News

Predictors of longitudinal cognitive ageing from age 70 to 82 … – Nature.com

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Carousel with three slides shown at a time. Use the Previous and Next buttons to navigate three slides at a time, or the slide dot buttons at the end to jump three slides at a time.
30 March 2021
Ida K. Karlsson, Margaret Gatz, … Chandra A. Reynolds
10 January 2018
Mark James Rawle, Daniel Davis, … Marcus Richards
26 September 2018
Petroula Proitsi, Diana Kuh, … Marcus Richards
13 January 2022
Deirdre M. A. O’Connor, Siobhan Scarlett, … Rose Anne Kenny
30 June 2020
Franziska Röhr, Nina Bucholtz, … Sandra Düzel
08 December 2020
Adrianna P. Kępińska, James H. MacCabe, … Olesya Ajnakina
19 August 2022
Annabell Coors, Mohammed-Aslam Imtiaz, … Monique M. B. Breteler
16 August 2022
Kristine B. Walhovd, Lars Nyberg, … Anders M. Fjell
25 February 2021
Katrin Wolfova, Zsofia Csajbok, … Pavla Cermakova
Molecular Psychiatry (2022)
905 Accesses
32 Altmetric
Metrics details
Discovering why some people’s cognitive abilities decline more than others is a key challenge for cognitive ageing research. The most effective strategy may be to address multiple risk factors from across the life-course simultaneously in relation to robust longitudinal cognitive data. We conducted a 12-year follow-up of 1091 (at age 70) men and women from the longitudinal Lothian Birth Cohort 1936 study. Comprehensive repeated cognitive measures of visuospatial ability, processing speed, memory, verbal ability, and a general cognitive factor were collected over five assessments (age 70, 73, 76, 79, and 82 years) and analysed using multivariate latent growth curve modelling. Fifteen life-course variables were used to predict variation in cognitive ability levels at age 70 and cognitive slopes from age 70 to 82. Only APOE e4 carrier status was found to be reliably informative of general- and domain-specific cognitive decline, despite there being many life-course correlates of cognitive level at age 70. APOE e4 carriers had significantly steeper slopes across all three fluid cognitive domains compared with non-carriers, especially for memory (β = −0.234, p < 0.001) and general cognitive function (β = −0.246, p < 0.001), denoting a widening gap in cognitive functioning with increasing age. Our findings suggest that when many other candidate predictors of cognitive ageing slope are entered en masse, their unique contributions account for relatively small proportions of variance, beyond variation in APOE e4 status. We conclude that APOE e4 status is important for identifying those at greater risk for accelerated cognitive ageing, even among ostensibly healthy individuals.
With advancing age, a pattern of decline is observed across a multitude of cognitive domains, though the magnitude differs across domains, and there are marked individual differences in rates of cognitive change in the population [1, 2]. Some cognitive abilities, such as vocabulary, remain relatively intact into later life. Other, complex cognitive processes such as processing speed, reasoning, and memory—which require the manipulating of mental data—begin to decline from early adulthood [3,4,5], and some of these changes are underpinned by a general factor of cognitive ageing [6,7,8]. Deterioration in cognitive abilities is linked to impairments in older adults’ everyday functions [9], quality of life [10], and health [11]. Better understanding of long-term cognitive trajectories and their determinants could inform public policy regarding targeted interventions for those adults at greatest risk of rapid decline, and of progression to Alzheimer’s Disease (AD) and other dementias [12], as well as protective factors for staying sharp in later life.
The determinants of individual differences in age-related cognitive decline are likely to include genetic and early-life factors, adult socio-economic status (SES), and health [13,14,15], though estimates differ with respect to their individual contributions. Risk of accelerated cognitive decline increases with age, cerebrovascular disease, cardiovascular risk factors (e.g. diabetes, obesity) and heart disease [16], but these factors only partially account for cognitive decline risk among the general population [14]. The APOE (apolipoprotein) e4 allele is a well-established genetic risk factor for AD [17, 18], however, the reported effects of APOE e4 across the full spectrum of cognitive functioning are highly inconsistent and there is disagreement about whether or not APOE e4 influences the rate of cognitive decline in healthy adults [19,20,21,22,23,24,25]. Despite a broad corpus of research literature on the role of behavioural risk factors in mitigating age-related cognitive decline, such as smoking, physical activity, alcohol, and diet [3, 26, 27], the evidence is patchy and often classed as low to moderate quality [10]. Importantly, many of the effect sizes are small, and findings are often partly, or wholly, attributed to reverse causation, where prior cognitive ability causes variation in the supposed cause of cognitive ability in later adult life [13].
Cognitive decline trajectories are likely to be the result of an accumulation of small effects from numerous individual genetic and environmental risk factors across the life-course [28]. Even smoking, for which there is consistent and demonstrable evidence of an adverse effect on cognitive and brain ageing [29,30,31], generally accounts for around only 1% of the variance in cognitive decline, similar in magnitude to the estimated effect size of APOE e4 on cognitive change from childhood to adulthood [32]. Given that many risk factors for cognitive decline are correlated [33], modelling these potential predictors together, i.e. simultaneously, may be a more valuable approach than focussing on single-candidate determinants (such as one individual lifestyle or health factor). Unlike univariate accounts of cognitive ageing, multivariate modelling acknowledges the multicollinearity among risk factors and provides more insight into their relative contributions to cognitive change. The very few studies to have tested multiple risk factor models of longitudinal (multi-domain) cognitive decline report few consistent correlates of cognitive change across abilities [34, 35]. In the same sample as in the current study—the Lothian Birth Cohort 1936—an earlier multivariate analysis by Ritchie et al. showed that faster rates of decline from age 70 to 76 years were observed in APOE e4 carriers, men, and those with poorer physical fitness for some, but not all, cognitive domains [36].
