Abstract
Background The group-level responsiveness of the original, 30-item Stroke Rehabilitation Assessment of Movement measure (STREAM-30) is similar to that of the simplified STREAM (STREAM-15), even though the STREAM-30 has twice as many items as those of the STREAM-15.
Objective The purpose of this study was to compare the responsiveness of the STREAM-30 and STREAM-15 at both group and individual levels in patients with stroke. For the latter level, the Rasch-calibrated 27-item STREAM (STREAM-27) was used because the individual-level indexes of the STREAM-30 could not be estimated.
Design A repeated-measurements design was used. In total, 195 patients were assessed with the STREAM-30 at both admission and discharge.
Methods The Rasch scores of the STREAM-27 and STREAM-15 were estimated from the participants' responses on the STREAM-30. We calculated the paired t-test value, effect size, and standardized response mean as the indexes of group-level responsiveness. The significance of change for each participant was estimated as the individual-level responsiveness index, and the paired t test and test of marginal homogeneity were used for individual-level comparisons between the STREAM-27 and STREAM-15.
Results At the group level, the STREAM-30, STREAM-27, and STREAM-15 showed sufficient and comparable responsiveness. At the individual level, the STREAM-27 detected significantly more participants with significant improvement and fewer participants with no change or deterioration compared with the STREAM-15.
Limitations Few patients with subacute stroke showed deterioration at discharge, so the abilities of the 2 measures to detect deterioration remain inconclusive.
Conclusions The STREAM-27 detected more participants with significant recovery compared with the STREAM-15, although the group-level responsiveness of the 2 measures was the same. The STREAM-27 is recommended as an outcome measure to demonstrate the treatment effects of movement and mobility for patients with stroke.
Responsiveness can be defined as the ability of a measure to detect change when it has occurred.1–3 Outcome measures should be responsive in order to demonstrate treatment effects and monitor patients' progress.4 Some authors have argued that the responsiveness of a measure should be examined at both the group level (detection of the average amount of change in a group) and the individual level (detection of change in an individual patient).2 To date, most studies examining the responsiveness of measures have used group-level indexes, which show the average treatment effect of a patient population.2,3 However, group-level indexes can hardly be applied to interpret change in individual patients in clinical or research settings,5 and individual-level indexes can help interpret individual patients' treatment effects that group-level indexes cannot. Therefore, both the group-level responsiveness and the individual-level responsiveness of an outcome measure are critical for researchers and clinicians to identify changes over time in patients.
Moreover, it has been argued that group-level comparisons of the responsiveness of competing measures (eg, long-form and short-form measures) potentially misrepresent the responsiveness of measures, possibly leading users to select a rather unresponsive outcome measure.6–8 In the past, short-form measures were recommended for efficiently quantifying patients' progress due to their briefness, speediness, and ability to detect change (ie, group-level responsiveness) similar to that of the long-form measures.9–12 Nevertheless, in these studies, only group-level comparisons were conducted, casting doubt on the implications of the findings. Previous studies compared the responsiveness of short-form and long-form measures (eg, the 5-item short-form Postural Assessment Scale for Stroke Patients [PASS] versus the 12-item PASS, the 12-item short-form Fugl-Meyer Motor Scale [FM] versus the 50-item FM) at both group and individual levels.6–8,13 The short-form measures were found to have inferior individual-level responsiveness, although both short-form and long-form measures have equivalent group-level responsiveness. Thus, the responsiveness of outcome measures should be compared at both group and individual levels to provide comprehensive and robust evidence for clinicians and researchers to select from among competing measures.
