Does APOE ɛ4 Status Change How 24-Hour Time-Use Composition Is Associated with Cognitive Function? An Exploratory Analysis Among Middle-to-Older Adults

Abstract

Background:

The 24 h time-use composition of physical activity, sedentary behavior, and sleep is linked to cognitive function in adults and may contribute to future dementia risk. However, the impact of reallocating time between behaviors may differ depending on an individual’s genetic dementia risk.

Objective:

To explore if there is an interaction between 24 h time-use composition and genetic dementia risk in relation to cognitive function, and to simulate how time-reallocations are associated with cognitive function across different levels of genetic dementia risk.

Methods:

Cross-sectional global cognition, executive function, genetic dementia risk (at least one apolipoprotein (APOE) ɛ4 allele versus none) and 7 days of 24 h accelerometry (average daily time-use composition of moderate-to-vigorous physical activity (MVPA), light physical activity, sedentary behavior, sleep) were collected from 82 adults (65.6±7.5 years, 49 females). Linear regression was used to explore the relationship between time-use composition and cognitive measures, testing for interaction between APOE ɛ4 status and time-use composition. The models were used to simulate time reallocations in both APOE ɛ4 status groups.

Results:

The 24 h time-use composition was associated with global cognition (F = 2.4, p = 0.02) and executive function (F = 2.6, p = 0.01). For both measures, the association differed according to genetic risk (interactions p < 0.001). In both APOE groups, reallocating time to MVPA was beneficially associated with measures of cognitive function, but associations were larger among those with at least one APOE ɛ4 allele.

Conclusion:

Genetic dementia risk may impact the effectiveness of activity interventions. Increasing MVPA may provide greater benefits among those with higher genetic dementia risk.

Keywords

Aging cognition compositional data analysis physical activity sedentary behavior sleep

INTRODUCTION

Dementia is one of the leading causes of disability and mortality worldwide and has been declared a global health priority by the World Health Organization [1]. About 60% of the risk of developing dementia is unknown or attributed to non-modifiable factors such as age and genetics, with the remaining 40% attributed to potentially modifiable factors [2]. The recent Lancet commission identified insufficient physical activity as an important modifiable risk factor during late life [2]. Physical activity is one part of a 24 h day, which cannot be changed without there being changes to other activities within that day [3]. Thus, this modifiable risk factor has implications for time use across all daily activities, and intervention to reduce this risk should be conceptualized as reallocation of time to physical activity from other activities (sedentary behavior and/or sleep) [4].

Studies considering the role of non-modifiable risk factors, specifically genetic risk, in the relationships between time-use behaviors and cognitive performance have reported mixed findings (see systematic review by de Frutos-Lucas et al. [5]). Yet, it is becoming increasingly acknowledged that a multifaceted approach incorporating both lifestyle and genetic risk may be required for dementia prevention [6]. A recent study of almost 200,000 participants in the UK Biobank identified no interaction between genetic risk (measured as polygenic risk scores) and a healthy lifestyle score (including smoking, physical activity, diet, alcohol) [7]. Nonetheless, they found that lifestyle mitigated genetic risk— among those with high genetic risk for dementia, fewer people with healthy lifestyles developed dementia than those with unhealthy lifestyles (1.13% versus 1.78%).

Similar findings have been reported in other studies conceptualizing genetic risk as a single gene called Apolipoprotein E (APOE), a cholesterol transport protein which exists as three alleles, APOE ɛ2, APOE ɛ3, and APOE ɛ4 [8]. The APOE ɛ4 allele has been identified as the strongest genetic risk factor for late onset Alzheimer’s disease (AD), the most common type of dementia accounting for ∼70% of cases [9]. Previous estimates suggest that in the Caucasian population, those carrying one APOE ɛ4 allele (heterozygotes) have a 3-fold increased risk of developing AD [9]. Dhana et al. [10] found no interaction effects between healthy lifestyle score and APOE ɛ4 status in cognitive decline, such that adherence to a healthy lifestyle was associated with slower cognitive decline in both APOE ɛ4 carriers and noncarriers. Jin et al. [11] reported APOE ɛ4 carriers with unhealthy or healthy lifestyles did not differ in their odds of cognitive impairment, whilst unhealthy APOE ɛ4 noncarriers had significantly higher odds of cognitive decline than healthy noncarriers.

