This review was preregistered on PROSPERO (CRD42024616523) and designed in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [61]. PRISMA 2020 checklist is available in Multimedia Appendix 1. Five databases were used for this search: Embase, MEDLINE, PsycInfo, Scopus, and Web of Science. The search criteria were restricted to studies conducted up to the date of the most recent search (January 2025), with no lower date limit. Keywords relevant to cognitive impairment, wearable digital technology, and SB formed a search string (Multimedia Appendix 2).

Articles were screened in accordance with the most generally adopted definitions of SB: any sitting, reclining, or lying behavior, while awake, characterized by an energy expenditure ≤1.5 METs [22-24] or any nonupright activity [24].

Articles were included if they were (1) peer reviewed; (2) published in the English language before 2025; (3) conducted within community-dwelling or aged residential care settings, including supportive living, assisted living, residential aged care, nursing homes, and care homes; (4) investigated individuals with MCI or dementia, inclusive of all subtypes to capture the full breadth of cognitive impairment presentations [62-65]; (5) involved participant samples with a mean age of ≥50 years to ensure the inclusion of MCI and young-onset dementia [62,63,66,67]; (6) reported SB metrics measured through quantitative wearable devices (eg, accelerometry); and (7) defined SB either in accordance with the established consensus definitions or through an alternative but clearly articulated and appropriate definition.

Articles were excluded if they were (1) published in a language other than English; (2) conducted in acute, in-patient hospital, or palliative care settings (as inactivity in these environments is often medically mandated and does not represent an individual’s usual, volitional behavior [68,69]); (3) focused solely on PA or sleep-related metrics; (4) assessed SB through subjective self-report measures without the inclusion of quantitative wearable-derived data; (5) involved children, adolescents, or adults without cognitive impairments, including those with subjective cognitive impairment; (6) included mixed populations without disaggregated SB data for individuals with cognitive impairment; or (7) were not primary research articles such as conference abstracts, conference proceedings, posters, study protocols, letters to the editor, reviews, meta-analyses, or gray literature.

All titles, abstracts, and full texts were independently screened by 2 reviewers (JLW and CH) using Rayyan software developed by Ouzzani et al [70]. Any disagreements were settled by a third reviewer (RMA).

Author JLW created data extraction forms using Excel (Microsoft Corp) and refined them in collaboration with 2 other authors (CH and RMA). One author (JLW) independently extracted data from all eligible articles, and the extracted data were verified by author CH.

The key measures of interest were: (1) study design, (2) participant characteristics and study setting, (3) stage of cognitive impairment (ie, MCI and mild, moderate, or severe dementia) and diagnostic criteria applied, (4) measure of outcome (SB), and (5) comparisons between groups (healthy controls or other forms of cognitive impairment). For outcome measures (SB), we extracted the (1) SB metrics and their values (eg, volume, pattern, and variability metrics); (2) method of assessment (eg, accelerometry); (3) definition of measurement (eg, body posture or cut-point thresholds); (4) device name and manufacturer; (5) data collection procedure (eg, device placement and length of assessment); (6) previously established validity and reliability of SB classification (eg, validation of cut point); and (7) data processing algorithms used to derive SB outcomes, including the classification approach (eg, cut-point–based or posture-based), algorithm source (eg, academic or proprietary), and evidence of algorithm validation.

The quality of study articles was assessed using an adapted version of the National Institute of Health Quality Assessment Tool for Observational Cohort and Cross-sectional Studies [71]. This adapted version has previously been used in similar reviews [72]. Study quality was assessed based on 7 questions, which can be found in Multimedia Appendix 3.

Two raters (JLW and CH) independently assessed the quality of the studies and reached consensus through discussion; average scores determined the overall quality of each study. A binary system was used to rate the studies (1=yes and 0=no), which enabled a summed score out of 7 to aid in determining the overall quality of each study (rated good: scores 5-7, moderate: scores 3-4, or poor: scores 0-2).

Narrative data synthesis was applied to summarize and integrate findings across the studies. Reflecting this systematic review’s aims, results were grouped by (1) synthesis of reported SB outcomes in people with cognitive impairment, considering disease severity and subtype; and (2) wearable technology methods and metrics used to measure SB, including device type and placement, characterization of SB, and processing techniques and validity. Due to heterogeneity in the data, meta-analysis was infeasible.

