Gait Variability in Older Adults: Observational Rating Validated by Comparison With a Computerized Walkway Gold Standard

Abstract

Background and Purpose: Gait variability has been measured with computerized technology–intensive techniques, which are not practical in clinical settings. The purpose of this study was to validate an observational rating of gait variability for routine clinical practice.

Subjects: Community-dwelling older adults aged 65 years and older (n=46; mean age=81.2 years, SD=6.8 years, range=66–91 years) participated in this study.

Methods: The standard deviation of stance time (stance time variability) derived from gait characteristics recorded by use of a computerized walkway was used as the gold standard for gait variability. The validity of the diagnostic test evaluated in this study (an observational rating of gait variability) was determined by comparison with the quantitative measure of stance time variability.

Results: Six validity indexes were defined for the observational rating of gait variability: sensitivity=81%; specificity=53%; positive predictive value=65%; negative predictive value=71%; positive likelihood ratio=1.72; and negative likelihood ratio=0.36.

Discussion and Conclusion: An observational rating of gait variability was validated by comparison with stance time variability derived from a computerized walkway. The concurrent validity of the 2 methods of determining gait variability provides support for the use of the observational rating as an alternative measure of gait variability for the purpose of identifying older adults at risk for mobility disability in clinical settings.

In older adults, gait variability is related to future mobility disability¹ and falls.^2–4 Gait variability also is related to balance, mental and functional status, and limitations of quality of life.² A valid measure of gait variability may help clinicians identify older adults at risk for falls and mobility disability. The majority of research studies that have examined the importance of gait variability have used computerized technology-intensive techniques to quantify gait variability.^2,4–7 The 2 main methods used to measure gait variability, foot switches² and instrumented walkways,⁵ are labor-intensive and require relatively expensive equipment; therefore, these methods are not practical for routine clinical practice.

The Modified Gait Abnormality Rating Scale (GARS-M), a 7-item observational scale, is a qualitative measure designed to identify abnormalities of gait characteristics that are associated with the risk of falling in older adults.⁸ Variability, an item of the GARS-M that is used to measure the inconsistency and arhythmicity of stepping, is relatively easy to administer and may be a low-cost alternative to the technology-intensive methods currently used to measure gait variability in clinical settings.

The purpose of this study was to validate an observational rating of gait variability by comparison with gait variability determined from gait characteristics recorded by use of a computerized walkway. We hypothesized that an observational rating of gait variability, that is, GARS-M item 1 (variability),⁸ is a valid alternative for the categorization of gait variability on the basis of stance time variability determined from a recording of footprints during gait on an instrumented walkway (a quantitative, evidence-based standard for predicting mobility disability).¹

Method

For this validation study of a clinical measure of gait variability, the stance time standard deviation (stance time variability) determined from gait characteristics recorded on a computerized walkway (GaitMat II^*)⁹ was used as the gold standard for gait variability. The observational rating of gait variability, that is, the GARS-M variability item, was determined from videotapes (recorded in a clinic) of the participants walking. The validity of this diagnostic test (the observational rating of gait variability) was determined by comparison with the quantitative measure of stance time variability.

Participants

Forty-six community-dwelling older men (n=10) and women (n=36) living independently or in assisted living at a senior continuing care residential community in Pittsburgh, Pennsylvania, volunteered to participate in this study (mean age, 81.2 years; SD, 6.8 years; range, 66–91 years). Inclusion criteria for the trial were as follows: (1) aged 65 years and older; (2) independent in ambulation with a straight cane or no assistive device; and (3) able to grant informed consent to participate in this research study, as indicated by a Mini-Mental State Examination score of ≥24. Exclusion criteria were as follows: (1) self-reported persistent lower-extremity pain; (2) presence of a progressive motor disorder, such as multiple sclerosis or Parkinson disease; (3) lower-extremity amputation; and (4) presence of medical instability (defined as an acute illness or cardiovascular disease, not well controlled with medication; hospitalization for cardiac reasons in the preceding 6 months; major thoracic or joint surgery; or a myocardial infarction, stroke, or transient ischemic attack in the preceding 6 months).