A further challenge in understanding the predictors of cognitive ageing trajectories is the difficulty in disentangling actual cognitive change from lifelong levels of performance (which are conflated in cross-sectional data) and partitioning the variance appropriately [8]. Longitudinal studies with repeated cognitive measures across an extended period in later life, paired with appropriate methodologies for modelling change, are crucial for characterising the progression of cognitive change and robustly identifying its correlates [15]. Ideally, studies should establish the extent to which potential determinants of differences in cognitive ageing are independent of prior cognitive ability differences.
In the current study, we address these issues using data from the Lothian Birth Cohort 1936, an extensively-phenotyped, community-dwelling sample of older adults in Scotland, for whom there are comprehensive cognitive data collected at five time-points across later life (age 70–82), cognitive ability scores from early-life, and data on a wide range of potential covariates (see Box 1 for a summary of study characteristics). Trajectories of cognitive function were evaluated using latent growth curve (LGC) modelling for four major domains of cognitive ability—visuospatial ability, processing speed, and memory (characterising fluid intelligence), and verbal ability (characterising crystallised intelligence). A main aim was to examine which putative cognitive ageing predictors from across the life-course survive simultaneous entry in multivariate cognitive models, using fifteen of the most commonly-used candidate risk factors in the field of cognitive ageing, covering: early-life (education, childhood IQ); demographic (age, sex, living alone, SES); lifestyle (smoking, physical activity, body mass index, alcohol), health (cardiovascular disease (CVD), diabetes, stroke); depressive symptoms; and APOE e4 carrier status. The present study doubles the time frame of the above-mentioned LBC1936 paper by Ritchie et al. [36] from 6 to 12 years of follow up, covering a more critical period for accelerated cognitive decline and dementia [37, 38], and includes several additional potential predictors (depression, living alone, physical activity, stroke). Having previously identified APOE e4 status as an independent predictor of cognitive change in this cohort, we perform separate trajectory analyses by APOE e4 carrier status. We also examine associations between predictors and a general factor of cognitive function which accounts for the shared variance across the cognitive domains.
1. Comprehensive cognitive battery with several high-quality tests for each cognitive domain.
2. 12-year follow-up—5 testing periods—using identical tests, equipment, and testing location on each occasion.
3. Cognitive testing across an important period from age 70, when cognitive ageing becomes pertinent, to age 82, when risk of rapid decline and dementia dramatically increases.
4. Record of general cognitive ability from a well-validated test at age 11.
5. Multiple (some correlated) candidate determinants are included in mutually-adjusted models, enabling estimates of relative contributions of each predictor to cognitive change.
6. 1,091 participants at baseline.
7. Sensitivity tests for incident dementia and death.
Participants were from the Lothian Birth Cohort 1936 (LBC1936) [39,40,41], a community-dwelling sample of 1091 men and women in Scotland, being studied in later life for the purposes of assessing the nature and determinants of cognitive and brain ageing. Most LBC1936 participants had taken part in a Scottish national intelligence test at age 11 years. The Scottish Mental Survey 1947 tested the cognitive ability of almost all Scottish children born in 1936, and attending school on 4 June 1947 (N = 70,805), using a validated test of general mental ability (The Moray House Test (MHT)) [42]. The first wave of the LBC1936 study was conducted between 2004 and 2007 at the age of ~70 years, and participants have been followed-up every 3 years at ages 73 (N = 866), 76 (N = 697), 79 (N = 550), and 82 (N = 431). Socio-demographic, medical history, physical function, blood-derived biomarkers, cognitive function, and lifestyle data were collected at all five waves of in-person testing. For the purposes of the current study, “completers” (N = 431) refer to participants who attended all five assessments at ages 70, 73, 76, 79, 82, and “non-completers” (N = 660) refer to the remaining participants those who took part in ≤4 assessments, and either withdrew or died before age 82 follow-up. All participants who completed at least the first wave of testing at age 70 were included in the main analyses (see Fig. S1 flowchart showing waves of testing, attrition and deaths).
Cognitive function was measured using a detailed battery of well-validated cognitive tests administered by trained psychologists at age 70 (baseline) and the same tests were repeated at ages 73, 76, 79, and 82 years [39]. Most of the cognitive tests derive from the Wechsler Adult Intelligence Scale III-UK edition [43] and the Wechsler Memory Scale III-UK edition (WMS-IIIUK) [44]. According to previous work examining their correlational structure [7], the cognitive tests were categorised into four domains of cognitive functioning. Visuospatial ability was measured using Block Design and Matrix Reasoning (WAIS-IIIUK) and Spatial Span (Forwards and Backward) (WMS-IIIUK). Processing Speed was measured using Digit-symbol Coding and Symbol Search (WAISIII-UK) and two experimental tasks: Choice Reaction Time [45]; and Inspection Time [46]. Memory was measured using Verbal Paired Associates and Logical Memory (WMSIII-UK) and Digit-span Backwards (WAIS-IIIUK). Verbal ability was measured using the National Adult Reading Test [47], the Wechsler Test of Adult Reading [48], and Verbal Fluency [49]. A general cognitive factor was constructed based on the shared variance between the four cognitive domains (see “Statistical analysis”). The Mini-Mental State Examination (MMSE) [50], widely used as a screening test for possible dementia, was administered at each wave of testing.
