Adherence to exercise interventions in older people with mild cognitive impairment and dementia: A systematic review and meta-analysis

Highlights • This literature review is on adherence to exercise interventions.• It focuses on people with dementia and mild cognitive impairment (MCI)• Forty-one studies are included in the review.• Mean adherence is 70%. It does not differ between participants with dementia and MCI.• Interventions with endurance/resistance elements yield higher adherence.


Introduction
The population is aging rapidly, with estimates reporting that by 2050, nearly 2 billion (22%) individuals worldwide will be 60 years old and over (World Health Organisation, 2017). These numbers represent a public health priority in view of the high prevalence of chronic disease, physical and mental health problems of aging individuals (Blondell et al., 2014). Cognitive decline associated with aging represents a major issue.
The association is not exclusive of normal brain deterioration typically occurring in healthy individuals, but it is also found in clinical conditions, such as Mild Cognitive Impairment (MCI) or dementia (Miller et al., 2012;Eric Ahlskog et al., 2011). Dementia is a syndrome causing deterioration in memory, thinking, behavior and the ability to perform everyday activities (World Health Organisation, 2017). MCI is characterized by deteriorated cognition without a significant impact on daily activities (Alzheimer's Society, 2019).
To obtain the continued health benefits associated with exercise, adherence is key (Robison and Rogers, 1994). Adherence can be intended as 'maintaining an exercise regimen for a prolonged period following the initial adoption phase' (Lox et al., 2016). A six-month home-based exercise intervention for people with dementia found that participants who adhered to ≥70% to the prescribed regime had significantly better balance at follow-up than those who adhered < 70% (Taylor et al., 2017).
Given its importance, adherence guidelines have been set around exercise for older adults. The World Health Organisation (WHO) recommend that older adults engage in at least 150 min of moderate-intensity aerobic exercise or 75 + minutes of vigorous-intensity aerobic exercise per week (World Health Organization, 2010). Older adults who cannot exercise due to health conditions, should engage in physical activity which is commensurate to their abilities as much as possible (World Health Organization, 2010). The UK Chief Medical Officers' Physical Activity Guidelines state that even minimal level of exercise (e.g. walking) generates some health benefits, as opposed to being sedentary (Gibson-Moore, 2019). However, research found poor adherence to exercise by older adults (Jancey et al., 2007;Nyman and Victor, 2011;Hawley, 2009).
In addition, adherence alone, does not necessarily produce positive intervention outcomes, which can be affected by a number of factors, including compliance and adverse events. Compliance is defined as 'conformity to a prescribed or self-prescribed fitness program' (e.g. whether the participants exercised at the prescribed intensity, such as heart rate) (Exercise compliance, 2012). Non-compliance can cause a lack of improvement in study outcomes, despite good adherence. Adverse events are defined as 'untoward medical occurrences that may present during treatment (…), but which do not necessarily have a causal relationship with this treatment' (e.g. physical ailments) (Uppsala Monitoring Center, 2020). Adverse events can cause the participants to withdraw from an intervention program before completion, a phenomenon defined as 'attrition' (Murray et al., 2013), and prevent them from obtaining the associated positive benefits, despite good adherence.
Although adherence has been investigated in a few studies focusing on exercise interventions for people with MCI and dementia (Lam et al., 2015;Tak et al., 2012), there is no literature review synthesizing the current evidence, which also identifies crucial factors such as compliance, attrition and adverse events. Considering this gap in research, the aim of this systematic review aims to fill this gap in research by investigating in exercise interventions studies for older people with MCI and dementia: 1) How adherence is defined, monitored and recorded; 2) Adherence rates; 3) Attrition, compliance and adverse events; 4) Intervention characteristics (i.e. type, length, format, intensity, frequency, duration, setting, incentives for participants) associated with adherence.

Methods
This review complied with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement (Moher et al., 2009). The review's protocol was published on the international database of prospectively registered systematic reviews in health and social care (PROSPERO) (Di Lorito et al., 2018). The search strategy (Appendix A) was based on the PICO (Population, Intervention, Comparison, Outcome) worksheet for conducting systematic reviews (Haynes et al., 1997), which identified three search domains: population (i.e. people with MCI or dementia), intervention (i.e. physical activity, exercise, or sport) and outcomes (i.e. adherence). In developing the search strategy, the research team was assisted by a librarian from the University of Nottingham, with expertise in systematic search of the literature. Minor changes to the search strategy were made to adapt it to the different characteristics of the databases.
Seven databases from relevant disciplines (i.e. medicine, sport, psychology, social sciences) were searched: Embase, Medline, PsychInfo, SPORTDiscus, AMED, CINAHL and the International Bibliography of Social Sciences. The searches were carried out in November 2018. The reference lists of the included studies and of the literature reviews retrieved through the database searches were screened to identify further eligible studies.

