Expanding the Scoring System for theDynamic Gait IndexAnne Shumway-Cook, Catherine S. Taylor, Patricia Noritake Matsuda,Michael T. Studer, Brady K. Whetten

Background. The Dynamic Gait Index (DGI) measures the capacity to adapt gaitto complex tasks. The current scoring system combining gait pattern (GP) and levelof assistance (LOA) lacks clarity, and the test has a limited range of measurement.

Objective. This study developed a new scoring system based on 3 facets ofperformance (LOA, GP, and time) and examined the psychometric properties of themodified DGI (mDGI).

Design. A cross-sectional, descriptive study was conducted.

Methods. Nine hundred ninety-five participants (855 patients with neurologicpathology and mobility impairments [MI group] and 140 patients without neurolog-ical impairment [control group]) were tested. Interrater reliability was calculatedusing kappa coefficients. Internal consistency was computed using the Cronbachalpha coefficient. Factor analysis and Rasch analysis investigated unidimensionalityand range of difficulty. Internal validity was determined by comparing groups usingmultiple t tests. Minimal detectable change (MDC) was calculated for total score and3 facet scores using the reliability estimate for the alpha coefficients.

Results. Interrater agreement was strong, with kappa coefficients ranging from90% to 98% for time scores, 59% to 88% for GP scores, and 84% to 100% for LOAscores. Test-retest correlations (r) for time, GP, and LOA were .91, .91, and .87,respectively. Three factors (time, LOA, GP) had eigenvalues greater than 1.3 andexplained 79% of the variance in scores. All group differences were significant, withmoderate to large effect sizes. The 95% minimal detectable change (MDC95) was 4 forthe mDGI total score, 2 for the time and GP total scores, and 1 for the LOA total score.

Limitations. The limitations included uneven sample sizes in the 2 groups. TheMI group were patients receiving physical therapy; therefore, they may not berepresentative of this population.

Conclusions. The mDGI, with its expanded scoring system, improves the range,discrimination, and facets of measurement related to walking function. The strengthof the psychometric properties of the mDGI warrants its adoption for both clinicaland research purposes.

A. Shumway-Cook, PT, PhD,FAPTA, Department of Rehabilita-tion Medicine, Division of PhysicalTherapy, University of Washing-ton, 1959 NE Pacific St, Box356490, Seattle, WA 98195(USA). Address all correspon-dence to Dr Shumway-Cook at:ashumway@u.washington.edu.

C.S. Taylor, PhD, College of Edu-cation, University of Washington.

P.N. Matsuda, PT, DPT, PhD,Department of RehabilitationMedicine, Division of PhysicalTherapy, University of Washington.

M.T. Studer, PT, MHS, NCS,Northwest Rehabilitation Associ-ates, Salem, Oregon.

B.K. Whetten, PT, DPT, GCS, North-west Rehabilitation Associates.

[Shumway-Cook A, Taylor CS,Matsuda PN, et al. Expandingthe scoring system for theDynamic Gait Index. Phys Ther.2013;93:1493–1506.]

© 2013 American Physical TherapyAssociation

Published Ahead of Print:June 27, 2013

Accepted: June 24, 2013Submitted: January 26, 2013

Research Report

Post a Rapid Response tothis article at:ptjournal.apta.org

November 2013 Volume 93 Number 11 Physical Therapy f 1493 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

The Dynamic Gait Index (DGI)is a commonly used clinicalmeasure that evaluates the

capacity to adapt gait to complexwalking tasks commonly encoun-tered in daily life.1 The DGI is basedon a person-environment model ofmobility disability (defined as reducedparticipation in the mobility domain)in which environmental demandsare categorized into 8 dimensions—distance, temporal, ambient, terrain,physical load, attention, posturaltransitions, and density—represent-ing the external demands that haveto be met for an individual to bemobile within a particular environ-ment.2 Item 1 of the DGI is consid-ered the reference or baseline taskand examines the ability to walkunder low-challenge conditions(self-paced, level surface, gait). Theremaining 7 items examine a per-son’s ability to adapt gait to taskdemands in 4 environmental dimen-sions: temporal (changing speed),postural transition (change in direc-tion using a pivot turn, change inhead orientation including horizon-tal and vertical head turns), terrain(climbing stairs), and density dimen-sion (stepping over and aroundobstacles for collision avoidance).Performance on each item is evalu-ated using an ordinal scale withcriteria based on a combination ofgait pattern (GP), speed, and levelof assistance (LOA). Ordinal scoresfor each item range from 0 (severeGP impairment, unable to performwithout the physical assistance ofanother person) to 3 (normal GP,in a timely manner, no assistancerequired), with a total score rangefrom 0 (worst/poor) to 24 (bestfunction).

The psychometric properties of theoriginal DGI have been studied in awide variety of patient populations,including older adults3–7 and thosewith stroke,8,9 Parkinson disease,10

multiple sclerosis,11–13 and vestibulardeficits.14,15 Interrater reliability isreported to be high.8,11,14 Test-retestreliability values also were high, withintraclass correlation coefficientsranging from .84 to .96.8–10 Concur-rent validity coefficients were foundto be moderate to high with theTimed “Up & Go” Test (TUG) andTimed Walking Test in people withstroke8 and the 10-Meter Walk Testand Postural Assessment Scale forStroke Patients.9 Performance on theDGI has been used as a measure offall risk; however, the cutpoint dif-ferentiating fallers from nonfallersvaries by patient population, rangingfrom 19 to 23.16,17

Two studies have used Rasch analy-ses to examine DGI function in adultmale veterans6 and individuals withvestibular deficits.15 Results suggestthe DGI is unidimensional, with arange of item difficulty from low(walking on level surface, changingspeed) to high (horizontal headturns, stairs). Multiple items (eg,walking at usual pace, walkingaround obstacles) have a comparablelevel of difficulty, leading to the sug-gestion that easier and psychometri-cally redundant items be dropped toshorten the test.6,15

Despite its psychometric strengths,the original DGI has a number oflimitations. First, performance israted using an ordinal score thatincorporates 3 facets of performance(assessment of GP relative to normal,LOA, and, in some items, time) with-out a clear explanation of the relativecontribution of each of these facetsof performance to the task score.Putting all 3 facets of performanceinto a single score prevents thera-pists from monitoring improvementor decline in mobility function rela-

tive to these different facets ofperformance. Second, a ceilingeffect has been reported when usingthe DGI to evaluate walking incommunity-dwelling older adultswith relatively high mobility func-tion, thus limiting responsivenessand sensitivity in this population.7

Both of these limitations reduce theutility of the DGI as an outcomemeasure.

