In the simplest terms, maintaining stable upright stance involves keeping the centre of mass (COM) of the body over the base of support (BOS) defined by the feet (and, in some situations, the arms). Hence, the essential defining feature of any balance perturbation is that it induces relative motion between the COM and the BOS. For stance perturbations, this relative motion can be induced by causing displacement of either the COM (e.g. via cable waist-pull systems) or the BOS (e.g. via motion platform systems). During gait, this relative motion can also be induced when the cyclic progression of the COM and BOS is disrupted due to perturbation of the BOS (e.g. due to a slip on a low-friction surface, a trip on an obstacle, or sudden acceleration or deceleration of a treadmill).

The various available perturbation methods may differ in a number of respects, such as the pattern of motion induced at specific joints, the evoked sensory inputs and the pattern of the early evoked muscle activation 40. Nonetheless, these various methods all fulfil the fundamental biomechanical requirement (disruption of the COM-BOS relationship) and can often elicit postural reactions that are similar in many respects 40. Hence, it is possible that the training benefits derived using one type of perturbation may generalise (at least to some extent) to the reactions evoked by other types of perturbations. This, however, remains to be established. If there is limited generalisability of training benefits, then it would be best to choose a perturbation method that closely emulates the perturbations that lead to loss of balance and falls in daily life; however, it is unlikely that this can be accomplished using any one single method of perturbation. The training perturbations would need to simulate both slips and trips, (report to occur in approximately 40–60% of falls in older adults 414243) as well as the sizeable proportion of falls that do not involve ambulation, (e.g. self-induced perturbation during leaning, turning, reaching or sit-stand transfer movements; support-surface motion that occurs when standing in a moving vehicle 4143).

For this study, we elected to develop a custom-designed pneumatic motion platform for the purpose of delivering the postural perturbations needed to evoke stepping and grasping reactions during training (Figure 2). One important feature of the moving-platform approach is the ease with which the direction of perturbation can be varied in an unpredictable manner 18. Unpredictability is a critical requirement, as the CNS can learn to recognise any features of the perturbation that are predictable (direction, magnitude, waveform or timing), and can use this information to improve the efficacy of the balance reactions in a predictive manner 4445464748. This type of predictive control is unlikely to be helpful in responding to the unpredictable perturbations that commonly lead to falls in daily life; hence, the adaptations that result from training with predictable perturbations may have limited generalisability 18. A second practical advantage of the motion-platform approach is that large numbers of perturbations can be delivered in a short span of time, hence maximising opportunity for training while minimising the risk of fatigue, which would be more likely to occur during gait-perturbation methods 3639.

During the training program, the surface of the platform is controlled to move suddenly and unpredictably in one of four directions: forward, backward, left or right. The first session will be a familiarisation session, in which the minimum perturbation magnitude that requires a change-in-support response will be determined. During this initial session, subjects will be instructed to respond to the platform motion naturally. For the initial stepping training session, handrails will be mounted on the platform to help subjects feel at ease, but will be removed for subsequent sessions. Subjects are instructed to respond to the platform movement by stepping, if required (Figure 2A). For grasping training, a handrail is mounted on one or both sides of the subject and foam blocks are placed around the feet to prevent stepping (Figure 2B). Subjects will be instructed to recover balance by grasping the rail or rails "as quickly as possible" in response to the platform movement. Subjects will wear a safety harness secured to the ceiling at all times during the training.

The training program adheres to the principles of exercise prescription, that is, specificity, progressive overload, individualisation and random variation. In addition, we applied principles that dictate how feedback and instructions should be provided in order to optimise motor learning.

Community-dwelling older adults, aged 64–80 years, will be recruited via advertisements in local newspapers and seniors' residences and by word of mouth. All subjects will be mobile (not dependent on mobility aids) and will be able to stand (≥ 1 minute) and walk (≥ 10 m) without any assistance. To simplify training and assessment, the study will be restricted to persons who are right upper- and lower- limb dominant (as defined by the preferred limb for writing and kicking). In order to target an 'at risk' population, volunteers who respond affirmatively to any of the following questions will be included in the study:

• Have you fallen in the past 5 years?