Study selection
After removing duplicates, title and abstract of all the records identified through the initial searches were independently screened by three authors (CDL, AB, VVDW), who eliminated clearly ineligible studies. Each of the three authors then independently screened the full texts of the remaining studies against the inclusion/exclusion criteria. Any disagreement in the selection process was resolved by consensus.

Inclusion criteria
• Empirical study collecting primary data; • Study involved people diagnosed with MCI or dementia (any type); • Study inclusion criteria for age was 65 + years old, or, if lower, the mean age of study participants was at least 70 years old; • Study tested the effectiveness of an intervention including exercise, defined as 'planned, structured and repetitive physical activity' (Caspersen et al., 1985). If the intervention included multiple components (e.g. cognitive stimulation + exercise), adherence rates must have been reported separately for exercise; • Study reported adherence to the intervention; • Any type of exercise intervention, any duration, frequency, intensity and mode of delivery (e.g. individual format, group format); • Any year and language; • Published or unpublished study (to reduce publication bias).

Exclusion criteria
• Non-empirical study (e.g. literature review), in the presence of which, its reference page is inspected, to identify any primary studies eligible for the review; • Study on stroke survivors or people with Parkinson's disease, HIV, Huntington's disease, multiple sclerosis or subjective memory complaint (i.e. not clinically diagnosed); • Study on people younger than 65 years old and with a mean age below 70 years old; • Study on functional ability (activities of daily living) interventions not including an exercise component; • Study on interventions with multiple components (e.g. exercise + cognitive training) that do not report adherence to the exercise component separately.

Study quality appraisal
Three independent raters (CDL, VVDW and AB) assessed the quality of the included studies. Each article was appraised by one rater only. The Critical Appraisal Skills Program (CASP) checklist for Randomized Controlled Trials (Program, 2019) was adapted, so that the items are relevant to a literature review around adherence. The total possible score of the tool was 13, with higher scores showing higher quality.

Data extraction
Data on study and intervention characteristics, adherence, attrition, compliance and adverse events were extracted into SPSS (Spss, 2016) using a custom designed form. The form was first piloted on a sample of three studies to ensure it captured the relevant information. The data were extracted by the main author (CDL) and checked by a second independent author (AB) to reduce error and bias.

Data analysis
Based on the study objectives, data analysis was carried out on: 1. How adherence was defined, monitored and recorded. This was reported through narrative synthesis and descriptive statistics. 2. Mean adherence weighted by study sample size. A test for heterogeneity was ran to determine whether a meta-analysis of the adherence rates from the individual studies was possible. This was carried out through Higgins' I 2 Test, which calculates the percentage of variation of adherence rates across studies due to heterogeneity rather than chance (Higgins and Thompson, 2002;Higgins et al., 2003). The thresholds used for the interpretation of I 2 , as per guidelines from the Cochrane handbook for systematic reviews of interventions (Higgins et al., 2019): 0-40% (heterogeneity not important); 30-60% (may represent moderate heterogeneity); 50-90% (may represent substantial heterogeneity); 75-100% (considerable heterogeneity). In addition, subgroup analyses were performed on a number of variables that may affect adherence, by selecting and meta-analyzing adherence rates from the studies with the relevant variables. The result was then compared with the original mean adherence (i.e. from all the studies), to determine whether the difference was statistically significant (i.e. p < 0.05). 3. Attrition, compliance and adverse events, analyzed through descriptive statistics. Parametric and non-parametric tests (as appropriate) were conducted to test a potential association between these variables and adherence, intervention characteristics (i.e. type, duration, frequency, setting, format of delivery, incentives to adherence), and participants' characteristics (i.e. cognitive scores, gender and age). P was considered statistically significant if < 0.05. 4. Characteristics (type, duration, frequency, intensity, format, setting, supervision, incentives to adherence) of interventions associated with adherence. These were identified through parametric and nonparametric tests (as appropriate). P was considered statistically significant if < 0.05.

