Postural Control Dysfunction and Balance Rehabilitation in Older Adults with Mild Cognitive Impairment

Older adults with mild cognitive impairment (MCI) are at an increased risk for falls and fall-related injuries. It is unclear whether current balance rehabilitation techniques largely developed in cognitively intact populations would be successful in older adults with MCI. This mapping review examined the available balance rehabilitation research conducted in older adults with MCI. Databases Medline, Cinahl, Cochrane, PubMed, Scopus, and PsycINFO were systematically searched from inception to August 2020. Twenty-one studies with 16 original randomized controlled trials (RCTs) involving 1201 older adults with MCI (>age 60) met the inclusion criteria, of which 17 studies showed significant treatment effects on balance functions. However, only six studies demonstrated adequate quality (at least single-blind, no significant dropouts, and intervention and control groups are equivalent at baseline) and evidence (medium or large effect size on at least one balance outcome) in improving balance in this population, and none of them are double- or triple-blind. Therefore, more high-quality RCTs are needed to inform future balance rehabilitation program development for older adults with MCI. Moreover, few studies examined the incidence of falls after the intervention, which limits clinical utility. Future RCTs should prospectively monitor falls or changes in risk of falls after the intervention.


Introduction
More than one in four adults aged 65 years and older fall each year [1,2]. Falls are the leading cause of fatal and non-fatal injuries among older adults [1,2]. In 2015, medical costs associated with older adult falls in the U.S. totaled USD 50 billion [3]. Cognitive impairment has been established as a significant risk factor for instability and falls among older adults [4]. Older adults with mild cognitive impairment (MCI) and dementia experience higher incidence of falls compared to cognitively intact older adults [5]. About 60% of older adults with MCI and dementia experience at least a fall every year, which is twice that of cognitively healthy older adults [6,7]. Cognitively impaired individuals are also more likely to experience injurious and non-accidental falls than their cognitively healthy peers [8,9].
MCI is a condition characterized by modest cognitive decline that does not yet significantly compromise everyday life independence. It is a reversible, transition stage between normal aging and dementia [10], with a 14.9% conversion rate to dementia among older adults with MCI monitored for network, which is normally activated at rest but deactivated during a task) has also been observed in elderly fallers with and without MCI [31,33].
It is unclear whether balance rehabilitation techniques used in populations with predominantly motor difficulties would apply to individuals with cognitive impairment. Cognitive impairment may interfere with the ability to benefit from training or generalize training to activities of daily living. The MCI stage may be an ideal time to intervene, as it is a transition stage between normal aging and dementia. Once the older adult meets the criteria for dementia, there may be fewer options available for intervention given the extent of neurodegeneration and reduced ability to learn from training. Therefore, the current review will examine the available balance rehabilitation research conducted in older adults with MCI, which will inform future research and clinical care (e.g., incorporating balance training into standard of care for older adults with MCI).

Materials and Methods
A mapping review approach was selected to give an overview of the published randomized controlled trials (RCTs) which reported rehabilitation intervention effects on balance function in MCI. The purpose of this review was to identify gaps in the balance rehabilitation research for MCI and inform more specific future reviews and/or research studies on this particular area. The following databases were systematically searched from inception to August 2020: Medline (via EBSCOhost), Cinahl (via EBSCOhost), Cochrane, PubMed, Scopus, and PsycINFO (via Ovid). Searches used the following combination of keywords: (mild cognitive impairment) AND (exercise OR cognitive training) AND (balance OR postural OR fall).
Risk of bias was evaluated based on treatment allocation blinding (participants, assessors, and trainers), equivalency in baseline characteristics among treatment groups (or if differences were adequately accounted for in the analyses), and dropout rates [36]. Studies with significant dropout rates (>20%) [36] were noted in Table 1. Effect sizes (Hedges' g) were calculated to determine the strength of the findings. For the effect sizes, pooled standard deviations from the experimental and control groups were weighted by sample sizes, and the baseline pooled standard deviations were used to calculate post-intervention effect size to minimize the influence of the interventions on the standard deviations [37]. Effect size was first calculated as the difference between the intervention and control groups at each time point (pre-and post-intervention). The difference between the two effect sizes were then derived as the final effect size for each comparison. Small sample sizes (total sample size [n] < 50) were corrected by multiplying the effect size with a bias correction factor ([n − 3]/[n − 2.25] × [n − 2]/n) [38]. Effect sizes were only calculated for studies purporting statistically significant group effects. 3 groups: an elastic band-base high-speed power training (HSPT), a low-speed strength training (LSST), or a control group; no blinding specified; significant dropout rates (30% in HSPT, 53% in LSST, and 63% in control); did not report whether baseline characteristics were equivalent among groups both exercise regimens were based on the use of elastic exercise bands (the elastic band-base HSPT included a contraction phase instructed to be carried out as quickly as possible) balance and tone exercises

