Preliminary Outcomes of the SmartWheel Users’ Group Database: A Proposed Framework for Clinicians to Objectively Evaluate Manual Wheelchair Propulsion

Standard Clinical Protocol 

The SWUG designed the standard clinical protocol to match requirements identified by member clinicians as being critical to clinical acceptance and implementation. Four requirements were identified: (1) use of surfaces common to clinics, (2) use of multiple surfaces representing varied resistance, (3) provision of useful information from a single module, and (4) adaptability to a clinic’s available space and time.

The standard clinical protocol is a modular assessment that requires users to propel a manual wheelchair across (1) level tile, (2) low pile carpet, (3) up a ramp that complies with the requirements of the Americans with Disabilities Act (ADA) (maximum rise to run, 1:12; grade, 8.3%; slope, 5°), and (4) perform a figure of eight on level tile with a SmartWheel attached unilaterally to the wheelchair.20 Use of a SmartWheel that matches the opposing wheel diameter maintains the user’s wheelchair configuration. A standard SmartWheel weighs 4.9kg (1.1lb), which increases the weight of the wheelchair but provides measures of stroke length and force that cannot be measured in any other manner in a clinic.

In all modules, data collection was initiated before users began to move. On tile and carpet, users began from a stationary position, accelerated to a comfortable self-selected velocity and pushed for a maximum of 10 seconds, or 10m, or to the end of the surface, whichever occurred first. Data collection was terminated before users left the surface or decelerated. From level ground directly in front of the ramp, and with casters touching the ramp threshold, users propelled up the full length until reaching a platform. Data collection was terminated before the user ascended onto the platform. Ramp length and slope varied, as is permitted by the ADA. The fourth module, the figure 8, assessed an individual’s ability to maneuver and is not included in this analysis.

By design, the standard clinical protocol does not require clinicians to randomize or prioritize the order of the modules. Randomization within a clinical environment may not be possible or reasonable. Additionally, definitions of surfaces were loosely constrained to maintain the practicality of implementation. Low pile carpet was defined as closed loop industrial type carpet that is often found in hospitals, clinics, and some businesses. Tile was defined as smooth, firm panels—often linoleum—lining the floors of hospitals and clinics. Tiled ramps with a maximum grade of 8.3%, per ADA definition, could be used for the protocol. Clinicians are encouraged to assess client mobility over any surface they believe will provide relevant information; however, only collections that match any module of the standard clinical protocol can be submitted to the central data pool (described below).

The SmartWheel clinical software and standard clinical protocol define steady state as being all strokes after the third stroke. If the target velocity has been achieved, this represents a state of propulsion inherently different from the acceleration phase described by “start-up” parameters. Restrictions in available space and the increasing difficulty of modules (ie, a ramp), may prevent achievement of a “steady-state” condition as it is traditionally defined. A minimum of 5 strokes is required for the SmartWheel clinical software steady-state calculations, although all available strokes taken, beginning with stroke 4, are included in steady-state calculations. It is incumbent on clinicians to compare start-up and steady state for each client and module to determine if a steady-state condition has been achieved.

Key Parameter Selection 

When a module was completed, the SmartWheel clinical software automatically generated 21 parameters to describe a client’s propulsion.21 Four parameters were identified by the SWUG as the most clinically important and relevant information provided by the SmartWheela (velocity, average peak resultant force, push frequency, stroke length). Clinician members of SWUG felt that all assessments should begin with velocity and all users should be able to achieve a minimum threshold velocity for safe and successful participation in their communities. A velocity of 1.06m/s, representing the average minimum velocity needed to safely cross an intersection,22 was chosen as the threshold for the purpose of discussion in this article. Force, push frequency, and stroke length were selected by the SWUG based on recommendations from the Clinical Practice Guidelines for the Preservation of Upper Limb Function Following Spinal Cord Injury (CPG).23 The CPG recommends the minimization of force and frequency of repetitive upper-limb tasks and the use of long strokes during propulsion.23

This analysis is restricted to the 4 parameters identified by the SWUG, plus time and distance for each module. Forces are weight normalized for a subset of statistics. Clinicians can generate weight-normalized forces by dividing the SmartWheel clinical software’s output by a client’s weight. To facilitate clinical application, all parameters presented in this analysis, except for distance covered in the module and time to complete the module, were calculated using MatLabb in the same manner as were parameters calculated by the clinical software. Distance and time were truncated when necessary to include only 5 strokes, which is the minimum needed by the SmartWheel clinical software to generate a full report that describes start-up, steady state, and summary results. We limited our analysis to 5 strokes to mimic what a clinician who can only collect 5 strokes can potentially expect to see as a result.