Study selection
The initial search retrieved 146 sources. Of these, 93 were clearly ineligible. Of the remaining 53 studies, five literature reviews were removed and 17 studies added after hand-searching the references of the included literature reviews. The full text of 65 articles was assessed for eligibility against the inclusion/exclusion criteria. Twenty-four of these were excluded and a final number of 41 articles selected for the review. The process is reported in Fig. 1 through a PRISMA flow diagram (Chu, xxxx).

Study quality appraisal
Results are reported in Table 1. The included studies had a quality score of 5-12 out of 13 (mean = 8; SD = 1). All included studies used an appropriate design and reported the duration of the intervention. Most of the studies did not provide a precise estimate of adherence (n = 10, 24%) or were inconsistent in reporting adherence (n = 11, 27%) (See Fig. 2).

Study characteristics
Study characteristics are reported in Table 2. The studies were conducted from 1995 to 2018. All were published in peer-reviewed journals, except for two doctoral theses (Chu, xxxx; Volkers, xxxx). Most studies were from the United States of America (n = 11; 27%), the Netherlands (n = 6; 15%) and the United Kingdom (n = 4; 10%). All the studies were in English, except for one (Brami et al., 2018), which was in French.
More than half of the studies were randomized controlled trials (RCTs) (n = 23; 56%), more than a quarter (n = 11; 27%) were feasibility studies, three studies were uncontrolled trials (7%), one study was longitudinal (2%), one a case-study (2%), one a cross-sectional study (2%) and one a follow-up to an RCT (2%). The sample size greatly varied, based on the study design. It ranged from eight participants from the only case-study included in the review (Kuiack et al., 2004) to 494 participants from a large RCT (Lamb et al., 2018). The mean sample size was n = 92 (SD = 92). The total number of participants with MCI was 970 and 2149 participants were living with dementia.
The eligibility criteria to take part in the studies usually included age, a formal (i.e. clinical) diagnosis of dementia or MCI, and the ability to engage in physical activity. The age of the samples ranged from 70 to 89 (x ̅ = 80; SD = 5). The sample under investigation included participants at different stages of any type of dementia (n = 34; 83%) or with MCI only (n = 7; 17%). Mini Mental State Examination (Folstein et al., C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139 1975) cores were not reported in 12 studies (29%). The overall mean MMSE score, weighted by the number of participants per study was 21/ 30 (SD = 5). The weighted MMSE score mean for participants with MCI only (n = 970) was 27/30 (SD = 2), while for participants with dementia (n = 2149) was 19/30 (SD = 5).

How adherence is defined, monitored and recorded
Results for adherence are reported in Table 4. Adherence was operationally defined in half of the studies (n = 20; 49%) as "The proportion between the number of sessions attended and the number of sessions offered × 100". However, not all studies conformed to this. One study (2%) (Steinberg et al., 2009) measured adherence through the percentage of (personal) goals achieved by the individual participants against the goals set at the beginning of the study. The remaining studies (n = 20; 49%) did not define adherence, but just reported adherence rates. All studies reported adherence rates at the end of the intervention period only (i.e. they did not report adherence at different time points during the intervention).

Adherence rates
Adherence rates for each study are reported in Table 4. Overall, adherence rates ranged from 16% to 100%, with a mean adherence of 70% (SD = 21). The Higgins' I 2 Test revealed a high level of heterogeneity (I 2 = 95%; 95% C.I. 94-96). Results from the subgroup analyses are reported in Table 5. None of the subgroup analyses evidenced any statistically significant difference with the original adherence rate mean (i.e. all studies). The highest adherence was found for studies which required participants to train more than three times a week (75.0%) and the lowest for studies with including participants exercising in nursing homes (65.4%).
Only one study reported adherence six months following the intervention period (Tak et al., 2012). The study found that more than half of participants had discontinued exercise after the end of the trial, and one quarter had continued. Health complaints, lack of time, injuries and lack of motivation were the most reported reasons for not continuing.

Attrition, compliance and adverse events
Attrition rates at the end of the study intervention were reported in 35 studies (85%). It averaged 17% (SD = 13) of the initial number of study participants. It ranged from 0% to 59%. We did not find any statistically significant association between attrition and adherence, intervention characteristics (i.e. type, duration, frequency, setting, format of delivery, incentives to adherence), and participants'

Fig. 2.
Studies included in meta-analysis on adherence rates at the end of the intervention NOTE: The value for Pitkälä 15 refers to adherence of participants to the group, as opposed to the individual intervention; the value for Tappen 82 refers to the adherence of participants to the walking plus conversation, as opposed to the walking only intervention.

