer (Terra Flex Model SA-102) measured the platform acceleration to within +0.0025 m s-’ (frequency response O-50 Hz, +_2 dB). The force plate and accelerometer signals were passed through lowpass anti-aliasing filters (second-order; - 3 dB at 10 Hz) and sampled by a PC(286)-based data acquisition system (Tecmar Labmaster) at a rate of 50 Hz. A visual surround (2 m wide, 2 m long, ceiling height 2.5 m, eye-surround distance approximately 1 m) was mechanically coupled to move with the platform, so that any dynamic visual cues resulted solely from the swaying motion of the subject’s head relative to the platform frame of reference. The white surfaces of the surround were covered with a pattern of markings, so as to provide a wide spatial frequency spectrum, and a visual target (a 15 cm black ‘X centered within a 40 cm x 40 cm frame, on a grey background) was mounted at eye height on the front wall (superimposed over the pattern of markings). The inside corners and edges of the surround were painted black to provide vertical and horizontal visual references. The enclosure was illuminated by four 33 W fluorescent fixtures, which provided an illuminance of 450 lux (measured at eye level). Further details regarding the platform design and performance are provided elsewhere (Maki, 1987; Maki et al., 1987; Maki et a[., 1991). The continuous perturbation had a pseudorandom waveform, composed of a sum of 15 sinusoids (equal in acceleration amplitude) with frequencies ranging from 0.13 to 5.0 Hz. The period of the waveform was 15.36 s and the total duration of each test was 62 s. The platform motion was controlled to start and end gradually, with no sudden changes in acceleration. Three root-mean-square (RMS) acceleration amplitudes were used: 0.09, 0.135 and 0.18 ms-*. The corresponding displacements and RMS velocities were 0.14, 0.21 and 0.28 m and 0.038, 0.057 and 0.076 m s- ‘, respectively. The peak acceleration was 2.3 times the RMS value. The transient perturbation was an approximately square acceleration pulse of duration 200 ms. This was followed by a 600 ms constant-velocity phase (during which the pulse response could be observed), after which the platform motion was stopped by a 200 ms deceleration pulse. Forward transient platform motions were administered at acceleration amplitudes of 0.4, 0.6 and 0.8 ms - 2. The corresponding displacements and peak velocities were 0.064,0.096 and 0.13 m and 0.08,0.12 and 0.16 m s- ‘, respectively. Backward transient motions were administered at acceleration amplitudes of 0.5,0.75 and 1.0 m s-*. The corresponding displacements and peak velocities were 0.08, 0.12 and 0.16 m and 0.1, 0.15 and 0.2 m s - ‘, respectively. For both continuous and transient perturbations, the maximum amplitudes were preselected so as to challenge stability close to the maximum limits that can be withstood without touching a handrail or moving the feet, based on the results of pilot tests on healthy young adults. The lower amplitudes were

Transient

and continuous

selected to be 50 and 75% of the maximum amplitudes. The bandwidth of the continuous waveform (0.13-5.0 Hz) and the ‘shape’of the transient waveform (200 ms pulsewidth, 600 ms interpulse interval) were selected so as to provide similar frequency content (e.g. see Doebelin, 1980). For each test, the subjects were instructed to stand relaxed, with feet comfortably spaced (in stocking feet) and with hands clasped together (arms hanging loosely in front of the abdomen), and to try to maintain balance without moving their arms or feet. To minimize the variations in visual and vestibular input, they were told to look straight ahead at the visual target. Monotonous music was played through headphones to mask any distracting noises and/or auditory cues from the platform motor. Prior to the first test, the heels were aligned against a vertical panel placed at right angles to the direction of platform motion (which was then removed before starting the tests), but the subjects were otherwise free to select the angulation and medial-lateral spacing of their feet, so as to simulate the base of support that they would normally choose. The outlines of the feet were then traced and the same foot positioning was maintained for all test conditions. The experimenter recorded any arm movements or changes in foot position that occurred, and the tests were videotaped to allow any such observations to be verified. If there was a change in foot position, then the feet were realigned with the tracing before performing the next test. The experimental factors of primary interest were: (1) waveform (continuous or transient), (2) visual condition (eyes open or blindfolded), and (3) perturbation amplitude (1 = low, 2 = medium, or 3 = high). A repeated-measures experimental design was employed, i.e. each subject was tested with all ‘treatments’ (i.e. all possible combinations of factor levels). Three replications of each treatment were performed. To allow possible order-of-testing effects to be assessed, the subjects were divided into four order-pattern groups: half of the subjects (groups 1 and 2) were tested with transient perturbations first and half (groups 3 and 4) were tested with continuous perturbations first; half (groups 1 and 3) were tested with eyes open first and haIf(groups 2 and 4) were tested blindfolded first. New subjects were assigned sequentially to one of the four order patterns until there was a total of three subjects in each order-pattern group. To control order effects further, the three replications were performed as three ‘rounds’; within each round, the tests were performed in random order. For the continuous perturbations. each round comprised three tests, i.e. one test at each amplitude. For the transient perturbations, each round comprised six tests, i.e. one test at each of the three amplitudes in both forward and backward directions. For each waveform, learning trials were performed prior to starting the experimental series. These trials were conducted with eyes open first and then repeated while blindfolded, and the amplitudes were adminisBH26:10-D

