Objectively assessed physical activity and sedentary behavior and global cognitive function in older adults: a systematic review

BACKGROUND
Both physical activity (PA) and sedentary behavior (SB) are important factors for healthy ageing. This systematic review aimed to determine the association of objectively assessed (instrumented) PA and SB with global cognitive function in older adults.


METHODS
PubMed, Embase, the Cochrane Library (via Wiley), CINAHL, PsychINFO, and SPORTDiscus (via EBSCO) were searched from inception to June 21, 2020 for articles that described associations of objectively assessed PA/SB with global cognitive function in older adults aged 60 years and older. Results were synthesized using an effect direction heat map and albatross plots portrayed estimated effect sizes (standardized regression coefficients (βs)), which were summarized in boxplots.


RESULTS
In total, 45 articles were included representing a total of 15,817 older adults (mean/median age ranged from 65 to 88 years; 49.5% female). Longitudinal studies (n = 7) showed that higher moderate-to-vigorous and light PA (MVPA and LPA, respectively) and lower SB were associated with better global cognitive function. Standardized βs of cross-sectional studies (n = 38) showed that lower SB (median [IQR], β = 0.078 [0.004-0.184] and higher LPA (β = 0.096 [0.046-0.188]), activity counts (β = 0.131 [0.049-0.224]), number of steps (β = 0.155 [0.096-0.246]), MVPA (β = 0.163 [0.069-0.285]) and total PA (TPA) (β = 0.174 [0.147-0.255]) were associated with better global cognitive function.


CONCLUSIONS
Higher PA and lower SB are associated with better global cognitive function in older adults. The greatest estimated effect sizes were found for moderate-to-vigorous and TPA, suggesting that greater duration of any PA, and high intensity PA could be most beneficial for global cognitive function.


Introduction
Physical activity (PA) reduces the risk of age-related diseases such as cardiovascular disease, type II diabetes mellitus (Wahid et al., 2016), and Alzheimer's disease (Stephen et al., 2017) and is pivotal for healthy ageing. Older adults are therefore recommended to perform 150 minutes a week of moderate-to-vigorous PA (Garber et al., 2011). However, many older adults spend at least ten hours a day in sedentary behavior (SB) (Arnardottir et al., 2013;Arnardottir et al., 2017;Fitzgerald et al., 2015;Ortlieb et al., 2014), which is defined as PA levels under 1.5 Metabolic Equivalent of Task (MET) (Pate et al., 2008). While PA and SB are related, they are not the inverse of each other, as an individual can be both sufficiently physically active and sedentary; for example, by meeting recommended PA guidelines, while being sedentary for the remaining time of the day.
Higher specific intensities of PA, namely light and moderate-tovigorous PA, have been associated with greater brain volume, higher levels of brain-derived neurotrophic factor, and greater synaptic plasticity (Brown et al., 2014;Cotman and Berchtold, 2002;Hillman et al., 2008;Spartano et al., 2019), which is known to be associated with cognitive function. Systematic reviews, including predominantly subjective (self-reported) as well as objectively (instrumented) assessed PA and SB measures, showed that higher PA and lower SB were associated with better cognitive function (Carvalho et al., 2014;Falck et al., 2019). However, subjective measures of PA and SB have been shown to both over-and underestimate actual levels of PA (Chinapaw and Slootmaker, 2009;Prince et al., 2008;Ryan et al., 2018;Tudor-Locke and Myers, 2001), indicating that these results should be interpreted with caution. Instrumented measures allow for the objective assessment of PA and SB (Ryan et al., 2018;Wullems et al., 2017) and can thereby accurately quantify the association between PA and SB with global cognitive function.
This systematic review aimed to determine the association between objectively assessed PA and SB and global cognitive function in older adults.