To support the interpretation of data and provide consistency across the literature, we categorized wearable technology metrics of SB into volumes, patterns, and variability of behavior. These categories were theoretically derived, drawing on established concepts in SB and PA literature [24,72-74], and were operationalized before the final data synthesis to ensure consistent classification of metrics across all included studies. Volume refers to the amount of time spent in SB during a specified period (eg, per day or full assessment period), expressed either as an absolute value (minutes or hours) or as a relative value (percentage of time) [22]. Where necessary, time-based data were converted to a common unit (eg, minutes) to ensure consistency in aggregation and comparison. Pattern explains the way sedentary time is accumulated, including the number and duration of SB bouts, the interruptions in SB, and the distribution of SB throughout the day [24]. Variability refers to the intraindividual fluctuations or day-to-day changes in sedentary time and captures how consistently (or inconsistently) a person engages in SB across different days or within various segments of the day (eg, morning vs evening) [75]. The categorization of metrics was performed by the primary researcher (JLW) and reviewed by all authors. Any amendments were discussed collaboratively, and all revisions were agreed upon by consensus.

Figure 1 describes the results of the search strategy for articles investigating SB using wearable digital technology in people with cognitive impairment. The search was conducted between November 2024 and January 2025 and initially yielded 2822 articles; 2 additional studies were identified through citation screening. Our final systematic review included 17 articles, all of which were published between 2016 and 2025.

Table 1 contains information relating to all the study characteristics. Studies took place in Germany (3/17, 18%), the Netherlands (3/17, 18%), the United States of America (3/17, 18%), Australia (2/17, 12%), Canada (2/17, 12%), Hong Kong (1/17, 6%), Japan (1/17, 6%), Norway (1/17, 6%), and Portugal (1/17, 6%). The sample size of participants with cognitive impairment ranged between 8 and 539 across all studies, with a mean age range of 68-91.7 years. Regarding participants with cognitive impairment, 59% (10/17) of studies reported ≥50% of participants as female. A total of 3 (18%) studies reported the ethnicity of participants with cognitive impairment, whereby more than 87% of the participants were White. Most studies were cross-sectional (14/17, 82%), the remaining studies (3/17, 18%) used baseline data from randomized controlled trials or intervention studies; there were no longitudinal observational studies. Of the 17 studies, 11 included only community-dwelling participants, 3 included only institutionalized participants, and 3 included a mix of both (Figure 2 [76-92]).

Levels of cognitive impairment described across the 17 studies included MCI (n=5, 25%), unspecified severity of dementia (n=4, 20%), unspecified level of cognitive impairment (n=3, 15%), mild-moderate dementia (n=2, 10%), mild-severe dementia (n=2, 10%), mix of MCI and dementia (n=2, 10%), probable MCI (n=1, 5%), and mild dementia (n=1, 5%) (Figure 2).

For the purposes of this review, cognitive impairment was categorized into 3 broader cohorts: MCI (n=6), dementia (n=11), and mixed cognitive impairment (n=5), spanning the 17 studies. Mixed cognitive impairment refers to cohorts that have been identified as cognitively impaired with no specific diagnosis or mixed groups of dementia and MCI. Overall, 12 studies (71%) included comparisons to controls, resulting in 16 control cohorts corresponding to the dementia (n=7), MCI (n=4), and mixed cognitive impairment (n=5) cohorts (Figure 2).

A total of 8 (47%) studies specified dementia disease subtypes, of which 88% (n=7) included participants with Alzheimer disease (AD) [77,80,85,87,91-93]; 25% (n=2) reported participants with MCI/dementia due to Parkinson disease [79,91]; 25% (n=2) involved participants with vascular dementia or mixed dementia [91,93]; and 13% (n=1) included participants with frontotemporal dementia, dementia with Lewy bodies, or Korsakoff syndrome [91]. Additionally, 15 (88%) studies explicitly described procedures to characterize cognitive impairment (eg, clinician review and cognitive score thresholds), and 80% (n=12) of these used validated diagnostic criteria.

Mostly, studies were of adequate quality, with 65% (11/17) being rated as “good” and 35% (6/17) being rated as “moderate.” No studies were rated “poor.” The most common reasons for reduced study quality were a lack of sample size justification (n=15) (Multimedia Appendix 3).