We planned a series of graphs to help clinicians understand and apply this analysis. Figure 1 shows a generalized representation of these graphs. Each graph contains 3 critical elements—a threshold velocity reference line, a line representing the linear regression between the parameters of interest, and the 75% and 95% covariance ellipses—that define 4 areas of interest. The purpose of the linear regression was to visually represent the significant correlation between velocities and push frequency/force documented during our preliminary analysis. We designated average trial velocity as the dependent variable for regression, based on a preliminary analysis that identified it as the strongest correlate of force and push frequency. If a regression was not significant (P<.05), a graph was unnecessary and therefore not constructed. The proposed application and interpretation of these areas is described in the Discussion section.

View Large Image Download to PowerPoint

Fig 1. Generalized regression plot. Abbreviations: Area A, above threshold velocity; below average force or push frequency; Area B, above threshold velocity; above average force or push frequency; Area C, below threshold velocity; above average force or push frequency; Area D, below threshold velocity; below average force or push frequency. Legend: The solid line is the linear regression line based on existing data. The dashed line is the extrapolated linear regression line. Threshold velocity is represented by a dash-dot line. Solid line ellipse is the 95% covariance ellipse. Dashed line ellipse is the 75% covariance ellipse.

Central Data Pool 

All data used in this analysis were contributed de-identified to a central data pool housed at the Human Engineering Research Laboratories (HERL), with the approval of HERL’s institutional review board (IRB). The HERL has an approved IRB to house the central data pool. All subjects in that pool were assigned a unique identifier and all met 2 inclusion criteria: (1) a minimum of 1 raw SmartWheel collection file representing 1 module and (2) local IRB approval for contribution to the central data pool. Exclusion criteria were the failure to meet the inclusion criteria. Restated, an individual was eligible for the central data pool if a single standard clinical protocol module was completed. Each submission included user demographics and his/her wheelchair characteristics, if that information was available, as well as the raw SmartWheel file generated by the module. User demographics included age (if age <89y), height, weight, sex, primary diagnosis, and the number of years a wheelchair had been the main means of mobility. Wheelchair characteristics included the manufacturer, the model, and the weight. Submission to the pool of multiple data collections for each module for a person was permitted. For this analysis, we ensured that each person was represented only once by demographics and 1 raw collection for each module of the standard clinical protocol.

Kinetic Data Reduction and Analysis 

This report focuses on the subset of subjects with SCI or dysfunction (paraplegia, tetraplegia, spina bifida). For each SCI subject, up to the first 5 strokes of a single data collection session on tile, carpet, or ramp that was submitted to the central pool was selected for analysis. The figure 8 was not included because it is a skill assessment of maneuverability and is not accurately described by the parameters. Five strokes were the minimum needed to complete an assessment and to generate a full report by the SmartWheel clinical software. Each trial was broken down into start-up and steady state, mimicking the clinical software. The clinical software defines “start-up” as the first 2 strokes from a stationary position. Strokes 1 and 2 were analyzed and presented separately. The clinical software bases steady-state analyses on the average of all strokes beginning with stroke 4 and requires a minimum of 5 strokes. The statistics showed that resultant force and velocity were different for start-up and steady state. We used Matlab to trim data to 5 strokes, identify the beginning and end of each stroke, define start-up and steady state, and generate the parameters.