Movement and mobility impairments are very frequent sequelae after stroke.14 To plan treatment and monitor progress, it is crucial to assess movement and mobility functions in patients with stroke. The original, 30-item Stroke Rehabilitation Assessment of Movement (STREAM-30) is one of the recommended objective and quantitative outcome measures for assessing voluntary upper and lower limb movement (UL and LL) and basic mobility (MO) difficulties after stroke.15,16 Its psychometric properties (reliability, validity, and group-level responsiveness) have been well supported by previous studies based on the classical test theory.16–20 The STREAM-30 is recommended and preferred over other related movement or mobility measures for measuring and monitoring motor function in patients with stroke.19,20 However, the STREAM-30 has a weakness in the construct validity revealed by Rasch analysis.21 Furthermore, the STREAM-30 requires 15 to 35 minutes to complete.22 Such a time-consuming administration limits the utility of the STREAM-30 in clinical settings. Accordingly, the 27-item STREAM (STREAM-27) and its short form, the 15-item simplified STREAM (STREAM-15), have been constructed to improve the psychometric properties and clinical utility of the STREAM-30.21
The STREAM-15 is brief and quick to administer, and its psychometric properties are similar to those of the STREAM-30, particularly the group-level responsiveness.10,21,23 However, as mentioned above, the similarity in group-level responsiveness does not guarantee similar abilities to detect change in an individual patient. Due to the lack of evidence supporting the individual-level responsiveness, clinicians and researchers cannot select a more responsive measure that can detect changes in movement and mobility in more patients. To ameliorate this situation, this study aimed to compare the responsiveness of the STREAM-30, STREAM-27, and STREAM-15 at both group and individual levels in patients with stroke. However, the STREAM-30 was not used in the individual-level comparison because the Rasch scores of the STREAM-30, used for calculating individual indexes, could not be estimated. Because the STREAM-27 has more items and thus provides more information for assessing movements and mobility, we hypothesized that the STREAM-27 would have superior individual-level responsiveness.
Method
Participants
The data used in this study were obtained from a previous longitudinal follow-up study.10 Participants were recruited from the departments of physical medicine and rehabilitation at 5 hospitals in Taiwan between April 2004 and October 2005. Patients were eligible for the study if they met the following criteria: (1) first or recurrent onset of cerebrovascular accident without other major diseases (eg, cancer, amputation, severe rheumatoid arthritis), (2) subacute stroke with hemiparesis or hemiplegia, (3) ability to follow instructions to complete the measure, and (4) informed consent given personally or by proxy. Patients discharged within 1 week of admission for rehabilitation were excluded.
Procedure
Each participant was assessed with the STREAM-30 at admission to the rehabilitation wards and reassessed at discharge. The assessments were administered by 1 of 12 therapists (3 physical therapists and 9 occupational therapists) who were experienced in using the STREAM-30. The interrater reliability of these raters on the 3 subscales of the STREAM-30 was sound (intraclass correlation coefficients >.96).10 The data of both the STREAM-27 and STREAM-15 were obtained from the STREAM-30 data of the same sample.
Measures
The STREAM-30 contains 30 items equally divided into 3 subscales (ie, 10 items in each of the UL, LL, and MO subscales).16 The items of the UL and LL subscales are scored on a 3-point ordinal scale (0–1–2) for rating the excursion and quality of limb movement as compared with the less impaired side. The items of the MO subscale are scored on a 4-point ordinal scale (0–1–2–3) for rating the quality, completion, and assistance required of the task. A total of 20 points can be obtained from each limb subscale, and 30 points can be obtained from the MO subscale.
The STREAM-27 was revised from the STREAM-30 based on Rasch analysis to improve the psychometric properties of the STREAM-30.21 The responses of 3 items in the STREAM-30 (2 UL susbscale items: scapular elevation and opposition; 1 LL subscale item: hip abduction) did not conform to the Rasch model's assumptions24 (ie, unidimensionality and equal item discrimination). Thus, the misfit items were deleted to obtain the advantages of Rasch analysis, such as the resulting scores can be viewed as intervals rather than ordinal scores, and the standard error (SE) of each score can be estimated for users.21 The STREAM-27 consists of 27 items: 8 UL subscale items, 9 LL subscale items, and 10 MO subscale items. The scaling of each subscale is the same as that of the STREAM-30. All 27 items in their respective subscales showed good fit to the Rasch model. The 3 subscales of STREAM-27 have good-to-excellent Rasch reliability.21
The STREAM-15 was constructed to improve the administrative efficiency of the STREAM-30. The STREAM-15 was developed from the STREAM-27 by retrieving 5 items in each of the subscales according to Rasch analysis and item representativeness.21 The scaling of the STREAM-15 is identical to that of the STREAM-30. The psychometric properties of the STREAM-15 are comparable to those of the STREAM-30 and STREAM-27, including the Rasch reliability, test-retest reliability, unidimensionality, concurrent validity, predictive validity, discriminative validity, and group-level responsiveness.10,21,23
The Barthel Index (BI) was developed to assess basic activities of daily living (ADL) function in patients with neurological or musculoskeletal problems.25 The total score ranges from 0 to 20. The BI has satisfactory psychometric properties in patients after stroke.26,27 We used the BI as the external criterion to investigate the external responsiveness of the 3 STREAM measures.
