Abstract
Background In the original and modified Dynamic Gait Index (mDGI), 8 tasks are used to measure mobility; however, disagreement exists regarding whether all tasks are necessary. The relationship between mDGI scores and Centers for Medicare & Medicaid Services (CMS) severity indicators in the mobility domain has not been explored.
Objective The study objectives were to examine the relationship between medical diagnoses and mDGI scores, to determine whether administration of the mDGI can be shortened on the basis of expected diagnostic patterns of performance, and to create a model in which mDGI scores are mapped to CMS severity modifiers.
Design This was a cross-sectional, descriptive study.
Methods The 794 participants included 140 people without impairments (control cohort) and 239 people with stroke, 140 with vestibular dysfunction, 100 with traumatic brain injury, 91 with gait abnormality, and 84 with Parkinson disease. Scores on the mDGI (total, performance facet, and task) for the control cohort were compared with those for the 5 diagnostic groups by use of an analysis of variance. For mapping mDGI scores to 7 CMS impairment categories, an underlying Rasch scale was used to convert raw scores to an interval scale.
Results There was a main effect of mDGI total, time, and gait pattern scores for the groups. Task-specific score patterns based on medical diagnosis were found, but the range of performance within each group was large. A framework for mapping mDGI total, performance facet, and task scores to 7 CMS impairment categories on the basis of Rasch analysis was created.
Limitations Limitations included uneven sample sizes in the 6 groups.
Conclusions Results supported retaining all 8 tasks for the assessment of mobility function in older people and people with neurologic conditions. Mapping mDGI scores to CMS severity indicators should assist clinicians in interpreting mobility performance, including changes in function over time.
The Dynamic Gait Index (DGI) is an outcome measure used to characterize mobility performance, specifically the ability to adapt gait to complex walking tasks associated with walking in community environments.1 A modified version of the DGI (mDGI), which retains the original 8 tasks but has a modified scoring system, was recently published.2,3
Prior research on the mDGI focused on evidence for the reliability and validity of mDGI scores in a large and diverse population of patients with mobility limitations.2 In addition, Matsuda et al3 provided evidence to support the invariance of internal structure mDGI scores across 5 diagnostic groups: stroke, vestibular dysfunction, gait abnormality, traumatic brain injury (TBI), and Parkinson disease (PD). Both of these studies recommended retaining the original 8 tasks of the DGI. In contrast, the results of other studies examining differences in task-specific performance on the original DGI in patients with specific neurologic conditions suggested modifying the DGI to eliminate tasks considered either redundant or too easy for particular populations.4,5 For example, DGI performance in patients with vestibular pathology was examined, and vertical and horizontal head turns were identified as the most difficult tasks in this population. On the basis of these results, Marchetti and Whitney4 suggested a 4-item version of the DGI, including gait on level surfaces, changes in gait speed, and horizontal and vertical head turns. However, the degree to which a shortened version of the DGI would effectively evaluate mobility function in all patients with a specific diagnosis, such as vestibular pathology, is uncertain. Therefore, the primary purpose of this study was to examine the relationship between medical diagnoses and task-specific patterns of performance on the mDGI.
Although we expected to find, on average, diagnosis-specific differences in performance on individual mDGI tasks, we expected that, because of the large heterogeneity within each diagnostic group, performance patterns based on group data could not be used to predict the pattern of mobility performance for any particular individual within a given diagnostic group. Such a finding would support the importance of retaining all 8 tasks of the mDGI in an evaluation of mobility performance.
