Critical periods for cognitive reserve building activities for late life global cognition and cognitive decline: the Sydney memory and aging cohort study

ABSTRACT

Cognitive, social, and physical activities, collectively linked to cognitive reserve, are associated with better late-life cognitive outcomes. To better understand the building of cognitive reserve, we investigated which of these activities, during which stages of life, had the strongest associations with late-life cognitive performance. From the Sydney Memory and Aging Study, 546 older Australians, who were community-dwelling and without a dementia diagnosis at recruitment (M_age 80.13 years, 52.2% female), were asked about their engagement in social, physical, and cognitive activities throughout young adulthood (YA), midlife (ML), and late-life (LL). Comprehensive neuropsychological testing administered biennially over 6 years measured baseline global cognition and cognitive decline. In our study, YA, but not ML nor LL, cognitive activity was significantly associated with late-life global cognition (β = 0.315, p < .001). A follow-up analysis pointed to the formal education component of the YA cognitive activity measure, rather than YA cognitive leisure activities, as a significant predictor of better late-life global cognition (β = 0.146, p = .003). YA social activity and LL cognitive activity were significantly associated with less cognitive decline (β = 0.023, p < .001, and β = 0.016, p = .022, respectively). Physical activity was not found to be associated with global cognition or cognitive decline. Overall, YA cognitive activity was associated with better late-life cognition, and YA social and LL cognitive activities were associated with less cognitive decline. Formal education emerges as the key contributor in the association between YA cognitive activity and late-life global cognition.

KEYWORDS:

• Social interaction
• physical fitness
• cognitive activity
• cognitive decline
• cognitive reserve

Introduction

As the population ages globally (United Nations, 2019) and life expectancy increases (World Health Organization [WHO], 2013), maintenance of cognitive function in old age is ever important. Dementia, a disorder of cognitive decline, is estimated to affect 5–8% of the global population aged over 60 (WHO, 2020). Furthermore, older people report being concerned about cognitive deterioration and dementia more than any other condition (L. A. Anderson et al., 2009). The economic burden of dementia globally exceeds US$1 trillion (Prince et al., 2015), and in Australia, dementia is the greatest cause of disability in older persons aged 65 and over (Australian Institute of Health [AIH], 2012).

Cognitive reserve refers to the ability of the brain to adapt, compensate, and maintain functional resilience to neuropathology without manifestation of cognitive symptoms (Fratiglioni et al., 2007). With a lack of disease-modifying treatments for neurodegeneration (O’brien et al., 2017), research has turned to prevention and symptom onset delay through modifiable risk factor reduction (Norton et al., 2014). Lifespan cognitive, social, and physical activities are the main cognitive-reserve-building activities that have been identified and investigated (Arenaza-Urquijo et al., 2015; Clare et al., 2017), with strong support for their protective benefits on late life cognitive functioning (Baumgart et al., 2015; Kuiper et al., 2016; Xu et al., 2017).

To better translate these findings into clinically useful recommendations and health policy, it is helpful to understand if there are critical windows of effect, or periods over the lifespan, where promotion of specific cognitive reserve-building activities provides the greatest benefit to late life cognition. Justification for this is provided by evidence of critical lifetime periods for other modifiable lifestyle factors that are implicated in cognitive decline. For example, reducing hypertension and obesity in midlife is strongly associated with better cognitive outcomes, including better cognitive performance and decreased risk of dementia in late life (Livingston et al., 2020).

Importantly, few studies have measured levels of cognitive, social, and physical activity in early, mid, and late life, and even fewer have compared the relative impacts of critical life activities, undertaken within critical life stages, on late life cognition and rates of cognitive decline. For cognitive activity, early life has been consistently identified as the most critical life stage to build cognitive reserve (Dekhtyar et al., 2015; Jefferson et al., 2011), though some studies have also identified midlife (Reed et al., 2011) as a critical life stage. However, these studies often do not encompass participants’ entire lifespans, or when they do, they often operationalize cognitive activity based on a single question at each life stage. For physical activity, very few studies have examined the comparative effect of different levels of engagement at different life stages. Of those that have, midlife physical activity has been linked to better late life cognition (Middleton et al., 2010). However, these studies rarely include participants’ entire lifespans. For the few studies investigating social activity, more engagement in midlife is shown to be most associated with better late life cognition (Sommerlad et al., 2019).

This area of research is also important to guide health policy and clinical recommendation guidelines. A better understanding of the optimal type and timing of cognitive reserve-building activities can help maximize return on investment for improving cognitive reserve and late life cognitive performance in real-world settings.