Results
The keyword search identified 655 articles, including 128 articles in EBSCOhost, 187 articles in Cochrane, 160 articles in PubMed, 98 articles in Scopus, and 82 articles in Ovid. After removing duplicates, 404 articles were checked for relevance based on the title and abstract using the following criteria: (1) the study sample included older adults with MCI; (2) a balance intervention was conducted; (3) the study design was RCT; (4) the outcomes included objective balance measures; (5) the article reported intervention results; (6) the article was written in English. The full texts of all articles that were deemed potentially relevant were then read by the authors using the same criteria. Additional articles were identified from references in relevant review articles that were found. The current review yielded a final sample of 21 RCTs.

Year of Publication
The inception of balance rehabilitation studies in MCI was in the beginning of the last decade (Table 1)

Group Design
Most original RCTs (44%, n = 7) had a two-arm design comparing the physical training paradigm of interest with an alternative physical training paradigm (n = 2) [40,53] or a no-intervention control group (n = 5) [43,47,54,57,58] (Table 1). Thirty-one percent (n = 5) had a two-arm design comparing a combined physical and cognitive training group with a physical training group (n = 3) [41,44,50], a cognitive training group (n = 1) [49], or a no-intervention control group (n = 1) [46]. Thirteen percent (n = 2) [42,56] had a four-arm design, which included a combined physical and cognitive training group, a physical training group, a cognitive training group, and a no-intervention control or waitlist control group. Six percent (n = 1) [39] had a three-arm design, which included two combined physical and cognitive training groups with different levels of cognitive demand and a cognitive training only group. Six percent (n = 1) [59] had a three-arm design comparing the physical training paradigm of interest with two alternative physical training paradigms.

