Measurement properties of observational gait analysis in patients with stroke: a systematic review

Abstract

[Purpose] This study aimed to conduct a systematic review the measurement properties of standardized observational gait analysis (OGA) tools for patients with stroke. [Participants and Methods] A systematically search was conducted in PubMed, Cochrane Library, and PEDro databases using stroke- and gait-related keywords. No time restrictions were applied. Studies evaluating OGA using gait analysis tools were included. The methodological quality of the selected studies was assessed using the COSMIN Risk of Bias checklist. [Results] A total of eighteen studies utilizing four gait analysis tools were identified. Sixteen studies employed video-based measurements. The GAIT and WGS tools were assessed for reliability, validity, responsiveness, and interpretability. However, the overall methodological quality of these assessments was not rated as high. [Conclusion] Video-based OGA holds significant potential for clinical applications, but several challenges remain to be met. Standardizing video recording protocols and measurement methods are essential and additional research is needed to determine the qualifications and expertise of the evaluators. Although some studies have shown video-based OGA are effective, determining clinically relevant indicators, such as the Minimal Clinically Important Difference (MCID), is necessary to enhance its applicability in clinical practice.

INTRODUCTION

Gait analysis plays a crucial role in physical therapy, aiding in patient assessment, treatment planning, and the selection of intervention methods¹⁾. The objectives of gait analysis include describing gait patterns, comparing them with normal gait, identifying the causes of functional disorders, evaluating assistive devices, and assessing intervention effects²⁾.

Various gait analysis methods have been developed, among which Instrumented Gait Analysis (IGA) is considered the gold standard for gait assessment and the development of gait indices^{3, 4)}. Three-dimensional motion analysis (3DMA) systems provide detailed gait data by measuring the spatial position and posture of body segments with high precision⁵⁾. However, 3DMA has limitations for routine clinical practice due to its high cost and the requirement for specialized expertise⁶⁾. Given these constraints, Observational Gait Analysis (OGA) is widely used in clinical practice⁷⁾. OGA is practical as it does not require specialized equipment and allows real-time gait analysis^{8, 9)}. However, OGA heavily depends on the evaluator’s observation and experience, making it subjective and potentially inconsistent¹⁰⁾. To reduce subjectivity and enhance the reliability and validity of OGA, standardized gait analysis tools have been developed. Ferrarello et al.¹¹⁾ conducted a systematic review of OGA tools for stroke patients and identified the Gait Assessment and Intervention Tool (G.A.I.T.) as a comprehensive and reliable assessment method¹²⁾. Furthermore, the application of OGA has expanded beyond stroke rehabilitation to include patients with neurological disorders¹³⁾, and pediatric populations¹⁴⁾, and patients with walking disorders¹⁵⁾, emphasizing the need for standardized tools. Previous reviews have assessed the psychometric properties of OGA and confirmed that standardized tools effectively identify abnormal gait^{11, 13,14,15)}. However, these reviews have not comprehensively examined implementation requirements such as evaluator conditions, assessment methods, and video recording standards. The feasibility of gait analysis influenced by factors related to the assessor and influenced by temporal, spatial and physical environmental conditions^{16, 17)}. Therefore, a systematic analysis of these factors is necessary.

This study aims to conduct a systematic review focusing on the measurement characteristics of standardized OGA tools and to verify the reliability and validity of assessments for stroke patients. Specifically, we will examine the impact of rater conditions (e.g. years of experience, expertise, training) and measurement methods (e.g. differences in assessment tools, measurement environment, procedures) on OGA results, and evaluate their characteristics and limitations. The significance of this study is that by clarifying rater conditions and measurement methods, it is expected to provide basic data that will contribute to improving the reliability of measurement methods using standardized OGA tools.