postural

control

1183

tered in ascending order, with one trial at each amplitude and (for the transient perturbations) in each direction (a total of six learning trials for continuousperturbation tests, and 12 for transient-perturbation tests). The subject was then informed that, in the succeeding tests, the same motion patterns would be administered several times in random order. The three rounds of tests (as detailed above) were then administered for the first waveform and visual condition (as determined by the subject’s assigned order pattern). Before the start of each test, the platform was controlled to move gently and smoothly (over an interval of 10 s) to the midpoint of the range of motion. There was then a delay of 3 s and, for each transient test, an additional random time delay of O-5 s before the start of the perturbation. Each subject was given a I5 min seated rest after completing the tests for the first visual condition. After completing both visual conditions. there was a 30 min seated rest period and the protocol was then repeated for the second waveform. Each of the four blocks of tests required the subject to stand on the platform for approximately 30 min. Including preparation and cleanup time, each testing session lasted approximately 4 h. Postural responses were characterized in terms of the anterior-posterior displacement of the center of pressure. The center ofpressure represents the effective location of the forces acting on the feet, and center-ofpressure displacement was selected as the performance measure because it has two useful interpretations. First, since anterior-posterior center-of-pressure displacement is approximately proportional to the net flexionextension moment of force (or ‘torque’) acting at the ankles, it is representative of the net mechanical outcome of the stabilizing ankle muscle activity. Secondly, since ‘loss of balance’ (i.e. the need to move the arms or feet) generally occurs at or near the time when the center of pressure reaches the limits of the base of support (i.e. the perimeter of the feet). center-ofpressure displacement can be interpreted in terms of relative postural stability. i.e. by indicating the degree to which stability limits are approached (Maki. 1987; Maki rt CL/.,1987). Postural performance was quantified in terms of the ‘unit-pulse response’, i.e. the peak center-of-pressure displacement in response to a unit-area acceleration pulse. The unit-pulse response is. in effect, a measure of a biomechanical ‘gain’ of the posture control system. To allow these values to be compared for individuals of differing body dimensions, each unit-pulse response was normalized with respect to (i.e. divided by) the anterior--posterior length of the base of support as measured from the foot tracing. For each transient-perturbation test, the unit-pulse response was determined by finding the peak anterior-posterior center-of-pressure displacement and dividing this value by the area (i.e. time integral) of the measured acceleration pulse (example responses are shown in Fig. 1). For each continuous-perturbation test, the unit-pulse response was estimated by

1184

B. E. MAKI and G. OSTROVSKI

0.d

r

I

8 I

I

-0.0

-0.5

,

,

,

I

,

,

1.0

0.5

TIME (SeoMlds)

,

,

,

1.5

,

c-1.0

20

0.al-l

I

-10

(a)

:

I

12

r

I

I

3 4 5 6 TlME(seconds)

7

,

8

9

1

1-0.5 la

(a)

0.30

1.0

0.25

!/

/ ‘s,; :

!

-0.5

;

0.20-

1

0.0 -0.2 I--"-

0.05-

I i;..' I' 0.001 I I 1, I I I, a.5 -0.0

04

-

I I q I,, 0.5 1.0 TlME(aecands)

oe”lm-of-pesa”re

....--- aarrlaralfon

I r r I, r-1.0 1.5 2.0

I (W

Fig. 1. Example responses to backward transient perturbations: (--) center-of-pressure displacement; (- - -) platform acceleration; the forward direction is positive; center-ofpressure=O.O at the back of heels. In (a), the center-ofpressure was actively controlled to move back toward the starting position after each of the acceleration and deceleration pulses, in contrast to the plateau seen in(b); this plateau may be indicative of an adaptation, whereby the subject learned to use the platform deceleration pulse to help restabilize the body.

fitting a linear, time-invariant transfer function model to the data, treating platform acceleration as the input and center-of-pressure displacement as the output, and then using this model to predict the center-ofpressure response to a unit-area acceleration pulse (Fig. 2 shows example test data and the resulting model prediction). To fit this model, the data were first used to estimate the nonparametric frequency response, by dividing the data into 15.36 s segments (50% overlap of adjacent segments), computing the fast Fourier transform for each segment, and then averaging the resulting ‘periodograms’ (the first data segment was discarded, to eliminate any transient effects or nonstationarity associated with the start of the platform motion). Each nonparametric frequency response was then fitted with a parametric transfer function, using the Akaike information criterion to determine the most appropriate model order, and the parametric transfer function was used to predict the unit-pulse response (Maki, 1986a; Maki, 1987; Maki et al., 1987; Maki and Fernie, 1988). In addition, to quantify the performance during continuous-perturbation tests independent of the modelling assumptions (i.e. linearity and time invaria-

