Stroke is the leading cause of disability in the United States, affecting approximately 5.6 million people in the United States, with an estimated 700,000 Americans experiencing a new or recurrent stroke each year.1 More than 1 million Americans with stroke report difficulties with basic activities of daily living (ADL) due to their stroke, and many also experience significant difficulty with mobility.1 Initially after stroke, two thirds of individuals are not able to walk or require assistance to walk.2 After three months, one third of individuals who experience a stroke still require assistance or are not able to walk.2 Many of these individuals will require rehabilitative services in order to optimize their level of recovery. It is therefore important for physical therapists to have useful and clinically meaningful measures to detect change in walking ability.
Gait speed measured over a short distance, ie, five or 10 meters, is one of the most widely used methods of measuring walking ability both in the clinic and in research.3,4 Overground gait speed can be measured in practically any setting and at all stages of recovery after stroke and has high clinical utility. The validity of gait speed as a method for measuring walking ability has been extensively studied.5–12 Gait speed has been found to be moderately to strongly related to paretic lower limb muscle performance.5,7,9 daCunha and colleagues6 found that gait speed was strongly related to energy expenditure and energy cost of gait. Gait speed is strongly related to balance after stroke as measured by the Berg Balance Scale.8 Perry and colleagues10 used gait speed to categorize individuals with stroke into different levels of home and community walking ability. Gait speed can also be used to predict discharge destination from inpatient rehabilitation after stroke.11
The reliability of gait speed has also been investigated by other authors.13–16 Test-retest reliability coefficients reported in the literature range from 0.9213 to 0.97.15 The variability associated with measuring gait speed has also been reported. In a group of patients who could walk independently without assistive devices undergoing rehabilitation, Evans and colleagues13 reported an estimate of error for measuring gait speed of −0.11 m/sec to +0.17 m/sec (95% confidence interval). Hill et al14 and Stephens et al15 reported similar findings with 95% confidence intervals for error estimates between −0.08 m/sec and +0.16 m/sec and −0.13 m/sec and +0.17 m/sec, respectively. In people with chronic stroke (>6 months), Flansbjer and colleagues16 reported a 95% confidence interval for the smallest real difference in gait speed of −0.15 m/sec to +0.25 m/sec.
The minimal detectable change (MDC) values reported above are important for both clinicians and researchers. For example, suppose a physical therapist is providing interventions designed to improve walking ability with a client with a chronic stroke and the physical therapist is using gait speed as an outcome measure to assess walking ability. Using the results from the study by Flansbjer et al,16 in order to be 95% confident that a true change in walking ability occurred after physical therapy, the client’s gait speed needs to increase by at least 0.25 m/sec. If the change in gait speed is <0.25 m/sec, then the change is likely due to error in the measurement and not a true change in walking ability.
MDC values are also important to consider when interpreting the results of clinical trials. Researchers often compare mean change values in gait speed between experimental and control groups. For example, Ada and colleagues17 explored the effect of a treadmill and overground walking program in community-dwelling people with chronic stroke. The mean change in gait speed after the four-week intervention for the group who received treadmill and overground walking training was 0.18 m/sec (standard deviation [SD] = 0.19) compared to 0.04 m/sec (SD = 0.07) in the control group who received a home exercise program. Although there was a statistically significant difference between the two groups, it is difficult to determine whether the change in gait speed in the experimental group was clinically significant. Since the mean change was 0.18 m/sec and the SD was 0.19 m/sec, some of the subjects did not exceed 0.25 m/sec. Thus, for at least some of the subjects, the change in gait speed did not exceed the potential error in the measurement.
In other cases, a small change in gait speed may be clinically important. Plummer et al18 examined the effect of locomotor training using a body weight support and treadmill system in seven people with subacute stroke. After 36 sessions, one subject’s gait speed improved from 0.18 m/sec to 0.29 m/sec, an improvement of 0.11 m/sec. While this change does not exceed the measurement error reported by Evans et al13 (0.17 m/sec), this change in gait speed may have indicated that the subject improved from being a physiologic ambulator to a full-time home ambulator as defined by walking categories developed by Perry and colleagues.10
When examining change scores in gait speed in both clinical practice and with research results, it is to important use MDC values that have been established in subjects with similar characteristics. A limitation of the findings described above to clinical practice is that the subjects in these studies were included based on their ability to walk without physical assistance13–16 or assistive devices13 or were in the chronic stages of recovery after their stroke.15,16 Since many of the patients who are undergoing rehabilitation after a stroke require physical assistance to walk and/or an assistive device, it is important to examine the reliability of gait speed in this population. The purpose of this study was to examine the test-retest reliability of gait speed in individuals undergoing rehabilitation after stroke who require varying levels of assistance to walk with varying locations of stroke.
