Essay: The minimal displacement principle

Abstract
The ‘minimal displacement’ principle (MDP) holds that obstacles are avoided with minimal foot displacement during walking. According to the MDP, obstacles are avoided using either a long stride strategy (LSS; extending the stride and placing the foot over the obstacle) or short stride strategy (SSS; shortening the stride placing the foot before of the obstacle), depending on the obstacle location relative to the planned stepping location. However, it is still unknown whether obstacle avoidance is performed according to the MDP and whether or not the MDP is influenced by age and attentional interference. This study examined the effect of age and attentional interference on stride adjustment strategies (SAS; LSS or SSS) when obstacles were suddenly presented during walking at various obstacle locations relative to the predicted foot placement position (pFP; that was predicted to occur in the absence of a stride adjustment). Fourteen healthy young (24.3 ?? 2.0 years) and thirteen healthy elderly adults (65.3 ?? 3.4 years) participated. The experiment was performed using the Interactive Walkway (IWW) that projected visual obstacles on the walkway and recorded the participant’s movements. Two conditions (with and without attentional interference, as induced by counting backwards) were performed, both in which obstacles were suddenly presented at seven locations relative to pFP (one per trial). No differences in success rates and secondary-task performance were found between groups and conditions. ??2-tests revealed that LSS was the preferred strategy, suggesting that during walking obstacles are not always avoided according to the MDP. Age and attentional interference both negatively affected adherence to MDP, showing an increased likelihood for performing LSS when SSS was expected. Future research should focus on other factors than MDP to unveil what underlies the choice of a particular SAS for obstacle avoidance during walking.
Keywords: minimal displacement, obstacle avoidance, dual-task, ageing, walking
1. Introduction
The goal of fall prevention programs for people who are prone to falling (e.g. due to ageing, previous falls, a stroke, Alzheimer’s or Parkinson’s disease) is to improve their walking abilities and to decrease their fall risk. The latter may benefit from training obstacle avoidance skills, for example by practicing daily life situations (e.g. crossing a door step) or by practicing obstacle-avoidance tasks (Weerdesteyn et al., 2006, 2008; Van Ooijen et al., submitted). Previous studies found that tripping over obstacles during walking in both young and elderly adults is dependent on the available response time (time from obstacle appearance until time of predicted foot placement; ART). Less time to respond has been demonstrated to result in higher obstacle avoidance failures (Chen et al., 1994; Weerdesteyn et al., 2005). When an obstacle has to be avoided, one can choose between two strategies: the long stride strategy (LSS) or short stride strategy (SSS) (Chen et al., 1994). The LSS means making one’s stride larger than in a situation without the obstacle. Making a shorter stride and crossing the obstacle with the contralateral leg is called the SSS. Previous studies showed that LSS is the favored strategy during walking (Chen et al., 1994; Moraes et al., 2004; Weerdesteyn et al., 2004, 2005; Roerdink et al., 2009; Bank et al., 2011). Overall, however, elderly adults were less successful in crossing obstacles when using LSS than young adults (Chen et al., 1994; Weerdesteyn et al., 2005).
The studies of Chen et al. (1994) and Weerdesteyn et al. (2005) investigated the influence of different ARTs on the stride adjustment strategies (SAS), but their experimental approaches were different (see intermezzo). Weerdesteyn et al. (2005) suggested that the avoidance strategy of young adults is more in line with the ‘minimal displacement’ principle (MDP; Patla et al., 1999). According to the MDP, the foot is minimally displaced when avoiding an obstacle, which means that an LSS is expected when the obstacle is located closer by the person’s predicted foot placement (i.e. in absence of an obstacle; pFP), and SSS when the obstacle is located farther away than pFP. However, for each age group Weerdesteyn et al. (2005) applied the MDP to the performed strategies of all trials of all ART conditions, thereby varying obstacle locations with respect to pFP (due to their experimental approach), without taking the ART and obstacle locations into account. Because conclusions in terms of MDP critically depend on the relation between pFP and actual obstacle location, the interpretation of SAS-results in this respect is questionable. Therefore, it is important to clarify whether SAS are in line with the MDP, taking both ART and obstacle location with respect to pFP into account.
It is of great interest to determine the SAS in elderly because most falls in elderly occur during walking (Markle-Reid et al. 2010) and 25% of the falls are due to tripping over an obstacle in the environment (Tinetti et al., 1988). A factor that increases the risk of falling is an impairment in avoiding obstacles while walking under dual-task conditions, which is suggested to be due to a decreased attentional capacity and difficulties in switching attention (Bock, 2008; Siu et al., 2008; Chen et al., 1996). A secondary task is often implemented to increase the attentional demand, and is comparable to daily life situations, e.g. talking on the phone while walking. Bock (2008) and Beurskens & Bock (2013) showed that walking and avoiding obstacles, while performing a visual secondary task resulted in a significant decrease in dual-task performance compared to situations in which other, non-visual secondary tasks were to be performed. Nonetheless, other studies that investigated the effects of non-visual secondary tasks during walking and/or avoiding obstacles found decreased dual-task performances, especially in elderly and patient groups (Chen et al., 1996; Kim & Brunt, 2007; Hausdorff et al., 2008; Hegeman et al., 2012; Smulders et al., 2012). Thus, identifying the effects of a secondary task on the SAS may provide additional information about the obstacle-avoidance skills.
The current study aimed to examine which SAS are used when obstacles are suddenly presented with a fixed ART (0.5 s) at varying obstacle locations with respect to pFP, and whether these responses differ when walking with a simultaneous secondary cognitive task (counting backwards in steps of three; dual-task condition) or without (single-task condition). Both young and elderly adults were included in the experiment to get more insight into age differences in SAS, since elderly have been shown to (1) use the avoidance strategies in a different manner (Chen et al., 1994; Weerdesteyn et al., 2005) and (2) have lower dual-task performance than young adults (Chen et al., 1996; Kim & Brunt, 2007). Virtual obstacles were presented at different locations (viz., multiple locations at shorter and longer distances than pFP position) to determine the influence of obstacle location on SAS and avoidance success rates. A maximum of one obstacle was presented per trial. The type of response (LSS/SSS), success rate, and secondary-task performance were determined.
