- Email: [email protected]
Shoulder pain and time dependent structure in wheelchair propulsion variability
Medical Engineering and Physics 38 (2016) 648–655
Contents lists available at ScienceDirect
Medical Engineering and Physics journal homepage: www.elsevier.com/locate/medengphy
Shoulder pain and time dependent structure in wheelchair propulsion variability Chandrasekaran Jayaraman a, Yaejin Moon b, Jacob J. Sosnoff b,∗ a b
Department of Industrial and Enterprise Systems Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
a r t i c l e
i n f o
Article history: Received 13 November 2015 Revised 26 February 2016 Accepted 3 April 2016
Keywords: Shoulder pain Nonlinear analysis Variation Compensatory mechanisms Entropy Manual wheelchair propulsion Complexity Human motor control Repetitive loading
a b s t r a c t Manual wheelchair propulsion places considerable repetitive mechanical strain on the upper limbs leading to shoulder injury and pain. While recent research indicates that the amount of variability in wheelchair propulsion and shoulder pain may be related. There has been minimal inquiry into the fluctuation over time (i.e. time-dependent structure) in wheelchair propulsion variability. Consequently the purpose of this investigation was to examine if the time-dependent structure in the wheelchair propulsion parameters are related to shoulder pain. 27 experienced wheelchair users manually propelled their own wheelchair fitted with a SMARTWheel on a roller at 1.1 m/s for 3 min. Time-dependent structure of cycle-to-cycle fluctuations in contact angle and inter push time interval was quantified using sample entropy (SampEn) and compared between the groups with/without shoulder pain using nonparametric statistics. Overall findings were, (1) variability observed in contact angle fluctuations during manual wheelchair propulsion is structured (Z=3.15;p<0.05), (2) individuals with shoulder pain exhibited higher SampEn magnitude for contact angle during wheelchair propulsion than those without pain (χ 2 (1)=6.12;p<0.05); and (3) SampEn of contact angle correlated significantly with self-reported shoulder pain (rs (WUSPI) =0.41;rs (VAS) =0.56;p<0.05). It was concluded that the time-dependent structure in wheelchair propulsion may provide novel information for tracking and monitoring shoulder pain. © 2016 IPEM. Published by Elsevier Ltd. All rights reserved.
1. Introduction Nearly 90% of the 3.6 million wheelchair users in the United States use a manual wheelchair for their mobility needs [1]. The use of a manual wheelchair places considerable mechanical strain on the upper limbs. Unfortunately, the human upper limb is not suited for such repetitive loading. The repetitive loading predisposes manual wheelchair users (mWCUs) for upper limb pathology. Indeed it has been shown that up to 70% of mWCUs report upper limb pain [2–5]. Furthermore, even in mWCUs who do not report pain, there is evidence of degenerative changes in the shoulder [6] suggesting that it is just a matter of time before these asymptomatic individuals will experience pain. Upper limb pain in mWCUs has been linked to difficulty performing activities of daily living [5], decreased physical activity and decreased quality of life [7]. Overall, any loss of upper limb function due to pain adversely impacts the independence and mobility of manual wheelchair users. It has been speculated that a decrease Abbreviations: SampEn, Sample Entropy; MWCUs, Manual wheelchair users; WUSPI, Wheelchair user shoulder pain index; VAS, Visual analog scale. ∗ Corresponding author. Tel.: +1 2173339472; fax: +1 217 2447322. E-mail address: [email protected] (J.J. Sosnoff). http://dx.doi.org/10.1016/j.medengphy.2016.04.005 1350-4533/© 2016 IPEM. Published by Elsevier Ltd. All rights reserved.
in independence and mobility results in greater health care costs and an increased risk for secondary morbidity (cardiovascular disease, obesity, etc.) [8, 9]. Although it has been proposed that propulsion biomechanics are related to shoulder pain in MWCUs, evidence is inconclusive [10]. Average spatiotemporal parameters of wheelchair propulsion (e.g. contact angle and push time) do not distinguish between those with and without pain. However, recent research indicates that spatial-temporal variability in wheelchair propulsion mechanics (e.g. push time and peak force during push) is related to shoulder pain [11]. This association between motor variability and pain is consistent with observations from occupation biomechanics and human motor control literature [12–15]. Although promising, a limitation of this research [11], is that it has only focused on the amount of variability in wheelchair propulsion. This approach seemingly ignores fluctuations over time (i.e. timedependent structure) inherent in physiological output [16]. Based on the tenets of the loss of complexity hypothesis of aging [16], it has been theorized that the time-dependent structure of motor output is a marker of physiological complexity and provides novel information concerning the health of the musculoskeletal system [14,17]. Specifically, it has been proposed that musculoskeletal injury leads to fluctuations in movement that are
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
649
Table 1 Demographic information. Comparison of demographic information between the groups with and without pain. Statistically significant group differences were observed for the pain scores between the groups (p<0.05). No significant group differences was observed between the groups in age, body weight, gender and experience using wheelchair. Demographic variables
Pain (n=13) Mean(SD)
No Pain (n=14) Mean(SD)
Age (Yrs) Body weight (LB) Experience using wheelchair (Yrs) Gender (F/M)
28.23(12.52) 164.50(56.39) 15.84(11.25) 6/7 Spina Bifida(n=5) T6-T12 (n=5) L1-L4 (n=1) Amputee- double (n=1) Sacral agenisis (n=1) 22.84(21.27) 4.24(2.8)
21.21(4.92) 131.17(36.96) 13.64(5.15) 6/8 Spina Bifida (n=4) T6-T12 (n=6) L1-L4 (n=1) Amputee-single (n=1) Arthrogryposis (n=1) C7 (n=1) 3.28(5.02) 0.36(0.96)
Injury demographics WUSPI∗ VAS∗ ∗
p<0.05
more structured. Two widely reported methods to measure the structure in fluctuation of movements are approximate entropy and sample entropy (SampEn) [18]. For instance, Tochigi et al., 2012, [19], utilized SampEn to analyze the structure of variability in gait in individuals with knee osteoarthritis with and without pain. Consistent with the loss of complexity hypothesis, it was reported that the pain group had significantly lower SampEn values when compared to those without pain. In a similar fashion, researchers have successfully employed approximate entropy to study the gait pattern related to anterior cruciate ligament (ACL) injury [20] and reported lower approximate entropy values in the knee with ACL deficiency compared to the healthy knee. Presently there is no information regarding the time-dependent structure of variability in the context of shoulder pain in mWCUs. Consequently the purpose of this investigation was to examine if the time-dependent structure in wheelchair propulsion parameters are related to shoulder pain. The goals of this investigation were twofold, (H1) to determine whether the time-dependent structure observed in wheelchair propulsion variability is structured, and (H2) to determine if individuals propelling a manual wheelchair with shoulder pain will demonstrate lower degree of time-dependent structure in fluctuations in their wheelchair propulsion parameters than those without shoulder pain. To test these hypotheses, two wheelchair propulsion parameters were analyzed, namely, contact angle and inter push time interval.
2. Methods
The participants were categorized as belonging to pain/no pain group based on the self-reported shoulder pain status (“Yes”/”No”). Participants also rated their current level of shoulder pain for each shoulder using a 10 cm visual analog scale (VAS) [21]. A VAS score of zero indicated no shoulder pain and higher score indicate greater shoulder pain at the time of testing. Participants also rated their shoulder pain using the wheelchair user’s shoulder pain index (WUSPI) [22]. WUSPI is a 15-item questionnaire. Each item is rated between 0 and 10, with 0 representing no interference with daily living functional activities and 10 representing complete interference with functional activities during the past week due to shoulder pain. The total score is the sum of scores of all the 15 items put together. Total score ranges from 0 (no pain)-150 (maximum limitations to daily activities due to pain) [22]. 2.2. Data collection and instrumentation 2.2.1. Kinetics On completion of the informed consent procedures, each participant’s wheelchair was fitted bilaterally with SMARTwheels (Three Rivers Holdings, LLC, Mesa, AZ, USA) and secured to a single roller dynamometer system using a four-point tie-down system and a flywheel system [23]. The SMARTwheels measures threedimensional forces and torques applied to the push rim. Participants were provided sufficient time to acclimate themselves with the dynamometer setup prior to the actual trial. Participants were asked to propel at constant speed of 1.1 m/s for 3 min [24,25]. A speedometer was used to provide real-time feedback to the participants during the three minute propulsion. The SMARTwheel data was post-processed using a custom developed MATLAB routine.
2.1. Participants 2.3. Data post-processing Wheelchair propulsion data from 27 experienced adult mWCUs was analyzed for this study. This dataset is a subset of data collected for a larger study focusing on wheelchair propulsion and shoulder pain [11]. Inclusion criteria for this secondary analysis were (1) more than one year of manual wheelchair experience; (2) between 18–64 years of age; and (3) trials data containing at least 106 wheelchair propulsion cycles. All procedures were approved by the local institutional review board at University of Illinois, Urbana-Champaign. Upon arrival to the laboratory, the data collection procedures were thoroughly explained and any questions the participants had regarding the protocol were addressed. Participants were informed that participation in the study was voluntary. The participants then provided their written informed consent and demographic information (Table 1).
