Shoulder Biomechanics During the Push Phase of Wheelchair Propulsion: A Multisite Study of Persons With Paraplegia

Participants 

A total of 61 subjects (21 from HERL, 20 from KMRREC, 20 from UW) volunteered and provided informed consent before participation in this study. All subjects used a manual wheelchair as their primary means of mobility, were over 18 years old, and had an SCI below T1 that had occurred more than 1 year before participation in the study. Each subject also had a wheelchair with quick-release wheels to ensure compatibility with the kinetic measurement device. People were excluded from this study if they had a history of fractures or dislocations in the arms including the shoulder, elbow, and wrist; upper-limb dysthetic pain as a result of a syrinx or complex regional pain syndrome type II; or if they had upper-limb pain that prohibited them from propelling a manual wheelchair. Nondominant-side data were used for all analyses. Five of the 61 subjects were left-handed; all others were right-hand dominant. Demographic information including height, weight, age, years since injury, injury level, and sex was collected from all subjects. Subjects were also asked 2 questions about shoulder pain: (1) Have you had any shoulder pain in the last month? (2) Does your shoulder hurt you while you are propelling your wheelchair?

Instrumentation and Data Collection 

Data Analysis 

Inverse dynamics 

Cooper et al1 previously described the anthropometric model used for this study. Segment lengths and upper-extremity circumferences of all subjects were measured as input to Hanavan’s mathematic model, which calculates the inertial properties of each body segment.19 Pushrim forces were transformed to the glenohumeral joint using a previously described inverse dynamics model.1 Calculations for the inverse dynamics model were performed using Matlab.e Shoulder joint forces were transformed to the anatomic coordinate system of the proximal segment of the shoulder joint, the trunk, as follows: anterior (x), posterior (−x), superior (y), inferior (−y), medial (z), and lateral (−z) (fig 1). Shoulder joint moments were calculated relative to the humeral local coordinate system described in previous work.1 The humeral and trunk local coordinate systems are coincident when the arm is in a neutral posture. Abduction (+) and adduction (−) moments occurred about the x axis, external (+) and internal (−) rotation produced moments about the y axis, and extension (+) and flexion (−) moments occurred about the z axis. HERL and KMRREC collected bilateral kinematic data, and therefore the trunk local coordinate system was calculated using markers at the acromions and greater trochanters as previously described.11 UW collected unilateral kinematic data, so a different method, using markers at the sternal notch, C7 vertebra, and T3 vertebra, was used to determine the trunk local coordinate system; however, both methods approximate the local coordinate systems recommended by ISB.15 To minimize differences between sites, a kinematic calibration trial was collected before testing. For this trial, subjects were instructed to sit in their wheelchairs such that the trunk was perpendicular to the ground aligned with the global coordinate system. A corrective transformation matrix was calculated for each subject to satisfy this condition. This transformation matrix was then applied to the trunk local coordinate system as calculated in all other trials.

View Large Image Download to PowerPoint

Fig 1. Trunk local coordinate system. Reprinted with permission.11

Forces and moments are presented with left shoulder sign convention to allow for averaging of all data regardless of each subject’s nondominant side. Peak values during the push phase of propulsion were calculated and averaged over all strokes of the 20-second trial. Push phase was defined as a deviation of pushrim force and moment data from baseline (0) and was determined through visual inspection kinetic data using Matlab. This represents the period in which the hand is contacting the pushrim. Resultant force at the shoulder was calculated by summing the directional force components. The timing of the peak loading, expressed as percentage of push cycle, was determined for the directional force and moments, as well as for resultant force. In addition, stroke frequency and mean velocity were calculated using data from the SmartWheel.