• Have you lost your balance or almost fallen in recent memory?

• Have you reduced your activities or changed your lifestyle because you were concerned that you might fall?

• Do you ever feel unsteady when: getting in or out of a chair, changing direction when you are standing or walking, reaching for something above your head, or walking and talking to someone at the same time?

• Has anyone, such as a friend or family member, expressed concern that you may have problems with your balance or are at risk of falling?

Additionally, volunteers who report a decline in their balance greater than a 'moderate amount' in recent years, or have low balance confidence (Activity-specific Balance Confidence score less than 85% 62), will be included in the study.

To avoid potential confounding effects on balance and to ensure a relatively homogeneous cohort, the following exclusion criteria will be applied: 1) current diagnosis of neurological or sensory disorders, diabetes, depression or osteoporosis; 2) uncorrected vision problems that impair ability to read, watch television, or drive a car; 3) recurrent dizziness; 4) cognitive impairment (standardized Mini-Mental State Examination 63 score < 24); and 5) joint replacement or joint fusion. To ensure that the subjects are sufficiently fit and healthy to tolerate the training program without adverse effects, volunteers who answer 'yes' to any items on the Physical Activity Readiness Questionnaire 64 will be asked to obtain approval from their doctor prior to participating in the study. To prevent potential confounding effects of other exercise programs, volunteers who regularly (≥ once per week) participate in a supervised exercise program will be excluded. Subjects who are unable to provide written informed consent (e.g. due to poor English language comprehension) will also be excluded.

Subjects will be randomly assigned to either the perturbation-based training group or a control group. Randomization will be stratified in blocks of two according to sex, age (64–72 or 73–80 years old), baseline compensatory stepping performance (average number of 'extra' steps required to recover balance: <1 or ≥ 1), and baseline compensatory grasping performance (reaction time: <120 ms or ≥ 120 ms). Random sequence generation will be performed by an individual who will not interact with subjects during the balance-testing sessions (AM).

Subjects will be informed that they will be randomly assigned to one of the two training programs, and will not be led to expect that either program will be more effective in improving their balance. The individual who administers the training programs (AM) will be the only member of the research team aware of the subjects' group allocation. A blinded research assistant will administer the balance tests and will perform any data processing that involves subjective judgements. Scripts will be used during testing to ensure that all subjects receive the same instructions.

A number of measures will be collected for purposes of characterising the cohort, in terms of sensory, musculoskeletal, neuromotor and cognitive function, balance and mobility, anthropometrics, psychometrics, health status and lifestyle. As indicated in Table 2, a subset of these measures will also be used to compute an overall falls-risk score (FallScreen^© 65), based on a database compiled from numerous prospective falls-prediction studies 66676869.

Each training program will last six weeks, with three 30-minute sessions per week. This duration and intensity is similar to that used in previous perturbation-based training studies 343539. All subjects will be asked to refrain from initiating any other new exercise programs, or otherwise consciously changing their activity levels, during their participation in the study. For the perturbation-based training program, each session will consist of 2–3 minutes of preparation and instruction, 10–12 minutes of step training, 3–5 minutes of rest, and 10–12 minutes of grasping training. We aim to complete at least 24 trials of stepping and 24 trials of grasping per session (three sets of eight trials for each).

The control subjects will undergo a flexibility and relaxation program, and will receive the same degree of interaction with experimenters as those in the perturbation-training group. The first session of each week will involve 30 minutes of passive muscular relaxation exercises 70. The second and third sessions will focus on lower- and upper-body flexibility exercises, respectively. In the flexibility sessions, a ten-minute warm-up (heart rate kept below 60% of maximum) will precede 15 minutes of stretching exercises, (two sets of 7–9 stretches for each muscle group, with each stretch held for 15 seconds). The control program is designed to involve a limited intensity of physical activity, so as to avoid bringing about any significant physiological changes. A similar, but more intense, flexibility and relaxation program failed to reduce falling 9, or brought about only moderate improvements in falls risk 71. Subjects will be asked not to practice the relaxation or flexibility exercises outside of the scheduled sessions during the study.