The purposes of this study were:(1) to develop a new scoring systemfor the DGI based on 3 facets ofperformance (ie, LOA, GP, and time)and (2) to examine the psychometricproperties of a modified version ofthe original DGI (mDGI). One goal ofthe new scoring system was to pro-vide a greater range of measurementso that scores could be used to reli-ably monitor improvement or deteri-oration of mobility in patients witha broader range of function than iscurrently possible using the originalscoring rubric. Additionally, the newscoring system would enable moni-toring of patients’ improvement ordeterioration in any 1 of the 3 iden-tified facets of performance (ie, LOA,GP, and time) in each walking task.A second goal of the study was toinvestigate the psychometric proper-ties of the mDGI scoring system,including dimensionality and poten-tial redundancy of measurement.

MethodIn order to conduct the plannedRasch analyses, sufficient numbers ofparticipants were needed to obtainstable item parameters. We esti-mated that 10 participants wereneeded for each score parameter;because the mDGI included 64 scoreparameters, a minimum of 640 par-ticipants was needed.

RecruitmentIdentification of patient participantsbegan with recruitment of potentialclinical testing sites. This processinvolved posting an e-mail on the

Available WithThis Article atptjournal.apta.org

• eAppendix: Modified DynamicGait Index

Modified Dynamic Gait Index

1494 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

Section of Neurology listserve forthe American Physical Therapy Asso-ciation (APTA) requesting assistancefrom clinical sites currently usingthe DGI as a part of current clinicalevaluation procedures. A letter ofagreement was signed by each par-ticipating center. Within each site,study participants were recruitedfrom among patients with neurolog-ical impairments, currently receivingphysical therapist services for bal-ance and mobility deficits, whowould be evaluated using the DGIas part of usual care. Additionalinclusion criteria included ability towalk 6.1 m (20 m) without the phys-ical assistance of another person,although use of an assistive devicewas permitted.

A convenience sample of adultswith no neurologic diagnosis wasrecruited from volunteers respond-ing to flyers posted at the Depart-ment of Rehabilitation Medicine,University of Washington. In addi-tion, letters were sent to retirementcommunities in the greater Seattleand Bellevue area. Criteria for inclu-sion in the control cohort were:age between 15 and 99 years, noneurologic diagnosis, ability to walkwithout the physical assistance ofanother person for a distance of6.1 m, and ability to give informedconsent. Potential participants wentthrough an initial telephone screento determine eligibility prior to beingtested at 1 of 5 community sites. Allparticipants gave informed consentprior to testing.

Modifying the DGITask modifications. The original8 tasks were retained in the mDGI;however, minor modifications weremade to facilitate timing and to clar-ify procedures for several of thetasks. To enable timing, a 6.1-m dis-tance was imposed for all tasks. Thechange pace task included an accel-eration phase (“walk as quickly asyou safely can”) but no deceleration

phase, as it was difficult to imposeboth of these task modifications ina distance of 6.1 m. The over obsta-cle task changed the dimensionsof the obstacles and specified thedimensions and placement of the 2obstacles. Obstacle dimensions were76 cm long, 12 cm wide, and 5 cmthick. The shoebox was eliminateddue to anecdotal reports that manypatients completed the task by slid-ing a foot around the obstacle ratherthat stepping over the obstacle(Anne Shumway-Cook, unpublishedreports). The first obstacle wasplaced 2.4 m (8 ft) from the startingpoint, with the 12-cm side flat on thefloor, requiring patients to increasestep length to clear the obstacle. Thesecond obstacle was placed withthe 12-cm side up, requiring partici-pants to increase their step height toclear the obstacle. This obstacle wasplaced 2.4 m past the first obstacle(about 4.9 m [16 ft] from the start).These modifications were based onresearch on obstacle crossing inpatients with hemiplegia, indicatingthat some patients have difficultyincreasing step length and that oth-ers have difficulty with step height18;thus, both conditions were includedin the obstacle task. The pivot turntask included a turn halfway throughthe course with a return to the start-ing position, rather than a stop, inorder to complete the 6.1-m dis-tance. The stairs task measured per-formance while going up but notdown stairs.

Scoring modifications. Scoringfor original DGI was limited to a sin-gle ordinal score for each item rang-ing from 0 to 3. For the mDGI, 3separate scores were applied to aparticipant’s performance. Thera-pists measured the time it took forthe individual to complete each taskperformed over a 6.1-m course.Level of assistance was scored usinga 3-level scale (2�no assistance;1�uses an assistive device [notincluding orthosis or braces]; and

0�requires the physical assistanceof another person, including contactguard). Gait pattern was scored on a4-level scale that differed slightly foreach task. For example, the 4-levelDGI scale for the usual pace walk onlevel surface task was:

(3) Normal: walks 6.1 m, normal GP,no evidence for imbalance

(2) Mild impairment: walks 6.1 m,mild gait deviations

(1) Moderate impairment: walks6.1 m, moderate gait deviations, evi-dence of imbalance but recoversindependently

(0) Severe impairment: cannot walk6.1 m or walks with severe gait devi-ations or imbalance

A copy of the mDGI is shown in theeAppendix (available at ptjournal.apta.org). Data for each participantwere collected for both the originalDGI and the mDGI. Prior to data col-lection, therapists participating inthis study received training on thenew scoring system for the DGI.Three patient training videos werecreated for this purpose, withdetailed verbal and written instruc-tions on the new scoring system. Alltherapists participating in this studycompleted these training videos andhad the opportunity to ask questionsregarding the new scoring system.In addition, they scored 1 trainingvideo and submitted their data tostudy coordinators, who providedfeedback to them regarding theirscores. Finally, all sites submitteddata on the first 5 patients for reviewby study investigators (P.N.M. andA.S-C.).

Converting time to an ordinalscore. In order to investigatedimensionality using factor analysisand Rasch analysis, all item scoreshad to be on a dichotomous or ordi-nal scale. Therefore, time measure-

Modified Dynamic Gait Index

November 2013 Volume 93 Number 11 Physical Therapy f 1495 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

ments were converted to a 4-levelordinal scale. All time measurementswere converted to meters per sec-ond. The time for the usual pacetask was converted to a 4-level scalebased on walking speeds docu-mented in the literature19–23:

(3) Normal walking speed: �1.1 m/s

(2) Community ambulatory: 0.8 to�1.1 m/s

(1) Limited ambulatory: 0.4 to �0.79m/s

(0) Household ambulatory: �0.39m/s

Time-level scores for tasks 2 through8 were determined through an exam-ination of the time distributions onthe remaining 7 tasks for individualsat each time level on the usual pacetask. Time distributions were exam-ined graphically through histogramsand numerically through descriptivestatistics. Initial cut scores for timelevels on the remaining 7 tasks werebased on the points of intersectionfor these distributions.

The points of intersection were mod-eled as follows:

1. For participants at each of the 4levels on the usual pace task,means and standard deviationsof time and speed (in meters persecond) were computed for tasks2 through 8.

2. A time band, based on mean (�1standard deviation), was com-puted for participants at eachusual pace time level on each ofthe other 7 tasks.

3. To set the cutpoint between timelevel 1 and time level 2 on tasks 2through 8, the midpoint was com-puted between the highest pointin the time band for participantsat level 1 and the lowest time inthe time band for participants atlevel 2. The midpoint became thetime level cutpoint between level1 and level 2 on the given task.

