The Effect of Therapeutic Exercise Interventions on Physical and Psychosocial Outcomes in Adults Aged 80 Years and Older: A Systematic Review and Meta-Analysis

This systematic review aimed to evaluate the effects of therapeutic exercise on physical and psychosocial outcomes in community-dwelling adults aged 80 years or older. Databases were searched from inception to July 8, 2020. Randomized controlled trials (RCTs) were screened by two reviewers who extracted data and assessed study quality. Sixteen RCTs (1,660 participants) were included. Compared to nonexercise controls there was no evidence of an effect of exercise on performance based (standardized mean differences: 0.58, 95% con ﬁ dence interval: [ − 0.19, 1.36]; I 2 : 89%; six RCTs; 290 participants; very low-quality evidence) or self-reported physical function (standardized mean differences: 1.35, 95% con ﬁ dence interval: [ − 0.78, 3.48]; I 2 : 96%; three RCTs; 280 participants; very low-quality evidence) at short-medium term follow-up. Four RCTs reporting psychosocial outcomes could not be combined in meta-analysis and reported varying results. Exercise appeared to reduce the risk of mortality during follow-up (risk ratio: 0.47, 95% con ﬁ dence interval: [0.32, 0.70]; I 2 : 0.0%; six RCTs; 1,222 participants; low-quality evidence).

The global population is progressively aging. Approximately 3 million people in the United Kingdom were aged 80 years or older in 2018, and this group is projected to increase to almost 6 million by 2043, making it the fastest growing population group (Office for National Statistics, 2018). Adults aged 80 years or older are the least physically active, and have the highest health care expenditure (England., 2018;Manini & Pahor, 2009). More than 85% of people aged 80 years or older in the United Kingdom reside in the community, rather than in nursing homes (Office for National Statistics, 2012). Optimizing physical function, quality of life, and psychosocial outcomes among this group is essential to facilitate ongoing independence.
Therapeutic exercise is participation in physical activity that is planned, structured, repetitive, and purposeful for the improvement or maintenance of a specific health condition (World Health Organization, 2010). The benefits of therapeutic exercise for all adults, and for many age-related conditions such as osteoarthritis and frailty, are well established (Fransen et al., 2015;Fransen, McConnell, Hernandez-Molina, & Reichenbach, 2014;Heyn, Johnson, & Kramer, 2008;Silva, Aldoradin-Cabeza, Eslick, Phu, & Duque, 2017). However, adults aged 80 years and older are significantly underrepresented in existing trials, and findings among adults aged in their 60s and 70s cannot necessarily be applied to those aged 80 years or older (Izquierdo, Morley, & Lucia, 2020;Witham et al., 2020).
Adults aged 80 years or older are a more heterogeneous population than younger older adults, with significant variability in magnitude of multimorbidity, disability, and physical and cognitive function (Collerton et al., 2016;Lafortune, Béland, Bergman, & Ankri, 2009;Santoni et al., 2015). In addition, loss of muscle strength accelerates from 10% to 15% per decade up to the age of 70 years, to 25% to 40% per decade beyond 70 years of age (Goodpaster et al., 2006;Hughes et al., 2001). Adults aged 80 years and older are often excluded from randomized controlled trials (RCTs) of exercise interventions for older adults, and clinically are less likely to be prescribed exercise than younger people (Smith, Collier, Smith, & Mansfield, 2019;Witham et al., 2020). A recent U.K. research priority setting exercise found that health professionals questioned whether exercise would have any impact on the health of people aged 80 years and older, and people aged 80 years or older raised concerns about the safety of exercise for them (James Lind Alliance, 2018).
To our knowledge no systematic reviews have been published evaluating the effects of therapeutic exercise specifically among community-dwelling adults aged 80 years or older. To inform patients, researchers, clinicians, and stakeholders about the effectiveness of therapeutic exercise for this rapidly growing population group a comprehensive review is required. Findings will also provide knowledge that will support the development of a therapeutic exercise intervention for people aged 80 years or older.
The objective of this systematic review was to evaluate the effects of therapeutic exercise interventions on physical function, health-related quality of life, and psychosocial outcomes in community-dwelling adults aged 80 years or older.

