This is the first systematic literature review and meta-analysis to examine the overall effects of BT on proxies of balance performance and to characterize and quantify the dose–response relationships of BT modalities (i.e., training period, training frequency, training volume) leading to balance improvements in healthy older adults. Analyses of BT data from 23 RCTs revealed that BT is an effective method to improve healthy older adults’ balance performance. However, the nature of these responses is nearly identical to those reported previously in young adults (Table 3). Against our hypothesis, the results raise the possibility that age does not affect BT parameters known to produce adaptations in static and dynamic measures of balance. We discuss these findings by interpreting the general effects of BT with reference to the already available literature and by taking potential age-specific dose–response relationships into account.
Effectiveness of Balance Training
A number of reviews and meta-analyses already examined the effects of different fall prevention programs in older adults [11, 14, 55–59] and revealed that among others BT is recommended if the main goal is to reduce risk and rate of falls in older adults [11, 14, 55, 56, 58]. However, there is no systematic review and meta-analysis available that examined the effects of BT on different measures of balance performance (i.e., static/dynamic steady-state balance, proactive balance, reactive balance, balance test batteries). Our analyses showed that BT is effective in improving measures of static/dynamic steady-state, proactive, and reactive balance as well as performance in balance test batteries in healthy old age. Thereby, the effects of BT on measures of static/dynamic steady-state balance are small to medium compared with large effects on proxies of proactive and reactive balance as well as on performance in balance test batteries. Potential ceiling effects may account for the lower effectiveness of BT regarding static/dynamic steady-state balance. Another factor contributing to the small to medium effect sizes is the large difference between the complex temporal and spatial structure of the BT stimuli delivered through the BT programs and the non-specific simple structure of the static balance tests. In terms of dynamic steady-state balance five of seven studies examined habitual gait speed pre- and post-BT. The subjects mean baseline gait speed (1.3 m/s) can be classified as high and is indicative that the included subjects were not mobility limited and had a low risk of falls . Despite the fact that the weighted mean SMDbs of 0.44 was small for proxies of dynamic steady-state balance, the absolute increase in gait speed of 0.07 m/s represents a small but meaningful improvement in gait speed, particularly for healthy older adults [60, 61].
Dose–response relationships following balance training
The scrutinized studies used a broad range of training periods (4–15 weeks), frequencies (1–7 times/week), number of total training sessions (6–84 training sessions), durations of single training sessions (15–90 min/session), and total durations of BT per week (20–210 min/week). Both the general as well as the specific dose–response relationships for overall balance performance and for measures of static steady-state balance revealed that a training period of 11–12 weeks, a frequency of three sessions per week, a total number of 36–40 training sessions, a duration of a single training session of 31–45 min, and a total duration of 91–120 min of BT per week is most effective to improve balance. Given that only a few included studies reported detailed information on training volume (i.e., the number of exercises per training session, number of sets and/or repetitions per exercise, duration of single BT exercises) as well as examined the effects of BT on measures of dynamic steady-state balance, proactive balance, and reactive balance as well as balance test batteries, we were not able to quantify dose–response relationships for each specific outcome category.
Our analysis illustrates that BT lasting between 11 and 12 weeks is most effective in enhancing both overall balance performance (mean SMDbs = 1.26; 23 studies) and static steady-state balance (mean SMDbs = 1.54; 12 studies). Figure 7 illustrates that less than 11 weeks of training resulted in lower effects on balance performance. This result is in accordance with Lesinski et al. , who quantified the dose–response relationships of BT in young adults (i.e., 18–40 years). Our findings agree with those for young adults in as much as a training period of at least 11–12 weeks is more effective to improve static steady-state balance as compared with shorter training periods (Table 3). Therefore, it seems that there is no age-effect in terms of training period because both meta-analyses observed largest effects when conducting BT for 11–12 weeks. Given that only few studies examined BT periods of more than 12 weeks, this result is preliminary.
A previous review that examined the efficacy of BT to reduce falls  concluded that training interventions that involved higher dose of exercise (>50 h) were more effective to reduce falls and recommended at least 2 h of training per week for a training period of 6 months. This might indicate that a BT period of more than 12 weeks could be even more effective in improving overall balance performance.
It is of interest to know whether training-induced adaptations are stable over time or whether they decline during detraining. In this regard, a previous study  investigated the effects of static/dynamic BT under single- and dual-task conditions during unipedal stance performance with eyes opened and closed in healthy elderly fallers (n = 8; mean age 71 ± 5 years) and non-fallers (n = 8; mean age 68 ± 5 years). A 3-month detraining period resulted in a significant decline in unipedal stance performance in fallers and non-fallers. Likewise, Rossi et al.  shows that perturbation-based BT for 6 weeks improved neuromuscular responses (e.g., reaction time) following perturbations (i.e., simulation of sudden forward and backward balance loss due to a sliding apparatus) in community-dwelling older women (n = 41; mean age 67 ± 3 years). However, the training-induced gains were not stable but declined after 6 weeks of detraining. With reference to the studies of Toulotte et al.  and Rossi et al.  and the recommendation of Sherrington et al. , we advise to conduct BT on a permanent basis to counteract age-related declines in balance performance.