4. This process was repeated to setthe cutpoints between subse-quent levels for the remainingtasks.

Proposed Scoring for the mDGITable 1 summarizes the proposedscoring for the mDGI. Scores are cal-culated at the task, facets of perfor-mance, and total score levels. Tocalculate a performance score at thetask level, scores for time, GP, andLOA within a task are summed, for ascore range of 0 to 8. An individualwho is unable or refuses to completea task is scored 0 for all 3 perfor-mance indicators. In addition to the8 individual task scores, a total scorefor each of the 3 facets of perfor-mance is calculated, enabling walk-ing performance to be characterizedwith respect to time (range�0–24),GP (range�0–24), and LOA (range�0–16). Finally, an mDGI total scoreis calculated by combining the 3 per-formance scores for a total scorerange of 0 to 64. Also included inTable 1 are the environmentaldimensions reflected in the mDGI.

ReliabilityScore reliability was investigated in 3ways: an interrater reliability (IRR)study, a test-retest reliability study,and an internal consistency reliabil-ity analysis. To investigate IRR, 9volunteer participants (7 in themobility-impaired group and 2 in thecontrol group) were videotapedperforming the DGI. Videos wereposted to the research website,where they were viewed and scoredby 8 physical therapists who partic-ipated in collecting data for thestudy. Time measures were con-verted to the time-level ordinal scalefor each of the tasks. Interrater reli-ability was evaluated for both indi-vidual task scores and total scores foreach of the 3 scoring facets. The per-centage of times physical therapistsgave exactly the same score (percentexact agreement) as the criterionscore (scores assigned by the testdeveloper [A.S-C.]) for time, LOA,and GP for each task were calcu-lated. Kappa coefficients were com-puted to evaluate the degree towhich agreement with criterion

Table 1.Proposed Scoring for the Modified Dynamic Gait Index

Task

Facets of Performance

TimeGait

PatternLevel of

AssistanceTotal

Task Score

Temporal

Usual pace 0–3 0–3 0–2 0–8

Change pace 0–3 0–3 0–2 0–8

Postural

Horizontal head turns 0–3 0–3 0–2 0–8

Vertical head turns 0–3 0–3 0–2 0–8

Pivot turn 0–3 0–3 0–2 0–8

Terrain

Stairs 0–3 0–3 0–2 0–8

Density

Stepping over obstacles 0–3 0–3 0–2 0–8

Stepping around obstacles 0–3 0–3 0–2 0–8

Total facet score 0–24 0–24 0–16

Total mDGI score 0–64

Modified Dynamic Gait Index

1496 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

scores was over and above whatwould be expected by chancealone.24 Interrater reliability at theperformance facet level also wasassessed by computing total facetscores for time, GP, and LOA andcorrelating each score with the cri-terion facet scores.

To investigate test-retest reliability,257 participants (253 in the mobility-impaired group and 4 in the controlgroup) were retested using boththe original DGI and the mDGI. Forpatients with rapidly changing func-tion (eg, patients in acute and sub-acute settings), the optimal time-frame for retesting was within 24hours. For patients whose functionwas changing at a slower rate (eg,outpatients with chronic condi-tions), the optimal timeframe forretesting was between 3 and 7 daysfollowing the initial examination.Internal consistency was computedfor task scores, 3 facets of perfor-mance scores, and the total scalescore from all 995 participants usingCronbach alpha coefficients.

Investigating Performance FacetsPerformance facets were investi-gated using both factor analysis andRasch analysis. Exploratory factoranalysis was conducted for all taskscores using direct Oblimin rotationto allow for correlated factors.Oblique rotation was applied in rec-ognition of the inevitable intercorre-lations among scores. We expectedthat both GP and LOA were likely toaffect the time taken for an individ-ual to complete a task.

The Rasch model was used for theitem response theory (IRT) analysis.The Rasch model focuses on only 1of several possible task score param-eters—that of the task difficulty. Fortasks with rubric scores, the Raschmodel is expanded to address thedifficulty of moving from a lowerscore to a higher score on a rubric,referred to as “step difficulties.”

Infit and outfit statistics were usedto evaluate the fit of a model to thedata. Infit statistics were used to indi-cate how well the model fit the datafor participants whose abilities weresimilar to the level of difficulty forthat task. Outfit statistics indicatedhow well the model fit the data forall examinees, even those with abil-ities very different from the difficultyof task scores. Fit statistics thatranged between 0.70 and 1.3 wereconsidered adequate fit to the data,that is, indicated the degree to whichscores “fit together” in a singlemeasure.25 A less conservative rangeof 0.5 to 1.5 also has beenrecommended.26

The first stage of IRT analyses wasintended to ascertain whether a par-tial credit model (PCM)27 or a ratingscale model (RSM) was a better fitto the data.28,29 The PCM allowsthe parameters from all task scoresto freely vary. In contrast, the RSMassumes that, when rating scaleshave the same meaning across items(eg, all time level scores), the dis-tances between score levels are thesame across items. For example, thedifficulty of moving from a time-levelscore of 0 to a time-level score of 1(or from time level 1 to time level 2)is the same across all tasks.

To assess the performance facets, allitem scores (time, GP, LOA) for eachtask were entered into the same PCMor RSM analyses. Fit statistics wereevaluated to select the model thatbest fit the data. Once the best-fittingmodel was selected, performancefacet scores were analyzed in sepa-rate analyses to evaluate whether asingle facet or a multifaceted modelof performance better fit the data.Finally, correlations were used toexamine the relationship betweenthe original DGI and the mDGI.

Comparisons of Mobility-Impaired and Control GroupsInternal evidence for the validity ofmDGI scores was determined bycomparing performance scores inthe adults with mobility impairmentsand the control cohort using multi-ple t tests. We expected that theadults with mobility impairmentswould have consistently lower scoresfor all score categories (task, perfor-mance facet, and total scale score).Finally, the minimal detectablechange (MDC) was calculated formDGI total score and for each ofthe 3 performance facet scores forthe control cohort and the mobility-impaired group using the reliabilityestimate for the alpha coefficientsfor each group.

ResultsParticipantsForty-two physical therapists from17 sites (5 university-affiliateddepartments, 5 medical centers, 3rehabilitation hospitals, and 4 outpa-tient centers) agreed to collect data.The mean age of the therapists was37 years (range�24–64), 85% werefemale, and 24% were board-certifiedclinical specialists (Neurology Certi-fied Specialists�8, Geriatric Certi-fied Specialists�3). A total of 995participants (140 controls and 855patients) participated. As shown inTable 2, the mean age of the entiresample was 65.3 years (range�15–99), 54.5% were female, and 84.9%were white. Half of the sample usedassistive devices when walking, andalmost half reported 1 or more fallsin the previous 6 months. The sam-ple included individuals with a widerange of diagnoses, with the largestgroup being those with stroke.