Participants were recruited from two inpatient rehabilitation facilities. Inclusion criteria were undergoing inpatient rehabilitation after stroke and being able to ambulate with at most maximal assistance of one person. The level of physical assistance was defined as follows: maximal assistance is when the participant expends <50% of the effort but >25%; moderate assistance is when the participant expends between 50% and 74% of the effort; and minimal assistance is when the participant expends between 75% and 99% of the effort; any physical contact by another person while walking was considered at least minimal assistance. Participants were excluded if they had a premorbid condition that affected their ability to walk independently prior to their stroke such as severe rheumatoid arthritis or Parkinson’s disease. Diagnosis and location of stroke were confirmed by reviewing the participants’ medical record and imaging study report if available. All participants provided informed consent, and the institutional review boards of all the involved facilities approved the study.
Four physical therapists and two physical therapist assistants who were working with the patients served as the raters. Experience of the raters ranged from six months to 15 years. None of the raters had experience using gait speed as an outcome measure prior to this study, but all underwent training in the data collection procedure described below.
Gait speed was measured over the middle five meters of a nine-meter walk using a stopwatch.19 Patients were instructed to walk at their comfortable pace. Patients started walking two meters before the start line and were timed from the moment their lead foot crossed the start line until the front foot crossed the finish line, five meters away. They continued walking for another two meters after crossing the finish line. Patients walked the additional two meters in front and at the end of the timed five meters to take into account acceleration and deceleration effects. The amount of physical assistance, assistive device, and/or orthotic used was determined by the patient’s primary physical therapist. Patients underwent two measurement sessions separated by one to three days during the week prior to their discharge from the inpatient rehabilitation hospital. One trial during each of the two sessions was used to determine the subject’s gait speed. If the amount of physical assistance, use of assistive device and/or orthotic, or Functional Independence Measure (FIM) locomotion score changed between the two measurement sessions, then the subject was excluded from study.
Test-retest reliability was assessed using the intraclass correlation coefficient (ICC2,1) with 95% confidence intervals as described by Shrout and Fleiss,20 and MDC at the 90% confidence level (MDC90) as described by Stratford21 and more recently by Haley and Fragala-Pinkham.22 The MDC90 determines the magnitude of change that must be observed before the change can be considered to exceed the measurement error and variability at the 90% confidence level. The MDC90 was calculated by determining the standard error of the measurement (SEM) using the formula (SD × √[1 − r]), where r is the test-retest reliability coefficient (ICC2,1 in this case) and SD is the SD of the measures.21,22 The SD was calculated from the combined gait speed measured in both sessions 1 and 2. The SEM was multiplied by 1.65 to determine the 90% confidence interval.21–23 The 90% confidence level was chosen as this appears to be most commonly reported in the literature.22 This value was multiplied by the square root of 2 to account for error associated with repeated measurements.21,22,24 The ICC is a reliability coefficient whose value ranges from 0.00 to 1.00. It is calculated using variance estimates from an analysis of variance; because of this, it reflects both the degree of correlation and level of agreement between measures. A correlation coefficient such as Pearson r only reflects the degree of the relationship between measurements and not the level of agreement.25
ICCs and MDC values were calculated for all subjects combined, for those who required physical assistance to ambulate, those who could walk without physical assistance, and those who used an assistive device to walk. Additionally, a Kruskal-Wallis one-way analysis of variance was performed to determine whether there were differences in gait speed between individuals with different locations of stroke, ie, cortical versus subcortical versus brainstem versus cerebellar stroke.
Thirty-five individuals agreed to participate. The mean age of the participants was 67.4 years (SD = 13.8), with a mean time of 34.5 days (SD = 17.7) post-stroke. The median score on the locomotion section of the FIM was 5 (range, 2–7). Thirteen of the 35 participants required physical assistance to walk, 10 required minimal assistance, and three required moderate assistance to walk. Twenty-two participants were able to ambulate without physical assistance. Twenty-eight individuals used an assistive device to walk (15 using a two-wheel rolling walker, four using a straight cane, four using a large-base quad cane, two using a small-base quad cane, two using a hemiwalker, and one using a standard walker) and six required an orthotic (ankle-foot orthosis [five], knee-ankle-foot orthosis [one]). Seventeen were diagnosed with a cortical stroke, eight with a subcortical stroke, seven with a brainstem stroke, and three with a cerebellar stroke. The Kruskal-Wallis test revealed no significant difference in gait speed among participants with different locations of stroke, H(3) = 2.10, P > 0.05.