Potential risk or protective factors for cognitive decline in later life were identified following a review of previous analyses of the cohort and other population studies; values were obtained from participants’ baseline assessment at age 70.
These predictors included age (in days), sex, age 11 IQ score, education (years of formal full-time schooling), living alone (yes/no), and SES. MHT scores from age 11 (SMS1947) were recorded and archived by the Scottish Council for Research in Education and were made available to the LBC1936 study. For the current study, MHT scores from age 11 were age corrected and converted into a standard IQ-type score for the sample (mean = 100, SD = 15)—henceforth referred to as age 11 IQ—and used a measure of childhood cognitive ability. SES was coded into six categories based on participants’ highest achieved occupation: 1 (highest professional occupations) to 5 (unskilled occupations), with 3 (skilled occupations) divided into 3N (non-manual) and 3M (manual), using the Classification of Occupations, 1980 [51].
Smoking was coded as current, former or never smoker. Physical activity was coded according to six categories: 1 (“moving only in accordance with household chores”; lowest level of activity) to 6 (“keep fit or aerobic exercise several times a week”; highest level of activity). Alcohol units per week were calculated using data collected at interview. Body mass index was calculated using height and weight measurements taken by trained nurses at the time of assessment.
APOE e4 carrier status (yes/no) was determined by genotyping at two polymorphic sites (rs7412 and rs429358) using TaqMan technology. Depressive symptoms were measured using the Depression Subscale of the Hospital Anxiety and Depression Scale [52] with a score range of 0–21. Health indicators included self-reported history (yes/no) of CVD, diabetes, and stroke.
Descriptive statistics are presented for the full sample, and ANOVA and Chi-square tests were used to identify differences in baseline characteristics between study completers vs. non-completers, and between deaths to follow-up vs. survivors.
We applied LGC modelling to the data to investigate level (i.e. intercept, age 70) and trajectories of change (i.e. slope, age 70–82) in cognitive functioning across all five waves of testing. Participants were included in the analytic sample even if they attended baseline-only, as the estimates for intercept (i.e. cross-sectional) and slope (i.e. longitudinal) associations are derived simultaneously from the same LGC model using all available data. A SEM-based “factor-of-curves” [53] approach was used, as has been done previously in this cohort [36, 54] which postulates the existence of common latent variables of cognitive change that underlie the distribution of explicit or observable variables (individual cognitive tests). In our models, we used the average time lag (in years) between the waves: (0, 2.98, 6.75, 9.81, 12.53) as the path weights for calculation of the slope factor. The path from the slope factor to baseline test score was set to zero. LGC analyses were conducted using the latent variable analysis package “lavaan” [55] in R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria) and the code is available online (https://www.ed.ac.uk/lothian-birth-cohorts/summary-data-resources). First, we fitted a single parallel process growth curve model at the level of the 13 individual cognitive tests; intercepts and slopes were correlated, but no hierarchical factor structure was imposed. Second, we fit separate growth curve models for each cognitive domain: visuospatial ability; processing speed; memory (Visual inspection of the fitted regression lines through the individual cognitive test scores at each wave indicated that memory might best be modelled using a non-linear factor of change (to account for the rise in mean test scores in the initial waves of testing, followed by a fall toward the end of the follow-up). To test for potential curvilinear trajectories for memory, we included a quadratic term in separate measurement models for the latent memory domain. However, these models did not converge successfully and are not discussed further.); and verbal ability. Here, the latent intercepts and slopes of each cognitive test load onto superordinate latent intercepts and latent slopes of their respective cognitive domains. The cognitive domain models were run for the full sample and also by APOE e4 carrier status (yes/no). Unstandardised (beta) estimates, standard errors, p values, and standard deviation (SD) change per year, are reported.
Next, we fit both univariate and multivariate risk factor models to the cognitive data to address which factors might contribute to individual differences in cognitive level (age 70) and slope (age 70–82). First, univariate LGC models were fit to test the associations of each life-course predictor (alongside age and sex) with each cognitive domain, i.e. without the other variables present in the model. For our main analyses, we fit multivariable LGC models which included all 15 predictor variables for each cognitive domain. By including all of the predictors simultaneously, we were able to compare the degree of variance in cognitive level and change accounted for by each risk factor, whilst controlling for the effects of all the other predictors in the same model. Our analysing the paper as we have done is in response to many papers in our field that tend to focus on a single predictor with a few basic covariates (age, sex, medical conditions, etc.) isolated from other predictors. Here, a main aim was to find out how many of the commonly-used cognitive ageing predictors survives simultaneous entry. We ran an additional model representing a general cognitive factor; this hierarchical model was fitted using the latent intercepts and slopes of each of the four cognitive sub-domains, and represents the shared (common) variance between them (Fig. 1 illustrates the hierarchical model framework for general cognitive function). Fully standardised estimates, obtained using the “standardizedSolution” function in lavaan, are presented.
