Abstract
Background The Community Balance and Mobility Scale (CB&M) is increasingly used to evaluate walking balance following stroke.
Objective This study applied Rasch analysis to evaluate and refine the CB&M for use in ambulatory community-dwelling adults following stroke.
Methods The CB&M content was linked to task demands and motor skill classifications. Rasch analysis was used to evaluate internal construct validity (structural validity) and refine the CB&M for use with ambulatory community-dwelling adults following stroke. The CB&M data were collected at 3 time points: at discharge from inpatient rehabilitation and at 6 and 12 months postdischarge (N=238). Rasch analysis evaluated scale dimensionality, item and person fit, item response bias, scoring hierarchy, and targeting. Disordered scoring hierarchy was resolved by collapsing scoring categories. Highly correlated and “misfitting” items were removed. Sensitivity to change was evaluated with standardized response means (SRMs) and one-way repeated-measures analysis of variance.
Results The CB&M was primarily linked to closed body transport task demands. Significant item-trait interaction, disordered scoring hierarchies, and multidimensionality were found. Scoring categories were collapsed in 15/19 items, and 5 misfitting items were removed. The resulting stroke-specific 14-item unidimensional CB&M (CB&MStroke) fit Rasch model expectations, with no item response bias, acceptable targeting (13% floor effects and 0% ceiling effects), and moderate-to-strong sensitivity to change at 6 months postdischarge (SRM=0.63; 95% confidence interval=−1.523, −0.142) and 12 months postdischarge (SRM=0.73; 95% confidence interval=−2.318, −0.760).
Limitations Findings are limited to a modest-sized sample of individuals with mild-to-moderate balance impairment following stroke.
Conclusions The CB&MStroke shows promise as a clinical scale for measuring change in walking balance in ambulatory community-dwelling adults poststroke. Future studies are recommended in a larger sample to validate and further refine the scale for use in this clinical population.
Following stroke, regaining dynamic balance and mobility, or community level “walking balance,” is essential to independence in daily living activities, community integration, and overall quality of life.1–3 Walking balance assessment incorporates a continuum of walking tasks to evaluate dynamic control of the center of mass over a changing base of support—from rudimentary walking activities, such as stepping, to more challenging walking activities, such as turning and stair negotiation—while integrating concurrent demands, such as carrying a bag or looking side to side while walking.2,4,5 The use of clinical measures with strong psychometric properties specific to a clinical stroke population is essential to evaluate changes in walking balance performance over time and in response to rehabilitation efforts.
The Community Balance and Mobility Scale (CB&M) is an ordinal scale developed to evaluate balance and mobility in ambulatory adults following traumatic brain injury.6 The test items are intended to reflect underlying balance and mobility skills that neurorehabilitation physical therapists and occupational therapists deemed important for safely returning to walking in the community.6 The CB&M has since been used to measure community-level walking balance abilities in people poststroke, where it was found to have excellent convergent validity and demonstrated greater sensitivity to change in balance performance than the Berg Balance Scale and the Timed “Up & Go” Test.7 Given these findings, further analysis was warranted to understand the psychometric properties, targeting, and responsiveness of the CB&M when applied to evaluate walking balance in ambulatory community-dwelling adults following stroke.
Rasch analysis has previously been applied to evaluate the internal construct validity (also termed “structural validity”) of ordinal clinical outcome scales based on fit to the Rasch measurement model assumptions.8–11 To achieve fit to the Rasch model, a primary assumption that must be satisfied is the scale must principally measure a single construct (in this case, walking balance). If the scale satisfies this requirement, the estimates of participant walking balance ability and difficulty of test items can be calibrated on the same interval scale: a log odds ratio scale in logits.12 The hierarchy of test item scoring structures is examined, as Rasch analysis assumes the easier the test item, the more likely it will be performed by participants with greater ability.13 Additionally, item response bias (differential item functioning [DIF]) based on participant attributes such as age or sex and the targeting of the scale item difficulty to the abilities of the clinical population of interest are examined.14 The fit of participants and items can be investigated for “misfit” to the Rasch model. Misfitting items demonstrating low levels of discrimination between different categories of participant ability can be removed to improve fit to the Rasch model, refining the scale content and reducing the overall burden of assessment.15,16 For clinical measures, such as the CB&M, it is important that removal of test items is informed not only by the interpretation of Rasch analysis results but also by the intended conceptual framework of the scale.8,17 Furthermore, the transformation of raw ordinal scale scores of patients to a linear log odds ratio (in logits) permits the calculation of change scores for between-group comparisons of patients.
