Interrater reliability of videotaped observational gait-analysis assessments.

Gait assessment has become an increasingly important part of physical therapy patient evaluations. Gait assessment is used to determine whether the patient's gait differs from "normal," to quantify the degree of abnormality, and to identify the causes of the abnormal gait patterns, and it is used as a reassessment tool to evaluate the efficacy of treatment. (1,2) Although some instrumented gait-analysis systems have been shown to give reliable and valid measurements, they are costly and may be impractical for most clinicians to use as an everyday assessment tool. (1-6)

We believe that some form of observational gait analysis (OGA) is the most widely used method of gait analysis. Observational gait analysis is used to make clinical judgments such as whether a patient needs bracing, how well a prosthesis fits, or whether an assistive device is needed. Observational gait analysis generally consists of joint displacement (kinematic) or temporospatial factor analysis, or both. The reliability of OGA assessments has generally been low. (6,7) Recently, videotaping has been used as an adjunct to OGA, because videotaping of the patient allows the therapist to view gait patterns repeatedly without inducing patient fatigue. (1,3) Videotaped observational gait analysis (VOGA) also allows the therapist to stop or slow the tape, which is thought to increase the precision of the assessment. (1,3) The reliability and validity of observations based on videotapes of different types of patient populations have not yet been demonstrated.

Krebs et al (3) examined the interrater and intrarater reliability of VOGA assessments in a pediatric population. Three trained observers rated videotaped joint kinematics in 15 children with lower-limb disabilities who were wearing knee-ankle-foot orthoses. Total agreement (identical ratings) between raters occurred for approximately 67.5% of the ratings. Furthermore, an additional 30% differed by one rating level. The authors concluded that the VOGA technique was "a convenient, but only moderately reliable" tool for evaluating their patient population. They did not, however, use stop- or slow-action videotaping.

Observational gait analysis and VOGA continue to be the most widely used methods for assessing gait in clinical settings. (1,3-4) The purpose of this study was to assess the interrater reliability of physical therapists' observations based on a videotape of the gait of three patients with rheumatoid arthritis. The null hypothesis for each of the 10 variables tested was that rater agreement would be no better than that attributable to chance alone.

Method

Patients

The three patients who were videotaped (ages 43, 45, and 61 years) were current or former patients at the National Institutes of Health, Bethesda, Md. Criteria for inclusion in the study were (1) an abnormal gait pattern secondary to stage 2 or 3 rheumatoid arthritis (8) as identified previously by the rehabilitation team and (2) the ability to walk at least eight lengths of a 10-m walkway (with or without rest periods) independent of an assistive device without complaining of fatigue or pain. Each patient read and signed a document of informed consent.

Raters

Fifty-four licensed physical therapists employed by seven different hospitals in the Washington, DC, metropolitan area were selected as raters for this study (see Appendix 1 for description of raters). No restrictions were placed on their physical therapy education, clinical experience, or gait-evaluation skills. (6) Each rater read and signed a document of informed consent.

Equipment and Procedure

Initial screening of the patients to determine whether they met the inclusion criteria was conducted by the investigators prior to the study. Videotaping was carried out using a Panasonic AG-HT3 video camera (1) and a Bogen 3033 tripod. (2) Patients were attired in a shirt and shorts and were barefoot. The patients were instructed to walk six lengths of the 10-m walkway at a self-selected pace while they were videotaped from anterior, posterior, and laterial views: 1 anterior pass, 1 posterior pass, and 4 lateral passes (2 passes on each side). The investigators reviewed each patient's segment of videotape to locate the points corresponding to the four gait events to be evaluated on each pass: initial contact, mid-stance, heel-off, and toe-off. The number on the videotape counter corresponding to each event was recorded. To further standardize the videotaping procedure, tripod placement, distances from lens to floor (91.4 cm [36 in]) and from lens to walkway (3.66 m [12 ft]), and zoom-lens setting were recorded and made uniform from patient to patient. All videotaping was carried out by the same individual (JA).

Rater training. The raters were oriented to the evaluation form prior to its use. All rating categories were defined, and the scoring protocol was explained. This orientation included a description of the mechanics of the rating session, but did not include a review of normal gait kinematics. A gait-analysis questionnaire was distributed to and completed by each of the raters at this time. Its purposes were to identify each rater's type of clinical experience, years of experience, and type of schooling and to obtain each rater's subjective self-assessment of his or her OGA capabilities (Appendix 1).