Search Strategy
The search strategy was developed in consultation with an academic librarian from the University of Oxford. Three components of the search strategy were developed separately (population, exercise, and RCTs), then combined using database-specific truncation terms. Search terms included controlled vocabulary (e.g., MeSH) and free-text terms. We searched the MEDLINE (via OVID), EMBASE (via OVID), and CINAHL (via EBSCOhost) databases from inception to July 8, 2020. No date or language limits were applied. The search strategy for MEDLINE is presented in Supplementary Table S1 (available online), and this was translated to the relevant syntax for each database. Supplementary searches of reference lists of included studies, relevant systematic reviews and the World Health Organization's International Clinical Trials Registry (http://apps. who.int/trialsearch) were also undertaken.

Study Selection
Randomized controlled trials involving therapeutic exercise for community-dwelling adults aged 80 years or older were eligible. Trials that included younger participants were eligible if the mean age minus 1 SD was 80 years or higher, or if there was clear presentation of results by strata for those aged 80 years or older. Trials with mixed community-dwelling and nursing home populations were eligible if <20% of the population resided in nursing homes, or if results were presented separately for communitydwelling participants. Interventions delivered to participants in inpatient hospital settings were not eligible. We excluded studies that only recruited participants affected by the following conditions: Parkinson's disease, Huntingdon's disease, Alzheimer's disease, advanced dementia, or stroke. Due to the nature of these neurological conditions, findings in these populations may not be applicable to adults aged 80 years or older without the specific disease. Some individuals with these (and other) health conditions may be included in studies of the general community; these trials were included.
Therapeutic exercise interventions were classified according to the ProFaNE taxonomy: aerobic (aimed at cardiovascular conditioning), resistance (contracting the muscles against external force such as weights, resistance band, or body weight), functional training (utilizing functional activities such as sit to stand as the training stimulus), balance training (challenging specific aspects of the balance systems), gait training (specific correction of walking technique), flexibility (stretching exercises which are practiced and progressed), or 3D (constant movement in a controlled, fluid, repetitive way through all three spatial dimensions, e.g., Tai Chi; Lamb et al., 2011). Interventions that included more than one type of exercise were classified as multicomponent. We included therapeutic exercise interventions with or without additional interventions (e.g., diet or pharmacotherapy) as long as the cointerventions were delivered to both intervention and control groups. We excluded interventions without a structured exercise program (e.g., providing pedometers without an exercise plan). Exercise could be delivered as a group intervention or individually, in an outpatient clinical setting, the participant's home or a community location (e.g., community center or gym). Interventions could be delivered in-person or via video consultation. Any comparator treatment, including usual care; no treatment; an alternative exercise treatment; pharmacotherapy, education or nutritional interventions were eligible.
Our primary outcomes were measures of physical function (performance-based or self-reported questionnaire) and healthrelated quality of life indices. Secondary outcomes were measures of psychosocial health (including anxiety, depression, and loneliness), falls, adverse events, and mortality. Trials were eligible if they reported one or more of these outcomes.
Search results were imported into Covidence (Veritas Health Innovation, Melbourne, VIC, Australia, www.covidence.org) for screening. A two-step process was used for screening and selection. In the first step, titles, and abstracts of all identified RCTs were independently screened by two reviewers (PN and VD). Following title and abstract screening, the full text of all potentially eligible articles was retrieved, and each screened independently for final inclusion by the same two reviewers (PN and VD). Disagreements were resolved through discussion, with an adjudicator (SH) available to address any unresolved disagreements.