A latent growth curve model in which predictors are associated with the intercept and slope of a latent factor of general cognitive function. A latent growth curve was estimated across five waves of data in a hierarchical model based on the intercepts and slopes of four cognitive domains. For illustrative purposes, not all tests are shown. The full model included at least three tests per domain. The regressions of predictors (represented by the dotted lines) on general cognitive function intercept (i) and slope (s) were the associations of interest.
With a large set of predictors, as in the current study, we increase the proportion of variance that can be explained in our cognitive outcomes by chance. In order to test whether or not the variance accounted for by the real predictors was comparable to a set of random predictors, we generated a set of Gaussian noise (and random binary) variables and entered them into the LGC models in place of the real predictors, and compared the model R2 for each domain. To optimise comparability, we ensured that the same number of continuous vs. binary variables were used, and that the patterns of missingness were matched with the real-world predictors.
We repeated the same baseline prediction models in three sensitivity analyses excluding: (1) individuals who reported a subsequent-to-baseline diagnosis of dementia (all participants were dementia-free at baseline); (2) individuals with an MMSE score <24 at any wave, as an indicator of possible pathological ageing; (3) deaths to follow-up (using linkage data obtained via National Health Service Central Register up to April 2021, provided by the National Records of Scotland).
In order to further examine the multivariate associations, a SEM-based path model was constructed with the latent variable of general cognitive function (g) intercept and slope as the dependent variables. The path diagram (Fig. S2) represents a life-course model with predictors from childhood to older age included. Specific assumptions regarding the direction of causal relationships were built into the model. We assumed chronological paths from childhood IQ → education → adult SES. Based on previous literature, we also assumed that childhood IQ, education, and adult SES might influence the lifestyle and health predictors, and that APOE e4 might influence CVD. All the predictors in the model have direct paths to g intercept and g slope. Direct pathways represent the unique contribution of each predictor to the outcome variable, which is not accounted for by other mediating pathways.
Models were run using full information maximum likelihood (FIML) estimation to ensure models used all available data to partially mitigate the bias of estimated trajectories and associations by participation bias. Instances of non-significant negative residual variance were set to 0 to allow models to converge upon within-bounds estimates. Model fit was tested using three indices of absolute fit: comparative fit index and Tucker-Lewis Index (values > 0.95 considered acceptable); and root mean square error of approximation (values < 0.06 considered acceptable). Correction for multiple testing was applied across LGC prediction models using the false discovery rate (FDR) [56] adjustment, and results marked in boldtype are FDR-significant.
Baseline characteristics and cognitive test scores for the LBC1936 sample (N = 1091) are shown in Table 1. Baseline age was 70 years (mean = 69.5, SD = 0.8), 49.8% of the sample were women, and mean number of years of education was 10.7 (SD = 1.1). APOE e4 allele carriers (N = 306) made up 28.0% of the overall sample. APOE e4 data were missing for 63 participants (5.8% of the sample). See Table S1 for missing covariate data. Characteristics are also presented according to completer status (completers vs. non-completers), and mortality status (deaths vs. non-deaths) by the end of the follow-up period. Compared with individuals who attended all five waves, non-completers had less education, lower childhood IQ, lower SES, lower physical activity, higher BMI, more depressive symptoms, and were more likely to be a smoker, have a history of CVD, diabetes, and stroke. Non-completers had significantly lower cognitive test scores at baseline than completers. Participants lost to follow-up as a result of death (N = 403) had a lower age 11 IQ, lower SES, lower physical activity, higher BMI, higher alcohol intake, more depressive symptoms, and were more likely to be male, a smoker and to have a medical history of CVD, diabetes and stroke, than those who survived to follow-up. Mean cognitive test scores at baseline were significantly lower in those who had died, compared with the survivors, except for Verbal Pairs (a memory test) and Verbal Fluency (a verbal ability test), for which the group differences were not significant. As noted above, we used FIML estimation in our LGC analyses to reduce any bias due to missingness.
A summary of the longitudinal cognitive test scores for the whole sample is presented in Table 2. Mean cognitive test scores declined between age 70-baseline and age 82 follow-up, except for two memory tests (Logical Memory and Verbal Pairs) and the verbal ability tests (NART, WTAR, and Verbal Fluency), which were marginally higher at age 82. Logical Memory and Verbal Pairs contain memorable material, which may have resulted in a rise in score in at least the second occasion of testing as a result of practice effects. All three verbal ability tests showed little change over time, and small increases in mean scores at age 82 compared with baseline. Further descriptive information about the cognitive tests scores for completers only, and by APOE e4 carrier status, is provided in the Supplementary Materials. In the subset of completers only (Table S2); this has the advantage that the same individuals appear at all waves, all of the mean cognitive test scores were lower at age 82 follow-up compared with baseline with the exception of WTAR (where the mean score was the same), and NART and Verbal Fluency which were slightly higher at follow-up. Note that Choice Reaction Time is scored negatively, such that a higher score indicates a slower reaction time. Mean cognitive test scores at age 70 and age 82 differed according to APOE e4 carrier status (Table S3). At age 70, APOE e4 carriers had significantly lower scores on Matrix Reasoning, Spatial Span and Inspection Time than non-carriers. By age 82, APOE e4 carriers had significantly lower scores on Block Design, Matrix Reasoning, Spatial Span, Digit-symbol Coding, Symbol Search, Choice Reaction Time, Logical Memory, Verbal Pairs, and Digits Backwards, and the differences were larger in magnitude than at age 70. Figure 2 plots the linear fitted regression lines through the raw test data for each of the cognitive tests by APOE e4 carrier status (non-linear fitted lines through the same data can be found in Fig. S3).