PARTICIPANTS AND METHODS

The review was conducted according to the PRISMA statement^{18, 19)}, the Reporting Guideline for Systematic Reviews, with the following procedures: preparation of a search, database search based on the search formula, primary screening, secondary screening, and analysis. This review was registered with PROSPERO (CRD 42023425209) at the University of York. The search strategy used the PubMed, Cochrane Library, and PEDro databases, and included papers published up to May 29, 2023. To ensure comprehensive literature collection, the reference lists of the included papers were screened and searched manually. After performing a scope search, terms such as (gait, walk, ambulation), (assessment, measurement, evaluation, observation, tool, scale, analysis), (stroke, hemiplegia) were combined using Boolean operators. The included articles were standardized according to the PICO framework (P: stroke patients, I: gait assessment by observation to identify gait disturbances, C: gait analysis tool/functional assessment, O: Psychometric properties) and were required to meet the following eligibility criteria: (1) focus on stroke patients aged 18 years or older, (2) utilize an assessment tool designed for gait analysis based on direct visual observation, (3) include scale properties such as validity, reliability, responsiveness, and interpretability, and (4) be written in English, while studies were excluded if (1) the full text was unavailable, (2) they were review papers, abstracts, or grey literature, or (3) they assessed functional gait using tests such as the Timed Up and Go test (TUG), the 6-min walk test (6MWT), the Functional Gait Assessment (FGA), or the Dynamic Gait Index (DGI). Papers retrieved from the database were imported into Rayyan²⁰⁾. Automatically identified duplicate papers were manually checked. The identified papers were compiled into an Excel spreadsheet (Microsoft Corporation, Redmond, WA, USA), and the paper screening process was initiated. For the primary screening of papers, two reviewers independently reviewed the title and abstract of the paper to determine whether it met the eligibility criteria. If there was a difference of opinion, the reviewers discussed the issue to determine whether the paper should be accepted. For secondary screening, two reviewers retrieved full-text articles and independently reviewed them for eligibility criteria. If there was disagreement, the issue was discussed, and if there was still disagreement, the consensus opinion of the third reviewer was used as the conclusion.

In this study, we used the Consensus-based Standards for the Selection of Health Measurement Instruments (COSMIN) Risk of Bias checklist^{21, 22)} to assess the quality of the identified scales and methodological quality of the studies that validated these scales. The COSMIN Risk of Bias checklist provides criteria for assessing the methodological quality of studies on the measurement properties of health measurement tools, and assessments are made on a four-point scale: inadequate, doubtful, adequate, and very good. The lowest assessment (worst score) in each box is applied to the final methodological quality score for each measurement property.

This checklist consists of 10 boxes covering 9 measurement properties across the three domains of validity, reliability, and responsiveness. In this study, Box 6 (reliability), Box 7 (Measurement Error), Box 8 (Criterion Validity), Box 9 (hypotheses testing for construct validity), and Box 10 (responsiveness) were used.

RESULTS

Database searches identified 2,402 articles 1,507 from PubMed, 774 from Cochrane Library, 116 from PEDro, and 5 from hand searches from which 79 articles were selected for primary screening (Fig. 1). After secondary screening, 18 studies were included, identifying four assessment tools: The Hemiplegic Gait Analysis form (HGAF) in 1 study²³⁾, The Rivermead Visual Gait Assessment (RVGA) in 1 study²⁴⁾, The Gait Assessment and Intervention Tool (G.A.I.T.) in 4 studies^{12, 25,26,27)}, and The Wisconsin Gait Scale (WGS) in 12 studies^{28,29,30,31,32,33,34,35,36,37,38,39)}.

Fig. 1.

PRISMA Flow diagram.

The results for the measurement properties are shown in (Table 1). The rater was identified as a physical therapist in 12 studies^{23, 25, 27, 29, 31,32,33,34,35,36,37, 39)}; However, no detailed description of the evaluator’s educational background or clinical experience was provided. While some studies included information on the evaluator’s prior practice time and methods, they did not specify the nature of the education provided or whether it was part of regular practice. Furthermore, many studies required patients to be capable of independent walking as a condition for participation^{12, 24, 27, 29, 32,33,34,35,36,37,38,39)}.

Table 1. Measurement properties and participant characteristics of studies using gait analysis tools.

