Setting
Community-dwelling adults (≥ 21 years) were recruited from a large north-eastern residential town of Yishun in Singapore, with a residential population of 220,320 (50.6% females), with 12.2% older adults (≥ 65 years). This is similar to the overall Singapore residential population [15] of 4,026,210 (51.1% females), with 14.4% older adults (≥ 65 years).
Participants
Random sampling was employed to obtain a representative sample of approximately 300 male and 300 female participants, filling quotas of 20–40 participants in each sex- and age-group (10-year age-groups between 21–60; 5-year age-groups after 60). Older adults (above 75 years old) were also additionally recruited through community and senior activity centres. Participants were excluded if they had physical disabilities that limited their activities of daily living; diagnosed with either cognitive impairment or any neuromuscular disorders; or suffering from more than five poorly controlled co-morbidities or chronic illness. Ethics approval was obtained from the National Healthcare Group Domain Specific Review Board (DSRB − 2017/00212) and written consent was obtained from all participants.
Methods
A 6 m instrumented walkway system, GAITRite® (CIR systems, USA, 120 Hz sampling rate) was used for the gait analysis. Participants were instructed to initiate their gait 1 m before and end 1 m after the walkway system, to account for any gait related accelerations or decelerations, respectively. Participants were instructed to walk barefoot at their self-selected (habitual) gait speed. After a practice trial, three valid trials were recorded. A trial was considered valid if at least 6 alternate footfalls were captured within the sensor platform. Spatiotemporal parameters [Fig. 1] were automatically calculated by the walkway software (Version 4.8.5). Gait speed was estimated from the mean stride velocity of the participants.
Prior to the gait assessments, participants performed two common functional mobility tests- the Timed-up and Go (TUG) [16] and the Five Times Sit to Stand Test (5XSTS). Participants performed the TUG twice and the mean value was used for analysis. As for the 5XSTS, participants were provided a single practice trial, after which the actual test was performed. Additionally, they also performed the short-Physiological Profile Assessment [17]. The short-PPA has been validated as an indicator of fall risk in older adults [17–19]. It consists of five sub-tests: (a) Melbourne Edge Test (b) hand reaction test (c) proprioception (d) knee extension strength (e) postural sway. However, only the last 3 sub-tests, which are related to the lower limb, are discussed in this study.
Participants aged 65 and above and with a fall risk score of 2.0 and above were categorized as the “High Risk” (HR) group and the rest as “Low Risk” (LR) group [17]. Those below the age of 65 were classified as the “Healthy Control” (HC) group and were included in the computation of the reference population’s EGVI.
EGVI Calculation
Alternative parameters [9], pn, which describes the intra-trial variability of step time (s), step length (cm), stance time (s), single support time (s), stride velocity (cm/s) from data of all subjects (n = 531) were included in the Principal Component Analysis (PCA) to compute the correlation coefficient, cn, that describes the contribution of each variable to the overall variability of the data. For the computation of alternative parameters, a macro [an Excel version was provided as a supplementary material by Gouelle et al. (2013)[9]] was implemented in R Studio (Version 3.6.1). The results of the PCA analysis suggested that close to 50% of the variance was explained by the first principle factor alone [Fig. 1]. However, this was substantially lower than those reported by Gouelle et al. (2013)[9]. Stride velocity and stance time contributed most to the overall gait variability, with all of the variables achieving correlation of at least 0.6 with the principle component [Fig. 1]. These coefficients were used as weights in the EGVI calculation as explained below.
The EGVI was calculated using a modified macro [an Excel version was also provided as a supplementary material by Gouelle et al. (2013)[9]] that was again implemented in R Studio (Version 3.6.1). Modifications were based on Gouelle et al. (2018) [12], and they primarily pertained to addressing issues related to (a) magnitude (b) direction (c) and redundancy. The details of the calculation and modifications are presented elsewhere [9, 12]. Here, we only highlight the key steps involved in the derivation of the EGVI.
First, the mean sum of product, sHP ,was calculated based on the 5 spatiotemporal parameters [see above] of 215 healthy participants [aged 21 to 65 and gait speed ≥ 100.0 cm/s] by matrix multiplication (see Gouelle et al., 2013 [9] for more details) of the weighted coefficient, cn, and the alternative parameters, pn. The sHP for this group was 18.05, which was close to Gouelle et al. 2013 [9]. Then, the sum of product of each participant, sα, was computed [again by matrix multiplication] and the absolute distance, dα,HP, between this participant (sα) and the healthy control group (sHP ) was calculated. An addition of 1 was added to dα,HP prior converting this value to the raw EGVI (
). If the sum of product of this participant, sα ,was lower than the control group (sHP = 18.05), the
was negated. If it was in the range of the mean raw EGVI of the healthy control group
, then a value of EGVIα = 100 was assigned to this participant, otherwise the z-score was computed, thereafter multiplied by 10 and add to a 100. This would be the participant’s EGVI. An EGVI score of 100 indicated that the participant’s gait variability was close to the healthy/control population and any deviations from the 100, more indicating greater and less indicating lower gait variability compared with the reference group [9, 12].
Statistical analysis
Mann–Whitney U non-parametric test was used to evaluate differences in raw EGVI, EGVI between populations. Student’s t-test was used to evaluate differences in all other continuous variables, including participant characteristics. Stepwise linear regression models for fall risk were built to examine the independent as well as combined effects of EGVI and gait speed. Linear modelling investigated the relationship between (a)EGVI and Age; (b)EGVI and gait speed. Whenever the scatterplots suggested possible quadratic relationship, the models were tested for significant improvement in adding a quadratic term to the model. Discriminatory power of EGVI was explored using ROC analysis. Pearson correlation was used to assess the relationship between EGVI and functional mobility and balance assessments. Significance level (α) was set to 0.05 for all statistical tests. Statistical analysis was performed in R Studio (Version 3.6.1).