One-hundred-and-six participants older than 65 years of age volunteered for this study (42 female, 64 male) with a mean ± SD age, weight and height of 75 ± 6 years, 78 ± 15 kg, 1.69 ± 0.09 m, respectively. Participants were a subset from a larger cohort (n = 252), and were included in the current study based on the Physiological Profile Assessment score (PPA, short form version) of Lord et al. (2003). To ensure a wide range of falls risk, participants were included in the current study if their PPA values were below 0.5 (n = 59) or above 1 (n = 47). All participants were recruited from the Brisbane metropolitan area via the Australian electoral role. A letter of invitation was sent with an information sheet, which outlined the potential risks and benefits of the research. Participants were excluded if they had a recent or recurrent history of surgery or musculoskeletal injury, any neurological impairment such as Parkinson's disease, were unable to ambulate independently without the use of a walking aid, or were cognitively impaired (i.e., Mini mental state exam score <24). The young group included 23 participants between 18 and 25 years of age (14 female, 9 male, 21 ± 2 years, 65 ± 10 kg, 1.72 ± 0.05 m) recruited from the student population of local universities and by word of mouth. All participants provided written informed consent. The experimental protocol was approved by the Institutional Human Research Ethics Committees and conformed to the Declaration of Helsinki.

Elderly participants were followed for 12 months after the balance assessment. They maintained a weekly falls diary, which they returned at the end of each month via reply paid post (Hannan et al., 2010). A fall was defined as: “an unintentionally coming to the ground or some other lower level, not as a result of a major intrinsic event (e.g., stroke) or overwhelming hazard. This included any slips, trips or accidents, which result in a fall onto a lower level, be it a chair, bed or the floor for example.” A “faller” was defined as a person who had one or more falls recorded within the 12-month follow-up period.

Participants stood barefoot on a force plate (Type 9286AA, Kistler Group, Winterthur, Switzerland), were blindfolded to exclude the contribution of vision to balance, and wore headphones playing white noise to limit distraction. Participants stood relaxed with arms hanging by their sides and data collection commenced after ~20 s to ensure balance had reached a steady state.

Participants stood for 135 s after commencement of data recording. Custom-made vibrators (Type YM2707, Electus Distribution, Sydney, Australia, ~1 mm, 60 Hz) were bilaterally attached halfway between the distal portion of the gastrocnemius muscle heads and the distal insertion of the Achilles tendon. Firm application was assured with pressure applied to the vibrators using a neoprene band wrapped around the ankle and vibrator. After 15 s, mechanical vibration was applied bilaterally for 15 s. Vibrators were switched on for a further 15 s period at 75 s from commencement of recording. This allowed 45 s after cessation of each vibration exposure to assess post-vibration effects on balance. The experimenter stood close to the participant to provide support in case of falling. If balance was assisted, data collection was stopped and restarted if the participant agreed.

Force plate data were amplified [Type 5233A, (range: Fx & Fy; 250N, Fz; 2500N), Kistler Group, Winterthur, Switzerland] and digitized with 16-bit precision at a sampling rate of 2,000 samples/s using a Power 1401 data acquisition system with Spike2 software (Cambridge Electronic Design Limited, Cambridge, UK).

Data were analyzed offline with Matlab (Mathworks inc., Natick, MA, USA). As calf vibration mainly perturbs balance in the anterior-posterior direction (Eklund, 1972), analyses were focused on CoP motion in this plane. CoP data were filtered using a second order low pass bi-directional Butterworth filter. Cut-off frequency was set at 20 Hz and bi-directional filtering increased the filter order to 4. After low-pass filtering, data were decimated to 100 samples/s. All measures were applied in a windowed (15-s epoch) manner (Riley et al., 1999; Webber and Marwan, 2014), to assess changes in CoP motion during and after calf vibration (see Figure 1 for details).

Matlab was used to perform the statistical analysis. The threshold for significance was set at P < 0.05.

Four elderly participants withdrew from the study (1 < 0.5 PPA, and 3 > 1.16 PPA), and no prospective falls data were available for these participants. Forty-two out of 102 elderly participants reported 1 or more falls. Of these participants, 2 reported 6 falls, 1 reported 5 falls, 3 reported 4 falls, 6 reported 3 falls, 9 reported 2 falls, and 21 participants reported 1 fall. Using the prospective falls data, the elderly were grouped into fallers (1 or more prospective falls) and non-fallers. Due to technical issues with data collection, data from 3 participants (1 faller and 2 non-fallers) were excluded from further analysis. Table 2 shows means (SD) of demographics and PPA values of all elderly participants included in the final analysis (n = 99). Age, height, weight, sex, and PPA values did not differ significantly between fallers and non-fallers (all, P > 0.09, Table 2).