To this end, the Interactive Walkway (IWW) was used. This is a new set-up that measures 3D-human-kinematics with Kinect’ cameras (Microsoft) during overground walking. Thus it allows participants to adapt their gait speed and walk as naturally as possible. This is an advantage over treadmill walking. Nagano et al. (2013) found that during treadmill walking especially elderly have increased gait variability compared to overground walking. The authors suggested that elderly change the way they walk on a treadmill to maintain their balance. Moreover, Wass et al. (2005) found that elderly were not able to get familiarized with walking on a treadmill after fourteen minutes (which was the duration of the measurement), because their lower limb kinematics were significantly different from overground walking. Besides, two-third of the subjects needed handrail support during the whole experiment. In contrast, the lower limb kinematics of young adults were comparable to overground walking after six minutes of walking on a treadmill (Matsas et al., 2000).
Regarding the MDP, the strong prediction for the used strategies (LSS/SSS) at varying obstacle locations with respect to pFP position is a block function (grey area in Figure 1). However, in view of natural variability this was approached using an S-shaped curve (or logistic function; red line in Figure 1). Since studies showed that LSS is the favored strategy, the curve was expected to shift towards the right (illustrated by the arrow). Elderly demonstrated more cautious crossing (Lowrey et al., 2009), and showed increased reaction times when avoiding obstacles compared to young adults (Weerdesteyn et al., 2005; George et al., 2007; Kim & Brunt; 2007). Since performing a SSS needs a quick reaction and abrupt deceleration in the walking direction compared to using the LSS, it was expected that elderly use more LSS than the young participants. Therefore, a greater shift to the right was expected for elderly participants than for young participants, implying that their obstacle avoidance would be less in line with the MDP. The reaction time was expected to increase for both groups in the dual-task condition compared to the single-task condition (Kim & Brunt, 2007; Hegeman et al., 2012; Smulders et al., 2012). Therefore, LSS could easier be implemented than SSS (Patla et al., 1999), hence decreasing adherence to the MDP. Overall, the performed strategies in the single-task condition were expected to be more in line with the MDP than in the dual-task condition. In addition using linear regression the relation between avoidance stride length and obstacle location was examined. The slopes and absolute intercepts provide information on the safety margin of successfully avoided obstacles, and would be exactly 1 if the obstacles would be avoided according to the MDP. Because some safety margin may be expected, deviations from this value were expected to be found. Intercepts were expected to be higher than |1|, since only successfully avoided obstacle trials were included in the regression. Slopes were expected to have a lower value than 1, because variation in obstacle location was expected to result in varying degrees of difficulty of the two strategies. Elderly participants were expected to have lower slope values (i.e. less consistency in safety margin over obstacle locations) and higher absolute intercepts values compared to young adults. The intercept and slope, and consequently the safety margin were expected to increase for both groups in the dual-task condition, assuming participants will increase their stride adjustment (Siu et al., 2008; Schulz, 2012). Furthermore, it was hypothesized that success rates were the lowest when the obstacle was located at pFP, since this location requires the largest foot displacement compared to locations closer by or farther away from pFP, assuming adherence to MDP. Regarding age differences, it was expected that young participants would have higher success rates, and better dual-task performance compared to elderly participants. In addition, the dual-task condition was expected to result in lower success rates compared to the single-task condition.
2. Methods
2.1. Participants
Fourteen healthy young (7 female, 24.3 ?? 2.0 years) and thirteen healthy elderly adults (9 female, 65.3 ?? 3.4 years) participated in this study. Beforehand, the elderly participants were interviewed by telephone about their mental and physical health (see Appendix I). They were included if their chance of falling during the experiment was considered minimal. None of the participants had any mobility problem, neurological or muscular disorder, or uncorrected visual impairment. All participants signed informed consent before participating. The local Ethics Committee had approved the experiment. Two female elderly quit during the experiment: one participant was not able to walk at a constant speed; the other participant was feeling dizzy during the first condition. Hence, the presented study explored the performance of fourteen young and eleven elderly participants (7 female, 64.7 ?? 3.4 years).
2.2. Experimental set-up
Participants were asked to walk multiple times over an 8.2-m long and 0.9-m wide walkway (Figure 2). Their movements were recorded using four Microsoft Kinect’ cameras (sample frequency: 30 Hz) that were placed along the left side of the walkway in such a way that the whole walkway was in view. Recordings were made in the x (anterior/posterior), y (medial/lateral), and z (inferior/superior) directions. A projector showed two lines on the floor that indicated the left and right border of the walkway. Since the Microsoft Kinect’ system was able to recognize a human body and its joints, no markers were needed. A videocamera (Kodak Playsport) was placed behind the end of the walkway to visually check the participant’s response to an obstacle.
2.2.1. Obstacle avoidance task
The software of the IWW allowed sudden presentation of obstacles at locations that depended on the predicted location of the individual’s pFP (i.e., predicted foot placement position in absence of an obstacle). The software determined timing and location of pFP online based on the ankle data. In this experiment, the distance between obstacle location and the pFP location was manipulated using seven different locations, either closer by (-21%, -14%, -7% of mean stride length), farther away (+7%, +14%, +21% of mean stride length), or at the position of pFP (0%; Figure 3).
The obstacle was colored red and had a length of 0.30 m. It was located at either the left or the right half of the walkway (obstacle width: 0.45 m) to activate a response in the corresponding leg. pFP was estimated using the ankle position data in the x-direction, assuming a foot outline of 0.05 m to the back and 0.25 m to the front of this ankle coordinate (hereafter referred to as TargetpFP) (Figure 2). Based on this information, an obstacle could be presented at an unpredictable location (for an unpredictable stride) in the second half of the walkway as a function of pFP. A maximum of one obstacle was presented per trial, at one of the seven locations relative to pFP (cf. Figure 3).
2.2.2. Secondary task
A cognitive secondary task (counting backwards in threes) was employed to distraction the attention of walking. Audio recordings were made with a memo recorder (Olympus). The initial 3-digit numbers were randomly chosen to minimize a learning effect.