Data Extraction
Data were extracted independently by two reviewers (PN and VD) using a customized piloted data extraction form in Microsoft Excel (Microsoft Corp., WA, DC, 2019). We extracted the following data: • Trial design (setting, sample size, inclusion and exclusion criteria, method of recruitment, length of follow-up, number of participants randomized, and number analyzed in the intervention, and comparator groups). • Characteristics of participants (age, gender, ethnicity, number randomized/analyzed, and dropouts in each arm). • Type of intervention (experimental/control components). Exercise details: type, supervised/unsupervised, group/individual, duration (weeks), frequency (per week), and intensity (subjective: e.g., self-rated scale or objective: e.g., heart rate monitor). • Adherence to intervention (method of assessment and reported data). • Outcomes measured (all time points). We extracted information on differences in outcomes between groups at follow-up only. We categorized follow-up as short-term (≥3 months postrandomization), intermediate term (4-11 months postrandomization), and long term (≥12 months postrandomization).
To avoid multiplicity, we selected one outcome measure when multiple outcome measures were reported for the same outcome in a trial. For performance-based physical function we prioritized: (a) the Timed Up and Go Test, (b) Sit to stand test (5× sit to stand or 30-s sit to stand), and (c) Tinetti test. If data were not reported in full in the published manuscript we emailed the corresponding author on two occasions, 1 month apart (if required) requesting the missing data.

Risk of Bias Assessment and Overall Evaluation of the Quality of the Evidence
Two reviewers (PN and VD) independently assessed the risk of bias of each included trial using the Revised Cochrane Risk of Bias Tool 2.0 (RoB 2; Sterne et al., 2019). Any disagreements in assessment between the two reviewers were discussed until consensus was reached.
We assessed the risk of bias for five domains: bias arising from the randomization process or lack of allocation concealment, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome (blinding), and bias in selection of the reported result. Within each domain, the two reviewers answered one or more signaling questions leading to judgments for each domain as "high risk of bias," "some concerns," or "low risk of bias." The judgments within each domain led to an overall risk of bias judgment (Higgins et al., 2020).
The overall quality of evidence was evaluated using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach (Cochrane Handbook for Systematic Reviews of Interventions, 2019). GRADE is a systematic approach to rate the certainty of evidence (high, moderate, low, or very low) across studies for each outcome of interest. Five domains are assessed, including methodological flaws of the included studies, heterogeneity of results across trials, generalizability of the findings to the target population, precision of the estimates, and risk of publication bias.

Data Synthesis
Descriptive characteristics of all included trials were summarized in tables and synthesized in narrative format by outcome. We performed meta-analyses using a random effects model as heterogeneity was expected in participant, intervention, and outcome characteristics. For continuous outcomes measured using the same scale we calculated the mean difference (MD) and for continuous outcomes measured using different scales we calculated the standardized mean differences (SMDs) with 95% confidence intervals (CIs). For binary outcomes we calculated the risk ratio (RR). Statistical heterogeneity across pooled studies was quantified using the I 2 statistic (Higgins et al., 2011). Separate meta-analyses were undertaken for studies using nonexercised comparators and trials using a different type of exercise as the comparator. Forest plots were presented by exercise type. Trials with short-medium term follow-up were combined due to the small number of studies, and those with long-term follow-up were analyzed separately. When trials reported multiple time points in one category the longest time point was included in meta-analysis. Meta-analyses were undertaken using Review Manager (Review Manager (RevMan) [Computer program]. Version 5.4, The Cochrane Collaboration, 2020).

Sensitivity Analysis
Where sufficient trials were identified, sensitivity analyses were planned to assess the effects of the exercise interventions excluding trials that included participants younger than 80 years of age; that had one or more "high risk" domains of risk of bias. Metaregression to explore the impact of trial level characteristics on outcomes was also planned where ≥10 studies provided data for each outcome. Due to the limited number of studies included in each of the meta-analyses, meta-regression was not undertaken.

Study Selection
The literature search identified 5,232 unique citations. Of these, 46 articles progressed to full-text eligibility review. We included 16 RCTs (reported in 20 articles) in narrative analysis, and 10 of these in quantitative analysis ( Figure 1).