It is important to note that samplesizes were comparable across 7 ofthe 8 mDGI tasks. However, 100cases were lost in the analyses of thestairs task because not all clinics had10 steps available for the standard-ized conditions of the task. Analyses

Modified Dynamic Gait Index

November 2013 Volume 93 Number 11 Physical Therapy f 1497 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

for the stairs task and total scoresreflect this decrease in number ofcases.

Conversion of Time toOrdinal ScoresPrior to conducting the performancefacet analysis, the interval time scale

was converted to an ordinal scalethat could be incorporated intofactor analyses and IRT analyses.Table 3 summarizes the distributionof time and speed for the ordinalscores related to time across theeight DGI tasks. Time ranges asso-ciated with the ordinal score of 3

are fairly consistent across alltasks, with the exception of chang-ing pace. Task-specific differences intime ranges (and speed) were pres-ent for the lower ordinal scores(0–2).

ReliabilityInterrater agreement at the tasklevel. Before the analysis relatedto performance facets could be con-ducted, it was important to obtainevidence to support the reliabilityof task scores through interrateragreement analysis. For all tasks,with the exception of stairs, the per-cent exact agreement between phys-ical therapists and the criterionscores was high for both time andLOA, ranging from 94% to 100%.Percent exact agreement for thestairs task was 75% for time and 97%for LOA. Percent exact agreementwas lower, although still acceptable,across all tasks for GP scores, rangingfrom 71% to 91%. Raters consistentlyrated GP slightly higher than the cri-terion scores. Kappa coefficientsalso were high for all 3 facets of scor-ing. Kappa coefficients ranged from.90 to .98 for time, from .59 to .88 forGP, and from .84 to 1.00 for LOA.These data suggest that task scoreswere sufficiently reliable to be usedto investigate facets of performanceof mDGI.

Interrater agreement of perfor-mance facet scores. Interrateragreement also was strong for all 3performance facet scores. For timetotal scores, the mean criterion scorewas 14.0 (SD�8.61) compared withthe raters’ mean score of 14.1(SD�8.3); the correlation betweenthese scores was r�1.0. For the GPtotal score, the criterion mean was15.8 (SD�7.2) compared with theraters’ mean score of 16.96 (SD�6.2);the correlation between these scoreswas r�.94. The average GP totalscore for the physical therapists was1.2 points higher than the mean cri-terion GP total score, and the stan-

Table 2.Sociodemographics of Sample (N�995)a

Measure Data

Age (y)

X 65.3

SD 18.0

Range 15–99

Sex, n (%)

Female 542 (54.5)

Male 453 (45.5)

Ethnicity, n (%)

White 845 (84.9)

African American 80 (8.0)

Latino 40 (4.0)

Asian 26 (2.6)

Other 2 (0.2)

Diagnosis, n (%)

Stroke 239 (24.0)

Vestibular dysfunction 140 (14.1)

Closed head injury, traumatic brain injury, brain tumor 100 (10.1)

Gait abnormality, imbalance, falls, geriatric weakness 91 (9.1)

Parkinson disease 84 (8.4)

Neuropathy 41 (4.1)

Multiple sclerosis 31 (3.1)

Brain tumor 24 (2.4)

Spinal cord injury 22 (2.2)

Cerebellar ataxia 10 (1.0)

Other 97 (9.7)

Control cohort 140 (14.1)

Gait device, n (%)

None 446 (47.4)

Cane 213 (22.7)

Walker 281 (29.9)

Falls, n (%)

None 446 (47.4)

1 213 (22.7)

2 or more 281 (29.9)

a Subtotals of less than 995 are due to missing data.

Modified Dynamic Gait Index

1498 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

dard deviation of the physical thera-pists’ GP scores was 1 point smallerthan that of the criterion GP scores.It appears from these data that thephysical therapists judged partici-pants to have better GPs with nar-rower distribution of scores com-pared with the criterion scores.Finally, the criterion mean scorefor LOA was 12.56 (SD�4.94), andthe raters’ mean score was 12.91(SD�4.62) with a correlation ofr�.90.

Test-retest reliability. To investi-gate test-retest reliability, 257 partic-ipants were retested an average of3.65 days (SD�5.20) after initial test-ing. The Pearson correlation for test-retest scores for time, GP, and LOAwas .91, .91, and .87, respectively.The test-retest correlation for theDGI total score was .92. Test-retestcorrelations for task total scoresranged from .86 to .90. These datasuggest that scores for the mDGIwere quite stable over time.

Performance Facet AnalysesFactor analysis. Table 4 presentsthe factor pattern matrix from thefactor analysis. Three factors witheigenvalues greater than 1.3 explained79% of the variability in scores.Factor 1 included all of the time-level scores (shaded scores), fac-

Table 3.Cutpoints (Time and Speed) for Time Levels for the 8 Dynamic Gait Index Tasks

MeasureUsualPace

ChangePace

HorizontalHead Turns

VerticalHead Turns

PivotTurn

SteppingOver Obstacles

SteppingAround Obstacles Stairs

Time (s)

3 �6.0 �4.9 �6.2 �6.0 �6.9 �6.0 �6.0 �6.1

2 6.0–7.6 4.9–6.8 6.2–8.5 6.0–8.2 6.9–9.4 6.0–8.5 6.0–8.2 6.1–9.0

1 7.7–15.2 6.9–11.7 8.6–14.5 8.3–13.9 9.5–16.9 8.6–17.4 8.3–14.5 9.1–19.7

0 �15.2 �11.7 �14.5 �13.9 �16.9 �17.4 �14.5 �19.7

Speed (m/s)

3 �1.01 �1.24 �0.98 �1.01 �0.88 �1.02 �1.03 �1.0

2 0.8–1.01 0.90–1.24 0.71–0.98 0.74–1.01 0.65–0.88 0.72–1.02 0.74–1.03 0.68–1.0

1 0.4–0.79 0.52–0.89 0.42–0.70 0.44–0.73 0.36–0.64 0.35–0.71 0.42–0.73 0.31–0.67

0 �0.4 �0.52 �0.42 �0.44 �0.36 �0.35 �0.42 �0.31

Table 4.Factor Pattern Matrix for Modified Dynamic Gait Index Task Scores

Measure Time Assistance Gait Pattern

Usual pace, time level .951 .023 �.022

Usual pace, gait pattern .126 �.058 .714

Usual pace, level of assistance .021 �.974 �.075

Change pace, time level .933 �.003 �.014

Change pace, gait pattern .201 �.045 .656

Change pace, level of assistance .011 �.993 �.075

Horizontal head turns, time level .887 .003 .028

Horizontal head turns, gait pattern �.096 .044 .912

Horizontal head turns, level of assistance �.066 �.943 .061

Vertical head turns, time level .905 .011 .035

Vertical head turns, gait pattern �.096 .014 .926

Vertical head turns, level of assistance �.063 �.930 .071

Pivot turn, time level .890 .013 .033

Pivot turn, gait pattern .107 �.099 .666

Pivot turn, level of assistance �.051 �.969 .021

Stepping over obstacles, time level .913 �.023 .007

Stepping over obstacles, gait pattern .341 �.142 .453

Stepping over obstacles, level of assistance .114 �.806 .012

Stepping around obstacles, time level .933 �.010 �.019

Stepping around obstacles, gait pattern .136 �.039 .684

Stepping around obstacles, level of assistance �.024 �.980 �.034

Stairs, time level .800 �.068 .026

Stair, gait pattern .392 �.156 .295

Stairs, level of assistance .117 �.698 .080

Modified Dynamic Gait Index

November 2013 Volume 93 Number 11 Physical Therapy f 1499 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