Table 2
Study characteristics (blank boxes indicate that the information was not provided).   C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139 C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139  Participants receiving the exercise intervention. C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139 Table 3 Intervention characteristics, as reported by the authors (Blank boxes indicate that the information was not provided).   C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139  (continued on next page) C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139 C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139  Suzuki (Bossers et al., 2015) Ninety-minute sessions including 10-min warm-up period, 20 min of muscle strength exercise, and 60 min of aerobic exercise, postural balance retraining (e.g. circuit training with stair stepping, endurance walking, and walking on balance boards) and dual-task (e.g. invent their own poem while walking) 48 2 Moderate (i.e. 60% of maximum heart rate)

Community
Group/Therapist Transportation to and from venue Tak (Brill et al., 1995) Two types: (1). Aerobic walking consisting of warm-up, moderate-intensity walking exercise, and a cool down; (2). Non aerobic exercise consisting of introduction, light range-of-motion movements and stretching, and a closing 48 2 Low or moderate (i.e. < 3 or > 3 metabolic equivalents [METs])

Group/Trainer
Tappen (Bossers et al., 2015) Thirty minutes of self-paced assisted walking interspersed with rest as needed (with vs. without conversation with supporter) 16 3 Nursing home Individual/

Student researchers
Taylor (Cancela et al., 2016) Exercises were predominantly balance focused, but also included strength and/or combined strength-balance exercises, e.g. tandem stance, knee extensions +/− weights, sit-to-stand, step ups on a block, Strength training focused on lower-body strengthening including dorsiflexion ("toe lifts"), knee extension and flexion ("knee straightening" and "back knee bends"), plantarflexion ("toe raises"), hip flexors ("marches"), abduction ("side lifts"), and extension ("back leg lifts"). Balance exercises including transfer exercises (chair stand), base of-support exercises (forward lean), and advanced walking skills (backwards walk   Hauer (Fratiglioni et al., 2000) ( Schwenk (Lautenschlager et al., 2008;Bossers et al., 2014)  Tak (Brill et al., 1995) (N sessions attended/N sessions offered) × 100 Instructor Not reported 53[46-60] (Kuiack et al., 2004) Tappen (Burgener et al., 2008)  (continued on next page) C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139 characteristics (i.e. cognitive scores, gender and age) (p > 0.05). Adverse events and serious adverse events were reported in 25 studies (61%). In those studies where they were reported, the data were extremely diverse, ranging from no adverse events at all to each study participant experiencing an average of 13 adverse events. Compliance was reported in seven studies only (17%). Again, the data were extremely diverse, ranging from 16% to 100%. The sparse data on adverse events and compliance did not allow us to test their association with adherence, intervention and participants' characteristics. Details on attrition, adverse events and compliance are fully reported in Table 6.

Characteristics of interventions associated with higher adherence
Non-parametric tests were conducted due to the non-normally distributed data associated with adherence rates. The only meaningful results were: • Adherence rates were found to be significantly associated with endurance/resistance training (U = 132, p = 0.05) and with interventions that did not include walking (U = 97, p = 0.01).
• A negative correlation, though not statistically significant, was found between adherence and intervention duration (Spearman's rank r s = −0.24, p = 0.11) and between adherence and frequency of training and adherence (r s = −0.10, p = 0.50). This suggests that when the intervention was shorter in duration or less frequent adherence was higher.
• No statistically significant effect was found regarding the use of incentives for adherence (U = 91, p = 0.48), although adherence was higher when the interventions used incentives (82%; SD = 14) compared to when they did not (72%; SD = 22).

