Approximately 50 million people are living with dementia worldwide, with numbers expected to increase to 82 million by 2030 [1]. There has yet to be a successful disease-modifying treatment for dementia. The World Health Organization (WHO) acknowledges the condition as a global public health priority [2] and recommends that people with dementia receive accessible, affordable person-centered support and care to maintain functional abilities and quality of life and remain living within the community [2]. People with mild cognitive impairment (MCI) should also be provided with such opportunities, particularly interventions focusing on modifiable lifestyle factors that could slow the progression of the disease [3,4]. Promoting habitual physical activity (HPA) may help to decelerate dependency and disability in people with cognitive impairment.

HPA refers to any movement produced by skeletal muscles that requires energy expenditure [5]. It does not always have to be planned, structured, or related to physical fitness (ie, exercise) but can be any physical activity such as walking or gardening. Supporting people with dementia and MCI to maintain their HPA may contribute to the WHO’s recommendations, with a recent meta-analysis suggesting that physical activity moderates the decline in cognitive abilities and reduces the risk of transitioning to more severe cognitively impaired states [6]. Participating in physical activity may also support functional abilities, reduce the progression of cerebrovascular pathology, decrease mortality risk, and attenuate behavioral and psychological symptoms of dementia such as depression and mood [7-9]. Conversely, physical inactivity is associated with an increased risk of comorbidities such as cardiovascular disease, which have knock-on effects in accelerating the course of disease and loss of functional independence. Given the potential health benefits, the first physical activity guidelines for older adults with MCI or subjective cognitive decline have recently been released, highlighting the growing clinical importance of this area for people with cognitive impairments [10].

However, prior research on people with cognitive impairment primarily focuses on exercise-based interventions, such as aerobic and anaerobic activities [11]. These may not be feasible for people with chronic health conditions, frailty, or those from low socioeconomic backgrounds who do not have access to appropriate interventional services [12]. People may not enjoy the types of exercise interventions offered, lack motivation, or feel uncomfortable or unconfident to participate [13]. Therefore, it may be more appropriate to consider all HPA that people with cognitive impairment participate in.

Capturing HPA behaviors in people with cognitive impairment can be challenging, as historically, this has relied on self-report methods, such as questionnaires. Self-report methods may provide inaccurate data, as they only provide a snapshot of an individual’s HPA and are susceptible to recall bias [14]. With technological advances such as wearable technology and ambient sensors [15-17], HPA can now be measured objectively and continuously, allowing a comprehensive approach that can capture the volume, intensity, pattern, and variability of HPA [18]. These HPA domains provide a framework to synthesize the current literature, consistent with a previous review quantifying HPA in aged residential care settings [19]. To truly understand the benefits of HPA, we must first summarize the current methods and metrics that have been used in the literature to capture HPA in people with cognitive impairment and how these might differ depending on the severity or type of cognitive impairment. Findings from this review will provide important insight into which methods and metrics are most common and feasible to characterize HPA in people with different degrees of cognitive impairment and will highlight how people with dementia and MCI may differ in discrete HPA metrics compared to normal aging. Given that ambulatory activities such as walking are the most accessible and common forms of HPA for older adults [20,21], this review will focus only on ambulatory physical activities.

The key aim of this review is to understand how HPA is currently measured and described in people with dementia and MCI in community-based settings and to quantify the volume, intensity, variability, and pattern of HPA in this population. To do this, this review has 4 core objectives, which are listed in Textbox 1.

A total of 6 databases were used for this search: Scopus, Web of Science, Psych Articles, PsychInfo, MEDLINE, and Embase. Key terms for the search strategy are detailed in Multimedia Appendix 1, with information regarding the full electronic search strategy. An initial search was conducted in September 2020, with a follow-up search conducted on November 2, 2021. Therefore, research articles published before November 2021 were considered for this review. This review was preregistered on PROSPERO (CRD42020216744) and designed following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines [22]. Multimedia Appendix 2 contains the completed PRISMA checklist.

Table 1 shows the eligibility criteria for article selection in this review.