We used previous research to formulate hypotheses regarding expected patterns of task performance specific to 5 medical diagnoses: stroke, vestibular dysfunction, gait abnormality, TBI, and PD. Expected performance on the mDGI in people with stroke was based on research by Robinson et al,6 who examined performance on 10 walking tasks in 30 participants with stroke (stroke group) and 30 participants without stroke (control group). Participants in the stroke group performed significantly worse (P<.001) than participants in the control group on the “usual pace” walking task and on all 9 other complex walking tasks. On the basis of this research, we expected significant differences between the control group and the stroke group on all 8 mDGI tasks. The InCHIANTI aging study7 reported that walking speed on 10 complex walking tasks was significantly lower in people aged 65 years and older than in those younger than 65 years. On the basis of this research, we expected that, compared with participants without a gait abnormality, participants with a gait abnormality (defined as patients who were referred for physical therapy under ICD-9 [International Classification of Diseases, 9th revision] code 781.2 [gait abnormality] and who did not have a neurologic diagnosis) would perform significantly worse on all 8 mDGI tasks. Previous research suggested that people with vestibular problems would perform significantly worse on both horizontal and vertical head turns than on other DGI tasks4 and that the frequency of freezing and other gait impairments would increase when people with PD changed the direction of walking, made turns, or were required to make changes in gait speed.8–10 On the basis of this research, we expected that participants with PD would have the most difficulty with the “around obstacles,” “pivot turns,” and “change pace” tasks and less difficulty with the remaining tasks. Finally, on the basis of research by Kleffelgaard et al,11 we expected that participants with TBI would have significantly lower scores on the “usual pace,” “change pace,” “horizontal head turn,” “vertical head turn,” and “pivot turn” tasks than participants without TBI.
In addition to determining whether all 8 tasks are needed for different populations of patients, it is important to consider how scores for mDGI tasks can be interpreted in terms of mobility function. In 2013, the Centers for Medicare & Medicaid Services (CMS) launched a reporting system to improve the documentation of various aspects of function, including mobility. The function codes are referred to as G codes, and the specific codes related to mobility and walking are G8978 through G8980. The goals of the new reporting system are to better understand function in CMS beneficiaries and the conditions, outcomes, and costs related to changes in function and to develop an “improved payment system.”12 In addition to the function G codes, the CMS has required the use of severity modifiers to characterize performance or “impairment limitation restriction” within all functional domains. The modifier scale varies from 0% impaired (CH) to 100% impaired (CN).12 A framework for mapping outcome measure scores from the mDGI (total, performance facet, and task scores) to CMS severity modifiers is needed. Therefore, another purpose of this study was to create a model in which mDGI scores are mapped to CMS severity modifiers (a new mDGI scoring system).
Method
An in-depth review of the methods used to investigate the reliability and validity of the mDGI was previously reported.2 A brief overview of the study methods is presented here.
Recruitment
An e-mail was sent via the American Physical Therapy Association's Section of Neurology listserve to recruit potential clinical sites. From within each of the volunteer sites, participants with neurologic impairments and currently receiving physical therapy for balance and mobility problems were evaluated with the DGI and the mDGI. The primary inclusion criterion was the ability to walk approximately 6 m (20 ft) without the physical assistance of another person. The use of an assistive device was permitted. As a control cohort, a convenience sample of adults was recruited from volunteers responding to a flyer posted in the Department of Rehabilitation Medicine at the University of Washington and in retirement communities in the greater Seattle-Bellevue area of Washington. Inclusion criteria were an age of between 15 and 99 years, no neurologic diagnosis, the ability to walk without the physical assistance of another person for a distance of approximately 6 m, and the ability to give informed consent. All participants provided informed consent before testing. The analyses reported here were done with a subset of data (794 participants) from a previous publication.2
Modification of the DGI
All 8 tasks from the original DGI were retained, with minor modifications made to distance ambulated (∼6 m [20 ft]) and 4 test items (change pace, over obstacles, pivot turn, and stairs).2 The original scoring system was modified to establish ordinal scores for 3 separate facets of walking performance: gait pattern (GP) (0–3), level of assistance (LOA) (0–2), and time; measures of time were converted to an ordinal scale (0–3).2 Performance scores at the task level were calculated by adding the scores for GP, LOA, and time, resulting in a score ranging from 0 to 8 for each of the 8 tasks. A total score for each of the 3 facets of performance was calculated to characterize walking performance with respect to GP (range=0–24), time (range=0–24), and LOA (range=0–16). The total score for the mDGI (sum of all 8 task scores) ranged from 0 to 64.