2.3. Procedure
Before the experiment started, participants practiced in five obstacle avoidance trials. Participants were instructed to walk at a constant gait speed during all walking trials. During practice trials, obstacles appeared at the moment of heel strike of the step preceding the targeted step, at the location of predicted foot placement (pFP) of the latter. Because stride times varied, ART was not controlled during the practice trials. After the practice trials, three pre-experimental obstacle avoidance trials were performed to determine mean stride length and stride time (excluding the avoidance stride) for personalisation of the ART (depending on participant’s stride time) and the seven obstacle locations relative to pFP (depending on participant’s stride length). See Appendix II.1 and II.2 for details about the associated calculations. The aim was to keep the ART constant (0.5 s) over all trials, but the degree to which this was achieved depended on the consistency of the participant’s stride times.
Next, the experiment proper was started. Participants first performed 70 single-task trials. Each obstacle location was presented eight times (8 repetitions x 7 locations = 56 obstacle trials). In one half of the obstacle trials, obstacle presentation was for the left leg, in the other half presentation was for the right leg (in random order). Dummy trials (trials without obstacles) were introduced to minimize predictability and anticipation (2 per obstacle location condition; 14 trials in total). Dummy and obstacle trials were randomly administered, as were the seven obstacle locations in the obstacle trials. Participants were instructed to walk at their preferred, constant gait speed from start to finish, and to avoid obstacles in the anterior-posterior direction. In order to see the obstacle they were advised to look at the walkway.
After a break, the secondary task (counting backwards in threes) was practiced five times and shortly after that three reference recordings of 15 s each were made while participants were sitting in a relaxed position. They were instructed to repeat the number given by the researcher and count backwards as fast as possible.
Next, the dual-task was practiced in the same way as the single-task. First, five trials were practiced, followed by three trials that were measured to calculate personalized ART and obstacle locations for this condition. Then 70 trials were performed, comprising 56 obstacle trials (8 per obstacle location, equally divided over the two legs) and 14 dummy trials (2 per obstacle-location condition) that were presented in random order. Participants were instructed to perform both tasks (avoiding obstacles and counting backwards) as well as possible. Again, participants were instructed to walk at their preferred gait speed, as was practiced in the five pre-experimental trials conducted for this task.
After the experiment was terminated, the distances from the ankle to the front/toe (??xANKLE-TOE) and rear/heel (??xANKLE-HEEL) of the shoe were determined, in order to establish whether the obstacle area (red area in Figure 2) was touched during the stance phase by the shoe (unsuccessful trial) or not (successful trial). Appendix II.3 provides a more detailed description of this procedure and analysis.
2.4. Pre-processing
2.4.1. Trial selection
Only the ankle’s x-position data were used for analysis. For each trial, the x-position data of both ankles were visually inspected in Matlab (version 7.9.0, The Mathworks Inc., US), complemented by inspection of the corresponding video recordings if necessary. In case the Kinect’s human-pose estimation algorithms had switched the left and right side, which sometimes happened in particularly the overlap region of two cameras, the corresponding samples were corrected or removed from the trial. Trials were excluded from further analysis if minimally 15 consecutive samples were missing (corresponding to at least half a second). Table 1 summarizes the percentages of excluded trials per Group and Task. For the remaining trials, ankle data were interpolated (if necessary) and filtered (second-order low-pass Butterworth filter, cut-off frequency: 5 Hz). Figure 4 shows a representative walking trial (interpolated and filtered) in which the participant started with the left leg, and illustrates how stride length and time were determined.
The ART of included obstacle trials was checked and only trials with an ART between 0.4 and 0.6 s were kept for further analysis (for a more detailed description, see Appendix III.1). Also the obstacle location (OL) relative to the position of pFP was evaluated prior to further analysis (for a more detailed description, see Appendix III.2). Although the variation is small when walking at a constant speed, every stride has a different length. Consequently, the location of the presented obstacle could differ a couple of centimeters from the pFP location as estimated by the IWW-software. Obstacles located at more than 25% distance from the absolute 21% location were excluded from further analysis (see Table 1).
Table 1. Percentages of excluded trials.
Group Single-task Dual-task
Young Error in data 2.07 2.07
ART < 0.4 s -25% <= OL <= 25% 25.64 5.07 0.78 40.90 6.47 0 OL < -25% OL > 25%
ART > 0.6 s -25% <= OL <= 25% 4.29 7.01 0 2.21 1.55 0 OL < -25% OL > 25%
0.4 s ‘ ART ‘ 0.6 s OL < -25% 12.36 10.90 OL > 25% 1.79 0
Total 59.01 64.10
Elderly Error in data 3.60 2.90
ART < 0.4 s -25% <= OL <= 25% 31.84 7.25 4.91 29.04 6.90 0.61 OL < -25% OL > 25%
ART > 0.6 s -25% <= OL <= 25% 9.22 6.58 0 5.26 7.68 0 OL < -25% OL > 25%
0.4 s ‘ ART ‘ 0.6 s OL < -25% 0 0 OL > 25% 0 0
Total 63.40 52.39
2.5. Data-analysis
2.5.1. Avoidance strategy
Both remaining ankle data and the corresponding video recordings were used to determine which avoiding strategy was used for all included obstacle trials. First, video recordings were examined to categorize the employed avoidance strategies: LSS when the target foot was placed after the obstacle, SSS when the stride target foot was placed in front of the obstacle, and ‘not avoided’ (NA) when the target foot was placed on the obstacle. Second, the ankle x-position data provided more precise information about the avoiding stride length with respect to stride lengths in the dummy trials. Because of the good between the IWW and video analysis (see Appendix IV), the further analyse were based solely on the ankle x-position data, as obtained for midstance after obstacle appearance (xAnkleAFTER). A stride was defined as LSS when the following requirements were met:
xAnkle_AFTER>Target_pFP (1)
The SSS was defined as:
xAnkle_AFTER MDLSS-NORM). Success rates (Success) were estimated for each of the two avoiding strategies (LSS and SSS), and was defined as:
Success= (# successfully avoided obstacles)/(# included trials with obstacles)*100% (10)
2.5.3. Secondary task (counting backwards in threes)
The audio files of all included trials were analysed in Matlab to determine the total time counted backwards. The participants started with the 3-digit number given by the researcher. Per trial the total time (s) was defined as:
t_TOTAL=’t_(LAST-END)- t’_(FIRST-END) (11)
where tLAST-END was the point in time at which the participant finished the last recorded response, and tFIRST-END was the point in time at which the participant was finished with the first response (i.e., repeating the number given by the researcher). In between these two time points, the total numbers of correct and incorrect responses were noted by hand (while listening to the recorded responses). Next, the total number of responses per trial was divided by the duration of the trial (yielding CBTEMPO, in responses per second) and the percentage of incorrect responses (CBERROR) was calculated over all obstacle trials.