The present study aimed to obtain measures of cognitive, physical, and social activities across the entire spectrum of early, mid, and late life, and to investigate their associations with global cognition and cognitive decline over 6 years, while also controlling for important confounding factors.

Materials and methods

Participants

The Sydney Memory and Aging Study (MAS) is a prospective cohort study of 1037 community-dwelling older Australians, aged 70–90 years and without dementia at baseline, who were randomly recruited from the electoral roll of two electorates in Sydney between 2005–07 (baseline). Exclusion criteria at baseline were insufficient English fluency to complete a comprehensive assessment, a diagnosis of dementia or Mini-Mental State Examination (MMSE) (Folstein et al., 1975) score of <24 (after adjustment for age, education, and non-English-speaking background), and specified medical and psychiatric conditions. More detailed methodology and baseline demographic information are available (Sachdev et al., 2010).

Participants in MAS were assessed every 2 years (called a Wave) by trained research assistants at the University of New South Wales (UNSW) or in participants’ homes. At each Wave, participants provided demographic and lifestyle data, including age, sex, and education, and completed a comprehensive neuropsychological assessment. Participants also completed a brief medical exam, and questionnaires about mood and personality, biennially, until 2020 (Wave 7).

For the present study, the sample comprised 546 MAS participants who completed the optional Lifetime Experiences Questionnaire (LEQ) (Valenzuela & Sachdev, 2007) out of the 889 participants in Wave 2 (2-year follow-up), the only year that it was administered. Demographic comparison between LEQ participants and the larger Wave 2 sample was performed.

The number of participants retained at each follow-up Wave in the study is presented in Figure 1.

Figure 1. Study flowchart showing n participants who completed LEQ at wave 2 and n of included LEQ participants who withdrew, were not assessed, or were deceased at each wave.

Participants gave written consent to participate, and the MAS study was approved by the Human Ethics Committee of the University of New South Wales (HC 05037, 09382, 14327).

Measures

The Lifetime Experiences Questionnaire (LEQ)

The Lifetime Experience Questionnaire (LEQ) is a validated tool for retrospectively assessing activities that build cognitive reserve over the lifespan (Valenzuela & Sachdev, 2007). The LEQ assesses individual engagement in cognitive, social, and physical activities across three life stages: young adulthood (YA; ages 13 = 29), midlife (ML; age 30–64), and late life (LL; age 65 and over). The LEQ is comprised of multiple-choice questions that ask participants about their level of engagement in various types of cognitive reserve-building activities across the different life stages. For each life stage, participants completed questions that were specific to activities typically undertaken at that age. For example, formal education in YA or occupation in ML. Participants also completed questions that were common across the life stages, like the amount of time spent reading or artistic pastimes, which were asked in YA, ML, and LL. (See Supplementary Table S1 for a list of all questions in the LEQ and their scoring method.)

The original LEQ scoring method did not code questions by the type of cognitive reserve activity (i.e., cognitive, social, or physical). For the purposes of this study, each LEQ question was assigned to one of those three activity categories by the authors, as outlined below.

Social activity

Social activity engagement was captured by the same question: “How often would you see a member of your family or friend during this time?.” Social activity engagement was scored on a 5-point Likert scale, with scores ranging from Never (0), Less Than Monthly (1), Monthly (2), Fortnightly (3), Weekly (4), to Daily (5).

Physical activity

Physical activity was captured by three questions: “How often would you take part in sports or activities that were (1) mild, (2) moderate, and (3) vigorous in intensity?.” Each question was scored on a 5-point Likert scale, with scores ranging from Never (0), Less Than Monthly (1), Monthly (2), Fortnightly (3), Weekly (4), to Daily (5).

In the original LEQ scoring criteria, scores from each question were summed. For the purposes of this study, metabolic equivalent (MET) values as per the International Physical Activity Questionnaire (IPAQ) guidelines (Craig et al., 2003) were used to provide comparative weighting based on intensity. A weight of 3.3, 4, and 8 for mild, moderate, and vigorous activity, respectively, was awarded. The frequency of participation was converted to a fraction relative to weekly participation (Daily = 7, Weekly = 1, Fortnightly = 0.5, Monthly = 0.23, Less Than Monthly = 0.12, Never = 0). Intensity scoring was multiplied by frequency and the three questions summed together to form a physical activity measure, with a possible score ranging from 0 to 107.1

Cognitive activity

The remaining questions in the LEQ were classified as cognitive activity. General cognitive leisure activity questions that were asked at all life-stage sections captured five leisure activities: (1) reading, (2) artistic pastimes, (3) travel, (4) playing a musical instrument and (5) practicing a second language. Frequency was graded on a Likert scale ranging from Never (0), Less Than Monthly (1), Monthly (2), Fortnightly (3), Weekly (4), to Daily (5).