Key parameters were as defined as follows. The resultant force (F), the vector sum of the force applied to the pushrim, was calculated by mathematically combining Fx, Fy, and Fz (in newtons).24, 25 Stroke length was defined as the distance traveled by the hand on the pushrim from the point of contact to the point of release (in degrees). Push frequency was calculated for the entire trial and defined as the frequency of pushrim contact (in contacts per second). Steady-state velocity is the average velocity during strokes 4 and 5 (in m/s). The clinical software uses all strokes, starting with stroke 4, in the steady-state average; here this is limited to strokes 4 and 5. Start-up velocity is defined as the peak velocity occurring during the beginning of contact of stroke 2 until the beginning of contact for stroke 3 (in m/s). Distance (in meters) covered during the assessment and time (in seconds) to complete the assessment were calculated from the beginning of the first contact to the release of the last contact (up to 5 strokes).

Statistical Analysis 

All analyses were completed with SPSS.c The distributions of the data were inspected and descriptive statistics were calculated for demographics and each parameter for each module. Given the substantial contributions of the HERL to the pool, we compared data from HERL with that from other participants to determine how the overall data set might be affected. It was decided a priori to include all data to capture the population variability, but we wanted to acknowledge potential skew resulting from a heavily represented facility. We used analysis of variance to separately compare subject demographics and steady-state average velocity between HERL and the remainder of the pool contributors. A separate analysis of covariance (ANCOVA) for each steady-state peak resultant force, stroke length, and push frequency was used to compare HERL with the remainder on all 3 modules, controlling for participant weight and velocity.

To examine the differences in all key variables between surfaces, we used a multivariate analysis of variance (MANOVA). Because we wanted to use all available data, we did not use repeated measures because that would have reduced our sample size to only those subjects who completed all 3 portions of the protocol. Use of an MANOVA decreases our power to detect differences and is a conservative approach. For the clinically oriented graphs, we used linear regression to investigate the relation between (1) weight normalized steady-state peak resultant force and steady-state velocity and (2) push frequency and steady-state velocity for each surface. Covariance ellipses at 75% and 95% were calculated to represent the variability in the parameters. Clinical reference graphs based on the linear regression and covariance ellipses were constructed using Matlab.

Demographics 

There were 128 subjects available for inclusion in the analysis. Maximum numbers available for analyses of the separate surfaces were as follows: 123 for tile; 94 for carpet; and 115 for ramp. Six facilities contributed subjects: HERL, Pittsburgh, PA (n=57), University College London, London, UK (n=22), Banner Good Samaritan Rehabilitation Institute, Phoenix, AZ (n=20), University of British Columbia, Vancouver, BC (n=17), Washington University, St. Louis, MO (n=6), and the Mayo Clinic, Rochester, MN (n=6). Table 1 lists the demographics of the cohort.

Table 1. Participant Demographics

	Characteristic	N	Mean ± SD or Count
	Age (y)	128	40.4±11.2
	Height (cm)	122	176.5±10.7
	User weight (kg)	85	80.8±19.9
	Wheelchair weight (kg)	87	13.2±10.2
	Duration of wheelchair use (y)	128	13.2±10.2
	Sex	128
	Male		102
	Female		26
	Diagnosis	128
	Paraplegia		88
	Tetraplegia		22
	SCI level unknown		11
	Spina bifida		7
	Wheelchair make	128
	Quickie		69
	Invacare		26
	TiLite		12
	Colours		6
	Other		15

Abbreviation: SD, standard deviation.

HERL Participants Compared With All Other Participants 

Participants submitted by HERL were significantly older, taller, and heavier than the remainder of the participants in the database. When weight was entered into the ANCOVA as a covariate, there were differences in kinetic parameters for stroke 1 peak force on carpet, stroke 2 and steady-state weight normalized forces on ramp, steady-state velocity on tile and carpet, and start-up velocity on ramp. A substantial number of the participants contributed by HERL were collected at the 2004 and 2005 National Veterans Wheelchair Games. As a group, these individuals tended to be older and heavier, but had high functional capabilities. All facilities were used in the analysis so that the reported values would capture the largest amount of variability found within the SCI population.