Study, reference	Tool	Country	Rater	Pre-measurement practice	Participants	Course of stroke	Level of walking independence
Hughes et al.23)	HGAF	UK	3 Physiotherapists (specializing in stroke rehabilitation)	Practice session conducted but unclear.	6	unclear.	unclear.
Rodriquez et al.28)	WGS	USA	2 Physiatrists	unclear.	18	Participants were at least 1 year poststroke (mean, 2.2; range, 1 to 5 year)	unclear.
Turani et al.29)	WGS	Turkey	1 Physical therapist	unclear.	35	The time from ounclear.et of hemiplegia to admission ranged from 2 to 40 weeks.	were able to walk independently for at least 15 m.
Pizzi et al.30)	WGS	Italy	1 Physiatrist (neurorehabilitative expertise, trained in WGS used)	unclear.	56	The mean time elapsed from stroke was 37 months (range 12–240).	ability to walk 10 m independently with or without a walking device.
Daly et al.12)	G.A.I.T.	USA	Inexperienced 1 clinician experienced 1 clinician	Inexperienced clinician (three, 1.5-h session) experienced clinician unclear, (academic background in gait assessment needed only 2-h).	29	12 months or more after stroke	unclear.
Zimbelman et al.25)	G.A.I.T.	USA	1 Physical therapist: (10 years of experience in clinical practice and gait assessment, and more than 3 years of experience in using the G.A.I.T. and TGS measures)	unclear.	44	greater than 6 months after a single stroke.	unclear.
Yaliman et al.31)	WGS	Turkey	2 Physiatrists2 Physical therapists	Received a training session about gait disorders in hemi-plegic patients and a copy of the WGS scoring criteria by principal investigator before they viewed the video- tapes.	19	The time from ounclear.et of CVA to admission ranged from 3 to 9 months.	Independent walking possible.
Lu et al.32)	WGS	China	2 Physical therapists (One was a novice with only two years’ working experience and the other was an experienced PT with 9 years’ work experience)	Before scoring, they were trained in the scales’ scoring instruction unclear., and to make common agreement on each item in one session.	20	unclear.	Independent walking possible.
Wellmon et al.33)	WGS	USA	14 Physical therapists. study enrollment required a minimum of 1 year of full-time clinical experience. (The raters had some experience examining gait in individuals with health condition unclear. that affected walking. However, they have never used WGS.	On the first day of the study, the raters were given time to review the WGS and its scoring criteria before using the instrument to rate the videotapes of the individuals post-stroke.	6	Chronic stroke mean 31.2 months (5–106 months).	Independent walking possible.
Guzik et al.34)	WGS	Poland	experienced 1 Physiotherapist (who had been trained in gait disorders affecting post-stroke hemiparetic patients and had knowledge of assessment criteria used in WGS)	unclear.	30	chronic stage after stroke, over 6 months.	Independent walking possible.
Guzik et al.35)	WGS	Poland	1 Physiotherapist (10 years of experience in working with patients who had experienced a stroke, The physiotherapist was trained in the use and interpretation of the WGS)	unclear.	50	Time from stroke month, mean 42.0, range, 8–120.	Independent walking possible.
Guzik et al.36)	WGS	Poland	1 Physiotherapist	unclear.	50	Time from stroke month, mean 42.0, range, 8–120.	Independent walking possible.
Guzik et al.37)	WGS	Poland	1 Physiotherapist	unclear.	100	Time from stroke month, mean 42.0, range, 8–120.	Independent walking possible.
Estrada-Barranco et al.38)	WGS	Spain	unclear.	unclear.	61	Patients whose stroke occurred within 8 weeks.	unclear.
Arya et al.24)	RVGA	India	2 raters had more than 20 years of experience in neurorehabilitation. 2 raters were novices in the field.	Prior study, three trial videos were scored by the raters to acquaint them with the method of assessment.	40	unilateral stroke of >6 months	able to walk for at least 10 m without any physical assistance, foot orthosis, or walking device.
Guzik et al.39)	WGS	Poland	1 Physical therapist (over 10 years of experience in working with patients post-stroke, and with expertise using the WGS and interpreting the scores)	unclear.	50	Time from stroke month, mean 42.0, range, 8–120.	Independent walking possible.
Saengsuwan and Vichiansiri26)	G.A.I.T.	Thailand	2 Physiatrists	unclear.	31	Median time after stroke was 45 days.	Able to walk 10 m independently with or without a gait aid.
Smith et al.27)	G.A.I.T.	Australia	1 Physiotherapist	unclear.	65	mean time post-stroke 12 weeks	unclear.

