Searching and screening

We conducted a systematic search in Pubmed and Embase using a protocol based on the PRISMA statement for conducting and reporting systematic reviews [16]. No language or time restrictions (up to April 15 2020) were applied. We identified additional articles by searching cited references of relevant articles. The detailed search strategy is shown in the Supporting information.

We included original and published articles that compared two methods of measuring gait speed over a maximum distance of 10 meters and that reported quantitative data on gait speed in m/sec or the equivalent in adults. We excluded articles that assessed only one measurement method, reported results of the 6-minute walk test, compared different computerized electronic walkway systems, or only reported comparision of usual versus maximal gait speed. Non-English articles, letters, reviews, and editorials were also excluded. Titles and abstracts were screened by two independent reviewers, who then performed a full-text screening based on the eligibility criteria. Discrepancies were resolved by a third reviewer.

Publications that included pairwise comparisons between testing methods for usual gait speed were identified. If a study made a comparison in the same study population more than once (e.g. for reliability purposes), the results of the first comparison were extracted for the systematic analysis.

Data extraction

Two reviewers independently extracted information using a standardized predefined data extraction tool. Additional methodological references and appendices included in the reviewed article were consulted as well. We extracted study and patient characteristics, descriptive data for each pairwise comparison and corresponding gait speed results. Two investigators independently extracted data, and disagreements were resolved by a third reviewer.

Risk of bias

We assessed the methodological quality of included studies based on the Agency for Healthcare Research and Quality (AHRQ) methodology checklist. The checklist domains assess inclusion of: the source of information, inclusion and exclusion criteria, time period, consecutive recruitment, masking, quality assurance, controlling of confounders, handling of missing data, response rate and completeness of data. This 11-item checklist is considered applicable for cross-sectional studies [17]. Each item is scored “yes” (1 point), “no” (0 point) or “not applicable”. Two independent reviewers used the checklist to score each study and a total score was calculated. Final scores ranged from 0–11, where 11 points suggested “lowest risk of bias” and 0 corresponded to “highest risk of bias”. A difference of 2 or more points was considered discrepant and decided by consensus. From the AHRQ checklist, we calculated a percentage for the overall risk of bias for each study dividing the number of items with 0 points by the total number of applicable items (i.e. total number of 11 items minus the number of non-applicable items for each study).

Analysis

For qualitative synthesis, stratification was performed by the type of pairwise comparison: 1) starting procedures (e.g. static (no acceleration) vs. dynamic start (distance for acceleration)), 2) distance procedures (e.g. 4 m vs. 10 m), 3) timing procedures (e.g. manual vs. automatic), 4) surface procedures (e.g. hard vs. soft ground), 5) walkway procedures (e.g. electronic walkway vs. overground surface) and 6) finishing protocols (deceleration vs. no deceleration).

In a summative approach, we combined results of included studies by calculating medians and ranges of absolute and relative differences of gait speed results and intraclass correlation coefficients, stratified by the type of pairwise comparison. Descriptive data are shown as numbers and percentages for categorical variables, and median and ranges for numerical variables. We considered a gait speed difference of 0.1 m/sec as clinically relevant, based on a systematic review on the minimal clinically important difference (MCID) of gait speed measurement [18].

Study selection

A total of 2109 studies were identified through the search strategy (Fig 1). Of these, 2030 studies were excluded after screening the abstract, leaving 79 studies for full-text screening to assess eligibility. Overall, 23 publications comprising 29 study reports and 53 pairwise comparisons were included in the systematic synthesis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart.

https://doi.org/10.1371/journal.pone.0234200.g001

Study report characteristics

Table 1 describes descriptive characteristics of included study reports (n = 29). Overall, 47% of participants were female with a mean age of 62.9 years (SD 16.0). The majority of reports recruited study participants from community settings (n = 23), although three reports were conducted with rehabilitation patients [19–21], and one with hospitalized patients [22]. Two reports did not specify the setting [23, 24]. Thirteen reports (45%) specifically targeted patients suffering from an underlying disease (stroke n = 5 [19, 20, 25], Parkinson disease n = 3 [24, 26, 27], and pulmonary disease n = 2 [21, 28]). The majority of study reports were observational cohort studies (n = 27), but two reports used a case-control design [29]. Overall, only four were multicenter studies [20, 30–32]. Thirteen reports investigated intertest (i.e. intra-rater) reliabilty and/ or agreement [19, 25, 27, 28, 31–36], while only two reports analyzed interobserver (.i.e. inter-rater) reliability and/or agreement [21, 37].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of included study reports (n = 29).