Our analysis revealed that a training frequency of three sessions per week is more effective to improve overall balance performance (mean SMDbs = 1.20; 23 studies) and static steady-state balance (mean SMDbs = 0.81; 12 studies) compared with BT comprising one to two sessions per week. In an intervention study, Maughen et al.  examined the specific effects of BT frequency on proxies of static/dynamic steady-state balance in healthy, physically active older adults (n = 60; mean age 73 ± 8 years). The authors were able to show that the group that conducted three sessions per week produced larger performance increases after 6 weeks of BT as compared with the group that executed BT once a week. However, the results from this study have to be interpreted with caution because it might be confounded by a higher number of total training sessions (18 vs. 6 training sessions). Still, our findings are confirmed by the recently published systematic review and meta-analysis on dose–response relationships of BT in young healthy adults  (Table 3). It appears that there is no age effect in terms of training frequency because both meta-analyses observed largest effects when conducting BT three times per week.
Training Volume (Number of Training Sessions)
Concerning the number of training sessions, our analysis revealed that an overall number of 36–40 training sessions produced the largest effects in terms of overall balance performance (mean SMDbs= 1.39; 23 studies) and static steady-state balance (mean SMDbs = 1.87; 12 studies). However, given that only one study examined the effects of more than 40 BT sessions, the result is preliminary. Sherrington et al.  showed that there are greater benefits from a higher dose of exercise (>50 h) that challenges balance and aims at reducing the risk of falls. Therefore, it might be that BT programs should contains at least 36–40 training sessions but indeed will obtain advantages of more than 40 training sessions in terms of training effects on overall balance performance.
Training Volume (Duration of a Single Training Session and Total Duration of Training per Week)
In terms of BT durations, our analyses highlighted that 31–45 min of a single BT session (mean SMDbs = 1.19; 22 studies) and 91–120 min of total BT per week (mean SMDbs = 1.93; 21 studies) seem to be most effective to improve overall balance performance. For improving proxies of static steady-state balance our analysis revealed that 31–45 min of a single BT session (mean SMDbs = 1.64; 11 studies) and 121–150 min (SMDbs = 3.19; one study only) of total BT per week produced the largest effects.
In accordance with the dose–response relationship of BT in young adults , there seems to be an inverse U-shaped relation between the effectiveness of BT and the duration of a single training session in old age. However, peak mean SMDbs values shifted to the right, to longer durations (i.e., 31–45 min) in older adults compared with young adults (i.e., 11–15 min). This shift in peak mean SMDbs can most likely be explained by the fact that most BT programs conducted in young adults (particularly in athletes) were either performed immediately before or after the sport-specific training. In older adults, training sessions consisted of BT only, included warm-ups and cool downs, and thus took more time. Taking this into account, the net balance training time appears to be almost similar in healthy older adults compared with young adults. Of note, our detailed analyses revealed that BT durations of more than 60 min produce no additional training effects in older adults. In fact, it seems to be more effective to split the total duration of BT per week (i.e., about 91–120 min) into more (i.e., three or more per week) and shorter (i.e., about 31–45 min) single training sessions, instead of longer single training sessions (i.e., ≥60 min) that are conducted one–two times per week only.
Given that only a few studies reported the number of exercises per training session, the number of sets and/or repetitions per exercise, and the duration of single-balance exercises, dose–response relationships were not computed for these training modalities. In addition, there is no methodological sound approach available in the literature on how to properly assess intensity during BT relative to the individual’s balance ability . Therefore, at this point, it is impossible to establish evidence-based guidelines for all BT modalities in healthy older adults (aged ≥65 years). However, with reference to the best practice recommendations of Sherrington et al. , it is possible to present qualitative recommendations on training intensity during BT. Sherrington and colleagues propagate that if the goal is to improve balance and to prevent risk of falling in older adults, moderate to high challenging balance exercises should be conducted in a sufficient dose (at least 50 h, this equate to around 2 h per week for 6 months). Furthermore, if the main aim is the prevention of falls in old age, practitioners should refer patients for the management of other risk factors where appropriate . Falls have multiple interacting predisposing and precipitating causes . Rubenstein  listed the important individual risk factor for falls according to 16 controlled studies and deduced the following order of priority: muscle weakness, balance deficit, gait deficit, visual deficit, mobility limitation, cognitive impairment, impaired functional status, and postural hypotension. Therefore, other intervention programs should be included in fall-preventive exercise program (e.g., strength or power training) to target a number of intrinsic fall-risk factors .
A limitation of this systematic review is the poor methodological quality of the included studies. Only 6 out of 23 studies were classified as high quality according to the PEDro Scale (PEDro score ≥6). In addition, many studies failed to report data necessary for computing SMD. Thus, future studies should report pre and post means and standard deviations for the investigated balance parameters. Moreover, further research of high methodological quality is needed to determine dose–response relationships of BT for specific training modalities such as training volume (i.e., number of exercises per training session, number of sets and/or repetitions per exercise, duration of a single balance exercise) in healthy older adults and to develop a feasible and effective method to regulate training intensity during BT. In addition, given that it is difficult to separate the impact of each training modality from that of others, that the heterogeneity between studies was considerable (i.e., I
2 = 76–92 %) and that we were not able to take the grade of instability/training intensity that has been trained into account, the present findings are preliminary and have to be interpreted with caution. Further, the highlighted comparisons of dose–response relationships in old vs. young adults are limited because we indirectly compared age-specific dose–response relationships across studies using SMDbs and not in a single controlled study. Finally, findings from this meta-analysis do not allow conclusions with regard to fall-prevention. In other words, our detailed analyses revealed effective BT modalities to improve overall balance performance as well as more specific measures of static steady-state balance. It is unclear how these performance enhancements translate into reduced fall rates.