CB_TEMPO=(total responses)/t_TOTAL (12)
CB_ERROR= (incorrect responses)/(total reponses) * 100% (13)
2.6. Statistical Analysis
2.6.1. Avoidance strategy
It was hypothesized that the percentage SSS of all obstacle trials followed a sigmoid curve with increasing normalized obstacle locations (as illustrated in Figure 1). Therefore, the overall percentage SSS was calculated for 15 consecutive, equally sized OLMD bins per Group x Task combination, and was fitted to the following logistic function:
%SSS= 100/(1+e^(-k(c-OL_MD)) ) (14)
where OLMD is the normalized obstacle location for both SSS and LSS (OLMD-SSS and OLMD-LSS), sorted from lowest (in front of the pFP-position) to highest (beyond the pFP-position). The variable c is the 50%-point in the curve, and variable k is an indicator of steepness at this point. If the curve would be completely in line with the MDP, c would be zero (i.e. at OLMD=0). Hypothetical values for k were calculated per Group x Task combination (young x single: -133.28; young x dual: -142.70; elderly x single: -160.29; elderly x dual: -151.92), assuming adherence to the MDP. These values differ between combinations, since the range of OLMD varied between groups.
??2-analyses were performed to determine overall goodness of fit between observed SAS frequencies and expected SAS frequencies based on the MDP, and to determine Group and Task effects on the goodness of fit. It was performed for the situations in which LSS was expected according to the MDP (i.e. when OLMD<0; observed number of LSS tested against total number of trials included for this situation), and when SSS was expected (i.e. when OLMD>0; observed number of SSS tested against total number of trials included for this situation). When significant, the employed SAS violated the MDP.
Since many trials had to be excluded (see Table 1), it was not possible to conduct analysis of variance with the factor obstacle location. Instead linear regression analysis was performed on ??SLMD-LSS and ??SLMD-SSS as a function of OLMD as obtained for all included trials in which the obstacle was successfully avoided. In total, eight linear regression analyses were performed: one for each Group (young, elderly) x Task (single-task, dual-task) x Strategy (LSS, SSS) combination. The intercept, slope and R2 were determined in Matlab.
2.6.2. Success rates
Success rates (Success) of participants with ‘ 15 included trials for the single-task and dual-task condition were checked for normality by examining the Shapiro-Wilk test, Z-scores of skewness and kurtosis, and histograms, which showed that the success rates were not normally distributed. Therefore, a Wilcoxon-rank sum test was chosen to determine the effect of Task (single-task, dual-task) on Success per Group. Next, Mann-Whitney U tests were performed to assess overall group effect on Success, and to assess differences between group per Task (single-task, dual-task) x Strategy (LSS, SSS). Alpha was 0.05 for all tests described above.
2.6.3. Secondary task
First, normality of CBTEMPO and CBERROR was checked in a similar fashion as the success rates. These tests showed that for both groups CBERROR violated the assumption of normality, while CBTEMPO did not. Next, median values of CBTEMPO (over trials per participant) were subjected to a 2×2 repeated measures ANOVA with Task (single-task, dual-task) as within-participants factor and Group (young, elderly) as between-participant factor. Homogeneity of variance between groups was checked using a Levene’s test.
A Wilcoxon rank-sum test was performed to determine the effect of Task (single-task, dual-task) on CBERROR, which was not normally distributed in both groups. A Mann-Whitney U test was assessed to determine an overall group effect on CBERROR. In addition, another Mann-Whitney U tests were performed to determine differences between groups per condition. Alpha was 0.05 for all tests described above.
3. Results
3.1. Avoidance strategy
The results of the logistic fit (Figure 6), ??2-tests, and regression analysis (Figure 7) are described in the sections below in relation to the factors Group and Task. The panels in Figure 6 show the distribution of SAS over the range of obstacle location, and the panels in Figure 7 show ??SLMD as a function of OLMD for both groups for each condition. The corresponding intercepts, slopes and R2 of the latter are summarized in Table 2. Overall, the R2-values for the logistic fit indicate high fits with the data, while R2-values for the regression analysis indicate low to medium fits of the data.
A ??2-test was performed to determine whether the use of a particular strategy (either LSS or SSS) was in line with the expectations based on the MDP. The test for of LSS was not significant, implying that no difference was demonstrated between the observed use of LSS (95%) and the expected use of LSS (100%) based on the MDP . In contrast, the test for SSS was significant, ??2=61.48, p<0.01, indicating the MDP did not apply when the use of SSS was expected (i.e. when OLMD>0). In the trials with OLMD>0, 56% of the avoidance strategies were SSS and 44% were LSS, while 100% SSS was expected.
3.1.1. Young vs. elderly participants
The logistic fits (Figure 6) showed high values for R2, indicating that the logistic curve provided a reasonable prediction of the type of strategy for specific obstacle locations. All values for the parameter c (50%-point) were positive, which is in agreement with the significant ??2-test for SSS. This suggests an overall preference for the use of LSS, and is not in agreement with the MDP. The 50%-point for the young participants was at a distance closer OLMD than the 50%-point for the elderly participants, which is an indication that the young participants’ avoiding strategies were more in line with the MDP than the strategies of the elderly. This is also underlined by the significant Group effect found for ??2-test for SSS, ??2(1)=57.24, p0, while elderly participants used LSS in 47% of the trials, indicating weaker adherence to the MDP for elderly participants. No significant Group effect was obtained for LSS (young: 96% LSS; elderly: 94% LSS; expected: 100% LSS).