SmartWheel data were collected at a sampling rate of 100 Hz and digitally filtered with an eighth-order, zero-phase, low-pass Butterworth filter with 20 Hz cutoff frequency [10]. The start and end of a propulsion cycle was defined when the push-rim moment (Mz) was above and below 1 Nm respectively [26]. To reduce the transient effects, data belonging to the first five propulsion cycles were not included for this analysis [27]. For consistency, the number of data cycles analyzed for each participant was maintained constant at 100 cycles (i.e. starting from the 6th cycle to 105th cycle of a SmartWheel data, Fig. 1(a)). The contact angle and resultant forces at hand-rim were extracted for each participant (Fig. 1(b)). Following this, the inter push time interval between peak resultant force between pushes were extracted (Fig. 1(b)). Thus, this process yielded two time series from each participant wheelchair
650
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
Fig. 1. Data processing and variable extraction for SampEn calculation. (a) resultant force time series at hand-rim showing the steady state portion of the time series used for all calculations; (b) a magnified view of two consecutive sample resultant force cycles at steady state to show the details of the variables extracted. The contact angle is defined as the angular measure between the start and end of contact. The inter push time interval between peak to peak resultant force was extracted from cycle-to-cycle during steady state to create the time series containing the inter push time interval fluctuations. The cycle-to-cycle contact angle magnitude during the steady state was extracted to create the time series of contact angle fluctuation.
propulsion data namely, (1) a time series of cycle-to-cycle contact angle fluctuation (Fig. 2(a)) and (2) a time series of cycle-to-cycle inter push time interval fluctuation between peak resultant force during consecutive push phases (Fig. 2(b)). For brevity, the variable, cycle-to-cycle inter-push time interval is termed inter push time interval within the manuscript. SampEn was then computed for each time series (see Section 2.4). A custom developed MATLAB code was used to accomplish all data post-processing. Based on the VAS scores, for the shoulder pain group, the data belonging to the side with greatest shoulder pain level was analyzed (right (n=11) and left (n=2)), while the data belonging to the dominant hand (right (n=12) and left (n=2)) was analyzed for the group without shoulder pain. 2.4. Time dependent structure in variability Complexity analysis quantifies the time dependent regularity and predictability of a time series. There are various nonlinear dynamics tools to measure the complexity of a physiological time series [28]. The choice of a specific tool depends on the characteristics of data being analyzed. In this investigation, we utilized SampEn, a widely utilized approach for studying time-dependent structure in short time series [18,19]. SampEn is a nonlinear metric used for measuring the regularity of a time series. The magnitude of the SampEn is an indication of the degree of regularity or irregularity of a particular time series. SampEn values typically ranges from 0 to 4. Values closer to 0 are consistent with greater regularity, such as a sinewave, while values nearing 4 represent greater irregularity such as pink noise [29]. Higher values of SampEn indicate that the time series is more unpredictable (i.e. unstructured). Details of the SampEn algorithms and input parameters used are available elsewhere [30,31]. MATLAB codes obtained from http: //www.physionet.org/physiotools/, [32] was used for SampEn cal-
culation. The performance (SampEn magnitudes) of the software program was tested with synthetically generated benchmark signals (for more details please see supplementary materials: Section A). This performance test was conducted to ascertain the validity of the SampEn values obtained from the physionet.com SampEn code. The tests revealed that the SampEn magnitudes obtained from the software program provided by physionet.com is comparable with those reported in literature (please see supplementary materials: Section A). Based on this validation, the SampEn code from physionet.org was used for calculating SampEn in this analysis. The parameter we choose for the SampEn analysis were (m=2, r=0.2) for the contact angle time series and (m=2,r=0.15) for the inter push time interval time series. The justifications for selecting these parameter choices are discussed in the supplementary materials (Section B). From each participant’s post-processed data, the SampEn for the original time series belonging to the contact angle and the inter push time interval were calculated. To statistically establish if the SampEn measure obtained from the original time series was indeed structured and not random, a simple surrogate analysis was carried out [33]. This surrogate analysis was conducted to test H1. Surrogate data sets were generated by randomly shuffling the original time series in order to determine whether the original time series was structured (Fig. 3). Such random shuffling preserves the distributional statistics (i.e., mean, standard deviation, and higher moments) between the original time series and the corresponding surrogate time series contain the same elements; however, the time-dependent sequential ordering is destroyed [34]. Each participant’s original time series was randomly shuffled 100 times to produce a pool of corresponding surrogate data. This surrogate procedure resulted in a pool of hundred surrogate time series for each original time series. The SampEn values were computed for each of the 100 surrogate time series in the pool and then averaged to generate a mean SampEn for the surrogate pool
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
651
Fig. 2. Sample contact angle and inter push time interval time series extracted from participants with and without shoulder pain from which the SampEn was calculated. (a) the cycle-to-cycle contact angle fluctuation at hand-rim for 100 pushes; (b) the cycle-to-cycle inter push time interval fluctuation at hand-rim for 100 pushes.
(Fig. 3). This procedure was repeated for every original time series. Finally non-parametric pair-wise tests comparing the SampEn values between each original time-series with the corresponding mean SampEn obtained from its surrogate pool was conducted. If the SampEn value of the original time series was significantly (statistically) smaller than its corresponding surrogate counterpart this indicates that the original data is not randomly derived and therefore its variability exhibits time-dependent structure [35]. Qualifying the surrogate test (H1), is a necessary step prior to testing H2. Additionally, the SampEn magnitudes can be influenced by nonstationarity in the original time series analyzed [18]. To check for this influence, the SampEn for both the contact angle and inter push time interval time series were also calculated from differenced versions of original time series. Pair-wise tests revealed no significant differences between the SampEn calculated from the original and their differenced versions (for more details please see supplementary materials: Section C). Thus the SampEn from the original time series were used for all the statistical analyses. 3. Statistical analysis IBM SPSS (version 21) was used for conducting all the statistical tests. The significance level for all the statistical tests were set at p≤0.05. All descriptive values are reported as mean (SD) unless otherwise noted. The between group factor for all the statistical tests was set as shoulder pain (with pain = 1; No pain = 0). Normality checks for data distribution using Shapiro–Wilk tests revealed that the age, self reported WUSPI scores, VAS scores, mean inter push time interval, mean speed and the SampEn required non-parametric statistical tests. 3.1. Demographics Participant demographics information (age, body weight, and manual wheelchair propulsion experience in years) and self reported scores (WUSPI & VAS) were treated as independent variables.
A series of two tailed Mann-Whitney U tests were conducted to check if demographic variables (age and self reported shoulder pain (WUSPI; VAS)) were significantly different between groups. A series of two tailed independent t-tests were conducted to check if significant between group differences in existed in body weight and wheelchair propulsion experience. A Chi-squared tests was conducted to determine if the sample gender composition significantly differed between the groups with and without shoulder pain. 3.2. Mean wheelchair propulsion variables In addition to the SampEn of contact angle and inter push time interval, mean propulsion variables that have been suggested to be associated with shoulder pain where also extracted from the SmartWheel data. Specifically, contact angle, peak resultant force at the hand-rim, mean inter push time interval between peak resultant force and mean push speed were determined with established procedures [36]. A series of two tailed Mann–Whitney U test were conducted to check if the mean inter push time interval and mean push speed were significantly different between groups with and without shoulder pain. A series of two tailed t-tests were performed to check if statistically significant group differences existed in other mean wheelchair propulsion variables (i.e. contact angle, peak resultant force at hand-rim). The mean wheelchair propulsion variables were verified to benchmark the data with previous literature. 3.3. SampEn To test if the SampEn of the original time series were significantly different from their surrogate counterpart (i.e. H1) separate pair-wise Wilcoxon signed rank tests were conducted for the contact angle and inter push time interval. To investigate the main effect of shoulder pain (i.e. H2), the SampEn obtained from the original time series for the contact angle and inter push time
652
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
Fig. 3. The surrogate analysis process. The mean(SD) of the original and the surrogated time series pool are same. The surrogation procedure used was a simple random shuffling of data. A pool of 100 randomly shuffled surrogates was created from the original time series (Surrogate (i=1 to 100) . If (SampEn)original is statistically lesser (p<0.05) than (MeanSampEn)Surrogate , then the time dependant structure observed in the original time series is not random and may have some meaningful insight.
interval were compared between the groups with and without shoulder pain using Kruskal–Wallis ANOVA’s. Spearman’s rank correlational analyses were conducted to investigate if the SampEn for the contact angle and inter push time interval was correlated with the self reported pain scores (WUSPI and VAS). 4. Results 4.1. Demographics No significant group differences in demographics were observed, (p’s>0.05; Table 1). Per design the shoulder pain group reported higher pain than the no shoulder pain group (WUSPI: [U=17; p<0.05]; VAS: [U=14.5; p<0.05];Table 1).
Table 2 Mean wheelchair propulsion variables at hand-rim. Comparison of mean wheelchair propulsion variables at hand-rim between the groups with and without pain. No significant group differences were observed between the groups. Mean propulsion parameters at handrim
Pain (n=13) Mean(SD)
No Pain (n=14) Mean(SD)
Peak resultant force (N) Contact angle (deg) Speed (m/s) Inter push time interval between peak resultant force (Sec)
69.82(23.84) 100.52(20.25) 1.1(0.04) 1.15(0.22)
60.23(19.72) 97.68(17.64) 1.1(0.06) 1.20(0.22)
4.3. SampEn of contact angle and inter push time interval
smaller than the SampEn from their corresponding surrogates: inter push time interval : [Z=2.59,p=0.009](m=2,r=0.15) ; [(mean(SD), IQR): (2.5(0.6),0.85)pain ; (2.2(0.4),0.63)No pain ]. contact angle:[Z=3.15,p=0.002](m=2, r=0.2); [(mean(SD), IQR): (2.1(0.2),0.27)pain ; (1.8(0.2),0.32)No pain ]. This indicates that the structure of variability found in wheelchair propulsion is not random but rather structured.