tor 2 included all of the LOAscores (shaded scores), and factor 3included all the GP scores (shadedscores) except for scores from thestairs and the over obstacles tasks.Correlations among the factors weremoderate. The Pearson correlationbetween time-level scores and LOAscores was .61; the correlationbetween time-level scores and GPscores was .70, and the correlationbetween GP and LOA scores was .57.After rotation, each factor explainedapproximately 11% to 13% of unique

variance. These results suggest thatthe mDGI measured 3 different butcorrelated facets of performance.The unique and relatively equal vari-ance explained by each facet sug-gested support for an additive modelfor generating the mDGI total score.

Rasch analysis. To evaluate rela-tive difficulty of task scores andscore dimensionality, all scores wereanalyzed together in a single analysis.Two strategies were used duringthe analysis. First, item scores for GP,

time, and LOA were grouped sepa-rately prior to analysis so that an RSMcould be established for each perfor-mance facet. For the second analysis,step parameters were allowed tovary across all scores using a PCM.During this stage of analysis, it wasdetermined that an RSM was a betterfit to the data; therefore, all subse-quent analyses reflect an RSM.

The Figure presents the relative dif-ficulty of each score when all facetsof performance are combined into a

Figure.Rasch person-item map. The underlying scale is represented as a vertical column of plus signs and vertical bars. Plus signs representinteger locations on the underlying scale. Vertical bars represent locations 0.25, 0.50, and 0.75 above whole positive numbers orbelow whole negative numbers. The scale ranges from negative 6.0 to 7.0. Item names are on the right-hand side of the scale. Personsare on the left-most column, with 7 people designated by a pound sign. Each location on the scale represents the point on the scalein which examinees have an equal probability of adjacent scores. For example, location �2.75 is the point on the scale whereexaminees have an equal probability of scoring 0 or 1 on the horizontal head turns gait pattern.

Modified Dynamic Gait Index

1500 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

single scale. In the Figure, the under-lying scale is represented as a verticalcolumn of plus signs and verticalbars. Plus signs represent integerlocations on the underlying scale.Vertical bars represent locations0.25, 0.50, and 0.75 above wholepositive numbers or below wholenegative numbers. The scale rangesfrom negative 6.0 to 7.0. Item namesare on the right-hand side of thescale. Persons are on the left column,with 7 people designated by a poundsign, and a dot representing 1 to 6

people. Each location on the scalerepresents the point on the scale inwhich examinees have an equalprobability of adjacent scores. Forexample, location �2.75 is the pointon the scale where examinees havean equal probability of scoring 0 or1 on the horizontal head turns GPscore. This person and item mapshows that the range of scores forthe mDGI is appropriate for individ-uals with a wide range of mobilityfunction and that there are very few

gaps in measurement along theRasch scale.

As shown in the Figure, when agroup of scores are located withinthe same scale band, they are consid-ered to have comparable level of dif-ficulty and thus are listed together.For example, earning a score of 3 fortime level on the pivot task is locatedin the same scale band as earninga score of 3 for time level on thechange of pace, vertical head turn,and stepping over obstacles tasks,

Figure.Continued

Modified Dynamic Gait Index

November 2013 Volume 93 Number 11 Physical Therapy f 1501 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

suggesting a comparable level of dif-ficulty. In contrast, earning a scoreof 3 for time level on pivot task islocated on a lower scale band than ascore for time level on stairs, suggest-ing a lower level of difficulty.

The Figure also shows the addedbenefits of measuring 3 facets of per-formance. As shown in the Figure, itis more difficult to earn a score of3 for time level on most tasks than itis to complete the task with a normalGP (score of 3). Thus, the additionof time extends the range of mea-surement of the original DGI (repre-sented by the range of scores of 0–3for GP for the 8 tasks) and, there-fore, capacity to discriminate levelsof mobility function. For example,among individuals who earn a scoreof 3 for GP on the usual pace task(item 1), the time-level score for thistask could conceivably range from 0to 3, providing further discrimina-tion among these individuals whowould have scored 3 on the originalDGI usual pace item.

When all task scores were combinedinto a single analysis, 3 scores hadinfit and outfit statistics greater than1.3 (GP scores for horizontal headturns, vertical head turns, and step-ping over obstacles). Two scores hadinfit statistics less than 0.70 (LOAscores for usual pace and change ofpace). One score (LOA on stairs)had an adequate infit statistic, butthe outfit statistic was greater than1.3. These incongruities in infit andoutfit statistics suggest that, althougha unidimensional scale (eg, DGI totalscore) is psychometrically accept-able and conceptually meaningful,further information about mobilityfunction is gained by examiningscores related to facets of perfor-mance (time, GP, and LOA), as wellas individual task scores, which carryinformation about mobility functionwith respect to demands in the 4environmental dimensions.

For the performance facet analyses,when scores for time level were ana-lyzed separately, all task scores fit 1RSM model except for the time-levelscores on the stairs task, suggestingthat individuals with impaired mobil-ity required more time to completethis task. For GP, all task scores fit1 RSM model. Finally, when LOAscores were analyzed together, 2scores did not fit the RSM model:LOA stepping over obstacles andLOA on stairs, suggesting that indi-viduals with impaired mobilityrequire more assistance for thesetasks. These results highlight the factthat environmental dimensions influ-ence the pattern of scores withineach facet of performance.

Internal Consistency Reliabilityfor Task, Performance Facet,and Total ScoresThe internal consistency of themDGI total score was quite strong,with an alpha coefficient of .97. Forthe 3 performance facet scores, thealpha coefficients also were high(.97 for time and LOA and .92 forGP), suggesting that the perfor-mance facet scores were internallyconsistent. The alpha coefficients forthe 8 DGI tasks ranged from .75 (hor-izontal head turns) to .85 (steppingover obstacles), suggesting moder-ately strong internal consistencywithin the tasks themselves. The factthat the alpha coefficients were notstronger supports the suggestionthat the 3 facets of performance pro-vide unique variance within eachtask.