Table 2 and Figure 7 show that all intercepts were larger than |1|, indicating a safety margin when OLMD is zero. Except for YDLSS all intercepts of the elderly were larger than those of the young participants, suggesting that safety margins were larger for elderly than for young adults. As was expected, all slopes were <1, implying that safety margins are higher when the SAS predicted by the MDP was used (i.e. the use of LSS (SSS) when LSS (SSS) is expected). The slopes for the young participants were steeper than the slopes for the elderly participants, except for the slope of YSLSS. This exception for YSLSS might be due to fact that the younger participants use LSS with high variability avoidance stride length (??SLMD), especially for obstacle locations when LSS is the preferred strategy as can been seen in Panel 7A. This may also explain the relatively low value for its corresponding R2. 3.1.2. Single- vs. dual-task For the logistic fits (Figure 6), only small differences between the two conditions were found. Whereas c in the dual-task was at an OLMD of 0.03 lower than in the single-task condition for the young participants, this point was 0.02 OLMD higher in the dual-task compared to the single-task condition for the elderly. Also the obtained values of k were marginally different between conditions. The ??2-test showed that Task had a significant effect on the use of SSS, ??2(1)=57.32, p0 the avoidance strategy was LSS, while this percentage was 47% for the dual-task. Again, no significant difference was found for LSS (i.e. when OLMD<0), indicating that for both single-task and dual-task conditions the use of LSS did not systematically deviate from MDP (single-task: 94% LSS; dual-task: 95% LSS; expected: 100% LSS). For the young participants, the results of the linear regression regarding the single- and dual-task conditions were largely in line with the expectations. Both single-task intercepts for the young participants were lower than the dual-task intercepts. Moreover, the regression line YDLSS in Panel 7B (dual-task condition) is almost parallel to the MD-line for LSS, whereas the slope YSLSS (Panel 7A; single-task condition) is much smaller. Comparison of Panels 7A and 7B suggests that this difference in slope was associated with an increased safety margin at obstacle locations farther from pFP (OLMD=0) in the dual-task condition. However, the slope of YDSSS was shallower than the slope of YSSSS, implying that the safety margin was more dependent on the obstacle location in the dual-task condition. The changes in intercepts and slopes were quite different for the elderly participants: LSS and SSS slopes decreased in the dual-task condition, whereas the LSS and SSS intercepts also decreased. However, these differences were minimal, suggesting that the additional task (i.e. counting backwards) hardly affected performance. 3.2. Success rate Figure 8 shows the mean percentage of successfully and unsuccessfully avoided obstacles per Group x Task x Strategy combination. Visual inspection indicates that elderly had lower success rates than the young participants for both conditions, which was expected. The non-parametric tests, however, demonstrated no significant effect for Group, Task, or Strategy on Success. ‘ Table 2. Linear regression analysis for successfully avoided obstacle trials. Group Task Strategy Name Intercept Slope R2 Young Single LSS YSLSS 1.26 0.71 0.55 SSS YSSSS -1.40 0.97 0.76 Dual LSS YDLSS 1.36 0.90 0.67 SSS YDSSS -1.42 0.82 0.54 Elderly Single LSS OSLSS 1.28 0.76 0.68 SSS OSSSS -1.56 0.77 0.48 Dual LSS ODLSS 1.27 0.73 0.62 SSS ODSSS -1.48 0.76 0.63 3.3. Secondary task 3.3.1 CBTEMPO Table 3 shows mean CBTEMPO for both groups for all conditions and the corresponding standard deviations. The ANOVA showed no significant effects. 3.3.2 CBERROR The means and standard deviations of CBERROR are summarized in Table 4. The Mann-Whitney U test did not yield an overall effect of Group on CBERROR. Moreover, no differences were obtained between the groups per Task (single-task, dual-task). The Wilcoxon rank-sum test revealed a tendency towards significance between the reference and obstacle conditions for the young participants, z= -1.84, p=0.07, r=-0.49. For the elderly participants, Task did not significantly affected CBERROR, probably due to high standard deviations. Table 3. Descriptives of CBTEMPO (total responses per second) for each group en experimental condition. ?? ”””; ””””””””””??. Group ?? ?? Reference Young 0.89 0.22 Elderly 0.83 0.22 Obstacle Young 0.93 0.14 Elderly 0.86 0.28 Table 4. Descriptives of CBERROR (error percentage) for each group and experimental condition. ?? = mean; ??= standard deviation. Group ?? ?? Reference Young 0.89 2.01 Elderly 1.25 2.31 Obstacle Young 2.23 2.49 Elderly 4.49 8.73 4. Discussion This study investigated the influence of obstacle location, dual-task interference and ageing on obstacle avoidance (SAS and success rate) during walking and the adherence of the applied SAS to the MDP. Overall, the distributions of SAS deviated from the distribution based on the MDP, showing a preference for LSS. The results also indicated that ageing and dual-task interference increased deviations from the MDP. Success rates and secondary-task performance did not differ between groups and tasks. The obstacle avoidance characteristics found in the present study are expected to match everyday life situations to a greater extent than characteristics obtained by research involving treadmill walking, because in the latter situations gait kinematics appear to be affected in elderly adults (Wass et al., 2005; Nagano et al., 2013). The IWW allows participants to walk overground, which is more natural than walking on a treadmill. Therefore, it was regarded a useful tool to investigate obstacle avoidance characteristics. 4.1. Adherence of obstacle avoidance to the MDP As was expected, all participants showed a preference for using LSS, which was demonstrated by the positive values found for c and the significant ??2-test for SSS. This latter result indicated that LSS was used significantly more frequently than expected at obstacle locations for which SSS was expected according to the MDP (i.e. when OLMD>0). At obstacle locations somewhat beyond zero, LSS remained the favoured strategy as indicated by the positive values for c of the logistic function. This implies that the participants adjusted their stride length more than necessary according to the MDP (supported the safety margin observed in the linear regressions), which led to the preference for using LSS over using SSS. At higher values for OLMD, however, only SSS was observed (cf. Figure 6 & 7).
These results are in contrast with the study of Weerdesteyn et al. (2005), who demonstrated that young adults used 50% LSS and 50% SSS. They argued that this result was perfectly in line with the expected distribution based on the MDP, although this interpretation was hampered due to methodological issues (see Introduction). The present study was to our knowledge the first study investigating the relation between MDP and SAS for multiple obstacle locations. It showed the importance of obstacle location with respect to pFP on the type of SAS, and therefore this should be taken into account when applying the MDP to the interpretation of the SAS. The way in which elderly participants avoided obstacles was less in line with the MDP than that of the young participants, as shown by the higher value found for c in the logistic functions of the elderly participants, and by the significant ??2-test that showed elderly participants used more LSS (47%) than young participants (41%) when 0% LSS was expected. This is in agreement with previous studies, which demonstrated that LSS was the favoured strategy for elderly adults (Weerdesteyn et al., 2005; Roerdink et al., 2009; Bank et al., 2011). An explanation for the decreased use of SSS for elderly participants is the need for a faster reaction time and quicker implementation compared to using LSS (Patla et al., 1999). Besides, the abrupt deceleration of the lower legs in the walking direction that is necessary to perform an SSS, is destabilizing the person if the rest of the body is not decelerated.