Statistical Analysis 

All statistical analyses were performed using SPSSf for Windows. Descriptive analysis including means, standard deviation, and frequencies for categoric variables was performed to summarize demographic (for each site individually and all sites combined) and biomechanic data. Demographic information was compared for differences between sites. Normally distributed, continuous variables (age, years since injury, height, weight) were compared using a 1-way analysis of variance (ANOVA). Median injury level, an ordinal variable, was compared using a Kruskal-Wallis test, and sex differences were examined using a Fisher exact test. Although not a major focus of this study, correlations between demographic variables were identified before creating linear regression models relating demographics to biomechanic variables. Pearson correlations were calculated between all demographic variables, excluding sex, which was examined using the Spearman ρ. The relationship between pain and demographics was investigated using the Fisher exact test for sex, Kruskal-Wallis H test for injury level, and independent samples t test for age, years since injury, height, and weight. Demographics were compared between subjects with shoulder pain and those without. Independent samples t tests were used to compare peak biomechanic variables (including stroke frequency, velocity, shoulder forces and moments, and shoulder kinematics) between subjects with shoulder pain and those without. An exploratory correlational analysis between demographics and biomechanic variables was performed to identify demographic variables most strongly related to propulsion biomechanics. These variables were then used in linear regression models to predict biomechanic variables (peak shoulder kinetics and kinematics) at each propulsion speed while controlling for the interrelation of demographic variables. Normalized group means for shoulder biomechanic variables including shoulder forces, moments, and Euler angles were calculated using 5 strokes from each subject to visually illustrate changes in biomechanics due to propulsion speed. A 1-way repeated-measures ANOVA was used to compare biomechanic variables between speed conditions. Bonferroni-adjusted pairwise comparisons were performed where appropriate. Nonparametric repeated-measures comparisons (Friedman test) were made between peak timing variables. The level of statistical significance was set at P less than .05 for all statistical analyses.

Demographics 

Table 1 summarizes subject characteristics by site, as well as for the overall subject group. Significant differences between sites were noted for years since injury (P=.008), height (P=.004), and sex (χ2=6.785, P=.034). Specifically, subjects at KMRREC were closer to the time of injury (P=.008) than those at HERL. Subjects from UW were shorter in stature than those from HERL (P=.006) and KMRREC (P=.025). UW recruited significantly more women than HERL (P=.023) or KMRREC (P=.028).

Table 1. Subject Characteristics

	Characteristics	HERL	KMRREC	UW	Overall
	n	21	20	20	61
	Age (y)	46.6±9.9	38.6±14.0	44.1±11.0	43.1±12.0
	Years since injury	19.4±9.0‡	9.5±11.0†	14.7±9.4	14.6±10.5⁎
	Height (m)	1.79±0.07§	1.78±0.07§	1.71±0.10†, ‡	1.76±0.09⁎
	Weight (kg)	77.5±11.0	79.8±16.2	70.4±13.4	75.9±14.0
	Injury level (median)	7.75(T7–8)	7.5(T7–8)	8.5(T8–9)	8.0(T8)
	Sex (male/female)	19/2§	18/2§	12/8†, ‡	49/12⁎

NOTE. Age, years since injury, height, and weight values are mean ± standard deviation (SD) and compared between sites using a 1-way ANOVA; median injury level is presented and compared for differences between sites using a Kruskal-Wallis test; sex, reported as n, was compared between sites using a Fisher exact test.

⁎ Significant difference between sites (P<.05). † Significantly different from HERL (Bonferroni-adjusted pairwise comparisons). ‡ Significantly different from KMRREC (Bonferroni-adjusted pairwise comparisons). § Significantly different from UW (Bonferroni-adjusted pairwise comparisons).

Correlation analysis showed that subjects who were older (r=.296, P=.021) and taller (r=.364, P=.004) tended to weigh more. Women tended to weigh less than men (ρ=−.253, P=.049) and were also shorter (ρ=−.559, P<.001). People who were further removed from injury also tended to be older (r=.604, P<.001). No other statistically significant relationships were identified between the variables listed in table 1.