Profiling Patterns of Performance on the Basis of the Medical Diagnosis
To investigate group-specific differences in performance on the mDGI, we compared mDGI total, performance facet, and task scores for the control cohort and the 5 diagnostic groups using analyses of variance (ANOVAs). One of the assumptions of an ANOVA is that the compared groups will have homogeneous variances. Given the likely heterogeneity of the variances between the control cohort and the diagnostic groups, we used the Levene test for homogeneity of variance before conducting the ANOVAs. We performed 2 ANOVAs, the Spearman F test and the Welch robust F test, in which the score means are weighted by the reciprocal of the group's mean variances to address issues of heterogeneity of variance. Because of multiple comparisons, a Bonferroni correction was used, with statistical significance set at P≤.001.13 The Cohen d was used to evaluate the effect size of the differences between means. To investigate whether patterns of scores based on medical diagnosis could be generalized to individuals, we used box and whisker plots to visually demonstrate the between- and within-group distributions of mDGI scores.
Mapping mDGI Scores to CMS Severity Indicators
The total score for the mDGI was composed of ordinal scores for time, GP, and LOA for each task. If CMS severity indicators were set on the basis of ordinal scores, then a score difference of, for example, 4 points at the top and bottom of the scale would likely represent much greater change or difference than a score difference of 4 points in the middle of the scale. Therefore, it was important to establish an interval scale before setting cutoff scores for the 7 levels of severity established by the CMS.
To set cutoff scores for the CMS severity indicators, we used underlying Rasch scales, which converted mDGI total and performance facet scores to interval (θ) scales.14,15 Shumway-Cook et al2 presented information regarding the Rasch analysis used in the development of the new mDGI scoring system.
The Rasch model is an item response theory (IRT) model that is frequently used for converting ordinal (raw score) scales to interval scales. The Rasch model also is useful for investigating the technical qualities of items and tasks in terms of difficulty, ability to discriminate between people with lower scores and those with higher scores, and effectiveness of scoring rubrics (such as those used for the mDGI tasks). During a Rasch analysis, item scores are located on an underlying difficulty (θ) scale, and these locations are used to generate a scale in which raw scores are associated with interval scale scores. This approach makes it possible to identify raw cutoff points that represent equal intervals on the underlying difficulty scale.
For the mDGI, the Rasch “items” were mDGI task subscores (eg, time, LOA, and GP scores for the “usual pace” task). The purposes of the original Rasch analysis were to investigate the technical quality of mDGI task subscores and to determine whether the new scoring system had addressed ceiling effects found on the original DGI. The Rasch analysis for the mDGI total score included all 24 subscores from the mDGI tasks. The Rasch analyses for the 3 mDGI performance facet scores included the subscores relevant to each performance facet from each mDGI task (eg, time subscores from the 8 mDGI tasks were included in the Rasch analysis for the time performance facet scale). The task subscore locations for the Rasch scale underlying the mDGI total score were reported by Shumway et al.2
Setting CMS cutoff scores required dividing the scale into intervals. Using the Rasch scale, we found the distance between the θ value associated with a raw score of 0 and the θ value associated with a raw score of 64 on the total score scale. We then divided this range by 7 to identify a distance on the Rasch scale, in θ units, that would represent one-seventh of the scale. This value was applied to the θ locations beginning with the lowest raw score on the test. Thus, we created 7 equal intervals of performance from a total score of 0 to a total score of 64. We repeated the process using Rasch scaling for each of the performance facet scores. Given the number of points in a task and the number of severity levels, equal intervals between cutoff scores at the task level were not possible; therefore, we set the severity cutoff scores using raw scores.