Modified DGI Score Comparisonsfor Participants With MobilityImpairments and the ControlCohortTable 5 compares performance onthe mDGI (8 tasks, 3 performancefacets, and the total score) betweenparticipants with mobility impair-ments and those in the controlcohort. Differences in means forall scores were significantly lower

for the participants with impairedmobility than for the control partici-pants (P�.001 for all comparisons).Effect sizes (d) suggest that all differ-ences were moderate (0.50�d�0.80) to large (d�0.80) except forthe LOA performance score (d�0.45).Given the fact that an inclusion cri-terion for both groups was the abilityto walk without the physical assis-tance of another on the usual pacetask, the lower effect size for LOAwas not surprising.

Minimal detectable change at the95% level of confidence for eachgroup was as follows: for the controlcohort, the mDGI total score was 4,the time total score was 2, the GPtotal score was 2, and the LOA totalscore was 1. For the mobility-impaired cohort, the MDC for themDGI total score was 5, the timetotal score was 2, the GP total scorewas 3, and the LOA total score was 2.

Relationship Between OriginalDGI and mDGICorrelations (r) between the originalDGI scores and the mDGI GP scoresranged from .90 to .99, with amedian correlation of .97. Correla-tions (r) between the original DGIscores and mDGI time-level itemscores ranged from .56 to .79 witha median correlation of .63. Correla-tions (r) between the original DGIitem scores and mDGI LOA scoresranged from .49 to .68, with amedian correlation of .53. The cor-relation (r) between original DGItotal scores and mDGI GP score was.99, suggesting that the original DGIscores captured primarily GP infor-mation. Correlations (r) betweenoriginal DGI total scores and mDGItime-level and LOA performancescores were .81 and .68, respec-tively. The Pearson correlationbetween DGI total scores and mDGItotal scores was .92.

These analyses, including factor anal-ysis, Rasch analysis, internal consis-

Modified Dynamic Gait Index

1502 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

tency, and correlation between orig-inal DGI and mDGI scores, suggestthat, although the original DGI itemscoring rubrics referred to time andLOA, the original DGI scores maynot have captured these 2 aspectsof performance. Instead, the originalDGI scores appear to have capturedprimarily information on GP.

DiscussionThe purpose of this study was todevelop and examine the psycho-metric properties of a new scoringsystem for the DGI. We expectedthat expanding the parameters forevaluating performance would enableclinicians to reliably monitor changein mobility in patients with a widerrange of function and to quantifyimprovement or deterioration in any1 of the 3 facets of performance (ie,GP, LOA, and time taken to completethe 8 walking tasks). The factor anal-ysis identified 3 separate factors—time, LOA and GP—that measuredunique but correlated aspects ofwalking performance. Rasch analy-ses verified that the mDGI allowedmeasurement of performance over awider range of mobility function

than the original DGI and supportedthe hierarchical structure of theitem tasks. The Rasch analysis showsgood measurement at all locationson the underlying scale. Previousstudies suggest that the original DGIscores resulted in a ceiling effectfor individuals with mobility impair-ments who were relatively highfunctioning, as well as fairly sizeablebreaks between scores, suggestinggaps in measurement.6–8,15 The orig-inal purpose for developing a revisedscoring system was to improve mea-surement capability of the DGI byeliminating both the ceiling effectsand the gaps in measurement. Theanalyses presented suggest thatthese goals have been achieved inthe mDGI. The mDGI showed soundpsychometric properties, includingstrong interrater reliability, strongtest-retest reliability, good internalconsistency, and evidence for dis-criminative validity at the individualtask, performance, and total scorelevels.

The major strength of this study wasthe large and diverse sample of par-ticipants, enabling the reliable use

of Rasch analysis to validate a newscoring system for the DGI. A total of995 adults participated (855 patientswith neurologic deficits currentlyreceiving physical therapy for mobil-ity deficits and 140 individuals with-out neurologic impairment). Only 2studies have used Rasch and factoranalyses to investigate propertiesof the original DGI. Chiu and col-leagues6 used a Rasch analysis to val-idate the 4 rating scale categoriesand to verify the single constructand hierarchical order of the scale.A limitation of that study was therelatively small and homogeneoussample (140 male veterans withimpaired balance); thus, there werenot sufficient cases for considerationof step parameters in their Rasch dif-ficulty estimations. This may havebeen the rationale for collapsing theDGI task scoring categories from 4to 2 levels. The study by Marchettiand Whitney15 included 123 partici-pants with balance and vestibulardisorders and 103 control partici-pants to construct and validate a4-item DGI using Rasch and factoranalyses. That study also was limitedby the relative homogeneity and size

Table 5.Description of Participants’ Performance on Modified Dynamic Gait Index (mDGI)

Measure

Mobility Impaired Group Control Group

dn Mininum Maximum X SD n Minimum Maximum X SD

mDGI total score 698 0 64 40.74 14.53 117 13 64 53.16 14.01 .87

Facets of performance scores

Time score 701 0 24 12.30 6.49 117 2 24 18.44 6.34 .96

Gait pattern score 702 0 24 15.94 4.78 118 5 24 20.10 5.16 .84

Level of assistance score 705 0 16 12.52 5.08 118 3 16 14.48 3.38 .45

Task scores

Usual pace 857 0 8 5.55 1.73 138 2 8 6.84 1.59 .78

Change pace 856 0 8 5.38 1.88 138 1 8 6.83 1.61 .83

Horizontal head turns 856 0 8 4.95 1.91 138 1 8 6.46 1.93 .79

Vertical head turns 857 0 8 5.08 1.92 138 1 8 6.46 2.08 .69

Pivot turn 857 0 8 5.22 1.89 138 1 8 6.53 1.89 .69

Stepping over obstacles 854 0 8 5.03 2.23 136 0 8 6.48 2.33 .64

Stepping around obstacles 856 0 8 5.53 1.80 138 2 8 6.92 1.50 .84

Stairs 702 0 8 4.75 2.19 118 0 8 6.31 2.36 .69

Modified Dynamic Gait Index

November 2013 Volume 93 Number 11 Physical Therapy f 1503 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

of the samples, which may havebeen the reason that only locationparameters (average of the step dif-ficulties) for the tasks were used.The current study has addressedboth of these limitations throughthe heterogeneity of the samples(participants representing a varietyof mobility disorders as well as par-ticipants without neurologic impair-ment) and the size of the samples,making the item and step difficultyestimates quite stable.

Rating ScaleThe current study suggests that therating scale developed for the origi-nal DGI captures the performancefacet of GP alone. In contrast, themodified scoring system, whichincludes time, GP, and LOA, mea-sures distinct but correlated facets ofperformance that operate togetheras an individual completes eachtask. These 3 facets of performance,although distinct, are nonethelesscorrelated and thus can be incorpo-rated into a single scale. Individually,each performance facet contributesabout 11% of unique variance, andtogether they explain about 80%of the total variance in scores.The new scoring system will enableclinicians to measure changes in 3aspects of walking function: theGP used for a walking task, theLOA needed, and the time requiredto complete each of the tasks. Mea-suring time and level of assistancein addition to GP is important inlight of research indicating that formany patients, improvement inwalking function is not always asso-ciated with changes in underlyingGP.30,31 Thus, locomotor rehabilita-tion may improve some aspects ofwalking performance, such as timeor LOA, but not others, such as thepattern used to walk. The new scor-ing system for the mDGI also willallow clinicians and researchers toexamine the association betweenchanges in these 3 facets of perfor-mance and changes in walking func-

tion. Finally, expanding the facets ofmeasurement expands the range ofability measured by the scale, thusreducing the likelihood of a ceilingeffect.