Data Analysis
In this study, we used the raw scores of the STREAM-30 and the Rasch scores of the STREAM-27 and the STREAM-15 for data analyses. We estimated the Rasch scores and standard errors of the 3 subscales for both the STREAM-27 and the STREAM-15 using the multidimensional Rasch rating scale model. The Rasch score for each subscale was linearly transformed to a range of 0 to 100 for ease of understanding. The linear transformation equation for each subscale was: Rasch score × 100 ÷ (maximum Rasch score − minimum Rasch score). All item parameter estimates (ie, difficulty logits) were fixed to the values obtained from the original data in the study by Hsueh et al.21 This process was used to ensure that the Rasch parameters of each item were consistent across assessments at the 2 time points.
Comparison of the group-level responsiveness.
To examine the group-level responsiveness of the 3 subscales of the STREAM-30, STREAM-27, and STREAM-15, both the internal and external responsiveness were inspected. Regarding the internal responsiveness, we calculated 3 indexes: paired t test, effect size (ES), and standardized response mean (SRM). Paired t tests were performed to determine the significance of the change scores in each subscale from admission to discharge. Two-sided P<.05 was considered statistically significant. Both the ES and SRM provided information on the magnitude of change. Effect size was calculated by dividing the mean change scores by the standard deviation of the baseline scores.28 The SRM was calculated by dividing the mean change scores by the standard deviation of the change scores.28 Effect size and SRM values ≥0.80 were considered large, values of 0.50 to 0.80 were considered moderate, and values of 0.20 to 0.50 were considered small.29,30 To investigate the external responsiveness, the correlation between change scores in each subscale of the 3 STREAM measures and those in the external criterion, the BI, was calculated using the Pearson correlation coefficient (r). Pearson coefficients ≥.60 were considered high, values of .30 to .60 were considered moderate, and values ≤.30 were considered poor.31 A moderate correlation was considered to indicate sufficient external responsiveness.
We used the bootstrap approach32 to compare statistical differences of the group-level internal and external responsiveness among the 3 STREAM measures. We drew 10,000 bootstrap samples, each equal in size to the number of participants observed. For each of the 10,000 samples, we calculated the ES, SRM, and r values for each subscale and calculated pair-wise differences of the ES, SRM, and r values among the 3 STREAM measures. We examined whether zero was included in the 98.4% bootstrap percentile confidence intervals (CIs) of differences. The critical alpha level was adjusted down to .016 (ie, α/n comparisons=.05/3) applying the Bonferroni correction for multiple comparisons. If zero was not included, the group-level internal or external responsiveness of those 2 measures was considered significantly different.
Comparison of the individual-level responsiveness.
Both the internal and external responsiveness were investigated at the individual level. To examine the individual-level internal responsiveness of the STREAM-27 and STREAM-15, we estimated the significance of change (SC) for each participant. The SC was adopted to quantify the size of an individual participant's Rasch change score between admission and discharge, which was expressed in units of his or her own SE of the Rasch score.6 The SE for each participant's score was estimated by Rasch analysis. The SC was calculated using the following formulas6,33:
A brief description of the deduction of the SEdifference formula is provided in the Appendix.
With respect to the external responsiveness, we calculated the percentages of the participants who had significant improvement in each STREAM measure and had improvement above the minimal clinically important difference (MCID) in the BI. Significant improvement in the STREAM measures was determined as an SC ≥1.96 × linear transformation equation. The MCID of the BI, the lowest benchmark to determine whether the changes are important to patients with stroke, has been estimated to be 1.85.34
To compare the individual-level responsiveness between the STREAM-27 and STREAM-15, we conducted paired t tests, marginal homogeneity tests (G2 statistics), and tests for difference of proportion (DP) in the comparisons of internal responsiveness, as well as DP in the comparisons of external responsiveness. In terms of the internal responsiveness, paired t tests were adopted to determine the statistical differences of the mean SCs of both measures. In addition, we used G2 statistics to examine whether the distributions of participants detected to achieve different levels of SC by the 2 measures were different. We categorized the SCs according to their sizes and directions into the following 3 groups: (1) significant improvement: SC ≥1.96 × linear transformation equation; (2) nonsignificant improvement: 0 < SC < 1.96 × linear transformation equation; and (3) no change or deterioration: SC ≤0. A P value of <.05 in G2 statistics was considered statistically significant. The 95% CI for the DP was used to compare the 2 proportions in each SC group. Zero in this interval meant that there was no difference between the 2 proportions. As for the comparisons of the external responsiveness, we examined whether zero was included in the 95% CI for the DP between the 2 STREAM measures in each subscale.
