Feedback-based perturbation balance training during stationary cycling improves reactive and proactive balance among older adults: a single-blinded randomised controlled trial

Abstract

Background

Perturbation balance training (PBT) is an effective regime that reduces fall rates by triggering and improving balance recovery skills. Controlling trunk movements consistently reflects effective reactive stepping, as it enhances proximal stability, providing a stable base for limb movements.

Objective

To demonstrate the effect of PBT during seated hands-free stationery cycling on objective balance parameters of reactive and proactive balance control in standing.

Design

Two-arm parallel-group, single-blinded randomised controlled trial with concealed allocation, blinded assessors and data analysers, with intention-to-treat analyses.

Participants

Fifty-six community-dwelling older adults, 70+ years of age (mean ± standard deviation: 76.43 ± 4.76 years, 39.3% of men and 60.7% of women), walking independently without assistive devices.

Interventions

The two groups performed twenty sessions of seated stationary cycling, 20 minutes each, over 12 weeks, while performing concurrent cognitive tasks: (i) cycling hands free, received perturbations with real-time implicit sensorimotor feedback (PBT during hands-free stationary cycling, n = 29); (ii) standard cycling training (SCT, n = 27) cycled using hands without perturbations.

Outcome measures

The primary outcome measures were the reactive balance measures in standing, e.g. single-step threshold, multiple-step threshold and the probability of stepping. Secondary outcomes were voluntary stepping Test and 6-Minute Walk Test (6MWT). Measures were taken at baseline and immediately postinterventions.

Results

The group-by-time interactions indicate that PBT during hands-free stationary cycling improved balance reactive responses i.e. increased single- and multiple-step thresholds in mediolateral perturbations (P = .001, effect size [ES] = 0.88, and P = .001, ES = 0.64, respectively) and multiple-step threshold in anteroposterior perturbations (P = .022, ES = 0.34) and decreased the probability of stepping compared to standard cycling training. PBT during hands-free stationary cycling also resulted in faster voluntary step reaction (P = .011, ES = −0.84) and foot contact times (P = .037, ES = −0.56). Both groups significantly improved their 6MWT results.

Conclusion

Feedback-based PBT during hands-free stationary cycling has the potential to improve reactive and proactive balance measures in standing.

Registration

clinicaltrials.gov, NCT03636672, https://clinicaltrials.gov/study/NCT03636672

Key Points

• Individualised feedback-based perturbation balance training during seated cycling improved standing balance.

• Generalisation of skills: Older adults transferred balance gains from seated cycling to fall resistance in standing.

• Optimised training: Adapting perturbation balance training to neuromotor capacities enhances real-life applicability.

Introduction

Falls are the leading cause of fatal and nonfatal injuries among older adults [1]. They reduce mobility and quality of life [1, 2] and increase morbidity and mortality [3, 4]. Reactive balance responses play a critical role in preventing falls [1, 2, 5, 6], as many of them arise from ineffective recovery reactions following an unexpected loss of balance [6–8].

Perturbation balance training (PBT) involves exposing participants to repeated, unannounced external balance disturbances in a safe, controlled setting. The goal is to specifically activate and train balance recovery skills (e.g. reactive step, reactive trunk and arms responses) to improve stability and prevent falls. Older adults who participated in PBT programmes were found to have reduced their rate of falls by 40%–46% compared to those in control groups [5, 9–11], whereas conventional balance exercise programmes, reduced fall incidence only by 21%–23% [12]. Participating in PBT programmes reduces laboratory-induced frequency of reactive step responses [9, 10, 13] and improves voluntary step performance [14–16], both balance-related fall risk factors [9, 10, 13–16]. However, PBT programmes are mostly conducted using different mechatronic systems that provide external perturbations during standing or walking [9–11, 13–16]. Consequently, such PBTs are only suitable for older adults who can stand and walk independently on a treadmill without relying on or holding onto the treadmill’s handrails [9–11, 13–16]. Previous research has shown that even light touch can diminish both reactive and proactive balance responses) [17, 18]. Unfortunately, since not all older adults are able to walk without holding the treadmill handrails during standard PBT, its accessibility remains limited.

Thus, in contrast to conventional PBT methods that elicit reactive balance responses during treadmill walking, we aimed to expose older adults to perturbations while seated and cycling on stationary bicycles without holding the handrails. This approach specifically targets trunk and arm reactive responses, with the goal of improving standing balance reactions. It may offer a more accessible alternative for older adults who are unable to walk independently on treadmill-based perturbation systems without relying on handrails. Even well-respected studies on balance training in older adults appear to have either ignored or misunderstood the concept of specificity of training. For example, the study of fall prevention performed training only in sitting, assuming that effects would directly transfer to standing [19]. Due to the lack of specificity in this model, it is not very surprising that no effects of the intervention were found. However, it is important to note that the ability to limit trunk motion has consistently distinguished successful balance recovery from laboratory-induced falls [20–22]. The lower trunk flexion angle and trunk velocity have been associated with successful balance recovery after underfoot perturbations [22–27] and following a trip during overground walking [21]. Several studies reported that trunk reactions are highly involved in controlling reactive fixed base-of-support (BoS) strategies [28, 29] as well as in stepping reactions following different methods of anteroposterior (AP) and mediolateral (ML) perturbations [23–30]. Smaller trunk movement (e.g. flexion and side flexion angles) at foot contact following unexpected balance perturbations in walking and in standing is a strong predictor of successful recovery by a single-step reaction to AP and ML perturbation [20, 23, 27, 31]. Moreover, the trunk angle at foot contact is also reported to be one of the principal mechanisms by which balance recovery is adapted by repeated exposure to balance perturbations [32, 33]. Additionally, older adults who participated in PBT programmes exhibited a significant reduction in their maximum trunk angle during balance recovery [25, 34]. Also, when compared to well-established active Tai Chi controls, the maximum trunk angle was smaller and a reactive balance rating was higher among PBT participants only [34, 35].

Therefore, given that the trunk parameters consistently reflect effective reactive stepping, we developed the Perturbation Stationary Bicycle Robotic (PerStBiRo) system that provides feedback-based balance perturbations during hands-free stationary cycling in sitting, aiming to improve the capabilities of trunk reactive responses [36]. A Microsoft Kinect system™ captures the participant’s trunk movements during the perturbation exercise. PerStBiRo’s software, which was specifically developed for this project, is able to identify a participant’s reactive balance responses. In case effective balance reactive trunk movements occurred (i.e. fast side-flexion and counter rotation of the trunk), the PerStBiRo system immediately stops and returns to its horizontal platform position, providing the trainee with real-time sensorimotor implicit feedback of an effective reactive trunk response (detailed description below) [36]. We were also inspired by the well-known health advantages older people gained from stationary cycling training programmes [37–40] and that pedalling and walking are lower-extremity rhythmic tasks with similar reflex modulation [40–43], sharing the operation of related neural circuitry [43, 44]. This study aims to evaluate whether an alternative feedback-based PBT programme, which delivers perturbations during hands-free stationary cycling in a seated position, can effectively transfer to improve balance function in standing and reduce fall rates among older adults. We hypothesised that three months of feedback-based PBT during hands-free stationary cycling would generalise to improvement in reactive and proactive balance control in a standing condition, as well as, on follow-up, reduce falls.