Influence of Shoulder Pain 

Forty-six percent of wheelchair users (n=28) in this study reported experiencing shoulder pain in the last month. Sixteen subjects reported bilateral shoulder pain, and 12 experienced unilateral pain. Twenty-three percent of subjects (n=14) reported shoulder pain while propelling the wheelchair. Of those reporting shoulder pain in the last month, over 60% (n=17) had seen a doctor for treatment. Sex, injury level, years since injury, height, and weight did not affect whether subjects reported shoulder pain in the last month or during propulsion. Using a 1-way ANOVA, we found that participants who reported pain during propulsion tended to be younger (36.6±8.8y vs 45.1±12.2y) than those who did not. The presence of shoulder pain in the last month was not influenced by subject age.

When comparing subjects who reported shoulder pain in the last month with those who reported no shoulder pain, no significant differences in propulsion biomechanics were found. The same was true when comparing propulsion biomechanics between persons who did and did not experience pain while propelling. Because subjects with and without shoulder pain were so similar in terms of demographics and biomechanics, the data were combined for all further analyses.

Demographics and Biomechanics 

A preliminary correlational analysis showed that men, taller people, and heavier people tended to use larger forces. Some significant correlations between kinematics and sex, height, weight, and age were noted. However, this was not seen at all speed conditions or in all directions of shoulder motion. The independent variables used in the linear regression models to predict shoulder kinetics were sex, height, and weight. Propulsion velocity influences biomechanics, and therefore this was controlled for in the regression analysis. Sex, height, weight, age, and velocity were used in linear regression models with shoulder kinematics as the dependent variables.

Sex, height, and weight were forced into regression models while controlling for velocity to predict shoulder kinetics. Sex was not a significant predictor of peak shoulder kinetics in any of the linear regression models. Subject height was only a significant predictor of shoulder extension moment at the self-selected and 0.9m/s speed conditions. However, subject weight seemed to be the primary factor contributing to shoulder kinetics, because it was the single significant (P<.05) demographic predictor of shoulder kinetics in many directions. As an example, for a regression model predicting resultant force (self-selected: r2=.462, P<.001; 0.9m/s: r2=.560, P<.001; 1.8m/s: r2=.501, P<.001), body weight was the only significant independent variable (self-selected: β=.646, P<.001; 0.9m/s: β=.67, P<.001; 1.8m/s: β=1.002, P<.001). Heavier subjects experienced a higher resultant force during propulsion. At all speed conditions, increased subject weight was predictive of higher anterior force, posterior force, inferior force, and lateral force. Prediction of shoulder moments was less consistent among speed conditions. At the self-selected speed condition, body weight was not a significant predictor of shoulder moments when entered into a regression model with velocity, sex, and subject height. At the 0.9m/s speed condition, higher body weight predicted larger internal and external rotation moments. At the 1.8m/s speed condition, higher abduction, internal rotation, external rotation, and flexion moments were predicted by body weight.

Shoulder kinematics during propulsion were largely independent of subject characteristics when age, height, weight, sex, and velocity were entered into a regression model. However, some statistically significant relationships were noted: at all speeds, greater subject weight predicted less shoulder extension, and during the 1.8m/s condition, older subjects used less external rotation and shoulder flexion during propulsion.

Biomechanics and Propulsion Speed 

Fifty subjects had data from all 3 speed conditions, and therefore this subset of subjects was used to analyze differences in shoulder biomechanics due to propulsion speed. Two subjects were unable to reach the 1.8m/s condition. One subject was almost double the target speed for the 0.9m/s speed, and therefore these data were excluded. The remaining 8 subjects had erroneous and/or noisy data for at least 1 speed condition, which affected the peak values. Table 2 presents temporal data for each of the speed conditions, which were compared using a 1-way repeated-measures ANOVA. Mean velocity, as expected, was significantly different between all speeds. Mean self-selected speed was 1.09m/s, which fell in between the mean velocities of the 2 target speed conditions. Subjects used a higher stroke frequency to propel at the 1.8m/s speed compared with the self-selected or 0.9m/s condition.