Discussion
This systematic review and meta-analysis investigated adherence to exercise intervention studies for older people with MCI and dementia using systematic means of investigations. It found that adherence was calculated similarly across the studies as 'the proportion between the number of sessions attended and the number of sessions offered, reported in percentage'. However, less than half of the studies provided a clear operational definition of adherence, which may be due to the fact that adherence was not the primary outcome in 98% (n = 40) of the included studies and as a result it was not discussed in depth. A lack of consensus around the concept of adherence has been reported in previous research (Hawley-Hague et al., 2016). Even more sparsely reported was how adherence was monitored. It is worth noting that among the few studies which discussed adherence monitoring, self-reports from study participants were quite frequent, bearing a potential risk for biased/inaccurate information (e.g. due to social desirable responses). This risk is particularly tangible in the context of people with dementia experiencing memory loss, thus urging adoption of more reliable measures in future research.
The weighted mean adherence for all the included studies was 70%. This is in line with the rate found for older people with chronic conditions and healthy older adults. Bullard et al., for example, found ∼77% adherence among adults with cancer, CVD, and diabetes (Bullard et al., 2019), while Nyman and Victor (Nyman and Victor, 2011) reported an adherence of ≥70% for walking and class-based exercise and 52% for individually targeted exercise in healthy older adults.
The subgroup analyses did not find any statistically significant differences with the original mean adherence. Interestingly, the same C. Di Lorito, et al. Preventive Medicine Reports 19 (2020) 101139 adherence was found for participants with dementia and MCI, potentially showing how progression of cognitive deterioration may not be accompanied by reduced adherence to exercise. It might be argued that, in order to exercise, participants with dementia need more supervision from others (e.g. carers, trainers), who may boost their motivation to adhere. This may also potentially explain why older participants (i.e. > 80 years), who may require greater support to exercise, had higher adherence rate than younger participants (i.e. ≤80 years). Findings around duration, frequency and intensity, though not statistically significant, suggest that shorter (in weeks) and less frequent (in weekly sessions) interventions might be easier to adhere to for people with dementia and MCI. We speculate that there might be issues in long-term interventions in the context of dementia, as the condition might entail dramatic changes/shifts in the person's wellbeing over a short period of time, thus resulting in barriers to adherence. This hypothesis warrants further exploring. Particularly relevant, in the context of MCI and dementia, might be issues such as compliance, adverse events and attrition, which, over time, might thwart willingness and ability to adhere to the prescribed exercise regime. Unfortunately, given the lack of systematic reporting of data around these crucial variables in the included studies, we could not explore further their mediation in adherence rates The factors or strategies used to promote adherence to exercise interventions were also sparsely reported. The review found that when used, these strategies were linked to higher adherence. We add that these might also be instrumental to motivate participants to remain active, to promote enduring lifestyle change and produce sustained health benefits, once the active intervention is over. Other than the incentives identified in this review, a number of other strategies have been identified in the literature (van der Wardt et al., 2017) include using established behavior change techniques (e.g. motivational interviewing), offering individual supervision/tailoring of interventions to meet participants' needs and preferences (e.g. enjoyable activities), setting SMART (i.e. specific, measurable, attainable, relevant, timebound) goals, providing booklets/guidance on exercises, giving phone calls or reminders to participants, addressing exercise barriers, sending out information/newsletters, offering continuous support to clinicians, and delivering the intervention in group settings using music. There is also accumulating evidence on the centrality of the role of carers in ensuring adherence for participants with MCI and dementia, particularly as the conditions progress (Peach et al., 2017). The effectiveness of these strategies, however, remains to be established.
This review found that the more challenging the intervention (i.e. including endurance/resistance training and not including walking), the higher the adherence. This might be explained by the fact that more able people sign up to challenging interventions and/or that these are delivered in more cognitively intact populations with dementia. It might also indicate that physical activities that are less demanding and more likely to be already part of the daily routines of participants (i.e. walking) might make participants less motivated to fully engage.
In line with a recent research on the effectiveness of interventions to prevent frailty in older adults (Apóstolo et al., 2018), this systematic review found that, though not statistically significant, adherence was higher when the intervention was delivered in a group format and in the community (as opposed to the participants' private home), suggesting that aspects including opportunities for socialisation, competitive behavior, social pressure (e.g. feeling under the scrutiny of others) and/or modelling might promote adherence. Research has found that a group format might also have other benefits on memory, attention and executive processing (Ybarra et al., 2008;Adolphs, 2009;Fratiglioni et al., 2004;Fratiglioni et al., 2000). However, there are potential barriers associated with community-based exercise delivered in group formats. For example, the review found higher adherence rates when participants did not have to travel to exercise venues in the community to participate in the intervention. This suggests that there might be factors impinging on the willingness and ability of people with these conditions to take part in group-delivered exercise programs in the community.
This review was characterized by certain strengths and limitations. To our knowledge, it is the first work summarising the existing evidence around adherence rates in exercise interventions with people with dementia and MCI. This investigation is timely and relevant, since any intervention program aimed at these populations cannot be successful, unless acceptable adherence from participants is achieved. This work was undertaken following standardised operating procedures and reporting systems (PRISMA), which ensure internal validity to study findings. It followed a protocol published in the International prospective register of systematic reviews (PROSPERO) (Di Lorito et al., 2018).