Unidimensional and HierarchicalOrder of TasksThe studies by Chiu et al6 and Mar-chetti and Whitney15 determinedthat the original DGI was unidimen-sional and measured a single factorreferred to as “dynamic balance.”These studies reported a hierarchicalorder of task difficulty, using the per-formance facet of GP. The specificorder of tasks based on difficulty var-ied slightly, probably due to differ-ences in samples tested. Both studiesconfirmed that the 3 most difficulttasks were walking with horizontalhead turns, walking up and downstairs, and walking with vertical headturns; the order of remaining tasksvaried by study. The 3 least difficulttasks were walking on level sur-faces,6 walking around obstacles,15

and walking and changing speeds.6,15

Both studies recommended adminis-trative changes in the DGI, includingshortening the test to eliminate psy-chometric redundancy and the needfor equipment or reordering and pos-sibly eliminating some tasks basedon both psychometric redundancyand hierarchical ordering.

Although the DGI was originallydesigned to capture the constructof dynamic balance, it was based ona conceptual framework that recog-nized the multidimensional natureof dynamic balance, including theability to adapt the sensory, motor,and cognitive systems controllingbalance and gait to specific demandswithin distinct environmental dimen-sions.1,2 The 8 tasks in the DGI weredesigned to assess the capacity toadapt gait to demands within 4 envi-ronmental dimensions (temporal,postural transitions, terrain, and den-sity) and thus to specify the constel-lation of environmental dimensions

contributing to mobility disability inan individual patient. This informa-tion was considered critical todeveloping targeted interventionsdesigned to reduce or reverse mobil-ity disability. Item response theoryidentifies score patterns based onlevel of difficulty; however, redun-dancy in measurement is not thesame as redundancy in construct.Just as a mathematics test mustassess all dimensions of the con-struct of mathematics (eg, algebra,geometry, probability, and statistics)to provide diagnostic informationabout students, each of the DGI taskswas designed to assess walking inrelationship to different environ-mental dimensions, thus providinginsight into the determinants ofmobility disability within an individ-ual patient. Although there may beredundancy in the level of Raschdifficulty of the scores, it does notmean that tasks or performancescores should be dropped. If 2 tasksdemonstrate the same level of diffi-culty but capture different dimen-sions of gait adaptation, they shouldbe retained because unique patternsof scores can help therapists identifyan individual patient’s response inthe different environmental dimen-sions that affect walking performance.

It is important to note that the Raschstep parameters represent theoreti-cal rather than empirical results andthe item difficulties found in thisstudy and previous studies representexpected rather than absolute differ-ences in difficulty for task scores.An individual patient could have ascore pattern that deviates fromthe theoretical model, which is animportant distinction when selectingthe appropriate therapy and measur-ing the success of therapy, particu-larly given the environmental dimen-sions of mobility function measuredby mDGI.

In addition, score patterns may beunique for different populations,

Modified Dynamic Gait Index

1504 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

based on the specific constellationof sensory, motor, and cognitiveimpairments that limit adaptationsto specific task and environmentaldemands. Previous research hasshown individuals with impairedmobility report avoiding specific fea-tures in the environment perceivedas challenging to safety when walk-ing in the community.32,33 How-ever, level of avoidance varies byenvironmental dimension; thus, notall aspects of the environment areperceived as equally challenging tosafe walking. Among people withstroke, avoidance in the temporaland postural dimensions was signifi-cantly associated with frequency ofcommunity walking.32 In contrast,avoidance in the distance, temporal,terrain, and postural dimensions wasassociated with frequency of partici-pation in community walking inolder adults with mobility disability.33

Even the fit statistics reflect theuniqueness of different tasks in time,GP, and LOA needed. The pattern offit statistics for the unidimensionalRasch analysis can be explained, inpart, by the different environmentaldimensions represented within theDGI. The LOA scores for 2 tasks(usual pace and change pace) had fitstatistics less than the criterion of.70. These tasks reflect the temporaldimension of the DGI. Because ofthe nature of these tasks, theyrequire less assistance than any ofthe other tasks in the other environ-mental dimensions. The GP scoresfor vertical head turns, horizontalhead turns, stepping over obstacles,and stairs tasks had fit statisticsgreater than the criterion of 1.30.These tasks reflect the postural, den-sity, and terrain dimensions of theDGI. The GP for each of these taskswould be expected to differ from theGPs used in the other 4 tasks. Thepattern of fit statistics for the Raschanalyses of the 3 facets of perfor-mance scores also pick up theunique contributions of each task.

Within the time facet, time-levelscores on the stairs task had infit andoutfit statistics that suggest the stairstask affects time-level scores differ-ently than other tasks. For the Raschanalysis of the LOA scores, fit statis-tics for the stepping over obstaclestask and for the stairs LOA differ fortasks measuring the terrain environ-mental dimension. These results pro-vide support for retaining the uniquecontributions of the different tasksfrom the original DGI in the mDGI.

LimitationsThe main limitations in this studywere the uneven sample sizesbetween the 2 groups and the sam-pling process. Individuals in thecontrol cohort were drawn fromamong volunteers between theages of 15 and 99 years, with noknown neurologic diagnosis. Thesample size of 140 was relativelysmall to be truly representative ofthis population. Participants withmobility impairments were recruitedfrom patients with neurologic diag-noses actively receiving physicaltherapy from clinical sites alreadyusing the DGI as part of their clini-cal assessment protocols. Therefore,these patients may not represent thebroader population of individualswith mobility impairments. Anotherlimitation was that a number of clin-ical sites did not have 10 steps forthe stair task; therefore, 100 caseswere lost in the analyses of the stairstask and the total score analyses.

Future ResearchThe next phase of analysis in thisstudy will focus on differential anal-ysis of the mDGI on the differentsample populations to determinewhether order of score difficulty var-ies in patients with different pathol-ogies. Future research is needed todetermine the responsiveness of thenew measure and the relationshipbetween the mDGI and fall risk.

ConclusionsThe mDGI retains the 8 items fromthe original test but modifies thescoring system, enabling measure-ment over a broader range of mobil-ity function and allowing cliniciansto characterize changes in 3 facets ofperformance: time, LOA, and GP.The expanded range and specificityof measurement should improvethe utility of the DGI as an outcomemeasure. We believe the strength ofthe psychometric properties of themDGI warrant its adoption for bothclinical and research purposes.