Results of the one-way ANOVA are presented in Table 3. At baseline (before vibration), compared to fallers, young had lower SP (moved slower), lower %DET (were more predictable), had longer L_mean (less sensitive to small perturbations, i.e., better balance performance) and had higher %LAM with longer TT (were more intermittent, with longer static episodes). Regarding DFA analysis, compared to fallers, young exhibited a larger DFA_tau and DFA₂ (a less conservative balance control strategy, lower anti-persistence: young>fallers) and had a larger DFA₁ (smoother and more persistent: young>fallers). Compared to non-fallers, young also had lower SP (moved slower), longer L_mean, longer TT, and were less anti-persistent (DFA₂: young<non-fallers), but %DET, %LAM, DFA₁, and DFA_tau were not significantly different between young and non-fallers.

No differences between fallers and non-fallers were observed at baseline.

Outcome data for CoP motion during and after vibration are shown in Figures 3–6, and summarized in Table 4. CoP mean displacement relative to baseline in response to calf vibration was not significantly different between the groups (Figure 3). In addition, CoP displacement after vibration was also not significantly different between the groups (Figure 3, Table 4).

For the young group, compared to fallers and non-fallers, across all epochs, %DET and L_mean (balance performance; young>fallers/non-fallers, Figure 4), %LAM and TT (intermittent control; young>fallers/non-fallers, Figure 5), and DFA₁ (smoothness/persistence; young>fallers/non-fallers at most epochs, Figure 6) values were higher, and SP (young<fallers/non-fallers, Figure 3) was lower (Table 4). DFA₂ was greater in the young than both fallers and non-fallers after removal of vibration at most epochs (Figure 6, Table 4). DFA_tau was higher in the young during vibration than both fallers and non-fallers, but, was similar for all groups directly after removal of vibration. DFA_tau values were higher in the young than both fallers and non-fallers after epoch 12 (Figure 6). This suggests that during vibration young were more persistent across more time-scales compared to both fallers and non-fallers (DFA_tau: young>fallers/non-fallers).

During ankle vibration, fallers and non-fallers were not significantly different in any of the linear or non-linear outcome variables (Figures 3–6). However, after removal of ankle vibration, CoP motion of the fallers had lower %DET (less predictable/more random; fallers<non-fallers, Figure 4), lower %LAM (less intermittent; fallers<non-fallers, Figure 5), and lower DFA₁ (less smooth/less persistent; Figure 6) values than those in non-fallers. See Table 4 for overview of the findings.

Our conclusions are based on interpretation of the non-linear measures of CoP motion, yet alternative interpretations of some RQA measures requires consideration (Negahban et al., 2010; Ramdani et al., 2013; Bernard et al., 2015). In contrast to our interpretation that a more deterministic structure infers more regular and stable balance control, this regularity has also been interpreted to reflect less behavioral flexibility and less complexity in CoP motion. That alternative interpretation was motivated by observations of increased regularity of CoP motion in Parkinson's disease (Schmit et al., 2006), retrospectively identified fallers (Ramdani et al., 2013), and in individuals with anterior cruciate ligament deficiency (Negahban et al., 2010). Although these interpretations might be reasonable in some cases (e.g., Parkinson's Disease), it might not apply to our observations.

Our interpretations and of others (Riley et al., 1999; Riley and Clark, 2003; Haddad et al., 2008; Cluff et al., 2009; Mazaheri et al., 2010) are based on the construct validity of RQA, methodological issues, and the nature and difficulty of our balance task. The deterministic structure is related to diagonal line feature in the recurrence plot. Both %DET and L_mean reflect CoP motions that run parallel in phase space because different CoP sections are spatially and temporally similar in shape (Figures 1C,D). This similarity reflects predictability, regularity and local stability of CoP motion. A lower %DET and shorter L_mean does not necessarily reflect more complex structure because random signals and noise also exhibit low %DET, yet those signals do not have a complex structure. Thus, although “predictability” can be determined, care is required to interpret “flexibility” and “complexity” from RQA outcome variables.