Measurement of Clinical Characteristics

A structured interview was used to assess each participant's age, race, mental status, and comorbid conditions. Mental status was assessed with the Mini-Mental State Examination (score range=0–30).¹⁰ Using the Comorbidity Index,¹¹ we asked all participants whether they were ever told by a physician that they had any of a list of medical conditions, including cardiovascular disease (angina, congestive heart failure, or heart attack), neurologic conditions (stroke or Parkinson disease), lung disease, musculoskeletal conditions (arthritis, osteoporosis, fracture, or joint replacement), general conditions (depression, sleep problems, or chronic pain syndrome), cancer, diabetes, or visual conditions (glaucoma or cataracts). The numbers (0–18) and the domains (0–8) of the comorbidities were recorded.

Measurement of Gait Variability

Computerized walkway.

The GaitMat II is a computerized system for measuring the spatial and temporal characteristics of gait. The GaitMat II consists of a 4-m walkway with pressure-sensitive switches embedded in the surface and a computer system for data collection and analysis. Additional 1-m sections at both ends of the walkway allow for the acceleration and deceleration of participants walking the length of the gait mat. The concurrent validity of the GaitMat II has been determined by comparison with the spatiotemporal characteristics of gait simultaneously recorded with a Vicon motion analysis system.^†,¹² On the basis of the intraclass correlation coefficient (ICC [2,1]) for the relationship between the GaitMat II and motion analysis recordings, the concurrent validity of the temporal characteristics of gait recorded with the GaitMat II (r=.99) is greater than that of the spatial characteristics of gait (r=.24).¹²

Gait characteristics, including gait speed, step length, step width, and stance time, were recorded. After 2 practice passes on the GaitMat II, each participant was asked to complete 2 passes on the GaitMat II at a self-selected walking speed. As defined for the GaitMat II automated analysis of recorded footprints, step length is the distance between 2 consecutive footprints, measured from the heel of one footprint to the heel of the next footprint; step width is the distance between the outermost borders of 2 consecutive footprints; and stance time is the time during which the foot is in contact with the floor (ie, from the initial foot-floor contact until the final foot-floor contact). Variability (step length, step width, or stance time) is defined as the standard deviation of the mean for all steps, right and left, recorded during 2 passes of walking on the GaitMat II or as the coefficient of variation, expressed as (SD/X̄) × 100.

To obtain the gait variability measure from the GaitMat II record, an evaluator observes the record of footprints and removes extraneous marks and incomplete footprints (such as half footprints for footfalls at the beginning or end of the recording area of the mat). Subsequently, the data for the gait characteristic of interest, such as stance time, for each step are extracted to a secondary database (such as Microsoft Excel^‡) to average the data across steps and derive the standard deviation, representing the variability of the gait measure of interest (stance time variability). Because the gold standard for stance time variability (stance time standard deviation) that was used to define gait variability was determined from all of the steps (right and left) from a single person in the study by Brach et al,¹ both right and left steps were combined in the present study for analysis of the gait characteristics from the GaitMat II record.

GARS-M.

The GARS-M is a 7-item observational scale that was developed to identify abnormalities of gait characteristics associated with the risk of recurrent falls in older adults.^8,13 The 7 items are gait variability, guardedness, staggering, foot contact, hip range of motion, shoulder extension, and arm–heel-strike synchrony.

The GARS-M is administered by videotaping of an individual walking at a self-selected pace approximately 8 m in one direction, turning, and walking back to the starting point for 8 m, for a total distance of about 16 m on a level surface. Videotaping was completed by one person, who moved a standard digital video camera to capture the anterior, posterior, and lateral aspects of walking. Videotaping usually took 1 to 2 minutes to complete. The digital videotape of the walk was replayed on a computer to be scored by use of standard digital video software. The videotaping allowed for repeated playback and slow- and stop-action viewing of the walk as needed for scoring, which required about 3 to 5 minutes per record; less time was needed as the evaluator gained experience. Each of the 7 items of the GARS-M was scored on a criterion-based rating scale of 0 to 3, with higher scores indicating poorer performance. Only gait characteristics during the straight portion of the walk were scored; abnormalities of gait during the turn were not evaluated.