Other cognitive activity questions in the LEQ were specific to a particular life-stage section. In YA, these questions were about years of secondary schooling and the number of full-time equivalent years of tertiary education studied. In ML, these questions asked about occupation every 5 years, which was scored according to the Australian Standard Classifications of Occupations (ASCO) Guide, managerial experience, and additional tertiary education (Valenzuela & Sachdev, 2007). In LL, these questions asked about daily activities, information sources, reading habits, community groups, entertainment events, and additional tertiary education. (See Supplementary Table S1 for a complete list of questions and scoring formulas. Scoring formulae for these questions were retained from the original and validated LEQ methods.)

Cognition and cognitive decline

Cognitive function was assessed using a comprehensive neuropsychological test battery comprised 10 tests that measured five major cognitive domains (attention/processing speed, language, executive function, visuospatial, and memory). The full table of tests and the cognitive domains to which they were assigned are presented in Supplementary Table 2.

Raw test scores were first converted to z-scores using the means and standard deviations (SDs) of a reference group comprised 723 MAS participants classified as cognitively healthy at baseline (i.e., native English speakers with a MMSE score of 24 or above, no evidence of dementia or current depression, no history of delusions or hallucinations, and no major neurological disease, or significant head injuries). Composite domain scores were formed by averaging the z-scores of the component tests, which were then standardized as z-scores using their means and SDs in the baseline reference group. Global cognition scores were calculated by averaging the standardized domain scores, and the global composite was then standardized as z-scores in a similar manner, using its mean and SDs in the baseline reference group.

As the LEQ was administered at Wave 2 (2-year follow-up), the global cognition scores at this wave were used as our cross-sectional outcome variables. For cognitive decline, global cognition data from Waves 1 to 4 (6 years of follow-up) were used, as Wave 4 was the last wave with available neuropsychological data and examining change from Wave 1 provided a longer follow-up period to examine decline.

Covariates

Covariates assessed for inclusion in the analyses were age, sex, years of education, English-speaking background, marital status, apolipoprotein E4 (APOE4) status, cardiovascular risk score measured according to the Framingham index (K. M. Anderson et al., 1991), depressive symptoms as measured by the Geriatric Depression Scale (GDS-15, short form) (Yesavage et al., 1982) and frailty status captured by a composite Fried Frailty score (Fried et al., 2001).

Statistical analysis

All cognitive, social, and physical activity variables were standardized by converting raw scores to z-scores to facilitate comparison of effects between predictors.

Linear regression analyses were used to investigate cross-sectional associations between cognitive reserve activity type (cognitive, social, physical) at each life stage (YA, ML, LL) and Wave 2 global cognition scores. Linear mixed models (LMM), with random intercepts for participants, and random slopes for time-in-study, were used to investigate longitudinal associations between the same variables and rate of change in global cognition scores over 6 years. A significant time-in-study × predictor interaction indicates a significant association between the predictor and linear changes in global cognition. For linear mixed models, time-in-study was first calculated as year(s) since the participant’s Wave 1 assessment then centered at the average follow-up time (i.e., 2.85 years since Wave 1) to reduce collinearity between this variable and the interaction terms. LMM is recommended for longitudinal analysis to reduce bias due to nonrandom attrition, compared with other traditional methods, which only use cases with complete data (Fitzmaurice et al., 2012). Additional post-hoc analyses were run for LMMs examining the associations between the LL predictors and changes in global cognition z-scores from Wave 2 (when the LEQ was administered) to Wave 4 (4-year follow-up), given that the LL variables might change in the period between Wave 1 and Wave 2.

For both the linear regressions and LMMs, partially and fully adjusted analyses were carried out for each of the three cognitive, social, and physical activity scores (totaling to 6 linear regression and 6 LMM models).

All partially adjusted models controlled for age, sex, and education, except for YA/Cog models, as years of formal education made up a large proportion of this LEQ score. In the fully adjusted models, other potential covariates were included in the models if they were associated with the outcome variable at p < 0.10 (see Supplementary Table S3) during initial exploratory analyses, in addition to the control variables from partially adjusted models.