Description of Key Parameters 

Table 2 shows the overall means, 95% confidence intervals (CIs) for the mean, and standard deviations (SDs) for each parameter for each module. Generally, stroke 2 peak resultant force was highest and steady-state average peak force was lowest. The start-up phase during the ramp condition represents the ascent onto the ramp from the flat area directly in front of the ramp. Steady-state average peak resultant force increased as module difficulty increased. Self-selected steady-state average velocity decreased as module difficulty increased. Start-up peak velocity was significantly different between tile and carpet and ramp, but not between carpet and ramp. Stroke length was similar on all surfaces. Push frequency also was similar on all surfaces, regardless of differences in self-selected velocity or resultant force. Summarized, users selected a lower velocity as surface difficulty increased, using increased forces at the same push frequency and stroke length. The increase in force applied apparently did not offset the increased resistance offered by the surface, resulting in a decrease in velocity.

Table 2. Descriptive Output for the Key Parameters

	Parameters	Tile	Carpet	Ramp
	Stroke 1 peak resultant force (N)⁎†	99.5±31.1	103.9±33.2	117.7±36.8
		93.9–105.2	97.1–110.7	110.9–124.5
		116	92	112
	Stroke 2 peak resultant force (N)⁎†‡	91.2±27.8	106.6±33.1	128.3±35.3
		86.3–96.1	99.8–113.3	121.9–134.8
		123	93	115
	Steady-state average peak resultant force (N)⁎†‡	72.3±25.3	87.5±28.5	126.2±34.0
		67.6–77.1	81.2–93.8	119.8–132.7
		110	79	106
	Start-up peak velocity (m/s)⁎‡	1.2±0.3	1.1±0.3	1.1±0.3
		1.2–1.3	1.0–1.1	1.0–1.1
		123	93	115
	Steady-state average velocity (m/s)⁎†‡	1.2±0.3	1±0.3	0.7±0.3
		1.1–1.2	0.9–1.0	0.7–0.8
		110	79	106
	Steady-state average stroke length (deg)	100.6±18.0	97.2±19.6	94.1±20.6
		97.2–104.0	92.9–101.5	90.2–98.0
		110	79	106
	Entire trial average push frequency (contacts/s)	1.0±0.2	1.0±0.2	1.0±0.2
		0.98–1.07	0.97–1.04	0.98–1.06
		123	94	115
	Trial time for 5 strokes (s)	5.1±1.1	5.2±1.0	5.2±1.3
		4.9–5.3	5.0–5.4	4.9–5.4
		110	79	106
	Trial distance for 5 strokes (m)⁎†‡	5.4±1.4	4.5±1.2	3.7±0.9
		5.2–5.7	4.2–4.8	3.5–3.8
		110	79	106

NOTE. Values are mean ± SD, 95% CI, and n.

⁎ Significant differences between tile and ramp (P≤.05). † Significant differences between tile and carpet (P≤.05). ‡ Significant differences between carpet and ramp (P≤.05).

Linear Regression and Clinical Graphical Reference 

Outliers were excluded from the linear regression and covariance ellipses. Outliers were defined as follows: less than −1.5 × the interquartile range (IQR) below quartile 1 or greater than 1.5 × IQR above quartile 3. For each regression and covariance ellipses, this resulted in exclusion of a maximum of 5% of each available dataset. Weight-normalized average peak resultant force was a significant predictor of self-selected steady-state velocity on tile and ramp conditions. Push frequency was a significant predictor of average self-selected velocity for the entire trial on all 3 surfaces. Table 3 shows β and R2 for each regression model.

Table 3. Regression Coefficients by Model to Predict Average Speed

	Predictor	Tile	Carpet	Ramp
	Steady-state weight-normalized average peak resultant force	.043 (.213) P≤.000, n=70	−.012 (.008) P≤.521, n=54	.030 (.151) P≤.001, n=70
	Entire trial push frequency	.384 (.110) P≤.000, n=119	.506 (.140) P≤.000, n=93	.857 (.395) P≤.000, n=111

NOTE. Values are β, (R2), P value, and n. Only calculated for trials with 5 strokes.

We plotted a corresponding regression line (Fig 2, Fig 3, Fig 4, Fig 5, Fig 6) for each significant relationship shown in table 2. Additionally, we plotted 75% and 95% covariance ellipses, which allows clinicians to determine where their clients fall in the variability of this population. A reference line was placed at 1.06m/s on the y axis for each regression. This line represents the average minimum walking velocity required to safely cross an intersection.22, 26

View Large Image Download to PowerPoint

Fig 2. Tile body weight−normalized average steady-state peak resultant force versus average steady-state velocity.