To our knowledge, no studies exploring genetic risk and lifestyle interactions have isolated 24 h time-use behaviors (physical activity, sedentary behaviors, and sleep) from other lifestyle behaviors, but a few have investigated physical activity in isolation. In an early study, Podewils et al. [12] reported that dementia risk differed between physical activity and APOE ɛ4 genotypes, such that any interaction between physical activity and dementia risk was limited to APOE ɛ4 noncarriers. Similar findings were reported in a recent study by Stringa et al. [13] who found no interaction between APOE ɛ4 status and physical activity for cognitive decline. In contrast, Schuit et al. [14] reported that in older men, the risk of cognitive decline among physically inactive APOE ɛ4 carriers was almost four times higher than the risk of physically active APOE ɛ4 carriers. It is not well understood whether the influence of 24 h time-use behaviors on cognitive function and dementia risk are more potent in APOE ɛ4 carriers versus noncarriers. Moreover, it is unclear whether targeted time-use interventions can attenuate or offset genetic risk.

To address this gap in the literature, this exploratory study aimed to investigate whether the relationship between 24 h time-use composition and cognitive function differed by APOE ɛ4 status in a sample of healthy adults.

MATERIALS AND METHODS

Participant recruitment and ethics

91 community-dwelling adults aged between 50 and 80 years from Adelaide, South Australia were recruited from an existing laboratory database, newspaper advertisements and aged care networks between February and October 2016. Through a selective recruitment process, the research group aimed to recruit an equal number of participants aged 50–59 years, 60–69 years, and 70–79 years. Additionally, an equal number of participants with ‘low’ and ‘elevated’ cardiovascular risk were also recruited in each age group in the interest of the broader study (detailed in Wade et al. [15]; Keage et al. [16]). Additional exclusion criteria included previous history of stroke or transient ischemic attack; diagnosis of dementia; blindness or vision difficulties without corrective lenses; diagnosis of intellectual disability; diagnosis of psychiatric disease; and history of unconsciousness for >5 min.

Procedure

Ethics approval was granted by the University of South Australia Human Ethics Committee (0000034635). All participants provided informed consent to take part in the study. Participants attended two 3 h appointments separated by ∼8–10 days. During session one, informed consent, general health, cognition, fasted blood tests (minimum 8 h fast), anthropometric assessments, blood pressure (arterial compliance measurement), and dietary assessments were conducted. EEG data, collected as a part of the larger study were recorded during the second session (see Keage et al. [16]). Participants wore the accelerometer for 7 days between session one and two.

Measures

24 h time-use composition

24 h time-use patterns (time spent in physical activity, sedentary behavior, and sleep) were measured using accelerometry. Participants were asked to wear a triaxial accelerometer (GENEActiv, Activinsights Ltd., UK) on their non-dominant wrist for 7 consecutive days, except during water-based activities (e.g., swimming, showering). Data were recorded at a sampling frequency of 100 Hz.

Raw acceleration data were downloaded using GENEActiv software (version 3.2) and processed using a customized MATLAB graphic user interface developed at the University of South Australia (COBRA; MATLAB 2018B). During processing, the vector magnitude of acceleration corrected for gravity (g) was calculated over 60 s epochs as described previously [17]).

Time spent in sleep was defined using a combination of participant recorded sleep logs and manual cross-checking of individual participant data by visually inspecting the acceleration trace averaged across 24 h. Waking day behaviors were classified as time in moderate-vigorous intensity physical activity (MVPA;>806 gravitational units per min, g-min), light intensity physical activity (LPA; 378–806, g-min) or sedentary behavior (<377 g-min, excluding sleep time) using previously published cut points [18].

Accelerometry data were considered a ‘valid wear day’ if the watch was worn for at least 10 waking hours. Participants were required to have three or more valid weekdays and one valid weekend day to be included in the analyses. Total time spent in each behavior was averaged across the recording period to provide an average daily time-use composition for each participant (time spent in MVPA, LPA, sedentary behavior and sleep in minutes per day).