Two balance-laboratory assessments will be performed to evaluate the effects of the training on compensatory stepping and grasping. The first session will occur not more than one week before the start of training and the second no more than one week after the end of training. Two different types of balance perturbation will be used: waist-pulls and support-surface translations. Because support-surface translations are also used during the perturbation-based training, testing with surface translations will allow us to identify whether the targeted aspects of the change-in-support reactions were improved by the training. The waist pulls involve perturbation of the centre of mass (rather than the base of support), elicit different patterns of body motion and sensory drive, and evoke stepping and grasping reactions that differ in some key respects in comparison to reactions evoked by platform perturbation 7273; hence, the waist-pull tests will allow us to begin to examine whether any of the benefits of the training are generalisable to other balance-loss situations.

All laboratory tests will be performed with the subject standing at the centre of a large (2 × 2 m) computer-controlled motor-driven motion platform 1774. The waist-pull system is mounted on the motion platform, thereby allowing the type of perturbation to be varied unpredictably during testing (Figure 4). Waist-pull perturbations will be applied by dropping a weight (20% of body weight), attached to a belt worn around the pelvis via a custom-built cable and pulley system. Four cables are attached to the belt to allow perturbations to be delivered unpredictably forward, backward, left or right. The weight is dropped 40 cm for stepping trials and 30 cm for grasping trials. For the support-surface translations, we will use a displacement, velocity and acceleration of 0.18 m, 0.6 m/s and 2.0 m/s², respectively, for forward translations (which evoke backward falling motion), and 0.27 m, 1.0 m/s and 3.0 m/s² for the other translation directions (backward, left, right). We know, from pilot tests and previous studies 222324, that the selected waist-pull and surface-translation parameters are tolerated by older adults and will consistently evoke stepping or grasping reactions.

Perturbation direction and type (i.e. waist pull or surface translation) will be varied in an unpredictable pseudo-random sequence so as to prevent subjects from pre-planning their response. An initial set of 12 familiarisation trials will be completed at the start of each laboratory session in order to dampen within-session habituation effects 47. Following this, three blocks of perturbation trials will be completed: 1) compensatory stepping during stance, 2) compensatory stepping while walking in-place, and 3) compensatory grasping during stance. Table 3 details the task conditions and numbers of trials completed in each trial block.

During compensatory stepping tests, subjects will be asked to respond to the balance perturbations naturally, but to minimise the number of steps taken if they do step. To restrict arm movement, they will hold a lightweight rod behind their back 74. To deter conscious efforts to pre-plan or modulate the reactions, subjects will be required to perform a concurrent distraction task (counting backward by 3's from a 3-digit number randomly assigned at the start of the trial) while waiting for the perturbation, during the stance trials (blocks 1 and 3). In the second block of trials, they will perform a walk-in-place task, (lifting the feet in time with a metronome, 100 beats per minute) to increase the probability that a foot collision will occur 22. For compensatory grasping trials, a handrail will be mounted to the right of the subject. Subjects will be instructed to grab the rail with the right hand as quickly as possible to recover balance, in response to the perturbation. They will be told not to step, and foam blocks will be placed around their feet to further deter stepping. A safety harness will be worn throughout all of the testing.

The four primary outcome measures correspond to the specific aspects of change-in-support reactions that were targeted in the perturbation-based training program: 1) average number of 'extra' steps taken to recover balance, 2) frequency of foot collisions when responding to lateral perturbations, 3) frequency of additional lateral steps when responding to antero-posterior perturbations, and 4) handrail-contact time for grasping reactions. Secondary outcome measures relating to other features of the balance reactions will also be analysed. For compensatory stepping, these include the frequency and magnitude of anticipatory postural adjustments (APA), foot-off time, foot-contact time, step length and frequency of arm reactions. For compensatory grasping, we will analyse onset timing and magnitude of arm-muscle activation (biceps, medial deltoid), arm trajectory, and frequency of grasping errors (e.g, undershoot, overshoot, collision of hand with rail). The early 'automatic postural response' (which occurs in both stepping and grasping trials) will be analysed using centre-of-pressure measures and the latency and magnitude of the initial ankle-muscle activation (tibialis anterior, medial gastrocnemius).