CVA: cerebrovascular accident; FAC: functional ambulation category; FIM: functional independence measure; BI: Barthel index; GDI: gait deviation index; GVI: gait variation; PASS: postural assessment scale for stroke patients; HGAF: hemiplegic gait analysis form; RVGA: Rivermead visual gait assessment; G.A.I.T.: gait assessment and intervention tool; WGS: Wisconsin gait scale.

The measurement Properties for OGA are shown in (Table 2). In the 16 studies that used gait analysis tools, video was used in^{12, 23, 24,25,26,27,28, 30,31,32,33,34,35,36,37, 39)}. However, none of the studies described details such as the focal length, which is important for determining the approximate distance between participants. In one study from 2019, a smartphone was used²⁴⁾. The gait distance required for OGA necessitated an environment with at least a 10 m straight line^{12, 23, 24, 26, 27, 31, 32, 35,36,37, 39)}, while shooting directions included the frontal and sagittal (front/back and side) planes^{23, 24, 30,31,32,33,34,35, 39)}. In the study by Guzik et al., the distance between the participant and the camera was set at 2 meter^{35,36,37,38,39)}. To capture continuous images of the patient’s entire body, some studies manually zoomed and, in the sagittal plane, moved with the patient to reduce parallax³³⁾. Regarding the projection method for evaluating gait, the study by Guzik et al.³³⁾ used a portable projection screen measuring 1.52 meter in length and width. Assessment times ranged from 20 to 25 min for WGS when videotaping patients³⁰⁾ and 10 to 30 min per patient for analysis^{30, 33, 37)}, while GAIT analysis time was 20 min³⁸⁾. However, the remaining 14 studies did not describe the time required for videotaping or assessment.

Table 2. Measurement properties of reviewed studies.

Study, Reference	Tool	Preferred speed or normal speed	No. of cameras	Video setup (How to take videos)	Video projection	Video views	Number of walking trials	Assessment time
Hughes et al.23)	HGAF	10 m/unclear.	2 VHS camcorders. (Panasonic)	Lateral and anteroposterior views.	unclear.	Split-screen film showed horizontal, lateral, and anteroposterior views simultaneously, with slow-motion and freeze-frame options.	4 trials.	unclear.
Rodriquez et al.28)	WGS	unclear.	Camera is used, but unclear.	unclear.	unclear.	4 times.	4repeated gait	unclear.
Turani et al.29)	WGS	unclear.	NA	NA	NA	NA	unclear.	unclear.
Pizzi et al.30)	WGS	10 m/comfortable speed.	2 VHS video-cameras. (Panasonic Digital Video-camera DS35)	One camera was placed 4 meters in front of the walking platform, while another on a trolley followed the patient laterally at a 2.5-meter distance, capturing footage of both sides in the lateral plane.	unclear.	unclear.	4 trials.	Video recording: about 20–25 min.Tape analysis: about 15–20 min one patient.
Daly et al.12)	G.A.I.T.	30 ft/unclear.	Camera is used, but unclear	unclear.	unclear.	unclear.	unclear.	20 min.
Zimbelman et al.25)	G.A.I.T.	unclear/unclear.	Camera is used, but unclear	unclear.	NA	unclear.	unclear.	unclear.
Yaliman et al.31)	WGS	10 m/unclear.	Camera is used, but unclear	The camera was placed 1 meter from the patient. The patient was then instructed to make three turns, which were recorded from the front, side, and rear.	unclear.	during the viewing of the video tapes and were given only 1 opportunity to view each patient during scoring.	unclear.	unclear.
Lu et al.32)	WGS	10 m/unclear.	2 cameras.	Patients were asked to walk three to four times to allow recording them from the sides, front and back.	unclear.	unclear.	3–4 trials.	unclear.
Wellmon et al.33)	WGS	7.62 m/Preferred speed or normal speed.	4 cameras.	We used fixed tripods for anterior and posterior views, manually adjusting zoom for continuous capture. For lateral views, rolling dollies were used, moving with participants to minimize parallax and optimize gait kinematics, spatial, and temporal parameters.	The participants’ walking video was projected onto a 1.52 m × 1.52 m portable screen.	The videotapes were scored using the WGS with multiple raters in the same room simultaneously viewing the screen.	4 trials.	Within 10 min.
Guzik et al.34)	WGS	unclear/self-selected speed.	2 video cameras.	right and left side, as well as back and front view.	unclear.	unclear.	6 trials.	unclear.
Guzik et al.35)	WGS	10 m/comfortable speed.	2 video cameras (BTS Vixta, BTS Bioengineering Corp).	We took photographs from both the frontal and sagittal planes. In the frontal plane, we used one camera aligned with the direction of the participants’ gait. In the sagittal plane, we positioned a second camera halfway along the walking path at a distance of 2 m from the path.	unclear.	unclear.	6 trials.	unclear.
Guzik et al.36)	WGS	10 m/comfortable speed.	2 video cameras.	Two cameras captured participants’ gait: one aligned with the frontal plane and the other in the sagittal plane, 2 m from the walking path.	unclear.	unclear.	6 trials.	unclear.
Guzik et al.37)	WGS	10 m/comfortable speed.	2 video cameras.	Two cameras captured participants’ gait: one aligned with the frontal plane and the other in the sagittal plane, 2 m from the walking path.	unclear.	unclear.	6 trials.	30 min.
Estrada-Barranco et al.38)	WGS	unclear/unclear.	unclear.	unclear.	unclear.	unclear.	unclear.	unclear.
Arya et al.24)	RVGA	15 m/unclear.	Video-cameras (smartphone, unclear).	Videotaped from the anterior aspect, posterior aspect, affected side, and less-affected side.	The rater played the coded video clip using VLC Media Player 2.2.1.	The playing speed of the clip was slowed to half.	4 trials.	unclear.
Guzik et al.39)	WGS	10 m/comfortable speed.	Two video cameras. (BTS Vixta, BTS Bioengineering Corp.)	We used two cameras to capture gait: one aligned with the frontal plane and another in the sagittal plane, 2 m from the path and halfway along it.	unclear.	unclear.	6 trials.	unclear.
Saengsuwan and Vichiansiri26)	G.A.I.T.	10 m/comfortable speed.	Camera is used, but unclear.	unclear.	unclear.	unclear.	2 trials.	unclear.
Smith et al.27)	G.A.I.T.	10 m/comfortable speed.	Camera is used, but unclear.	video was recorded as per the GAIT instructions.	unclear.	unclear.	3 trials.	unclear.