Information sources and search strategy
The protocol of this review was registered in PROSPERO International prospective register of systematic reviews, with registration number CRD42018103910. The electronic databases PubMed, Embase, the Cochrane Library (via Wiley), CINAHL, PsychINFO, and SPORT-Discus (via EBSCO) were searched from inception to June 21 st , 2020 for articles describing associations of objectively measured PA and SB with any health outcome in older adults using the following search terms: 'active or inactive lifestyle', 'motor activity', and 'people over 60 years of age'. The full search strategy is provided in Appendix A. Endnote (Version X8.2 Clarivate Analytics, Philadelphia, USA) and Rayyan QCRI (Ouzzani et al., 2016) were used to organize and manage articles that specifically reported associations of PA and SB with global cognitive function.

Inclusion criteria
Full-text articles published in English or Dutch were considered eligible if the following criteria were met: 1) observational or experimental study, 2) mean or median age of cohort greater than or equal to 60 years, 3) PA and SB were measured objectively (using an instrument i.e. an accelerometer or pedometer), 4) a measure of global cognitive function defined as Mini-Mental State Examination (MMSE) (Folstein et al., 1975), Montreal Cognitive Assessment (MoCA) (Nasreddine et al., 2005), Modified Mini-Mental State (3MS) Test (Teng and Chui, 1987), Six Item Screener (SIS) (Callahan et al., 2002), Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-Cog) (Skinner et al., 2012), Ascertain Dementia 8-Item Questionnaire (AD8) (Galvin et al., 2005), or any other validated assessment tool explicitly described as evaluating global cognitive function was reported, and 5) PA and SB were studied in relation to global cognitive function. Intervention studies were included if the association at baseline was reported or if data from the control group data could be used.

Article selection
Title and abstract screening and the subsequent screening of full-text articles were performed by two independent assessors (LDA and AGMR) and discrepancies regarding inclusion and exclusion judgements were settled by another assessor (EMR). References of included articles were checked for additional eligible articles.

Data extraction
Two independent assessors (LDA, AGMR or KAR) completed data extraction and disagreements were resolved by an additional assessor (ABM). The following information was extracted: first author, publication year, country, cohort, study design and follow-up (if applicable), characteristics of study cohort (population selection), sample size, age (in years), sex (number and percentage of females), objective PA and SB measuring instrument (accelerometer or pedometer), device name, wearing location of the device, number of monitor days, mean device wear time, minimum wearing duration to define a valid day (in hours per day), number of valid days required for analysis, reported measures of PA and SB and their definition, PA and SB score, tool and definition used for the global cognitive function assessment, the score of global cognitive function, adjustment model(s), statistical test(s) used to study the association of interest, effect size(s) with 95% confidence interval(s) (CI) or standard error(s) (SE), and significance level(s) (p-value).

Quality assessment
Each article was assessed for study quality and risk of bias by two assessors (LDA and AGMR) using modified versions of the Newcastle-Ottawa Scale (NOS) for cross-sectional and longitudinal studies (Wells et al., 2000), tailored to this systematic review; presented in Appendix B. Articles were assessed based on the following domains: 1) selection: representativeness of study cohort and ascertainment of exposure, 2) comparability: adjustment model and statistical test, and 3) outcome: assessment of the outcome and, in case of a longitudinal study design, adequacy to follow-up. The median of total stars (points) was set as the cut-off to discriminate between high and low quality (i.e., low and high risk of bias), defined as ≥ or < 4 out of 7 and ≥ or < 5 out of 9 stars for cross-sectional and longitudinal studies, respectively.