Interpretation of RQA can be aided by concurrent assessment of other non-linear features using additional measures such as DFA, as we have done. Our observation that young people have higher long-range correlations than elderly concurs with other observations of similar differences between young and old (Norris et al., 2005; Amoud et al., 2007; Duarte and Sternad, 2008; Kim et al., 2008). Observed CoP fluctuations reflect both the perturbations [internal (self-generated) and external] and the postural responses to perturbations. The combination of these perturbations is resolved by the postural system. When the postural system appropriately responds to perturbations, then successive CoP positions relate to previous positions and correlation is strong (Bruijn et al., 2013). Successful balance corrections to perturbations would not likely result in additional changes of CoP direction, but instead induce smooth CoP motion that tends to persist in the same direction. Alternatively, more changes in direction would increase the relative contribution of CoP fluctuations at shorter time-scales (DFA₁) and reduce the strength of long-range correlations. CoP motion with fewer directional changes/smoother appearance would more likely be more deterministic than CoP with more directional changes. Thus, DFA outcomes aid interpretation of whether the deterministic structure of signals arise from regular signals containing fluctuations at a single or limited number of time scales (such as a sine wave), or from regular signals with long-range correlations across multiple time scales. Taken together, concurrent observations of RQA and DFA strengthen our interpretation that high %DET and long L_mean observed in CoP motion of the young compared to the elderly is not related to reduced complexity as underlying CoP fluctuations were evident at a range of time scales.

Some RQA settings require consideration as they could impact findings. Without an appropriate corridor (Theiler window, shown in Figure 1C) along the line of identity (i.e., self-recurrences of CoP states), diagonal line measures could be overestimated by the inclusion of recurrences that are temporally close (Marwan, 2011). This is important as the threshold, below which a recurrence is defined, depends on the phase space diameter (i.e., the amplitude of CoP movements) or depends on the fixed amount of recurrences in the recurrence plot. In both cases, higher amplitude CoP motion, resulting in a larger phase space diameter, would require a higher recurrence threshold. This would increase the neighborhood to find recurrences. Greater amplitude of CoP motion is usually observed in elderly (Hageman et al., 1995; Gill et al., 2001; Abrahamová and Hlavačka, 2008). In these cases, RQA would be biased to observe longer diagonal lines and higher determinism if an appropriate Theiler window was not used (Marwan, 2011) allowing inclusion of temporally close neighbors. We used a 1-s Theiler window to minimize this bias. Use of a Theiler window was not reported in previous studies (Seigle et al., 2009; Ramdani et al., 2013; Bernard et al., 2015), the max diagonal lines reported in some studies (Riley and Clark, 2003; Schmit et al., 2006) generally approach the length of the delay embedded signals, which implies that a Theiler window was not applied and findings may have been biased.

The difficulty of the task used to assess balance control requires consideration. In the current study, participants maintained balance while somatosensory information from the calf muscle was suddenly altered by addition and removal of vibration in the absence of vision. Although aware of the experimental conditions, participants were in an unfamiliar balance situation and none had previously experienced muscle vibration. When balance is challenged, more predictable balance control could be beneficial. Increased regularity and smoothness of CoP motion could be interpreted as an appropriate adaptation to a more challenging balance task (Riley et al., 1999).

The postural vertical is modified when sensory input is increased by muscle vibration (Eklund, 1972). The impact of vibration depends on both peripheral functioning of the muscle spindles and the weighting of this sensory input. Our results showed, on average, that linear measures of the CoP displacement caused by addition and removal of calf vibration did not differ between the young group and both older groups. This suggests the peripheral functioning of the calf proprioceptors and the weight assigned to the sensory information were not different between the groups. Although, Brumagne et al. (2004) also found similar CoP displacement for older and healthy young individuals, some authors (Pyykkö et al., 1990; Quoniam et al., 1995) reported less CoP response to calf vibration with increased age. Pyykkö et al. (1990) and Quoniam et al. (1995) used older individuals from a non-community dwelling setting which might explain the difference in outcome. Further, Quoniam et al. (1995) used 3 s vibration (15 s used here), which might be too short (Capicíková et al., 2006) to affect their elderly group's CoP.