A follow-up analysis was also performed to investigate the two individual components of the YA cognitive activity variable. That is, we examined YA years of formal education (primary, secondary, and tertiary schooling summed), and cognitive leisure activities (defined earlier, summed score ranging from 0 to 35) separately as predictors for late life global cognition. This allowed further examination of the predictive utility of formal education versus leisure activities that are cognitively stimulating, undertaken in YA, as per their associations with global cognition.

Statistical significance was set at p < .05. Analyses were conducted using IBM SPSS Statistics 27 (Field, 2013).

Results

Participant and variables description

Participants who completed the LEQ at Wave 2 (n = 545) did not significantly differ from those who did not complete the LEQ at Wave 2 (n = 344) in terms of sex, APOE4 status, cardiovascular risk scores, and frailty scores (see Table 1). However, participants who completed the LEQ were on average, younger, more educated, more likely to be from an English-speaking background, more likely to have been married, were less depressed, had higher MMSE scores and global cognition scores, and were less likely to have MCI.

Table 1. Wave 2 demographics and characteristics of MAS participants who completed the LEQ compared to participants who did not complete the LEQ.

Demographics (Wave 2)	Non-LEQ participants (N = 344)	LEQ Participants (N = 545)	P value
Sociodemographic Factors
Age M (SD)	81.14 (4.81)	80.13 (4.66)	.002*
Median (IQR)	81.23 (77.01–84.66)	79.65 (76.30–83.34)
Sex (Female) %	57.1% (n = 196)	52.2% (n = 285)	.167
Education M (SD)	11.22 (3.34)	11.98 (3.55)	.002*
Median (IQR)	11 (9–13)	11 (9–15)
Non-English-Speaking Background (NESB) %	19.5% n = 67	14.1% n = 77	.039*
Marital Status (Never Married) %	14.3% n = 49	11.4% n = 62	.012*
Genetic Factors
Apolipoprotein E4 Status (E4 Carrier) %	20.4% n = 68	23.9% n = 129	.244
Health Factors
Framingham Cardiovascular Risk Score^c M (SD)	3.95 (3.27)	3.92 (3.11)	.885
Median (IQR)	4 (2–6)	4 (2–6)
Depression (GDS Score) M (SD)	2.70 (2.52)	2.21 (2.10)	.002*
Median (IQR)	2 (1–3.16)	3 (1–3)
Frailty Status^c (Robust) %	31.8% n = 85	36% n = 176	.411
Cognition (at Wave 1)
Cognitive Status Normal %, MCI %, Dementia %	52.1% (n = 174) 47.9% (n = 160) 0.0% (n = 0)	59.0% (n = 311) 41.0% (n = 216) 0.0% (n = 0)	.002
(Adjusted)^ߙ MMSE Score [M (SD)]	28.72 (1.33)	28.77 (1.32)	.798
Median (IQR)	29 (28–30)	29 (28–30)
Global Cognition (z-score) [M (SD)]	−0.11 (1.00)	0.19 (0.92)	.245
Median (IQR)	−0.06 (−0.77–0.61)	0.22 (−0.45–0.88)
Cognition (at Wave 2)
Cognitive Status Normal %, MCI %, Dementia %	45.5% (n = 138) 48.8% (n = 148) 5.6% (n = 17)	59.0% (n = 304) 39.6% (n = 204) 1.4% (n = 7)	<.001*
(Adjusted)^ߙ MMSE Score [M (SD)]	28.52 (2.08)	28.89 (1.39)	.001*
Median (IQR)	29 (28–30)	29 (28–30)
Global Cognition (z-score) [M (SD)]	−0.60 (1.37)	−0.10 (1.07)	.037
Median (IQR)	−0.16 (−0.98–0.51)	0.16 (−0.48–0.82)
Cognition (at Wave 3)
Cognitive Status Normal %, MCI %, Dementia %	49.2% (n = 93) 34.4% (n = 65) 16.4% (n = 31)	70.4% (n = 323) 25.3% (n = 116) 4.4% (n = 20)	<.001*
(Adjusted)^ߙ MMSE Score [M (SD)]	28.31 (2.61)	29.14 (1.26)	<.001*
Median (IQR)	29 (28–30)	30 (28–30)
Global Cognition (z-score) [M (SD)]	−0.28 (1.06)	0.12 (−0.92)	.021
Median (IQR)	−0.27 (−0.98–0.51)	0.24 (−0.39–0.74)
Cognition (at Wave 4)
Cognitive Status Normal %, MCI %, Dementia %	39.1% (n = 81) 34.8% (n = 72) 21.7% (n = 45)	55.8% (n = 273) 35.6% (n = 174) 7.4% (n = 36)	<.001*
(Adjusted) ^c MMSE Score [M (SD)]	27.93 (3.30)	28.80 (1.91)	<.001*
Median (IQR)	29 (28–30)	29 (28–30)
Global Cognition (z-score) [M (SD)]	−1.00 (1.46)	−0.41 (1.31)	.014
Median (IQR)	−0.34 (−0.87–0.42)	0.24 (−0.39–0.74)

Note MCI = Mild Cognitive Impairment.