The choice of classification approach appeared to influence the types of metrics reported across the included studies. Studies using posture-based classification were more likely to report pattern-related metrics (3/4, 75%) compared with those using CPM thresholds (4/12, 33%). The MET-based study did not report any pattern-related metrics. Day-to-day variability was the least frequently captured dimension, reported in only 1 study [76], which used a posture-based approach. The specific parameters used by each study are detailed in Multimedia Appendix 5, and definitions for all identified metrics are provided in Table 2. As SB volume was the only outcome reported by all studies, it was used to examine how methodological choices shaped the resulting estimates. Volume metrics reported by at least 2 studies are synthesized below and in Table 1, with a complete overview provided in Multimedia Appendix 4.

All CPM-based studies (n=12) applied devices on the wrist (n=7 [78,81,82,85,90,91,93]) or hip (n=5 [80,86,88,89,92]). Wrist-worn SB time ranged from 510 to 1203 minutes per day (n=4 [78,83,90,91]), and the proportion ranged from 43.55% to 63.2% per day (n=4 [78,81,83,85]). Hip-worn SB time ranged from 565.6 to 622.2 minutes per day (n=3 [80,86,88]), and proportion ranged from 60.94% to 87.2% per day (n=4 [86,88,89,92]). Posture-based studies (n=4) applied devices on the lower back (n=2 [79,87]), thigh (n=1 [84]), or chest (n=1 [76]). Lower‑back placement produced sedentary proportions of 78% to 89% per day [79]. Thigh placement reported 637 minutes of SB time per day [84], while chest reported 1224 minutes per day [76]. The single MET-based study (n=1 [77]) used a waist-mounted device and reported 476.2 minutes of SB time per day.

Among the studies adopting the lowest cut-point thresholds of ≤99-100 CPM, SB time per day ranged from 565.6 to 1038 minutes (n=4 [80,86,88,91]), while SB proportion per day ranged from 72% to 87.2% (n=3 [86,88,89]). In studies adopting mid-range thresholds of ≤145-149 CPM, SB proportion ranged from 57% to 60.94% (n=2 [92,93]). For those using a higher threshold of ≤178.5 CPM, SB time per day ranged from 626.9 to 1203 minutes (n=2 [78,90]), with SB proportion per day ranging between 43.55% and 61.65% (n=2 [78,81]).

Posture-based studies generally provided more granular data than threshold-based methods. Two such studies reported daily SB time (637-1224 minutes per day) and further distinguished between sitting (558-678 minutes per day) and lying time (21-546 minutes per day) [76,84]. Additionally, a study using body-posture classification estimated the proportion of SB spent at ≤1.5 METs to be between 84% and 89% per day [79]. The single MET-based study estimated SB time ≤1.5 METs (476.2 minutes per day) [77].

Of the 10 studies that quantified SB proportion, a total of 6 [82,85,86,88,92,93] calculated proportion relative to wear or wake time, and 4 [78,79,81,89] calculated proportion relative to a full day (24 hours). Specifically, within CPM-based studies, using wear or wake time as a denominator yielded higher SB proportions (range: 57%-87.2%) compared to those calculated relative to a 24-hour day (range: 43.55%-72%).

Of the 12 studies (16 cohorts) using proprietary algorithms, 8 reported SB proportion per day [78,79,81,85,86,89,92,93] (11 cohorts; range: 43.55%-89%), and 5 reported SB time [78,83,84,86,90] (range: 510-1203 minutes per day). Both studies using academic algorithms [76,77] reported SB time (range: 476.2-1224 minutes per day). The 2 studies using custom algorithms [80,88] also reported SB time (range: 565.6-622.2 minutes per day). No clear patterns emerged across algorithm types, as estimates overlapped considerably.

This systematic review is the first to characterize SB across cognitive impairments (ie, MCI and dementia) and critically appraise the wearable digital methods used to measure it. A key finding is that while total SB volume did not consistently differ between those who are cognitively impaired and controls, their pattern appeared to. Specifically, individuals with dementia exhibited significantly fewer but longer sedentary bouts. However, this review also highlights significant methodological inconsistencies in how SB is measured. Variations in device type and placement, wear-time protocols, cut-point thresholds, proportion denominators, and reported outcomes limit cross-study comparisons and, consequently, our understanding of SB and its influencing factors.