Plots of the regression lines fitted through the raw data, normalised for baseline score, to illustrate the differences in trajectories of cognitive change with age by APOE e4 carrier status (with shaded 95% confidence intervals). Red = non-carrier, blue = carrier.
First, we tested whether there was significant ageing-related mean change in each of the 13 individual cognitive tests in a single parallel process LGC model (Table S4). There was a significant, negative mean slope for all tests (p < 0.001 except WTAR (p < 0.05)), with the exception of NART where the slope was non-significant. SD change per year was calculated for each cognitive test score and ranked in order of most change (1) to least change (13). The four individual processing speed tests showed the largest SD declines over time (range, −0.120 to −0.072), followed by the three visuospatial tests (range, −0.055 to −0.038), the three memory tests (range −0.038 to −0.027), and the three verbal ability tests (range, −0.010 to 0.0001) which showed the least decline. SD change in NART scores was marginally positive but not significantly different from zero (SD change/year = 0.0001).
Second, we tested whether there was significant ageing-related mean change in each of the four latent cognitive domains for all participants, and then separately by APOE e4 carrier status in LGC models (Table 3). In the full sample, there was a significant, negative mean slope of ageing-related change across all four cognitive domains. The latent variable of processing speed showed the greatest SD decline per year between age 70 and 82 (SD change/year = −0.088), followed by visuospatial ability (SD change/year = −0.054), memory (SD change/year = −0.028), and verbal ability (SD change/year = −0.003). Model fit indices for Table 3 are shown in Table S5, alongside those for Tables 4 and 5.
In the APOE e4 non-carriers sub-group, the slopes, indicating negative mean change over time, were significant for processing speed (SD change/year = −0.068) and visuospatial ability (SD change/year = −0.033) only, but there was little (and non-significant) change in memory (−0.010) or verbal ability (−0.004). In the APOE e4 carrier sub-group, the mean slopes were negative and significant for all but verbal ability. Compared to the APOE e4 negative group, APOE e4 carriers showed greater SD decline in processing speed (SD change/year = −0.106 vs. −0.068), visuospatial ability (SD change/year = −0.065 vs. −0.033), and memory (SD change/year = −0.072 vs. −0.010). The difference was most marked in the slope for memory; APOE e4 carriers showed a seven-fold greater SD decline per year compared with APOE e4 non-carriers (and in the non-carrier group the slope for memory is non-significant). In contrast with the full sample and the APOE e4 non-carriers, memory decline was steeper than visuospatial ability decline in the APOE e4-positive group. Figure 3 presents horizontal bar plots illustrating the SD change/year in each cognitive test for all participants, and in each cognitive domain for all participants, APOE e4 carriers, and APOE e4 non-carriers. Formal tests of intercept and slope differences for APOE e4 carriers and APOE e4 non-carriers are carried out below.
Standard deviation (SD) change per year in a each cognitive test (grouped by cognitive domain), and b each cognitive domain (grouped by all participants, and by APOE e4 non-carriers and carriers). SD change per year was derived from latent growth curve models, by calculating the slope mean divided by the intercept SD. SD change per year was converted to +ve values for illustrative purposes, with the exception of NART (National Adult Reading Test) which became –ve. Error bars represent the standard error of SD change per year.
First, we performed univariate analyses which regressed the intercepts and slopes at the level of each cognitive domain, and then general cognitive function, on all of the predictor variables individually. These univariate (partially-adjusted) models are distinct from the later models featuring multiple risk factors (fully-adjusted) which are the main models of interest. In the univariate models for cognitive ability level at age 70, all of the predictors except living alone were significantly associated with scores on at least one cognitive domain (full results are shown in Table 4). In the univariate models for cognitive slope, only APOE e4 status, alcohol, smoking, and age 11 IQ were significant predictors of decline across selected domains. APOE e4 carriers were more likely to show decline between age 70 and age 82 in visuospatial ability (β = −0.185, p = 0.005), speed (β = −0.215, p < 0.001), memory (β = −0.235, p < 0.001), and general cognitive ability (β = −0.233, p < 0.001). Smoking was associated with more decline in verbal ability (β = −0.203, p = 0.004) only, and a higher alcohol intake was associated with more decline in visuospatial ability only (β = −0.183, p = 0.015). Finally, a higher childhood cognitive ability (β = −0.252, p = 0.001) was associated with more decline in visuospatial ability only.