The mean gait speed for the first session for all participants was 0.45 m/sec (SD = 0.30) and the mean for the second session was 0.54 m/sec (SD = 0.39). The ICC2,1 was 0.862 and the MDC90 was 0.30 m/sec. Mean gait speed, ICC2,1, and MDC90 results for all subjects and subgroups are presented in Table 1, raw data for gait speed for sessions 1 and 2 are presented in Figure 1.
Gait speed measured over a short distance in the clinic is a reliable measure of walking ability in people who are undergoing inpatient rehabilitation after stroke. The MDC90 of 0.30 m/sec indicates that the gait speed of 90% of persons with stroke demonstrating characteristics similar to those of the subjects in this study will vary by <0.30 m/sec. This implies that a change ≥0.30 m/sec is necessary in an individual patient in order to be 90% certain that the change is not due to intertrial variability. The ICC of 0.862 is indicative of good reliability.25 The MDC is a measure of sensitivity to change and is useful for interpreting change scores in individual patients, whereas the ICC calculated here is a test-retest reliability index, which is useful for examining the reliability of group data.
When selecting outcome measures for assessing change in individuals after stroke, clinicians and researchers should use tools with sound psychometric properties. However, outcome measurement can be difficult in people post-stroke due the heterogeneity of symptoms, variability in severity, and various etiologies.26 The results of this study can be used to measure change in walking ability in three different subgroups of patients post-stroke: patients who require physical assistance to walk (MDC90 = 0.07 m/sec), patients who can walk without physical assistance (MDC90 = 0.36 m/sec), and patients who use an assistive device to walk (MDC90 = 0.18 m/sec) who are two to six weeks post-stroke. When attempting to determine whether a true change in gait speed has occurred, researchers and clinicians should use MDC values established from subjects with similar characteristics. An area of further research would be to determine MDC values for other subpopulations of patients post-stroke, ie, patients who use an orthotic to walk.
One interesting finding from this study is that gait speed is more reliable and sensitive to change in individuals who require physical assistance to walk than in patients who can ambulate without physical assistance. The ICC2,1 for the patients who required physical assistance to walk was 0.971 and the MDC90 was 0.07 m/sec as compared to an ICC2,1 of 0.800 and an MDC90 of 0.36 m/sec in patients who could ambulate without physical assistance. There was greater variability in gait speed for participants who could walk without physical assistance. The SD for the two walks in these patients was 0.30 and 0.39 m/sec. In the patients who required physical assistance to ambulate, the variability between the two walks was much less (SD = 0.17 and 0.18 m/sec).
One possible explanation for this finding is that patients who require physical assistance to walk are not able to vary their gait speed by as much because they rely on another person to assist them. The physical therapist, or physical therapist assistant, who was working with the patient throughout his or her rehabilitation, provided the assist during the walk. Because of their familiarity with the patient, the therapist may have been able to consistently provide the amount of physical assistance required to walk, thus reducing variability in gait speed while walking. Patients who can ambulate without physical assistance are not limited by the physical assistance of another person. This may allow for greater variability in walking performance at this stage of recovery. Another possible explanation for this finding is that there is less capacity for change in gait speed in people who require physical assistance to walk after stroke. As a person progresses in his or her rehabilitation and is able to ambulate without physical assistance, there may be a much greater capacity for change in gait speed and this may be reflected in the larger MDC90 value.
The results of this study indicate that there is slightly more error and variability when measuring gait speed compared to previous studies.13–16 Evans and colleagues13 reported an ICC2,1 of 0.92 and a 95% confidence interval for error estimate of −0.11 m/sec to +0.17 m/sec. Hill et al14 reported an ICC2,1 of 0.95 and 95% confidence interval for error estimate of −0.08 m/sec to +0.16 m/sec. Stephens and Goldie15 reported an ICC2,1 of 0.97 and 95% confidence interval for error estimate of −0.13 m/sec to +0.17 m/sec. Flansbjer et al16 reported an ICC2,1 of 0.94 and a 95% confidence interval for the smallest real difference of −0.15 m/sec to +0.25 m/sec.
