The Development and Validity of the Salford Gait Tool: An Observation-Based Clinical Gait Assessment Tool

Initial Design of the SF-GT 

The initial structure of the SF-GT was based on our previous experience,11 a review of existing tools,5 and reviews of video images of 10 children (6 boys, 4 girls; mean age, 7y; age range, 5−10y) with hemiplegic (n=3), diplegic (n=3), and quadriplegic (n=4) CP gait. All 10 gave their written consent to participate in the research and ethics approval was granted. The initial tool was designed to assess sagittal plane hip, knee, and ankle angular positions at 6 specific events during the gait cycle (initial contact, end double support, mid stance, start double support, toe-off, mid swing). We selected 6 events as a compromise between using a large number of events that would enable a comprehensive assessment but would be time consuming, or using a small number of events that would be quicker to complete but might provide too little detail. Moreover, the gait events needed to be precisely and repeatedly identifiable visually.

We devised a 5-point category scoring system (2, 1, 0, −1, −2) to describe the positions of the hip, knee, and ankle at the 6 gait events (3 joints, 6 gait events = 18 total assessments). Each scoring category would correspond to a specific range of angular positions. Importantly, we felt that the range of angular positions defining each category must be small enough to be sensitive to differences between different gait styles and changes resulting from clinical interventions, but large enough to be identified by the naked eye when observing gait on a video screen.

The sum of the 6 category scores for each joint would then represent that joint’s function over the entire gait cycle and provide a qualitative description of the entire joint pathology.

It was our intention that the final definitions of the boundaries between 1 scoring category and the next, as well as the boundaries between the summed category scores, would be based on quantitative kinematic gait data that described different types of gait pathology. To enable us to obtain early user feedback on the tool’s overall design, however, the boundaries between categories were provisionally set based on gait literature12, 13, 14 and clinical observation.

Development of Construct and Content Validity 

This initial version of the tool was then evaluated for its user friendliness, physical layout, wording, and construct and content validity by 9 successive focus groups over a 1-year period. Eight groups had an average of 19 physical therapy students per group (total N=152) and 1 group included 4 physical therapists who were specialists in the fields of gait assessment, pediatrics, neurology, and musculoskeletal physical therapy. There were several testing, redesign, and testing iterations as changes were made to the tool at each stage.

Development of Criterion Validity 

In this development stage, we used quantitative kinematic gait data to adjust the upper and lower boundaries of the scoring categories for each of the 3 joints at the 6 gait phases of the SF-GT. Such boundaries in existing tools appeared to be defined based on clinical experience6, 7, 8, 9, 10 rather than on formal evaluation of the gait patterns the tools would subsequently be used to evaluate. We assumed that careful adjustment of these boundaries would improve the SF-GT’s validity, sensitivity, and specificity.

In previous work15 we defined 13 gait styles by using cluster analysis of kinematic gait data from 56 children with a mixture of CP types and 11 neurologically intact children. Details of the kinematic data collection protocol and statistical analysis have been reported.15 We adjusted the SF-GT so that the upper and lower boundaries of each scoring category and the boundaries of the summed category scores reflected the mean joint positions and standard deviations (SDs) of the kinematic data of the children with each of the 13 distinct gait types. The calibration procedure was completed for the hip, knee, and ankle at all 6 gait phases. It was stopped when the scoring of the kinematic data of all gait types resulted in an adequate numeric and qualitative description of the entire joint pathology that was different from the other gait types.

Clinical Evaluation 

To evaluate the clinical validity of gait assessments made with the SF-GT, the gait of 13 children was assessed by 23 pediatric physical therapists. The visual assessments were then compared with the quantitative assessment of the children’s gait. The observers were recruited from 9 National Health Service Trusts from the Greater Manchester area in the United Kingdom and their clinical expertise ranged from junior (n=1), senior II (n=2), senior I (n=18), to superintendent (n=2). All observers gave their written consent before participating. The 13 children each represented 1 of the 13 gait styles previously defined with cluster analysis (9 boys, 4 girls; age range, 6−16y; mean age, 9.5y).15 Eleven children had hemiplegia (n=4), diplegia (n=6), or quadriplegia (n=1) CP and 2 children were neurologically intact. The video data were collected during the same laboratory visit at which the kinematic gait data were collected. We used that data, which were part of a larger clinical gait analysis database, to define the 13 gait styles.15 Parents and children had previously given their consent to use the images for research purposes and the research was approved by the appropriate ethics committees.

Before making the gait assessments, the 23 observers were given general training on gait, video assessment of gait, and the use of the SF-GT. Observers were each allocated a workstation with a DVD player and television screen and did a trial gait assessment. The observers then worked individually without discussion to assess 1 gait cycle of 1 leg of each of the 13 children, using the SF-GT. They were permitted to work at their own speed, to review the gait cycles as often as required, and there was no time limit. The assessments were completed in from 3 to 5 hours.