APOE ɛ4 status

Approximately 23 mL of whole blood was collected from each participant. Samples were collected by trained phlebotomists into 2×9 mL ethylenediaminetetraacetic acid (EDTA) (18 mg) anti-coagulant and 1×4 mL sodium fluoride vacuette tubes (grenier bio-one, Kresműnster, Austria). Blood samples were first centrifuged at 4000 rpm for 10 min at 4°C. The plasma layer was aliquoted into 1.8 mL microfuge tubes. Then, vacuette tubes with the remaining sample were re-centrifuged at 2500 RPM at 4°C for another 10 min, resulting in three distinct layers (clear upper layer of plasma, thin cloudy layer of buffy coat and a large viscous layer of concentrated erythrocytes). The buffy coat layer was removed with a blunt end 1mL pipette tube and samples were initially stored at –20°C for up to 1 week before being transferred to –80°C until analysis.

Deoxyribonucleic acid (DNA) was extracted from the stored buffy coat using a commercial assay kit (Blood Mini DNA Kit, Qiagen, Valencia, CA). Single nucleotide polymorphisms (SNPS) were identified in duplicate for APOE 158 and APOE 112 using quantitative PCR TaqMan^TM gene expression assay (Thermofisher Scientific, USA) on a 96 well plate with the following conditions: initial holding stage at 95°C for 10 min, 40 cycles of 15 s at 95°C followed by 60°C for 1 min, holding stage of 60°C for 1 min. Negative controls were performed in triplicate for each SNP and Plate. Participants were classified as APOE ɛ4 carriers if they were heterozygous or homozygous (i.e., carrying one or both APOE ɛ4 alleles) as per Ghebranious et al. [19]. Participants who did not carry an APOE ɛ4 allele were classified as noncarriers [20].

Cognitive function

Global cognition

Addenbrooke’s Cognitive Examination III (ACE-III) was used to measure global cognition. The ACE-III is a paper-and-pencil style cognitive assessment which assesses function across memory, fluency, attention/orientation, language, and visuospatial domains. Participants can score a maximum of 100, with higher scores indicating higher global cognitive function. The ACE-III has demonstrated high specificity (0.96), sensitivity (1.00) and strong criterion validity and reliability [21] when screening for MCI.

Executive function

The Cambridge Neuropsychological Test Automated Battery (CANTAB) was used to measure executive function. CANTAB is a computerized test battery which contains a range of cognitive assessments which have demonstrated discriminant validity between clinical populations and healthy controls, moderate correlations with traditional neuropsychological tests [22], and sufficient test-retest reliability [23].

An executive function composite score was created as the sample-specific z-score of the arithmetic mean of the Attention Switching Task (AST) Scores (AST percent correct; AST congruency cost; AST switch cost) and the Spatial Working Memory (SWM) Score (SWM strategy). The raw test scores were reversed where necessary, so that a higher score indicated better performance.

Covariates

Age, sex, and education level (classified as having completed primary, secondary, or tertiary education as the highest level of education) were included as covariates in the analyses, as they are associated with cognitive function and future dementia risk [2]. All three measures were obtained using self-report questionnaires.

Statistical analysis

24 h time-use composition

All analyses were performed in R (version 4.1.0), using the compositions [24] package for the compositional data analysis. Daily activity was conceptualized as an average 24 h time-use composition consisting of sleep, sedentary behavior, LPA and MVPA. As a 24 h time-use composition consists of mutually exclusive and exhaustive categories of activities (in this case, sleep, sedentary behavior, LPA and MVPA) that always sum to the same total of 24 h, the parts are perfectly multi-collinear. In other words, it is not possible to increase one of the activities without also reducing other activities collectively by the same amount to compensate. Accordingly, time-use activities are co-dependent and violate the assumption of independence required by linear regression models [4]. Compositional data analysis removes the perfect multi-collinearity issue via a log-ratio transformation of the time-use activities, enabling the complete 24 h time-use composition to be included in the same linear regression model [25]. Following compositional data analysis procedures, the time-use composition was expressed as a set of isometric log-ratio co-ordinates [25]. There were no zeros present in any of the behavior variables.

Relationship between cognitive function and time-use composition by APOE ɛ4 status

Linear regression models with robust MM-type estimators [26] were used to explore the relationship between the time-use composition (expressed as isometric log ratios) and the cognitive measures (Model 1). In Model 2, interaction between time-use composition and APOE ɛ4 status was tested and retained if statistically significant (p < 0.05). Second-order polynomial terms were retained for the time-use composition isometric log ratios if they improved the predictive ability of the models, as determined by comparing the root mean squared errors and mean absolute errors of models with/without the polynomial terms. All models were adjusted for age, sex, and education. The statistical significance of the overall time-use composition (Model 2) and its interaction with APOE ɛ4 status (Model 2) was assessed using Analysis of Deviance Tables with Type II tests.