Characteristics of Included Studies
Characteristics and details of the 16 included RCTs (1,660 participants) are presented in Table 1. Studies were published between 1997 and 2020. Trials were conducted across 13 countries, most commonly in Northern Europe (four studies; Bårdstu et al., 2020;Bechshoft et al., 2017;Hvid et al., 2016;Luukinen et al., 2006;Luukinen et al., 2007) or Australia/New Zealand (three studies; Campbell et al., 1997;Hamdorf & Penhall, 1999;Rosie & Taylor, 2007). Three RCTs were conducted in communities specifically developed for older adults (retirement villages or senior communities where older people lived independently; Bonnefoy et al., 2003;de Bruin & Murer, 2007;Hartshorn, Delage, Field, & Olds, 2002). All other RCTs were conducted with people aged 80 years or older who resided independently in the wider community.
Therapeutic exercise compared to nonexercise comparators.
Performance-based physical function: Nine RCTs reported on performance-based physical function, most commonly using the Timed Up and Go Test (n = 7). There was no evidence of a difference in effect between therapeutic exercise and nonexercise comparators on performance-based physical function at shortmedium term follow-up (SMD: 0.58, 95% CI [−0.19, 1.36]; I 2 : 89%; 6 RCTs; 290 participants; very low-quality evidence; Figure 2). This finding was consistent across exercise types. No data were reported for long-term follow-up.
Self-reported physical function: Five RCTs reported selfreported measures of physical function. There was no evidence of a difference in effect between therapeutic exercise and nonexercise comparators on self-reported physical function at shortmedium term follow-up (SMD: 1.35, 95% CI [−0.78, 3.48]; I 2 : 96%; three RCTs; 280 participants; very low-quality evidence; Figure 3). None of the RCTs included in meta-analysis used the same outcome measure, contributing to the observed high level of heterogeneity. This finding was consistent across exercise types. No data were reported for long-term follow-up.
Health-related quality of life: The single RCT that reported health-related quality of life (Gine-Garriga et al., 2013;n = 51) found statistically significant benefits of a multicomponent exercise intervention compared to a nonexercise control on the physical function subscale of the 12-Item Short Form Survey, and the physical and mental composite scores at both short-and mediumterm follow-ups (Table 2). Psychosocial outcomes: Data from the four RCTs reporting psychosocial outcomes (Ansai & Rebelatto, 2015;Hamdorf & Penhall, 1999;Hartshorn et al., 2002;Luukinen et al., 2006) were not able to be combined in meta-analysis due to significant heterogeneity in the constructs being measured and reporting of results (Table 2).
Findings varied between the included RCTs. Ansai et al. (Ansai & Rebelatto, 2015) reported no differences in scores on the Geriatric Depression Scale at medium-term follow-up between participants who undertook 16 weeks of multicomponent training or no intervention control. Luukinen et al. (Luukinen et al., 2006) also used the Geriatric Depression Scale but used it to categorize if participants were depressed or not, and found no difference in the proportion of participants classified as being depressed at long-term follow-up between those who undertook a 16-month multicomponent program and those who continued routine care. Hamdorf et al. (Hamdorf & Penhall, 1999) found that participants who completed a 6-month aerobic exercise program had significant improvements on the Philadelphia Geriatric Morale Scale compared to a nonexercise control at medium-term follow-up (Table 2).