The primary purpose of this study was to use Rasch analysis to evaluate the internal construct validity of the CB&M and to refine the scale for use in the evaluation of walking balance in ambulatory community-dwelling adults following stroke. The secondary aim was to examine the responsiveness of the resulting stroke-specific scale (ie, CB&MStroke) to change in balance performance over time in a cohort of participants evaluated at 3 time points in their first year following discharge from inpatient stroke rehabilitation.
Method
Participants
A secondary analysis was undertaken on CB&M data from 2 clinical trials (ClinicalTrials.gov NCT00400712 and NCT01573585). The participants were ambulatory English-speaking community-dwelling adults who had been discharged home from 1 of 4 inpatient stroke rehabilitation programs in Ontario and British Columbia, Canada, after they had sustained their first major unilateral stroke from 2007 to 2014. Participants with serious comorbidities (eg, mobility-limiting arthritis) or who were unable to follow simple verbal instructions were excluded. Data were collected at 3 time points: within 1 month of discharge from inpatient stroke rehabilitation (discharge, n=100), 6 months postdischarge (6 months, n=71), and 12 months postdischarge (12 months, n=67), providing a total of 238 sets of CB&M scores (Tab. 1). Test items were recategorized based on whether the paretic or nonparetic limbs were evaluated, as this was thought to be of greater relevance than whether the task was performed with the right or left limb.
Participant Characteristicsa
Outcome Measure
The CB&M is a 19-item ordinal scale scored out of 96 points, with higher scores representing better walking balance performance.6 Each test item is scored from 0 to 5 (with the exception of item 12: descending stairs, rated from 0 to 6) to reflect a hierarchy of task difficulty based on established criteria such as time, distance, and quality of performance.
Data Analysis
To establish the initial conceptual framework of the CB&M, 2 physical therapists (K.J.M. and C.L.P.) independently classified the content of the CB&M test items, with a third investigator (S.J.G.) available to adjudicate discrepancies based on the motor skill classification (modified based on taxonomy of motor skill tasks proposed by Gentile18). Each item was further described based on the task demands and key scoring criteria.
Rasch analysis (RUMM2030 version 5.1e 2010, RUMM Laboratory Pty Ltd, Duncraig, Western Australia, Australia) was then used to evaluate the internal construct validity and targeting of the CB&M, refining the scale for use with ambulatory community-dwelling survivors of stroke. Good overall model fit was indicated by a nonsignificant chi-square test (Bonferroni-adjusted P≥.003) and mean standardized fit residual statistics approximating 0.0 (SD=1.0).19 Misfitting items and participants and scale dimensionality issues can lead to misfit of the overall model to Rasch assumptions; therefore, a stepwise approach to the evaluation and refinement of the CB&M was undertaken in accordance with recommendations made by Pallant and Tennant19 and Hagquist et al.20
Individual item fit.
First, scoring hierarchy was examined. A threshold represents a point between 2 adjacent scoring categories, where there is a 50–50 likelihood that either score will be assigned. Disordered thresholds can occur when thresholds for scoring are inconsistent (eg, participants scoring 2 are not consistently more capable in their walking balance abilities than participants scoring 1 on a test item). Disordered thresholds were resolved by collapsing scoring (eg, collapsing scores 1 and 2 together), creating fewer scoring categories within the item, and analyses were rerun. When scoring categories were collapsed, we reworded the descriptors for the new scoring categories but retained the scoring criteria from the original CB&M (eg, time, distance, and movement quality indicators; see eAppendix 1).