Statistical analysis
Associations of PA and SB with global cognitive function were selected according to the following hierarchy of adjustment: 1) age and sex, 2) age and sex, and other factors, 3) age or sex, and other factors, 4) neither age nor sex, other factors only, and 5) unadjusted. When multiple statistical analyses were used to describe the association of interest, the following order was considered: 1) adjusted linear regression, 2) adjusted logistic regression, 3) partial correlation, 4) unadjusted linear regression, including Pearson's and Spearman's correlation, 5) ANOVA, and 6) Mann-Whitney test, student's t-test, or chi-squared test.
When the same measure of PA or SB was reported in different units the following continuous measures were preferred: step counts (#/day), activity counts (#/day), and duration of PA (total PA, TPA; moderate to vigorous PA, MVPA; and light PA, LPA) and SB (unit of time/day). Intensity measures included energy expenditure (EE) and metabolic equivalent of task (MET). Frequency and accumulation of PA and SB, respectively, were characterized as the number and duration of bouts, as well as (long) breaks in sedentary time (BST). Where p-values were not reported or reported using cut-off (e.g., reported as p < 0.05), p-values were calculated using different methods. For linear regression, the upper and lower limit of the 95% confidence interval (CI) were used to calculate the standard error (SE), SE=(upper limit -lower limit of 95% CI)/(2*1.96), which was then used to obtain the absolute (abs) z-statistic (z) value, z = abs(regression coefficient/SE) to acquire the calculated pvalue (p(calc)), p(calc)=exp((-0.717*z) -(0.416*z 2 )). Where ratio measures (odds ratio (OR), risk ratio (RR)) were reported, the previously mentioned calculations were applied except for that the upper and lower limit of the 95% CI as well as the regression coefficient were first transformed into logarithms using natural log (ln) (Altman and Bland, 2011). For correlations, the sample size (n) and coefficients (including Pearson's R, Spearman's Rho) were used to obtain the t-statistic (t), t = R*√((n-2)/(1-R)) of which the absolute value was compared to a two-sided t-distribution using Microsoft Excel's T.VERD.2 T function to work out p(calc). Where mean scores were compared between groups, the means and standard deviations (sd) were used to acquire t via (mean 1 -mean 2 )/√((((n 1 -1)*sd 1 2 )+((n 2 -1)*sd 2 2 ))/(n 1 +n 2 -2))*((1/n 1 )+ (1/n 2 ))), which was compared to a two-sided t-distribution using the above-mentioned function in Microsoft Excel (ESAG, unpublished observations, 2020) (Ramsey et al., 2021). Associations for which the p-value could not be calculated were conservatively estimated as ≥0.25 (for non-significant associations) or 0.01 ≤ p < 0.05 (for significant associations) and included in the effect direction heat map but excluded from the albatross plots, as further described below.

Data visualization
Following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) (Moher et al., 2009) and Synthesis Without Meta-analysis (SWiM) (Campbell et al., 2020) guidelines, results on the associations of PA and SB with global cognitive function were synthesized using an effect direction heat map (qualitative overview) and albatross plots (quantification of estimated effect sizes).

Effect direction heat map
All associations of PA and SB with global cognitive function were ordered by sample size. The observed effect direction was derived from whether higher PA and lower SB were associated with better (positive effect) or worse (negative effect) global cognitive function, which was signified by an upwards or downwards triangle, respectively (Thomson and Thomas, 2013). Statistical significance level was indicated using a color scheme: p < 0.001 (dark blue filled triangle), 0.001 ≤ p < 0.01 (blue filled triangle), 0.01 ≤ p < 0.05 (light blue filled triangle), 0.05 ≤ p < 0.1 (light grey empty triangle), 0.1 ≤ p < 0.25 (grey empty triangle), and p ≥ 0.25 (dark grey empty triangle).