Although addition of vibration did not differently perturb balance in fallers and non-fallers, removal of the vibratory stimulus revealed differences. This implies that dynamical integration of sensory information is more challenging when sensory input is reduced than when it is augmented. This appears to concur with observations that, in some contexts, addition of a subthreshold stochastic stimulus (e.g., vibration shoe soles) can improve balance (Niemi et al., 2002; Priplata et al., 2003). Although we did not observe improvement of balance in response to sudden unfamiliar addition of a supra-threshold vibratory stimulus to the muscle, our data did show that both fallers and non-fallers responded equally well to this perturbation (across our suite of measures). Spindle activity related to changes in calf muscle length is likely to be masked during vibration (Roll et al., 1989) indicative of reweighting of the calf proprioception by the CNS away from the additional inaccurate component provided by the vibration. Although inaccurate information is provided, the additional input was integrated by the CNS for balance control evident by the backwards CoP shift in all groups. Cessation of vibration ceases the vibration related discharges immediately (Roll et al., 1989). Because the proprioceptive information was down-weighted during vibration, sudden reduction of proprioceptive information may be more difficult to accommodate to than adding proprioceptive information using vibration. The period after removal of vibration would require fast reweighting to use available somatosensory information (van der Kooij and Peterka, 2011) and the proprioceptive information from the calf might contain greater noise (Rogers et al., 1985) and less useful somatosensory information than during vibration. Further, analysis using the sliding window after removal of vibration may have reduced the variance leading to a greater probability to observe significant differences between groups than the analysis of a single epoch during vibration.

SP is known to increase, together with L_mean and %DET, in balance tasks with greater sensory challenge (e.g., eye closure, compliant surfaces; Riley et al., 1999; Riley and Clark, 2003). Increased sway would generate more proprioceptive information and potentially compensate for reduction/removal of sensory information from other sources (Carpenter et al., 2001, 2010). This “self-generated” proprioceptive information is dynamically integrated to guide balance control leading to more deterministic CoP motion and has been referred to as “perceptually guided control” (Riley et al., 1999). Although our elderly group had longer SP than the young group, this was not associated with more deterministic CoP motion. A similar observation was made by Seigle et al. (2009); removal of vision in quiet stance led to greater SP but less %DET in elderly than younger individuals. This highlights a potential compromise in perceptually guided control in the elderly; particularly for fallers after removal of vibration.

Lower DFA₂ and DFA_tau in all groups after vibration removal suggest that fluctuations at longer timescales were relatively small (~>0.8 s from DFA_tau). Proprioceptive information that establishes sense of upright functions at lower frequencies (Diener et al., 1986; Diener and Dichgans, 1988), and reduced fluctuation at longer time scales could reflect that the vertical upright of the participants was affected after removal of calf vibration. The observation of greater DFA₂ and DFA_tau for young than older participants after ~15 epochs (window starting 15 s after cessation of vibration) suggests that young more successfully established an upright subjective vertical. As we assessed fluctuation at a maximum time scale of 4.42 s our data might partly reflect recalibration of upright sense by dynamic reweighting of sensory information that establishes upright sense (i.e., proprioceptive information from the calf and vestibular input).

Poorer balance control after removal of vibration in fallers might be related to an inferior capacity to dynamically reweight the sources of somatosensory information to minimize the perturbation effects of vibration removal. The ability to flexibly explore somatosensory information is affected by increased age (Teasdale et al., 1991; Hay et al., 1996; Doumas and Krampe, 2010; Eikema et al., 2013, 2014). Reduced functioning of somatosensory systems other than the calf muscle spindles might also affect balance steadiness. Physiological aging affects all somatosensory systems (Sturnieks et al., 2008), particularly the vestibular system (Sloane et al., 1989; Strupp et al., 1999) as evidenced by greater dependence of elderly individuals on visual information to establish vertical upright (Sundermier et al., 1996; Simoneau et al., 1999), and inferior capacity to align themselves with the vertical after being tilted (Menant et al., 2012). Because sensory information from the calf was unreliable and vision was unavailable in the present study, reliance on vestibular information would have been increased. Compromised function of the vestibular system would also render the somatosensory information less useful, as the internal upright reference frame against which somatosensory information is compared (Horak et al., 1989) is less accurate. Accurate sense of vertical in combination with optimal balance control would be expected to produce CoP motion that exhibits periods of minimal CoP movements and high %LAM as observed in the young group. It follows that lower %LAM observed in the older group, and lower %LAM observed post vibration in fallers than non-fallers might reflect a compromise of this combination. Confirmation that %LAM relates to inaccurate upright sense, poorer balance control, or both cannot be derived from the present data and requires further investigation.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.