Second, individual item fit was then examined using the same criteria as for overall model fit. Test items with significant chi-square values (P<.003) and mean standardized fit residual statistics >2.5 displayed low levels of discrimination between differing ability categories. These “misfitting,” and potentially redundant, items were sequentially removed only if: (1) removal of these items improved fit to the overall model15 and (2) removal of the items did not substantially change the intended conceptual clinical framework of the scale.8,17 The latter judgments were based on our determination of the task demands and motor skill classifications of the CB&M items in the context of a physical therapy walking balance assessment.2,4,18
Third, DIF indicative of response bias based on sex, age (≤64/≥65 years), and paretic side (dominant/nondominant) was evaluated using analysis of variance (ANOVA) with the significance level set at 5% and applying a Bonferroni correction. Differential item functioning also was evaluated over the 3 time frames (discharge, 6 months, and 12 months) to assess the invariance of scores over repeated assessments of participants. Invariance over repeated assessments is an important attribute of the scale when measuring change in performance over time in these community-dwelling survivors of stroke.21
Participant fit.
Misfit to the Rasch model based on participants' scoring in an unusual manner was identified based on criteria of chi-square (P<.003) and standardized fit residuals of >2.5.
Scale dimensionality.
Response dependency was evaluated by inspecting the residual correlation matrix for pairs of items with r>.3, indicating scoring for items might be “dependent” or closely related to each other.22 “Testlets” were run with the item pairs, and the resulting person separation index (PSI) values were compared with original values for the full scale to evaluate whether these item pairs inflated scale internal consistency. Using a similar approach to that used in the handling of misfitting items, decisions to remove highly correlated items that inflated PSI values were based on overall model fit and informed by the conceptual clinical framework of the scale.
Finally, principal component analyses were undertaken to identify subsets of items with positive and negative loadings on the first unrotated component. We then applied t tests to compare Rasch-derived person estimates from these subsets. Scale unidimensionality (measurement of a single construct) was confirmed if t-test comparisons were significant (P<.05).
Targeting of the scale.
Targeting of the scale to the walking balance abilities of the community-dwelling participants with stroke was evaluated based on mean participant ability in relation to overall test item difficulty of the scale. Floor and ceiling effects over 15% were considered indicative of an inappropriate measurement model.23 Furthermore, PSI values were established to estimate the capacity of the CB&M to distinguish or stratify groups of participants with different walking balance abilities. For group use, a value ≥0.70 was considered acceptable, and a value of ≥0.85 was taken to be indicative of suitability for individual use.24
Scale responsiveness.
The sensitivity to change of the resulting CB&MStroke was estimated using 2 approaches using the data from 62 participants assessed at all 3 time points. First, the standardized response mean (SRM) was calculated by dividing the mean of change scores (in logits) by the standard deviation of change scores from discharge to 6 months and from discharge to 12 months.25 These data were interpreted using Cohen's criteria for effect size, with 0.2 to 0.49 considered small, 0.5 to 0.8 moderate, and >0.8 indicative of a large effect size.26 Second, scores at discharge, 6 months, and 12 months were compared using a one-way repeated-measures ANOVA with post hoc pair-wise comparisons and the method of least significant difference. Effect sizes derived from partial eta-square statistics were interpreted using guidelines proposed by Cohen (0.1=small, 0.6=moderate, 1.4=large).26
Role of the Funding Source
The data used for this secondary analysis were obtained from studies supported by a Heart and Stroke Foundation of BC and Yukon grant awarded to Dr Garland (grant number 000227) and a Heart and Stroke Foundation of Ontario grant awarded to Dr Brouwer and Dr Garland (SRA 5974). Dr Miller was supported by a Heart and Stroke Foundation Junior Personnel Research Fellowship, and Dr Pollock was supported by a Canadian Institutes of Health Doctoral Fellowship. Dr Garland and Dr Brouwer were the principal investigators for grant funding of the respective clinical trials providing the data used in this secondary analysis.