Table 2. Temporal Data for 3 Speeds of Wheelchair Propulsion

	Temporal Data	SS Speed (n=50)	0.9m/s (n=50)	1.8m/s (n=50)	Significance of Speed (P)
	Mean velocity (m/s)	1.09±0.31‡, §	0.95±0.14†, §	1.74±0.18†, ‡	<.001⁎
	Stroke frequency (1/s)	0.95±0.25§	0.99±0.24§	1.29±0.30†, ‡	<.001⁎

NOTE: Values are mean ± SD. Abbreviation: SS, self-selected.

⁎ Significant difference between speed conditions (P<.05) detected using a 1-way repeated-measures ANOVA. † Significantly different from self-selected speed condition (Bonferroni-adjusted pairwise comparisons). ‡ Significantly different from 0.9m/s speed condition (Bonferroni-adjusted pairwise comparisons). § Significantly different from 1.8m/s speed condition (Bonferroni-adjusted pairwise comparisons).

Table 3 summarizes peak shoulder kinetics for the 3 propulsion speeds. Joint kinetics are strongly affected by propulsion velocity. All shoulder forces and moments followed the same pattern observed for the mean velocities, with the lowest values occurring for the 0.9m/s and the highest values resulting during the 1.8m/s trial. The kinetic values for the self-selected speed condition fell between these extreme values and tended to be closest to the quantities measured for the 0.9m/s trial. Significant differences due to speed, evaluated using a 1-way repeated-measures ANOVA, were noted for all shoulder joint forces and moments except the abduction moment. Specifically, the shoulder kinetics for the 1.8m/s speed were significantly higher than those for both the self-selected speed and the 0.9m/s trial. The only exception to this was inferior force; only the 0.9m/s and 1.8m/s values were statistically different. Statistically significant but small differences were noted for some variables between the self-selected and 0.9m/s conditions: posterior force, superior force, resultant force, adduction moment, internal rotation moment, and extension moment. All of these kinetic variables are active joint forces, meaning they occur as a result of the force applied to the wheelchair. The largest directional force component was the posterior force, and the internal rotation moment was the largest directional moment.

Table 3. Peak Shoulder Kinetics for 3 Speeds of Wheelchair Propulsion

	Kinetics	SS Speed (n=50)	0.9m/s (n=50)	1.8m/s (n=50)	Significance of Speed (P)
	Forces (N)
	Anterior force	17.0±10.6§	15.9±10.4§	29.5±15.9†, ‡	<.001⁎
	Posterior force	40.9±14.0‡, §	35.6±10.4†, §	55.2±17.7†, ‡	<.001⁎
	Superior force	6.7±20.6‡, §	−0.44±13.0†, §	21.2±25.8†, ‡	<.001⁎
	Inferior force	47.6±12.5	46.4±13.5§	52.4±19.3‡	.003⁎
	Medial force	9.2±7.5§	7.2±6.8§	14.2±11.3†‡	<.001⁎
	Lateral force	17.6±9.7§	16.1±8.4§	24.6±10.3†‡	<.001⁎
	Max resultant force	59.1±15.8‡§	54.4±13.5†§	75.7±20.7†‡	<.001⁎
	Moments (Nm)
	Abduction moment	3.2±1.7	2.9±1.7	3.4±2.3	NS
	Adduction moment	7.1±4.2‡§	5.4±3.4†§	10.6±5.3†‡	<.001⁎
	External rotation moment	5.1±2.8§	5.1±3.3§	7.4±5.1†‡	<.001⁎
	Internal rotation moment	15.3±6.1‡§	13.5±5.2†§	20.6±8.0†‡	<.001⁎
	Flexion moment	5.4±3.2§	4.8±2.7§	7.3±3.9†‡	<.001⁎
	Extension moment	10.8±6.0‡§	9.5±4.7†§	14.3±6.0†‡	<.001⁎

NOTE: Values are mean ± SD. Abbreviation: NS, not significant.

⁎ Significant difference between speed conditions (P<.05) detected using a 1-way repeated-measures ANOVA. † Significantly different from self-selected speed condition (Bonferroni-adjusted pairwise comparisons). ‡ Significantly different from 0.9m/s speed condition (Bonferroni-adjusted pairwise comparisons). § Significantly different from 1.8m/s speed condition (Bonferroni-adjusted pairwise comparisons).

