Clinical library · Outcome measures
Functional Gait Assessment
Ten walking tasks scored out of 30, built specifically to remove the ceiling effect that limits the Dynamic Gait Index. It is the measure to reach for when a patient walks well enough to top out the Berg and still falls.
In one line. Ten gait tasks — level walking, changing speed, head turns in both planes, pivot turn, stepping over an obstacle, narrow base of support, eyes closed, backwards walking and stairs — each scored 0 to 3, totalling 30. It takes around ten minutes and needs a marked walkway, an obstacle and a staircase.
It exists because the Dynamic Gait Index ran out of headroom. Three harder tasks were added and one easier one removed, and the result is a walking-balance measure that keeps discriminating in people who are doing relatively well.
The numbers you actually need
| Property | Value | Source and caveat |
|---|---|---|
| Falls cut-off, community-dwelling older adults | 22 or below out of 30 | 35 adults aged 60 to 90, falls tracked prospectively for 6 months. 100% sensitivity, 72% specificity, LR+ 3.6, LR- 0. Small sample, short follow-up [1] |
| Cut-off, stroke versus healthy | 20 or above | 63 people with stroke and 30 healthy older adults; effectively differentiated the groups [4] |
| Minimal detectable change, stroke | 6 points (SEM 2) | 52 people, median 6 weeks post-stroke, independent walkers [3] |
| Minimal detectable change, community stroke | 5.02 points (SEM 1.81) | Inter-rater ICC 0.94, test-retest ICC 0.85 [4] |
| Reliability, incomplete spinal cord injury | Inter-rater ICC above 0.92; intra-rater above 0.91 | Only 12 participants, chronic motor incomplete injury, AIS C or D [5] |
| Concurrent validity | r = 0.84 with Berg; r = -0.84 with TUG | Community-dwelling older adults [1]. In stroke, correlations of 0.61 to 0.83 with timed walking tests, Berg and Stroke Impact Scale mobility [3] |
| Ceiling effect | None demonstrated | Rasch analysis of 179 older adults found no ceiling effect [2]; no relevant ceiling effect after stroke either [3] |
What it measures
Walking balance under progressively harder conditions: changing speed, moving the head while walking, turning, negotiating an obstacle, narrowing the base of support, removing vision, reversing direction, and stairs. Unlike the Berg Balance Scale, which is largely static and anticipatory, the tasks here are dynamic and gait-based.
Rasch modelling confirmed that the task hierarchy broadly matches clinical expectation, with one exception worth knowing: "walking on level surface" located in the middle of the difficulty continuum rather than at the easy end. Two of the three added tasks were the most difficult items. [2]
Where it misleads
1. The 22/30 cut-off comes from 35 people
The falls cut-off quoted everywhere derives from a prospective cohort of 35 community-dwelling older adults followed for six months. The performance figures are striking — 100% sensitivity, 72% specificity, negative likelihood ratio of 0 — and the authors themselves list the small sample and short follow-up as limitations, noting the sample did not allow criterion scores to be evaluated by decade. [1] A likelihood ratio of 0 from 35 participants is a small-sample artefact as much as a finding.
2. It is not one cut-off, it is several
22 or below identifies falls risk in community-dwelling older adults; [1] 20 or above separates people with stroke from healthy adults. [4] These answer different questions in different populations and are not interchangeable. Applying the older-adult falls threshold to a stroke caseload is a category error.
3. A low score does not tell you why
Total score correlates with knee extensor strength — both maximum torque and rate of torque development — and with reduced postural sway in static balance tasks, though not with weight-shifting performance. Walking backwards and stair climbing had the highest correlations with strength. [6] So a poor total may reflect weakness rather than a balance deficit, and the pattern across items is more informative than the total.
4. The spinal cord injury data rest on twelve people
Reliability in incomplete spinal cord injury is reported as excellent, with inter-rater ICCs above 0.92 and validity against the 10-Metre Walk Test of -0.90. The study had 12 participants, 11 of them male, all with chronic motor incomplete injury. [5] That is a promising signal, not an established property.
What the evidence supports — and what it does not
Supported
- Assessing walking balance without a ceiling effect, including after stroke. [2][3]
- Higher-functioning patients where the Berg saturates. [2][3]
- Excellent reliability across older adults, stroke and incomplete spinal cord injury. [3][4][5]
- Strong concurrent validity against Berg and timed walking tests. [1][3]
- A measurement error of about 5 to 6 points after stroke when interpreting individual change. [3][4]
Not supported
- Transferring the 22/30 falls cut-off between populations. [1][4]
- Treating the 35-person cut-off study as definitive. Its own authors list sample size and follow-up as limitations. [1]
- Reading the total score as a pure balance measure. It is influenced by knee extensor strength. [6]
- Generalising the spinal cord injury findings. Twelve participants. [5]
- Assuming item difficulty is intuitive. Level walking sits in the middle of the hierarchy, not at the bottom. [2]
How certain is this?