Rating procedure. Viewing of the videotape took place at the individual facilities for the convenience of the raters. The videotape was shown to small groups of raters. One of the investigators (JA) was present throughout each session. No discussion regarding the assessment was allowed among the raters during the session. While viewing the videotape, the raters evaluated the patients' knee joint displacement and gait temporospatial factors as being inadequate, normal, or excessive. Knee joint displacement variables were analyzed from lateral (flexion/extension), anterior (genu valgum, step width), and posterior (genu valgum, step width) views at each of the four subphases of stance. Initial contact was defined as the event that occurred when the heel first touched the floor. Mid-stance was defined as the event that occurred when the knee and hip joints were directly over the ankle. Heel-off was defined as the event that occurred when the heel started to lift from the floor. Toe-off was defined as the event that occurred when the toe left the floor.

Four passes were analyzed from a lateral view. The videotape was slowed and uniformly stopped for 5 to 7 seconds at premeasured and predesignated points during the middle 3 m of passes 1 and 3. (During these passes, the reference leg, or most severely involved leg, was closest to the camera.) The stop-action points corresponded to the four particular subphases of stance being analyzed (ie, initial contact, mid-stance, heel-off, and toe-off) and were identified and marked using the counter on the videocassette recorder as described previously. Slow-and stop-action techniques were used in an attempt to improve the reliability of the gait-analysis assessments, as suggested by Krebs et al. (3) Once the middle 3 m was viewed, videotape speed was returned to normal for the purpose of analyzing cadence, stance time, step length, and stride length. The videotape speed was controlled by the same investigator (JA) throughout the study. Once the four lateral passes were completed, the raters were given the opportunity to view the patient again at slowed or normal speed. The camera angle then was adjusted for anterior-posterior viewing, with one pass viewed posteriorly and one pass viewed anteriorly. To facilitate these judgments, the videotape was again slowed or stopped during the first three steps of the posterior view and the last three steps of the anterior view, which corresponded to the time when the subject was closest to the camera. As with the lateral views, after the anterior-posterior viewing, raters were given the opportunity to review the segment, if needed. No rater required more than two reviews.

Kinematic judgments were recorded on the Knee Joint Displacement Evaluation Form. Evaluation of the temporospatial factors was carried out using a Temporospatial Factors Assessment Form. These forms were derived from common OGA forms (2-5,9,10) and combined into a single document (Appendix 2).

Data Analysis

Interrater agreement for this study was assessed using the generalized kappa (K). This statistic was developed by Cohen (11) to assess agreement among two or more raters after taking the effect of chance agreement into account. The formulas for polychotomous, multirater data (9) were used. Agreement among raters, for the purposes of this study, were considered equivalent to interrater reliability.

A wide range of variability in patient performance for each variable lessens the possibility of chance agreement among raters. (9,11) For reliability measurements to be meaningful, patient variability must be present. (10,12) To assess the presence of patient variability, we examined the modal distributions of the ratings of the patients by category. Frequency counts for each of the 10 variables by patient are summarized in Tables 1 and 2.

Relative agreement strengths for kappa values have been classified as follows: <0=poor, 0-.20=slight, .21-.40=fair, .41-.60=moderate, .61-.80=substantial, and .81-1.00= almost perfect. (13) Significance for kappa values (agreement among raters significantly greater than chance) was tested by calculating a kappa variance and using a Z test. (9,11,14)

Kappa coefficients were calculated for each of the 10 gait variables across the three patients. In addition, pair-wise kappa coefficients were determined for each of the three two-patient

     Table 1. Frequency of Responses for Kinematic Variables
                             Rating
Variable           Patient   Inadequate   Normal   Excessive
Knee flexion at      A           18         24        12
 initial contact     B            4         33        17
                     C           35         14         5
Knee flexion at      A            32        18         4
  mid-stance
                     B            6         44        4
                     C            45        4         5
Knee flexion at      A           45         9         5
  heel-off
                     B           17         33        4
                     C           45         7         2
Knee flexion at      A           25         25        4
  toe-off
                     B            5         43        6
                     C           42         11        1
Genu valgum          A            4         1         49
                     B (a)        1         36        16
                     C           44         8         2
  (a) One rater did not rate this patient for this variable.