-

_._-I\ unit“area” acooleration pulsw

cemer-of-prersure ....... &xelera5on

Fig. 2. Example response to continuous perturbation: (a) first 10s of data from a 62s test; (b) unit-pulse response predicted from a linear transfer function model fitted to the data from the continuous-perturbation trial; (--) center-ofpressure displacement; (- - -) platform acceleration; the forward direction is positive; center-of-pressure = 0.0 at the back of heels in (a); in Fig. 2b, the predicted change in center-ofpressure displacement is shown relative to an arbitrary starting point. A general similarity is noticed in the form of the predicted pulse response (b) and the measured pulse response (i.e. the response to the initial acceleration pulse in Fig. la).

rice),, a ‘RMS gain’ was calculated for each of these tests, i.e. the RMS anterior-posterior center-of-pressure displacement (relative to the mean) divided by the RMS acceleration. Response atnptitudes (as opposed to gains) were also analyzed, i.e. the RMS amplitude of center-of-pressure displacement, for continuous-perturbation tests, and the peak center-ofpressure displacement, for transient-perturbation tests. Separate analyses were performed for data from the backward transient, forward transient and continuous-perturbation tests. In each case, a four-way analysis of variance (ANOVA) was performed using a repeated-measures model (i.e. subjects treated as random blocking variables). The ‘within-subject’ factors were visual condition, perturbation amplitude and round number. Order pattern was the only ‘betweensubjects’ factor. In addition to the ANOVA, analysis of covariance (ANCOVA) was performed in an attempt to reduce the influence of order-of-testing effects by treating test number as a covariate. Except for the addition of the covariate term, the ANCOVA and ANOVA models were identical.

1185

Transient and continuous postural control Any tests in which the subjects were observed to touch the handrail or take a step were excluded from the analyses. This criterion resulted in the rejection of four forward transient tests, one backward transient test and two continuous-perturbation tests, out of a total of 648 tests. In order to perform the ANOVA and ANCOVA analysis, each missing value was replaced with an average value (Little and Rubin, 1987) calculated using data from the other tests performed on the same subject at the same amplitude and visual condition. Because of the relatively large number of F-tests performed in using these high-order ANOVA and ANCOVA models, the criterion level of significance for hypothesis testing was set, a priori, at a conservative level (IX= O.Ol), to reduce the likelihood ofcommitting Type I errors (i.e. false detection of differences between means). The assumptions of the ANOVA and ANCOVA models were checked by examining plots of the model residuals (Montgomery, 1984). In order to ‘stabilize’ the variance, it was necessary to perform logarithmic transformations on the data prior to analysis (Neter et al., 1985).

RESULTS

The analyses of the unit-pulse response scores failed to support the hypothesis that the responses elicited by the transient and continuous perturbations would show similar behavior in terms of their dependence on perturbation amplitude (Table 1). For both forward and backward transient perturbations, amplitude had a highly significant influence on the unit- pulse response scores (p < O.OOl), which were seen to decrease with increasing amplitude (Fig. 3). In contrast, for the continuous-perturbation tests, the results showed only weak evidence of an amplitude effect (p = 0.07). Noting once again that the unit-pulse response is essentially a ‘gain’ (i.e center-of-pressure displacement per unit acceleration), the changes in unit-pulse response with

perturbation amplitude seen in the transient tests are indicative of a nonlinear system, whereas the nearconstant unit-pulse responses seen in the continuousperturbation tests are indicative of greater linearity. None of the ANOVAs showed significant interaction between perturbation amplitude and any of the other factors, i.e the influence of perturbation amplitude was not confounded by the effects of vision or order of testing. The analyses of the unit-pulse response scores, using the ANCOVA statistical model, did appear to support the hypothesis that the responses to the transient and continuous perturbations would show similar vision dependence, in that the responses to both types of perturbation were unaffected by deprivation of visual inputs. The results of the initial ANOVAs of these data, however, were confounded by order-of-testing effects (Table 1). Although there were no significant differences between testing rounds (~20.06) or between order-pattern groups (p > 0.15), there was, for the continuous-perturbation tests, evidence of a significant interaction between vision and order pattern (p = 0.003). The influence of vision was confounded by the order in which the tests were performed: center-ofpressure displacement tended to be larger in whichever visual condition happened to be tested first (Fig. 4). For the transient-perturbation tests, similar trends appeared to occur, although the differences were not statistically significant (p > 0.06). The order-of-testing effects were eliminated in the ANCOVAs in which test number was included as a covariate (Table 1). In these analyses, responses to both transient and continuous perturbations failed to show a significant vision effect (p20.13). Analysis of the RMS gain values calculated directly from the continuous-perturbation data (as opposed to the unit-pulse response scores which were estimated by modelling these data) led to the same conclusions with regard to the influence of vision, but showed somewhat different results with respect to the influence of perturbation amplitude. Specifically, ANOVA