^aFramingham Cardiovascular Risk Score is comprised the following information: sex, smoking status, diabetic status, systolic blood pressure, total cholesterol level, high-density lipoprotein level, and current antihypertensive medications (See Supplementary Table S3 for a comprehensive explanation of the Framingham Index Score in this study.).

^bFrailty Status was determined using the FRIED framework which combined the following information: body mass index, timed sit to stand, self-reported exhaustion, time to complete a 6 m walk and self-reported physical activity. Participants were categorized as Robust, Pre-Frail, and Frail in order of increasing frailty (See Supplementary Table S3 for a comprehensive explanation of Frailty Status in this study.).

^cAdjusted as per (Anderson et al, 2007) allowing a maximum of 4 additional points.

*p-value < .005.

We did not observe any substantial skewness in the cognitive variables (absolute value of skewness less than one). However, some of the LEQ predictors were skewed. Specifically, physical activity scores were positively skewed (most scores were in the lower range), whereas social activity scores were negatively skewed (most participants scored high).

Covariates that met the inclusion criterion (p < .10) and were thus retained in cross-sectional models of global cognition were English-speaking background, depression scores, and frailty status. Covariates retained in longitudinal models of cognitive decline were APOE4 status and frailty status. These data are presented in Supplementary Table 3.

Cross-sectional association between activity type, life stage, and global cognition

As seen in Table 2, engaging in more cognitive activities in YA was significantly associated with better global cognition scores at Wave 2 in both partially and fully adjusted models (β = 0.279, p < .001 and β = 0.315, p < .001, respectively; Model 1). Greater cognitive activity in ML and LL was not significantly associated with better global cognition in partially or fully adjusted models. Social activity in YA was significantly associated with global cognition in the partially adjusted model but was nonsignificant in the fully adjusted model (Model 2). Social activity in ML and LL was not significantly associated with better global cognition in partially or fully adjusted models. Physical activity was not significantly associated with global cognition at any life stage in both partially and fully adjusted models (Model 3).

Table 2. Linear regression models examining the associations between (1) cognitive activity, (2) social activity, and (3) physical activity at three life stages, and late life global cognition (z-scores) at Wave 2.

Cognitive Activity
	Partially Adjusted^b (Model 1a)					Fully Adjusted^b (Model 1b)
	Beta		CI		P-value	Beta			CI		P-value
Young Adulthood	0.279		0.151–0.332		<.001*	0.315			0.187–0.361		<.001*
Mid Life	0.019		−0.077–0.110		.734	0.004			−0.085–0.092		.935
Late Life	0.026		−0.064–0.110		.539	0.061			−0.030–0.137		.210
Social Activity
	Partially Adjusted ^a (Model 2a)					Fully Adjusted ^b (Model 2b)
	Beta	CI		P-value			Beta	CI		P-value
Young Adulthood	0.087	0.005–0.157		.036*			0.046	−0.033–0.121		.266
Mid Life	0.038	−0.046–0.118		.392			0.056	−0.030–0.135		.213
Late Life	−0.022	−0.101–0.060		.616			−0.027	−0.107–0.055		.529
Physical Activity
	Partially Adjusted ^a (Model 3a)					Fully Adjusted ^b (Model 3b)
	Beta	CI		P-value			Beta	CI		P-value
Young Adulthood	−0.020	−0.125–0.086		.717			−0.010	−0.117–0.098		.861
Mid Life	0.039	−0.069–0.146		.485			−0.019	−0.133–0.095		.748
Late Life	−0.080	−0.169–0.020		.120			−0.054	−0.153–0.051		.324

Note CI = Confidence Interval.

^aRegressions were controlled for age, sex, and education. In Model 1, education was not controlled as previously specified.

^bRegressions were controlled for age, sex, education, NESB, depression, frailty. In Model 1, education was not controlled as previously specified.

*P-values<0.05.