All titles, abstracts, and full texts were independently screened by 2 reviewers (authors RMA and KJB) using Rayyan software developed by Ouzzani et al [23], with a third reviewer settling disagreements (author MC). Information about the decision-making process can be found in Figure 1. Data extraction forms were developed on Excel (Microsoft Corp) by author RMA and refined in consultation with 2 other authors (MC and KJB). Data were extracted from all eligible articles, with key measures of interest as follows: (1) diagnosis and diagnostic criteria applied; (2) method of HPA assessment (eg, technology, device location, time period, and criteria used to characterize HPA); (3) HPA metrics assessed and their values (eg, volume, intensity, pattern and variability metrics; see Table 2 for definitions); and (4) main finding of the articles in respect to HPA. A quality assessment was conducted independently by 2 reviewers (authors RMA and KJB) using the National Institute of Health Quality Assessment Tool for Observational Cohort and Cross-sectional Studies [24]. The quality assessment was adapted for this review by removing questions relating to the measurements of exposures of interest and by adding the following question: “Were clinical diagnostic criteria and severity ratings for dementia reported and adhered to?” This adapted version has previously been used in similar reviews [19,25]. Average scores determined the overall quality of each study.

Narrative data synthesis was conducted to answer our research aims. This primarily considered the key outcomes of this review—the methods and metrics used to assess and characterize HPA in people with cognitive impairment and significant differences in HPA metrics across the cognitive spectrum. To support data interpretation and provide consistency across the literature, we adopted the same HPA framework as Mc Ardle et al [19] in their review of HPA assessment in aged care facilities, which groups HPA metrics into 4 domains: volume, intensity, pattern, and variability of HPA. Due to significant heterogeneity across the methods, protocols, metrics, and populations in the studies included in this review, a meta-analysis was deemed inappropriate.

The initial search conducted in September 2021 identified 2880 papers (Figure 1). The updated search strategy conducted between September 2020 and November 2021 identified 514 articles (Figure 1). A total of 33 articles were included in this review following data extraction. All papers were published between 2008 and 2021.

Multimedia Appendix 3 contains information relating to all the study characteristics and key results. Studies took place in Germany (n=7, 21%), the United States (n=6, 18%), the United Kingdom (n=5, 15%), Japan (n=4, 12%), the Netherlands (n=4, 12%), Belgium (n=2, 6%), Israel (n=2, 6%), Italy (n=2, 6%), Hong Kong (n=2, 6%), Taiwan (n=1, 3%), France (n=1, 3%), Canada (n=1, 3%), Brazil (n=1, 3%), Singapore (n=1, 3%), Australia (n=1, 3%), and Norway (n=1, 3%). One (3%) study did not specify which country it took place in. The sample size of participants with cognitive impairment ranged between 7 and 323 across all studies, with mean age ranging from 63 to 89 years. Regarding participants with cognitive impairment, 61% (n=20) of studies reported ≥50% of participants as female. Only 3 (9%) studies reported the ethnicity of participants with cognitive impairment; in all 3 (100%) studies, >85% of the participants were White. Most studies were cross-sectional (n=20, 61%), with 11 (33%) studies using baseline data from randomized controlled trials/interventions and 2 (6%) being feasibility/pilot studies. There were no longitudinal observational studies.

Levels of cognitive impairment described included MCI (n=13, 39%), mild dementia (n=2, 6%), mild-moderate dementia (n=5, 15%), unspecified level of dementia (n=12, 36%), a mix of MCI and dementia (n=2, 6%), and unspecified level of cognitive impairment (n=2, 6%). Of the 12 (36%) studies that specified dementia disease subtypes, all (n=12, 100%) reported participants with Alzheimer disease; 25% (n=3) reported participants with dementia with Lewy bodies, vascular dementia, or MCI/dementia due to Parkinson disease; 17% (n=2) reported participants with frontotemporal dementia, mixed dementia, or amnestic MCI; and 8% (n=1) reported participants with early-onset dementia, nonamnestic MCI, or Korsakoff syndrome. Moreover, 28 studies (85%) explicitly described procedures to characterize cognitive impairment (eg, clinician review, cognitive score thresholds), and 64% (n=18 studies) of these used validated diagnostic criteria (eg, [26-29]). Multimedia Appendix 4 reports on the quality of the studies.

Out of 33 studies, 31 (94%) employed wearable technology to measure HPA in people with cognitive impairment, and 2 (6%) used ambient home-based sensors. The most popular device used was an accelerometer (n=23, 70%), while the most common device location was the wrist (n=13, 39%) followed by the lower back (n=8, 24%). Most study protocols asked participants to wear/use the device for seven days (n=17, 52%). Figure 2 provides further details regarding the measurement of HPA.

A total of 14 (42%) papers addressed reasons for loss of data and issues with adherence to protocol, including technical issues (n=5, 36%) [30-34], participants removing or refusing to wear devices (n=6, 43%) [30,33,35-38], insufficient data collected (n=8, 57%; <10 hours per day for 8/14 days [39]; <80% daily wear time [40]; <3 days [31]; <10 hours per day for at least 3 days [41]; <7 days [15]; <6 consecutive days [33]; incomplete recorded days [34,40]), lost devices (n=2, 14%) [15,33], organizational issues (n=2, 14%) [15,33], forgetting to wear the device (n=1, 7%) [42], and hospitalization during the data collection period (n=1, 7%) [38].