Table 1. Results of ANOVA and ANCOVA on unit-pulse response scores. p-values* Forward transient perturbations ANOVA ANCOVA

Factor 0 (order pattern) V (vision) A (amplitude) R (round Interactions: 0-l’ All other interactions

#)

Backward transient perturbations ANOVA ANCOVA

Continuous perturbations ANOVA ANCOVA

0.66 0.17
0.62 0.13
0.86 0.32
0.86 0.32
0.15 0.52 0.07 0.06

0.06 > 0.08

0.40 >O.lO

0.43 > 0.02

0.98 > 0.02

0.003 > 0.09

0.15 0.52 0.08 0.51

0.38 > 0.09

*p-values: F-test for differences between means due to the factors listed, four-way ANOVA and ANCOVA; ANCOVA included test number as covariate; dependent variable = unit-pulse response; unit-pulse response determined directly for transient-perturbation data and estimated from model fitted to continuous-perturbation data; log transformation performed prior to analysis.

1186

B. E.

0.0

MAKI and G. OSTROVSKI

TRANSIENT FORWARD CONTINUOUS TRANSIENT BACKWARD

Fig. 3. Influence of changes in perturbation amplitude on the unit-pulse response (i.e. peak center-of-pressure displacement, expressed as a fraction of base-of-support length, divided by the time integral of the acceleration pulse; to facilitate physical interpretation, the unit-pulse response values have been divided by 5 to show the response to a pulse with area = 0.2 m s-r, i.e. backward transient amplitude # 3); for the continuous-perturbation tests, the unit-pulse responses are the values predicted using the transfer function model fitted to the test data; the bars represent the grand mean scores for the 12 subjects, and the error flags represent the standard deviation. The changes in unit-pulse response seen in the transient-perturbation tests are indicative of a nonlinear system, whereas the near-constant unit-pulse response values seen in the continuous-perturbation tests suggest greater linearity. These differences would appear to reject the hypothesis that the responses to the two types of perturbation would show similar amplitude dependence. 0.7

P

0.6.

0’

0.5

b g

0.4-

y

0.3.

g

0.2-

3

1

DISCUSSION

The present study investigated a hypothesis that postural responses to transient and continuous perturbations would show similar behavior in terms of their dependence on visual inputs and on perturbation amplitude. An underlying objective was to provide some evidence that would help to evaluate the degree to which transient balance recovery and continuous postural regulation might be governed by the same control mechanisms, although a definitive examination of this issue is well beyond the intended scope of this study. A second, more practical objective was to provide data to help assess the feasibility of using

0.1.

0.0)

BF FIRST EO FIRST CONTINUOUS FIRST

BF FIRST 0 FIRST TRANSIENT FIRST

10

eyes open (EO) m

Mindfolded WI

Fig. 4. Influence of visual conditions on the unit-pulse response scores predicted from the continuous-perturbation tests (data scaled as in Fig. 3); results are shown separately for the four order-pattern groups; for each group, the left bar represents the visual condition that was tested first; EO = eyes open (empty bars), BF = blindfolded (filled bars); each bar represents the grand mean score for the three subjects in the order-pattern group, and each flag represents the standard deviation. These data illustrate the confounding influence of order of testing, in that the unit-pulse response values were higher in whichever visual condition happened to be tested first; similar trends were seen in the transient-perturbation data. When these order effects were taken into account, the analysis failed to show evidence of a vision effect for either type of perturbation, thereby supporting the hypothesis that the responses to transient and continuous perturbations would show similar vision dependence.

of the RMS gain values from the continuous-perturbation

significant, decreases as perturbation amplitude was increased (p < 0.001) (Fig. 5). In order to investigate further the nature of the amplitude dependence of the postural responses, the response amplitudes (as opposed to unit-pulse response or RMS gain values) were also analyzed. ANOVA of the response amplitudes showed significant increases in response with increasing perturbation amplitude, for both transient and continuous perturbations (p
data

failed

to show

any

influence

of vision

(p = 0.26) nor any vision-order interaction (p =0.29). However, the RMS gain did show some small, but

EYES OPEN

BLINDFOLDED

Fig. 5. Influence of changes in perturbation amplitude on the RMS gain for the continuous-perturbation tests (RMS gain = RMS center-of-pressure displacement, expressed as a fraction of base-of-support length, divided by RMS acceleration); each bar represents the grand mean score for the 12 subjects, and each flag represents the standard deviation. In contrast to the results for the continuous-perturbation unit-pulse response data shown in Fig. 3, these data indicate a nonlinear amplitude dependence and, hence, might be interpreted as supporting the hypothesis that the responses to continuous and transient perturbations would show similar (nonlinear) amplitude dependence. However, the contribution of spontaneous postural sway to the RMS center-of-pressure displacement may have confounded these results (see Discussion).