Results
Classification of the Content of the CB&M
There was 100% agreement between the 2 physical therapist reviewers in linking the 19 items of the CB&M to task demands and motor skills classifications (Tab. 2). Seventeen of the 19 items were classified as body transport tasks involving a range of dynamic unipedal or bipedal balance task demands, such changes in base of support, direction, or speed. Some test items, such as 11P (walking, carrying, and looking to the paretic side) and 11NP (walking, carrying, and looking to the nonparetic side), included concurrent task demands involving manipulation and diversion of visual gaze. The remaining 2 items, 1P (unilateral stance on the paretic limb) and 1NP (unilateral stance on the nonparetic limb), were classified as body stability tasks requiring static unipedal balance.18
Classification of Content of Community Balance and Mobility Scale Itemsa
Overall Model, Individual Item, and Participant Fit
The 19-item CB&M did not fit the Rasch model expectations (Tab. 3, analysis 1), as the scale demonstrated significant item-trait interaction (P<.0001) and multidimensionality (15.5% of t-test comparisons were significant). The PSI of 0.86 was sufficiently high for individual use, and mean standardized fit residuals for participant data fit the model expectations. However, deviation of test item data from the model expectations was evident, as the mean standardized fit residual statistics did not approximate 0.0 (SD=1.0). On inspection, disordered thresholds were found for 15/19 test items.
Model Fit Statistics for Original and Revised Versions of the CB&Ma
Test items with disordered thresholds were sequentially rescored. The resulting scoring structure and individual item fit statistics are presented in Table 4. The rescored 19-item CB&M was reanalyzed (Tab. 3, analysis 2), and significant item-trait interaction (P<.0001) and multidimensionality (14.1% of t-test comparisons were significant) remained. The PSI value improved, and the mean standardized fit residuals indicated that participant data fit the model expectations; however, the test item data (item fit residual statistics) again deviated from model expectations.
Initial Tests of Fit and Item Difficulty for the Community Balance and Mobility Scalea
Overfitting of test items 1P (unilateral stance on the paretic limb) and 1NP (unilateral stance on the nonparetic limb) was evidenced by large mean standardized fit residual statistics (5.749 and 6.609, respectively) and significant chi-square statistics (P<.0005, Tab. 4). Test items 1P and 1NP were removed improving the model fit and PSI, but not dimensionality of the scale (Tab. 3, analysis 3).
Scale Dimensionality
Potential artificial inflation of internal consistency was noted between items 11P (walking, carrying, and looking to paretic side) and 11NP (walking, carrying, and looking to nonparetic side). The residual correlation for these 2 items was r=.536, with inflation of the PSI from 0.88 to 0.97. These test items were similar in walking balance requirements to items 8P (walking and looking to paretic side) and 8NP (walking and looking to nonparetic side), with the additional demand of carrying shopping bags. Participants with stroke were often observed to have difficulty with carrying grocery bags if they had poor motor recovery or shoulder pain in their hemiparetic upper limb. The item needed to be adapted to accommodate these individuals, leaving the nonparetic upper limb carrying a greater weight and potentially threatening the test item validity. The level of residual correlation suggests that the addition of carrying loads while walking did not represent an extension of task, which further challenged walking balance during the walk and look task and, therefore, is unlikely to further stratify participants' abilities. For these reasons, items 11P and 11NP were removed from the scale, and analysis of fit was rerun (Tab. 3, analysis 4), finding improvement in the model fit but not in the dimensionality of the scale, as 17.2% of t-test comparisons were significant (P<.05). A strong residual correlation (r=.929) was observed between items 4NP (lateral foot scooting on nonparetic limb) and 5NP (hopping forward on nonparetic limb), with modest inflation of the PSI values from 0.91 to 0.96. Both items demanded dynamic unilateral nonparetic limb support. Item 5NP was retained to evaluate activity limitations associated with jumping and hopping. Removal of item 4NP maintained good overall model fit and improved dimensionality of the scale, with an identical PSI value of 0.97 (Tab. 3, analysis 5).