4.3.1. SampEn comparison between original and surrogated time series Wilcoxon signed rank pairwise tests revealed that the SampEn obtained from the original time series were statistically
4.3.2. Shoulder pain and time dependent structure of propulsion variability As illustrated in Fig. 4(a) the SampEn for contact angle was higher in the pain group than the no-pain group. This was
4.2. Mean wheelchair propulsion variables No significant differences were observed in mean propulsion variables between groups (p’s>0.05; Table 2).
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
Fig. 4. Groupwise mean (SD) comparison for SampEn. (a) the SampEn for contact angle was significantly greater (p<0.05) for the group with shoulder pain; (b) the SampEn for inter push time interval at hand-rim between groups with and without shoulder pain was not significantly different (p>0.05) between the groups with shoulder pain.
confirmed by Kruskal–Wallis test, χ 2 (1) = 6.12, p = 0.013. No significant main effect of shoulder pain was observed for the SampEn of inter push time interval (Fig. 4(b); p>0.05). A statistically significant moderate positive correlation was observed between the self-reported shoulder discomfort scores (WUSPI and VAS) and the SampEn of contact angle, [rs (25)=0.41, p<0.05]WUSPI ; [rs (25)= 0.56, p<0.05]VAS. There was no association between SampEn of inter push time interval and self reported pain scores (p>0.05). 5. Discussion This investigation explored the relation between shoulder pain and the structure of motor variability in spatiotemporal variables of manual wheelchair propulsion. Three novel observations were made, (1) variability observed in the fluctuations in contact angle and inter push time interval during manual wheelchair propulsion is structured, (2) individuals with shoulder pain exhibited higher SampEn during wheelchair propulsion compared to those without pain; and (3) SampEn measures correlated significantly with the amount of self-reported shoulder pain scores. Overall the observations suggest that SampEn analysis of wheelchair propulsion may provide novel insights for monitoring the development and treatment of shoulder pain in mWCUs. While the majority of wheelchair research focuses on mean kinematic and kinetics during propulsion [10,37] growing evidence suggests that variability measures could be a sensitive marker of shoulder pain in mWCUs [11]. Consistent with previous literature the mean propulsion variables were not able to identify any difference between mWCUs with and without shoulder pain [10,37].To date, the few wheelchair propulsion studies which incorporate
653
measures of intra-individual variability have exclusively focused on linear distributional statistics (standard deviation and coefficient of variation) and not measures of time dependent structure. The current investigation used SampEn, to quantify the time dependent structure in two propulsion parameters that are closely implicated with shoulder pain in mWCUs, namely, contact angle and inter push time interval between peak resultant forces, [10,37]. Surrogate analyses revealed that the time dependent structure observed in both measures were not random. This observation is consistent with reports from occupational ergonomics and motor control research which supports the tenet that the time dependent structure of variability in physiologic output contains meaningful insights concerning health and function [35,38,39] It is important to note that differences in wheelchair users with and without shoulder pain in SampEn were only observed in fluctuations in contact angle. Individuals with shoulder pain exhibited higher SampEn for fluctuations in contact angle during wheelchair propulsion in comparison to those without pain. This is distinct from examinations of the magnitude of variability which revealed that individuals with shoulder pain had lower amounts of variability in spatial and temporal parameters during propulsion as indexed by coefficient of variation [40]. The divergent results between time-dependent and magnitude metrics are consistent with the view that they are at least partially independent [41]. The null effect between pain groups for SampEn of inter push time interval could have resulted as an artifact of the experimental set-up. It is well established that the temporal structure of rhythmic movement is impacted when individuals are externally paced [39]. Consequently, the lack of group difference in the degree of time-dependent structure in fluctuations in inter push time interval could have been due to the real time visual feedback of speed provided. It is important to note that the current observations are counter to the original hypothesis (H2). The original hypothesis that individuals with shoulder pain would have lower degree of timedependent structure in fluctuations in their wheelchair propulsion variables was based on the tenets of the loss of complexity hypothesis. The loss of complexity hypothesis maintains that with senescence and pathology there is a decline in physiological complexity (i.e. lowered degree of time-dependent structure in fluctuations of physiologic time series) resulting from reductions in control process and their interaction [16]. A criticism of this theoretical framework is that it does not take into the intrinsic dynamics (e.g. fluctuation around a steady fixed state [42], rhythmic limit cycle attractor like in a homeodynamic process [43]) and environmental demands of the motor task being performed [17,44]. Indeed previous research demonstrates that observed differences in time-dependent structure between pathological groups can be bidirectional and their directionality is influenced by task constraints [38,41,44,45] Our observation that SampEn of contact angle was higher for the shoulder pain group is in agreement with occupation ergonomics research [38]. For instance, individuals with shoulder/neck pain performing repetitive upper limb tasks exhibited greater degree of time-dependent structure in movement than healthy controls [38]. Such time-dependent structure in the movement has been implicated as compensatory strategies employed by the neuromuscular system to mitigate discomfort arising from musculoskeletal pain [12, 38]. Along these lines, the higher SampEn magnitude observed for the contact angle at the hand-rim in mWCUs with shoulder pain could be a manifestation of the compensatory strategy they employed to minimize shoulder discomfort when performing the repetitive propulsion task. It is important to note that the current observations of higher SampEn magnitude in the shoulder pain group during wheelchair propulsion is contrasting with observations reported in gait re-
654
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
search which associates lower entropy magnitudes to injury or musculoskeletal disorders [19,20]. There are three potential explanations for this contrast. Perhaps, the most direct explanation could be that the relatively higher degree of freedom of the shoulder musculature leading to abundance in solution strategies to mitigate the discomfort caused due to pain [46,47]. An other explanation could be the choice to study the effect of shoulder pain using the propulsion variables at distal joint (i.e. hand), while the gait research on knee musculoskeletal injuries studied the timedependent structure in fluctuations at the site of injury (i.e. knee) [19,20]. Lastly, the divergent results could have simply resulted from the inherent differences in walking versus wheelchair propulsion. Consistent with previous research self-reported pain was positively associated with the degree of time-dependent structure in fluctuations in contact angle [38]. This observation suggests that, in addition to being able to differentiate between mWCUs with and without shoulder pain, SampEn may also be a potential biomarker to track shoulder pain progression in mWCUs. This exciting possibility warrants further investigation. Finally, while SampEn is one among the many available nonlinear dynamics metrics, numerous researchers have used other techniques such as approximate entropy (ApEn), de-trended fluctuation analysis (DFA) and Lyapunov exponent (Ly Ep) to study the time dependent structure of human movement [28,35,48,49]. All these studies reported that physiologic time series complexity is sensitive to age, pathology and functional state in humans. Nevertheless in context to wheelchair propulsion literature, this investigation is a first step towards showing that pursuing further research in this direction could be beneficial and improve our knowledge to provide better health, diagnosis, and intervention for shoulder pain related to propulsion in mWCUs.
tact angle during manual wheelchair propulsion is structured, (2) individuals with shoulder pain exhibited higher SampEn magnitude for contact angle during wheelchair propulsion compared to those without shoulder pain; and (3) SampEn measure correlated significantly with the amount of self reported shoulder pain. The higher SampEn in shoulder pain group may be a compensatory mechanism adopted to minimize discomfort to shoulder while performing the propulsion task. Overall, we conclude that incorporating non-linear dynamics based measures in wheelchair propulsion analyses may provide new knowledge and can be an important tool for better health, diagnosis, and therapeutic interventions to prevent shoulder pain pathology in mWCUs.
6. Limitations
Supplementary materials
Despite being novel this study is not devoid of limitations. The study is cross-sectional in nature and hence any cause/effect inference cannot be draw. The choice of metrics utilized was hampered by the short time series length which prevented other complementing non-linear analyses such as de-trended fluctuation analysis. In general, adopting a combination of non-linear methods is recommended as best practice [26,41]. The data were collected in a laboratory setting with roller dynamometer setup (akin to a passive treadmill), so it is not clear if these results would occur in real life propulsion environment. It could be possible that SampEn measures of wheelchair propulsion variables other than ones considered in this analyses are more sensitive to shoulder pain in mWCUs [11,37,50]. The association between varying propulsion speeds and SampEn could not studied. These open questions warrants future analyses. The injury demographics of our sample were diverse. Consequently, it is possible that differences in SampEn between pain groups were due to different disability. Though significant, these demographic and environmental limitations are relatively common to wheelchair propulsion research. Finally, despite these limitations, this investigation provides novel contributions that are important and compliments the research findings from previous research [11,12,38].
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.medengphy.2016.04. 005.
7. Conclusions This cross sectional study provided novel information regarding the relationship between the time-dependent structure of wheelchair propulsion and shoulder pain in mWCUs. Examining the time-dependent structure of wheelchair propulsion variables using SampEn, this investigation produced three novel observations, namely, (1) variability observed in the fluctuation in con-
Conflict of interest The authors have no conflict of interest to disclose. Ethical approval All procedures were approved by the local institutional review board at University of Illinois, Urbana-Champaign (IRB #09,424). Acknowledgments This project was funded in part by the National Institute of Health (#1R21HD066129-01A1). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. No additional external funding was received for this study. The authors thank Ms. Iris M-K Hsu, Mr. Mike Socie, Ms. Shawna Culp and Dr. Karla Wessels for their help with data collection. The authors extend their gratitude to the Beckman Institute Illinois Simulator Laboratory.