Methods

Study design

This two-arm parallel-group randomised controlled trial followed the study protocol presented in Batcir et al. [45] We used the Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines [46]. All participants signed an informed consent form approved by the Helsinki ethics committee at Barzilai Medical Center in Ashkelon, Israel (BARZI0104), and registered in the U.S. Clinical Trials Registry (NCT03636672). After baseline testing, Participants were randomly assigned to either Perturbation Balance Training during hands-free Stationary Bicycling Riding (Group PBT during hands-free stationary cycling) or Standard Cycling Training (Group SCT) during hands-on Stationary Bicycling Riding without perturbations. Reactive and Proactive Balance in standing (see below) was assessed pre and immediately post all interventions. Fall incidence was followed prospectively for 1 year, performing a monthly phone call. Since the intervention was conducted at two locations, participants were randomly assigned in two blocks of 34 each using computer-generated randomisation (Random Allocation Software version 1.1, Isfahan, Iran). Block randomisation was computer-generated, ensuring that equal numbers were allocated to each group, and performed by the principal investigator who was not involved in participant assessments or training. Staff performing participant assessments as well as analyses were blinded to group allocation and the training process.

Participants

We recruited older volunteers (3.1.2019–23.2.2021) using advertisements, flyers and personal contacts from two sites: Beit Yona retirement home for older adults, Beer-Sheva (n = 16), and Kibbutz Hazerim, Israel (n = 40). At enrollment via face-to-face screening interviews, we verified our eligibility criteria: 70+ years of age, the ability to walk independently without assistive devices, independence in daily living activities, Mini-Mental State Examination (MMSE) < 24 [47], and weight < 120 kg (The weight limit of the safety harnesses used in training and testing devices) [45]. Then, participants completed a medical history questionnaire. Exclusion criteria included: (i) the inability to ambulate independently or without walking aids, (ii) a serious vestibular disorder (e.g. Meniere’s disease), (iii) serious visual impairment (e.g. blindness in one eye), (iv) symptomatic cardiovascular disease, (v) neurological disorders (e.g. Parkinson’s disease, stroke), (vi) orthopaedic disorders (e.g. joint replacement in the past year), (vii) peripheral neuropathy of the lower extremities, (viii) severe arthritis or (ix) cancer (metastatic or active treatment). Prior to participation, qualified individuals provided a medical waiver signed by their primary-care physician, allowing them to participate in PBT.

Intervention programmes

Training interventions took place approximately twice a week over a 12-week period. Due to the COVID-19 pandemic, which led to national lockdowns, various restrictions and the quarantine of several participants during the intervention, the number of training sessions outlined in the study protocol [45] was reduced. We excluded participants who were unable to complete 20 training sessions within 13 weeks. The duration of our programme is still consistent with applications of lower-magnitude perturbation training during standing or walking [9]. Both groups were trained by the same PerStBiRo system [36, 45] (Figure 1), and the difficulty of the training is individualised to the participant’s ability by expert physiotherapists and based on the participant’s performance.

Figure 1.

The Perturbation Stationary Bicycle Robotic (PerStBiRo) system—A feedback-based perturbation device. (a–c) represent a sample of trunk and shoulder lateral angles (c, shoulder angle—dark gray line; head and neck angle—light gray line) and the PerStBiRo system bicycle platforms horizontal angle (c, bicycle angle—black line) by time during perturbation exercises, following a programmed unannounced 15° right and left tilting perturbations (a–c—dotted grey timeline). Note: The ‘sinusoidal’ green line observed before the first perturbation (around second 48) at approximately one cycle per second represents the head and neck angle during voluntary pedalling.

All participants performed 20-minute sessions that began with a 3-minute warm-up period and ended with a 2-minute cool-down period of self-paced pedalling without perturbation. This was followed by a 15-minute exercise period in which the PBT during hands-free stationary cycling group was exposed to mediolateral tilting perturbations during hands-free stationary bicycling and the SCT group performed hands-on stationary bicycling without perturbations. Medio-lateral trunk control was chosen since impaired ability to control lateral balance during step execution even in forward or backward stepping appears to be a particular problem. In one study [48], over 30% of the initial forward or backward stepping reactions in older adults were followed by steps that were directed so as to recover lateral stability, a tendency that was rarely seen in young adults. In addition, older adults with a history of falling tended to include additional lateral foot movement in the initial step of the reaction, when responding to forward instability [49], would also appear to indicate difficulty in controlling lateral stability during anteroposterior step execution. Furthermore, a prospective study found that the tendency to follow an initial forward or backward step with a lateral step predicted an increased risk of falling laterally [50]. In addition, during the execution of a voluntary forward or backward step, an anticipatory postural adjustment (APA) is first made in the mediolateral direction to shift the body weight onto the standing foot, followed by the actual stepping movement [51].

During the exercise period, participants in both groups received the same cycling intensity and cognitive dual tasks at increasing levels of difficulty set according to their training programme (Table 1). The cycling intensity is also a physical endurance training, which individually raises the heart rate to a range of 60%–80% of the maximum heart rate, as monitored and alerted using a smartwatch (TomTom Runner Cardio [52]). The concurrent interactive visual cognitive tasks were provided to distract the trainee, presented on a TV screen and required the trainee’s verbal response, then checked by the physiotherapist for accuracy [36].

Table 1.