Data Analysis 

We analyzed the level of agreement between clinicians’ gait assessments using the SF-GT and the criterion standard kinematic gait assessments in 2 ways. First, we compared the frequency of agreement between the scoring categories estimated by each observer with the scoring categories assigned by the kinematic data (eg, if 5° of knee flexion at initial contact by the kinematic data were category 1, there was agreement if the observer also rated the knee as category 1). We also analyzed the extent of agreement by counting the frequencies of observer category scores that deviated by 1, 2, or more categories from the categories assigned by the kinematic data. The results of this analysis would enable direct comparison with the results of other OGA tools.

A disagreement, however, between observer category scores with categories assigned by the kinematic data may not always reflect the true closeness (or difference) between the observer assessment and kinematic data. For example, if an observer estimated the hip flexion to be 46° (category 2) and the actual kinematic data were 45° (category 1), this would have resulted in a disagreement by 1 category, even though there would have been only 1° difference between the assessments. Conversely, if an observer estimated the hip flexion to be 45° (category 1) and the actual kinematic degrees were 16° (also category 1), this would have resulted in a perfect agreement, even though the estimations would have differed by 29°. Therefore, the observer derived angles in degrees (recorded to 1°) were compared with the joint positions derived from the kinematic gait data (recorded to 0.1°). For this analysis, we used the least significant difference (LSD) method.10, 16, 17 The LSD defines a “proximity” range above and below the target kinematic data. These upper and lower boundaries define what would be considered acceptable deviations from a perfect match to the kinematic data. We assumed that if the observers stated a number of degrees that was within this range then it was close enough to the actual kinematic data to be valid. This approach has been used in a similar way to assess intraobserver repeatability of the EVGS.10 The LSD represents a statistically significant difference between observer and kinematic data and occurs because of interclinician variation. It considers the distance between all the observations and the target kinematic data and is derived by multiplying the SD between 2 assessments (observer and kinematic data) with the relevant t distribution from the Student t test (with 1 degree of freedom, 2-tailed t test, .05 level of significance). If the observer’s estimated degree value lies outside the LSD range, then the difference between observers and kinematic degrees is statistically significant and too large to be considered acceptable. We used Excela for the statistical analysis.

Criterion Validity 

We set the final definitions of the boundaries between each scoring category at the end of the calibration procedure. An example of the calibration result for the knee at initial contact for the 13 gait styles is given in table 1. The final gait assessment tool is detailed in appendix 1, which also provides information on normative gait data and guidelines for use of the tool.

Table 1. Allocation of the 13 Gait Clusters to the Knee Grade Category at Initial Contact After Calibration of the SF-GT

	Outcome of Knee Calibration at Initial Contact
	Gait Cluster	Category −2 (−16° E or more)	Category −1 (−6° to −15° E)	Category 0 (−5° E to 10° F)	Category 1 (11° to 45° F)	Category 2 (46° F or more)
	1: mild crouch	NA	NA	NA	27.01±5.11	NA
	2: mobile crouch	NA	NA	NA	33.31±5.75	NA
	3: moderate crouch	NA	NA	NA	NA	48.49±7.44
	4: severe crouch	NA	NA	NA	36.89±2.89	NA
	5: mild equinus	NA	NA	NA	20.83±4.66	NA
	6: moderate equinus/knee extension	NA	NA	NA	33.44±5.71	NA
	7: moderate equinus/knee flexion	NA	NA	NA	29.79±3.33	NA
	8: severe equinus	NA	NA	NA	20.33±5.23	NA
	9: stiff leg	NA	NA	4.57±4.45		NA
	10: weak plantarflexion	NA	NA	NA	15.50±5.05	NA
	11: ankle double bump	NA	NA	NA	30.27±4.96	NA
	12: near normal	NA	NA	−2.48±7.67	NA	NA
	13: normal	NA	NA	1.90±3.05	NA	NA

NOTE. Values are mean degrees ± SD. Abbreviations: E, extension; F, flexion; NA, not applicable.

Clinical Evaluation 

Agreement between category scores 

Observers agreed on 2900 category scores (58%; range, 28%−75%) of a possible 5004 category scores with the kinematic gait data (table 2). Regarding the magnitude of disagreement, of the 2104 category scores that disagreed, 95% (1991 category scores) disagreed by 1 category, 4% (93 category scores) disagreed by 2 categories, 1% (19 category scores) differed by 3 categories, and 1 observer’s category score differed by 4 categories from the kinematic category score.

Table 2. Level of Agreement, Computed With Percentage Agreements of Categories, Between Mean Observer Data and Kinematic Data for Hip, Knee, and Ankle, and for All Gait Clusters

	Gait Cluster	Hip Agreement (mean %)	Knee Agreement (mean %)	Ankle Agreement (mean %)	Mean Agreement (%) for Cluster
	1: mild crouch	83⁎	63	53	66
	2: mobile crouch	62	71	73	69
	3: moderate crouch	39	24	22†	28†
	4: severe crouch	26	82	49	52
	5: mild equinus	80	75	44	66
	6: moderate equinus/knee ext	30	49	24	34
	7: moderate equinus/knee flex	49	74	63	62
	8: severe equinus	78	46	63	62
	9: stiff leg	32	66	65	54
	10: weak plantarflexion	68	80	77	75⁎
	11: ankle double bump	72	68	44	61
	12: near normal	77	40	64	60
	13: normal	63	52	81	65
	Mean agreement for joint (%)	58	61	56	Total average
					58

⁎ Highest agreement † lowest agreement.