References [1] LaPlante MP, Kaye HS. Demographics and trends in wheeled mobility equipment use and accessibility in the community. Assist Tech 2010;22:3–17. [2] Curtis KA, Drysdale GA, Lanza RD, Kolber M, Vitolo RS, West R. Shoulder pain in wheelchair users with tetraplegia and paraplegia. Arch Phys Med Rehabil 1999;80:453–7. [3] Gironda RJ, Clark ME, Neugaard B, Nelson A. Upper limb pain in a national sample of veterans with paraplegia. J Spinal Cord Med 2004;27:120–7. [4] Nichols PJ, Norman PA, Ennis JR. Wheelchair user’s shoulder? Shoulder pain in patients with spinal cord lesions. Scand J Rehabil Med 1979;11:29–32. [5] Dalyan M, Cardenas DD, Gerard B. Upper extremity pain after spinal cord injury. Spinal Cord 1999;37:191–5. [6] Lal S. Premature degenerative shoulder changes in spinal cord injury patients. Spinal Cord 1998;36:186–9. [7] Gutierrez DD, Thompson L, Kemp B, Mulroy SJ. The relationship of shoulder pain intensity to quality of life, physical activity, and community participation in persons with paraplegia. J Spinal Cord Med 2007;30:251–5. [8] Pentland WE, Twomey LT. Upper limb function in persons with long term paraplegia and implications for independence: Part II. Paraplegia 1994;32:219–24. [9] Silfverskiold J, Waters RL. Shoulder pain and functional disability in spinal cord injury patients. Clin Orthop Relat Res 1991:141–5. [10] Collinger JL, Boninger ML, Koontz AM, Price R, Sisto SA, Tolerico ML, et al. Shoulder biomechanics during the push phase of wheelchair propulsion: a multisite study of persons with paraplegia. Arch Phys Med Rehabil 2008;89:667–76. [11] Sosnoff JJ, Rice IM, Hsiao-Wecksler ET, Hsu IMK, Jayaraman C, Moon Y. Variability in wheelchair propulsion: a new window into an old problem. Front Bioeng Biotechnol 2015;3:105. [12] Srinivasan D, Mathiassen SE. Motor variability in occupational health and performance. Clin Biomech (Bristol, Avon) 2012;27:979–93. [13] Madeleine P, Voigt M, Mathiassen SE. The size of cycle-to-cycle variability in biomechanical exposure among butchers performing a standardised cutting task. Ergonomics 2008;51:1078–95.
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655 [14] Stergiou N, Leslie MD. Human movement variability, nonlinear dynamics, and pathology: is there a connection? Hum Mov Sci 2011;30(869):88. [15] Hamill J, Palmer C, Van Emmerik RE. Coordinative variability and overuse injury. Sports Med Arthrosc Rehabil Ther Technol 2012;4. [16] Lipsitz LA, Goldberger AL. Loss of ‘complexity’ and aging. Potential applications of fractals and chaos theory to senescence. JAMA. 1992;1806:9. [17] Sosnoff JJ, Newell KM. Are age-related increases in force variability due to decrements in strength? Exp Brain Res Exp Hirnforsch Exp Cereb 2006;174:86–94. [18] Yentes JM, Hunt N, Schmid KK, Kaipust JP, McGrath D, Stergiou N. The appropriate use of approximate entropy and sample entropy with short data sets. Ann Biomed Eng 2013;41:349–65. [19] Tochigi Y, Segal NA, Vaseenon T, Brown TD. Entropy analysis of tri-axial leg acceleration signal waveforms for measurement of decrease of physiological variability in human gait. J Orthop Res 2012;30(897):904. [20] Georgoulis C, Moraiti S, Ristanis Stergiou N. A novel approach to measure variability in the anterior cruciate ligament deficient knee during walking: the use of the approximate entropy in orthopaedics. J Clin Monit Comput 2006;20:11–18. [21] Campbell WI, Lewis S. Visual analogue measurement of pain. Ulster Med J 1990;59:149–54. [22] Curtis KA, Roach KE, Applegate EB, Amar T, Benbow CS, Genecco TD, et al. Development of the wheelchair user’s shoulder pain index (WUSPI). Paraplegia 1995;33:290–3. [23] Koontz AM, Worobey LA, Rice IM, Collinger JL, Boninger ML. Comparison between overground and dynamometer manual wheelchair propulsion. J Appl Biomech 2012;28:412–19. [24] Beekman CE, Miller-Porter L, Schoneberger M. Energy cost of propulsion in standard and ultralight wheelchairs in people with spinal cord injuries. Phys Ther 1999;79:146–58. [25] Tolerico ML, Ding D, Cooper RA, Spaeth DM, Fitzgerald SG, Cooper R, et al. Assessing mobility characteristics and activity levels of manual wheelchair users. J Rehabil Res Dev 2007;44:561–71. [26] Jayaraman C, Moon Y, Rice IM, Hsiao Wecksler ET, Beck CL, Sosnoff JJ. Shoulder pain and cycle to cycle kinematic spatial variability during recovery phase in manual wheelchair users: a pilot investigation. PLoS One 2014;9:e89794. [27] Koontz AM, Roche BM, Collinger JL, Cooper RA, Boninger ML. Manual Wheelchair Propulsion Patterns on Natural Surfaces During Start-Up Propulsion. Arch Phys Med Rehabil 2009;90:1916–23. [28] Stergiou N. Innovative analyses of human movement- analytical tools for human movement research. Human Kinetics; 2004. [29] Richman JS, Moorman JR. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol 20 0 0;278:H2039–49. [30] Pincus S.M., Goldberger A.L. Physiological time-series analysis: what does regularity quantify? Am J Physiol Heart Circ Physiol Meas. 1994;266:H1643–H56. [31] Lake D, Richman JS, Griffin MP, Moorman JR. Sample entropy analysis of neonatal heart rate variability.. Am J Physiol Regul Integr Comp Physiol 2002;283:R789–97.
655
[32] Goldberger A.L., Amaral L.A.N., Glass L., Hausdorff J.M., Ivanov P.C., Mark R.G., et al. PhysioBank, physiotoolkit, andphysionet: components of a new research resource for complex physiologic signals. Circulation. 20 0 0: e215–e20. [33] Shelhamer M. Nonlinear dynamics in physiology. World Scientific; 2006. [34] Jordan K, Newell KM. The structure of variability in human walking and running is speed-dependent. Exerc Sport Sci Rev 20 08;36:20 0–4. [35] Hausdorff JM, Mitchell SL, Firtion R, Peng CK, Cudkowicz ME, Wei JY, et al. Altered fractal dynamics of gait: reduced stride-interval correlations with aging and Huntington’s disease. J Appl Physiol 1997;82:262–9. [36] Boninger ML, Souza AL, Cooper RA, Fitzgerald SG, Koontz AM, Fay BT. Propulsion patterns and pushrim biomechanics in manual wheelchair propulsion. Arch Phys Med Rehabil 2002;83:718–23. [37] Boninger ML, Koontz AM, Sisto SA, Dyson-Hudson TA, Chang M, Price R, et al. Pushrim biomechanics and injury prevention in spinal cord injury: recommendations based on CULP-SCI investigations. J Rehabil Res Dev 2005;42:9–19. [38] Madeleine P, Madsen TMT. Changes in the amount and structure of motor variability during a deboning process are associated with work experience and neck–shoulder discomfort. Appl Ergon 2009;40:887–94. [39] Hausdorff JM, Purdon PL, Peng C-K, Ladin Z, Wei JY, Goldberger AL, et al. Fractal dynamics of human gait: Stability of long-range correlations in stride interval fluctuations. J Appl Physiol 1996;80:1448–57. [40] Rice IM, Jayaraman C, Hsiao-Wecksler ET, Sosnoff JJ. Relationship Between Shoulder Pain and Kinetic and Temporal-Spatial Variability in Wheelchair Users. Arch Phys Med Rehabil 2013. [41] Sosnoff JJ VA, Newell KM. Independence between the amount and structure of variability at low force levels. Neuroscience Letters 2006;392:165–9. [42] Kelso J. Dynamic patterns: the self-organization of brain and behavior. Cambridge: MIT Press; 1995. [43] Yates F. The dynamics of aging and time: how physical action implies social action, New York: Springer; 1988. 198. [44] Vaillancourt DE, Newell KM. Changing complexity in human behavior and physiology through aging and disease. Neurobiology of Aging 2002:1–11. [45] Vieluf STJ, Berton E, Jirsa VK, Sleimen-Malkoun R. Effects of task and age on the magnitude and structure of force fluctuations: insights into underlying neuro-behavioral processes. BMC Neurosci 2015;16(1):1–17. [46] Bernstein N. The co-ordination and regulation of movements. Oxford: Pegamon Press; 1967. [47] Latash ML. The bliss (not the problem) of motor abundance (not redundancy). Exp Brain Res 2012;217:1–5. [48] Cavanaugh J, Guskiewicz KM, Giuliani C, Marshall S, Mercer VS, Stergiou N. Recovery of postural control after cerebral concussion: New insights using approximate entropy. J Athl Train 2006;41:305–13. [49] Sethi A, Davis S, McGuirk T, Patterson TS, Richards LG. Effect of intense functional task training upon temporal structure of variability of upper extremity post stroke. J Hand Ther 2013;26:132–8. [50] Jayaraman C, Beck CL, Sosnoff JJ. Shoulder pain and jerk during recovery phase of manual wheelchair propulsion. J Biomech 2015.