Post-hoc investigation of YA cognitive activity components

We investigated the relationship between global cognition and the different components of YA cognitive activity, specifically the contribution of formal education versus cognitive leisure activity. To do this, we repeated Models 1a and 1b from Table 2 but replaced YA cognitive activity with its two component scores; YA formal education and YA cognitive leisure activities, as described earlier, and retained ML and LL cognitive activity variables as covariates.

Table 3 shows YA formal education was not significantly associated with late life global cognition at Wave 2 in the partially adjusted model but was significantly associated with global cognition at Wave 2 in the fully adjusted model (β = 0.146, p = .003; see Table 3 Model 4a). YA cognitive leisure activities were not significantly associated with global cognition at Wave 2 in either partially or fully adjusted models (see Table 3 Model 4b).

Table 3. A linear regression model examining the associations between young adulthood (YA) cognitive leisure activities, YA years of education and global cognition (z-scores) at Wave 2.

	Partially Adjusted^b (Model 4a)			Fully Adjusted^b (Model 4b)
	Beta	CI (95%)	P-value	Beta	CI (95%)	P-value
YA Formal Education	0.097	−0.002–0.173	.055	0.146	0.045 – 0.220	.003*
YA Cognitive Leisure Activities	0.038	−0.052–0.118	.446	0.062	−0.029–0.137	.202

Note YA = Young Adulthood.

^aAge and sex were controlled for.

^bAge, sex, non-English speaking background, depression, and frailty were controlled for.

*P-values<0.05.

Longitudinal association between activity type, life stage and global cognition decline

As seen in Table 4, higher levels of LL cognitive and YA social activities were significantly associated with slower rates of decline in global cognition between Wave 1 to Wave 4, in both partially and fully adjusted models (LL: β = 0.019, p = .007and β = 0.016, p = .022, and YA: β = 0.024, p < .001and β = 0.023, p < .001, respectively; see Models 5 and 6). No physical activity variable was significantly associated with the rate of global cognitive decline at any life stage (Model 7).

Table 4. Linear mixed models examining the associations between (1) cognitive activity, (2) social activity, and (3) physical activity at three life stages, and the rate of change in global cognition (z-score) over time (wave 1–4).

Cognitive Activity
	Partially Adjusted ^a (Model 5a)					Fully Adjusted ^b (Model 5b)
	Beta ^c		CI		P-value	Beta ^c			CI		P-value
Young Adulthood x time	−0.001		−0.015–0.013		.902	−0.001			−0.015–0.014		.926
Mid Life x time	−0.001		−0.015–0.013		.891	0.006			−0.008–0.021		.381
Late Life x time	0.019		0.005–0.032		.007*	0.016			0.002–0.030		.022*
Social Activity
	Partially Adjusted ^a (Model 6a)					Fully Adjusted ^b (Model 6b)
	Beta ^c	CI		P-value			Beta ^c	CI		P-value
Young Adulthood x time	0.024	0.012–0.036		<.001*			0.023	0.010–0.035		<.001*
Mid Life x time	0.004	−0.009–0.017		.536			0.006	−0.008–0.019		.401
Late Life x time	−0.000	−0.013–0.012		.953			−0.001	−0.014–0.012		.840
Physical Activity
	Partially Adjusted ^a (Model 7a)					Fully Adjusted ^b (Model 7b)
	Beta ^c	CI		P-value			Beta ^c	CI		P-value
Young Adulthood x time	0.007	−0.009–0.023		.392			0.004	−0.013–0.022		.589
Mid Life x time	−0.005	−0.022–0.011		.545			−0.004	−0.023–0.014		.625
Late Life x time	0.003	−0.011–0.017		.644			0.009	−0.006–0.024		.251

Note CI = Confidence Interval.

^aAge, sex, and education were controlled for. In Model 5, education was not controlled as previously specified.

^bAge, sex, education, non-English speaking background, depression, and frailty status were controlled for. In Model 5, education was not controlled as previously specified.

^cBeta value represents change in standardized cognition score per year for every one unit change in predictor.

*P-values<0.05.

Post hoc analyses were conducted to examine whether a change in cognitive scores from Wave 2 (when the LEQ was administered) to Wave 4 (4-year follow-up) produced similar results to the model examining Wave 1 to Wave 4 cognition scores (6-year follow-up). For late life physical activity and social activity scores, the pattern remained the same; late life cognitive activity scores were, however, non-significant over the shorter follow-up period (see Supplementary Table S4).

Discussion

Our study aimed to examine the life-stage timing of known cognitive reserve-building activities and their comparative associations with cognition and cognitive decline in late life, to determine whether our data identified any emergent life stages during which building cognitive reserve is most critical.