Albatross plots
Albatross plots present the estimated magnitude of reported associations (effect sizes) as a function of their sample size against two-sided pvalues, stratified by the observed effect direction. More specifically, data points represent results of individual articles transformed into hypothetical effect sizes, namely standardized regression coefficients (βs), derived from the following equation: N=(1-β 2 /β 2 ) Z p (where Z p symbolizes the z-value associated with two-sided p-values (Harrison et al., 2017). These data points are located on the right or left side of the plot, indicating a positive or negative effect direction, using the same strategy as for the effect direction heatmap, respectively. Three contour lines were superimposed onto the plot for reference to evaluate standardized regression coefficient at β = 0.1, β = 0.2, and β = 0.3. Albatross plots were generated using the Stata Statistical Software: Release 16.0 (Sta-taCorp LLC, College Station, Texas, United States) for each measure of PA or SB with global cognitive function if that measure was reported in greater than or equal to five articles. Subgroup analyses were performed to assess the influence of population selection (general versus disease population), and adjustments (adjusted versus unadjusted associations) on estimated βs.
Boxplots were generated to summarize estimated βs, but due to the number of included articles longitudinal studies were depicted individually in contrast to cross-sectional studies. Boxplots were made using Plotly (Plotly Technologies Inc., Montreal, Québec, Canada) to recapitulate the obtained β coefficients.

Cognitive parameters
Global cognitive function was assessed in the included studies using the following tools: Mini-Mental State Examination (MMSE), (Alosco et al., 2012;Amagasa et al., 2020;Barnes et al., 2008;Brown et al., 2012;Eggermont and Scherder, 2008;Hartman et al., 2018;Kimura et al., 2013;Koohsari et al., 2019;Kurose et al., 2019;Liguori et al., 2020;Manas et al., 2020;Marmeleira et al., 2017;Razjouyan et al., 2020;Siddarth et al., 2018;Terashi et al., 2019;Thapa et al., 2020;Umegaki et al., 2018;van Alphen et al., 2016;van Uem et al., 2018) Table C2). Table 2 describes all associations between PA/SB and global cognitive function. These associations are visualized by effect directions heatmaps in Fig. 2. Estimated βs obtained from albatross plots in Fig. 3 are compared in Fig. 4 via boxplots providing the median and        Zhu et al., 2017) out of four articles (Manas et al., 2020;Wondergem et al., 2020), for LPA in two Wondergem et al., 2020) out of three articles, Zhu et al., 2017) and for SB in one  out of three articles (Wondergem et al., 2020;Zhu et al., 2017). In one article each, higher steps, (Chen et al., 2020) higher EE  and shorter SB bouts (Wondergem et al., 2020) were associated with better global cognitive function. In cross-sectional studies, higher steps, higher activity counts, higher TPA, higher MVPA, higher LPA and lower SB were associated with better global cognitive function. Intensity-based accelerometer measures, reported as MET or EE, were associated with better global cognitive function. Measures of PA/SB accumulation and frequency (bouts) indicated that number and duration of PA bouts were not associated with cognitive function nor were number SB bouts, but shorter duration of SB bouts were associated with better global cognitive function (Fig. 2)

Sub-group analyses
In Appendix Figure D1, sub-group analysis by stratification for population selection (general vs. disease) showed a larger median effect size for general populations in comparison to disease-populations with median β [IQR] of 0.140 [0.034-0.225] and 0.120 [0.052-0.196], respectively. Larger effect sizes were observed for unadjusted associations compared to adjusted associations with β = 0.175 [0.090-0.248] assessment scale, N/R: not reported, SIS: Six Item Screener. Q: quartile, vs: versus, M: males, F: females, SE: standard error. P(calc): calculated p-value using different methods referred to in method section.

Article quality
According to the NOS, 36 out of 45 articles were classified as high quality (low risk of bias), including all studies with a longitudinal design (Appendix Table C3).