Best agreement (83%) was for the hip joint of cluster 1 (gait style is mild crouch gait) and the lowest agreement (22%) was for the ankle joint of cluster 3 (gait style is moderate crouch gait). The knee joints were assessed more accurately than the hip and ankle joints (mean agreement: hip, 58%; knee, 61%; ankle, 56%). Cluster 10 (weak plantarflexion gait) was the gait style with the best agreement (75%). Cluster 3 (moderate crouch gait) had the least agreement (28%).

Agreement between degrees 

We computed the LSD on a total of 4446 pairs of observer and kinematic data for all gait events (n=6) of all joints (n=3) and clusters (n=13). The degree data from 2 observers were not recorded on the SF-GT and therefore were not available for analysis. The mean LSD was 16.25° (range, 2.31°−63.20°). This means that the mean observations differed by 16.25° from the kinematic data. An average of 80% (range, 68%–91%) of gait observations lay within the LSD range of the kinematic data at the 6 phases of gait for all joints and clusters (table 3). Best agreement (91%) was for the hip joint of cluster 4 (gait style is severe crouch) and the lowest agreement (68%) was for the hip joint of cluster 2 (gait style is mobile crouch). The mean agreement between joints was similar (mean agreement: hip, 80%; knee, 80%; ankle, 81%). The clusters that demonstrated best agreement (both 84%) were clusters 6 (gait style is moderate equinus/knee extension) and 13 (gait style is normal gait). The cluster with the least agreement was cluster 2 (gait style is mobile crouch). Regarding the magnitude of agreement between the observers and the kinematic data, 86% of observations were within 25° of kinematic data, 79% of observations were within 20° of kinematic data, 62% of observations were within 15° of kinematic data, 24% within 10°, and 2.6% within 5°. Table 4 shows the data from 1 observer for mild equinus gait (cluster 5); this gait cluster was representative of a mean agreement of 80% between all observers.

Table 3. Level of Agreement, Computed Using LSD and Degree Estimates, Between Mean Observer Data and Kinematic Data for Hip, Knee, and Ankle, and for All Gait Clusters

	Gait Cluster	Hip Agreement (mean %)	Knee Agreement (mean %)	Ankle Agreement (mean %)	Mean Agreement (%) for Cluster
	1: mild crouch	73	74	76	74
	2: mobile crouch	68†	69	78	72†
	3: moderate crouch	77	83	78	79
	4: severe crouch	91⁎	72	84	82
	5: mild equinus	85	77	79	80
	6: moderate equinus/knee ext	83	83	86	84⁎
	7: moderate equinus/knee flex	77	79	80	79
	8: severe equines	78	83	77	79
	9: stiff leg	76	83	86	82
	10: weak plantarflexion	85	82	82	83
	11: ankle double bump	82	87	81	83
	12: near normal	76	83	81	80
	13: normal	83	83	87	84⁎
	Mean agreement (%) for joint	80	80	81	Total average
					80

⁎ Highest agreement † lowest agreement.

Table 4. Results From 1 Observer Showing the Differences Between Kinematic Data and Observer Data, SD, t Value, Individual LSD, and Mean LSD (all observers) for All Gait Phases of Cluster 5 (mild equinus)

	Gait Cluster 5: Mild Equinus	Kinematic Data	Observer Data	SD	t=39n	LSD	Mean LSD
	Hip
	Initial contact	19	20	0.71	2.02	1.43	31.85
	End double support	18	30	8.49	2.02	17.14	19.78
	Mid stance	−2	10	8.49	2.02	17.14	17.14
	Start double support	−19	−10	6.36	2.02	12.86	13.57
	Toe-off	−3	0	2.12	2.02	4.29	11.28
	Mid swing	27	45	12.73	2.02	25.71	22.21
	Knee
	Initial contact	20	30	7.07	2.02	14.28	16.07
	End double support	22	20	1.41	2.02	2.86	14.21
	Mid stance	12	10	1.41	2.02	2.86	9.57
	Start double support	14	15	0.71	2.02	1.43	11.28
	Toe-off	61	50	7.78	2.02	15.71	13.43
	Mid swing	71	90	13.44	2.02	27.14	22.07
	Ankle
	Initial contact	−3	−5	1.41	2.02	2.86	16.00
	End double support	−1	−10	6.36	2.02	12.86	5.50
	Mid stance	1	−10	7.78	2.02	15.71	8.28
	Start double support	2	−5	4.95	2.02	10.00	9.57
	Toe-off	−33	−30	2.12	2.02	4.29	17.85
	Mid swing	−15	−10	3.54	2.02	7.14	15.35