Author, Year
Training Duration Training Intensity Progression

Risk of Bias
Six percent (n = 1) [53] of the original RCTs had a double-blind design, with the trainers and assessors blinded to the treatment allocation; participants were unblinded. Sixty-three percent (n = 10) [40,41,43,44,46,47,54,[56][57][58] used a single-blind design, of which, all but one RCT blinded the outcome assessors from treatment allocation (the other RCT [54] blinded the statistician). Thirty-one percent (n = 5) [39,42,49,50,59] did not specify whether they masked the treatment allocation in intervention delivery or data analysis. All but one study reported no significant differences in baseline characteristics between the experimental and control groups, or accounted for baseline differences in their primary analyses. The remaining study [59] did not compare baseline characteristics in the manuscript (Table 1).
Twenty-five percent (n = 4) [39,43,53,59] of the original RCTs noted significant dropout rates (>20%) and differential dropout patterns among the treatment arms (one of the studies [39] changed their randomization scheme and had subjects self-select into the treatment arm with an especially high dropout rate). Although dropout rate was less than 20%, one study [49] reported that those who dropped out had significantly lower cognitive scores than those who stayed in the study (Table 1).
Of the 17 studies that reported significant intervention effects, four studies yielded small effect sizes (Hedge's g ≈ 0.2) [39,52,53,57], three studies yielded medium effect sizes (Hedge's g ≈ 0.5) [43,48,58], and four studies yielded large effect sizes (Hedge's g ≈ 0.8 or larger) [40,50,54,55]. The remaining six studies [41,42,46,51,56,59] had variable effects sizes depending on the balance outcome and/or treatment group. Three studies [41,46,51] had a range of small to large effect sizes among their balance outcomes (in one study [51], the large treatment effects were primarily driven by pre-intervention differences rather than post-intervention differences between groups). One study found a medium effect in the physical training only group and a large effect in the combined physical and cognitive training group [42]. Another study found a small effect in the cognitive training only group and a medium effect in the combined physical and cognitive training group [56]. The last study yielded large effect sizes when the two intervention groups were compared against the control group, but the difference between the two intervention groups was small [59].
Only six studies [40,41,46,48,56,58] had at least the assessors blinded (single-blind), without significant dropout rates, and yielded a medium or large effect size on at least one balance measure. Choi and Lee (2018) [40] compared a ground kayak paddling exercise intervention (paddling while sitting on a chair following directions of trainers) with a home exercise control group. They found a large treatment effect in the ground kayaking group on the FRT relative to control. However, there were no significant group differences in other balance measures (TUG and BBS). The same research team followed up with a similar RCT, in which the intervention group performed kayak paddling exercises in a virtual environment (following movements of a pre-recorded kayak on a screen) [41]. In this study, statistically significant effects were found across all balance measures relative to control, and effect sizes ranged between small and large (with most in the medium range).
Delbroek et al. [46] reported that virtual reality physical and cognitive dual-task training on the BioRescue posturography platform significantly improved TUG performance, in comparison to a no-training control group. Three out of five subscores for TUG showed statistically significant effects (two with small or negligible effect sizes and one with a large effect size); five other reported measures were not statistically significant (Tinetti, four subscores for dual-task version of TUG). Another RCT compared a combined physical and cognitive training with physical training only, cognitive training only, and waitlist control [56]. No significant difference was found among groups in fall rate and PPA fall risk index post intervention, although TUG was improved in the combined training (medium effect compared to waitlist) and cognitive training only (small effect compared to waitlist) groups at follow-up. Although the study did not directly compare among the non-waitlist groups, our calculation of the effect sizes found small to medium effect (Hedges ≈ 0.4) sizes when comparing the combined training group to physical training only and cognitive training only groups. Of note, the effect seen in the combined versus cognitive groups was primarily driven by a pre-intervention difference between groups rather than post-intervention difference.
Doi et al. found a medium treatment effect on the smoothness of trunk movement using a multicomponent intervention program consisting of aerobic, strength, balance, and endurance exercises, compared to an educational control group among older adults with aMCI [48]. However, there were no statistically significant difference between the intervention and control groups on the OLS and a dual task with balance demand, which the authors attributed to the balance exercise being a small part of this multicomponent intervention program [47]. Finally, Sungkarat et al. found that Tai Chi significantly improved PPA fall risk index and postural sway relative to an educational control, with medium effect sizes, among older adults with aMCI [58].