The total GARS-M score showed interrater reliability (ICC=.97) and intrarater reliability (ICC=.97).⁸ Concurrent validity was determined by comparison with temporal and spatial gait characteristics, and construct validity was determined on the basis of the ability of the measure to distinguish older adults with a history of falls from older adults without a history of falls.⁸

The variability item is a measure of the inconsistency and arhythmicity of stepping and arm movements.⁸ Scores were assigned as follows: 0 for fluid and predictably paced limb movements; 1 for occasional interruptions (changes in velocity), approximately 25% of the time; 2 for unpredictability of rhythm, approximately 25% to 75% of the time; and 3 for random timing of limb movements. Substantial agreement for scoring of this item (variability) was reported, with kappa values of .635 for multiple raters and .676 for a single rater (the same rater who scored all of the GARS-M measures for the present study).⁸

Procedure

During a single clinic visit, gait variability was determined by use of a computerized walkway (GaitMat II), and each participant was videotaped walking for later scoring by the observational rating evaluated in the present study (the variability item of the GARS-M). The order of testing for gait variability with the GaitMat II and the GARS-M measure was not defined, and no attempt was made to randomize the testing order. Participants were given several minutes of rest between gait tests and additional time as needed to recover to their baseline self-determined state of rest. All testing was supervised by physical therapists experienced in assessment of and intervention for older adults with mobility problems. To minimize bias, the 2 measures of gait variability were analyzed independently of each other. One clinician analyzed the computerized walkway data, and a different clinician used the GARS-M measure to rate the videotapes of participants walking.

Data Analysis

For validation of the observational rating of gait variability, participants were dichotomized into 2 groups (variable and nonvariable) on the basis of 2 methods of gait variability assessment: (1) stance time variability (SD) derived from steps recorded during 2 passes on the computerized walkway (gold standard) and (2) an observational rating of gait variability, that is, the GARS-M variability item score (diagnostic test). Older adults with a stance time variability (SD) of ≥0.0365 second, a value for stance time variability that has been determined to be an indicator of mobility disability (ie, self-reported difficulty walking 0.5 mile) in older adults,¹ were categorized into the variable group (VG-GM). Older adults with a stance time variability (SD) of <0.0365 second were categorized into the nonvariable group (NVG-GM). Older adults with a GARS-M variability item score of 1, 2, or 3 (diagnostic test) were categorized into the variable group (VG-GARSM), and older adults with a GARS-M variability item score of 0 were categorized into the nonvariable group (NVG-GARSM).

For establishment of the concurrent validity of the observational rating of gait variability, that is, the GARS-M variability item, 6 validity indexes were determined. The 6 validity indexes for identifying older adults with or without stance time variability (determined from the computerized walkway record of stance time during gait) were sensitivity, specificity, positive and negative predictive values, and positive and negative likelihood ratios.

The Mann-Whitney test was used to compare values for the rating of variability (GARS-M item 1 [variability]) between adults known to demonstrate (VG-GM group) and adults known not to demonstrate (NVG-GM group) stance time variability. The same comparisons were made between the VG-GM and NVG-GM groups for the 6 other GARS-M items and the total GARS-M score to explore differences in observed gait characteristics for groups of older adults defined by differences in stance time variability.

Results

The majority of the older adults (mean age=81.2 years, SD=6.8) participating in the present study were white, female, and walked slowly compared with the typical adult walking speed of 1.2 to 1.3 m/s¹⁴ (Tab. 1). Participants stratified on the basis of stance time variability (VG-GM versus NVG-GM) did not differ by age, race, sex, cognitive status, total comorbidity score, or median score in each comorbidity domain. The numbers of participants reporting comorbid conditions were as follows: cardiac (7), neurologic (6), respiratory (13), musculoskeletal (29), general (14), cancer (14), diabetes (3), and visual (22). Compared with older adults categorized into the NVG-GM group (with respect to stance time variability), older adults categorized into the VG-GM group walked more slowly and with shorter step length, greater step width, and longer stance time. The VG-GM and NVG-GM groups differed in gait variability, with step length and stance time variability being larger and step width variability being smaller for older adults in the VG-GM group than in the NVG-GM group (Tab. 1).

Table 1.

Characteristics of Older Adults Stratified on the Basis of Stance Time Variability^a

^aData are X̄ (SD) unless otherwise indicated. VG-GM=older adults with stance time variability (SD) of ≥0.0365 s (variable group), NVG-GM=older adults with stance time variability (SD) of < 0.0365 s (nonvariable group), MMSE=Mini-Mental State Examination, COV=coefficient of variation.