Two possible explanations for this discrepancy are the differences in the subject characteristics and the methods used to calculate the amount of variability in gait speed. In all the studies listed above, the subjects were capable of independent ambulation and were in a more chronic phase of recovery than the subjects in this study. The mean length of time post-stroke in these studies was 1.5 months,13 2.8 months,14 3.7 months,15 and 18 months, respectively.16 As more time elapses post-stroke, it is likely the more stable a person’s walking ability becomes. The Copenhagen Stroke Study reported that walking ability plateaus in 95% of patients by 11 weeks post-stroke.2 The two studies13,14 in which the time since stroke was nearest the time post-stroke in this study used a foot switch system to time the patients. This automated system may have less error than a hand-held stopwatch. Another automated data collection system that may improve accuracy of gait speed data collection is the GAITRite portable walkway system (CIR Systems, Inc., Havertown, PA). The GAITRite walkway has sensors embedded in a portable mat that can accurately gather temporospatial measures of gait such as speed, cadence, step length, and single limb support time. Although these automated systems (foot switch or GAITRite) may improve accuracy, they take time to gather data and require special equipment and training. The simple stopwatch method used in this study is easy to perform in the clinic and requires very little time or training to perform.
Three of the studies13–15 determined the 95% confidence interval for error estimate in gait speed by determining the mean change in gait speed from one trial to the next and the mean of the SD of the change scores. They multiplied the SDs of the change scores by 1.96 (Z score for 95% confidence level) and added or subtracted this from the mean change in gait speed to arrive at the 95% confidence interval for error estimate. By not taking into account the reliability coefficient and error associated with multiple measures, the error estimate may be underestimated.
Flansbjer and colleagues16 used the Bland Altman27 method of determining the smallest real difference (SRD) in gait speed. The SRD was defined as the smallest change that indicates real improvement or deterioration in gait speed.16 Their findings were the closest to the findings of this study, a 0.05-m/sec difference (0.30 m/sec compared to 0.25 m/sec). This difference is likely due to the fact that participants in the study by Flansbjer et al16 were 16 to 18 months post-stroke and were all independent community ambulators. The walking ability of people with chronic stroke is likely to be more stable and less variable than patients undergoing rehabilitation within one month after a stroke.
A potential limitation in the findings of this study is the time between successive measures of gait speed of one to three days was not standardized for each patient. This time range was chosen due to the time constraints of collecting data during clinical practice and other studies had used a similar time range.14,16 The measures were taken over the last week of their inpatient stay in an attempt to minimize any potential changes in walking ability over the one- to three-day time frame. However, during this one- to three-day period, it is possible that the gait speed of some participants may have actually improved. If this was the case, then this true change would be erroneously included in the measurement variability calculated by the MDC90 and the ICC2,1. Thus, the MDC90 would actually be less and the ICC2,1 would be greater than the data indicate.
However, these data were collected while the subjects were undergoing inpatient rehabilitation under conditions similar to physical therapy practice, which suggests that the results are appropriate for interpreting change in gait speed in patients with similar characteristics in the clinic. Despite any changes in walking ability that may have occurred, the results demonstrate that variability in gait speed is quantifiable for people undergoing rehabilitation in the first month post-stroke with different levels of walking ability. These results can be used to interpret change in walking ability in people with stroke who are undergoing active rehabilitation shortly after stroke.
Although the MDC is an important characteristic to know in order to determine whether the amount of change that occurred in a patient exceeded measurement error, it does not let us know whether the change was clinically meaningful.21,22 Further research is necessary to determine the minimal clinically important difference in gait speed.
The results of this study may be used to interpret findings from research studies that examine the efficacy of physical therapy interventions designed to improve walking ability after stroke in patients with similar characteristics as those in this study. In addition to presenting their findings in terms of statistically significant differences between groups, researchers could express the results in terms of the proportion of patients in the experimental group who exceeded the amount of change that is necessary in order to be 90% certain that the change is a true amount of change compared to the same proportion of subjects in the comparison group. From these percentages, the number needed to treat could be calculated, which may provide a more clinically relevant method of examining the differences between interventions strategies.28
Gait speed is a reliable measure of walking ability for a wide variety of patients undergoing rehabilitation after stroke. The results of this study can also be used to interpret change in walking ability in individual patients. Our results suggest that a change of >0.30 m/sec in patients undergoing inpatient rehabilitation after stroke may be necessary in order to determine whether a change in gait speed exceeds measurement error and patient variability. As an outcome measure, gait speed is more sensitive to change in patients who require physical assistance to walk than in patients who can walk without physical assistance during the first two to six weeks after stroke. The measurement error and variability associated with gait speed are only 0.07 m/sec in patients who require physical assistance to walk as compared to 0.36 m/sec in patients who are able to walk without physical assistance during the relatively early stages of rehabilitation. This information is useful for the clinician trying to interpret gait speed measures in patients two to six weeks post-stroke.