Evidence grade: Low to moderate.
The strongest element is the absence of a ceiling effect, which is the reason the measure exists. It is supported by Rasch analysis of 179 older adults [2] and independently by a stroke cohort that found no relevant ceiling effect where other capacity measures showed one. [3] That is the claim this page would defend most firmly.
Reliability is consistently excellent, but each estimate comes from a modest single-centre sample: 52 people after stroke, [3] 63 in the community stroke study, [4] and 12 with incomplete spinal cord injury. [5]
The falls cut-off is the weakest widely quoted figure on this page and is presented here with its sample size attached for that reason. [1] The strength relationship is a cross-sectional correlation study in 46 healthy older adults and describes association, not mechanism. [6]
What would change the grade: a large prospective validation of the falls threshold with follow-up beyond six months, and MDC values in populations other than stroke.
Common questions
When should I choose this over the Berg?
When the patient walks well enough that the Berg is likely to saturate, or when the clinical question is walking balance rather than static and anticipatory balance. Rasch analysis found no ceiling effect in older adults, [2] and after stroke it yielded no relevant ceiling effect in contrast to other capacity measures. [3] The two correlate strongly (r = 0.84), [1] so they are measuring related constructs at different difficulty levels.
What change can I defend in one patient?
After stroke, about 5 to 6 points. One study reported a standard error of measurement of 2 and a minimal detectable change of 6 points, and advises taking a 6-point measurement error into account when interpreting change. [3] A community stroke sample gave SEM 1.81 and MDC 5.02. [4] Outside stroke, published MDC values are not available.
Is 22 out of 30 the falls threshold?
In community-dwelling older adults, that is the published figure: a score of 22 or below gave 100% sensitivity and 72% specificity for prospective falls, with LR+ 3.6. But it comes from 35 people followed for six months, and the authors flag the sample size and follow-up as limitations. [1] Treat it as a reasonable flag rather than a validated decision rule, and do not carry it into other populations — in stroke the discriminating cut-off was 20 or above. [4]
Can I use it in incomplete spinal cord injury?
The available evidence is encouraging and very small. Inter- and intra-rater reliability were excellent, with ICCs above 0.92 and 0.91 respectively, and validity against the 10-Metre Walk Test was -0.90. There were 12 participants. [5] Reasonable to use; not reasonable to quote as established.
My patient scores badly. Is that balance or strength?
Look at which items they lost. Total score correlates with knee extensor strength, and walking backwards and stair climbing had the highest correlations with strength measures, while the item challenging proprioceptive input was associated with static balance. [6] The item pattern is the useful clinical information; the total is a summary.
References
- Wrisley DM, Kumar NA. Functional gait assessment: concurrent, discriminative, and predictive validity in community-dwelling older adults. Physical Therapy. 2010 May;90(5):761–73. doi:10.2522/ptj.20090069 PMID 20360052 Prospective cohort study
- Beninato M, Ludlow LH. The Functional Gait Assessment in Older Adults: Validation Through Rasch Modeling. Physical Therapy. 2016 Apr;96(4):456–68. doi:10.2522/ptj.20150167 PMID 26337259 Rasch analysis
- Van Bloemendaal M, Bout W, Bus SA, et al. Validity and reproducibility of the Functional Gait Assessment in persons after stroke. Clinical Rehabilitation. 2019 Jan;33(1):94–103. doi:10.1177/0269215518791000 PMID 30084264 Cross-sectional study
- Ng SSM, Chen P, Li S, et al. Psychometric properties of functional gait assessment in people with stroke. BMC Neurology. 2026 Jan 3;26(1):83. doi:10.1186/s12883-025-04591-w PMID 41484708 Cross-sectional study
- Kahn JH, Ohlendorf A, Olsen A, et al. Reliability and Validity of the Functional Gait Assessment in Incomplete Spinal Cord Injury. Topics in Spinal Cord Injury Rehabilitation. 2020;26(4):268–274. doi:10.46292/sci19-00069 PMID 33536732 Reliability and validity study
- Karabin MJ, Sparto PJ, Rosano C, et al. Impact of strength and balance on Functional Gait Assessment performance in older adults. Gait & Posture. 2022 Jan;91:306–311. doi:10.1016/j.gaitpost.2021.10.045 PMID 34800923 Cross-sectional study
About this resource
Using this in clinic
Every figure here is traceable to its source.
Every cut-off and measurement error on this page carries the sample size it was derived from, because several of the most quoted figures for this measure come from very small studies. Where a value could not be verified against the paper it came from, it is not on this page, and the omission is stated rather than filled with a number from a secondary source.