The regression models were used to estimate cognitive measures for a series of different predictive time-use compositions. Estimates were made separately for APOE ɛ4 carriers and non-carriers. The predictive time-use compositions were selected to represent the change of one ‘dominant’ time-use behavior (e.g., sleep), relative to all the remaining behaviors (e.g., sedentary behavior, LPA and MVPA). Each time-use behavior was iteratively considered to be the dominant behavior. Following published procedures [27, 28], the associations of reallocating up to 30 min from the dominant time-use behavior to the remaining behaviors (and vice versa) relative to the sample mean, were computed and plotted to aid interpretation. As described in Dumuid et al. [28], the absolute change in the dominant behavior was compensated for by multiplicative change in the remaining behaviors (e.g., if sleep was increased by 30 min, the remaining behaviors were all collectively reduced by 30 min by applying the same constant to all the remaining behaviors). Subsequently 95% bootstrap confidence intervals were generated for the estimated differences in cognition, using 1000 resamples.

RESULTS

Of 91 participants, 84 had complete APOE ɛ4 data. Of these, 82 had valid 24 h time-use composition, cognitive function (ACE-III) and covariate data. Fewer (78) participants had complete data for executive function (Table 1). 22 participants (27%) carried at least one APOE ɛ4 allele, and of these, only 1 participant had two APOE ɛ4 alleles. There were no significant differences in any of the participant characteristics presented in Table 1 depending on APOE ɛ4 status (t-tests for continuous variables and Chi-square tests for categorical variables, all p > 0.05).

Table 1

Participant characteristics

Characteristic	APOE ɛ4 status:	APOE ɛ4 status:
Non-carrier (n = 60)	Carrier (n = 22)
Age, y mean (SD)	65.1 (7.7)	67.1 (7.1)
Sex, n (%)
Male	25 (42)	8 (36)
Female	35 (58)	14 (64)
Education, n (%)
Primary	14 (23)	7 (32)
Secondary	12 (20)	2 (14)
Tertiary	34 (57)	12 (54)
Time-Use Composition, min/d arithmetic mean (SD)
Sleep	503 (61)	492 (50)
Sedentary behavior	660 (83)	662 (78)
Light physical activity	202 (49)	220 (50)
Moderate-to-vigorous physical activity	72 (45)	64 (30)
Time-Use Composition, min/d compositional mean
Sleep	510	497
Sedentary behavior	669	667
Light physical activity	200	217
Moderate-to-vigorous physical activity	60	59
ACE-III, mean (SD)	91.0 (5.4)	92.0 (5.6)
Executive function score components: mean (SD) (n = 78)	(n = 57)	(n = 21)
Attention Switching Task: Percent correct	89.1 (12.0)	86.6 (17.9)
Attention Switching Task: Congruency cost	127.0 (97.5)	114.6 (95.3)
Attention Switching Task: Switch cost	–123.1 (123.4)	–108.7 (119.1)
Spatial Working Memory: Strategy score	66.4 (11.9)	69.8 (10.3)

APOE ɛ4, apolipoprotein ɛ4 allele; ACE-III, Addenbrooke’s Cognitive Examination-III. Compositional mean is calculated as the geometric mean of each time-use behavior, linearly adjusted so that together all behaviors sum to 1440 min/d.

Associations of time-use composition with cognitive measures and interaction by APOE ɛ4 status

Type II Analysis of Deviance tests of Model 1 showed that the 24 h time-use composition of sleep, sedentary behavior, LPA and MVPA was associated with global cognition (F = 2.4, p = 0.02) and executive function (F = 2.6, p = 0.01) (Table 2). For both cognitive measures, Model 2 showed there were significant interactions between 24 h time-use composition and APOE ɛ4 status (both interactions p < 0.001) (see Table 2).