HGAF: hemiplegic gait analysis form; RVGA: Rivermead visual gait assessment; G.A.I.T.: gait assessment and intervention tool; WGS: Wisconsin gait scale.

Psychometric properties are presented in (Table 3). Seven studies focused on reliability^{12, 23, 24, 31,32,33)} and 14 studies investigated validity^{12, 23,24,25, 27,28,29, 32, 34,35,36,37,38,39)}. Additionally, five studies^{12, 25, 28,29,30)} tested responsiveness, and three of his studies^{26, 27, 36)} measured the minimum clinically important difference MCID.

Table 3. Outcome measures of reviewed studies.

Study, Reference	Tool	Reliability	Validity	Responsiveness, MCID
Hughes et al.23)	HGAF	intra-rater reliability: KCC=0.40–0.95 (Total Scores)	speed: r=0.60
		inter-rater reliability: KCC=0.94–0.95 (Total Scores)	step length symmetry: r=−0.09
			single support symmetry: r=0.94
Rodriquez et al.28)	WGS	Inter-rater reliability: r=0.44–0.85	Physical functioning score (HSQ subscale): r=0.64	WGS score: before and after intervention: 26.8/23.9 (p<0.05)
Turani et al.29)	WGS		Brunnstrom Recovery stages	WGS admission/ WGS discharge: 29.1/25.9 (p<0.007)
			WGS A: r=−0.46, WGS D: r=0.54	WGS admission/ WGS discharge: r=0.78 (p<0.01)
			velocity A
			WGS A: r=−0.45, WGS D: r=−0.54
Pizzi et al.30)	WGS			Wilcoxon signed-rank test,
				28 before intervention, 26.5 after intervention: p<0.003
Daly et al.12)	G.A.I.T.	Intra-rater reliability: ICC (0.98, 95% CI=0.95–0.99)	G.A.I.T. item 26 score and the motion capture data r=0.65	PoLytomous Universal Model. For comprehensive GAIT training without FES-IM, z=−2.93, p<0.003.
		Inter-rater reliability: ICC (0.83, 95% CI=0.32–0.96)	G.A.I.T. item 27 score and the motion capture data r=0.76	For comprehensive GAIT training, with FES-IM, z=−3.3, p<0.001.
		experienced clinician and inexperienced: ICC (0.99, 95% CI=0.98–0.99)
Zimbelman et al.25)	G.A.I.T.		G.A.I.T. and speed: r=0.73	G.A.I.T.: (pretreatment minus posttreatment; pretreatment minus midtreatment; midtreatment minus posttreatment)
				Wilcoxon signed-rank test, p<0.005. According to the G.A.I.T., 40 participants (91%) showed improved scores, 2 (4%) no change, and 2 (4%) a worsening score.
Yaliman et al.31)	WGS	Cronbach’s α was 0.91 (Day 1) and 0.94 (Day 3).
		Inter-rater reliability: ICC=0.91–0.96
		Intra-rater reliability: ICC=0.75–0.90
Lu et al.32)	WGS	WGS total score	WGS and speed: r=−0.81