Table 2. Frequency of Responses for Temporospatial Variables

                         Rating
Variable       Patient   Inadequate   Normal   Excessive
Stride length    A (a)       14         30         9
                 B (a)       5          42         6
                 C           2          2         50
Step length      A           16         22        16
                 B           6          36        12
                 C           5          2         47
Stance time      A           15         20        19
                 B           6          33        15
                 C           5          0         49
Cadence          A (a)       5          30        18
                 B           17         35        2
                 C           9          1         44
Step width       A (a)       33         10        10
                 B           2          32        20
                 C           31         10        13
(a) One rater did not rate this patient for this variable.

pairs to determine whether any one of the three patients was noticeably more difficult to evaluate by the raters and to assess effects of order. Raters were divided into two groups based on physical therapy experience: (1) raters with [is less than or equal to]3 years of experience and (2) raters with >3 years of experience. Separate kappa coefficients were calculated for each of these two groups to determine whether interrater reliability (agreement) changed with experience.

Intraclass correlation coefficients (ICCs) (9,15,16) were used to provide a weighting of differences within each variable, with the added assumption that the three variable levels were equally spaced (ie, inadequate=-1, normal=0, excessive=+1). Although a large sample of raters (N=54) from several different hospitals (N=7) was used, the selection procedure did not strictly meet the requirements of randomness. For this reason, both ICC (2,1) and ICC (3,1) were computed.

Results

A summary of the data accumulated by the Gait-Analysis Questionnaire is presented in Appendix 1. Fifty-three raters completed the questionnaire. (One rater took the questionnaire with her when she left the rating session, and we were unable to retrieve it.) Not all raters answered all questions. If a rater qualified an answer by adding any comments or modifiers, it was eliminated from the count.

All therapists completed the gait-analysis rating forms. Frequency counts for therapists' ratings of each of the 10 variables are presented in Tables 1 and 2. Generalized kappa and ICC (2,1) and (3,1) values are provided in Table 3. All of the generalized kappa coefficients were significantly greater than 0 (z>1.96, P<.01), and the null hypothesis of no rater agreement was rejected. These kappa values, however, were only high enough to indicate slight to moderate agreement. (15) The ICCs were significant (F>1.46; df=53, 106; P<.05) for the following variables: knee flexion at initial contact, knee flexion at mid-stance, knee flexion at toe-off, cadence, and step width.

Results of the pair-wise kappa analysis (Tab. 4) demonstrate that when the ratings for patient A, the first patient on the videotape, were included in the data analysis, the kappa coefficients were smaller. Furthermore, the kappa coefficients when the ratings for patient A were not included in the data analysis were the highest of the three pairs for all but two variables (ie, genu valgum and step width) and the kappa coefficients for these two variables were within .03. All the kappa coefficients for patient pair B/C were significant (z>1.96, P<.01), but 11 of the 20 kappa coefficients for patient pairs A/B and A/C were not significant.

Table 3. Generalized Kappa

Coefficients (K) and Intraclass Correlation

Coefficients (ICCs) (2,1) and (3,1)

                           ICC         ICC
Variable          K (a)    (2,1)       (3,1)
Knee flexion at
 initial contact  .11      .26 (a)     .34 (a)
Knee flexion at
 mid-stance       .31      .28 (a)     .33 (a)
Knee flexion at
 heel-off        .18       .19         .19
Knee flexion at
 toe-off         .23       .35 (b)     .38 (b)
Genu valgum       .52      .69         .65
Stride length     .40      .49         .52
Step length       .24      .28 (b)     .32 (a)
Stance time       .23      .26 (a)     .31 (a)
Cadence           .29      .32         .34
Step width        .16      .23         .24
(a)P<.01.
(b)P<.05.

Table 4. Pair-Wise Kappa Coefficients

                   Patient Pair
Variable           A/B      B/C (a)  A/C
Knee flexion at
 initial contact   .02      .18       .04
Knee flexion at
 mid-stance        .20 (a)  .46       .05
Knee flexion at
 heel-off          .15 (a)  .22      -.01
Knee flexion at
 toe-off           .10      .36       .07
Genu valgum        .39 (a)  .38       .57 (a)
Stride length      .03      .55       .38 (a)
Step length        .03      .35       .21 (a)
Stance time        .02      .34       .21 (a)
Cadence            .05      .38       .22 (a)
                                     -.02
Step width         .19 (a)  .16
(a)p<.01.

Kappa coefficients comparing the reliability of raters with [is less than or equal to]3 years of clinical experience with raters with >3 years of clinical experience are presented in Table 5. Sixteen of these 20 kappa coefficients were significant (z>1.96, P<.01).