Only 1 (3%) study reported acceptability information, with 83% of participants in the study by Rawtaer et al [43] finding the use of multiple remote monitoring devices, including a wrist-worn sensor, acceptable, saying it was comfortable and enjoyable to wear but inconvenient to charge or wear out of the home in some cases. No studies considered levels of cognition or dementia subtype when reporting compliance or acceptability.

Table 2 provides definitions of all HPA metrics included in this review. Of the 33 studies included, 24 (73%) reported HPA metrics relating to volume, 15 (46%) relating to intensity, 14 (42%) relating to pattern, and 9 (27%) relating to the variability of HPA.

This is the first systematic review to report the digital methods, protocols, and metrics used to capture HPA in people with cognitive impairment and to explore differences in HPA between levels of cognitive impairment and disease subtypes. Key findings highlighted that accelerometers worn on the wrist or lower back were the most prevalent method, while metrics relating to volume (eg, steps per day) were the most common characterization of HPA in this population. People with dementia are less physically active than those with normal cognitive function, showing lower volumes and intensity, with significant differences in variability metrics and daytime patterns of HPA. Findings in people with MCI varied, but they may have different patterns of HPA compared to those with normal cognition. We highlight key recommendations for measuring and reporting HPA and to guide future directions for research in this area, as summarized in Figure 4.

Consistent with a recent review quantifying HPA in aged care, a variety of protocols, measurement devices, placement of devices, assessment periods, metrics, and criteria to characterize HPA (eg, cutoff thresholds) were used across the studies [19]. Limited information was provided regarding participant compliance and acceptability regarding these methods and protocols. All factors relating to methods and protocols can majorly impact results, and the interpretation of research findings is limited by the lack of standardization. For example, 3 (9%) studies reported daily step count as >10,000, significantly higher than most of the studies that recorded this metric [15,38,57]. Of these, 2 (66%) studies suggested that this was due to a lower cutoff threshold for quantifying HPA (ie, anything over 3 continuous steps) [15,38]. These findings highlight how the algorithm applied can alter results and emphasize the importance of validating the devices and methods used within a population with cognitive impairments. It should be noted that over half (n=17, 51%) of the studies included in this review did not report or cite information regarding the validity of the devices or methods used [31,33,35,39,40,42,43,46,48, 49,51,53,54,59,61-63]. Of those that did, devices and methods were validated in older adults [36,37,41,52,56,60], general adults [15,34,38,44,45,57,64], middle-aged females [50], people with Parkinson disease [37], and people with dementia [55]. Studies should cite or report validation results when possible to facilitate a greater understanding of discrepancies across the literature. Further work is required to ensure commonly used devices and methods are valid, acceptable, and accurate for people across the spectrum of cognitive impairment.

Of the 44 metrics captured across these studies, 28 (64%) were used only once (Table 2). Data synthesis is therefore difficult and reduces our understanding of how people with cognitive impairment participate in HPA. Metrics relating to volume (ie, steps, walk time, bouts, and activity counts per day) and intensity (ie, light physical activity or MVPA) were most commonly applied across multiple studies. When considering future studies, these metrics may be useful to employ for comparative purposes. Certain HPA metrics, such as root mean square difference, are not readily understandable and require an understanding of accelerometry, which may limit their use for clinical purposes. Similarly, a number of studies (n=7, 21%) linked global cognition and discrete cognitive domains (ie, information processing, executive function, visual perceptual abilities) with volume, intensity, and pattern of HPA [15,30,38,40,45,46,56], but there was a limited number exploring relationships between HPA and other clinically meaningful measures, such as functional independence and psychological or social outcomes. There were also no longitudinal observational studies identified for this review, which prevents minimal clinically meaningful differences and sensitivity to change measures to be quantified. Given that the loss of HPA has significant impacts on independence, health, socialization, and mental health [11,65] and is influenced by multiple socioecological factors [13], we suggest that a core set of interpretable HPA metrics should be identified which are associated with clinically important outcomes for people with cognitive impairment and their clinicians [66]. These should be monitored longitudinally to improve our understanding of trajectories of HPA change in people with cognitive impairment and how this change reflects the loss of independence, social networks, health, and psychological well-being.