^bComorbidity data for only a subset of the sample (n=34: VG-GM group [n=18] and NVG-GM group [n=16]).

With stance time variability as the gold standard and the observational rating as the diagnostic test to be evaluated, 6 validity indexes were calculated (Tab. 2). Of 21 older adults with stance time variability, 17 were correctly identified by the observational rating of gait variability (sensitivity=81%). Of 19 older adults without stance time variability, 10 were correctly identified by the observational rating of gait variability (specificity=53%). The positive predictive value of the observational rating of variability was 65%; of the 26 older adults identified as having gait variability by the observational rating, 17 were in the VG-GM group. The negative predictive value of the observational rating of variability was slightly better, at 71%; of the 14 older adults identified as not having gait variability by the observational rating, 10 were in the NVG-GM group. The positive and negative likelihood ratios were 1.72 and 0.36, respectively.

Table 2.

Validity Indexes for Observational Rating of Variability for Stance Time Variability Groups^a

^aData are reported as number of older adults, unless otherwise indicated. GARS-M=Modified Gait Abnormality Rating Scale; VG-GM=older adults with stance time variability (SD) of ≥0.0365 s (variable group); NVG-GM=older adults with stance time variability (SD) of <0.0365 s (nonvariable group); VG-GARSM=older adults with a GARS-M variability item score of 1, 2, or 3 (variable group); NVG-GARSM=older adults with a GARS-M variability item score of 0 (nonvariable group).

^bNumber for specificity, negative predictive value, and negative likelihood ratio.

Compared with older adults classified into the NVG-GM group on the basis of stance time variability, older adults classified into the VG-GM group differed with regard to the GARS-M variability item score, with median scores of 1 (mean=0.95; range=0–2) and 0 (mean=0.53; range=0–2), respectively. The measurable difference in the stance time variability of the older adults corresponded to an observable difference in gait variability noted by a physical therapist rating the gait abnormality.

A comparison of the GARS-M item and total scores for participants stratified on the basis of stance time variability illustrated differences in observed gait abnormalities for groups defined by values for stance time variability associated or not associated with mobility disability. Differences between the groups of older adults with regard to variability, guardedness, foot contact, and hip range of motion (GARS-M items 1, 2, 4, and 5) and total GARS-M scores were seen (Tab. 3).

Table 3.

Differences in Modified Gait Abnormality Rating Scale (GARS-M) Item and Total Scores for Stance Time Variability Groups^a

^aVG-GM=older adults with stance time variability (SD) of ≥0.0365 s (variable group), NVG-GM=older adults with stance time variability (SD) of <0.0365 s (nonvariable group.

Discussion

In this cross-sectional study of the concurrent validity of an observational rating of gait variability for older adults, the observational rating yielded a relatively high sensitivity in comparison with the specificity and a relatively high negative predictive value in comparison with the positive predictive value for recognizing older adults with stance time variability. Because a condition is likely to be absent when a test is negative and the sensitivity is high,¹⁵ the observational rating of gait variability may be most useful in ruling out the diagnosis of gait variability when the GARS-M variability item rating indicates no gait variability. According to the guide proposed by Jaeschke and colleagues,¹⁶ a positive likelihood ratio of 1.72 and a negative likelihood ratio of 0.36 indicated small changes between probability before testing and probability after testing. Because the negative likelihood ratio of 0.36 was more meaningful than the positive likelihood ratio of 1.72, the observational rating of gait variability may be more useful in ruling out the diagnosis of gait variability. Although the likelihood ratios for the observational rating of gait variability are not promising, older adults classified by the observational rating of gait variability into the variable group (VG-GARSM) walked more slowly than those classified into the nonvariable group (NVG-GARSM), with mean (SD) gait speeds of 0.81 (0.24) and 1.04 (0.25) m/s, respectively (P<.005), and had greater gait variability, with mean (SD) stance time variabilities of 0.05 (0.04) and 0.03 (0.01) seconds, respectively (P<.05).