Discussion

Generally, the agreement coefficients were in the low-to-moderate range for the 10 variables tested. The greatest agreement among the raters for all the variables was in the assessment of genu valgum (K=.52). The modal response of the ratings suggested that each of the three patients demonstrated one of the three possible conditions. The raters agreed 91% of the time on the presence of excessive genu valgum, or valgus (in patient A); 68% of the time on the presence of normal genu valgum, or no valgus (in patient B); and 81% of the time on the presence of inadequate genu valgum, or varus (in patient C). With patient B, 30% of the raters noted excessive genu valgum.

Raters had more difficulty with the kinematic assessment than with the temporospatial factor assessment. In the kinematic assessment, the raters were apparently not sure of what constituted a normal amount of knee flexion for each subphase of stance. The highest kappa value for these four variables was .31, which occurred for the assessment of knee flexion at mid-stance. This finding is perhaps due to the raters' greater familiarity with how much flexion should be present during mid-stance than during any other subphase of stance.

Rater agreement for the adequacy of stride length was .40, among the highest ratings in this study. This value, however, is somewhat inflated by the fact that there was 93% agreement on the presence of a shortened stride length in patient C, who exhibited a stride length that was clearly shorter than that of patients A and B. Most raters identified patients A and B as being normal (57% and 79%, respectively).

In the evaluation of the adequacy of step length, the raters were asked to compare the involved-side step length with that of the contralateral side. This

[TABULAR DATA OMITTED]

comparison seemed difficult for the raters. Two thirds (67%) of the raters observed the step lengths to be equal (normal difference) in patient B, although 23% assessed step length as decreased. In patient C, however, the raters overwhelmingly (87%) noted a decreased step length. We believe that structural deformities and other abnormalities in the gait pattern that were present in patient C misled the raters into consistently rating this patient as abnormal.

The raters had equal difficulty in assessing the adequacy of stance time. Again, the raters' assessments were spread among the three rating categories for patient A (28% excessive, 37% normal, and 35% inadequate). Although the modal response was "normal," a significant percentage of raters (28%) evaluated patient B as having a decreased stance time on the involved side.

Rater agreement for cadence was fair (K=.29). As in the assessment of the adequacy of step length and stance time, the raters were divided between at least two responses for patients A and B. The exception was patient C, for whom there was 81% agreement that she demonstrated a decreased cadence. Patient C, however, also exhibited decreased stance time on the involved side and walked more slowly than the other female patient. These deviations were readily observable and may have swayed the raters into overwhelmingly reporting a decreased cadence.

There was very little agreement in the assessment of the adequacy of step width (K=.16). The ratings were evenly distributed across the three categories for all three patients.

There seem to be two main reasons for the slight to moderate interrater reliability. First, the therapists did not seem to be familiar with the normative values for the tested gait variables. (9,17-19) Second, there was a tendency for some of the raters to identify an abnormality as being present in the two patients with severe joint deformities. Conversely, the raters tended to rate the patient with the least apparent joint deformities as normal.

Unfortunately, because of time constraints, patients were presented to all raters in the same order. This procedure resulted in an ordering bias, as demonstrated by the pair-wise kappa values. Ratings for patient A were highly variable among the raters, with ratings split among all three categories for each variable. The pair-wise kappa values suggest that the raters then judged patient B against patient A and patient C against patient B.

Despite the orientation session, difficulty understanding the terminology or the evaluation form may also have contributed to the low reliability coefficients. A practice test segment to be viewed prior to the three graded test segments might be used in the future to address this issue. No practice segment was used during this study for two reasons: (1) The limited time allotted for testing by the facilities precluded the incorporation of any more subjects onto the videotape, and (2) in the clinical setting, raters do not usually use a patient to "refresh" their skills immediately prior to analyzing the videotape of another patient. We felt this was a more clinically applicable approach.

The ICCs (2,1) and (3,1) allowed the weighting of the various responses to account for the levels of disagreement. Disagreement of one rating level was assumed to be less clinically meaningful than disagreement at two rating levels at which responses are totally contradictory. The ICCs were slightly higher than the kappa coefficients, indicating that the disagreements were predominately of one rating level.

There was no consistent difference in reliability between groups based on years of clinical experience. There was no difference in kappa values, regardless of the raters' years of clinical experience. Neither formal educational training nor clinical training appeared to provide assurance of reliability in gait evaluation as measured in this study. Many therapists felt comfortable with their ability to analyze gait observationally (Appendix 1), yet our data showed only fair reliability among their judgments.