Details of the PBT during hands-free stationary cycling programme and the SCT training programme (the SCT group followed the pedalling intensity and cognitive tasks only)

Training level	Platform tilting (deg)	Platform peak vel. (deg/s)	Platform peak acc. (deg/s2)	Type of training (announced/blocked/random)	Type of perturbations (external, internal)	Type of external cues (none, visual, sensorimotor)	Number of perturbations per minute	Pedalling intensity (bicycle resistance)	Cognitive task type and difficulty
1	0°	Hands-free cycling practice, no perturbations				None	0	0	None
2	2°	10	10	Announced onset blocked Lt	External only	None	4	0	None
3	3°	20	20	Announced onset blocked Rt	External only	Visual	4	1	None
4	4–5°	25	25	Unexpected onset blocked Rt-Lt	External only	Visual	4	1	None
5	5–7°	30	30	Unexpected onset blocked Rt-Lt	Internal & external	Visual	4	1	None
6	5–8°	30	30	Unexpected onset blocked Rt-Lt	Internal & external	Sensorimotor	4	1	None
7	6–9°	30	30	Random	Internal & external	Sensorimotor	5	1	Find the odd one out of 1
8	5–10°	30	30	Random	Internal & external	Sensorimotor	5	1	Find the odd one out of 2
9	7–11°	30	30	Random	Internal & external	Sensorimotor	5	2	Find the odd one out of 3
10	8–12°	30	30	Random	Internal & external	Sensorimotor	5	2	Find a specific object 1
11	5–13°	30	30	Random	Internal & external	Sensorimotor	5	2	Find a specific object 2
12	5–13°	30	30	Random	Internal & external	None	6	2	Find the differences 1
13	9–14°	30	30	Random	Internal & external	Sensorimotor	5	2	Find the differences 1
14	10–15°	30	30	Random	Internal & external	Sensorimotor	5	2	Find the differences 2
15	5–16°	30	30	Random	Internal & external	Sensorimotor	5	2	Find the differences 2
16	5–16°	30	30	Random	Internal & external	None	6	3	Find the differences 3
17	11–17°	30	30	Random	Internal & external	Sensorimotor	5	3	Find the differences 3
18	12–18°	30	30	Random	Internal & external	Sensorimotor	5	3	Find the differences 4
19	5–19°	30	30	Random	Internal & external	Sensorimotor	5	3	Find the differences 4
20	5–19°	30	30	Random	Internal & external	None	6	3	Find the differences 5
21	13–20°	30	30	Random	Internal & external	Sensorimotor	5	3	Find the differences 5
22	14–20°	30	30	Random	Internal & external	Sensorimotor	5	3	Find the differences 5

Abbreviations: PBT, perturbation balance training; SCT, standard cycling training; deg, degrees; s, seconds; s², second ^*second; vel., velocity; acc., acceleration.

The perturbation system

The PerStBiRo system (Figure 1a–c) comprises a stationary bicycle that is mounted on a moving platform connected to a servo-motor that generates right and left platform rotation, i.e. balance perturbation tilts. The tilting axis passes under the bicycle seat and the participant’s centre of mass (CoM) located in the pelvis. Therefore, when the participant’s CoM rapidly tilts aside, the participant is forced to respond reactively in the same direction of the perturbation in order to quickly recover balance using hip, trunk, and arm movements and even ankle and leg responses (Figure 1a). The PerStBiRo system provides a maximum right and left perturbation tilt angle of 20° (on each side) with a maximum acceleration and deceleration of 30 m/s^2 and a maximum velocity of 30 m/s. Perturbations are provided in two forms: (i) programmed machine-induced (external) perturbations delivered as unexpected medio-lateral platform tilting. These situations fall under a reactive balance control mechanism and (ii) self-induced (internal) perturbations provided by the unfixed ‘floating’ mode of the platform, which is subjected to the forces applied during pedalling. These situations fall under a proactive balance control mechanism. The Microsoft Kinect system™ captures the participant’s movements during the perturbation exercise (Figure 1b). The PerStBiRo’s software (Figure 1c) is able to identify a participant’s proactive and reactive balance responses. By capturing the upper body joints, its software calculates two key-factor angles: (i) the shoulder line angle—the angle of the participant’s shoulder line related to the ground (Figure 1c, horizontal purple line) and (ii) the head–neck angle—the angle of the participant’s head-to-neck line related to the vertical line (Figure 1c, horizontal green line). The programme relies on both angles to determine the initial trainee’s sitting posture and to distinguish the trunk response type during three situations: (i) normal oscillations during pedalling, or (ii) anticipatory postural trunk adjustments (proactive balance movements) following self-induced perturbations, or (iii) reactive balance responses after programmed unexpected external perturbations. Following external perturbations (Figure 1c humps in horizontal black line), when a rapid change occurs in both upper-body angles (Figure 1c, horizontal green and purple lines), an effective balance reactive response is detected (as presented in Figure 1c, vertically dotted grey timeline). This causes the perturbation to stop immediately, and the PerStBiRo system returns to its horizontal platform position. Providing the trainee with these real-time sensorimotor implicit feedback cues enhances the motor learning of reactive trunk responses. It should be noted, the hands-free cycling condition, performed without any external support, including foot strap assistance (Figure 1), revealed two additional types of trunk movement also in the anteroposterior plane: (i) normal trunk counter-rotation during cycling and (ii) consistent oscillations in the anteroposterior plane. These movements were further intensified during the mediolateral perturbations. For a detailed description of the PerStBiRo system, see references 36, 45.

The perturbation training programme

The PBT during the hands-free stationary cycling programme (Table 1) contains 22 levels of exercise that gradually increase in difficulty in several exercise components with respect to motor learning, strength, endurance and especially balance control principles:

A) Hands-free cycling practice since external support (handlebars) significantly reduces postural responses [17, 18] (training level 1 and above).

B) Increasing perturbation challenge over time (displacement, velocity and accelerations in platform tilting and number of perturbations per minute),

C) Shifting from announced-onset blocked-direction training for beginners (training levels 2 + 3) to unexpected-onset blocked-direction training for advanced beginners (training levels 4–6), and random perturbation method at advanced levels (random in onset, direction and magnitude, training levels 7 and above). Better motor learning was achieved by varied practice in a random order [53].

D) Shifting from participant exposure to external machine-induced perturbations at the beginning (training levels 2–4) to combined external and internal self-induced perturbations providing an unstable ‘floating’ condition (training levels 5 and above). Advanced practice in an unstable environment contributed to improve skill acquisition [54].

E) Altering the external cue type given the participant from no cue at all for beginners (training levels 2 & 3), to a visual cue for advanced beginners (real-time self-viewing of a mirror on the TV screen, training levels 3–5), to an implicit sensorimotor cue for stopping the perturbation following an effective balance response (training levels 7 and above). This led to improving intrinsic sensorimotor feedback and provided the best possible motor learning implementation [55].

F) Improving trunk range of motion by providing the full pre-planned tilting angle (levels 12, 16 and 20) in a few training sessions, when no sensorimotor cues were given in order to maximise the participants’ upper body movements to improve their upper body range of motion.

G) Increasing the pedalling intensity by increasing the bicycle resistance over time.

H) Cognitive tasks begin at level 7, and difficulties also increase along the training process to distract the participants from balance tasks. All of these provided the best possible motor learning implementation of reactive balance responses. [17, 18, 53–55].