		Intra-rater reliability ICC=0.96 (95% CI=0.90–0.98)	WGS and FMA: r=−0.67
		Intar-rater reliability ICC=0.94 (95% CI=0.77–0.96)	WGS and MI: r=−0.68
			WGS and CSI: r=0.30
Wellmon et al.33)	WGS	Intra-rater reliability: ICC=0.91 (95% CI=0.85–0.94)
		Inter-rater reliability: ICC=0.83 (95% CI=0.63–0.97)
		SEM=1.47
		MDC95=4.24
Guzik et al.34)	WGS		WGS and velocity: r=−0.39
			GAIT cycle and WGS, unaffected: r=0.36, affected: r=0.37
Guzik et al.35)	WGS		WGS and 3D, (0.7≤ |R|<0.9)
Guzik et al.36)	WGS		WGS and BI: r=0.63	MCID: 2.25 points
Guzik et al.37)	WGS		Total score
			GDI affected leg and WGS: r=−0.87
			GVI affected leg and WGS: r=−0.93
			GVI unaffected leg and WGS: r=−0.88
			spatiotemporal parameters
			GVI affected leg and WGS (items: 2, 3, 5, 6): r=−0.81
			GVI unaffected leg and WGS (items: 2, 3, 5, 6): r=−0.81
			GDI affected leg and WGS (items: 4, 7–14): r=−0.85
Estrada-Barranco et al.38)	WGS		First assessment;
			FAC: r=−0.77, BBS: r=−0.67, PASS: r=−0.64,
			BI: r=−0.65, FIM: r=−0.59
			3 months;
			FAC: r=−0.87, BBS: r=−0.88, PASS: r=−0.84,
			BI r=−0.81, FIM r=−0.69
			6 months;
			FAC: r=−0.90, BBS: r=−0.81, PASS: r=−0.89,
			BI: r=−0.86, FIM: r=−0.80
			one year;
			FAC: r=−0.88, BBS: r=−0.90, PASS: r=−0.89,
			BI: r=−0.81, FIM: r=−0.82
Arya et al.24)	RVGA	Inter-rater reliability:	RVGA Vs BBS
		experienced therapists: r=0.94 (0.88–0.96)	experienced therapists: r=0.47–0.48 (95% CI=0.19–0.69)
		novice therapists: r=0.94 (0.89–0.96)	novice therapists: r=0.38–0.46 (95% CI=0.08–0.67)
		Intra-rater reliability:	No correlation was observed between RVGA and FMA-LE, 10MWT, and TUG (p>0.05)
		experience rater: r=0.95 (0.90–0.97)
		novice rater: r=0.95 (0.91–0.97)
Guzik et al.39)	WGS		3D symmetry indexes: Stance Time (s), Stance %, Hip FE ROM, and Knee FE ROM (0.7 ≤|R|<0.9; 0.9 ≤|R|<1)
			Step Length (m) SI (0.3≤|R|<0.5)
Saengsuwan and Vichiansiri26)	G.A.I.T.			G.A.I.T. score: Initial, 30 days (16.9 ± 10.2, 13.1 ± 9.5, p<0.001)
				CGS≥0.06 m/s: MCID=2.5, AUC=0.76 (95% CI=0.58–0.95)
				Clinicians GROC≥+3: MCID=4.0, AUC=81.8 (95% CI=48.2–97.7)
				Participants GROC≥+3: MCID=1.5, AUC=0.71 (95% CI=0.31–1.00)
Smith et al.27)	G.A.I.T.		G.A.I.T. Vs FAC: r=−0.73	FAC level 3 or HA, the MCID is 11.48 (CI=7.59–15.36).
			G.A.I.T. Vs ambulatory level: r=−0.69	FAC level 4 and 5 or LCA and CA the MCID is 5.19 (CI=2.01–8.37)
			G.A.I.T. Vs speed: r=−0.79