Despite the poor reliability estimates obtained in this study, they may actually be greater than those that can be obtained clinically. Because we controlled the videocassette recorder, we chose to slow down the tapes for analysis at the same points for all observers. We therefore allowed or, to some extent, even forced all raters to focus on the same portions of gait during each pass. In clinical practice, the decision to slow down a tape for further analysis must be made by each rater. The possibility exists that not all therapists would choose the same exact portion of the tape to analyze at slow motion. The segments we chose were based on our identification of the occurrence of a phase of gait we wanted evaluated. Based on our methods, we cannot be sure that every rater would similarly identify the time periods when these phases of gait occur.

In the only other published VOGA clinical reliability study, Krebs et al (3) found ICCs (2,1) on average of .73; however, they stated the range of ICC values was .40 to .94. They concluded that, in general, sagittal-plane assessment was more reliable than either transverse- or frontal-plane assessment. We found that the frontal-plane assessment (used for the genu valgum evaluation) produced the most reliable kinematic results. Direct comparison of our study with that of Krebs et al is difficult for three reasons. First, they did not describe the results for each variable rated. Second, they also used Pearson Product-Moment Correlation Coefficients and percentages of agreement to analyze their data, which we believe can yield inflated reliability coefficients. Third, they studied only kinematics. We used two of their suggestions for improving reliability--stop-action videotapes and defining discrete events in the gait cycle for analysis. We also limited our analysis to a single joint. None of these changes appeared to improve the reliability of the raters' observations.

Clinical Implications

As with any clinical test, the most significant sequela of inaccurate measurement is the possibility of improper treatment. In the rating of patient A's knee flexion during initial contact, heel-off, and toe-off, the distribution of responses was almost equally divided among the possible responses: 33% excessive, 44% normal, and 33% inadequate. Some of these therapists would presumably prescribe a therapeutic intervention that would promote greater knee flexion throughout stance, whereas others would do the opposite and some would do nothing.

Although this argument relates to the validity of raters' judgments as well as to reliability, the two are inextricably linked. In the clinical setting, if two therapists see different deviations in rating the same variable of a patient's gait, how does one ascertain which is correct? Any measurement must be reliable in order to be valid.

Based on our interpretation of the results of this study, combined with the informal and subjective reports of several therapists who took part in the study, we suggest the following: (1) Greater standardization of gait-analysis training protocols is needed, and (2) instruction should emphasize that the degree of joint deformity does not necessarily correlate with more significant gait deviations.

Suggestions for

Further Research

In the only other study to date that has examined the reliability of ratings obtained using VOGA, Krebs et al (3) found only moderate reliability. Similar studies should be undertaken to address the findings of the study by Krebs et al (3) and of this study.

We believe that the raters judged patients B and C with respect to patient A, and our analysis of order effects supports this hypothesis. The possibility of a learning effect should be accounted for, perhaps by allowing the raters to view a videotaped gait sequence of a non-test patient and to practice using the gait-analysis assessment form prior to evaluation of the test patients. We believe that referenced norms should be incorporated into evaluation forms, as is done in the Rancho Los Amigos Normal and Pathological Gait Syllabus, (9) to reacquaint therapists with normal ranges of values. Therapists then would not have to use the patient as a learning tool during each gait evaluation.

This study focused only on the involved knee during the subphases of stance. Analysis should be extended to include swing phases, in addition to analysis of all other joints involved in gait.

Conclusions

Physical therapists demonstrated slight to moderate reliability in assessment of the adequacy of temporospatial and kinematic gait variables when focusing on the most severely affected knee in three patients with rheumatoid arthritis. The results of this study suggest that VOGA has potential as a clinical assessment technique but that improvements are needed.

There appears to be a discrepancy between the raters' self-assessment of their gait-analysis capabilities and this ability as reflected by the data. We feel that frequent exposure to VOGA, stressing a more standardized approach to teaching, and implementing VOGA with greater emphasis on the referenced normative values could increase interrater reliability.

Acknowledgments

We thank Dr Charles L Eastlack and Dr John J Bartko for their assistance with statistical problems. We also thank David Krebs, PhD, PT, Ken Holt, PhD, PT, and Susan Daleiden, ScD, PT, for their editorial comments on an earlier version of this article.