https://doi.org/10.1371/journal.pone.0234200.t001

Most reports evaluated the impact of one testing procedure (e.g. timing) on the outcome of gait speed. However, some reports evaluated pairwise comparisons of different test procedures in the same study population (e.g. one pairwise comparison on timing methods and another pairwise comparison on distance).

Morever, Table 1 depicts the risk of bias for each report. Overall, we found the largest median risk of bias in surface test procedures (75.0%) and the lowest median risk of bias in timing test procedures (41.7%) (Table 2). Overall, 17 of 29 reports had a high risk of bias of >50% and 12 reports had a low risk of bias of less than 50%.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Impact of testing procedures on gait speed results of pairwise comparisons (n = 53).

https://doi.org/10.1371/journal.pone.0234200.t002

Characteristics of pairwise comparisons

Overall, 29 study reports had information on 53 pairwise comparisons. Among these pairwise comparisons, 14 (26%) compared different approaches for the starting procedure including data of 3814 participants. Pairwise comparisons on starting procedures compared a static start (0 m acceleration distance) with a dynamic start (range: 0.5 to 3 m) (S1 Table). Fourteen (26%) pairwise comparisons compared the impact of the length of the testing distance on resulting gait speed including data of 2695 participants. Thereby, a longer distance (range: 6.1 to 10 m) was compared with a shorter distance (range: 2.4 to 5 m) (S2 Table). Twelve (23%) pairwise comparisons compared automatic timing system (sensor-based) with a manual timing (e.g. stopwatch) including data of 3797 participants (S3 Table). Six (11%) pairwise comparisons compared surface (soft vs. hard surface) test procedures including data of 222 participants (S4 Table). Seven (13%) compared walkway procedures with usual surface including data of 182 participants (S5 Table).

None of the identified pairwise comparisons evaluated only the impact of finishing procedures (Table 2).

Pairwise comparisons were heterogenous related to the methodologic approach used, with important differences on the characteristics of the study, and elements of the measurement protocols, that are summarized in S1–S5 Tables.

Effect of testing procedures on gait speed

Table 2 summarizes the effect of the different testing procedures on resulting gait speed. In the Supplementary information, S6 to S10 Tables depict the mean values of the difference in gait speed observed in each pairwise comparison. These Tables also demonstrates that most studies reported an exact p-value for this mean difference, but 95% confidence intervals of the mean difference were often not reported.

Starting test procedures: In studies with a static start, participants are instructed to walk at their usual pace and time is measured immediately upon initiation of walking and includes the acceleration phase in the overal time measure. In contrast, for a dynamic start, timing usually begins 2 m after walking is initiated, so the acceleration phase is not included in the overall time. Among the 14 observations comparing the effect of starting test procedures, the median of mean differences in observed gait speed was 0.06 m/sec, with a higher gait speed for dynamic start as compared to static start. The MCID of 0.1 m/sec was exceeded in 4 of the 14 pairwise comparisons (S6 Table). Five comparisons also found a statistically significant higher mean gait speed for dynamic start, but below the MCID of 0.1 m/sec. In three comparisons the difference was not statistically significant, probably related to small sample sizes. Finally, as an exception, one study [27] found a statistically significant lower gait speed (-0.02 m/sec) for dynamic start as compared to static start in a sample of patients with mild Parkinson disease. The 95% confidence interval of this mean difference did not exceed the MCID (–0.008 to -0.026 m/sec). Of the 14 pairwise comparisons, only one reported an intraclass correlation coefficient (ICC 0.98) (Lindholm [27]).