Table 2

Analysis of Deviance Table (Type II tests)

	Global cognition z-score (n = 82)		Executive function z-score (n = 78)
	F	p	F	p
Model 1: Time-Use Composition vs Cognition
Time-Use Composition^a	2.4	0.02	2.6	0.01
Sex	0.0	0.96	3.3	0.07
Age	11.4	0.001	5.3	0.02
Education	4.8	0.01	5.0	0.01
Model 2: Model 1 + APOE ɛ4 interaction with Time-Use Composition
APOE ɛ4 status	5.7	0.02	0.2	0.68
Time-Use Composition^a	40.9	<0.001	5.0	<0.001
APOE ɛ4* Time-Use Composition^a	19.6	<0.001	4.2	<0.001
Sex	0.0	0.99	5.3	0.02
Age	11.4	0.001	1.8	0.19
Education	6.3	0.003	7.2	0.002

APOE ɛ4, apolipoprotein ɛ4 allele; Global cognition from Addenbrooke’s Cognitive Examination III, and executive function from the Cambridge Neuropsychological Test Automated Battery. ^aTime-use composition is expressed as a set of isometric log ratios. For all models, polynomial terms are included for the time-use composition isometric log ratios. Bold denotes statistical significance at p < 0.05.

The tests in Table 2 show statistically significant interactions between APOE ɛ4 status and time-use behaviors but they cannot provide any indication of the direction of relationships with cognitive performance or effect sizes. To visualize how the relationship between time-use behaviors and cognitive measures differ by APOE ɛ4 status, model-estimated cognitive response curves were plotted for a series of time-reallocations in APOE ɛ4 carriers and non-carriers (Fig. 1). The time reallocations from the mean composition are shown on the x-axis. A positive reallocation indicates more time in the respective behavior, drawing time from all remaining behaviors pro-rata, to compensate. Thus, +30 minutes in the Sleep column means 30 min have been reallocated to sleep (from the sample mean of 508 min, to 508+30 = 538 min), from the remaining behaviors (sedentary behavior, LPA and MVPA) in equal relative reductions. Figure 1 shows very similar sleep and sedentary behavior response curves for both APOE ɛ4 status groups (both flat, suggesting no relationship between them and cognitive measures for either group). There appear to be APOE ɛ4 status differences in response curves for LPA and MVPA against both cognitive measures. While LPA was negatively associated with cognitive function, and MVPA positively associated with cognitive function, the estimated differences in cognitive measures were larger among APOE ɛ4 carriers versus non-carriers. For example, reallocating 30 min to MVPA from the remaining behaviors (going from 0 to +30 on the x-axis) was associated with about half a standard deviation increase in executive function (y-axis) for those carrying an APOE ɛ4 allele (red curve), versus only about a 0.2 standard deviation increase among those without an APOE ɛ4 allele (green curve). However, the 95% confidence interval bands for the response curves were wide and overlapping, therefore any between-group comparisons are descriptive and should be made with caution.

Fig. 1

Estimated differences (and 95% confidence interval bands) in cognitive measures when time is reallocated to/from the behavior in the header, relative to all remaining behaviors. A zero reallocation (x-axis) represents the sample mean duration of the respective behavior in (min/d): Sleep = 508; SB = 668; LPA = 204; MVPA = 60. Estimates are averaged across age, sex, and education levels. SB, sedentary behavior; LPA, light physical activity; MVPA, moderate-to-vigorous physical activity; ACE, Addenbrooke’s Cognitive Examination III.

DISCUSSION

This exploratory study aimed to investigate the relationship between 24 h time-use composition and cognitive function in a sample of healthy adults, and to examine the role of APOE ɛ4 status. Our findings indicate time-use composition is associated with cognitive function in both APOE ɛ4 carriers and non-carriers. Among both genetic risk groups, model-based estimates showed that time reallocations away from light physical activity (to the remaining behaviors) and to moderate-to-vigorous physical activity (away from the remaining behaviors) were favorably associated. On visual inspection, the dose-response curves were steeper among APOE ɛ4 carriers versus non carriers. Although 95% confidence intervals were wide, estimated differences in executive function associated with reallocations to/from moderate-to-vigorous physical activity were approximately twice as large among APOE ɛ4 carriers versus non carriers. For example, among non-carriers, a 30-min reduction in MVPA was associated with a 0.5 standard deviation reduction in executive function, while among carriers the same reduction in MVPA was associated with a 1 standard deviation reduction in executive function. Our descriptive findings suggest that maintaining or increasing engagement in moderate-to-vigorous physical activity may be more beneficial for cognitive function in those with higher genetic predisposition to cognitive decline or dementia (i.e., APOE ɛ4 carriers). At a population level, a 0.5 standard deviation difference between APOE ɛ4 status groups suggests the role of genetic dementia risk warrants further investigation in future studies.