Discussion
In this systematic review, longitudinal associations studies a positive association between higher physical activity (PA) and lower sedentary behavior (SB) and better global cognitive function. Cross-sectional studies showed higher PA and lower SB to be associated with better global cognitive function. The strongest cross-sectional associations were found for moderate-to-vigorous PA (MVPA) and total PA (TPA). Effect sizes of associations were larger for the general population when compared to disease populations, and larger for unadjusted analyses.
Higher PA was associated with better global cognitive function, which is in line with previous systematic reviews investigating similar topics. Previous systematic reviews describing the association between PA and cognitive function, which included self-reported PA found similar results (Carvalho et al., 2014;Falck et al., 2019). Notably, a recent scoping review investigated the association between self-reported and objectively assessed LPA and cognitive function in older adults and suggested LPA to be beneficial (Erlenbach et al., 2021). This is in line with our results, which is an important finding, as functional and physical limitations increase with age, preventing older adults from participating in higher intensity levels of physical activity (Schutzer and Graves, 2004). Systematic reviews investigating the impact/effectiveness of PA interventions (specifically exercise interventions) and cognitive function in older adults, (Falck et al., 2019;Klimova and Dostalova, 2020;Sanders et al., 2019) found that PA has the potential to improve and maintain cognitive function.
Our findings reported the greatest effect sizes for MVPA and TPA, indicating that not only PA at a higher intensity but in fact a greater duration of PA at any intensity might be beneficial for the preservation of global cognitive function in older adults. This is in line with both recommendations that encourage increasing MVPA as well as encouraging the principle that any PA is better than none (World Health Organization, 2019). Not only were associations identified for measures of PA but additionally associations between SB and global cognitive functions were found, indicating that SB could affect global cognitive function. However, it is important to note the possibility of reverse causation, in which disentangling cause and effect is difficult. Possible mechanisms by which SB could affect global cognitive function are suggested to be both cellular and systematic and include hippocampal neurogenesis, modulation of endogenous growth factors as brain-derived neurotrophic factor (BDNF) and inflammation and oxidative stress (Engeroff et al., 2018;Olanrewaju et al., 2020;Voss et al., 2014). Cerebral perfusion is one of the mechanisms suggested to be essential for preserving cognitive function (Leeuwis et al., 2017;Wolters et al., 2016). Higher PA is known to increase cardiac output and positively affecting cerebral perfusion (Jefferson, 2010). On the other hand, a recent study in older adults did not find an association between SB and decreased short-term cerebral perfusion after three hours of sitting (Maasakkers et al., 2020). However, increased blood pressure and cerebrovascular resistance were present, which are two other factors that are known to negatively impact brain health in the long term (Maasakkers et al., 2020). Furthermore, these effects were not mitigated with regular breaks in SB (Maasakkers et al., 2020), suggesting a specific detrimental effect of SB. Other mechanisms explaining the potential benefits and deleterious effects of PA and SB, respectively, on global cognitive function could be found in increases in exercise-induced metabolic factors and muscle-derived myokines, which can stimulate the production of neurotrophins and thereby the promotion of neurogenesis. In addition, anti-inflammatory effects of PA can ameliorate the pathophysiological hallmarks of Alzheimer's Disease, although these mechanistic pathways need further confirmation in human studies (Valenzuela et al., 2020). However, studies combining objective assessments of PA/SB, cognitive function and mechanistic pathways to monitor the specific effect of PA/SB on the human brain are still lacking and need to be encouraged.
Relatively more included articles described non-significant associations between PA/SB and global cognitive function, although consistent effect directions were found. These findings could be explained by the fact that risk factors for cognitive decline are multifactorial, (World Health Organization, 2019) and PA/SB behaviors only represent two of these factors while other influential factors such as management of cardiovascular risk factors may confound associations (Baumgart et al., 2015). Estimated standardized regression coefficients for the association between SB and global cognitive function were found to be relatively small, according to cut-offs of Cohen's guidelines of Pearson's R designated as R = 0.10; 0.30, and 0.50 for small, medium, and large effects, respectively (Cohen, 1992). However, Cohen's guidelines have been found to overestimate effect sizes in the field of gerontology, and rather cut-offs of R = 0.10; 0.20, and 0.30 have been suggested to estimate small, medium, and large effects more accurately (Brydges, 2019). In case these proposed cut-offs are used, negligible (<small) to approximately medium effects were identified in this review.