Our study found that YA cognitive activity was significantly associated with late-life global cognition, and further that LL cognitive activity and YA social activity were significantly associated with reduced cognitive decline over the 6-year follow-up period. Cognitive and social activities undertaken in other life stages were not found to be significantly associated with late life global cognition or cognitive decline. Physical activity, in any life stage, was not found to be significantly associated with late life global cognition or cognitive decline.

Few studies have investigated the relationships between different types of cognitive reserve-building activities undertaken at different life stages and their association with cognitive performance in later life (Gow et al., 2017; Jefferson et al., 2011). Previous studies have investigated the associations between cognitive reserve-building activities and cognitive outcomes in late life; however, these studies have tended to focus on only one activity type, or a single life stage, which may miss the complex interplay of activities across the lifespan (Akbaraly et al., 2009; Chang et al., 2010; Suo et al., 2012).

Our findings are consistent with previous studies demonstrating isolated associations between YA cognitive reserve-building activities and late life cognition (Mantri et al., 2019; Snowdon et al., 1996) and LL cognitive activity and reduced cognitive decline (Dartigues et al., 2013). Within our older cohort, we found an association between YA cognitive and social activities and cross-sectional global cognition and longitudinal cognitive decline, respectively, which suggests YA could be a key time for maximizing cognitive reserve building, especially in terms of activities that promote social engagement and cognitive stimulation. Such findings encourage further investigation into intervention programs that particularly target this life stage for the prevention or delay of cognitive decline.

Our follow-up analysis focused on the component variables of our YA cognitive activity measure, namely YA formal education and YA cognitive leisure activities, which very few prior studies, if any, have examined independently. Previous studies have measured cognitive leisure activity engagement, but typically only focus on a specific activity such as reading or bilingualism (Wilson et al., 2015). A small subset of studies have measured a larger range of leisure activities and controlled for education (Chan et al., 2019), though not examining the two independently.

Our study showed that more years of formal education in YA, rather than cognitively stimulating leisure activities, was significantly associated with better global cognition scores in later life. The mechanism and direction of the relationship between education and better cognitive performance in LL is debatable. It is well established that years of education are strongly associated with other factors such as intelligence, financial stability, lifetime occupation, and health literacy (Clouston et al., 2017; OECD, 2022), all of which may be, at least partially, mediating this relationship.

Nonetheless, our findings are supportive of translational implications that educational attainment in YA could provide the greatest protective benefit against late-life cognitive decline for general or at-risk populations. Importantly, our study’s formal education measure encompasses a broad spectrum of educational environments, including university, trade apprenticeships, and technical courses, to name a few examples, which suggests the positive effect we found between YA education and late life cognition is not necessarily limited to traditional academia.

Our study also found that participants who engaged in more LL cognitive activity had less cognitive decline over 6-years. This finding builds upon previous studies supporting an association between LL cognitive activity as being protective against cognitive decline (Dartigues et al., 2013). Interestingly, our study found cognitive activity across the three life stages was differentially associated with cross-sectional cognitive performance compared to cognitive decline. For example, while YA cognitive activity was significantly associated with cross-sectional cognitive performance, less LL cognitive activity was significantly associated with cognitive decline over 6 years. The reason for these differences is not obvious. It can be speculated that while YA cognitive activity may contribute to cognitive reserve level, the trajectory of subsequent decline in late life is perhaps influenced by cognitive activities undertaken during those later periods. Late life could potentially be one of the most important and efficacious life stages to engage in cognitive activity to slow the rate of cognitive decline. If these findings were to be corroborated in future studies, this would lend support to the notion that it may never be too late for older adults to improve their cognitive outcomes and would resolve the question of reverse causality. That is, whether older adults who experience less cognitive decline are more inclined to undertake cognitively demanding leisure activities, such as learning a new language or instrument. However, this interpretation should be considered with caution as in additional analyses, where our study follow-up period was restricted from Wave 2 to Wave 4 (4-year follow-up) only, and LL cognitive activity was no longer significantly associated with cognitive decline. The meaning of this difference in findings is also uncertain, and whether, for example, the additional analysis provided insufficient data points to appropriately demonstrate significant predictors that emerge only over a longer follow-up period.