Contents lists available at ScienceDirect
Medical Engineering and Physics journal homepage: www.elsevier.com/locate/medengphy
Shoulder pain and time dependent structure in wheelchair propulsion variability Chandrasekaran Jayaraman a, Yaejin Moon b, Jacob J. Sosnoff b,∗ a b
Department of Industrial and Enterprise Systems Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
a r t i c l e
i n f o
Article history: Received 13 November 2015 Revised 26 February 2016 Accepted 3 April 2016
Keywords: Shoulder pain Nonlinear analysis Variation Compensatory mechanisms Entropy Manual wheelchair propulsion Complexity Human motor control Repetitive loading
a b s t r a c t Manual wheelchair propulsion places considerable repetitive mechanical strain on the upper limbs leading to shoulder injury and pain. While recent research indicates that the amount of variability in wheelchair propulsion and shoulder pain may be related. There has been minimal inquiry into the fluctuation over time (i.e. time-dependent structure) in wheelchair propulsion variability. Consequently the purpose of this investigation was to examine if the time-dependent structure in the wheelchair propulsion parameters are related to shoulder pain. 27 experienced wheelchair users manually propelled their own wheelchair fitted with a SMARTWheel on a roller at 1.1 m/s for 3 min. Time-dependent structure of cycle-to-cycle fluctuations in contact angle and inter push time interval was quantified using sample entropy (SampEn) and compared between the groups with/without shoulder pain using nonparametric statistics. Overall findings were, (1) variability observed in contact angle fluctuations during manual wheelchair propulsion is structured (Z=3.15;p<0.05), (2) individuals with shoulder pain exhibited higher SampEn magnitude for contact angle during wheelchair propulsion than those without pain (χ 2 (1)=6.12;p<0.05); and (3) SampEn of contact angle correlated significantly with self-reported shoulder pain (rs (WUSPI) =0.41;rs (VAS) =0.56;p<0.05). It was concluded that the time-dependent structure in wheelchair propulsion may provide novel information for tracking and monitoring shoulder pain. © 2016 IPEM. Published by Elsevier Ltd. All rights reserved.
1. Introduction Nearly 90% of the 3.6 million wheelchair users in the United States use a manual wheelchair for their mobility needs [1]. The use of a manual wheelchair places considerable mechanical strain on the upper limbs. Unfortunately, the human upper limb is not suited for such repetitive loading. The repetitive loading predisposes manual wheelchair users (mWCUs) for upper limb pathology. Indeed it has been shown that up to 70% of mWCUs report upper limb pain [2–5]. Furthermore, even in mWCUs who do not report pain, there is evidence of degenerative changes in the shoulder [6] suggesting that it is just a matter of time before these asymptomatic individuals will experience pain. Upper limb pain in mWCUs has been linked to difficulty performing activities of daily living [5], decreased physical activity and decreased quality of life [7]. Overall, any loss of upper limb function due to pain adversely impacts the independence and mobility of manual wheelchair users. It has been speculated that a decrease Abbreviations: SampEn, Sample Entropy; MWCUs, Manual wheelchair users; WUSPI, Wheelchair user shoulder pain index; VAS, Visual analog scale. ∗ Corresponding author. Tel.: +1 2173339472; fax: +1 217 2447322. E-mail address: [email protected] (J.J. Sosnoff). http://dx.doi.org/10.1016/j.medengphy.2016.04.005 1350-4533/© 2016 IPEM. Published by Elsevier Ltd. All rights reserved.
in independence and mobility results in greater health care costs and an increased risk for secondary morbidity (cardiovascular disease, obesity, etc.) [8, 9]. Although it has been proposed that propulsion biomechanics are related to shoulder pain in MWCUs, evidence is inconclusive [10]. Average spatiotemporal parameters of wheelchair propulsion (e.g. contact angle and push time) do not distinguish between those with and without pain. However, recent research indicates that spatial-temporal variability in wheelchair propulsion mechanics (e.g. push time and peak force during push) is related to shoulder pain [11]. This association between motor variability and pain is consistent with observations from occupation biomechanics and human motor control literature [12–15]. Although promising, a limitation of this research [11], is that it has only focused on the amount of variability in wheelchair propulsion. This approach seemingly ignores fluctuations over time (i.e. timedependent structure) inherent in physiological output [16]. Based on the tenets of the loss of complexity hypothesis of aging [16], it has been theorized that the time-dependent structure of motor output is a marker of physiological complexity and provides novel information concerning the health of the musculoskeletal system [14,17]. Specifically, it has been proposed that musculoskeletal injury leads to fluctuations in movement that are
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
649
Table 1 Demographic information. Comparison of demographic information between the groups with and without pain. Statistically significant group differences were observed for the pain scores between the groups (p<0.05). No significant group differences was observed between the groups in age, body weight, gender and experience using wheelchair. Demographic variables
Pain (n=13) Mean(SD)
No Pain (n=14) Mean(SD)
Age (Yrs) Body weight (LB) Experience using wheelchair (Yrs) Gender (F/M)
28.23(12.52) 164.50(56.39) 15.84(11.25) 6/7 Spina Bifida(n=5) T6-T12 (n=5) L1-L4 (n=1) Amputee- double (n=1) Sacral agenisis (n=1) 22.84(21.27) 4.24(2.8)
21.21(4.92) 131.17(36.96) 13.64(5.15) 6/8 Spina Bifida (n=4) T6-T12 (n=6) L1-L4 (n=1) Amputee-single (n=1) Arthrogryposis (n=1) C7 (n=1) 3.28(5.02) 0.36(0.96)
Injury demographics WUSPI∗ VAS∗ ∗
p<0.05
more structured. Two widely reported methods to measure the structure in fluctuation of movements are approximate entropy and sample entropy (SampEn) [18]. For instance, Tochigi et al., 2012, [19], utilized SampEn to analyze the structure of variability in gait in individuals with knee osteoarthritis with and without pain. Consistent with the loss of complexity hypothesis, it was reported that the pain group had significantly lower SampEn values when compared to those without pain. In a similar fashion, researchers have successfully employed approximate entropy to study the gait pattern related to anterior cruciate ligament (ACL) injury [20] and reported lower approximate entropy values in the knee with ACL deficiency compared to the healthy knee. Presently there is no information regarding the time-dependent structure of variability in the context of shoulder pain in mWCUs. Consequently the purpose of this investigation was to examine if the time-dependent structure in wheelchair propulsion parameters are related to shoulder pain. The goals of this investigation were twofold, (H1) to determine whether the time-dependent structure observed in wheelchair propulsion variability is structured, and (H2) to determine if individuals propelling a manual wheelchair with shoulder pain will demonstrate lower degree of time-dependent structure in fluctuations in their wheelchair propulsion parameters than those without shoulder pain. To test these hypotheses, two wheelchair propulsion parameters were analyzed, namely, contact angle and inter push time interval.
2. Methods
The participants were categorized as belonging to pain/no pain group based on the self-reported shoulder pain status (“Yes”/”No”). Participants also rated their current level of shoulder pain for each shoulder using a 10 cm visual analog scale (VAS) [21]. A VAS score of zero indicated no shoulder pain and higher score indicate greater shoulder pain at the time of testing. Participants also rated their shoulder pain using the wheelchair user’s shoulder pain index (WUSPI) [22]. WUSPI is a 15-item questionnaire. Each item is rated between 0 and 10, with 0 representing no interference with daily living functional activities and 10 representing complete interference with functional activities during the past week due to shoulder pain. The total score is the sum of scores of all the 15 items put together. Total score ranges from 0 (no pain)-150 (maximum limitations to daily activities due to pain) [22]. 2.2. Data collection and instrumentation 2.2.1. Kinetics On completion of the informed consent procedures, each participant’s wheelchair was fitted bilaterally with SMARTwheels (Three Rivers Holdings, LLC, Mesa, AZ, USA) and secured to a single roller dynamometer system using a four-point tie-down system and a flywheel system [23]. The SMARTwheels measures threedimensional forces and torques applied to the push rim. Participants were provided sufficient time to acclimate themselves with the dynamometer setup prior to the actual trial. Participants were asked to propel at constant speed of 1.1 m/s for 3 min [24,25]. A speedometer was used to provide real-time feedback to the participants during the three minute propulsion. The SMARTwheel data was post-processed using a custom developed MATLAB routine.
2.1. Participants 2.3. Data post-processing Wheelchair propulsion data from 27 experienced adult mWCUs was analyzed for this study. This dataset is a subset of data collected for a larger study focusing on wheelchair propulsion and shoulder pain [11]. Inclusion criteria for this secondary analysis were (1) more than one year of manual wheelchair experience; (2) between 18–64 years of age; and (3) trials data containing at least 106 wheelchair propulsion cycles. All procedures were approved by the local institutional review board at University of Illinois, Urbana-Champaign. Upon arrival to the laboratory, the data collection procedures were thoroughly explained and any questions the participants had regarding the protocol were addressed. Participants were informed that participation in the study was voluntary. The participants then provided their written informed consent and demographic information (Table 1).