To reflect Aim 1, this review suggests that the relationship between cognitive impairment and SB may be characterized less by a simple increase in volume and more by a potential shift in pattern. The available evidence indicates that people with dementia tend to exhibit longer and fewer sedentary bouts compared with those with MCI and cognitively intact controls [79,85,93], a profile that could be indicative of more prolonged, uninterrupted SB, which is often considered the most hazardous type [94]. However, no longitudinal research was identified in the review, meaning we cannot ascertain how SB patterns change with disease progression or if it predicts worse outcomes. Furthermore, our synthesis points to a nuanced conclusion: a significantly greater volume of SB does not appear to be a consistent hallmark of cognitive impairment. Among the 12 studies (comprising 18 cohorts) that considered control comparisons, only 4 identified a significant increase in SB volume [85,90,91,93]. Of these, 3 (75%) were specific to dementia cohorts, while 1 pertained to a group with mixed cognitive impairment. The existing self-report literature, which focuses on physical inactivity, provides a relevant context for interpreting the current findings on SB. Studies have found that physical inactivity is not directly associated with lower global cognition in community-dwelling adults with dementia [95-100]. Instead, inactivity appears to be more closely linked to other factors, including depression, a history of falls, fewer waking hours, and environmental barriers [95,99-102]. This suggests that cognitive impairment may not inevitably lead to pervasive inactivity, which may also extend to SB, pointing to the influence of other mediating factors.

The apparent discrepancy between inconsistent volume and consistent pattern findings may be explained by neuropsychiatric symptoms and their interaction with the environment, rather than the disease pathology itself. For instance, apathy, a prevalent symptom in dementia, varies significantly between individuals and may manifest not as a greater total sedentary time, but as a reduced frequency of interruption, leading to the observed pattern of prolonged bouts [32,103]. This perspective is supported by Alonzo’s [104] theory, which posits that the impact of an illness emerges from its interaction with the individual’s environment. In this view, symptoms like apathy and disorientation interact with environmental contexts (eg, institutional settings and lack of stimulation) to promote sedentariness. Qualitative evidence corroborates this mechanism, identifying symptoms such as exhaustion and disorientation as common barriers to activity [103]. Therefore, future qualitative research is warranted to further elucidate the lived experience of sedentariness in this population.

It is important to note, however, that this interpretation is constrained by the methodological limitations identified in this review. The widespread use of wrist-worn devices and unvalidated cut points [37,105] is likely to impair the detection of short, nonpostural movements that break up sedentary time. Therefore, the current evidence base may underestimate the true extent of sedentary patterns in cognitive impairment, and the potential role of symptoms warrants further investigation with more precise and standardized measurement tools.

Furthermore, the generalizability of these findings across dementia subtypes remains uncertain. The available evidence is predominantly derived from cohorts with Alzheimer disease (88% of studies that reported dementia subtypes). Other dementia subtypes were included but not analyzed separately [91,93], preventing insights into how their unique symptomatic profiles might influence distinct SB patterns [106,107]. The current literature’s focus on Alzheimer disease underscores a significant evidence gap regarding SB in non- Alzheimer disease dementias, which future research should devote attention to.

Aligned with Aim 2, key findings in this review indicate that, in keeping with existing literature [108], studies quantified SB through 2 distinct methods: identifying body posture or applying cut-point thresholds. These approaches also shape the type of outcomes that can be derived; posture-based methods more readily capture pattern-related metrics (eg, bouts and breaks), whereas cut-point threshold approaches primarily quantify overall sedentary volume. This methodological divergence is compounded by differences in device placement and detection mechanisms. For posture-based measurements, thigh-worn devices struggle to distinguish sitting from lying [43], while lower-back devices struggle to distinguish sitting from standing [109]. In contrast, cut-point thresholds do not distinguish between postures and are highly sensitive to nonambulatory arm movements. Consequently, wrist-worn devices may misclassify sedentary fidgeting as nonsedentary time [110,111].

This pattern is reflected in the findings of the present review: trunk placements (eg, hip, lower back, and waist) generally reported higher and more stable estimates of SB time and proportion, whereas wrist-worn devices produced the lowest SB proportions and the greatest variability in SB time. However, these differences may also be partially explained by how SB is characterized, as wrist-worn studies demonstrated greater variability in cut-point thresholds compared with more standardized placements such as the hip. Collectively, these findings suggest that differences in SB estimates may be driven less by device placement alone and more by underlying methodological choices (eg, CPM thresholds), which reflect differing conceptualizations of SB.