In summary, the use of the GARS-M item 1 observational rating of gait variability has good potential for recognizing older adults who have increased stance time variability and who may be at risk for mobility disability and falls; however, the use of this method may also result in the identification of some older adults without gait variability (false-positive results), who would not have the same risk for mobility disability. We believe that the appropriate follow-up for older adults recognized as being at risk for mobility disability and falls is a comprehensive geriatric medical and physical evaluation of factors contributing to mobility and balance problems and the establishment of an integrated plan of care to address the physical, mental, social, and environmental factors contributing to the risk of falling. Despite some health care costs and some burden of time and effort for the older adults and clinicians involved, the costs of such geriatric evaluation and intervention seem unlikely to exceed the costs of health care for older adults with a loss of independence secondary to mobility disability or injuries associated with falls or the personal distress of older adults with reduced independence and their caregivers. We believe that the observational rating of gait variability is a conservative approach to the management of mobility disability in older adults.

The VG-GM and NVG-GM groups of older adults also exhibited other observable differences in gait abnormalities recognized using the GARS-M. Specifically, older adults with variable gait demonstrated greater forward trunk flexion (eg, head, arms, and trunk in front of the base of support) associated with diminished propulsion in stepping, a reduced foot-floor angle at heel strike, and reduced extension of the hip of the trailing limb at push-off. The “trunk leading strategy,” altered heel strike pattern, and minimal or no hip extension are all gait abnormalities^17–19 that could contribute to disruption of the automated locomotor (stepping) pattern²⁰ and gait variability. The higher median total GARS-M score for older adults in the VG-GM group than for those in the NVG-GM group indicates that adults with variable gait have a greater risk for recurring falls.¹³

Gait variability has been measured on the basis of spatial and temporal characteristics of gait. In the present study, stance time variability was chosen as the representative (gold standard) measure for gait variability because of the greater reliability of temporal characteristics of gait than of spatial characteristics of gait¹² and because stance time variability is an independent predictor of mobility disability.¹ Stance time variability was shown to be negatively correlated with gait speed.^2,5 However, the relationship between stance time variability and gait speed was only moderate in the current study, suggesting that both high stance time variability and low stance time variability exist in older adults who walk near or at the typical gait speed as well as those who walk slowly.^1,21 When the participants were stratified on the basis of gait speed, we found that for older adults walking near or at typical gait speed (≥1.0 m/s), 1 of 6 of those in the VG-GM group and 7 of 12 of those in the NVG-GM group were correctly identified by the observational rating of gait variability. In contrast, for older adults who walked slowly (<1.0 m/s), 16 of 19 in the VG-GM group and 3 of 7 in the NVG-GM group were correctly identified by the observational rating of gait variability. Visual recognition of gait variability appears to be more difficult for older adults who walk near or at the typical gait speed.

The present study has several limitations. The observational rating of gait variability as measured by GARS-M requires videotaping, which may not be efficient or feasible in some clinical settings. However, in contrast to the computerized walkway, which involves more time for cleaning and extracting and calculating gait variability measures, the variability item of the GARS-M represents a less-expensive clinical method of determining gait variability. Future studies could focus on determining the reliability and validity of the direct observation of gait variability. The validation of such a scale could lead to a more efficient and cost-effective alternative clinical measure of gait variability. The gait variability measure obtained from the GaitMat II in the present study was calculated from a limited number of steps and therefore may not reflect the true value of gait variability. Future investigations validating observational rating with gait variability measures derived from foot sensor-based systems would allow for data collection over ground for a greater distance and could address the problem of a limited number of steps.

In the present study, stance time variability was derived by combining left and right steps recorded on the computerized walkway, in the same way in which stance time variability was determined and used to predict mobility disability.¹ Measures of stance time variability based on only right or only left steps may have a different clinical meaning. Similarly, measures of stance time variability determined from only right or only left steps also may not relate to the observational rating of gait variability, that is, GARS-M item 1 (variability), as reported in the present study. Additional studies are needed if the specific intent is to understand the relationship of observational gait ratings and the clinical meaning of only right or only left stance time variability in older adults.

Conclusion

In the present study, we validated an observational rating of gait variability, the GARS-M item 1, by comparison with the standard method for determining gait variability, an instrumented walkway (the GaitMat II), in older adults who were independent in ambulation with a straight cane or no assistive device. The concurrent validity of the methods of determining gait variability in older adults provides support for the use of the observational rating by physical therapists in clinical settings as an alternative measure of gait variability for identifying older adults at risk for mobility disability.