Funding for this research was provided by the New York Physical Therapy Association.
1.Rosamond W, Flegal K, Friday G, et al. Heart disease and stroke
statistics—2007 update: a report from the American Heart Association Statistics Committee and Stroke
Statistics Subcommittee. Circulation.
2.Jorgensen HS, Nakayama H, Raaschou HO, Olsen TS. Recovery of walking function in stroke
patients: the Copenhagen Stroke
Study. Arch Phys Med Rehabil.
3.Richards CL, Malouin F, Dean C. Gait in stroke
: assessment and rehabilitation. Clin Geriatr Med.
4.Richards CL, Olney SJ. Hemiparetic gait following stroke
. Part II: Recovery and physical therapy. Gait Posture.
5.Bohannon RW, Walsh S. Nature, reliability
, and predictive value of muscle performance measures in patients with hemiparesis following stroke
. Arch Phys Med Rehabil.
6.daCunha IT, Lim PA, Henson H, et al. Performance-based gait tests for acute stroke
patients. Am J Phys Med Rehabil.
7.Maeda A, Yuasa T, Nakamura K, et al. Physical performance tests after stroke
and validity. Am J Phys Med Rehabil.
8.Wolf SL, Catlin PA, Gage K, et al. Establishing the reliability
and validity of measurements of walking time using the Emory Functional Ambulation Profile. Phys Ther.
9.Hsu AL, Tang PF, Jan MH. Analysis of impairments influencing gait velocity and asymmetry of hemiplegic patients after mild to moderate stroke
. Arch Phys Med Rehabil.
10.Perry J, Garrett M, Gronley JK, et al. Classification of walking handicap in the stroke
11.Rabadi MH, Blau A. Admission ambulation velocity predicts length of stay and discharge disposition following stroke
in an acute rehabilitation hospital. Neurorehabil Neural Repair.
12.Roth EJ, Merbitz C, Mroczek K, et al. Hemiplegic gait. Relationships between walking speed and other temporal parameters. Am J Phys Med Rehabil.
13.Evans MD, Goldie PA, Hill KD. Systematic and random error in repeated measurements of temporal and distance parameters of gait after stroke
. Arch Phys Med Rehabil.
14.Hill KD, Goldie PA, Baker PA, et al. Retest reliability
of the temporal and distance characteristics of hemiplegic gait using a footswitch system. Arch Phys Med Rehabil.
15.Stephens JM, Goldie PA. Walking speed on parquetry and carpet after stroke
: effect of surface and retest reliability
. Clin Rehabil.
16.Flansbjer UB, Holmback AM, Downham D, et al. Reliability
of gait performance tests in men and women with hemiparesis after stroke
. J Rehabil Med.
17.Ada L, Dean CM, Hall JM, et al. A treadmill and overground walking program improves walking in persons residing in the community after stroke
: a placebo-controlled, randomized trial. Arch Phys Med Rehabil.
18.Plummer P, Behrman AL, Duncan PW et al. Effects of stroke
severity and training duration on locomotor recovery after stroke
: a pilot study. Neurorehabil Neural Repair.
19.Salbach NM, Mayo NE, Higgins J, et al. Responsiveness and predictability of gait speed
and other disability measures in acute stroke
. Arch Phys Med Rehabil.
20.Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability
. Psychol Bull.
21.Stratford P. Getting more from the literature: estimating the standard error of measurement from reliability
studies. Physiother Can.
22.Haley SM, Fragala-Pinkham MA. Interpreting change scores of tests and measures used in physical therapy. Phys Ther.
23.Fritz JM, Irrang JJ. A comparison of a modified Oswestry Low Back Disability Questionnaire and the Quebec Back Pain Disability Scale. Phys Ther.
24.Wyrwich K, Tierney W, Wolinsky F. Further evidence supporting a SEM-based criterion for identifying meaningful intra-individual changes in health related quality of life. J Clin Epidemiol.
25.Portney LG, Watkins MP. Foundations of Clinical Research: Applications to Practice
, 2nd ed. Upper Saddle River, NJ: Prentice Hall Health, 2000.
26.Barak S, Duncan PW. Issues in selecting outcome measures to assess functional recovery after stroke
27.Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res.
28.Dalton GW, Keating JL. Number needed to treat: a statistic relevant for physical therapists. Phys Ther.
Keywords:© 2008 Neurology Section, APTA
gait speed; reliability; stroke