Next, we ran multivariate models to simultaneously estimate the associations of multiple risk factors on cognitive level at age 70. We ran collinearity diagnostics and inspected tolerance and variance inflation errors. Variance inflation factor and tolerance levels were within acceptable limits (tolerance > 0.10 and variance inflation factors < 10.0 [57]; and thus did not indicate multicollinearity. When all 15 predictors were modelled at the same time, 13 (not living alone or alcohol intake) made a significant contribution to the variability in cognitive ability level at age 70 (i.e. the intercept) in at least one of the cognitive domains (upper section, Table 5). Performance on all four cognitive domains and the general factor of cognitive function was associated with age (within-wave differences) (range, standardised beta (β) = −0.089 to −0.157, p < 0.001) and age 11 IQ (range β = 0.442 to 0.668, p < 0.001); age 11 IQ accounted for the most variance in cognitive level of any of the predictors, with the largest effect size (β = 0.668) for general cognitive function. Education and SES predicted performance in the general factor, and three out of four of the domains (no association between education-speed and between SES-memory), with an average (β) effect size across the four domains of −0.176 and −0.123, respectively. The directions of associations were as expected, such that individuals with better age 70 cognitive ability level were younger, had a higher childhood intelligence, were more educated, and were from more professional occupational classes. Male sex (β = 0.261, p < 0.001) was a predictor of better visuospatial ability level, and female sex was a predictor of better memory level (β = 0.121, p < 0.001), but sex was not a significant predictor of general cognitive function.
Healthy lifestyle factors were selectively associated with better cognitive ability at age 70: more physical activity (β = 0.082, p = 0.009) and less smoking (β = −0.095, p = − 0.001) with better processing speed. A higher BMI (a measure of obesity) was associated with a lower verbal ability (β = −0.053, p = 0.01) but conversely with higher visuospatial ability (β = 0.084, p = 0.003). Alcohol intake did not significantly predict age 70 cognitive ability in any domain. None of the lifestyle factors measured were significantly associated with general cognitive function in the multivariate model. APOE e4-positive carrier status predicted poorer visuospatial ability (β = −0.100, p < 0.001), processing speed (β = −0.103, p < 0.001) and general cognitive function (β = −0.056, p = 0.009) at age 70. History of disease was associated with lower cognitive scores but not consistently across domains: CVD (β = −0.069, p = 0.013) and stroke (β = −0.071, p = 0.011), were associated with lower processing speed, in addition to a non-FDR-significant association with diabetes (β = −0.057, p = 0.04). Diabetes was associated with lower verbal ability (β = −0.053, p = 0.01) and general cognitive function (β = −0.055, p = 0.01). Depressive symptoms were associated with lower processing speed (β = −0.101, p < 0.001) and general cognitive function (β = −0.066, p = 0.002). Notably, many of the previous univariate associations between individual predictors and cognitive level at age 70 (across selected domains) became non-significant in the multivariate models.
In contrast to cognitive level at age 70, we found that few predictors were associated with longitudinal cognitive change between age 70 and 82 (as shown in Table 5 for slope, lower section) once all 15 predictors were entered simultaneously. APOE e4 carrier status accounted for the most variability in cognitive slopes. Possessing the APOE e4 allele was associated with significantly steeper decline in visuospatial ability (β = −0.170, p = 0.009), processing speed (β = −0.211, p < 0.001), memory (β = −0.234, p < 0.001), and general cognitive function (β = −0.246, p < 0.001), but not with verbal ability (β = −0.058, p = 0.35). Moreover, APOE e4 was the only unique significant predictor of cognitive change in processing speed, memory, and general cognitive function, with resultant effect sizes markedly larger in magnitude than any of the other variables. Other than being an APOE e4 allele carrier, a steeper slope in visuospatial ability was also associated with a having a higher age 11 IQ (β = −0.272, p < 0.001). The only predictors of a steeper verbal ability slope were more smoking (β = −0.192, p = 0.007), and contrary to expectations, a lower age (β = 0.262, p < 0.001). Comparisons between the univariate and multivariate predictor models for cognitive slope indicate that the univariate association between higher alcohol intake and greater decline in visuospatial ability (β = −0.183, p = 0.015) was non-significant in the multivariate model (β = −0.146, p = 0.05).
Figure 4 illustrates the unique variance (R2) accounted for by the 15 predictor variables in Table 5 for each cognitive domain, vs. a matched set of simulated random variables. These comparisons allow us to check whether our predictor group performed better than the same number of null variables, and are presented as stacked barplots showing the real data (in colour) and random data (in grey). The overall R2 for the set of real predictors was significantly larger than the null scenario across the domains: visuospatial ability (real = 20%, null = 4%); processing speed (real 8% = null = 2%); memory (real = 8%, null = 1%); verbal ability (real = 16%, null = 4%); general cognitive function (real = 9%, null = 2%).
Stacked barplots showing the unique variance (R2) in cognitive domain slopes explained by the predictor variables in the multivariate models (Table 5). Grey columns show the R2 explained by the same number of simulated (random) variables in each cognitive domain as a comparison.
We performed three sensitivity analyses to determine whether our results were driven by: participants who developed dementia by the age 82 assessment (N = 24); low MMSE scorers at one or more testing waves (N = 46); or deaths (N = 403). We found no substantive differences between the results of the sensitivity analyses (reported in Tables S68) and those reported above. The only notable result of these exclusions was an attenuation in effect sizes for the APOE e4 associations with visuospatial ability slope, of 46%, 22%, and 34%, respectively, across the three analyses, which were no longer significant at p < 0.05.