SmartWheel data were collected at a sampling rate of 100 Hz and digitally filtered with an eighth-order, zero-phase, low-pass Butterworth filter with 20 Hz cutoff frequency [10]. The start and end of a propulsion cycle was defined when the push-rim moment (Mz) was above and below 1 Nm respectively [26]. To reduce the transient effects, data belonging to the first five propulsion cycles were not included for this analysis [27]. For consistency, the number of data cycles analyzed for each participant was maintained constant at 100 cycles (i.e. starting from the 6th cycle to 105th cycle of a SmartWheel data, Fig. 1(a)). The contact angle and resultant forces at hand-rim were extracted for each participant (Fig. 1(b)). Following this, the inter push time interval between peak resultant force between pushes were extracted (Fig. 1(b)). Thus, this process yielded two time series from each participant wheelchair
650
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
Fig. 1. Data processing and variable extraction for SampEn calculation. (a) resultant force time series at hand-rim showing the steady state portion of the time series used for all calculations; (b) a magnified view of two consecutive sample resultant force cycles at steady state to show the details of the variables extracted. The contact angle is defined as the angular measure between the start and end of contact. The inter push time interval between peak to peak resultant force was extracted from cycle-to-cycle during steady state to create the time series containing the inter push time interval fluctuations. The cycle-to-cycle contact angle magnitude during the steady state was extracted to create the time series of contact angle fluctuation.
propulsion data namely, (1) a time series of cycle-to-cycle contact angle fluctuation (Fig. 2(a)) and (2) a time series of cycle-to-cycle inter push time interval fluctuation between peak resultant force during consecutive push phases (Fig. 2(b)). For brevity, the variable, cycle-to-cycle inter-push time interval is termed inter push time interval within the manuscript. SampEn was then computed for each time series (see Section 2.4). A custom developed MATLAB code was used to accomplish all data post-processing. Based on the VAS scores, for the shoulder pain group, the data belonging to the side with greatest shoulder pain level was analyzed (right (n=11) and left (n=2)), while the data belonging to the dominant hand (right (n=12) and left (n=2)) was analyzed for the group without shoulder pain. 2.4. Time dependent structure in variability Complexity analysis quantifies the time dependent regularity and predictability of a time series. There are various nonlinear dynamics tools to measure the complexity of a physiological time series [28]. The choice of a specific tool depends on the characteristics of data being analyzed. In this investigation, we utilized SampEn, a widely utilized approach for studying time-dependent structure in short time series [18,19]. SampEn is a nonlinear metric used for measuring the regularity of a time series. The magnitude of the SampEn is an indication of the degree of regularity or irregularity of a particular time series. SampEn values typically ranges from 0 to 4. Values closer to 0 are consistent with greater regularity, such as a sinewave, while values nearing 4 represent greater irregularity such as pink noise [29]. Higher values of SampEn indicate that the time series is more unpredictable (i.e. unstructured). Details of the SampEn algorithms and input parameters used are available elsewhere [30,31]. MATLAB codes obtained from http: //www.physionet.org/physiotools/, [32] was used for SampEn cal-
culation. The performance (SampEn magnitudes) of the software program was tested with synthetically generated benchmark signals (for more details please see supplementary materials: Section A). This performance test was conducted to ascertain the validity of the SampEn values obtained from the physionet.com SampEn code. The tests revealed that the SampEn magnitudes obtained from the software program provided by physionet.com is comparable with those reported in literature (please see supplementary materials: Section A). Based on this validation, the SampEn code from physionet.org was used for calculating SampEn in this analysis. The parameter we choose for the SampEn analysis were (m=2, r=0.2) for the contact angle time series and (m=2,r=0.15) for the inter push time interval time series. The justifications for selecting these parameter choices are discussed in the supplementary materials (Section B). From each participant’s post-processed data, the SampEn for the original time series belonging to the contact angle and the inter push time interval were calculated. To statistically establish if the SampEn measure obtained from the original time series was indeed structured and not random, a simple surrogate analysis was carried out [33]. This surrogate analysis was conducted to test H1. Surrogate data sets were generated by randomly shuffling the original time series in order to determine whether the original time series was structured (Fig. 3). Such random shuffling preserves the distributional statistics (i.e., mean, standard deviation, and higher moments) between the original time series and the corresponding surrogate time series contain the same elements; however, the time-dependent sequential ordering is destroyed [34]. Each participant’s original time series was randomly shuffled 100 times to produce a pool of corresponding surrogate data. This surrogate procedure resulted in a pool of hundred surrogate time series for each original time series. The SampEn values were computed for each of the 100 surrogate time series in the pool and then averaged to generate a mean SampEn for the surrogate pool
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
651
Fig. 2. Sample contact angle and inter push time interval time series extracted from participants with and without shoulder pain from which the SampEn was calculated. (a) the cycle-to-cycle contact angle fluctuation at hand-rim for 100 pushes; (b) the cycle-to-cycle inter push time interval fluctuation at hand-rim for 100 pushes.
(Fig. 3). This procedure was repeated for every original time series. Finally non-parametric pair-wise tests comparing the SampEn values between each original time-series with the corresponding mean SampEn obtained from its surrogate pool was conducted. If the SampEn value of the original time series was significantly (statistically) smaller than its corresponding surrogate counterpart this indicates that the original data is not randomly derived and therefore its variability exhibits time-dependent structure [35]. Qualifying the surrogate test (H1), is a necessary step prior to testing H2. Additionally, the SampEn magnitudes can be influenced by nonstationarity in the original time series analyzed [18]. To check for this influence, the SampEn for both the contact angle and inter push time interval time series were also calculated from differenced versions of original time series. Pair-wise tests revealed no significant differences between the SampEn calculated from the original and their differenced versions (for more details please see supplementary materials: Section C). Thus the SampEn from the original time series were used for all the statistical analyses. 3. Statistical analysis IBM SPSS (version 21) was used for conducting all the statistical tests. The significance level for all the statistical tests were set at p≤0.05. All descriptive values are reported as mean (SD) unless otherwise noted. The between group factor for all the statistical tests was set as shoulder pain (with pain = 1; No pain = 0). Normality checks for data distribution using Shapiro–Wilk tests revealed that the age, self reported WUSPI scores, VAS scores, mean inter push time interval, mean speed and the SampEn required non-parametric statistical tests. 3.1. Demographics Participant demographics information (age, body weight, and manual wheelchair propulsion experience in years) and self reported scores (WUSPI & VAS) were treated as independent variables.
A series of two tailed Mann-Whitney U tests were conducted to check if demographic variables (age and self reported shoulder pain (WUSPI; VAS)) were significantly different between groups. A series of two tailed independent t-tests were conducted to check if significant between group differences in existed in body weight and wheelchair propulsion experience. A Chi-squared tests was conducted to determine if the sample gender composition significantly differed between the groups with and without shoulder pain. 3.2. Mean wheelchair propulsion variables In addition to the SampEn of contact angle and inter push time interval, mean propulsion variables that have been suggested to be associated with shoulder pain where also extracted from the SmartWheel data. Specifically, contact angle, peak resultant force at the hand-rim, mean inter push time interval between peak resultant force and mean push speed were determined with established procedures [36]. A series of two tailed Mann–Whitney U test were conducted to check if the mean inter push time interval and mean push speed were significantly different between groups with and without shoulder pain. A series of two tailed t-tests were performed to check if statistically significant group differences existed in other mean wheelchair propulsion variables (i.e. contact angle, peak resultant force at hand-rim). The mean wheelchair propulsion variables were verified to benchmark the data with previous literature. 3.3. SampEn To test if the SampEn of the original time series were significantly different from their surrogate counterpart (i.e. H1) separate pair-wise Wilcoxon signed rank tests were conducted for the contact angle and inter push time interval. To investigate the main effect of shoulder pain (i.e. H2), the SampEn obtained from the original time series for the contact angle and inter push time
652
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
Fig. 3. The surrogate analysis process. The mean(SD) of the original and the surrogated time series pool are same. The surrogation procedure used was a simple random shuffling of data. A pool of 100 randomly shuffled surrogates was created from the original time series (Surrogate (i=1 to 100) . If (SampEn)original is statistically lesser (p<0.05) than (MeanSampEn)Surrogate , then the time dependant structure observed in the original time series is not random and may have some meaningful insight.
interval were compared between the groups with and without shoulder pain using Kruskal–Wallis ANOVA’s. Spearman’s rank correlational analyses were conducted to investigate if the SampEn for the contact angle and inter push time interval was correlated with the self reported pain scores (WUSPI and VAS). 4. Results 4.1. Demographics No significant group differences in demographics were observed, (p’s>0.05; Table 1). Per design the shoulder pain group reported higher pain than the no shoulder pain group (WUSPI: [U=17; p<0.05]; VAS: [U=14.5; p<0.05];Table 1).
Table 2 Mean wheelchair propulsion variables at hand-rim. Comparison of mean wheelchair propulsion variables at hand-rim between the groups with and without pain. No significant group differences were observed between the groups. Mean propulsion parameters at handrim
Pain (n=13) Mean(SD)
No Pain (n=14) Mean(SD)
Peak resultant force (N) Contact angle (deg) Speed (m/s) Inter push time interval between peak resultant force (Sec)
69.82(23.84) 100.52(20.25) 1.1(0.04) 1.15(0.22)
60.23(19.72) 97.68(17.64) 1.1(0.06) 1.20(0.22)
4.3. SampEn of contact angle and inter push time interval
smaller than the SampEn from their corresponding surrogates: inter push time interval : [Z=2.59,p=0.009](m=2,r=0.15) ; [(mean(SD), IQR): (2.5(0.6),0.85)pain ; (2.2(0.4),0.63)No pain ]. contact angle:[Z=3.15,p=0.002](m=2, r=0.2); [(mean(SD), IQR): (2.1(0.2),0.27)pain ; (1.8(0.2),0.32)No pain ]. This indicates that the structure of variability found in wheelchair propulsion is not random but rather structured.