Floor and ceiling effects.
The floor and ceiling effects of the 3 subscales in each STREAM measure were examined. We calculated the percentages of the participants who obtained the lowest and highest possible scores to indicate floor and ceiling effects, respectively. Floor and ceiling effects >15.0% were considered significant.35
Role of the Funding Source
This study was supported by research grants from the E-Da Hospital (EDAHT102015 and EDAHT103027).
Results
A total of 388 participants were assessed using the STREAM-30 at admission, but 193 of these participants were lost at second assessment either because they were in an unstable condition or were discharged without prior notification. The remaining 195 participants (50.3%) completed the assessments at both admission and discharge, and their data were used for further analyses. The baseline scores of the 3 subscales of the STREAM-30 of the remaining 195 participants were not significantly different (P>.05) from those of the 193 participants lost at second assessment. The scores of the 195 participants were scattered across the entire ranges of the 3 subscales' possible scores.10 As expected, deterioration during hospitalization was detected by the STREAM-30 in a few of the participants (10, 12, and 6 participants for the UL, LL, and MO subscales, respectively). The demographic characteristics of the participants and the raw scores and Rasch scores of the STREAM-30, STREAM-27, and STREAM-15 are shown in Table 1.
Demographic Characteristics and Descriptive Statistics of the STREAM-30, STREAM-27, and STREAM-15 in Participants With Stroke (n=195)a
Comparison of the Group-Level Responsiveness
Table 2 presents the scores for group-level internal and external responsiveness of the STREAM-30, STREAM-27, and STREAM-15. Regarding the internal responsiveness, the changes in score between admission and discharge of each subscale in the 3 measures were all significant (t=7.89–15.05, all P values <.001). The ES and SRM values showed the 3 STREAM measures to have similar and small-to-large group-level internal responsiveness. As to the external responsiveness, the correlation between the change scores of each subscale and those of the BI were similar among the 3 STREAM measures. The UL and LL subscales demonstrated moderate external responsiveness, and the MO subscale showed high external responsiveness.
Group-Level Internal and External Responsiveness of the STREAM-30, STREAM-27, and STREAM-15 (n=195)a
Table 3 shows the results of comparisons of group-level internal and external indexes among the 3 STREAM measures. In terms of the internal indexes (ES and SRM), zero was included in most 98.4% CI values of the differences. Particularly, the STREAM-15 had ES and SRM values similar to those of the STREAM-30 and STREAM-27, except for the comparisons in UL and LL subscales between the STREAM-15 and STREAM-30. That is, the ES and SRM values of the UL subscale and the SRM values of the LL subscale of the STREAM-15 were significantly higher than those of the STREAM-30, but the magnitudes of difference were small. Concerning the external indexes (r values), all of the 98.4% CI values embraced zero, indicating that the external responsiveness did not differ significantly among the 3 STREAM measures.
Bootstrap Analyses for Comparisons of Group-Level Internal and External Responsiveness Among the STREAM-30 (Raw Score), STREAM-27 (Rasch Score), and STREAM-15 (Rasch Score) in Participants With Stroke (n=195)a
Comparison of the Individual-Level Responsiveness
The results for individual-level internal and external responsiveness of the STREAM-27 and STREAM-15 are listed in Table 4. For internal responsiveness, the mean SCs of the STREAM-27 were higher than those of the STREAM-15, and the differences were significant in all 3 subscales (t=3.56–6.00, all P values <.001). The G2 statistics show that the distributions of participants in the 3 SC groups were significantly different between the STREAM-27 and STREAM-15 (P=.002–.011). In terms of the proportions in each SC group, more participants were detected to achieve “significant improvements” of LL and MO function by the STREAM-27 (22.6%–35.9%) than by the STREAM-15 (18.5%–29.2%). The UL subscales of both measures exhibited trends similar to those of the LL and MO subscales. On the other hand, a significantly smaller proportion of participants were categorized as having “no change or deterioration” in each of the 3 subscales by the STREAM-27 (12.8%–18.5%) than by the STREAM-15 (18.5%–24.1%). Regarding the external responsiveness, significantly more participants were found to have significant/important improvement by both the STREAM-27 and BI (20.0%–32.8%) than by both the STREAM-15 and BI (17.4%–26.7%) in the MO subscale. The UL and LL subscales also showed similar trends.