Discussion
The aim of this mapping review was to evaluate the efficacy of balance rehabilitation research conducted in older adults with MCI on their balance function, in order to inform future research and clinical care. Twenty-one studies reporting on 16 original RCTs met the inclusion criteria and were analyzed. Four studies [47][48][49]58] included only older adults with aMCI, while the rest included mixed MCI subtypes. Studies varied in their determination of MCI diagnosis, but all used acceptable criteria, such as clinician diagnoses or validated cutoff scores on established neuropsychological measures. Although the majority of the studies reported beneficial effects on one or more balance measures, many studies suffered from a myriad of methodological limitations (e.g., inadequate masking of treatment allocation, significant dropout rates) that increased their risk of bias. Two-thirds of the RCTs had a single-blind design, with the outcome assessors blinded to treatment allocation. The only double-blind RCT blinded the trainers and assessors but not the participants; this trial also had significant dropouts which limited its validity [52,53]. The remaining RCTs did not specify any masking of treatment allocation, which made them highly biased. The gold standard for RCTs is to have a triple-blind design, where trainers, assessors, and participants are blinded to treatment allocation [36]. Understandably, this may be difficult to achieve in physical training interventions as opposed to a pharmaceutical trial. However, researchers should still use an active control group (instead of no treatment) and conceal study hypotheses from the participants and study personnel to minimize placebo effects and other sources of bias.
Moreover, many studies yielded small or variable effect sizes, potentially minimizing their clinical utility. Only six studies demonstrated adequate quality evidence, with at least a single (assessor)-blind design, no significant dropouts (>20%), and reported a medium or large effect size on at least one balance outcome measure [40,41,46,48,56,58]. Among these six studies, only two studies showed consistently high treatment effects above and beyond an active control group; one used a virtual kayaking intervention [41] and the other used Tai Chi [58]. That being said, neither of these studies masked the treatment allocation from the participants or the trainers, therefore increasing their degrees of bias.
Importantly, many studies did not report effect sizes, which limited the clinical interpretability of their results. In studies that did report effect sizes, most of them only calculated the post-treatment effect for each treatment arm separately, which did not elucidate whether the intervention of interest yielded a clinically relevant effect relative to control groups. The current review calculated effect sizes comparing the intervention and control groups, which often showed vastly different effect sizes from what the original trial papers reported. Taken together, more work is needed to establish an efficacious balance intervention for older adults with MCI. Future studies need to report the magnitude of treatment effects relative to control groups.
A major limitation of the literature base is the omission of falls in study outcomes. Only three studies (two original RCTs [44,56] and one secondary report [45]) measured the incidence of falls. None of these studies found a significant treatment effect in reducing falls relative to control in older adults with MCI. Thus, the currently limited literature has not demonstrated that balance interventions can reduce the fall incidence in older adults with MCI. Given the elevated risks of falls and fall-related injuries among older adults with MCI [5,6,8,9], more high-quality RCTs focusing on improving balance and reducing falls in this population is strongly needed. To accomplish this, prospective monitoring of falls after the intervention is recommended. However, because the number of falls need to be tracked prospectively and that may not always be easy to accomplish, examining the treatment effect on lowering the risk of falls in future studies may be a good alternative as long as the estimate of "risk of falls" is objective and accurate.
As mobility and cognition are inter-related based on common neural pathways [12], combining physical and cognitive training is a promising approach in balance rehabilitation for older adults with MCI. Only two studies utilizing a combined approach demonstrated adequate quality evidence [46,56]. Both studies showed highly variable effects on the balance outcomes included, each with multiple outcomes showing no statistically significant treatment effects. The probability for obtaining a type I error increases as the number of outcomes (thus number of tests) increases. In fact, in order to obtain the highest level of evidence (Class I) according to the AAN criteria for therapeutic trials, a RCT cannot have more than two primary outcome measures [36]. This prevents studies from examining a large number of outcome measures without a priori hypotheses on gold standard measures for the constructs of interest. Nevertheless, in the limited pool of adequate quality studies, it appears that combined physical and cognitive interventions may be more efficacious than physical or cognitive training alone. These will need to be confirmed with higher quality RCTs comparing a combined intervention with physical only and cognitive only control groups to elucidate the marginal efficacy of the combined intervention. Moreover, some of the studies we examined focused on cognition rather than balance as the primary construct of interest. Therefore, the researchers may not have designed the studies in order to maximize treatment effects on balance functions. Taken together, given the limited good quality RCTs in this area, future RCTs should target balance as the construct of interest with just one or two gold standard measures as primary outcomes, in order to detect true treatment effects.
In terms of training intensity, future studies may consider designing the interventions with increasing motor and cognitive demands, as older adults with MCI may not be able to tolerate high cognitive demands at the beginning of training. Retention effects should be examined after the interventions, as most studies did not conduct any long-term follow-ups. Possible mechanisms linking rehabilitation intervention to enhanced balance function were not demonstrated in the literature. Future studies may examine neuroimaging and/or electrophysiological outcomes before and after interventions to investigate possible neural mechanisms underlying the intervention-induced balance improvements.

Conclusions
The current mapping review identified 17 studies which showed significant treatment effects on improving balance function relative to control groups among older adults with MCI. However, only six studies demonstrated adequate quality (at least single-blind, no significant dropouts, and intervention and control groups are equivalent at baseline) and evidence (medium or large effect size on at least one balance outcome) in improving balance in this population, and none of them were double-or triple-blind. Therefore, more high-quality RCTs are needed to establish an efficacious balance intervention for older adults with MCI. Moreover, few studies examined training effects on the incidence of falls, which limits clinical utility. Future RCTs should prospectively monitor falls or changes in risk of falls after the intervention. Publisher's Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.