The life-course path model, showing significant associations (standardised beta regression weights) among the variables, is illustrated in Fig. 5 and full results can be found in Supplementary (Table S9). “Living alone” was not included as it was not associated with any of the cognitive domains in the LGMs. Direct paths to g intercept from the earlier-life factors were significant: childhood IQ (0.666, p < 0.001); education (0.199, p < 0.001); mid-life SES (−0.117, p < 0.001), as were the direct paths from depressive symptoms (−0.066, p < 0.001) and diabetes (−0.060, p < 0.001) to g intercept. The path model did indicate mediation paths from age 11 IQ → depressive symptoms and diabetes → g intercept. The % of the direct effect from age 11 IQ → g intercept mediated by these two health factors was minimal, at 1.0% and 0.8%, respectively. As in the multivariate LGM, the sole predictor of cognitive slope was APOE e4 carrier status (−0.240, p < 0.001). None of the lifestyle or health predictors had significant paths to cognitive slope. The association of APOE e4 with general cognitive function decline was not mediated by an increased risk of CVD as hypothesised, neither was there any indication in the model of any other mediator effects by the other life-course variables, on cognitive change. In response to a reviewer, we also tested for an interaction effect of APOE e4 × CVD on g intercept and slope in a separate model, given the role of APOE in CVD prevalence, and neither path was significant (0.012, p = 0.674; 0.005, p = 0.911, respectively). The path model demonstrates that APOE e4 status uniquely, among this set of predictors, influences cognitive change from age 70 to 82 years in the LBC1936, even when the variance from the other predictor variables is accounted for.
Path (SEM) model showing significant (*p < 0.05, **p < 0.01, ***p < 0.001) associations between early-, mid-, and later-life factors. All model predictors were regressed on a latent variable of general cognitive function (g) intercept and slope, estimated within the model. The numbers accompanying the paths are the beta values (standardised regression weights). Age (in days) and sex were included in the path model but not shown to reduce visual clutter. The covariances between lifestyle and health factors are not shown here. Full results are presented in Supplementary Table S9. SES is coded negatively (lower values=higher (more professional) social class).
We examined 12-year trajectories of cognitive functioning, using multiple measurement points across later life, in a birth cohort of community-dwelling older adults for whom childhood cognitive ability scores are available. Five waves of cognitive assessments were used to model change in visuospatial ability, processing speed, memory, and verbal ability from age 70 to 82 years, allowing a robust examination of rates of cognitive decline. Using a multivariate approach, we examined the relative contributions of determinants of individual differences in age 70-cognitive level and age 70 to 82-cognitive change, using 15 of the most commonly used candidate risk factors in the field of cognitive ageing. Our key finding is that APOE e4 status was the single most important factor determining longitudinal cognitive decline when all of the predictors were modelled simultaneously. Carriers of the APOE e4 allele show significantly steeper declines across the three “fluid” domains of memory, processing speed, and visuospatial ability, compared to non-carriers, even after adjusting for many other potential predictors which were strong correlates of age 70 cognitive level (including childhood IQ, education, adult socio-economic status, lifestyle, and health). APOE e4 status was the sole predictor of decline in general cognitive function—with a moderate to large effect size of 0.25 [58]—comparable in magnitude, for instance, to the reduction in risk of dying from head injuries associated with wearing a cycling helmet [59]. This contrasts with the relatively modest cross-sectional associations between APOE e4 and cognitive functioning at age 70 which suggests that the effect of APOE e4 on cognitive deficits becomes more manifest in later life. These findings are striking given that when many other candidate predictors of cognitive ageing slope are entered en masse, their unique contributions account for relatively small proportions of variance, beyond variation in APOE e4 status, and might indicate an increasing genetic influence on cognitive outcomes as individuals’ progress into their eighth and ninth decades of life.
The presence of faster rates of decline in APOE e4 carriers, across several different domains of cognitive functioning, adds valuable new data to the debate on whether APOE e4 influences “normal” cognitive ageing. Our findings stand in contrast with some studies reporting null findings such as the Australian PATH study [60], and the HALCyon programme which provided only very limited evidence of an effect of APOE e4 on a test of word recall, but not on other cognitive measures [19]. Discrepancies in findings may reflect differences in sample age; both samples were considerably younger than the present study, perhaps too young to show e4-related decrements. Our results extend prior work that does find an effect of APOE e4 in the following ways. First, we report that APOE e4 exerts broad and general adverse effects on cognitive functioning, typically only reported in cross-sectional meta-analytic data across many piecemeal studies [25] but not in a single longitudinal analysis. Second, we found a particularly deleterious effect of APOE e4 on memory decline, consistent with two single-candidate studies using a single memory test [23, 61]. Here, we show this association is robust to simultaneous adjustments in a multi-candidate study, and reliable across a broad cognitive trait of memory, captured by the latent domain. Third, we show that the relationship between APOE e4 and long-term cognitive decline is largely independent of childhood cognitive ability, an important confound (but rarely available measure) in studies of cognitive ageing [62]. Fourth, we were able to show that the APOE e4 allele affects age-related cognitive decline independently of possible cognitive impairment, dementia, and deaths to follow up, suggesting that this relationship is present, not just in dementia and AD [17, 63], but in cognitively “healthy” individuals.