Role of the Funding Source
This study was supported by a grant from the Walter C. and Anita C. Stolov Research Fund, Department of Rehabilitation Medicine, University of Washington.
Results
Participant Sociodemographics
The analyses reported here were done with a subset of data (794 participants) from a previous publication.2 The subset included the control cohort (140 people without neurological impairment) and people with the following diagnoses: stroke (n=239), vestibular dysfunction (n=140), TBI (including head injury and concussion) (n=100), gait abnormality (n=91), and PD (n=84). Table 1 shows the sociodemographics of the participants by group.
Sociodemographics of Participants by Group
Patterns of Performance as a Function of Medical Diagnosis
Table 2 shows the means, standard deviations, and 95% confidence intervals for the mDGI total scores and the 3 performance facet scores for the control cohort and the 5 diagnostic groups. Table 3 shows the means, standard deviations, and 95% confidence intervals for the mDGI task scores. Tables 2 and 3 also include significance values from the test of homogeneity of variance. Finally, effect sizes for each of the comparisons are also shown. Despite statistically significant differences between variances for some of the comparisons, the results for the traditional F test and the Welch robust F test (both shown in Tabs. 2 and 3) were nearly identical.
Modified Dynamic Gait Index (mDGI) Total and Facet Scores and Effect Sizes (Cohen d) for Control Cohort and Five Diagnostic Groupsa
Modified Dynamic Gait Index Task Scores and Effect Sizes (Cohen d) for Control Cohort and Five Diagnostic Groupsa
Total mDGI Scores
As shown in Table 2, the total mDGI scores for 3 of the 5 diagnostic groups (stroke, gait abnormality, and TBI) were significantly lower than those for the control cohort. However, there was large variability in scores within each of the participant groups, as shown in the Figure; a box and whisker plot (panel A) compares total mDGI scores (median, 25th and 75th percentiles, and total range) in each of the 6 groups. The total mDGI scores ranged from 4 to 64 in all of the participant groups.
Box and whisker plots comparing total modified Dynamic Gait Index (DGI) scores (A), total time (TimLvlTotal) scores (B), total “usual pace” task scores (C), and total “horizontal head turn” task (HHeadTotal) scores (D) by group. PD=Parkinson disease, TBI=traumatic brain injury. Circles represent outliers, with participant outlier numbers.
Performance Facet Scores
As shown in Table 2, time facet scores were significantly lower for 4 diagnostic groups (vestibular dysfunction, TBI, gait abnormality, and stroke) than for the control cohort. Total GP scores were significantly lower for all 5 mobility-impaired groups. Finally, significantly lower LOA scores were found only for the groups with stroke and TBI. As with the total mDGI scores, the range of scores for each of the 3 performance facets was large, suggesting that the performance of any particular individual could not be inferred from the group performance. These findings are evident in the Figure; a box and whisker plot (panel B) compares time facet scores (median, 25th and 75th percentiles, and total range) in each of the 6 groups.
Task-Specific Patterns of Performance
As expected, task-specific patterns of performance were found (Tab. 3); for the most part, these were consistent with the patterns of performance expected on the basis of the medical diagnosis. For example, all task scores were significantly lower for the groups with stroke and gait abnormality than for the control cohort, as expected, with effect sizes of 0.91 to 1.31. Also as expected, the group with vestibular dysfunction scored significantly lower than the control cohort on both “horizontal head turn” and “vertical head turn” tasks (effect sizes of 0.49 and 0.42, respectively), but performance on the remaining tasks was not significantly different. For the group with PD, lower task scores were found on the “change pace” and “around obstacles” tasks (effect sizes of 0.47 and 0.52, respectively), as predicted. However, in contrast to our expectations, scores on the “pivot turn” task were not significantly lower (F=7.98, P=.005, effect size=0.45). Finally, in contrast to the expected pattern of performance, the group with TBI had lower scores on all 8 DGI tasks than the control cohort, with effect sizes ranging from 0.53 to 0.72.