4.3.1. SampEn comparison between original and surrogated time series Wilcoxon signed rank pairwise tests revealed that the SampEn obtained from the original time series were statistically
4.3.2. Shoulder pain and time dependent structure of propulsion variability As illustrated in Fig. 4(a) the SampEn for contact angle was higher in the pain group than the no-pain group. This was
4.2. Mean wheelchair propulsion variables No significant differences were observed in mean propulsion variables between groups (p’s>0.05; Table 2).
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
Fig. 4. Groupwise mean (SD) comparison for SampEn. (a) the SampEn for contact angle was significantly greater (p<0.05) for the group with shoulder pain; (b) the SampEn for inter push time interval at hand-rim between groups with and without shoulder pain was not significantly different (p>0.05) between the groups with shoulder pain.
confirmed by Kruskal–Wallis test, χ 2 (1) = 6.12, p = 0.013. No significant main effect of shoulder pain was observed for the SampEn of inter push time interval (Fig. 4(b); p>0.05). A statistically significant moderate positive correlation was observed between the self-reported shoulder discomfort scores (WUSPI and VAS) and the SampEn of contact angle, [rs (25)=0.41, p<0.05]WUSPI ; [rs (25)= 0.56, p<0.05]VAS. There was no association between SampEn of inter push time interval and self reported pain scores (p>0.05). 5. Discussion This investigation explored the relation between shoulder pain and the structure of motor variability in spatiotemporal variables of manual wheelchair propulsion. Three novel observations were made, (1) variability observed in the fluctuations in contact angle and inter push time interval during manual wheelchair propulsion is structured, (2) individuals with shoulder pain exhibited higher SampEn during wheelchair propulsion compared to those without pain; and (3) SampEn measures correlated significantly with the amount of self-reported shoulder pain scores. Overall the observations suggest that SampEn analysis of wheelchair propulsion may provide novel insights for monitoring the development and treatment of shoulder pain in mWCUs. While the majority of wheelchair research focuses on mean kinematic and kinetics during propulsion [10,37] growing evidence suggests that variability measures could be a sensitive marker of shoulder pain in mWCUs [11]. Consistent with previous literature the mean propulsion variables were not able to identify any difference between mWCUs with and without shoulder pain [10,37].To date, the few wheelchair propulsion studies which incorporate
653
measures of intra-individual variability have exclusively focused on linear distributional statistics (standard deviation and coefficient of variation) and not measures of time dependent structure. The current investigation used SampEn, to quantify the time dependent structure in two propulsion parameters that are closely implicated with shoulder pain in mWCUs, namely, contact angle and inter push time interval between peak resultant forces, [10,37]. Surrogate analyses revealed that the time dependent structure observed in both measures were not random. This observation is consistent with reports from occupational ergonomics and motor control research which supports the tenet that the time dependent structure of variability in physiologic output contains meaningful insights concerning health and function [35,38,39] It is important to note that differences in wheelchair users with and without shoulder pain in SampEn were only observed in fluctuations in contact angle. Individuals with shoulder pain exhibited higher SampEn for fluctuations in contact angle during wheelchair propulsion in comparison to those without pain. This is distinct from examinations of the magnitude of variability which revealed that individuals with shoulder pain had lower amounts of variability in spatial and temporal parameters during propulsion as indexed by coefficient of variation [40]. The divergent results between time-dependent and magnitude metrics are consistent with the view that they are at least partially independent [41]. The null effect between pain groups for SampEn of inter push time interval could have resulted as an artifact of the experimental set-up. It is well established that the temporal structure of rhythmic movement is impacted when individuals are externally paced [39]. Consequently, the lack of group difference in the degree of time-dependent structure in fluctuations in inter push time interval could have been due to the real time visual feedback of speed provided. It is important to note that the current observations are counter to the original hypothesis (H2). The original hypothesis that individuals with shoulder pain would have lower degree of timedependent structure in fluctuations in their wheelchair propulsion variables was based on the tenets of the loss of complexity hypothesis. The loss of complexity hypothesis maintains that with senescence and pathology there is a decline in physiological complexity (i.e. lowered degree of time-dependent structure in fluctuations of physiologic time series) resulting from reductions in control process and their interaction [16]. A criticism of this theoretical framework is that it does not take into the intrinsic dynamics (e.g. fluctuation around a steady fixed state [42], rhythmic limit cycle attractor like in a homeodynamic process [43]) and environmental demands of the motor task being performed [17,44]. Indeed previous research demonstrates that observed differences in time-dependent structure between pathological groups can be bidirectional and their directionality is influenced by task constraints [38,41,44,45] Our observation that SampEn of contact angle was higher for the shoulder pain group is in agreement with occupation ergonomics research [38]. For instance, individuals with shoulder/neck pain performing repetitive upper limb tasks exhibited greater degree of time-dependent structure in movement than healthy controls [38]. Such time-dependent structure in the movement has been implicated as compensatory strategies employed by the neuromuscular system to mitigate discomfort arising from musculoskeletal pain [12, 38]. Along these lines, the higher SampEn magnitude observed for the contact angle at the hand-rim in mWCUs with shoulder pain could be a manifestation of the compensatory strategy they employed to minimize shoulder discomfort when performing the repetitive propulsion task. It is important to note that the current observations of higher SampEn magnitude in the shoulder pain group during wheelchair propulsion is contrasting with observations reported in gait re-
654
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655
search which associates lower entropy magnitudes to injury or musculoskeletal disorders [19,20]. There are three potential explanations for this contrast. Perhaps, the most direct explanation could be that the relatively higher degree of freedom of the shoulder musculature leading to abundance in solution strategies to mitigate the discomfort caused due to pain [46,47]. An other explanation could be the choice to study the effect of shoulder pain using the propulsion variables at distal joint (i.e. hand), while the gait research on knee musculoskeletal injuries studied the timedependent structure in fluctuations at the site of injury (i.e. knee) [19,20]. Lastly, the divergent results could have simply resulted from the inherent differences in walking versus wheelchair propulsion. Consistent with previous research self-reported pain was positively associated with the degree of time-dependent structure in fluctuations in contact angle [38]. This observation suggests that, in addition to being able to differentiate between mWCUs with and without shoulder pain, SampEn may also be a potential biomarker to track shoulder pain progression in mWCUs. This exciting possibility warrants further investigation. Finally, while SampEn is one among the many available nonlinear dynamics metrics, numerous researchers have used other techniques such as approximate entropy (ApEn), de-trended fluctuation analysis (DFA) and Lyapunov exponent (Ly Ep) to study the time dependent structure of human movement [28,35,48,49]. All these studies reported that physiologic time series complexity is sensitive to age, pathology and functional state in humans. Nevertheless in context to wheelchair propulsion literature, this investigation is a first step towards showing that pursuing further research in this direction could be beneficial and improve our knowledge to provide better health, diagnosis, and intervention for shoulder pain related to propulsion in mWCUs.
tact angle during manual wheelchair propulsion is structured, (2) individuals with shoulder pain exhibited higher SampEn magnitude for contact angle during wheelchair propulsion compared to those without shoulder pain; and (3) SampEn measure correlated significantly with the amount of self reported shoulder pain. The higher SampEn in shoulder pain group may be a compensatory mechanism adopted to minimize discomfort to shoulder while performing the propulsion task. Overall, we conclude that incorporating non-linear dynamics based measures in wheelchair propulsion analyses may provide new knowledge and can be an important tool for better health, diagnosis, and therapeutic interventions to prevent shoulder pain pathology in mWCUs.
6. Limitations
Supplementary materials
Despite being novel this study is not devoid of limitations. The study is cross-sectional in nature and hence any cause/effect inference cannot be draw. The choice of metrics utilized was hampered by the short time series length which prevented other complementing non-linear analyses such as de-trended fluctuation analysis. In general, adopting a combination of non-linear methods is recommended as best practice [26,41]. The data were collected in a laboratory setting with roller dynamometer setup (akin to a passive treadmill), so it is not clear if these results would occur in real life propulsion environment. It could be possible that SampEn measures of wheelchair propulsion variables other than ones considered in this analyses are more sensitive to shoulder pain in mWCUs [11,37,50]. The association between varying propulsion speeds and SampEn could not studied. These open questions warrants future analyses. The injury demographics of our sample were diverse. Consequently, it is possible that differences in SampEn between pain groups were due to different disability. Though significant, these demographic and environmental limitations are relatively common to wheelchair propulsion research. Finally, despite these limitations, this investigation provides novel contributions that are important and compliments the research findings from previous research [11,12,38].
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.medengphy.2016.04. 005.
7. Conclusions This cross sectional study provided novel information regarding the relationship between the time-dependent structure of wheelchair propulsion and shoulder pain in mWCUs. Examining the time-dependent structure of wheelchair propulsion variables using SampEn, this investigation produced three novel observations, namely, (1) variability observed in the fluctuation in con-
Conflict of interest The authors have no conflict of interest to disclose. Ethical approval All procedures were approved by the local institutional review board at University of Illinois, Urbana-Champaign (IRB #09,424). Acknowledgments This project was funded in part by the National Institute of Health (#1R21HD066129-01A1). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. No additional external funding was received for this study. The authors thank Ms. Iris M-K Hsu, Mr. Mike Socie, Ms. Shawna Culp and Dr. Karla Wessels for their help with data collection. The authors extend their gratitude to the Beckman Institute Illinois Simulator Laboratory.