Primary outcome measures

Reactive balance

Participants stood wearing their own shoes with feet placed together (heels and toes touching). They were exposed to random onset and direction of right/left/anterior/posterior unannounced platform translations [56] that systematically increased and were controlled at ten perturbation levels, from very low (3 cm platform displacement in velocity of 0.06 m/s and acceleration of 0.09 m/s²) to the highest perturbation level (18 cm platform displacement in velocity of 1.2 m/s and acceleration of 3.0 m/s²), for a total of 40 perturbation trials (see Table 2) [45]. Participants were instructed to prevent themselves from falling. In case a participant fell into the safety harness or asked at any point to stop the experiment, it was stopped immediately and did not continue to the next perturbation trial.

Table 2.

Surface translation parameters for pre- and post-tests

Perturbation level	Platform displacement (cm)	Platform velocity (m/sec)	Platform acceleration (m/s2)	Perturbation random order
Level 1	3	0.06	0.09	B, R, F, L
Level 2	4	0.06	0.18	B, F, R, L
Level 3	5	0.11	0.35	L, B, F, R
Level 4	6	0.16	0.52	F, R, B, L
Level 5	7	0.22	0.7	B, F, R, L
Level 6	8	0.33	1.1	B, F, L, R
Level 7	9	0.44	1.5	B, R, F, L
Level 8	12	0.66	2.0	B, F, R, L
Level 9	15	0.88	2.5	R, B, F, L
Level 10	18	1.20	3.0	B, R, L, F

Note: Perturbations were induced through a mechatronic device that provides controlled and unexpected anteroposterior and mediolateral platform translations during standing (random in onset and direction) [45].

Abbreviations: R, right; L, left; F, forward; B, backward; cm, centimetre; m, metre; sec, seconds; s², second ^*second.

The existence of fixed-BoS, single-step or multiple-step responses was identified observationally off-line, trial by trial, by a research physiotherapy expert using a 3D Vicon system that allowed image pauses, slow motion, back and forth image viewing, and changes in the viewing angle. Each trial was evaluated for each of the three reaction types (fixed-BoS, single-step or multiple-step reactions) [45, 57, 58]. We next determined the single-step and multiple-step thresholds, defined as the lowest perturbation level to evoke a single-step response and a multiple-step response, respectively, at two consecutive magnitudes for the AP and ML directions. We also calculated the probability of stepping for the entire protocol for ML and AP perturbations (# stepping trials/# trials) and for each perturbation level separately. Using this method, we previously reported excellent interobserver reliability for the total of single-step and multiple-step trials and for stepping thresholds with older adults (ICC_2,1 = 0.974–0.987, P < .001) [57, 58].

Secondary outcome measures

Proactive balance

Proactive balance control was assessed by the Voluntary Step Execution test [51, 59]. Participants stood barefoot on a single Kistler 9287 force platform (Kistler Instrument Corp., Winterthur, Switzerland) with their feet abducted 10°, and their heels separated medio-laterally by 6 cm, and viewed an ‘X’ displayed on a screen 3 m in front of them. They were instructed to voluntarily step as quickly as possible and to remain stable after stepping, following a somatosensory cue given randomly on one of their heels. The steps were standardised to a length of 50–60 cm, targeted at a square sign on the ground. Ground reaction force and centre of pressure movement data were collected from the force platform and sampled at a frequency of 100 Hz. Five forward and five sideways stepping trials were conducted and calculated together. Outcomes included voluntary step reaction time, preparation time, swing time and foot contact time [51, 59]. The foot contact time and the reaction time were able to predict future falls [59] and injury from falls [60].

In addition, we assessed the Berg Balance Scale (BBS) [61], 6MWT [62], MMSE [47], self-reported function [Late Life Function and Disability Instrument (LLFDI)] [63], Fall Efficacy Scale (FES-I) [64], physical level of activity [Habitual Physical Activity Questionnaire (HPAQ)] [65] and satisfaction with the training programme. Falls (# in the last 12 months) were reported at baseline and then tracked after completion of training for up to one year via monthly phone calls.

Sample size estimation

Sample size requirements were calculated using PS Power and Sample Size Calculation Software Version 3.1.2 [66]. Calculations were performed separately based on a single-step threshold in standing (reactive balance) and voluntary step execution times (i.e. foot contact time, proactive balance). Both indicate impaired balance in older compared to younger adults [50, 51, 57, 59, 67] and fallers than nonfallers [50, 58, 59, 67–71]. Since we previously found that both measures are normally distributed [51, 57–59, 60, 72, 73], calculations were performed using a t-test study design. A Type I error probability of 0.05 and a Type II error probability of 0.2 were applied for the two 2-sided test calculations. Based on our preliminary results, older cyclists compared to age-, gender- and health-matched noncyclists exhibited an ML single-step threshold of 10.8 cm versus 7.9 cm, respectively [72], and a foot contact time of 921 ms versus 1025 ms in voluntary stepping, respectively [73]. Using net reduction values (2.9 cm, 8.2 mm, and 104 ms, respectively), in combination with the initial variance estimates (standard deviations of 3.45 cm, 7.8 mm, and 134 ms, respectively), we determined that 27 participants per group who complete the intervention would be required. To account for reported attrition rates of ~25% in studies involving older adults [19], initially, we decided to include ~34 participants in each group for a total of 68 (27 × 1.25 = 34) [45]; however due to a COVID pandemic, we encountered great difficulties in recruiting subjects, and were able to recruit only 56 older adults (29 for PBT during hands-free stationary cycling and 27 for SCT).

Statistical analysis

Data were analysed using PASW, version 29.0 (Somers, NY, USA). Intention-to-treat analyses were performed for all parameters, except for reporting fall rates, by carrying the last assessment obtained forward for those subjects who did not complete all aspects of the study. Prior to conducting the analysis, we assessed demographic variables as well as primary and secondary outcome measures for normality using Shapiro–Wilk tests. Baseline characteristics were compared between groups using t-tests [age, weight, height, body mass index (BMI), a Mann–Whitney U test (MMSE, FES-I, LLFDI, HPAQ, health questionnaire and self-reported past falls] and a chi-square (gender and last-year fall status).

Reactive stepping thresholds (AP and ML single- and multiple-step thresholds), voluntary step times (reaction time, preparation time, swing time and foot contact time), the BBS and 6MWT were analysed by 10 two-way repeated-measures analysis-of-variance models using the general linear model (GLM) procedure for time (baseline vs. postintervention), group (experimental vs. control) and group-by-time interactions. Since the existence of each step reaction was determined dichotomously trial by trial, to assess changes in the probability of reactive stepping, mixed-effect logistic regression models were used to account for repeated-measured hierarchical multiple observations obtained for each participant (10 perturbation levels, 4 perturbation directions, pre- and post-tests). We computed mixed-effect logistic regression models for ML and AP analysis, separately, with ‘subjects’, ‘perturbation level’ and ‘perturbation direction’ as random effects and ‘group’ (PBT during hands-free stationary cycling and SCT), ‘time’ (pre- and post-test), ‘perturbation direction’ and ‘group-by-time interaction’ as fixed-effect factors. Meeting the perturbation protocol requirements was also assessed between groups at baseline and within each group pre- to post-test. We used Poisson regressions to compare the number of fall incidents. Statistical significance for all parameters was set a priori at P < 0.05. The effect size (ES) between the two independent groups was calculated using Cohen’s d formula: ES = [(Δ Experimental training − Δ Control training)/(mean std. deviation of both groups)]. Values of 0.2–0.49, 0.5–0.79 and 0.8≤ present, respectively, small, moderate and large effects [74].