Although, for the most part, task-specific patterns of performance on the basis of the medical diagnosis were found, the ranges of total scores within each of the 8 tasks varied from 0 (100% impairment) to 8 (0% impairment), supporting the expectation that mobility performance would vary greatly among people with the same medical diagnosis. This within-group variability in task scores is shown in the Figure; box and whisker plots compare performance on the “usual pace” task (panel C) and the “horizontal head turn” task (panel D) for the 6 groups.
Mapping mDGI Scores to CMS Severity Indicators
To facilitate the interpretation of the severity of mobility limitation, we mapped mDGI scores to the CMS severity modifier classification system. Table 4 summarizes the proposed model, showing the relationship between total mDGI, performance facet, and task scores and the severity of mobility disability classified with the CMS framework.
Modified Dynamic Gait Index (mDGI) Score Mapping to Severity Indicators
Discussion
The primary aim of the present study was to examine the relationship between medical diagnoses and mDGI scores to determine whether the administration of the mDGI could be changed on the basis of expected patterns of performance in an individual with a specific medical diagnosis. Another aim of the present study was to map CMS severity indicators to mDGI scores (total, performance facet, and task) to provide guidelines for interpreting changes in mobility function with mDGI scores.
Patterns of Performance as a Function of Medical Diagnosis
The results of this research confirm the presence of diagnosis-specific differences in performance on individual mDGI tasks; however, as we expected, there was large heterogeneity within each diagnostic group. All task scores were significantly lower for both the group with stroke and the group with gait abnormality than for the control cohort. This finding was consistent with previous research reporting impaired performance on complex walking tasks in both people with stroke6 and older adults with mobility disability.7
Consistent with previous reports, the group with vestibular dysfunction scored significantly lower than the control cohort on both “horizontal head turn” and “vertical head turn” tasks, but the results were not significantly different on the remaining tasks.4,5 However, we also found that the range in performance on all tasks in participants with vestibular dysfunction was 0 to 8, suggesting that although some participants showed the expected pattern, others did not. For the group with PD, lower task scores were found for the “change pace” and “around obstacles” tasks, as predicted; however, in contrast to our expectations, “pivot turn” task scores were not significantly lower in participants with PD. This finding may have been due to our stringent criteria for determining the level of significance. The effect size for the difference between the group with PD and the control cohort in the “pivot turn” task scores was 0.45. Finally, compared with the control cohort, the group with TBI had lower scores on all 8 DGI tasks. This finding was not consistent with the study of Kleffelgaard et al,11 who reported that after TBI, patients with self-reported balance problems had significantly lower scores on “usual pace,” “change pace,” “horizontal head turn,” “vertical head turn,” and “pivot turn” tasks than people with no balance problems. These differences may be due to random differences in the TBI samples in the 2 studies, differences in the measures used, or differences in the inclusion criteria (ie, Kleffelgaard et al11 omitted patients who needed a gait aid, whereas participants who needed gait aids [but not the physical assistance of another person] were included in the present study).
In summary, the present study provided some support for the patterns of performance expected at the task level on the basis of the medical diagnosis. However, as expected, the range of performance within each diagnostic group was large; therefore, an individual's performance could not be predicted on the basis of his or her medical diagnosis. The results are not surprising because differences between groups represent a central tendency of a large group but do not provide information about an individual's performance. This information is critical to the development of treatment strategies for improving walking function. On the basis of the results, we suggest that it is not reasonable or advisable to drop or skip any task on the basis of the medical diagnosis because the performance of individual patients frequently deviates from expected performance patterns. The results support the importance of completing all 8 tasks in every patient regardless of the diagnosis to fully characterize mobility function.