KCC: Kendall’s coeficient of concordance; CSI: composite spasticity index; MI: motricity index; FMI: Fugl Meyer assessment; GROC: global rating of change; AUC: area under the curve; HA: household ambulator; LCA: limited community ambulator; CA: community ambulator; BBS: Berg balance scale; FE: flexion/extension; ROM: range of motion; A: admission; B: discharge, FAC: Functional Ambulation Category; MCID: minimal clinically important difference; ICC: intraclass correlation coefficients; MCD95: minimal detectable change 95; HGAF: hemiplegic gait analysis form; RVGA: Rivermead visual gait assessment; G.A.I.T.: gait assessment and intervention tool; WGS: Wisconsin gait scale.

The results of the COSMIN Risk of Bias checklist are presented in (Table 4). Box 6 (Reliability) was assessed in seven studies^{12, 24, 28, 31,32,33)} and was rated as “very good” in one study³³⁾. The primary reason for not achieving a “very good” rating in other studies was the lack of a detailed description of the model and calculation formula used for ICC. Box 8 (Criterion Validity) was rated as “very good” in four studies^{12, 34, 35, 39)} and “inadequate” in two studies^{32, 38)}. Box 10 (Responsiveness) was rated as “inadequate” in five studies^{12, 25, 29, 30)}.

Table 4. Methodological quality results with COSMIN risk of bias checklist.

Study	Box 6	Box 7	Box 8	Box 9	Box 10
Hughesl et al.23)	Doubtful			Doubtful
Rodriquez et al.28)	Doubtful			Inadequate	Doubtful
Turani et al.29)				Inadequate	Inadequate
Pizzi et al.30)					Inadequate
Daly et al.12)	Adequate		Very good		Inadequate
Zimbelman et al.25)					Inadequate
Yaliman et al.31)	Adequate
Lu et al.32)	Doubtful			Adequate
Wellmon et al.33)	Very good	Very good
Guzik et al.34)			Very good
Guzik et al.35)			Very good
Guzik et al.36)					Adequate
Guzik et al.37)				Adequate
Estrada-Barranco et al.38)				Very good
Arya et al.24)	Inadequate			Adequate
Guzik et al.39)			Very good
Saengsuwan and Vichiansiri26)					Adequate
Smith et al.27)					Very good

DISCUSSION

This systematic review focused on the measurement properties of gait analysis in stroke patients. The review by Ferrarello et al.¹¹⁾ included eight papers describing five gait analysis tools (NYMSOGA⁴⁰⁾, HGAF, WGS, GAIT, and RVGA). In this review, NYMSOGA was excluded because it did not meet the eligibility criteria as it was poorly constructed and mostly incomplete. Furthermore, the same gait analysis tools have been used for stroke patients, and no new tools have been developed. Of the 18 studies selected for this study, WGS was selected in 12 studies, and GAIT was selected in 4 studies. GAIT uses 31 items to measure gait impairment related to the coordinated elements of gait¹²⁾. The WGS consists of 14 items that assess clinically relevant gait components and is a low-burden option that is expected to be adopted in clinical practice. This approach was extensively validated by Guzik et al^{34,35,36,37, 39)}. Sixteen OGA studies using video cameras have been conducted, and these studies have shown that the complexity and breadth of the gait phenomenon makes it difficult to interpret through direct observation and real-time analysis. Video recording is required for GAIT analysis, and the conditions for this are clearly described in the literature¹²⁾. On the other hand, different video recording protocols were used in each study included in the WGS.