Results

Participants

After screening for eligibility, 31 applicants were excluded, and 56 were randomly assigned to PBT during hands-free stationary cycling (n = 29) or SCT (n = 27) groups and participated in the baseline assessment (Figure 2, Study flowchart diagram). Nine participants dropped out [PBT during hands-free stationary cycling (n = 5) and SCT (n = 4)] for reasons unrelated to training. Another seven participants who completed the interventions [PBT during hands-free stationary cycling (n = 3) and SCT (n = 4)] were lost during the fall tracking monitoring period.

Figure 2.

Study flowchart diagram. Note: This study was conducted during the COVID-19 pandemic, and we encountered great difficulties in recruiting and following up subjects. Therefore, we were unable to reach the original sample size of 27 in each group who completed the training programs, as reported in Batcir et al. [45]. Abbreviations: PBT = perturbation balance training.

Characteristics

There were no significant differences in baseline characteristics between the PBT during hands-free stationary cycling and SCT participants (Table 3).

Table 3.

Baseline characteristics of PBT during hands-free stationary cycling and standard cycling training (SCT) control groups

Characteristic	PBT during hands-free stationary cycling (n = 29)	Standard cycling training (n = 27)	Statistic	P -value
Age (year)	76.8 ± 4.6	76.0 ± 4.9	t = −0.632	.532
% Female (number)	58.6 (n = 17)	62.9 (n = 17)	X2 = 0.11	.740
Height (cm)	166.4 ± 10.1	162.7 ± 9.4	t = −1.43	.158
Weight (kg)	75.3 ± 11.7	72.9 ± 12.9	t = −0.71	.480
BMI (kg/m 2 )	27.1 ± 2.9	27.5 ± 4.6	t = 0.44	.660
Number of diagnosed diseases	2.0, 2.0 (0,4)	2.1, 2.0 (0,4)	z = −0.31	.755
Number drugs/day	3.6, 3.0 (0,9)	3.8, 4.0 (0,9)	z = −0.59	.555
Number of falls in the last year	0.70, 0.0 (0,3)	0.6, 0.0 (0,4)	z = −1.12	.261
% Recurrent fallers (number)   % One-time fallers (number)   % Non-fallers (number)	20.6 (6) 27.5 (8) 51.7 (15)	18.5 (5) 11.1 (3) 70.3 (19)	X2 = 2.76	.251
Visits to a family doctor (past 3 months)	0.9, 1.0 (0,3)	0.9, 1.0 (0,4)	z = 0.26	.788
Visits to a specialist doctor (past year)	2.5, 2.0 (0,6)	2.5, 2.0 (0,7)	z = 0.12	.897
Habitual Physical Activity Questionnaire (score)	5.5, 5.2 (3.5,8.5)	5.5, 5.7 (3.25,8.5)	z = −0.04	.961
MMSE	28.9, 29 (26,30)	29.0, 29.0 (27,30)	z = 0.15	.876
FES-I	25.2, 23.0 (16,57)	25.7, 23.0 (16,47)	z = −0.34	.730
LLFDI
Overall function (score)	65.7, 66.1 (53.8,73.3)	68.0, 66.7 (43.48,100)	z = −0.53	.593
Basic lower extremity function (score)	77.3, 74.3 (61.8100)	82.5, 81.1 (51.2100)	z = −1.20	.230
Advanced lower extremity function (score)	60.0, 62.1 (39.5,77.3)	61.5, 59.4 (27.8100)	z = −0.06	.948

Values are means ±1 SD where a variable was normally distributed using t-tests for group comparisons (indicated by t); means, medians (Minimum value, Maximum value) where a variable was not normally distributed using Mann–Whitney U tests (indicated by z); and percentage % (numbers) for qualitative variables where group comparisons were based on chi-square (indicated by X²). Note: the habitual physical activity questionnaire consists of three sections (occupation, sport and leisure activity indices) and a total score. We used the total score excluding the occupation activity because all participants were retirees.

Abbreviations: PBT, perturbation balance training; SCT, standard cycling training %, percentage; cm, centimetres; BMI, body mass index; kg, kilograms; kg/m², kilograms per metre squared; MMSE, Mini-Mental State Examination; FES-I, Fall Efficacy Scale; LLFDI, Late Life Function and Disability Instrument.

Primary outcome measures

Reactive balance

We found main effects of group-by-time interactions for single-step and multiple-step thresholds in ML perturbations (F = 11.960, P = .001, ES = 0.88 and F = 12.037, P = .001, ES = 0.63, respectively Table 4), indicating that these thresholds increased significantly with PBT during hands-free stationary cycling compared to SCT. PBT during hands-free stationary cycling also resulted in a significant group-by-time increase in the multiple-step threshold in AP perturbations (F = 5.598, P = .022, ES = 0.34) compared to SCT. In addition, the PBT during hands-free stationary cycling resulted in a significant increase in the highest perturbation-level achieved (see Table 5—Meeting the perturbation protocol requirements).

Table 4.

Balance measures in experimental PBT during hands-free stationary cycling and SCT control training programs