Deviations from the MDP are further underlined by the results of the regression analysis (Figure 7). All slopes were lower than 1 and all intercepts were higher than |1|, indicating the presence of a safety margin when obstacle avoidance was successful. This indicated reduced adherence to the MDP, according to which the following results were expected: slope=1, intercept=|1|, safety margin=0. When comparing the two groups, elderly participants showed higher absolute intercepts and smaller slopes, deviating more from the expectations according to the MDP. Lower slopes indicate that the safety margin was increased when using the expected SAS. Higher slopes that approach 1 (for example 0.97 found for SSS of the young adults in the single-task condition) indicated that a more constant safety margin was used over the whole range of obstacle locations. This may be partly explained by the slower coupling of visual information to a corresponding stride adjustment for the elderly adults (Weerdesteyn et al., 2005). When an obstacle is presented without a restricted response time, elderly adults look at the obstacle for a longer time than young adults do (Chapman & Hollands, 2006; Keller Chandra et al., 2011). Hence, elderly adults might need more response time than 0.5 s to estimate the necessary adjustment to avoid an obstacle (Uemura et al., 2011), and by a way of precaution they increase their stride adjustment (Siu et al., 2008; Schulz, 2012).
Contrary to the hypothesis and the Task effect revealed by the ??2-test, differences between the single- and dual-task regression lines and logistic functions were minimal for the elderly participants. Besides, the logistic functions of the young participants showed minimal differences, indicating that the preference of LSS was not increased further in the dual-task condition. The regression analysis of the young participants showed that the slope was increased for the LSS while it was decreased for the SSS in the dual-task condition. This latter observation was in line with the expectation and suggested that the SSS becomes more difficult to perform with attentional interference, which could be due to increasing reaction times with increasing task demands (Peper et al., 2012; Mazaheri et al., 2014). As previous studies showed, dual-tasking leads to behaviour that increases stability and decreases the risk of obstacle contact, suggesting a ‘posture-first’ strategy (Chen et al., 1996; Schrodt et al., 2004; Siu et al., 2008; Smulders et al., 2012). This was found for the young participants, since the safety margin increased in the dual-task condition. In contrast, since the present results indicated that the elderly participants showed such behaviour already in the single-task condition, it might explain the absence of a dual-task effect on type of strategy and safety margin for this group.
4.2. Success rates & secondary-task performance
The findings regarding success rates and secondary-task performance are not consistent with most previous studies, which showed decreased success rates and secondary-task performance due to ageing and/or dual-task interference (Chen et al., 1996; Weerdesteyn et al., 2005; Hausdorff et al., 2008). Compared to those studies, the elderly adults participating in this study were relatively young (??=64.7 vs. ??>70 years), showing similar success rates and secondary-task performance as the young participants. On the other hand, in the study of Smulders et al. (2012) healthy elderly adults (??=54 years) also did not show decreased success rates in the dual-task condition, but did show a decrease in secondary-task performance (reaction time). Possibly, the secondary-task used in the present study was too easy for some participants, especially for those with affinity for numbers and mathematics. A reaction-time task would be more neutral in this respect, and may therefore be recommended as secondary-task for future research.
Furthermore, the lack of a Task effect on the success rate and secondary-task performance could be due to a decreased preferred gait speed in the dual-task condition . With a decreased gait speed, the attentional demand of walking is reduced (Dubost et al., 2006), thereby increasing the attentional capacity available for other tasks. However, the effect of gait speed was not analysed in the current study, and its effect on success rate remains unknown. Nevertheless, the lack of difference in success rate and secondary-task performance might be the consequence of the given instructions to perform both tasks (i.e. obstacle avoidance and secondary-task) as good as possible to minimize prioritization for either task.
4.3. Limitations
A major limitation of this study was the large amount of excluded trials, because of data errors or deviations in ART and obstacle location from the prescribed values. This impaired the quality of the analysis, since an ANOVA for SAS with the factor obstacle location could not be performed . Furthermore, the single-task walking trials were always presented prior to the dual-task walking trials. These conditions were not randomized because it was expected that elderly would have difficulties when starting with the dual-task condition (i.e. resulting in very low dual-task performance and relatively high single-task performance). However, because the dual-task condition was always presented secondly, a potential difference between the two conditions may have been obscured due to a learning effect.
4.4. Conclusion
The present study showed the importance of obstacle location with respect to pFP when analysing the performed SAS and corresponding safety margins. Obstacle avoidance during walking was found to deviate from the expectations based on the MDP, suggesting foot placement is driven by other factors than its minimal displacement. In addition, deviations from the MDP were increased by ageing and dual-task interference. Future research should focus on other factors than MDP to unveil what underlies the choice of a particular SAS for obstacle avoidance during walking.
5. References
Bank, P. J. M., Roerdink, M. & Peper, C. E. (2011). Comparing the efficacy of metronome beeps and stepping stones to adjust gait: steps to follow! Experimental Brain Research, 209(2), 159’169.
Beurskens, R. & Bock, O. (2013). Does the walking task matter? Influence of different walking conditions on dual-task performances in young and older persons. Human Movement Science, 32(6), 1456’1466.
Bock, O. (2008). Dual-task costs while walking increase in old age for some, but not for other tasks: an experimental study of healthy young and elderly persons. Journal of Neuroengineering and Rehabilitation, 27(5) .
Chapman, G. J. & Hollands, M. A. (2006). Evidence for a link between changes to gaze behaviour and risk of falling in older adults during adaptive locomotion. Gait & Posture, 24(3), 288’294.
Chen, H. C., Ashton-Miller, J. A., Alexander, N. B. & Schultz, A. B. (1994). Age effects obstacles on strategies used to avoid. Gait & Posture, 2(3), 139’146.
Chen, H. C., Schultz, A. B., Ashton-Miller, J. A., Giordani, B., Alexander, N. B. & Guire, K. E. (1996). Stepping over obstacles: dividing attention impairs performance of old more than young adults. Journals of Gerontology, 51A(3), 116’122.