Clinical implications
With an aging population, there is concern over the number of older adults with cognitive impairments as incidence increases with age (World Health Organization, 2012). The lack of an effective treatment for dementia underpins the urgency to understand the modifiable lifestyle factors influencing cognitive function. Subsequently, identifying PA and SB as associated with cognitive impairment using objective measures is not only clinically relevant but also provides a foundation for developing actionable targets. Evidence-based guidelines from the World Health Organization recommend physical activity interventions to people with normal cognition to reduce their risk of cognitive decline (World Health Organization, 2019). However, clear recommendations with regard to dose and duration of PA are missing, as evidence is based upon self-reported measures of PA or specific exercise interventions, whereas favorable specific daily PA patterns could be targeted through lifestyle interventions. Future research should also focus on specific (resistance) exercise training and its effect on global cognitive function, rather than solely the total duration of PA/SB behaviors.
A crucial issue that needs to be resolved to inform PA/SB guidelines is the standardization of assessment methods of PA/SB. Most commonly used devices are considered valid to assess PA/SB in older adults (Cavanaugh et al., 2007;Clarke et al., 2017;Dijkstra et al., 2010;Heesch et al., 2018;Murphy, 2009;Taraldsen et al., 2011;Valkenet and Veenhof, 2019). However, other factors are known to affect PA/SB Fig. 4. Effect sizes of physical activity and sedentary behavior with global cognitive function derived from albatross plots, expressed as standardized regression coefficients (βs). assessment (Murphy, 2009]. Specifically, cut-off points, location of instrument to measure PA/SB, and the comparison of inertial sensors with different data processing methods need further attention since large variations between sensors and processing algorithms are present, which can result in different conclusions (Gorman et al., 2014). Standardization of assessment methods will guide future research to unravel dose-response relationships between PA/SB and global cognitive function, informing healthcare professionals on how to prevent, delay, and partly counteract declines in cognitive function.

Strengths and limitations
A strength of this study is the use of objective measurement methods to assess PA and SB, as subjective measures tend to overlook unstructured PA and LPA (Amagasa et al., 2017;Manns et al., 2012). In addition, older adults are more likely to overestimate PA and underestimate SB if self-reported (Dyrstad et al., 2014;Van Cauwenberg et al., 2014). The inclusion of populations of individuals with specific disease characteristics is also a strength, as comorbidity is highly prevalent in older adults (Barnett et al., 2012), thereby increasing our study's generalizability. It is important to note that although standardized effect estimates in albatross plots were presented, we were unable to perform a meta-analysis due to heterogeneity in assessment methods at the article level, which would have provided a more precise pooled estimate of associations. Another limitation is the inclusion of relatively few longitudinal studies. With the inclusion of cross-sectional studies, the probability of reversed causation is high and future studies should focus on longitudinal study designs to further unravel the causal direction between PA/SB and global cognitive function. In addition, the specific detrimental effects of SB behavior alone, even in individuals who are physically active, need to be studied. Another essential factor that could not be taken into account was the level of PA during the night, which can often be present in cognitively impaired older adults and does not represent favorable PA behavior. In addition, information on sleep duration and quality were not available in the studies included.

Conclusions
Objectively assessed higher PA and lower SB are associated with better global cognitive function, with MVPA and TPA providing the strongest evidence for beneficial effects. Future research, specifically larger longitudinal studies taking standardization of PA/SB measures into account, must be conducted to unravel dose-response relationships and the direction of causality to inform PA/SB related guidelines and establish evidence-based targets for older adults.

Financial disclosure
No financial disclosures were reported by the authors of this paper.

Funding
This work was supported the European Union's Horizon 2020 research and innovation programmes under the Marie Curie Skoldowska-Curie grant for the PANINI project (675003); and the European Union's Horizon 2020 research and innovation programme PreventIT (689238).

Declaration of Competing Interest
The authors report no declarations of interest.