Transient

and continuous

models derived from continuous-perturbation balance tests to predict responses to transient perturbations. In designing the experiment, it was assumed, based on the well-known ability of humans to adapt their postural responses (e.g. Nashner, 1976), that perturbation predictability could be a major confounding factor. To minimize adaptations, efforts were directed at making both the transient and continuous perturbations as unpredictable as possible. Furthermore, potential confounding effects of any adaptations that did occur were minimized by explicitly accounting for order of testing in the statistical analyses. These analyses did, in fact, show evidence of significant order effects. For example, in the continuous-perturbation tests, center-of-pressure displacement tended to be larger in the first set of tests, regardless of the visual condition, and the transientperturbation tests also showed possible evidence of a similar trend. Although the order effects could be due to fatigue-related changes, several seated rest periods were provided in order to prevent fatigue. More likely, the order effects are a result of progressive changes learned through repeated exposure to the same perturbation waveform. Such adaptations might represent a gradual ‘finetuning’ and optimization of control strategy, and may have resulted from a progressive shift to more of a predictive control strategy. Although the continuous pseudorandom waveform was in fact periodic, the period was chosen to be relatively long (15 s) in order to prevent subjects from ‘remembering’ the pattern. Nonetheless, subjects may have been able to learn some of the features of the pseudorandom pattern (e.g. amplitude constraints) and thereby developed predictive strategies to improve their performance. The transient perturbations were unpredictable in terms of amplitude, direction and time of onset. yet there was one predictable feature in that each acceleration pulse was always followed, after a fixed time interval, by a deceleration pulse. Subjects may have adapted by learning to use the platform deceleration to help return the body to an upright posture. This point is illustrated in Fig. 1. In the test shown in Fig. l(a), the center of pressure was actively controlled to move back toward the starting position during the constantvelocity phase following the acceleration pulse. In contrast, in Fig. l(b), the center of pressure simply remained at its displaced value until the deceleration pulse occurred, suggesting perhaps the use of a control strategy in which the subject generated just enough ankle moment to arrest the sway and then maintained the leaning posture until the platform deceleration acted to return the body to an erect position. In addition to the issue of predictability, it was anticipated that it was necessary to control other potentially confounding characteristics of the transient and continuous waveforms, i.e. frequency content and level of challenge to stability. Frequency content is particularly important in studying vision effects, because previous studies have suggested that visual

postural

control

I IX7

contributions to postural stabilization are slow (Lestienne et al., 1977) and, therefore, limited to low frequencies, e.g. 0.034.3 Hz (Diener er al., 1982). In the present study, the transient and continuous perturbations both included substantial energy at low frequencies: the bandwidth of the continuous waveform was 0.1-5 Hz and the transient waveform also provided substantial excitation over the same frequency range. With regard to level ofchallenge, failure to control this factor could well lead to differences in postural strategy. For example, Horak and Nashner (1986) have proposed that an ‘ankle strategy’ is used to correct for small postural disturbances, whereas a ‘hip strategy’ is used when the limits of stability are challenged to a greater extent. In addition, ‘central set’ could play a confounding role, wherein, for example, subjects might adopt a ‘sloppy’ control strategy when their balance is not adequately threatened (Maki and Fernie, 1988). In the present study, we attempted to match the two types of perturbation for level of challenge (or ‘difficulty’) by selecting, as the maximal perturbations, amplitudes that were found to be nearly large enough to elicit consistently stepping responses in pilot tests. In analyzing the center-of-pressure data, however. it became apparent that while the continuous perturbation was successful in challenging stability at occasional points during the test. the challenge was less substantial during a large portion of the test. The occasional extreme center-of-pressure excursions that occur limit the maximum amplitude that can be tested without evoking stepping; however, for most of the test, the center-of-pressure displacement is much smaller. For example, during the maximum-amplitude continuous-perturbation tests, the average RMS center-of-pressure displacement was only about f3% of the base-of-support length. As discussed in more detail below, this factor may have been responsible, at least in part, for the greater linearity seen in the responses to continuous perturbations. In future, it might be possible to achieve more persistent large-amplitude excitation by changing the amplitude distribution of the continuous waveform, e.g. using a binary (squarewave) sequence rather than a sum of sinusoids. Another aspect of the current study which might be thought to limit generalizability relates to the visual field conditions, i.e. the fact that the visual surround was controlled to move with the platform. This testing situation is analagous to standing in a windowless vehicle moving in a horizontal plane. In this situation. the objective of the posture control system is to maintain an upright posture with respect to the platform, and the most relevant visual ‘error signal’ for the control system is the relative motion between the head and the platform. Moving the visual surround with the platform allows the visual inputs to provide a direct measure of this error signal. Therefore, from this point of view, this testing paradigm is expected to enhance the importance and effectiveness of the visual feedback. On the other hand, this paradigm may lead to an incongruence between the visual inputs. which record