Parameter		Baseline mean ± 1 SD	Postintervention mean ± 1 SD	Time (pre- to post-test)	Group * time interaction	Effect size
A. Reactive stepping
i. Single-step threshold
ML direction (perturbation level)	Experimental	3.93 ± 0.96	4.41 ± 1.80	F = 0.415 P = .522	F = 11.960 P = .001*	0.88
	Control	4.15 ± 1.35	3.44 ± 1.36
AP direction (perturbation level)	Experimental	5.34 ± 1.38	5.69 ± 1.85	F = 0.063 P = .803	F = 3.140 P = .082	0.39
	Control	5.04 ± 1.48	4.78 ± 1.52
i. Multiple-step threshold  N
ML direction (perturbation level)	Experimental	6.34 ± 1.44	7.07 ± 1.71	F = 0.943 P = .336	F = 12.037 P = .001*	0.64
	Control	6.33 ± 2.09	5.93 ± 1.99
AP direction (perturbation level)	Experimental	8.10 ± 1.78	8.41 ± 1.50	F = 0.044 P = .836	F = 5.598 P = .022*	0.34
	Control	7.74 ± 2.49	7.37 ± 2.45
B. Voluntary stepping
Reaction time (milliseconds)	Experimental	220 ± 38	191 ± 28	F = 11.357 P = .001*	F = 6.934 P = .011*	−0.84
	Control	204 ± 32	202 ± 32
Preparation time (milliseconds)	Experimental	389 ± 79	381 ± 73	F = 0.092 P = .763	F = 0.322 P = .573	−0.12
	Control	365 ± 72	367 ± 72
Swing time (milliseconds)	Experimental	285 ± 39	277 ± 48	F = 0.351 P = .556	F = 0.908 P = .345	−0.21
	Control	284 ± 58	286 ± 46
Foot contact time (milliseconds)	Experimental	893 ± 111	850 ± 116	F = 4.202 P = .045*	F = 4.563 P = .037*	−0.56
	Control	853 ± 69	854 ± 113
C. Functional tests
BBS (score)	Experimental	53.52 ± 2.47	54.48 ± 2.26	F = 22.870 P < .001a	F = 3.124 P = .083	0.52
	Control	53.63 ± 3.81	53.07 ± 3.64
6MWT (metres)	Experimental	471 ± 83	493 ± 86	F = 26.148 P < .001*	F = 0.037 P = .848	0.02
	Control	460 ± 136	480 ± 149
D. Fall tracking (#falls per person/year)
	One-year retrospective group fall rate (95% CI)	One-year prospective group fall rate (95% CI)		Between-group prospective fall rates	Within-group pre–post comparison
Experimental (n = 21)	0.76 (0.44–1.31)b	0.38 (0.19–0.76)		P = .948	P = .099
Control (n = 19)	0.29 (0.12–0.70)	0.37 (0.18–0.77)			P = .566

A–C: Comparisons between the groups based on a generalised linear model (ANOVA repeated-measure 2 ^* 2). Effect size (ES) for each parameter using Cohen’s d formula: ES = [(Δ Experimental training – Δ Control training)/(mean SD of both groups)].

D: Comparisons between-group prospective fall rates and within-group (retrospective vs. prospective fall rates) performed on the 40 participants who completed fall tracking based on Poisson regressions.

^N—Note: All 56 participants reached their single-step threshold in ML and AP perturbation directions in pre- and post-tests. Fifty-three participants reached their own multiple-step threshold level in the ML perturbation direction in pre- and post-tests. Fifty participants also reached their own AP multiple-step threshold level in both tests. However, a few subjects who completed the entire protocol did not reach their multiple-step threshold in either the ML or AP direction (ceiling effect). Therefore, in these cases, we determined the multiple-step threshold as the 11th level, one level above the maximal perturbation level.

Abbreviations: PBT, perturbation balance training; SCT, standard cycling training; %, percentage; SD, standard deviation; ML, mediolateral; AP, antero-posterior; BBS, Berg Balance Scale; 6MWT, 6-Minute Walk Test; #, number; CI, confidence interval; ES, effect size.

^aSignificant difference set at P = .05.

^bSignificant difference in the retrospective self-reported fall rate prior to the intervention among the 40 follow-up participants.

Table 5.

Meeting the protocol requirements—global reactive behaviour and performance in experimental PBT during hands-free stationary cycling and SCT control training

A. Global reactive behaviour	Baseline number (%)	Postintervention number (%)	Between-group comparison at baseline	Within-group comparison baseline to postintervention
Number of participants completed the protocol (%)
Experimental (n = 29)	15 (51.7)	19 (65.5)	χ2 = 0.668 P = .413	χ2 = 1.119, P = .290
Control (n = 27)	11 (40.7)	15 (55.6)		χ2 = 1.178, P = .277
Number of participants that asked to stop before the experimental protocol ended (%)
Experimental (n = 29)	9 (31.0)	8 (27.6)	χ2 = 2.456 P = .117	χ2 = 0.0.080, P = .779
Control (n = 27)	14 (51.8)	10 (37.0)		χ2 = 1.176, P = .278
Number of participants that stopped the experiment due to a fall (%)
Experimental (n = 29)	5 (17.2)	2 (6.9)	χ2 = 1.208 P = .271	χ2 = 1.496, P = .232
Control (n = 27)	2 (7.4)	2 (7.4)		χ2 = 0, P = 1
B. Global reactive performance	Baseline mean, median (Min, Max)	Post-intervention mean, median (Min, Max)	Between-group comparison at baseline	Within-group comparison baseline to postintervention
Total number of perturbation trials
Experimental (n = 29)	35.83, 40 (19,40)	37.45, 40 (24,40)	z = −1.247 P = .212	z = 2.048, P = .041a
Control (n = 27)	33.30, 36 (17,40)	34.04, 40 (16,40)		z = 1.482, P = .138
Highest perturbation level achieved by the participant
Experimental (n = 29)	9.07, 10 (5,10)	9.48, 10 (6,10)	z = −1.422 P = .155	z = 2.058, P = .040*
Control (n = 27)	8.52, 9 (5,10)	8.67, 10 (4,10)		z = 1.069, P = .285

(A) Values are numbers of participants (percentages of each group) for global reactive behaviour (the reason for stopping the perturbation protocol), which were compared between groups and within each group based on chi-square tests (χ²) for column proportions. (B) Values are means, medians (minimum value, maximum value) for global reactive performance (total number of perturbation trials and the maximal perturbation level achieved), which were compared between groups at baseline by a Mann–Whitney U test and within each group from baseline to postintervention by a Wilcoxon signed-rank test.

Abbreviations: PBT, perturbation balance training; SCT, standard cycling training; %, percentage; Min, minimum; Max, minimum.

^aSignificant differences within each group from baseline to postintervention (P = .05).

In practice, a total of 3419 perturbation trials were performed as pre- and post-tests, of which 1798 resulted in reactive stepping responses: 1036 in ML and 762 in AP perturbations. Performing intention-to-treat analysis, The model for probability of stepping following ML perturbations (F = 30.889, P < .001) revealed main effects for ‘group-by-time interaction’ (F = 10.650, P = .001) and ‘perturbation level’ (F = 39.584, P < .001), but not for ‘group’ (F = 0.305, P = .581), ‘time’ (F = 3.327, P = .068), or ‘perturbation direction’ (F = 0.202, P = .653). The model for AP direction (F = 34.268, P < .001) revealed main effects for ‘group-by-time interaction’ (F = 4.404, P = .036), ‘perturbation level’ (F = 39.784, P < .001), and ‘group’ (F = 4.390, P = .036), but not for ‘time’ (F = 0.486, P = .486) or ‘perturbation direction’ (F = 2.031, P = .154). Figure 3 presents the probability of stepping by level of perturbation for both groups.

Figure 3.