Fig 2, Fig 3 depict the group means for shoulder forces and moments normalized to a single propulsion cycle. Both push and recovery phases are shown in the figures, although we focused on the push phase in this study. The transition between phases is represented by a band, rather than a single line, because the mean time of transition varied between speed conditions. The transition occurred at 42%, 46%, and 49% of the propulsion cycle for the 0.9m/s, self-selected, and 1.8m/s trials, respectively. Maximum joint loading tends to occur during the push phase in most directions; however, Fig 2, Fig 3 show some exceptions, such as inferior (see fig 2B) and medial force (see fig 2C). As the hand applies force to the pushrim to propel the wheelchair forward, the shoulder is loaded in the posterior direction (see fig 2A) and the humerus is pushed in the superior direction (see fig 2B), counteracting the weight of the arm. Through the push phase, the humerus also applies a more medially directed force (see fig 2C) to the glenoid. The application of force to the pushrim, along with the flexion, adduction, and external rotation of the humerus (fig 4) during the push phase, generates adduction (see fig 3A), internal rotation (see fig 3B), and extension (see fig 3C) moments about the shoulder.

View Large Image Download to PowerPoint

Fig 2. Group mean shoulder forces—(A) Fx, anterior and posterior force; (B) Fy, superior and inferior force; (C) Fz, medial and lateral force; and (D) resultant force—during 3 speeds of propulsion. The transition from push phase to recovery is shaded because it differs slightly between speed conditions.

View Large Image Download to PowerPoint

Fig 3. Group mean shoulder moments during 3 speeds of propulsion. (A) Mx, abduction and adduction moment, (B) My, external and internal rotation moment, and (C) Mz, flexion and extension moment. The transition from push phase to recovery is shaded because it differs slightly between speed conditions.

View Large Image Download to PowerPoint

Fig 4. Group mean shoulder Euler angles—(A) plane of elevation, (B) elevation, and (C) rotation—during 3 speeds of propulsion. The transition from push phase to recovery is shaded because it differs slightly between speed conditions.

Table 4 summarizes peak kinematics of the shoulder for the 3 propulsion speeds. Near the beginning of the push phase, the shoulder is maximally extended (see fig 4A), abducted (see fig 4B), and internally rotated (see fig 4C). Only small differences (<2°) were noted between speeds at this end of the shoulder range of motion (ROM). The shoulder is maximally flexed and minimally abducted and internally rotated near the end of the push phase. The shoulder had greater flexion and less internal rotation (closer to a neutral posture) at the end of the stroke with increasing speed. Minimum abduction was unaffected by propulsion speed. The lack of speed dependence is further illustrated in figure 4, which depicts group means for shoulder Euler angles normalized over a single push phase. Although the ROM is slightly larger for the fastest speed condition, the actual angle difference is minimal (only a few degrees).

Table 4. Peak Shoulder Kinematics for 3 Speeds of Wheelchair Propulsion

	Kinematics	SS Speed (n=50)	0.9m/s (n=50)	1.8m/s (n=50)	Significance of Speed (P)
	Shoulder angles
	Max extension (deg)	47.1±10.5	45.9±10.2§	47.1±9.3‡	.030⁎
	Max flexion (deg)	23.7±18.6‡§	15.6±17.8†§	27.6±17.3†‡	<.001⁎
	Max abduction (deg)	52.4±7.5	52.6±8.2	53.3±7.8	.042⁎
	Min abduction (deg)	30.5±6.2	30.8±6.6	31.4±6.2	NS
	Max internal rotation (deg)	83.9±11.5	83.1±12.0	83.7±10.4	NS
	Min internal rotation (deg)	9.8±23.0‡	15.6±20.0†§	6.9±21.0‡	<.001⁎

NOTE: Values are mean ± SD.

⁎ Significant difference between speed conditions (P<.05) detected using a 1-way repeated-measures ANOVA. † Significantly different from self-selected speed condition (Bonferroni-adjusted pairwise comparisons). ‡ Significantly different from 0.9m/s speed condition (Bonferroni-adjusted pairwise comparisons). § Significantly different from 1.8m/s speed condition (Bonferroni-adjusted pairwise comparisons).