Falls: Data from the three RCTs reporting falls could not be combined in meta-analysis due to heterogeneity in follow-up periods, measurement and reporting (Table 2; Ansai et al., 2016;Campbell et al., 1999Campbell et al., , 1997Luukinen et al., 2007). Two RCTs (Ansai et al., 2016;Luukinen et al., 2007) reported no difference in the risk of falls at short or medium follow-up between those who completed multicomponent exercise programs or no intervention control. In contrast, Campbell et al. (1999Campbell et al. ( , 1997 reported significant reductions in the relative hazard for a first fall with injury (heart rate: 0.61, 95% CI [0.39, 0.97]) and for all falls following a multicomponent home exercise program at long-term follow-up (heart rate: 0.69, 95% CI [0.49, 0.97]).
Between-exercise comparisons. Performance-based physical function: Four RCTs (n = 163; four comparisons) compared aerobic, functional, or multicomponent exercise interventions to resistance exercise comparators. There was no evidence of a difference in effect between therapeutic exercise and resistance exercise comparators on performance-based physical function at shortmedium term follow-up (SMD: 0.03, 95% CI [−0.28, 0.33]; four RCTs; 163 participants; I 2 : 0.00% ; low-quality evidence; Figure 5). This finding was consistent across exercise types. No data were reported for long-term follow-up.
Self-reported physical function: Two RCTs (n = 193; two comparisons) compared aerobic or functional exercise interventions to resistance exercise comparators. There was no evidence of a difference in effect between therapeutic exercise and resistance exercise comparators on self-reported physical function at shortmedium term follow-up (SMD: 0.14, 95% CI [−0.55, 0.27]; two RCTs; 193 participants; I 2 : 0.00%; low-quality evidence; Figure 6). This finding was consistent across exercise types. No data were reported for long-term follow-up.
Health-related quality of life: None of the included RCTs comparing therapeutic exercise to other exercise reported healthrelated quality of life outcomes.
Psychosocial outcomes: One RCT reported psychosocial outcomes (Ansai & Rebelatto, 2015). Ansai and Rebelatto (2015)      found no significant differences in scores on the Geriatric Depression Scale at medium-term follow-up between participants who undertook a multicomponent training or resistance training (mean difference: 0.00, 95% CI [−1.55, 1.55]); Table 2). Falls: One RCT reported falls. Rosie and Taylor (2007) reported no difference in the number of falls short-term between participants who undertook a functional training program and those who completed resistance exercises (RR: 0.67, 95% CI [0.12, 3.74]; Table 2).
Adverse events and mortality: Adverse event data were reported in two RCTs (n = 113; Ansai et al., 2016;Rosie & Taylor, 2007). No serious adverse events were reported. All nonserious adverse events reported were mild muscle pain. Meta-analysis showed no difference in risk of nonserious adverse events between therapeutic exercise and resistance exercise comparators (RR: 0.79, 95% CI [0.40, 1.57]; I 2 : 0.00%; two RCTs; 113 participants; moderate-quality evidence; Figure 7). Mortality data were reported in one RCT that compared multicomponent exercise to resistance exercise (de Bruin & Murer, 2007). No deaths were reported in either exercise group.