Two additional item pairs had residual correlations of greater than .30; however, these items were not removed, as artificial inflation of internal consistency was not evident. A small rise in PSI value from 0.96 to 0.97 was associated with a residual correlation of r=.575 between items 8P (walking and looking to paretic side) and 8NP (walking and looking to nonparetic side), whereas no difference in PSI values between subtests and the original scale was detected despite the strong residual correlation (r=.730) between items 13P (step-ups × one step with the paretic foot leading) and 13 NP (step-ups × one step with the nonparetic foot leading). Hence, analysis 5 presented the final solution, the CB&MStroke, a 14-item scale (/46) that fits with Rasch assumptions tailored to evaluate the walking balance abilities of ambulatory community-dwelling people following stroke (Tab. 5, eAppendix 1). Review of the standardized fit residuals and chi-square statistics revealed no misfit of items or participants in the CB&MStroke to the model. In addition, all ANOVA statistics were nonsignificant (Bonferroni-adjusted P=.0012) for all items, indicating no evidence of response bias based on sex, age (≤64/≥65 years), paretic side (dominant/nondominant), or when the scale was used for repeated assessments of participants over time (discharge, 6 months, 12 months).
Stroke-Specific Community Balance and Mobility Scale (CB&MStroke) Ordered by Item Difficultya
Scale Responsiveness
In the Figure, the targeting of the original CB&M (graph A) and the CB&MStroke (graph B) to the walking balance abilities of the community-dwelling adults poststroke is presented. The mean participant ability estimate was −1.039 (SD=3.046) for the original CB&M and −1.910 (SD=5.4000) in the CB&MStroke. The observed floor effects were 13%, and the ceiling effects were 0%, fitting within the established criteria (<15% floor and ceiling effects) for an “appropriate” measurement model.23 Item thresholds in the CB&MStroke (bottom half of graph B in the Figure) ranged from −13.20 to 6.80. Similar to the original scale, relatively large gaps can be observed in the estimated item threshold difficulties in the lower half of the CB&MStroke scale. The PSI of 0.97 suggests that the CB&MStroke had an excellent capacity to distinguish/stratify individuals with different walking balance abilities.13
Targeting (A) the original Community Balance and Mobility Scale (CB&M) and (B) the stroke-specific 14-item Community Balance and Mobility Scale (CB&MStroke) to walking balance abilities of community-dwelling adults poststroke. These person-item distribution graphs represent the number of participants at each ability level (top half of graph) and number of item thresholds (bottom half of graph) at each difficulty level on the same scale for the original CB&M (A) and the CB&MStroke (B). Negative logit values indicate easier item categories and participants with poorer balance ability. Clustering of participants unable to perform any test items is evident on the far left side of both graphs, representing the floor effects of each of the 2 scales. In addition, large gaps in the spread of test item difficulty in the lower half of the measurement scale continuum are evident in both scales.
As the CB&MStroke adhered to Rasch model assumptions, logit values corresponding to the raw CB&MStroke scores obtained by participants were determined using RUMM2030. Conversion of the ordinal scores to interval scores permitted the calculation of change scores and the estimation of the sensitivity to change for the 62 participants assessed at all 3 assessment time frames. The CB&MStroke demonstrated moderate responsiveness between the discharge and 6-month assessments (SRM=0.63) and between the discharge and 12-month assessments (SRM=0.73). The one-way repeated-measures ANOVA comparison of CB&MStroke scores converted to logits revealed large, significant effects for assessment time (Wilks lambda=0.79; F2,62=8.173, P=.001; partial eta-square=0.206). Significant changes in CB&MStroke scores were observed from discharge to 6 months (95% confidence interval [95% CI]=−1.523, −0.142; P=.019), from 6 months to 12 months (95% CI=−1.209, −0.204; P=.007), and from discharge to 12 months (95% CI=−2.318, −0.760; P<.0005). A nomogram table is provided in eAppendix 2 for conversion of total CB&MStroke ordinal scores to transformed values (in logits). This conversion provides a strategy to overcome limitations associated with the comparing changes in raw ordinal scores where a change in total score from 24 to 25 is not equivalent to a change from 20 to 21.