Over half (n=17, 51%) of the studies measured HPA continuously for 7 days. However, protocols varied from 2 days to 3 months of continuous data collection within this review. In the general older adult population and in aged care, up to 5 days of data capture is considered sufficient for reliably estimating HPA outcomes using wearable devices worn on the trunk [67,68]. The most common continuous time period is generally 7 days, as it captures both weekdays and weekends, where structural differences are more likely to occur [60]. Key findings in this review indicate that the HPA of people with dementia varies between days, suggesting that they may have dynamic daily routines and habitual patterns, which may be affected by their degree of dependency [60,62]. Individuals with more severe cognitive impairment will have a greater reliance on social support, while those with less severe cognitive impairment may manage more of their daily activities independently, leading to structural differences in HPA (eg, between weekdays and weekends) [60]. This has important considerations for the interpretation of data and HPA assessment protocol, as averaging across days may overlook nuances in people’s behaviors [68]. There is a clear need for research to establish the required number of days for HPA assessment in people with cognitive impairment in the community to ensure that the information contributing to clinical decision-making is accurate and reliable.

Based on the studies included in this review, HPA is significantly different across all domains of measurement described in people with dementia, while the groups with MCI appeared to have different patterns of daytime activity and higher intraindividual and intradaily variability. This suggests that while the amount of HPA does not significantly change in the early stages of cognitive impairment, the way it is carried out is different. For example, patterns of daytime activity in HPA may reflect an individual’s routine, which may change over the course of the disease due to a range of socioecological factors [13]. Time-series analysis may be useful to capture the loss of functional independence; for example, a person with cognitive impairment may show a decrease in peak activity times if they are no longer responsible for key household tasks, such as cleaning or grocery shopping [38,64]. As previously highlighted in this paper, establishing associations between different HPA metrics and clinically or functionally relevant outcomes may provide new insights into the hierarchy of HPA loss across the cognitive spectrum.

Key limitations and suggestions for future research have already been highlighted regarding the lack of validity and consistency in methods and metrics reported, the lack of longitudinal research, and the limited associations between HPA metrics and clinically meaningful outcomes. A further limitation is the lack of information regarding the representativeness and inclusivity of the participant samples included in the studies in this review. Only 3 (9%) studies reported ethnicity [45,47,52], and 2 (6%) compared HPA in different dementia subtypes [33,38]. Moreover, only 1 (3%) study recruited early onset dementia, limiting our understanding of HPA in people with younger onset cognitive impairment [59]. It is unclear whether discrete dementia disease stages (ie, MCI vs dementia, mild vs moderate dementia) participate differently in HPA due to the limited research [69]. It is important to ensure the recruitment of participants from underserved groups, such as ethnic minorities, lower socioeconomic status, rarer or more advanced dementias, or people living in remote areas, so that results are generalizable and inform the development of inclusive interventions for health and social care. We need to understand physical activity participation in underserved groups like these to ensure that we are creating inclusive support strategies and interventions. Feasibility studies should be prioritized within these groups to ensure protocols, devices, and metrics used are acceptable and usable across the general population with cognitive impairments.

This systematic review had several strengths, including a comprehensive search strategy and the use of multiple databases used to screen potential articles for inclusion. Independent title, abstract, and full-text screening was carried out, with a third reviewer adjudicating disagreements. Our quality assessment suggested that most studies were moderate-to-good quality. However, based on our resources, this review was limited due to the inclusion of only articles written in English, which may have led to the exclusion of relevant studies published in other languages. Additionally, we only included HPA metrics that captured ambulatory activities, excluding functional HPA characteristics, such as standing or sitting time. These were beyond the scope of our review, but understanding how people with cognitive impairment participate in these should also be considered, as changes in ambulatory HPA will impact functional HPA (ie, more ambulatory movement may lead to less time sitting or standing) [70]. Because this is a systematic review, we were unable to account for confounding variables that may impact physical activity in people with cognitive impairment, such as multimorbidity or poor physical health; however, this should be acknowledged in future research studies. Due to the huge range of metrics, protocols, and criteria to characterize HPA, a meta-analysis was not an appropriate way to synthesize these data but should be considered as the field develops and standardizes.

Wearable technology (eg, accelerometers) is the most common HPA assessment tool, while metrics relating to volume are the most prevalent to describe HPA in people with cognitive impairment. People with dementia have lower volumes and intensities of HPA compared to normal aging, and people with both MCI and dementia show different patterns and higher variability of HPA compared to controls. The lack of standardization across methods and metrics limits our understanding; therefore, more inclusive recruitment strategies are needed in future studies to establish how people with cognitive impairment participate in HPA. Future research needs to consider longitudinal studies to understand HPA change in people with cognitive impairment and how this reflects clinically relevant outcomes.