For distance test procedures we observed an overall median absolute difference of 0.04 m/sec with a trend towards higher gait speed measured over the longer distance as compared to the shorter distance. Test distances varied between 2.4 and 10 m in these comparisons. There were 5/14 pairwise comparisons demonstrating that the difference in measured gait speed exceeded the MCID, all reporting higher gait speed results for the longer distance compared with the shorter distance (S7 Table). Two studies found statistically significant differences in measured gait speed below the MCID, but these studies did not report confidence intervals. The remaining 7 observations found no statistically significant differences. Three pairwise comparisons reported an intraclass correlation coefficient (median 0.80 (range 0.79 to 0.93).

For timing test procedures, the median absolute difference was 0.00 m/sec. Overall, MCID was exceeded in none (0%) of the pairwise comparisons on timing procedures (S8 Table). Three pairwise comparisons reported a statistically significant difference ([31], [31], [37]) in resulting gait speed in both directions, but 95% confidence intervals of mean differences did not exceed MCID. The remaining comparisons did not find statistically significant differences in measured gait speed for automatic versus manual timing. Five of the 12 pairwise comparisons on timing test procedures reported intraclass correlation coefficients with a median of 0.99 (range 0.91 to 1.00).

The absolute median difference in gait speed was -0.11 m/sec for surface test procedures. Most comparisons found slower gait speed for soft surfaces (e.g. carpet or grass) compared to hard surfaces. Overall, MCID is exceeded in 4/6 of surface test procedures (S9 Table). These studies hypothesized that soft or uneven surfaces may cause the test person to reduce usual walking speed. Interestingly, one study (Wilmott[22]) found a statistically signifcant correlation in the other direction, which was below the MCID. In this study, geriatric inpatients exhibited a slower gait speed (mean difference 0.08 m/sec) on a vinyl floor as compared to a carpeted floor. The authors suggested that patients overall felt unsafe while walking. Consequently, they may have been more cautious on a hard surface than on the softer carpeted surface. No intraclass correlations were reported for surface test procedures.

For walkway test procedures, the absolute median difference in gait speed was 0.04 m/sec. Overall, all pairwise comparisons were statistically significant, but in both directions. Five pairwise comparisons found higher gait speed results and two found slower results when electronic walkways were compared with usual surfaces. MCID was only exceeded in one pairwise comparison (Peters 2014 [25]) showing higher gait speed results for the electronic walkway than the usual walk test, but without reporting a 95% confidence interval. Moreover, the 95% confidence interval of the mean difference was exceeded in two pairwise comparisons: Cleland et al. [23] found lower gait speed results for the electronic walkway than usual surface of -0.08 m/sec (95% CI -0.05 to -0.1 m/sec), and Sustakoski et al. [41] found higher gait speed results on the electronic walkay compared with the usual surface of 0.07 m/sec (0.04 to 0.1 m/sec). Of note, Peters et al. (2014) [25] investigated three study subpopulations when comparing a usual walk test (3m) with a GAITrite walkway and found results in both directions. Among participants with good mobility, gait speed was significantly faster on the GAITRite walkway. In contrast, participants with impaired mobility had significantly slower gait speeds on the GAITRite walkway. Six of seven pairwise comparisons described an intraclass correlation coefficient with a median of 0.81 (range 0.49 to 0.96).

We did not find any studies that reported pairwise comparisons of finishing test procedures. However, with the exception of one pairwise comparison on starting test procedures (Lindholm[27]), all starting test procedures also had a distance for decleration upon dynamic starting protocols, but no distance for deceleration was provided on static starting protocols. Overall, most pairwise comparisons reported whether participants were told to stop at the finishing line (thereby including the deceleration phase in the timing of gait speed), or to stop after the finishing line (excluding deceleration from the gait speed measure). Distances of 1.5 m (24) and 2 m (16, 34) were reported for deceleration after the finishing line.