Discussion
This study is unique in applying Rasch analysis to evaluate the internal construct validity of the CB&M and to refine the scale for use in the evaluation of walking balance in ambulatory community-dwelling survivors of stroke. The original CB&M test items were linked to dynamic unipedal or bipedal balance task demands, with most items classified on Gentile's taxonomy of motor skill tasks as closed body transport activities.18 The internal construct validity of the original CB&M did not meet Rasch model assumptions when it was used to measure walking balance in people poststroke. Significant item-trait interaction, disordered scoring hierarchies, and multidimensionality were found; therefore, scoring categories were collapsed in 15/19 items, and 5 misfitting items were removed. The resulting 14-item CB&MStroke (/46) adhered to Rasch model assumptions (eAppendix 1). The CB&MStroke demonstrated no ceiling and acceptable floor effects in the community-dwelling survivors of stroke. It showed an excellent capacity to stratify individuals based on their balance recovery, and it had moderate-to-strong sensitivity to change in their walking balance performance over the first year following discharge from inpatient stroke rehabilitation. With further validation in a larger sample, the CB&MStroke shows promise as a clinical scale for measuring change in walking balance in ambulatory community-dwelling adults poststroke.
In this study, the internal construct validity (also termed “structural validity”) of the scale was evaluated based on the fit of data for ambulatory community-dwelling survivors of stroke to the Rasch measurement model requirements.27,28 To fit to the Rasch model, CB&M test items were rescored and removed, while balancing the need to provide clinically meaningful information about walking balance performance. In addition to changing item scoring, we removed 5 test items because they were misfitting or highly correlated; therefore, they potentially inflated the internal consistency of the scale. Rasch analysis provided an objective means of identifying “overfitting,” potentially redundant test items that did not substantively contribute to differentiating walking balance abilities in the community-dwelling individuals following stroke.11,15,16 Removal of these 5 items resulted in an improvement in overall model fit to Rasch assumptions, scale dimensionality, and PSI in the sample of ambulatory community-dwelling survivors of stroke. The removed items were 1P (unilateral stance on the paretic limb), 1NP (unilateral stance on the nonparetic limb), 11P (walking, carrying, and looking to paretic side) and 11NP (walking, carrying, and looking to nonparetic side), and 4NP (lateral foot scooting on nonparetic limb). Items 1P and 1NP were the only items in the original CB&M with a body stability motor skill classification.18 These test items evaluated static unipedal balance, potentially encompassed in the dynamic unipedal balance body transport activities, such as items 4P (lateral foot scooting on paretic limb), 5P (hopping forward on paretic limb), 5NP (hopping forward on nonparetic limb), 13P (step-ups × 1 step with paretic foot leading), and 13NP (step-ups × 1 step with nonparetic foot leading). Items 11P (walking, carrying, and looking to paretic side) and 11NP (walking, carrying, and looking to nonparetic side) were arguably similar in task demand to items 8P (walking and looking to paretic side) and 8NP (walking and looking to nonparetic side) in assessing dynamic bipedal balance while diverting visual gaze to a target. It is acknowledged that carrying items while walking is a potentially important activity in a community setting; therefore, adaptation and validation of this item to one-arm carrying should be considered.
Relatively high correlations (>r=.30) were found between pairs of test items that assessed the paretic and nonparetic sides of the body (eg, items 8P [walking and looking to paretic side] and 8NP [walking and looking to nonparetic side], items 13P [step-ups × 1 step with paretic foot leading] and 13NP [step-ups × 1 step with nonparetic foot leading]); however, these items were associated with little, if any, artificial inflation of the reliability estimates, supporting their inclusion in the scale. The retention of bilateral assessment is consistent with clinical practice in rehabilitation poststroke, providing a comprehensive understanding of patient abilities and addressing the need for clinically meaningful information.