In this review, walking distance was standardized to approximately 10 m in most studies. The distance between the camera and the participant in the sagittal plane was usually 1 to 2.5 meters^{30, 31, 33, 35,36,37, 39)}. In some studies, strategies have been employed to minimize camera parallax, such as moving with the patient during the gait test or using a zoom function to capture the entire body^{30, 33)}.

No previous studies have offered comprehensive descriptions of camera performance. Although establishing a universal guideline for object distance is challenging, it is essential to consider these considerations when determining the measurement range. Many studies lack documentation regarding the projection method and viewing procedures for video-based OGA. Some studies using video projection used a screen measuring 1.52 meters in both the vertical and horizontal dimensions. In addition, some studies have allowed unrestricted use of slow-motion and freeze-frame techniques during video viewing²³⁾. When conducting OGA through video recording, mitigating measurement errors and inadvertent mistakes without clearly defined video shooting and projection conditions becomes challenging. In this review, Wellmon³³⁾ was the sole contributor that calculated the Standard Error of Measurement (SEM) and Minimal Detectable Change at the 95% confidence level (MDC95). Recently, there have been reports of utilizing the slow-motion function of smartphones for OGA in cerebral palsy patients^{41, 42)}. It is anticipated that gait analysis will be conducted on diverse patient groups using the camera sensor function of smartphones. Therefore, the development of a standardized measurement protocol is imperative to ensure consistency and accuracy of these assessments.

The COSMIN risk of bias checklist predominantly resulted in evaluations of “Inadequate” or “Doubtful”, indicating concerns about the methodological quality of the study. In the reliability section of Box 6, only Wellmon et al.³³⁾ demonstrated a “very good” rating, whereas others lacked a presentation of the ICC model, contributing to this variation. Four studies^{12, 34, 35, 39)} compared OGA and 3DMA gait analysis to assess their criterion-related validity. The absence of a clearly defined gold standard presents a significant challenge. The conceptual validity of scales measuring gait speed, motor function, and functional disability has been evaluated. However, the validation of the four extracted gait analysis tools remains limited, as previous studies have primarily focused on specific assessment scales.

A review published in 2013¹¹⁾ compared only the GAIT analysis with three-dimensional motion analysis. More recently, WGS has been validated for assessing spatiotemporal gait parameters through three-dimensional gait analysis, demonstrating its reliability^{34, 35, 39)}. In addition, the MCID for walking ability in stroke patients has been estimated using both WGS and GAIT in recent years^{26, 27, 36)}. According to Guyatt et al.⁴³⁾ the MCID is defined as the change in outcome score corresponding to a clinically significant change recognized subjectively by the patient.

The MCID for walking assessment tools is a crucial indicator of the threshold of substantial and meaningful improvement or deterioration as perceived by patients and clinicians. This step helps determine therapeutic efficacy based on assessment tool scores surpass this threshold. In the gait analysis of stroke patients, it is essential to use assessment tools that are not only applicable in clinical settings and possess well-validated psychometric properties, rather than relying solely on subjective or real-time assessment. One such example is video-based OGA. Furthermore, clarifying specific indicators such as the MCID can enhance the utility of these assessment tools and further promote their practical application in clinical settings.

A limitation of this study was that the inclusion criteria relied solely on papers written in English, and the use of a limited set of search databases may have restricted the comprehensive inclusion of all relevant studies. Furthermore, although rigorous training was conducted for methodological assessment, the evaluation was based on a consensus among multiple reviewers, and the possibility of inconsistencies cannot be entirely ruled out because of the lack of rigorous uniformity in the review process.

This study focused on OGA in stroke patients, identifying four gait analysis tools, among which 16 studies utilized videography for assessment, and both GAIT and WGS were assessed for reliability, validity, responsiveness, and interpretability. However, neither of the methodological quality tools used in this study was rated as high-quality, emphasizing the need for further refinement.

Future research should standardize the measurement properties and videography protocols to enhance the accuracy and reliability of the gait analysis tools. Further clarification of the MCID and other psychometric properties is essential to improve the clinical utility of these assessment tools and to facilitate their practical application in clinical settings.

Conflict of interest

The authors have no conflicts of interest.