All authors provided concept/idea/researchdesign. Dr Shumway-Cook, Dr Taylor, andDr Matsuda provided writing and data anal-ysis. Dr Shumway-Cook, Dr Matsuda, MrStuder, and Dr Whetten provided datacollection. Dr Shumway-Cook provided fundprocurement. Mr Studer and Dr Whettenprovided study participants. Mr Studer pro-vided facilities/equipment. Dr Matsuda pro-vided institutional liaisons. Dr Taylor providedconsultation (including review of manuscriptbefore submission). The authors thank thetesting sites, physical therapists, and all par-ticipants for their contributions to this study.

All procedures were approved by the Univer-sity of Washington Human Subjects Division.

A summary of this research was presented inposter format at the Combined SectionsMeeting of the American Physical TherapyAssociation; January 21–24, 2013; SanDiego, California.

This study was supported by a grant fromthe Walter C. and Anita C. Stolov ResearchFund, University of Washington.

DOI: 10.2522/ptj.20130035

References1 Shumway-Cook A, Woollacott M. Motor

Control: Translating Research Into Clini-cal Practice. 4th ed. Baltimore MD: Lippin-cott Williams & Wilkins; 2011.

2 Patla A, Shumway-Cook A. Dimensions ofmobility: defining the complexity and dif-ficulty associated with community mobil-ity. J Aging Phys Act. 1999;7:7–19.

3 Shumway-Cook A, Gruber W, Baldwin M,Liao S. The effect of multidimensionalexercises on balance, mobility, and fallrisk in community-dwelling older adults.Phys Ther. 1997;77:46–57.

Modified Dynamic Gait Index

November 2013 Volume 93 Number 11 Physical Therapy f 1505 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from

4 Badke MB, Sherman J, Boyne P, et al.Tongue-based biofeedback for balance instroke: results of an 8-week pilot study.Arch Phys Med Rehabil. 2011;92:1364–1370.

5 Herman T, Inbar-Borovsky N, Brozgol M,et al. The Dynamic Gait Index in healthyolder adults: the role of stair climbing, fearof falling, and gender. Gait Posture.2009;29:237–241.

6 Chiu Y, Fritz SL, Light KE, Velozo CA. Useof item response analysis to investigatemeasurement properties and clinical valid-ity and data for the Dynamic Gait Index.Phys Ther. 2006;66:779–787.

7 Pardasaney PK, Latham NK, Jette AM, et al.Sensitivity to change and responsivenessof four balance measures for communitydwelling older adults. Phys Ther. 2012;92:388–397.

8 Jonsdottir J, Cattaneo D. Reliability andvalidity of the Dynamic Gait Index in per-sons with chronic stroke. Arch Phys MedRehabil. 2007;88:1410–1415.

9 Lin JH, Hsu MJ, Hsu HW, et al. Psychomet-ric comparisons of three functional ambu-lation measures for patients with stroke.Stroke. 2010;41:2021–2025.

10 Huang S-L, Hsieh C-L, Wu R-M, et al. Min-imal detectable change of the Timed “Up& Go” Test and the Dynamic Gait Index inpeople with Parkinson disease. Phys Ther.2011;91:114–121.

11 McConvey J, Bennett SE. Reliability of theDynamic Gait Index in individuals withmultiple sclerosis. Arch Phys Med Reha-bil. 2005;86:130–133.

12 Cattaneo D, Jonsdottir J, Repetti S. Reli-ability of four scales on balance disordersin persons with multiple sclerosis. DisabilRehabil. 2007;29:1920–1925.

13 Cattaneo D, Regola A, Meotti M. Validity ofsix balance disorders scales in personswith multiple sclerosis. Disabil Rehabil.2006;28:789–795.

14 Wrisley DM, Walker ML, Echternach JL,Strasnick B. Reliability of the Dynamic GaitIndex in people with vestibular disorders.Arch Phys Med Rehabil. 2003;84:1528–1533.

15 Marchetti GF, Whitney SL. Constructionand validation of the 4-item Dynamic GaitIndex. Phys Ther. 2006;86:1651–1660.

16 Shumway-Cook A, Baldwin P, Polissar N,Gruber WM. Predicting the probability forfalls in community dwelling older adults.Phys Ther. 1997;77:812–819.

17 Dibble LE, Lange M. Predicting falls in indi-viduals with Parkinson disease: a recon-sideration of clinical balance measures.J Neurol Phys Ther. 2006;30:60–67.

18 Said CM, Goldie PA, Culham E, et al. Con-trol of lead and trail limbs during obstaclecrossing following stroke. Phys Ther.2005;85:413–427.

19 Perry J, Garrett M, Gronley JK, Mulroy SJ.Classification of walking handicap in thestroke population. Stroke. 1995;26:982–989.

20 Bohannon RW. Comfortable and walkingspeed of adults aged 20–79 years: refer-ence values and determinants. Age Aging.1997;26:15–19.

21 Perera S, Mody SH, Woodman RC, Studen-ski SA. Meaningful change and responsive-ness in common physical performancemeasures in older adults. J Am GeriatrSoc. 2006;54:743–748.

22 Schmid A, Duncan PW, Studenski S, et al.Improvements in speed-based gait classifi-cations are meaningful. Stroke. 2007;38:2096–2100.

23 Studenski S, Perera S, Patel K, et al, Gaitspeed and survival in older adults. JAMA.2011;305:50–58.

24 Cohen J. A coefficient of agreement fornominal scales. Educational and Psycho-logical Measurement. 1960;20:37–46.

25 Wright BD, Linacre JM. Reasonable mean-square fit values. Rasch MeasurementTransactions. 1994;8:370–376.

26 Linacre JM. Winsteps help for Rasch anal-ysis: misfit diagnosis. Winsteps Manual.2005. Available at: http://www.winsteps.com/winman/index.htm?diagnosingmisfit.htm. Accessed January 3, 2013.

27 Masters GN. A Rasch model for partialcredit scoring. Psychometrika. 1982;47:149–174.

28 Andrich D. A rating formulation forordered response categories. Psycho-metrika. 1978;43:561–573.

29 Wright BD, Masters GN. Rating ScaleAnalysis. Chicago, IL: MESA Press; 1982.

30 Den Otter AR, Geurts AC, Mulder T,Duysens J. Gait recovery is not associatedwith changes in the temporal patterningof muscle activity during treadmill walkingin patients with post-stroke hemiparesis.Clin Neurophysiol. 2006;117:4–15.

31 Burke JH, Nene AV, Kwakkel G, et al.Recovery of gait after stroke: whatchanges? Neurorehabil Neural Repair.2008;22:676–683.

32 Robinson C, Matsuda PN, Ciol M,Shumway-Cook A. Participation in com-munity walking following stroke: the influ-ence of self-perceived environmental bar-riers. Phys Ther. 2013;93:620–627.

33 Shumway-Cook A, Patla A, Stewart A, et al.Environmental components of mobilitydisability in community-living older per-sons. J Am Ger Soc. 2003;51:393–398.

Modified Dynamic Gait Index

1506 f Physical Therapy Volume 93 Number 11 November 2013 by Evan Prost on November 15, 2013http://ptjournal.apta.org/Downloaded from