References [1] LaPlante MP, Kaye HS. Demographics and trends in wheeled mobility equipment use and accessibility in the community. Assist Tech 2010;22:3–17. [2] Curtis KA, Drysdale GA, Lanza RD, Kolber M, Vitolo RS, West R. Shoulder pain in wheelchair users with tetraplegia and paraplegia. Arch Phys Med Rehabil 1999;80:453–7. [3] Gironda RJ, Clark ME, Neugaard B, Nelson A. Upper limb pain in a national sample of veterans with paraplegia. J Spinal Cord Med 2004;27:120–7. [4] Nichols PJ, Norman PA, Ennis JR. Wheelchair user’s shoulder? Shoulder pain in patients with spinal cord lesions. Scand J Rehabil Med 1979;11:29–32. [5] Dalyan M, Cardenas DD, Gerard B. Upper extremity pain after spinal cord injury. Spinal Cord 1999;37:191–5. [6] Lal S. Premature degenerative shoulder changes in spinal cord injury patients. Spinal Cord 1998;36:186–9. [7] Gutierrez DD, Thompson L, Kemp B, Mulroy SJ. The relationship of shoulder pain intensity to quality of life, physical activity, and community participation in persons with paraplegia. J Spinal Cord Med 2007;30:251–5. [8] Pentland WE, Twomey LT. Upper limb function in persons with long term paraplegia and implications for independence: Part II. Paraplegia 1994;32:219–24. [9] Silfverskiold J, Waters RL. Shoulder pain and functional disability in spinal cord injury patients. Clin Orthop Relat Res 1991:141–5. [10] Collinger JL, Boninger ML, Koontz AM, Price R, Sisto SA, Tolerico ML, et al. Shoulder biomechanics during the push phase of wheelchair propulsion: a multisite study of persons with paraplegia. Arch Phys Med Rehabil 2008;89:667–76. [11] Sosnoff JJ, Rice IM, Hsiao-Wecksler ET, Hsu IMK, Jayaraman C, Moon Y. Variability in wheelchair propulsion: a new window into an old problem. Front Bioeng Biotechnol 2015;3:105. [12] Srinivasan D, Mathiassen SE. Motor variability in occupational health and performance. Clin Biomech (Bristol, Avon) 2012;27:979–93. [13] Madeleine P, Voigt M, Mathiassen SE. The size of cycle-to-cycle variability in biomechanical exposure among butchers performing a standardised cutting task. Ergonomics 2008;51:1078–95.
C. Jayaraman et al. / Medical Engineering and Physics 38 (2016) 648–655 [14] Stergiou N, Leslie MD. Human movement variability, nonlinear dynamics, and pathology: is there a connection? Hum Mov Sci 2011;30(869):88. [15] Hamill J, Palmer C, Van Emmerik RE. Coordinative variability and overuse injury. Sports Med Arthrosc Rehabil Ther Technol 2012;4. [16] Lipsitz LA, Goldberger AL. Loss of ‘complexity’ and aging. Potential applications of fractals and chaos theory to senescence. JAMA. 1992;1806:9. [17] Sosnoff JJ, Newell KM. Are age-related increases in force variability due to decrements in strength? Exp Brain Res Exp Hirnforsch Exp Cereb 2006;174:86–94. [18] Yentes JM, Hunt N, Schmid KK, Kaipust JP, McGrath D, Stergiou N. The appropriate use of approximate entropy and sample entropy with short data sets. Ann Biomed Eng 2013;41:349–65. [19] Tochigi Y, Segal NA, Vaseenon T, Brown TD. Entropy analysis of tri-axial leg acceleration signal waveforms for measurement of decrease of physiological variability in human gait. J Orthop Res 2012;30(897):904. [20] Georgoulis C, Moraiti S, Ristanis Stergiou N. A novel approach to measure variability in the anterior cruciate ligament deficient knee during walking: the use of the approximate entropy in orthopaedics. J Clin Monit Comput 2006;20:11–18. [21] Campbell WI, Lewis S. Visual analogue measurement of pain. Ulster Med J 1990;59:149–54. [22] Curtis KA, Roach KE, Applegate EB, Amar T, Benbow CS, Genecco TD, et al. Development of the wheelchair user’s shoulder pain index (WUSPI). Paraplegia 1995;33:290–3. [23] Koontz AM, Worobey LA, Rice IM, Collinger JL, Boninger ML. Comparison between overground and dynamometer manual wheelchair propulsion. J Appl Biomech 2012;28:412–19. [24] Beekman CE, Miller-Porter L, Schoneberger M. Energy cost of propulsion in standard and ultralight wheelchairs in people with spinal cord injuries. Phys Ther 1999;79:146–58. [25] Tolerico ML, Ding D, Cooper RA, Spaeth DM, Fitzgerald SG, Cooper R, et al. Assessing mobility characteristics and activity levels of manual wheelchair users. J Rehabil Res Dev 2007;44:561–71. [26] Jayaraman C, Moon Y, Rice IM, Hsiao Wecksler ET, Beck CL, Sosnoff JJ. Shoulder pain and cycle to cycle kinematic spatial variability during recovery phase in manual wheelchair users: a pilot investigation. PLoS One 2014;9:e89794. [27] Koontz AM, Roche BM, Collinger JL, Cooper RA, Boninger ML. Manual Wheelchair Propulsion Patterns on Natural Surfaces During Start-Up Propulsion. Arch Phys Med Rehabil 2009;90:1916–23. [28] Stergiou N. Innovative analyses of human movement- analytical tools for human movement research. Human Kinetics; 2004. [29] Richman JS, Moorman JR. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol 20 0 0;278:H2039–49. [30] Pincus S.M., Goldberger A.L. Physiological time-series analysis: what does regularity quantify? Am J Physiol Heart Circ Physiol Meas. 1994;266:H1643–H56. [31] Lake D, Richman JS, Griffin MP, Moorman JR. Sample entropy analysis of neonatal heart rate variability.. Am J Physiol Regul Integr Comp Physiol 2002;283:R789–97.
655
[32] Goldberger A.L., Amaral L.A.N., Glass L., Hausdorff J.M., Ivanov P.C., Mark R.G., et al. PhysioBank, physiotoolkit, andphysionet: components of a new research resource for complex physiologic signals. Circulation. 20 0 0: e215–e20. [33] Shelhamer M. Nonlinear dynamics in physiology. World Scientific; 2006. [34] Jordan K, Newell KM. The structure of variability in human walking and running is speed-dependent. Exerc Sport Sci Rev 20 08;36:20 0–4. [35] Hausdorff JM, Mitchell SL, Firtion R, Peng CK, Cudkowicz ME, Wei JY, et al. Altered fractal dynamics of gait: reduced stride-interval correlations with aging and Huntington’s disease. J Appl Physiol 1997;82:262–9. [36] Boninger ML, Souza AL, Cooper RA, Fitzgerald SG, Koontz AM, Fay BT. Propulsion patterns and pushrim biomechanics in manual wheelchair propulsion. Arch Phys Med Rehabil 2002;83:718–23. [37] Boninger ML, Koontz AM, Sisto SA, Dyson-Hudson TA, Chang M, Price R, et al. Pushrim biomechanics and injury prevention in spinal cord injury: recommendations based on CULP-SCI investigations. J Rehabil Res Dev 2005;42:9–19. [38] Madeleine P, Madsen TMT. Changes in the amount and structure of motor variability during a deboning process are associated with work experience and neck–shoulder discomfort. Appl Ergon 2009;40:887–94. [39] Hausdorff JM, Purdon PL, Peng C-K, Ladin Z, Wei JY, Goldberger AL, et al. Fractal dynamics of human gait: Stability of long-range correlations in stride interval fluctuations. J Appl Physiol 1996;80:1448–57. [40] Rice IM, Jayaraman C, Hsiao-Wecksler ET, Sosnoff JJ. Relationship Between Shoulder Pain and Kinetic and Temporal-Spatial Variability in Wheelchair Users. Arch Phys Med Rehabil 2013. [41] Sosnoff JJ VA, Newell KM. Independence between the amount and structure of variability at low force levels. Neuroscience Letters 2006;392:165–9. [42] Kelso J. Dynamic patterns: the self-organization of brain and behavior. Cambridge: MIT Press; 1995. [43] Yates F. The dynamics of aging and time: how physical action implies social action, New York: Springer; 1988. 198. [44] Vaillancourt DE, Newell KM. Changing complexity in human behavior and physiology through aging and disease. Neurobiology of Aging 2002:1–11. [45] Vieluf STJ, Berton E, Jirsa VK, Sleimen-Malkoun R. Effects of task and age on the magnitude and structure of force fluctuations: insights into underlying neuro-behavioral processes. BMC Neurosci 2015;16(1):1–17. [46] Bernstein N. The co-ordination and regulation of movements. Oxford: Pegamon Press; 1967. [47] Latash ML. The bliss (not the problem) of motor abundance (not redundancy). Exp Brain Res 2012;217:1–5. [48] Cavanaugh J, Guskiewicz KM, Giuliani C, Marshall S, Mercer VS, Stergiou N. Recovery of postural control after cerebral concussion: New insights using approximate entropy. J Athl Train 2006;41:305–13. [49] Sethi A, Davis S, McGuirk T, Patterson TS, Richards LG. Effect of intense functional task training upon temporal structure of variability of upper extremity post stroke. J Hand Ther 2013;26:132–8. [50] Jayaraman C, Beck CL, Sosnoff JJ. Shoulder pain and jerk during recovery phase of manual wheelchair propulsion. J Biomech 2015.