Following the reduction in test items and the streamlining of scoring categories in the CB&MStroke, the scale retained moderate-to-strong sensitivity to change in performance over time in community-dwelling adults following stroke. An established strength of the original CB&M and the revised CB&MStroke presented here is the absence of a ceiling effect compared with other commonly used clinical measures, such as the Berg Balance Scale, in this clinical population.2,7,29 The spread of item difficulty in the upper half of the CB&MStroke evident in part B of the Figure enables greater precision and sensitivity to change in higher-level community walking balance performance.
The participant data were “stacked” according to previously published guidelines and evaluated for response bias using DIF analysis.21 The lack of response bias found when the CB&MStroke was used for repeated assessments of participants at discharge, 6 months, and 12 months is an important attribute if the scale is to be used by clinicians and researchers as an outcome measure to monitor change in balance performance over time. Although the repeated measures violate the assumption of independence observations in Rasch analysis, in this case, the benefits of establishing the invariance of the scale over time to support its repeated use in clinical practice as an outcome measure were balanced against the relatively minor impact of this violation on person dependency estimates.21
Although the CB&MStroke shows promise, there are limitations that must be considered prior to using this clinical scale for measuring change in walking balance in ambulatory community-dwelling adults poststroke. In this study, we found floor effects that were acceptable but still considerable and relatively larger than the floor effects reported when a different sample of community-dwelling survivors of stroke was evaluated using the original CB&M.23 This difference in floor effect may be attributed to differences in the inclusion criteria for participants in the 2 studies. Nonetheless, a large gap was observed in the spread of test item difficulty in the lower half of the measurement scale continuum of both the original CB&M and the revised CB&MStroke (Figure). This gap potentially diminishes the sensitivity of both scales in individuals with poorer walking balance. Finally, the findings of this study are based on a relatively modest-sized sample of community-dwelling survivors of stroke with mild-to-moderate balance impairment. Rasch analysis on a sample of 250 or more individuals would be necessary to draw robust conclusions about the psychometric properties of the CB&MStroke.30
In conclusion, using Rasch analysis, we have refined the CB&M and developed a promising stroke-specific CB&M (CB&MStroke) that demonstrates sound internal construct validity and moderate-to-strong sensitivity to change in the walking balance abilities of ambulatory community-dwelling survivors of stroke. Further research is recommended in a larger sample to validate and further refine the scale for use in this clinical population. These studies should investigate the addition of items in the lower half of the measurement continuum to enhance sensitivity to change in individuals with lower-level balance abilities. In addition, the reliability, relative responsiveness, minimal clinically important difference, and clinical utility should be established to inform use of the CB&MStroke by clinicians and researchers.
Footnotes
Dr Miller planned and executed the analyses. All authors contributed to interpreting the results, writing, and approval of the final manuscript. The authors acknowledge the support of Dr J. Pallant in providing Rasch analysis methodology materials and advice.
The preliminary findings from this research were presented at the Canadian Stroke Congress; October 4–7, 2014; Vancouver, British Columbia, Canada.
The data used for this secondary analysis were obtained from studies supported by a Heart and Stroke Foundation of BC and Yukon grant awarded to Dr Garland (grant number 000227) and a Heart and Stroke Foundation of Ontario grant awarded to Dr Brouwer and Dr Garland (SRA 5974). Dr Miller was supported by a Heart and Stroke Foundation Junior Personnel Research Fellowship, and Dr Pollock was supported by a Canadian Institutes of Health Doctoral Fellowship. Dr Garland and Dr Brouwer are the principal investigators for grant funding of the respective clinical trials providing the data used in this secondary analysis.
This study was a secondary analysis of data from 2 clinical trials. Ethics approval for ClinicalTrials.gov NCT00400712 was obtained from the human research ethics boards at Western University and Queen's University. The Human Research Ethics Board at the University of British Columbia approved both ClinicalTrials.gov NCT00400712 and ClinicalTrials.gov NCT01573585.
- Received July 30, 2015.
- Accepted March 28, 2016.
- © 2016 American Physical Therapy Association