Starting procedures

In the majority of reports, we found faster gait speed results for dynamic start protocols versus static start protocols. It is reasonable to conclude that a person needs time to accelerate before reaching usual gait speed. In a static start protocol, the acceleration phase is measured as part of the gait speed result, which can underestimate usual gait speed. In contrast, a dynamic start protocol removes the acceleration phase from the gait speed result and provides a more accurate measure of usual gait speed. Distances and times for acceleration and deceleration are of the upmost importance when measuring and comparing skills of atheletes. A study of elite female soccer players demonstrated that the mean distance to achieve a desired speed was 1.4 m for both accelerations (±0.15) and decelerations (±0.14) [42]. Further analysis showed that the mean and maximum distance per effort varied according to rate of acceleration and deceleration. In older people, starting procedures may have a clinically significant effect on gait speed results. The impact of static versus dynamic starting procedures may be even become more pronounced in patients with underlying diseases (e.g. an advanced Parkinson disease) that are marked by progressive decline in gait and muscle function. In these patients, additional time and distance will probably be needed to attain usual gait speed.

Based on these considerations, the acceleration phase (usually estimated at 1.5 to 2 m) should not be included in the measuremet of usual gait speed.

Distance procedures

In our systematic review, overall median absolute gait speed results were faster for longer distances (ranging from 6.1 to 10 m) than for shorter distances (2.4 to 5 m). However, this difference might be minimal. In one of the included studies with low risk of bias, Bohannon et al. [19] found a statistically significant, but clinically irrelevant, difference of 0.01 m/sec between gait speed measured over 2.4 m and 6.1 m distances. The authors concluded that the shorter distance of 8 feet (= 2.4 m) was justifiable. From a clinical perspective, shorter distances are more feasible and preferrable. Thus, a distance of 2.4 to 3 m is probably sufficient to measure time of usual gait speed.

Timing procedures

The median difference in gait speed results between automatic and manual timing protocols was zero, but some studies noted relevant differences. For example, Kim et al. [31] described a significantly slower gait speed result using manual timing compared to automatic timing. The authors argued that manual timing using a stopwatch might lead to misclassification of mobility status and suggested that automatic timing be used instead. A possible reason for this effect might be, that the handling of stopwatches is based on subjectivity, as assessors manually trigger the stopwatch too early after the starting line or stop the timing too late, suggesting that establishing clear criteria for stopping the test might reduce measurement variation of manual timing. On the basis of these reports, automatic timing seems to be preferrable as a standardized method of measuring gait speed.

Surface procedures

Surface test procedures revealed the largest overall median effect on gait speed. Rough or soft surfaces resulted in significantly slower gait speed compared to hard walkway surfaces. These effects are plausible as it is a natural phenomenon that irregular surfaces result in a slowing of pace to avoid loss of balance or falling. Use of challenging surfaces are more appropriate for assessing balance and functional gait. If the goal of the test is to measure usual gait speed, a hard surface walkway would be the most appropriate and reliable choice.

Walkway procedures

Effects for walkway procedures showed mixed results. We found reports of both faster and slower gait speed results for electronic walkways compared to usual walk tests. Peters [40] explained these variable effects based on the level of mobility status of individuals tested. While older people with mobility impairment had slower gait speed results on the electronic walkway, older people without relevant mobility impairment had faster gait speed results on the electronic walkway. Thus, the authors argued that people with mobility restriction may have perceived the mat of the electronic walkway GAITRite as a potential tripping hazard and therefore decreased their walking speed as a precautionary measure. This explanation would be analogous to the effect of hard versus soft (carpet) surfaces.

Moreover, underlying distances of the electronic walkway and the usual walk test could have biased the results of pairwise comparisons of walkways. While the electronic walkway GAITRite is a mat of 4.42 m, the usual walk test was variously defined among reports. For example, Peters [40] defined the usual walk test as a distance of 3 m. Thus, participants potentially had more distance to accelerate using the electronic walkway (4.42 m) and therefore potentially achieved a faster peak speed compared to the shorter usual walk test of 3 m.

Thus, electronic walkways do not reflect the patterns of usual gait speed, and results may vary according to type of electronic walkway.

Finishing procedures

We did not find any study reporting on results of pairwise comparison of finishing procedures. We assume that in older patients, the effect of deceleration on gait speed could be similar to the effect seen with acceleration during starting procedures (i.e., as older patients may need distance to achieve their usual pace, they may also need distance for deceleration at the end of the test). Without taking this natural deceleration into account, the resulting gait speed may be underestimated. Therefore, similar to starting procedures, a distance of 1.5 to 2 m after the finishing line should be sufficient to account for deceleration.