The kinematic measures (step length, arm trajectory) will be determined by using a six-camera infrared motion-analysis system to record displacement of reflective markers placed bilaterally on the feet and arms. The behavioural measures (foot collisions, additional steps) will be determined using a computer-based algorithm to process the foot-marker kinematics; results will be verified by manual inspection of these data and associated video recordings by a blinded investigator. In addition, effectiveness of the stepping and grasping reactions in preventing "falls" will be analysed by using a load cell to monitor the force applied to the safety harness. Step timing, APAs and centre-of-pressure will be characterized using three force plates built into the surface of the motion platform. Handrail-contact timing will be determined from force-sensing resistors mounted on the rail. Muscle activation will be recorded bilaterally using surface electromyography (EMG). All timing measures will be defined in relation to onset of platform motion (measured with accelerometers) or onset of weight-drop (measured with a load cell). See previous articles for more details about the measurement and analysis methods 747576.

Other secondary outcome measures will be analysed to evaluate whether the training had any benefits extending beyond effects on balance reactions. We will explore indirectly the effect on overall falls risk using the FallScreen^© battery noted earlier. In addition, we will explore effects on: 1) balance confidence (Activity-specific Balance Confidence questionnaire 62); 2) mobility ('Timed Up & Go' test 77); 3) flexibility (modified 'sit and reach' test 78); and 4) lower-limb power (using the repeated step-up test 79 as a proxy measure 80).

The effect of the two training interventions on the outcome measures will be determined using repeated-measures analysis of variance (ANOVA) with two factors: 1) group (perturbation-based training versus control group), and 2) session (pre- versus post-training). Any of the cohort descriptors (measured at baseline, as detailed earlier) that show significant inter-group differences in preliminary univariate analyses will be included as covariates in the main analysis. As noted earlier, the control training program was specifically designed not to improve balance control; therefore, changes in the control group (pre- versus post-training) are expected to represent typical learning effects resulting from repeated exposure to the balance-assessment procedure, as well as possible 'placebo' effects associated with the interventions. Analysis of the group-by-session interaction will therefore reveal any real benefits resulting from the perturbation-based training intervention.

To date, five subjects have completed the perturbation-based training and five have completed the control intervention. Sample size calculations for repeated measures ANOVA 81 were performed using the variance from these initial ten subjects. Separate calculations were performed using two of the primary outcome measures, so as to determine the sample-size requirements based on stepping reactions (average number of 'extra' steps taken to recover balance) and grasping reactions (handrail-contact time). For both calculations, probability of Type I error was 0.05, and probability of a Type II error was 0.2.

A previous study of older adults showed that the average number of 'extra' steps taken to recover balance was 1.21 in 'fallers' and 0.825 in 'non-fallers' 25. This suggests that a reduction equal to 0.5 steps, due to the perturbation-based training, will have functional significance. Using this value, in combination with the initial variance estimate (SD = 0.44 steps), it was determined that 15 subjects per group will be required. For grasping responses, the previous study 25 indicated that handrail-contact time was delayed by 50 ms, on average, in the 'fallers'. Detection of a training effect of this size, given the initial variance estimate (SD = 39 ms), requires a sample of only 12 subjects per group. Thus, to ensure sufficient statistical power for both stepping and grasping measures, it appears that 15 subjects per group will be required. However, sample-size requirements will be re-evaluated at two intermediate stages, i.e. after completing 10 and 15 subjects in each group. If any subjects are still enrolled in either training program when the new estimate for the required number of subjects is reached, these remaining subjects will complete the study and the tested sample will exceed the estimated requirement.