The probability of stepping (# stepping trials/# trials) by level of perturbation for PBT during hands-free stationary cycling (represented in the legend by PBT) and standard stationary cycling training without perturbations (represented in the legend by SCT), following mediolateral (ML) (a, b) and anteroposterior (AP) (c, d) perturbations at baseline (a, c) and post-intervention (b, d) assessments. Comparisons between groups based on mixed-effects logistic regression models for each perturbation level in the ML and AP directions adjusted for perturbation direction. Abbreviations: PBT = perturbation balance training, SCT = standard cycling training, ML = mediolateral perturbations, AP = antero-posterior perturbations. ^*Significant group-by-time interaction (P < .05).

Secondary outcome measures

Proactive balance

Regarding voluntary stepping, the main effects of the ‘group-by-time interaction’ for the reaction time (F = 6.934, P = .011, ES = −0.84) and foot contact time (F = 4.563, P = .037, ES = −0.56) indicate that these stepping times shortened significantly with PBT during hands-free stationary cycling compared to SCT (Table 4). There were no main effects of ‘time’ or ‘group-by-time interactions’ for the preparation and swing times.

Functional measures

We found main effects of ‘time’ for the BBS and 6MWT (F = 22.87, P < .001, and F = 26.14, P < .001, respectively), but not for ‘group-by-time interactions’ (F = 3.12, P = .083, and F = 0.037, P = .848, respectively). Both interventions resulted in an increase in 6MWT distances (Table 4).

Fall monitoring

In a 1-year prospective fall monitoring on 40 participants who completed both interventions and follow-ups (21 from PBT during hands-free stationary cycling group and 19 SCT participants), there were eight fall incidents among the PBT during hands-free stationary cycling group (7 fallers) versus seven among the SCT group (6 fallers). We found no differences between the PBT during hands-free stationary cycling and SCT trainings in the 1-year prospective group fall rates [38% (19%–76%) vs. 37% (18%–77%), respectively, Table 4]. These follow-up participants showed significant differences in the retrospective self-reported fall rate prior to the intervention (P = .38, Table 4), but no significant differences were observed in the rest of baseline characteristics. However, the PBT during hands-free stationary cycling group insignificantly reduced their one-year prospective fall rates by 38% compared to their own retrospective fall data (P = .099).

Adverse events

Muscle soreness with localised pain was experienced by seven participants of the PBT during hands-free stationary cycling, and nine SCT participants, especially in the early stages of training. One participant of PBT during hands-free stationary cycling reported an increase in chronic back pain. These effects were managed by adjusting the training intensity, and the symptoms disappeared. One SCT individual lost balance while dismounting from the PerStBiRo system but was stopped safely by the harness and was able to stand up unassisted without injury.

Satisfaction questionnaire

Twelve participants of PBT during hands-free stationary cycling who completed the intervention answered a 5-point-Likert-scale satisfaction questionnaire (5 = strongly agree) addressing the intervention safety (4.8 ± 0.4), difficulty (4.2 ± 0.7), enjoyment (4.5 ± 0.3), their interest in continuing (4.2 ± 1.1), recommendations to other to participate (4.9 ± 0.5) and positive effects on their balance (3.8 ± 1.0).

Discussion

The main findings of this study are that PBT during hands-free stationary cycling in a sitting position generalised to effectively improve reactive balance, particularly in lateral direction (i.e. ML single-step and multiple-step thresholds, AP multiple-step threshold) and proactive stepping abilities (i.e. step reaction and foot contact times in voluntary stepping) in standing. These results are important since previous work showed that low single-step and multiple-step thresholds and more laboratory-induced reactive stepping indicate poorer reactive balance control for older compared to young adults [50, 52, 67, 70, 71, 75, 76] and for fallers compared to non-fallers [58, 69–71, 75–79].

More specifically, well-established research showed that the lateral balance function declines with age [9–11, 14, 25, 50, 57, 67, 69–71, 75, 80–84] and is highly related to falling among older adults [9–11, 14, 22, 50, 58, 67, 69–72, 77, 80–84]. Lateral balance control is important even following AP perturbations where older adults tend to fall sideways [67, 69, 70, 75, 76, 82] is also a prospective predictor of falls [69, 79, 80, 83, 84] and is related to injurious falls in older adults [80, 81]. We found significant group-by-time interactions and large ESs showing increases in ML single-step thresholds (ES = 0.88) and ML multiple-step thresholds (ES = 0.63) among PBT during hands-free stationary cycling compared with SCT participants, suggesting that training trunk reactive abilities while PBT during hands-free stationary cycling helps to control the CoM even in standing. Moreover, this aligns with their decrease in the probability of stepping at low-to-moderate perturbation levels. These findings are consistent with previous work showing that participating in PBT improved lateral balance function [5, 9, 10, 13, 14, 85]. As we hypothesised, the participants of PBT during hands-free stationary cycling recovered balance more effectively by fixed-BoS strategy components (i.e. those requiring reactive trunk and arm movements) in the ML plane, given that our balance training specifically targeted these particular movements. Our findings suggest that the balance skills participants developed throughout PBT during hands-free stationary cycling led to improved balance recovery in standing and may stem from enhanced control of CoM motion over the BoS, particularly in the ML direction following perturbations. Barrett et al. [32] suggested that the adaptation from a multiple- to single-step reactive response in older adults was related to improving control of the whole-body CoM, rather than adjustments in the BoS, i.e. by step length or step timing. Pai and Patton [86] and Pai et al. [87] found that the likelihood of taking a step following a perturbation depended on the stability limits determined according to the interaction between the position and velocity of the whole body’s CoM related to the BoS. According to this model, a reactive stepping response is necessary once a sufficiently high velocity of CoM displacement exists, even if the CoM is located within the limits of the BoS [86, 87]. Based on this model, and on the importance of controlling the whole-body CoM motion after loss of balance [32], the occurrence of a no-step, single-step or multiple-step reaction is determined by one’s ability to control the velocity of the CoM after a perturbation. Given the similar BoS, and under the same perturbation conditions, our results of shifting stepping thresholds to higher perturbation levels and the reduction in stepping responses in low-to-moderate perturbation levels indicate that PBT during hands-free stationery cycling improved participants’ ability to decelerate and control the CoM motion over the BoS following unexpected ML and AP balance perturbations through a better trunk control. Inability to control trunk motion after a loss of balance is consistently discriminated against older adults who fall from both younger adults and older adults who have been able to avoid falling [20]. They also showed that the ability to decrease in maximum trunk angle can be rapidly learned, by older adults after PBT and has been shown to effectively decrease fall risk due to unannounced perturbations [20]. Collectively, we believe that our findings strongly suggest that PBT during hands-free stationery cycling task-specific trunk training can improve trunk control and CoM control. Participants in PBT during hands-free stationery cycling could restore balance by fixed-BoS strategies rather than by taking a reactive step in standing (i.e. a higher single-step threshold) and could recover by taking a single step, rather a sequence of steps (i.e. a higher ML and AP multiple-step threshold), suggesting that PBT during hands-free stationery cycling may transfer to or be used during the standing balance tasks. This is supported by a previous study that found a clear association between the ability to control CoM path displacement and the capacity to recover balance with a single step in older adults without a history of falls, compared to those who reported recurrent falls (i.e. more than one fall in the past year) [58]. It was found that older adults with a history of falls had difficulty in ‘catching’ the CoM while moving over the base of support immediately after balance loss led to the execution of more extra steps to try to avoid an imminent fall. Furthermore, the study reported that during multi-step balance recovery, CoM displacement was nearly twice as large among recurrent fallers compared to nonfallers (23.53 cm vs. 14.80 cm) [58]. Thus, improvements in step thresholds clearly indicate that participating in PBT during hands-free stationary cycling enhances reactive balance function.