[TABULAR DATA OMITTED]

[TABULAR DATA OMITTED]

(1) Panasonic Industrial Co, One Panasonic Way, Secaucus, NJ 07094.

(2) Bogen Photo Corp, 565 E Cresent Ave, PO Box 506, Ramsey, NJ 07446.

References

[1] Craik Rl, Oatis CA. Gait assessment in the clinic: issues and approaches. In: Rothstein JM, ed. Measurement in Physical Therapy. New York, NY: Churchill Livingstone Inc; 1985:169-205.

[2] Gronley JK, Perry J. Gait analysis techniques: Rancho Los Amigos gait laboratory. Phys Ther. 1984;64:1831-1838.

[3] Krebs DE, Edelstein JE, Fishman S. Reliability of observational kinematic gait analysis. Phys Ther. 1985;65:1027-1033.

[4] Robinson JL, Smidt GL. Quantitative gait evaluation in the clinic. Phys Ther. 1981;61:351-353.

[5] Holden MK, Gill KM, Magliozzi MR, et al. Clinical gait assessment in the neurologically impaired: reliability and meaningfulness. Phys Ther. 1984;64:35-40.

[6] Miyazaki S, Kubota T. Quantification of gait abnormalities on the basis of continuous footforce measurement: correlation between quantitative indices and visual rating. Med Biol Eng Comput. 1984;22:70-76.

[7] Goodkin R, Diller L. Reliability among physical therapists in diagnosis and treatment of gait deviations in hemiplegics. Percept Mot Skills. 1973;37:727-734.

[8] Guccione AA. Rheumatoid arthritis. In: O'Sullivan SB, Schmitz TJ, eds. Rehabilitation: Assessment and Treatment. Philadelphia, Pa: FA Davis Co; 1988:435-460.

[9] Normal and Pathological Gait Syllabus. Downey, Calif: Professional Staff Association of Rancho Los Amigos Hospital Inc; 1978.

[10] Bartko JJ, Carpenter WT. On the methods and theory of reliability. J Nerv Ment Dis. 1976;163:307-317.

[11] Cohen J. A coefficient of agreement for nominal scales. Educational and Psychological Measurement. 1960;20:37-46.

[12] Stuberg WA, White PJ, Miedaner JA, Dehne PR. Item reliability of the Milani-Comparetti Motor Development Screening Test. Phys Ther. 1989;69:328-335.

[13] Landis JR, Koch GG. The measurement of observer agreement for categorial data. Biometrics. 1977;33:159-174.

[14] Fleiss JL. Statistical Methods for Rates and Proportions. 2nd ed. New York, NY: John Willey & Sons Inc; 1981:212-236.

[15] Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull. 1979;86:420-428.

[16] Rothstein JM, Miller PJ, Roettger RF. Goniometric reliability in a clinical setting: elbow and knee measurements. Phys Ther. 1983;63:1611-1615.

[17] Murray MP, Kory RC, Sepic SB. Walking patterns of normal women. Arch Phys Med Rehabil. 1970;51:637-650.

[18] Murray MP, Drought AB, Kory RC. Walking patterns of normal men. J Bone Joint Surg [Am]. 1964;46:335-360.

[19] Kettlekamp DB, Johnson RJ, Smidt GL, et al. An electrogoniometric study of knee motion in normal gait. J Bone Joint Surg [Am]. 1970;52:775-790.

ME Eastlack, MSPT, PT, is Staff Therapist, National Rehabilitation Hospital, 102 Irving St NW, Washington, DC 20010.

J Arvidson, MSPT, PT, is Staff Therapist, Center for Health Promotion, Portland, ME 04101.

L Snyder-Mackler, ScD, PT, is Assistant Professor of Physical Therapy and of Anatomy and Physiology, University of Delaware, 309 McKinly Laboratory, Newark, DE 19716 (USA).

JV Danoff, PhD, PT, is Associate Professor, Howard University, Washington, DC, and Research Consultant, Department of Rehabilitation Medicine, Warren G Magnusen Clinical Center, National Institutes of Health, 9000 Rockville Pike, Bethesda, MD 20892.

CL McGarvey, MS, PT, is Chief of Physical Therapy, Department of Rehabilitation Medicine, Warren G Magnusen Clinical Center.

Ms Eastlack and Ms Arvidson were students in the Master of Science in Physical Therapy Program, Sargent College of Allied Health Professions, Boston University, when this study was conducted in partial fulfillment of the requirements of their Master of Science in Physical Therapy degrees.