Physical activity was not found to be significantly associated with late life cognition or cognitive decline in any life stage in our study. There are mixed results in previous studies reporting on the relationship between physical activity and late life cognitive outcomes (Xu et al., 2017; Young et al., 2015). Our findings align with some studies that have reported no association between physical activity and late-life cognition (Young et al., 2015), but our findings do not corroborate other studies that have found a significant association (Xu et al., 2017). This difference may be due to heterogeneity in the characterization of physical activity between studies. For physical activity, our measure assessed activity levels by intensity, which was coded according to evidence-based MET values and combined with the frequency of participation. In other studies, physical activity measures were variably designed, for example using walking distances, activity checklists, or arbitrary cutoffs between high and low activity groups (Chang et al., 2010; Ravaglia et al., 2008).

Strengths of our study include the large, well-characterized sample with detailed demographic and medical histories and the relatively long follow-up period of 6 years. We also controlled for important covariates, such as non-English-speaking background, APOE4 status, and frailty. An additional strength is our extensive neuropsychological test battery, which allowed us to compute comprehensive global cognition scores where many other studies examining cognitive reserve rely on short tests such as the MMSE as a proxy for cognitive ability (Lee & Kim, 2016). We also benefitted from the relatively extensive measure of cognitive activity captured by the LEQ at each life stage, especially for YA. This allowed us to further investigate individual components of cognitive activity in follow-up analyses. Other studies are limited by single proxy measures of cognitive activity (e.g., years of schooling or occupation) captured at a single point (e.g., age 40 of ML), which provides limited opportunity for more considered follow-up analyses (Reed et al., 2011).

However, our study has several limitations. First, our measures of social and physical activity at each life stage are limited in number and scope. For social activity, recent studies use measures compromised by social network size, composition, quality of interactions, and other key features (Sharifian et al., 2019), which were not captured by the LEQ. For physical activity, our measure assessed activity levels with intensity that were coded according to evidence-based MET values and combined with their frequency. However, only three questions were used to assess physical activity across the entire life stage and the 5-point response scale was limiting. Second, our data were collected using a retrospective, self-assessment tool. While many previous studies have utilized a similar methodology, important limitations include memory bias and reverse causation. Only 7 of 548 participants (1.3%) received a dementia diagnosis at Wave 2 shortly after completing the LEQ. In the case of those participants, the self-report limitations of the LEQ may have had a heightened impact. Clear-cut questions such as years of schooling completed or formal education qualifications were less likely to be affected by memory bias, but questions such as frequency of reading or physical activity in each life stage may have been over- or under-reported. Additionally, the completion of the LEQ was voluntary and only half of MAS participants at Wave 2 opted to complete it. Group-level analyses showed that LEQ completers were less likely to have MCI and were more educated than LEQ non-completers, meaning our LEQ sample was not representative of the greater MAS sample, which itself aimed to be representative but was biased to a single area of Sydney that was more educated than the national average (Sachdev et al., 2010).

Our findings would benefit from replication in other cohorts, which will help determine whether critical life stages exist for cognitive reserve-building activities. Future studies should include a more comprehensive characterization of social and physical activity to determine whether the association between YA social activity and reduced late life cognitive decline and the null associations we found between physical activity and global cognition and cognitive decline, respectively, are replicated. Future research may consider examining the interactions between activity life stages and their impact on cognition or cognitive decline. Future research may also use neuropathology markers or imaging techniques to determine if our study’s findings reflect true cognitive reserve, which stipulates better cognitive functioning accounting for neuropathology. Our study had a relatively short follow-up period of 6-years, after which, comprehensive neuropsychological testing was not available. Ideally, future studies would examine cognitive trajectories over a longer period to determine the predictive utility of the LEQ.

In conclusion, our study found that YA cognitive activity was significantly associated with better LL global cognition and that YA social activity and LL cognitive activity were significantly associated with slower rates of cognitive decline in later life. Analysis of the components of our YA cognitive activity measure found formal education appeared to be driving the association with late life global cognition, as opposed to engaging in cognitively stimulating leisure activities. Our findings encourage future studies to continue investigating cognitive reserve using a lifespan approach, with the eventual aim of translational recommendations that optimize the timing of intervention strategies and population health programs for building cognitive reserve and preventing cognitive decline.

Acknowledgments

We thank the participants and their informants for their time and generosity in contributing to this research. We also acknowledge that the MAS research team has contributed to the creation and maintenance of the MAS dataset: https://cheba.unsw.edu.au/research-projects/sydney-memory-and-aging-study

Disclosure statement

HB has been an advisor or consultant to Biogen, Lily, Roche, Skin2Neuron and Cranbrook Care. MV is CEO of Skin2Neuron.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/13825585.2023.2181941

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

The work was supported by the National Health and Medical Research Council [APP1093083, ID350833, ID568969]; National Health and Medical Research Council of Australia