Overall Evaluation of the Quality of Evidence
The overall quality of the evidence assessed using GRADE, including reasons for downgrading, is summarized in Table 3. The certainty of evidence for all outcomes was downgraded for risk of bias and inconsistency, resulting in low confidence in the effect estimate.

Discussion
We identified 16 RCTs including 1,660 participants evaluating therapeutic exercise interventions among community-dwelling  adults aged 80 years or older. The trials included a range of exercise interventions, most commonly multicomponent or resistance exercise, with most trials using a nonexercise control group. Due to significant heterogeneity across RCTs, we were only able to conduct a limited meta-analysis. We found no evidence of a difference in effect between therapeutic exercise and nonexercise comparators on performance-based or self-reported physical function at short-medium term follow-up. There was some evidence that therapeutic exercise reduced the risk of mortality during follow-up compared to nonexercise controls. Individual trials reported significant benefits of therapeutic exercise on healthrelated quality of life, psychosocial outcomes and falls, however these findings have to be considered uncertain as many of these measures were secondary outcomes.

Results in Context
To our knowledge, no systematic reviews have been published evaluating the effects of therapeutic exercise on physical function and psychosocial outcomes among community-dwelling adults aged 80 years or older. The small number of RCTs that were included in our review was unsurprising, given that people aged 80 and over have been deemed "the great forgotten" in clinical studies (Izquierdo et al., 2020).
Our finding of a lack of evidence for the effectiveness of therapeutic exercise interventions on physical function, healthrelated quality of life, and psychosocial outcomes highlights the gap between the known physiological benefits of exercise and observed clinical benefits in published RCTs. Among the nine included RCTs that reported exercise intensity, both heart rate and resistance intensities were most commonly moderate. We cannot be certain that the therapeutic exercise was sufficient to cause physiological changes needed to improve strength and function. Previous reviews have reported that exercise is commonly prescribed to older adults at an insufficient dose to improve strength and function (Singh, 2002;Steib, Schoene, & Pfeifer, 2010). Future research examining the effects of different dosages of therapeutic exercise is required.
The variation in results of the included RCTs is consistent with a recent systematic review of meta-analyses on exercise interventions in older adults aged ≥65 or mean age ≥70 (Di Lorito et al., 2021). Di Lorito et al. also found variable results and no significant effects for performance-based physical function (Di Lorito et al., 2021). We noted that health-related quality of life was only measured in one included RCT and psychosocial outcomes were only reported in 25% of included RCTs in our review, a finding also noted by Di Lorito et al. (2021).
Our findings suggest that therapeutic exercise may reduce mortality for community-dwelling adults aged 80 years or older. This is consistent with a recent systematic review and metaanalysis of the safety and effectiveness of long-term exercise interventions among older adults (n = 28,532, mean age 74.2 years), which found that long-term exercise was associated with reduced risk of serious adverse events (including mortality; García-Hermoso et al., 2020). Generally high rates of completion and exercise adherence across the included studies also suggest that exercise interventions are acceptable and feasible for people aged 80 years or older. This supports the similar finding from an earlier systematic review of exercise interventions for frail older people living in the community (Clegg, Barber, Young, Forster, & Iliffe, 2012).

Implications for Clinical Practice
High-quality evidence demonstrating the effectiveness of therapeutic exercise for adults aged 80 years or older is lacking. While future RCTs are undertaken to improve our confidence in the effects of exercise for this population, given that exercise seems safe, it should be recommended in clinical practice. Which type of therapeutic exercise is most beneficial remains unclear, so exercise selection should be based on an individual's deficits, goals, and preferences.

Implications for Future Research
This systematic review highlights the need for high-quality RCTs evaluating the effects of well-designed therapeutic exercise interventions among community-dwelling adults aged 80 years or older. Identifying how best to facilitate participation of this population in such trials is an important issue. The Innovations in Clinical Trial Design and Delivery for the Under-served project seeks to address this issue, and the resources produced from this work should guide the design of future RCTs (Witham et al., 2020). Future RCTs should assess health-related quality of life, psychosocial outcomes, and cost-effectiveness analyses in addition to physical function, and should include long-term follow-up. Care should also be given to the content of the comparator intervention (Freedland et al., 2019). Variation in comparator interventions may in part explain the inconsistent effects observed in our review.
Given that adults aged 80 years or older are a highly heterogenous population interventions should be individualized and tailored to the goals and preferences of each participant. Exercise intervention design should be based on behavior change theory, and should include behavior change techniques such as goal setting, motivational strategies, and booster sessions, to optimize exercise uptake and adherence long term (Nicolson et al., 2017).

Strengths and Limitations
The strengths of this review include the comprehensive search, explicit eligibility criteria, duplicate assessment of eligibility, risk of bias assessment, and use of the GRADE approach to rate the overall quality of evidence.
The limitations of our review relate largely to the underlying evidence. First, the heterogenous nature of the included RCTs reflected the large inconsistency in the effect estimates of the metaanalyses. Second, the sample sizes of included RCTs were generally small, and many included outcomes were secondary. Finally, the quality of evidence for the outcomes of interest was low to very low, suggesting that the true effect may differ from the findings reported in this review.

Conclusions
Results of this systematic review and meta-analysis suggest that evidence for the effectiveness of therapeutic exercise interventions on physical function, health-related quality of life and psychosocial outcomes in community-dwelling adults aged 80 years or older is limited, and of low quality. Therapeutic exercise may reduce mortality in this population. Well-designed RCTs should be a research priority.
(Ahead of Print) Exercise Interventions for Adults Aged ≥80 Years