1188

B. E. MAKIand ci.

the head motion relative to the platform, and the vestibular otolith inputs, which record the head motion in an absolute reference frame. This incongruence, if unanticipated, might affect the performance in initial tests (Nasher and Berthoz, 1978) and could, in fact, account for some of the order effects discussed earlier, as subjects may gradually adapt their control strategy to deal with what is initially, perhaps, an unfamiliar and/or unexpected situation. The possibility that subjects simply disregarded visual inputs because of the unfamiliarity of the visual field conditions cannot be ruled out, although it can be argued that the conditions are in fact familiar to anyone who has ever stood in a moving vehicle. It should be pointed out that the alternative approach, i.e. having the subjects stand on the moving platform while viewing a stationary visual field, is likely to lead to predictive responses and adaptations that would tend to confound seriously the influence of vision on continuous-perturbation responses. Under such conditions, subjects could estimate the limits of the platform range of motion by viewing this motion relative to stationary reference points in the laboratory and thereby predict when changes in direction are imminent. Possible limitations due to the modelling assumptions should also be noted. To facilitate comparison with the transient-perturbation responses, data from each continuous-perturbation test were fitted with a linear, time-invariant input-output model and this model was used to predict the response to transient perturbation. Even though the posture control system may involve significant nonlinearities, a linear model may provide a reasonable approximation of the system dynamics over a limited operating range while avoiding the complexities of nonlinear analysis. The linearized models did, in fact, show reasonable ‘goodness of fit’, as indicated by the coherence function (Maki, 198613,1988): mean coherence, averaged over the perturbation frequencies, typically exceeded 0.9. Furthermore, some preliminary work has suggested that addition of nonlinear terms to the model results in relatively little improvement in fitting the model to the data or in predicting responses to transient perturbations (unpublished data). Previous analyses have supported the assumption of time invariance, in that responses to tests of even longer duration failed to show strong evidence of nonstationarity (Maki, 1986a). The initial 15 s data segment, where nonstationarity is most likely to occur (Werness, 1982), was not used in fitting the model. Notwithstanding the above, there was concern that the modelling approximations might have influenced the findings, and it was for this reason that the RMS gain values were also calculated (directly from the response data). Analysis of these data substantiated the finding of no vision effects, but did provide somewhat different results with respect to amplitude dependence. A possible explanation for the discrepancy is discussed later. The present results failed to demonstrate any influence of vision on the dynamic biomechanical re-

USTROVSKI

sponses to either transient or continuous platform translations, once order-of-testing effects were taken into account. Using platform-tilt perturbations, Diener et al. (1986) and Diener and Dichgans (1988a) likewise found that responses to transient perturbations (toes-up tilts) were unaffected by vision. However, in contrast to the present results, they reported that changes in visual conditions did have a significant influence on the responses to continuous perturbations (sinusoidal tilts). Presumably, visual inputs are not normally used in dynamic stabilization because the long latency of the visual responses (Lestienne et al., 1977) greatly reduces the probability of generating adequate stabilizing responses before the limits of stability are exceeded. This argument applies regardless of whether the perturbation is transient or continuous, as long as the waveform is unpredictable and has amplitude and frequency characteristics sufficient to challenge stability. On the other hand, if the perturbation is predictable, as in the case of the sinusoidal platform motion used by Diener et al. (1986) and Diener and Dichgans (1988a), then the posture control system may learn to compensate for the visual latency in a predictive manner and thereby use the visual inputs to help synchronize anticipatory responses with the perturbation (‘phase locking’). This type of anticipatory adaptation to sinusoidal perturbation has, in fact, been documented (Hocherman et al., 1988), and we suggest that this is the main reason for the discrepancy between the present results and those of Diener et al. (1986) and Diener and Dichgans (1988a). However, it should be noted that differences in the method of perturbation (platform tilt versus translation), as well as differences in visual field conditions and perturbation frequency content and level of challenge (as discussed earlier), could also have played a role. With regard to the influence of perturbation amplitude, the present results showed highly significant amplitude effects in the transient-perturbation tests, as the unit-pulse response scores decreased markedly with increasing perturbation amplitude. This is indicative of nonlinear performance, since the unit-pulse response is, in effect, a measure of a biomechanical ‘gain’ of the posture control system (i.e. center-ofpressure displacement per unit acceleration). If the system were linear, the gain would be unaffected by changes in perturbation amplitude. Although other investigators have also reported a scaling of responses to transient platform-translation perturbations (e.g. Diener et al., 1988b), it is difficult to compare these findings with those of the present study, where the perturbations were presented in a much less predictable fashion. Another difficulty is that the magnitude of the perturbation is usually quantified in terms of platform displacement or velocity, rather than acceleration. Of course, changes in acceleration amplitude must necessarily produce concomitant changes in velocity and displacement (unless the ‘shape’, i.e. the frequency content, of the waveform itself is altered).