Individual-Level Internal and External Responsiveness of the STREAM-27 (Rasch Score) and STREAM-15 (Rasch Score) (n=195)a
The Figure shows the distributions of the SE values (vertical axis) across UL, LL, and MO subscale functions (horizontal axis) for the STREAM-27 and STREAM-15 at admission and discharge. The curves of the SE values of both measures appeared to be parallel. Particularly, at every time point and subscale, a majority of the SE values of the STREAM-15 were larger than those of the STREAM-27, and the mean SE values of the STREAM-15 were significantly larger than those of the STREAM-27 (t=22.46–46.56, all P values <.001).
Comparison of standard errors across function location on the upper limb movement (UL), lower limb movement (LL), and basic mobility (MO) subscales between the 27-item Stroke Rehabilitation Assessment of Movement (STREAM-27) and 15-item STREAM (STREAM-15) at admission and discharge. The standard error for each participant's Rasch score was estimated by Rasch analysis. The standard errors showing U-shaped distribution, as shown in the other Rasch-calibrated measures,37–39 indicate that the measurement errors of 2 extreme levels of function were larger. The large measurement errors were due to the smaller number of items developed for both extremely difficult levels of function and led to less precise measurement.
Floor and Ceiling Effects
Table 5 shows that the STREAM-30 demonstrated both significant floor effects (21.0%–31.3%) and ceiling effects (19.0%–26.2%) in the UL subscale at both time points, a significant floor effect (20.5%) at admission, and a significant ceiling effect (19.5%) at discharge in the LL subscale and no significant floor or ceiling effects in the MO subscale. Neither a floor effect nor a ceiling effect was found in the 3 subscales of the STREAM-27 and those of the STREAM-15.
Floor and Ceiling Effects of the STREAM-30, STREAM-27, and STREAM-15 at Admission and Discharge (n=195)a
Discussion
The main purpose of this study was to investigate whether the ability of the STREAM-15 to detect patients' progress is similar to those of the STREAM-30 and STREAM-27 at both the group and individual levels. We found that the STREAM-15 showed sufficient and similar group-level responsiveness, both in internal and external indexes, to the STREAM-30 and STREAM-27; however, the STREAM-27 had greater individual-level internal and external responsiveness than the STREAM-15. These results demonstrate that the STREAM-30, STREAM-27, and STREAM-15 are equally able to identify changes in movement status and mobility when used in a group and that the group changes in the 3 STREAM measures can comparably reflect the corresponding changes of ADL function assessed with the BI. As for detection of an individual patient's movement and mobility recovery, the STREAM-27 can detect more patients with significant movement and mobility improvement than the STREAM-15. In addition, the STREAM-27 revealed a greater number of patients achieving both significant recovery in movement and mobility and clinically important improvement in ADL function. Accordingly, the STREAM-27 is recommended as an outcome measure to demonstrate treatment effects of movement and mobility for patients with stroke in both clinical and research settings.
The superior ability of the STREAM-27 over that of the STREAM-15 to detect change in movement and mobility in individuals was supported by 2 individual-level indexes. First, all of the SCs of the STREAM-27 were significantly higher than those of the STREAM-15. That is, the STREAM-27 could identify greater amounts of change in movement and mobility for individual patients compared with the STREAM-15. Second, the DP showed that the STREAM-27 detected significantly more patients with significant movement and mobility recovery and fewer patients with no change or deterioration of movement and mobility function compared with the STREAM-15. The reason for the better individual-level responsiveness of the STREAM-27 could be that the STREAM-27 yields smaller measurement errors (ie, SE values) of the scores for individual patients, as shown in the Figure. With a narrower distribution (smaller SE values) of estimated scores, the score differences between admission and discharge of each patient are more likely to be significant. These observations indicate that the STREAM-27 is a better outcome measure than the STREAM-15 at the individual level.