Dubost, V., Kressig, R. W., Gonthier, R., Herrmann, F. R., Aminian, K., Najafi, B. & Beauchet, O. (2006). Relationships between dual-task related changes in stride velocity and stride time variability in healthy older adults. Human Movement Science, 25(3): 372-382.
Hausdorff, J. M., Schweiger, A., Herman, T., Yogev-Seligmann, G. & Giladi, N. (2008). Dual-task decrements in gait: contributing factors among healthy older adults. Journal of Gerontology, 63A(12), 1335’1343.
Hegeman, J., Weerdesteyn, V., van den Bemt, B., Nienhuis, B., van Limbeek, J. & Duysens, J. (2012). Dual-tasking interferes with obstacle avoidance reactions in healthy seniors. Gait & Posture, 36(2), 236’240.
Keller Chandra, S., Bockisch, C. J., Dietz, V., Hegemann, S. C. A., Straumann, D. & van Hedel, H. J. A. (2011). Gaze strategies for avoiding obstacles: Differences between young and elderly subjects. Gait & Posture, 34(3), 340’346.
Kim, H. D. & Brunt, D. (2007). The effect of a dual-task on obstacle crossing in healthy elderly and young adults. Archives of Physical Medicine and Rehabilitation, 88(10), 1309’1313.
Lowrey, C. R., Watson, A. & Vallis, L. A. (2007). Age-related changes in avoidance strategies when negotiating single and multiple obstacles. Experimental Brain Research, 182(3), 289’299.
Markle-Reid, M., Browne, G., Gafni, A., Roberts, J., Weir, R., Thabane, L., Miles, M., Vaitonis, V., Hecimovich, C., Baxter, P. & Henderson, S. (2010). A cross-sectional study of the prevalence, correlates, and costs of falls in older home care clients ‘at risk’ for falling. Canadian Journal on Ageing, 29(1), 119’137.
Matsas, A., Taylor, N. & McBurney, H. (2000). Knee joint kinematics from familiarised treadmill walking can be generalised to overground walking in young unimpaired subjects. Gait & Posture, 11(1), 46’53.
Mazaheri, M., Roerdink, M., Bood, R. J., Duysens, J., Beek, P. J. & Peper, C. E. (2014). Attentional costs of visually guided walking: effects of age, executive function and stepping-task demands. Gait & Posture, 40(1), 182’186.
Nagano, H., Begg, R., Sparrow, W. & Taylor, S. (2013). A comparison of treadmill and overground walking effects on step cycle variability and asymmetry in young and older individuals. Journal of Applied Biomechanics, 29(2), 188-193.
Patla, A. E., Prentice, S. D., Rietdyk, S., Allard, F. & Martin, C. (1999). What guides the selection of alternate foot placement during locomotion in humans. Experimental Brain Research, 128(4), 441’450.
Peper, C. E., Oorthuizen, J. K. & Roerdink, M. (2012). Attentional demands of cued walking in healthy young and elderly adults. Gait & Posture, 36(3), 378’382.
Roerdink, M., Lamoth, C., van Kordelaar, J., Elich, P., Konijnbelt, M., Kwakkel, G. & Beek, P. J. (2009). Rhythm perturbations in acoustically paced treadmill walking after stroke. Neurorehabilitation and Neural Repair, 23(7), 668’678.
Schrodt, L. A., Mercer, V. S., Giuliani, C. A. & Hartman, M. (2004). Characteristics of stepping over an obstacle in community dwelling older adults under dual-task conditions. Gait & Posture, 19(3), 279’287.
Siu, K. C., Lugade, V., Chou, L., Van Donkelaar, P. & Woollacott, M. H. (2008). Dual-task interference during obstacle clearance in healthy and balance-impaired older adults. Ageing Clinical and Experimental Research, 20(4), 349’354.
Smulders, K., van Swigchem, R., de Swart, B. J., Geurts, A. C. & Weerdesteyn, V. (2012). Community-dwelling people with chronic stroke need disproportionate attention while walking and negotiating obstacles. Gait & Posture, 36(1), 127’132.
St George, R. J., Fitzpatrick, R. C., Rogers, M. W. & Lord, S. R. (2007). Choice stepping response and transfer times: effects of age, fall risk, and secondary tasks. Journal of Gerontology, 62A(5): 537-542.
Tinetti, M., Speechley, M. & Ginter, S. (1988). Risk factors for falls among elderly persons living in the community. New England Journal of Medicine, 319(26), 1701’1707.
Uemura, K., Yamada, M., Nagai, K. & Ichihashi, N. (2011). Older adults at high risk of falling need more time for anticipatory postural adjustment in the precrossing phase of obstacle negotiation. Journals of Gerontology, 66A(8), 904’909.
Van Ooijen, M. W., Heeren, A., Smulders, K., Geurts, A. C. H., Janssen, T. W. J., Beek, P. J., Weerdesteyn, V. & Roerdink, M. (2013). Gait adjustments and associated attentional demands after gait adaptability training after stroke. (under review)
Wass, E., Taylor, N. F. & Matsas, A. (2005). Familiarisation to treadmill walking in unimpaired older people. Gait & Posture, 21(1), 72’79.
Weerdesteyn, V., Nienhuis, B., Hampsink, B. & Duysens, J. (2004). Gait adjustments in response to an obstacle are faster than voluntary reactions. Human Movement Science, 23(3-4), 351’363.
Weerdesteyn, V., Nienhuis, B. & Duysens, J. (2005). Advancing age progressively affects obstacle avoidance skills in the elderly. Human Movement Science, 24(5-6), 865’880.
Weerdesteyn, V., Rijken, H., Geurts, A. C. H., Smits-Engelsman, B. C. M., Mulder, T. & Duysens, J. (2006). A five-week exercise program can reduce falls and improve obstacle avoidance in the elderly. Journal of Gerontology, 52A(3), 131’141.
Weerdesteyn, V., Nienhuis, B. & Duysens, J. (2008). Exercise training can improve spatial characteristics of time-critical obstacle avoidance in elderly people. Human Movement Science, 27(5), 738’748.