Our findings appear to contradict several previous studies which reported no modification of the associations between physical activity and cognition in APOE ɛ4 carriers versus noncarriers [5, 7]. There are several important differences between the current study and previous studies which may contribute to the contradictory findings. Firstly, most previous studies have assessed physical activity using subjective questionnaires and have subsequently grouped light and moderate-vigorous intensity activities together. However, there is some evidence to suggest that physical activities of different intensities may vary in their effect on cognitive function in adult populations, such that moderate-to-vigorous physical activity may have more potent effects on cognitive function compared to physical activity of lighter intensities [17, 29]. In the current study, the associations with objectively measured moderate-to-vigorous physical activity appeared to be much larger and in the opposite direction to those of light physical activity. However, it is unclear why the beneficial associations of moderate-to-vigorous physical activity for cognitive function were more evident in APOE ɛ4 carriers compared to APOE ɛ4 non-carriers in the current study. One possible explanation is that APOE ɛ4 carriers in this study had slightly lower moderate-to-vigorous physical activity compared to non-carriers (8 min less if considering arithmetic means), hence they are lower on the dose-response curve where the angle is steeper.

There are a number of strengths to this study. We used objective measures of time use and a compositional data analysis approach which enabled all 24 h time-use behaviors to be considered simultaneously. Previous studies have focused either on just one single time-use behavior or self-reported activity levels. In addition, the present study employed a tiered recruitment strategy among those of high and low cardiovascular disease risk to ensure representation across cardiovascular dementia risk profiles. Several limitations should be acknowledged. First, the small sample size limits the strength of conclusions that can be made and increases susceptibility to outlying observations. This is reflected in the relatively wide confidence intervals around the estimates in Fig. 1. To overcome the potential impact of outlying observations we used models with robust estimators. Second, the cross-sectional nature of the data means that any re-allocation of time analyses are purely associations and causality cannot be implied. Only one-for-remaining time re-allocations were considered, but time can be reallocated in many ways between daily behaviors (one-for-one swaps, or more complex reallocations). Third, only one participant in the current study was identified as a homozygous APOE ɛ4 carrier (ɛ4/ɛ4). Homozygous APOE ɛ4 carriers have a stronger risk of developing dementia compared to heterozygous ɛ4 (ɛ4/ɛ2 or ɛ4/ɛ3) carriers (15-fold versus 3-fold, respectively in a previous study) [9]. A further complication is that the ɛ2 allele has been shown to be protective against cognitive decline [30 –32] and therefore may mitigate the risk of the ɛ4 allele in the heterozygous ɛ4/ɛ2 presentation. In this study, 8 of the 22 APOE ɛ4 carriers (36%) had the heterozygous ɛ4/ɛ2 combination. The role of APOE allele profile could not be investigated due to the restricted sample and should be further investigated in future studies. Fourth, while the APOE ɛ4 allele constitutes the largest genetic risk for dementia in most polygenic risk scores, future studies could consider taking either an a-priori pathway association risk factor approach or a polygenetic risk score approach [33]. Fifth, because we excluded any participants with a diagnosis of dementia, our sample missed those who may have been most affected by having an APOE ɛ4 allele (i.e., survival bias). Finally, establishing greater diversity in cohorts of dementia risk is a key focus of dementia researchers worldwide. This small study included predominantly Caucasian participants with high levels of education which limits the conclusions that can be made for more genetically and culturally diverse populations.

Conclusion

This study provides preliminary descriptive evidence that the relationship between 24 h time-use composition and cognitive function is magnified by APOE ɛ4 status and highlights the importance of considering all time-use behaviors together to understand the role of lifestyle behaviors for cognitive function in adults with greater genetic risk of dementia. It is recommended that similar research is conducted in larger samples with greater spread of allele profiles (greater proportion of homozygous ɛ4 carriers).

Footnotes

ACKNOWLEDGMENTS

DD is supported by the Australian National Health and Medical Research Council (NHMRC) Early Career Fellowship APP1162166. MLM is supported by a Dementia Australia Research Foundation PhD scholarship.

Authors’ disclosures available online ().

References

World Health Organization, Global action plan on the public health response to dementia 2017-2025, https://www.who.int/publications/i/item/global-actionplan-actionplan-on-the-public-health-response-to-dementia-2017—2025, Accessed 10 February 2022.