Kinetic and Kinematic Timing 

Table 5 summarizes the timing of peak shoulder forces and moments during the push phase of propulsion. For all speed conditions, the peak resultant force at the shoulder occurred during the first half of the push phase. The same is true for most directional forces and moments. Figure 4 shows that the humerus is extended, abducted, and internally rotated throughout the push phase, with the most extreme postures occurring at the time of contact. Peak resultant force occurred later as speed increased. As a result, the humerus was less extended, abducted, and internally rotated (closer to neutral) at the time the shoulder experienced maximal joint force.

Table 5. Timing of Shoulder Kinetics for 3 Speeds of Wheelchair Propulsion

	Biomechanics Variables	Median Timing of Peak (% Push Phase)				Significance of Speed (P)
		SS Speed	0.9m/s	1.8m/s
	Anterior force	0.4	0.3	1.4	4.093	.129
	Posterior force	50.3§	43.7§	66.3†‡	23.510	<.001⁎
	Superior force	55.4‡	36.5†§	45.5‡	9.918	.007⁎
	Inferior force	17.2	10.8§	22.2‡	7.260	.027⁎
	Medial force	70.6§	70.8§	66.6†‡	10.082	.006⁎
	Lateral force	38.8‡	34.5†§	41.6‡	11.878	.003⁎
	Resultant force	38.7	25.9§	42.5‡	16.449	<.001⁎
	Abduction moment	9.8	12.7	11.9	2.358	.308
	Adduction moment	78.1	77.5	74.9	3.435	.180
	External rotation moment	0.8	0.6	1.3	2.667	.264
	Internal rotation moment	43.8‡	39.1†	37.1	9.702	.008⁎
	Flexion moment	6.8	4.5	6.4	5.144	.076
	Extension moment	36.9	30.3	37.9	4.667	.097

⁎ Significant difference between speed conditions (P<.05) detected using the Friedman test. † Significantly different from self-selected speed condition (Bonferroni-adjusted pairwise comparisons). ‡ Significantly different from 0.9m/s speed condition (Bonferroni-adjusted pairwise comparisons). § Significantly different from 1.8m/s speed condition (Bonferroni-adjusted pairwise comparisons).

Study Limitations 

One limitation of studies of wheelchair biomechanics, including the current study, is that testing is completed in a simulated environment, either on a dynamometer or wheelchair ergometer. Testing on a dynamometer allowed us to control velocity while testing each subject in his/her own wheelchair. Also, kinematic data can be collected for an extended (unlimited) period of time, and variability in propulsion surfaces and slopes is eliminated, which is particularly important in a multisite study design. However, the self-selected velocity observed in the current study is slower than would be expected for overground propulsion.29 Although these simulated testing environments afford many advantages for conducting research, future efforts should incorporate overground testing to fully describe wheelchair propulsion in real-world environments. Instrumentation can also affect the results in a study of wheelchair biomechanics. The SmartWheel is a widely used, validated device that allows researchers to monitor pushrim kinetics wirelessly. However, the SmartWheel may increase rolling resistance because it has a solid insert, but this ensures that tire pressure does not vary between subjects or sites, as is possible with pneumatic tires.

Although this study is descriptive in nature, future work will take further advantage of the large sample size afforded by a multisite recruitment effort. The relationship between shoulder biomechanics and injury, measured by magnetic resonance imaging, radiography, and physical examination, will be investigated. The effect of propulsion pattern and wheelchair setup on shoulder biomechanics may also provide insight into potential interventions to reduce shoulder loading during this repetitive task. The ultimate goal of this large, multisite study is to identify risk factors for upper-limb repetitive strain injuries resulting from wheelchair propulsion. This will lead to the development of interventions, such as propulsion training or changes to wheelchair equipment, that we hope will reduce the high incidence of shoulder pain and injury among manual wheelchair users.