View Large Image Download to PowerPoint

Fig 3. Ramp body weight−normalized average steady-state peak resultant force versus average steady-state velocity.

View Large Image Download to PowerPoint

Fig 4. Tile push frequency versus average velocity.

View Large Image Download to PowerPoint

Fig 5. Carpet push frequency versus average velocity.

View Large Image Download to PowerPoint

Fig 6. Ramp push frequency versus average velocity.

Study Strengths and Limitations 

Clinical application and interpretation of our results requires an understanding of the study’s limitations and strengths. Inherent variability in participants, protocol administration, surface selection, and equipment modification across facilities increases the data variability. Such variability obscures relationships between parameters and differences within a parameter when modules are compared. We believe the large numbers of participants has minimized this limitation and the impact of variability will decrease as the database matures. Furthermore, we believe this variability is a key strength because it allows this database to encompass the natural variability in users and environments. Approximately half of the current database was contributed by HERL, which may have skewed the results. Any skew is expected to decrease as the database expands through submissions from additional members, thus limiting the impact of any single facility’s contribution.

Limiting our steady-state analysis to 5 strokes increases variability, which possibly obscures relationships and differences. Increasing the number of strokes used in the analysis would decrease the variability. Moreover, individuals may not reach “steady state” by strokes 4 and 5. In a confined space, clinicians should consider comparing multiple collections over a single surface to determine what is “typical” for a client. In an attempt to mimic the bare minimum that a clinician might have on which to base their decisions, we limited our analysis to 5 strokes, which is the minimum required by the SmartWheel to generate a clinical report.

In comparing the start-up and steady state between various modules, clinicians should be aware that start-up parameters describing the ramp portion of the protocol represent the transition from level ground to a ramped surface. Clinicians interested in propulsion purely on a ramped surface should restrict their inspection to steady-state parameters. The transition from a level surface onto a ramp captured by start-up parameters may be a unique challenge for some users, which represents a point of evaluation in select instances.

Our analyses were restricted to subjects with SCI, who represent a unique group among the community of manual wheelchair users. Clinicians who evaluate non-SCI manual wheelchair users should be aware that their clients’ propulsion may differ substantially from this cohort. Use of velocity to evaluate the potential of a manual wheelchair user to achieve successful community function is not diagnosis specific. Any manual wheelchair user should be able to achieve a minimal velocity for functional purposes, regardless of his/her diagnosis. This is consistent with the CMS national coverage determination, which bases coverage of power and manual mobility on function, independent of diagnosis.35

Proposed Clinical Application Framework 

We present a proposed framework with which to guide clinicians to intervention opportunities through evaluation of velocity in context with push frequency and force (fig 7). Each clinical reference graph (see Fig 2, Fig 3, Fig 4, Fig 5, Fig 6) is divided into 4 areas by a threshold velocity line and the regression line for the force/push frequency and velocity regression (see fig 1). Covariance ellipses allow clinicians to visualize variability in this population and determine how their clients compare. Reference values in the absence of a velocity context provide a general comparison point (see table 2). Clinicians can generate the body weight−normalized force shown in Fig 2, Fig 3, Fig 4, Fig 5, Fig 6 by dividing output of the SmartWheel by a client’s body weight.

View Large Image Download to PowerPoint

Fig 7. Clinician decision-making flowchart. Abbreviations: see figure 1; MWU, manual wheelchair user.

Based on clinical guidance, the proposed framework prioritizes velocity over force/push frequency. Research indicates forces and push frequency are related to upper-extremity injury and minimizing these parameters is recommended to delay upper-extremity deterioration.16, 23, 36 Our clinicians report, however, that low forces were a trademark characteristic of low self-selected speed. In this situation, the clinicians’ priority was to increase a user’s ability to self-select higher speeds. After a “threshold” speed was achieved, clinicians sought to minimize force and push frequency. The idealized goal for any user is an above threshold velocity coupled with low force and push frequency on multiple surfaces (see fig 7, diamond box).