The CPM approach was the most common method adopted to measure SB in cognitive impairment, with 71% (12/17) of studies using this method. The CPM approach is widely used in SB research [112] and functions as a proxy measure for energy expenditure (METs) [40], aligning with the standard SB definition of ≤1.5 METs. However, its application is highly inconsistent; 6 different CPM thresholds were identified in this review, none of which were validated in cognitively impaired populations. Although 2 thresholds (178.5 CPM [78,81,90] and 1853 vector magnitude CPM [85]) were validated in healthy older adults, the remainder lacked validation in any older population, significantly undermining the validity and comparability of findings.

The consequences of this inconsistency were most apparent in estimates of sedentary proportion. Unlike the broader literature, where lower cut points typically yield lower sedentary estimates [43,113], studies in this review using lower thresholds (≤99–100 CPM) reported the highest sedentary proportions, whereas higher thresholds (≤178.5 CPM) produced lower estimates. This inverse pattern suggests that lower cut points validated in younger populations may overestimate sedentary time when applied to older adults with cognitive impairment, while thresholds derived from older populations may provide more accurate estimates. These trends were not observed for absolute sedentary time (minutes per day) or by algorithm type (proprietary, academic, and custom), indicating that proportional differences may instead reflect denominator choice. Studies using waking or wear-time denominators reported higher sedentary proportions than those using a 24-hour denominator, as excluding sleep and nonwear reduces the denominator and inflates the proportion of time classified as sedentary. This aligns with methodological evidence demonstrating that accelerometer processing decisions, including wear-time definitions, substantially influence sedentary estimates and complicate comparisons across studies [114-117]. Together, inconsistencies in cut-point selection and denominator choice limit comparability across studies and hinder evidence synthesis.

While less common, posture-based methods (used by 24% of studies) offer a more direct assessment of SB. This approach circumvents the need for population-specific energy expenditure validation and may offer more accurate detection of SB [24,43]. Matthews et al [29] have called for such objective definitions to incorporate posture, arguing that energy-based criteria alone may not fully capture the nature of SB. However, this method introduces a different set of limitations; crucially, it cannot estimate METs and is therefore limited to classifying posture rather than intensity [118]. This is a significant constraint, as it diverges from the consensus energy-based definition of SB (≤1.5 METs). Therefore, the definition provided by Chastin and Granat [24] (“any nonupright activity”) may offer a practical alternative. Thus, the choice of method to measure SB presents a fundamental trade-off: prioritizing conceptual alignment with the energy-based definition of SB, often via CPM, at the potential cost of measurement accuracy (due to the current lack of validated CPM thresholds), or opting for the practical accuracy of a posture-based definition (“any nonupright activity”) that may neglect the MET component of SB.

It is important to recognize that SB, like all behaviors, is multifaceted and challenging to fully capture with a single measurement [38]. Therefore, future research should carefully consider the limitations of existing measurement tools and select those best aligned with specific research objectives and target populations. Crucially, measures must adhere to core principles of good measurement: demonstrating reliability, validity, population specificity, and sensitivity to change [119,120]. By emphasizing such criteria, researchers can enhance the accuracy and relevance of SB assessment, particularly in populations with unique physiological profiles such as cognitive impairment.

To aid future research and advance the field, we propose a set of recommendations to guide more meaningful and standardized approaches to measuring SB in people with cognitive impairment (Textbox 1).

This review has several strengths. It is the first to systematically examine the wearable digital technology methods and metrics used to assess SB in individuals with cognitive impairment, offering a comprehensive overview of current practices and highlighting key methodological inconsistencies. The study also extends its scope to consider the influence of cognitive impairment severity and disease subtype on SB, which has received limited attention in previous research. A rigorous search strategy, transparent inclusion criteria, and quality appraisal of included studies further strengthen the reliability of the findings.

However, this review’s findings are subject to important limitations. The categorization of SB into volume, pattern, and variability, while necessary for synthesis and aligned with existing frameworks [72,121], may not capture all relevant dimensions of SB. Furthermore, the deliberate exclusion of home-based technologies, though methodologically justified by their current lack of standardized validation and restricted ecological scope [48,122], delimits the evidentiary basis exclusively to wearable-derived data. Consequently, the conclusions presented here should be interpreted within the context of these methodological boundaries.

This review establishes that while preliminary evidence suggests that SB patterns differ across cognitive impairment, differences in SB volume are not consistently observed. Furthermore, the increasing use of wearable technology to assess SB in cognitive impairment is hampered by significant methodological heterogeneity and a critical lack of population-specific validation. This heterogeneity underscores the critical need for a consistent methodology to advance the field and establish robust, reproducible evidence on the role of SB in cognitive impairment.