Transient and continuous postural control However, for translating-platform paradigms (where the subject and platform are both initially at rest and the subject’s feet remain ‘fixed’ with respect to the platform), it can be argued that acceleration is the most meaningful parameter, in that the destabilizing joint moments that are induced by the platform translation are actually caused by the platform acceleration; free-body analysis demonstrates that constant-velocity platform translation does not induce any perturbing joint moments (Maki, 1987). The nonlinearity seen in the responses to transient perturbations might suggest that the control system is operating in some nonlinear fashion, which could involve nonlinear feedback control and/or nonlinear scaling of preprogrammed feedforward responses. Another possibility would be amplitude-dependent changes in control strategy, e.g. a shift from an ‘ankle strategy’ to more of a ‘hip strategy’ at larger perturbations. This could, in fact, explain the nonlinear effects, because a hip strategy would tend to reduce the required center-of-pressure displacement (Horak and Nashner, 1986). It is also possible that the nonlinearity is intrinsic to the biomechanics of the system. One important biomechanical nonlinearity is ‘base-of-support saturation’, in which the center of pressure simply reaches the functional limits of the base of support (i.e. near the end of the feet) and cannot be displaced any further (Maki et al., 1987). When center-of-pressure displacement saturates, attempts to generate further stabilizing ankle moment will cause the heel or toes to leave the ground. Subjects may be able to maintain balance (without touching a handrail or stepping) for brief periods beyond this point, but the center-ofpressure displacement will not change substantially. Many tests at higher amplitudes did, in fact, show a plateau in center-of-pressure displacement which might be attributed to base-of-support saturation [e.g. see Fig. l(b)]. The present study is, to our knowledge, the only study to have examined the scaling of responses to continuous perturbations. The results showed no significant amplitude dependency in the unit-pulse response estimates (derived by modelling the data from the continuous-perturbation tests), in contrast to the substantial nonlinearity seen in the transient-perturbation tests. As mentioned earlier, this may be so because the control system was simply operating in a range of greater linearity during much of the continuous perturbation, as indicated by the relatively small RMS center-of-pressure excursions that occurred. Alternatively, it could be that the unit-pulse response analysis simply failed to detect an amplitude dependence, since analysis of the RMS gain values calculated directly from the continuous-perturbation data did show small, but statistically significant, decreases with increasing perturbation amplitude; however, another explanation in possible. Assume that the overall center-of-pressure displacement, C,, can be represented as the sum of two independent variables: (I) a spontaneous-sway component, C,, that has a constant root-

1189

mean-square value independent of the perturbation amplitude (i.e. Cs = K), and (2) a perturbation-induced component, C1, that is linearly related to the perturbation amplitude (i.e. C, = kA). In this case, the measured RMS gain, i.e. RMS gain = Co/A = [Ci + Cf]“.5/A

will necessarily decrease as the perturbation amplitude, A, is increased. Substitution of typical values into this equation yields predictions of amplitude-related changes in RMS gain that are similar in magnitude to the experimentally observed amplitude dependence. Although the possibility that transient and continuous perturbation elicit different modes of control cannot be ruled out, the foregoing discussion suggests that the observed differences in amplitude dependence may instead have been a result of waveform-dependent differences in the degree to which biomechanical nonlinearities were excited (i.e. rather than the transient or continuous nature of the waveform per se). Excitation of biomechanical nonlinearities could affect the linearity of the measured center-of-pressure displacement directly, but could also affect linearity indirectly by forcing the control system to adopt nonlinear control mechanisms in order to deal with the biomechanical constraints (e.g. a switch from an ‘ankle’ to a ‘hip’ strategy in order to circumvent limitations on ankle moment generation imposed by base-of-support saturation). In summary, the present study failed to demonstrate significant differences between responses to transient and continuous perturbations with regard to the influence of vision. The failure to see a vision effect in either task supports the hypothesis that the responses to the two types of perturbation would show similar vision dependence, and also serves as evidence to counter previous claims that responses to continuous perturbations are necessarily influenced by vision (Diener er al., 1986; Diener and Dichgans 1988a). It is suggested that the discrepancy in findings is primarily due to the greater unpredictability of the perturbations used in the present study. A finding of greater linearity in the continuous-perturbation data would appear to reject the hypothesis that the responses to transient and continuous perturbations would show similar dependence on perturbation amplitude. However, it is suggested that the amplitude distribution of the particular pseudorandom waveform used in this study, rather than the continuous nature of this waveform per se, may account for the observed differences. More work is needed to resolve this issue. Although the present results cannot be used unequivocally to support the hypothesis that the same control mechanisms may be utilized in transient and continuous postural control, they do demonstrate that distinctions between these two types of postural task are not nearly so clear cut as has been suggested. Nonetheless, from a practical perspective, the finding of substantial