Application Process 

Figure 7 pictures the clinical progression through velocity, force, and push frequency evaluations and intervention opportunities. Self-selected velocity is a traditional indicator of present and future function in the ambulatory population.37, 38, 39, 40, 41 We selected the average walking velocity required to safely cross an intersection (1.06m/s) as our threshold. Clinicians may modify the threshold as needed. If a manual wheelchair user propels below threshold velocity on 1 or multiple surfaces (see fig 1, areas C and D), a clinician’s initial goal is to design an intervention that will help the client achieve threshold velocity (see fig 1, areas A and B). Such interventions could include combinations of strength training, propulsion training, and alterations in the client’s current chair set-up, or use of a lighter weight, more adjustable chair. Initial interventions may also include power mobility options if, through experience, a clinician believes that the above interventions are inappropriate for his/her client. Ultimately, it is the user who must choose between manual or power mobility.

Once threshold velocity is reached, clinicians then attempt to preserve velocity while minimizing force and push frequency to help delay the onset of upper-extremity pain and dysfunction (see fig 1, area A), in accord with the recommendations of the Consortium for Spinal Cord Medicine.23 Velocity achieved at the expense of high force or push frequency may unnecessarily increase the risk of upper-extremity pain and dysfunction (see fig 1, areas B and C). The ideal goal is for the user to propel above threshold velocity at below average force or push frequency across all surfaces (see fig 1, area A). Users pushing with above average force or push frequency at below threshold velocity may require powered mobility (see fig 1, area C).

Assessment Without a SmartWheel 

Clinicians without a SmartWheel may still use the push frequency graphical references. They can mark a 10-m path on a tile, carpet, or ramped surface and record the time taken by their clients to complete the path. Users start from a stationary position and accelerate to a comfortable velocity, pushing through the finish line. The number of times the client pushes in covering the distance is recorded. Velocity and push frequency is calculated as follows:

As an estimate, these numbers can be used to compare users with those in the central data pool. In addition, clinicians can utilize the proposed framework. Velocity and push frequency assessments and intervention paths are useful even without knowledge of force. Clinicians can advise clients to use long, smoothly applied strokes at a low push frequency to minimize force at any velocity.23

Suggestions for Determining Important Clinical Changes 

Clinicians may wonder about the importance of a change in velocity. Unfortunately, we do not yet have concrete numbers, but the literature indicates small changes in velocity could have a functional impact. There are small differences in self-selected velocity between SCI levels on a given surface or condition.29, 31, 42 Self-selected velocity differed by 0.4ms between tetraplegia (0.8ms) and paraplegia (1.2ms) subjects in Beekman et al,42 a difference of 50%. The absolute and relative difference between a preferred walking velocity (1.22m/s) and the minimum velocity (1.06m/s) needed to safely cross an intersection is even smaller, 0.16m/s,22, 26 a 15% difference. Because of these small, absolute but functionally important, differences, clinicians may argue that a small, consistent increase in self-selected velocity is important.

Similarly, research has yet to identify absolute or relative force/push frequency thresholds linked to the development or prevention of upper-extremity pain. The amount of force needed to propel a wheelchair, however, is small, highly repetitive, and related to upper-extremity injury.19, 36, 43, 44 The CPG recommends minimizing push frequency and force to cumulatively decrease exposure; reducing the development of upper-extremity dysfunction.23 Until thresholds are identified, small systematic reductions in force or push frequency are considered beneficial. Therefore, it is reasonable for clinicians to argue that consistent but small force or push frequency reductions at a given velocity postintervention represent the objective success of an intervention. Interventions that reduce both force and push frequency while maintaining velocity provide the strongest evidence of success.

Future Directions 

As the central data pool grows, sex, age, and diagnosis-specific reference values can be defined. Reference values to evaluate manual wheelchair propulsion without a SmartWheel should be developed for use by all clinicians. A biomechanical-focused exploration of the data is warranted. Values for all parameters should be periodically recomputed to ensure that the largest possible population is represented in the database. Eventually, these values could become normative, providing a criterion standard for clinicians.