‘
APPENDIX I QUESTIONNAIRE INCLUSION/EXCLUSION CRITERIA ELDERLY
(Dutch version was used in the present study)
Research title Obstacle avoidance in elderly during walking: step adjustment strategies, success rates, attentional interference
Name participant
Date of Birth
Telephone number
Researchers Anne van den Berg (supervisors: Melvyn Roerdink & Lieke Peper)
Date
Exclusion after considering the safety of the participant, when one or more questions are answered with ‘yes’.
Question No Yes Remarks
1 What is your age? (exclusion if 75) Age =
2 Do you have osteoarthritis?
3 Do you have rheumatoid arthritis in your legs?
4 Do you have an artificial joint (hip of knee)?
5 Do you have difficulties to fully extend your knees?
6 Do you have difficulties to fully bend your hips and move your thigh forward or sideward?
7 Do you have Parkinson’s disease?
8 Do you suffer from twinkles or insensibility in your hands, feet or legs (e.g. because of neuropathy or diabetes)?
9 Did you break your leg last year? (exclusion if 1 a week and < 3 month) 19 Did you fainted in the past 6 months? (exclusion if this was > 1 with possible vestibular, neurological or cardiovascular)
20 Did you use sleeping pills in the last week (no use of sleeping pills 48 hours before experiment)
21 Did you use antidepressants last week?
22 Did you use beta blockers last week?
23 Do you have hearing problems?
‘
APPENDIX II PRE-EXPERIMENTAL PROCEDURES
1. ART
The IWW software used online determined heel-strike events to trigger obstacle presentation. A delay in samples was used to personalize the ART. A Matlab function estimated the sample of peaks of subtracted left and right ankle interpolated and filtered (second order Butterworth filter, 5 Hz low-pass) x-position data (??x, Figure 4A). Figure 4 shows a walking trial of a participant whose initial step was with his left leg. The interval between two subsequent maximum peaks (MAX) in Figure 4A roughly indicate the stride time of the left leg, and two subsequent minimum peaks (MIN) indicate the stride time of the right leg. Thereafter, the following equations were used to calculate the sample delay:
ST(i)=((MAX(i+1)- MAX(i)) +(MIN(i+1)- MIN(i)))/(2*fs) (a)
of which i is the number of a peak
The stride time (ST(i)) of one left and right side were calculated for multiple strides (minimal 5) that were not influenced by obstacles. So dividing the sum of the two strides times by two times the sample frequency (fs; 30 Hz) results in a stride time in seconds. In equation (b) mST equals the mean of all ST(i) calculated in equation (a).
Samples= (mST-ART)/(1/fs)-6 (b)
The IWW-software used a sensitivity of 6 samples for online identification of a heel-strike. These 6 samples needed to be subtracted in equation (b). The number of samples calculated using equation (b) was an input for the IWW-software to standardize ART (under the assumption that stride times would remain similar throughout the experiment).
2. Obstacle location
The obstacle locations were at a distance of -21%, -14%, -7%, 0%, +7%, +14%, +21% of mean stride length from the pFP position. First, stride lengths (minimal 5) were calculated in a similar fashion as ST(i), which were used to calculate mean stride length (mSL). However, in this case the x-positions at the maxima and minima were used (cf. Figure 4B):
SL(i)=(‘(x’_(MAX(i+1))-x_MAX(i) ) + ‘(x’_(MIN(i+1))-x_MIN(i) ))/2 (c)
Next, the seven obstacle locations that are farther away by or closer by from pFP position were determined as percentages of mSL.
3. Outline shoe
To calculate the outline of the shoes, four measurements consisted of stepping into pre-programmed stepping stones with the length of the person’s shoes. Recordings started when the participant stood still with two shoes in two separate stepping stones, which were 0.30 m apart from each other. It was instructed to fit the shoes as well as possible in the stepping stones. One measurement was in view of one Kinect camera. Thereafter, the distance of the ankle to the front/toe (??xANKLE-TOE) and rear/heel (??xANKLE-HEEL) of the shoe was calculated. These distances were used to determine for all the obstacle trials whether the shoe was in the obstacle area or not, and subsequently to determine the individual’s success rates.
‘
APPENDIX III PRE-PROCESSING
1. ART
The mean (mSTdummy) and standard deviation (sdSTdummy) of the left and right stride times were calculated, using only data of dummy trials (in which no obstacle appeared). The nearest sample of a peak (MAX or MIN: respectively when obstacle was located at the left or right side, see Figure 4A) before and after obstacle appearance was assessed. The ART was then estimated using equation (d).
ART= t(‘PEAK’_BEFORE)+ mST_DUMMY-t_START- 6*1/fs (d)
The ART was calculated in seconds; t(PEAKBEFORE) is the time point of either MAX or MIN before obstacle appearance; tSTART is the time point of obstacle appearance. Trials with an ART below 0.4 s and above 0.6 s were not used for data-analysis (see Table 1 for percentages of excluded trials).
2. Obstacle location
Before estimating the obstacle location, the mean (mSLDUMMY) and standard deviation (sdSLDUMMY) of all stride lengths in the dummy trials were calculated. In equation (e) the obstacle location (OL, in m) relative to pFP is calculated.
OL=Target_pFP- x’Ankle’_BEFORE+ mSL_DUMMY (e)
where xAnkleBEFORE is the ankle position during midstance before obstacle appearance. Trials with an obstacle location at a distance of 25% farther than the absolute 21%-location were excluded (see Table 1).
‘
APPENDIX IV CORRESPONDENCE BETWEEN VIDEO ANALYSIS AND IWW-DATA ANALYSIS
To validate the use of IWW-data, adjustment strategies and whether the obstacle was successfully avoided were noted by hand during inspection of the video recordings. Per group mean correspondence percentages were calculated (see Table 5 below). When ‘NA’ was noted during video inspection, either LSS or SSS was performed according to Equations (1) and (2). Therefore, ‘NA’ of the recordings matched with both LSS and SSS as was calculated using the IWW-data.
Correspondence=(# matches video and IWW)/(# total included obstacle trials )*100% (f)
Note that only one video camera was used, and distinguishing successful from unsuccessful avoidance based on visual inspection was questionable for some trials (i.e. when foot placement position was close to the obstacle location). High correspondence percentages were found, and therefore IWW-data was used for the analysis.
Table 5. Resemblance percentages for IWW- and video data of the avoidance (successful/unsuccessful) and strategy (LSS/SSS).
Avoidance Strategy
Young Elderly Young Elderly
Correspondence (%) 97 95 100 100