The observed improvement in AP multiple-step thresholds following PBT during hands-free stationery cycling compared with SCT participants is not surprising, given the critical role of ML stability even during forward or backward balance recovery [48, 50]. Although the initial perturbations occur in the AP direction, older adults often exhibit a tendency to lose balance sideways immediately after completing a first recovery forward step. This is largely due to the foot being placed in a near-tandem position, which inherently reduces ML stability and increases the likelihood of requiring an additional lateral step to prevent a fall. In fact, McIlroy and Maki [48] reported that ~30% of forward stepping responses in older adults involve a secondary lateral step. Therefore, improvements in ML stability—particularly the ability to stabilise the body after the initial step—can reduce the need for additional recovery steps. This likely explains the observed improvement in AP stepping thresholds: better ML control allows for successful balance recovery with a single, well-placed step, thereby reducing the overall number of steps required.

Our findings add to previous work exhibiting the positive effects of PBT on voluntary step times. The PBT during hands-free stationary cycling resulted in faster reaction and foot contact times compared to the SCT training. Slower voluntary step times reflect an increased risk of falls in older adults [51, 59] and distinguish between fallers and nonfallers [59]. Furthermore, these step times have been shown to predict future falls [59, 88] and injurious ones [60]. Interestingly, voluntary stepping, which was not trained in our programme, may further suggest that generalisation of balance abilities occurred throughout the PBT during hands-free stationary cycling. The PBT during hands-free stationary cycling reaction times decreased by 14%, from 220 to 191 ms, with a large effect size (−0.84). Voluntary step reaction time was suggested to be a traditional and basic measure of central neural processing speed [51, 59, 88, 89] and is sensitive to older age and cognitive load [51]. Specifically, Woods et al. [90] found that an increase in reaction-time latency was due to slowed motor output. Rogers et al. [16] suggested that faster voluntary reaction time after PBT may indicate enhanced neural processing speed [88, 89] and motor output of initiation and velocity of movement execution [90, 91], which are major components of the motor skills required for successful balance recovery [88–91]. These findings suggest that the central nervous system makes adaptive improvements in trunk stability as a result of trial-and-error perturbation practice in a sitting position. Repeated exposure to unpredictable perturbations in sitting may have enhanced proximal stability (i.e. trunk) and can facilitate quicker and more effective distal responses including volitional stepping, even if stepping wasn’t explicitly practised. The trunk provides a stable base for limb movement, enhancing neuromuscular coordination and balance. Improved trunk control has been shown to enhance stepping and gait performance in older adults [92], in poststroke patients [93] and even in Parkinson disease patients [94]. A systematic review highlighted that trunk muscle strength is closely linked to balance, functional performance and fall prevention in older adults [95]. These findings suggest that interventions focusing on trunk control may be beneficial for improving step performance. Not surprisingly, the 6MWT performance improved in both groups, suggesting that both cycling training programmes that also provide resistance to lower limb musculature were equally effective in enhancing lower limb muscle strength and endurance among older adults. This was supported by previous studies that found older adults who participated in bicycling training programmes experienced increases in lower limb muscle strength and endurance [38–40].

With respect to the clinical feasibility of applying the training approach used in the present study, the adverse events reported from the 980 training sessions were mainly mild in severity and participant dropout reasons were unrelated to the intervention protocols. In addition, the study’s 16% attrition rate during exercise is much lower than the 25% that has been reported in interventions involving older adults [19]. Therefore, this perturbation approach appeared to be generally well tolerated and easily accessible by older adults. Also, the satisfaction questionnaire reflected that the training challenged their balance and was enjoyable and safe.

This study has several limitations: First, the data came from a fairly small sample of independent older adults; thus, the results cannot be generalised to frail or institutionalised older persons. Second, the study ended before reaching the pre-calculated sample size (n = 27) because of difficulties in recruiting volunteers during the COVID-19 crisis, as older adults were particularly at risk during this time period. Third, the 1-year follow-up showed no significant fall reduction in either training programme, but the sample size was too small to draw definitive conclusions. Using G^*Power 3.1.9.4, the required sample size to detect a significant difference in fall rates over 1 year was estimated at 163 participants per group, based on a 76% baseline fall rate, a 38% expected reduction, 80% power and a 0.05 alpha level. Future studies should incorporate a higher number of participants, particularly frail and prefrail older adults, in which the presented training programme was designed.

Conclusions

The findings touch on an important key point in the fields of fall prevention, rehabilitation and motor learning, indicating, for the first time, that feedback-based PBT during hands-free stationary cycling has the potential to improve reactive and proactive balance measures in standing. A key innovation of this PBT programme is the integration of both motorised external and self-induced (internal) perturbations with real-time sensorimotor feedback to enhance reactive upper-body movements, within a cycling-based protocol, delivered in a controlled and safe seated environment using a safety harness. This makes the intervention particularly suitable for frail older adults, individuals with limb amputation and stroke survivors who may not tolerate upright PBT during walking. Moreover, several studies have shown that even short-duration PBT can produce meaningful improvements in reactive balance and reduce fall risk [5, 96], supporting its cost-effectiveness. Future studies should examine whether feedback-based PBT during hands-free stationary cycling can also contribute to reduced healthcare costs.

Declaration of Conflicts of Interest

Authors S.B., Y.L., R.L.L., A.S. and I.M. own a patent on some of the technology (the PerStBiRo system) used in the perturbation training.

Declaration of Sources of Funding

This study was partially supported by the ‘Helmsley Charitable Trust through the Agricultural, Biological, and Cognitive Robotics Initiative of Ben-Gurion University of the Negev’ for the development, design and building of the technology used in the interventions (the PerStBiRo system) and partially supported by a grant from the Recanati School Foundation of the Faculty of Health Sciences, Ben-Gurion University of the Negev for employing physiotherapists to conduct the interventions.

Data Availability

Data will be available on request from the author of the correspondence.

Declaration of generative AI and AI-assisted technologies

We have nothing to declare.