Mapping mDGI Scores to CMS Severity Modifiers
The framework proposed here relates mDGI scores to the severity modifier scale developed by the CMS. A framework for mapping the original DGI total score (range=0–24) to severity modifiers was proposed; in that framework, the total score range of 0 to 24 was divided by 7 to establish the CMS impairment categories.16 Here we propose a different framework for mapping mDGI total, performance facet, and task scores to the 7 CMS impairment categories: the use of an underlying Rasch scale to convert raw scores to an interval (θ) scale. We believe that this approach is more appropriate for establishing cutoff points for impairment categories. The Rasch scales developed and used by Shumway-Cook et al2 and Matsuda et al3 demonstrated that the true intervals between raw scores differed at different points on the mDGI raw score scale. Therefore, the use of an interval scale helps to ensure that score differences are neither overinterpreted nor underinterpreted on the basis of their location on the scale. Given that 7 severity indicator levels are required for reporting, the division of the Rasch interval scale into 7 equal intervals serves as an appropriate starting point for using mDGI scores to assess the severity of impairment. However, further research is needed to verify the validity of the proposed CMS severity indicator cutoff points for the mDGI.
Interpreting changes in mDGI scores relative to shifts in CMS impairment categories could be used to infer improvement or deterioration of mobility performance in patients. However, information on minimal detectable changes (MDCs) for mDGI scores also must be considered. For example, an mDGI raw score change of 1 point could move a patient from one CMS impairment category to another; however, a patient with a borderline score could easily be on either side of the cutoff score by chance alone on the basis of the standard error of measurement (the basis for MDCs). Therefore, when using mDGI scores to interpret changes in mobility function, clinicians should consider both whether the score change is sufficient to change the CMS impairment category and is equal to or greater than the published MDC for that score. Matsuda et al3 recommended that the 95% MDCs for scores on the mDGI be set at 7 for the total score, 4 for the gait pattern score, 2 for the level of assistance score, and 3 for the time score. Because the present study excluded people who required the physical assistance of another person to complete the “usual pace” task, the proposed model for mapping mDGI scores to CMS severity indicators should be considered preliminary and should be re-examined with a population of participants that includes those with a low level of functioning.
Limitations
The present study has several limitations. First, the sample sizes of the groups included in the present analysis were different. The results may have been biased by the particular people in each group, particularly for the groups with smaller sample sizes. The people recruited for the study were volunteers and were actively receiving physical therapy. In addition, people who required the assistance of others to complete the baseline “usual pace” task were excluded. Therefore, the results may not be representative of the performance of a larger population, including people with low mobility function. Another limitation was that several groups included in the original sample were not included in the present analysis because of insufficient sample size. Therefore, the results cannot be generalized to all patient populations. An additional limitation is that specific information related to each medical diagnosis was not available (eg, time since onset and severity of condition).
Future Research
Future research should focus on the sensitivity and specificity of the mDGI for predicting fall risk with prospective fall data. In addition, research establishing the minimal clinically important difference is needed. An increased understanding of performance facet scores, the effects of treatment strategies on each aspect of performance, how a change in one aspect of performance affects the others, and the total score also is needed.
In conclusion, the results of this research support the importance of maintaining all 8 tasks in the assessment of mobility function in older people and people with neurologic conditions. The preliminary model for mapping mDGI scores to CMS severity indicators should assist clinicians in interpreting mobility performance, including changes in function over time, changes associated with treatment, or both.
Footnotes
Dr Taylor and Dr Shumway-Cook provided concept/idea/research design and data analysis. All authors provided writing. Dr Matsuda and Dr Shumway-Cook provided data collection. Dr Matsuda provided project management. The authors thank the testing sites, physical therapists, and all of the participants for their contributions to this study.
The University of Washington Human Subjects Division approved all procedures used in this study.
This study was supported by a grant from the Walter C. and Anita C. Stolov Research Fund, Department of Rehabilitation Medicine, University of Washington.
- Received July 18, 2014.
- Accepted November 23, 2014.
- © 2015 American Physical Therapy Association
References
Issue highlights