Floor and ceiling effects limit a measure's ability to detect changes in individuals who score the minimum and maximum scores, respectively,36 thereby decreasing both the group-level and the individual-level responsiveness of a measure. Our results imply that both the STREAM-27 and STREAM-15 can differentiate among individuals with low or high levels of movement and mobility function, but the STREAM-30 cannot. Such an observation results from the fact that more measurement information from the related subscales was taken into account using multidimensional Rasch scaling model in the STREAM-27 and STREAM-15. That is, a patient's Rasch score on one subscale (eg, UL) was estimated by considering the same patient's responses on the other 2 subscales (ie, LL and MO). Thus, the STREAM-27 and STREAM-15 were more likely than the STREAM-30 to reveal the fine distinctions among patients with very low or high levels of function, and the floor and ceiling effects of the STREAM-27 were the smallest. These findings may be one of the reasons why the STREAM-27 and STREAM-15 had similar but slightly higher group-level responsiveness than the STREAM-30 and why the STREAM-27 had the best individual-level responsiveness.
Previously, the STREAM-15 was recommended as an efficient and reliable substitute for the STREAM-30 because its group-level psychometric properties were shown to be as robust as those of the STREAM-30 and STREAM-27.10,21,23 However, we found that the STREAM-15 is less able than the STREAM-27 to demonstrate treatment effects for individual patients in clinical trials or clinical settings. Our findings were consistent with those of previous studies comparing both the group-level and individual-level responsiveness of long-form and short-form outcome measures.6–8,13 Long-form measures are expected to be more responsive than any of their short-form measures because more items with different difficulty levels would provide more useful information to estimate patients' abilities. Thus, the measurement errors (SE) would decrease, which would yield better individual-level responsiveness. These results imply that using only the group-level responsiveness is insufficient, and the misguided belief in its sufficiency might have led clinicians and researchers to choose a less responsive outcome measure. Furthermore, the individual-level responsiveness of outcome measures can provide individual-level effects (eg, individual patient's progress, number of patients with significant improvement) that the traditional indexes of group-level responsiveness cannot provide. This information can help researchers to interpret the treatment effects and assist clinicians in translating the evidence from research to clinical settings (eg, how many patients could benefit from the treatment protocol). Thus, we strongly suggest that individual-level responsiveness needs to be included in the examination of responsiveness of outcome measures in order to provide comprehensive empirical evidence.
This study had 3 limitations. First, only 3% to 6% of the participants, as expected, deteriorated during hospitalization. Therefore, the abilities of the STREAM-27 and STREAM-15 to detect deterioration of movement and mobility in patients with stroke remain unknown. Second, the participants were in the subacute stage. Consequently, the individual-level responsiveness of the STREAM-27 and STREAM-15 may not be generalized to the patients with acute or chronic stroke. Last, the high attrition rate (49.7%) might limit the generalizability of our findings. However, the limitation of the generalizability might be trivial for the following 2 reasons. First, we found that the differences in movement and mobility impairments between the remaining 195 patients and the 193 participants lost at second assessment were not significant. In addition, the 195 participants covered a wide range of movement and mobility deficits.
In summary, our findings demonstrate that the STREAM-27 is better able than the STREAM-15 to detect movement and mobility improvement for an individual patient, although the STREAM-15 and the STREAM-27 are equally able to detect change in a group. Thus, the STREAM-27 is recommended as an outcome measure to demonstrate treatment effects of movement and mobility for patients with stroke in both clinical and research settings.
Appendix.
Deduction of SEdifference Formulaa
a The abbreviation “SE” here means measurement error, not standard error of the mean. The SE represents the measurement error of an individual patient's Rasch score, and the SEdifference means the measurement error of an individual patient's Rasch change score between repeated assessments.
Footnotes
All authors provided concept/idea/research design. Ms Huang and Dr Hsieh provided writing. Dr Hsieh provided data collection, project management, fund procurement, and administrative support. Mr Chou provided data analysis. Mr Hou provided study participants. Dr Chen and Ms Hsueh provided consultation (including review of manuscript before submission).
This study was approved by the Institutional Review Board of National Taiwan University Hospital.
This study was supported by research grants from the E-Da Hospital (EDAHT102015 and EDAHT103027).
- Received August 1, 2014.
- Accepted February 17, 2015.
- © 2015 American Physical Therapy Association