B. E. MAKI and G. USTROVSKI

1190

nonlinearity indicates that attempts to predict responses to transient perturbations from continuousperturbation tests should be performed with caution. Acknowledgements-This work was supported by an operating grant (#MT-10576) from the Medical Research Council ofCanada. B. E. Maki’ held a research fellowship from the Ontario Ministry of Health and is currently supported by the Research Program in Aging at Sunnybrook Health Science Centre. G. Ostrovski was supported on sabbatical leave by Rafael-ADA, Haifa, Israel. The authors thank G. Griggs for his technical assistance, J. P. Szalai and M. Katie for their statistical advice, and G. Fernie, P. Holliday and W. McIlroy for their comments and suggestions regarding the manuscript. REFERENCES

Diener, H. C., Dichgans, J., Bruzek, W. and Selinka, H. (1982) Stabilization of human posture during induced oscillations of the body. Exp. Brain Res. 45, 126-132. Diener, H. C., Dichgans, J., Guschlbauer, B. and Bather, M. (1986) Role of visual and static vestibular influences on dynamic posture control. Hum. Neurobiol. 5, 105-l 13. Diener, H. C. and Dichgans, J. (1988a) On the role of vestibular, visual and somatosensory information for dynamic postural control in humans. Progress in Brain Research 76, 253-263.

Diener, H. C., Horak, F. B. and Nashner, L. M. (1988b) Influence of stimulus parameters on human postural responses. J. Neurophysiology 59, 1888-1905. Doebelin, E. 0. (1980) System Modeling and Response: Theoretical and Experimental Approaches, p. 253. Wiley, New York. Hocherman, S., Dickstein, R., Hirschbiene, A. and Pillar, T. (1988) Postural responses of normal geriatric and hemiplegic patients to a continuing perturbation. Exp. Neural. 99, 388402. Horak, F. B. and Nashner, L. M. (1986) Central programming of postural movements: adaptation to altered support-surface configurations. J. Neurophysiol. 55, 1369-1381. Horak, F. B., Diener, H. C. and Nashner, L. M. (1989) Influence of central set on human postural responses J. Neurophysiol. 62, 841-853. Lee, W. A. (1989) A control systems framework for understanding normal and abnormal posture. Am. J. Occupational Therapy 43, 291-301.

Lestienne, F., Soechting, J. F. and Berthoz, A. (1977) Postural readjustments induced by linear motion of visual scenes. Exp. Brain Res. 28, 363-384. Little, R. J. and Rubin, D. B. (1987) Statistical Analyses with Missing Data. Wiley, New York. Maki, B. E. (1986a) Selection of perturbation parameters for identification of the uosture-control svstem. Med. Biol. Engng Comput. 24, 581-568.

Maki, B. E. (1986b) Interpretation of the coherence function when using pseudorandom inputs to identify nonlinear systems. IEEE Trans. Biomed. Engng BME33, 775-779. Maki, B. E. (1987) A posture control model and balance test for the prediction of relative postural stability. Doctoral dissertation, University of Strathclyde, Glasgow. Maki, B. E. (1988) Addendum to ‘Interpretation of the coherence function when using pseudorandom inputs to identify nonlinear systems’. IEEE Trans. Biomed. Engng BME-35, 279-280. Maki, B. E., Holliday, P. J. and Fernie, G. R. (1987) A posture control model and balance test for the prediction of relative postural stability. IEEE Trans. Biomed. Engng BME-34, 797-810. Maki, B. E. and Fernie, G. R. (1988) A system identification approach to balance testing. Progress in Brain Res. 76, 297-306.

Maki, B. E., Holliday, P. J. and Topper, A. K. (1991) Fear of falling and postural performance in the elderly. J. GerontoI. Med. Sci. 46, M123-131. Maki, B. E. and Whitelaw, R. S., (1993) Influence of expectation and arousal on center-of-pressure responses to transient postural perturbations. J. Vest&. Res. 3, 25-39. Montgomery, D. C. (1984) Design and Analysis of Experiments (2nd Edn), Ch. 4, Wiley, New York. Nashner, L. M. (1976) Adapting reflexes controlling the human posture. Exp. Brain Res. 26, 59-72. Nashner, L. M. (1981) Analysis of stance posture in humans. Handbook of Behavioral Neurobiology: Motor Coordination, Vol. 5 (Edited by Towe, A. L. and Luschei, E. S.), pp.

527-565. Plenum, New York. Nashner, L. and Berthoz, A. (1978) Visual contribution to rapid motor responses during postural control. Brain Res. 150,403-407.

Neter, J., Wasserman, W. and Kutner, M. H. (1985) Applied Linear Statistical Models (2nd Edn). vu. 615-617. Richard D. Irwin, Inc., Homewooh, Illinois: - _ Werness, S. A. S. (1982) Parametric analysis of dynamic postural responses. Doctoral dissertation, University of Michigan, Ann Arbor.