Livingston

, Huntley

, Sommerlad

, Ames

, Ballard

, Banerjee

, Brayne

, Burns

, Cohen-Mansfield

, Cooper

, Costafreda

, Dias

, Fox

, Gitlin

, Howard

, Kales

, Kivimäki

, Larson

, Ogunniyi

, Orgeta

, Ritchie

, Rockwood

, Sampson

, Samus

, Schneider

, Selbæk

, Teri

, Mukadam

(2020) Dementiaprevention, intervention, and care: 2020 report of the LancetCommission. Lancet 396, 413–446.

Dumuid

, Pedisić

, Palarea-Albaladejo

, Martín-Fernández

, Hron

, Olds

(2020) Compositionaldata analysis in time-use epidemiology: What, why, how. Int JEnviron 17, 2220.

Pedisić

, Dumuid

, Olds

(2017) Integrating sleep,sedentary behaviour, and physical activity research in the emergingfield of time-use epidemiology: Definitions, concepts, statisticalmethods, theoretical framework, and future directions. Kinesiology 49, 18.

de Frutos-Lucas

, Frost

, Erickson

, Serrano

, Maestu

, Laws

, Brown

(2020) Does APOE genotype moderate the relationship between physical activity, brain health and dementia risk? A systematic review. Ageing Res Rev 64, 101173.

Licher

, Ahmad

, Karamujić-Comić

, Voortman

, Leening

, Ikram

(2019) Genetic predisposition,modifiable-risk-factor profile and long-term dementia risk in thegeneral population. Nat Med 25, 1364–1369.

Lourida

, Hannon

, Littlejohns

, Langa

, Hyppönen

, Kuźma

, Llewellyn

(2019) Association of lifestyle andgenetic risk with incidence of dementia. JAMA 322, 430–437.

Crean

, Ward

, Mercaldi

, Collins

, Cook

, Baker

, Arrighi

(2011) Apolipoprotein E ɛ4 prevalence in Alzheimer’s disease patients varies across global populations: A systematic literature review and meta-analysis. Dement Geriatr Cogn Disord 31, 20–30.

Gharbi-Meliani

, Dugravot

, Sabia

, Regy

, Fayosse

, Schnitzler

, Kivimäki

, Singh-Manoux

, Dumurgier

(2021) The association of APOE ɛ4 with cognitive function over the adult life course and incidence of dementia: 20 years follow-up of the Whitehall II study. Alzheimers Res Ther 13, 1–11.

10.

Dhana

, Aggarwal

, Rajan

, Barnes

, Evans

, Morris

(2020) APOE4, lifestyle factors, and cognitive decline: A population-based cohort study: Prevention: ADRD prevention research: Who is at risk and who benefits more? Alzheimers Dement 16, e043228.

11.

Jin

, He

, Zhang

, Gong

, Niu

, Ji

, Li

, Zeng

, Yan

(2021) Association of APOE ɛ4 genotype and lifestyle with cognitive function among Chinese adults aged 80 years and older: A cross-sectional study. PLoS Med 18, e1003597.

12.

Podewils

, Guallar

, Kuller

, Fried

, Lopez

, Carlson

, Lyketsos

(2005) Physical activity, APOE genotype, and dementia risk: Findings from the Cardiovascular Health Cognition Study. Am J Epidemiol 161, 639–651.

13.

Stringa

, van Schoor

, Milaneschi

, Ikram

, Del Panta

, Koolhaas

, Voortman

, Bandinelli

, Wolters

, Huisman

(2020) Physical activity as moderator of the association between APOE and cognitive decline in older adults: Results from Three Longitudinal Cohort Studies. J Gerontol A Biol Sci Med Sci 75, 1880–1886.

14.

Schuit

, Feskens

EJM

, Launer

, Kromhout

(2001) Physicalactivity and cognitive decline, the role of the apolipoprotein e4allele. Med Sci Sports Exerc 33, 772–777.

15.

Wade

, Tregoweth

, Greaves

, Olds

, Buckley

, Keage

, Coates

, Smith

(2020) Long-chain omega-3 fatty acid intake is associated with age but not cognitive performance in an older australian sample. J Nutr Health Aging 24, 857–864.

16.

Keage

, Feuerriegel

, Greaves

, Tregoweth

, Coussens

, Smith

(2020) Increasing objective cardiometabolic burden associated with attenuations in the P3b event-related potential component in older adults. Front Neurol 11, 643.

17.

Smith

, Dumuid

, Goldsworthy

, Graetz

, Hodyl

, Thornton

, Ridding

(2021) Daily activities are associated with non-invasive measures of neuroplasticity in older adults. Clin Neurophysiol 132, 984–992.

18.

Eslinger

, Rowlands

, Hurst

, Catt

, Murray

, Eston

(2011) Validation of the GENEA accelerometer. Med Sci Sports Exerc 43, 1085–1093.

19.

Ghebranious

, Ivacic

, Mallum

, Dokken

(2005) Detection of ApoE E2, E3 and E4 alleles using MALDI-TOF mass spectrometry and the homogeneous mass-extend technology. Nucleic Acids Res 33, e149–e149.

20.

Lim

, Villemagne

, Pietrzak

, Ames

, Ellis

, Harrington

, Snyder

, Martins

, Masters

, Rowe

(2015) APOE ɛ4 moderates amyloid-related memory decline in preclinical Alzheimer’s disease. Neurobiol Aging 36, 1239–1244.

21.

Hsieh

, Schubert

, Hoon

, Mioshi

, Hodges

(2013) Validation of the Addenbrooke’s Cognitive Examination III in frontotemporal dementia and Alzheimer’s disease. Dement Geriatr Cogn Disord 36, 242–250.

22.

Smith

, Need

, Cirulli

, Chiba-Falek

, Attix

(2013) A comparison of the Cambridge Automated Neuropsychological Test Battery (CANTAB) with “traditional” neuropsychological testing instruments. J Clin Exp Neurophysiol 35, 319–328.

23.

Goncalves

, Pinho

, Simões

(2016) Test–retest reliability analysis of the Cambridge Neuropsychological Automated Tests for the assessment of dementia in older people living in retirement homes. Appl Neuropsychol 23, 251–263.

24.

Van den Boogaart

, Tolosana-Delgado

(2008) “compositions”: A unified R package to analyze compositional data. Comput Geosci 34, 320–338.

25.

Chastin

, Palarea-Albaladejo

, Dontje

, Skelton

(2015) Combined effects of time spent in physical activity, sedentary behaviors and sleep on obesity and cardio-metabolic health markers: A novel compositional data analysis approach. PLoS One 10, e0139984.

26.

Yohai

(1987) High breakdown-point and high efficiency robust estimates for regression. Ann Stat, 15, 642–656.

27.

Dumuid

, Stanford

, Martin-Fernández

J-A

, Pedisić

, Maher

, Lewis

, Hron

, Katzmarzyk

, Chaput

J-P

, Fogelholm

(2018) Compositional data analysis for physicalactivity, sedentary time and sleep research. Stat Methods MedRes 27, 3726–3738.

28.

Dumuid

, Pedisić

, Stanford

, Martín-Fernández

J-A

, Hron

, Maher

, Lewis

, Olds

(2019) The compositional isotemporal substitution model: A methodfor estimating changes in a health outcome for reallocation of timebetween sleep, physical activity and sedentary behaviour. StatMethods Med Res 28, 846–857.

29.

Fanning

, Porter

, Awick

, Ehlers

, Roberts

, Cooke

, Burzynska

, Voss

, Kramer

, McAuley

(2017) Replacing sedentary time with sleep, light, or moderate-to-vigorous physical activity: Effects on self-regulation and executive functioning. J Behav Med 40, 332–342.

30.

Conejero-Goldberg

, Gomar

, Bobes-Bascaran

, Hyde

, Kleinman

, Herman

, Chen

, Davies

, Goldberg

(2014) APOE2 enhances neuroprotection against Alzheimer’s disease through multiple molecular mechanisms. Mol Psychiatry 19, 1243–1250.

31.

Corder

, Saunders

, Risch

, Strittmatter

, Schmechel

, Gaskell

PC Jr

, Rimmler

, Locke

, Conneally

, Schmader

, Small

, Roses

, Haines

, Pericak-Vance

(1994) Protective effect of apolipoprotein E type 2 allele for late onset Alzheimer disease. Nat Genet 7, 180–184.

32.

Bertram

, McQueen

, Mullin

, Blacker

, Tanzi

(2007) Systematic meta-analyses of Alzheimer disease genetic association studies: The AlzGene database. Nat Genet 39, 17–23.

33.

Lewis

, Vassos

(2020) Polygenic risk scores: From research tools to clinical instruments. Genome Med 12, 44.