Our results suggest that differences in cognitive functioning between e4 and non-e4 carriers become more pronounced with advancing age, regardless of any pathological changes. This finding aligns with earlier reports of an age effect of APOE e4 on cognition across the lifespan in single-determinant studies, with associations rarely seen in those <70 years [19, 23]. Age effects are consistent with theories that APOE e4 carriers are more vulnerable to damage accumulated over their lifetime, via reductions in neural protection and repair [64]. The APOE e4 allele is implicated in exacerbating neurodegeneration, tau pathology and inflammation; all pathological hallmarks of AD [65, 66]. Yet, the precise mechanisms by which APOE e4 exerts a deleterious effect on brain health in non-pathological ageing is currently unclear. In some studies, common neuropathologies including B-amyloidosis and tau tangle densities account for nearly all age-related cognitive decline [67, 68], raising the possibility that estimates of cognitive decline may be inflated by undiagnosed AD. However, residual effects of APOE e4 on cognition in cognitively-normal individuals have been reported even after controlling for AD pathology [69]. A recent neuroimaging study in UK Biobank has found that APOE e4 genotype associates with an increased burden of white matter hyperintensities, a marker of poor cerebrovascular health [70].
The presence of preclinical dementia may account for observed associations between APOE e4 and cognitive function [21, 71] leading to an overestimation of the effect of APOE e4 in age-associated, non-pathological cognitive decline. In the current study, the associations remained robust even after the exclusion of individuals with low MMSE scores indicating impaired cognition. With the exception of visuospatial ability, the effect sizes were of similar magnitude, indicating that the APOE e4-cognition associations were not driven by a sub-group who subsequently developed dementia. Our results are consistent with those of another study involving our sister cohort, the LBC1921, with whom we share similar methodology. Addressing a common criticism of studies investigating “normal” cognitive ageing—lack of diagnostic follow-up for dementia ascertainment—the authors used evidence from medical records, deaths certificates and clinical reviews to ascertain dementia status after 16 years of follow-up. They found that unrecognised dementia at baseline (age 79 years) had a small or no effect on the determinants of cognitive ageing including APOE e4 [72]. Given their conclusions, we judge that prodromal or undiagnosed dementia had little influence on our findings of a robust association of APOE e4 status and cognitive slope.
We found limited evidence in the LBC1936 that individual health behaviours alter rates of decline between ages 70–82 years when modelled in tandem with other life-course predictors. Those with a history of smoking showed faster declines in verbal ability, consistent with prior work documenting the detrimental effects of smoking on cognition and brain health [27, 29, 30], though the change in this crystallised domain was minimal over time. One major question for the field of cognitive ageing is whether various lifestyle choices all compete for a limited opportunity to enhance cognitive function or whether the effects could be additive, as part of a synergistic lifestyle pattern [73, 74]. While there were few individual effects, Fig. 4 makes it clear that together, lifestyle predictors account for a greater amount of the variance in cognitive decline than might be attributed to chance. In accordance with a “marginal gains” theory of cognitive ageing [28], individual differences in cognitive trajectories among our sample, probably reflect an accumulation of many small influences from numerous lifestyle (and other) factors. Though the magnitude of the observed associations between the various individual lifestyle factors and cognitive change were mostly small, if these associations represent a causal effect, their cumulative efforts are likely to have significance for cognitive health at the population level.
The presence of a significant intercept but not slope relationship with some past or premorbid factors supports a “passive” model of cognitive reserve [75]. That describes the situation, for instance, where highly-educated individuals continue to perform at a higher level of cognitive functioning than their less educated peers (i.e. influencing baseline differences, which we found), rather than having the ability to compensate for deficits (i.e. differential rates of cognitive decline over time, which we did not find). Other studies on cognitive decline show comparable findings for early-life socio-economic advantage [76] and education [77]. Here, this finding extends to early cognitive ability. Consistent with previous studies [36, 78], a higher childhood IQ—the strongest predictor of higher cognitive level in our sample—did not confer an advantage in terms of protection from steeper declines in the long-term. In fact, higher early-life cognition was associated with greater decline in visuospatial ability. This counterintuitive finding was surprising but not unusual, and may indicate regression to the mean, that is, a consequence of higher ability individuals performing relatively more poorly on tests with known ceiling effects when followed longitudinally [79]. Nevertheless, the current study benefits from knowing individuals’ cognitive starting point in order to ascertain degree of decline and to rule out confounding or reverse causation. Early-life cognition is associated with a subsequent cascade of social, behavioural and clinical effects [80], such that children with higher cognitive ability tend to become brighter and healthier adults [28], thus being able to remove this confound is important to reduce the likelihood of the observed associations being artefacts of the relationship between childhood IQ and healthy life markers. In doing so, our findings help to address an important issue in cognitive ageing research, namely, distinguishing differential preservation from preserved differentiation [8, 81]. With the clear exception of APOE, our results support the preserved differentiation of cognitive function only—whereby level of ability is a manifestation of prior ability—but not differential preservation (which leads to differences in subsequent rates of decline).
Finally, we observed that declines in processing speed between age 70 and 82 were greater than those of the other domains which supports the theory that processing speed is the core issue responsible for deficits in performance on complex cognitive measures in ageing populations [82,83,84]. Memory declined less steeply, across the whole sample, than processing speed and visuospatial ability, even in the ninth decade when one might expect to see more pronounced changes in this domain [85]. However, memory tests repeated longitudinally are subject to practice effects, whereby participants may improve or maintain their tests scores in spite of a cognitive decline [86]. Despite the potential of practice effects to obscure the variance in memory performance measured over time (e.g. in tests containing memorable information in stories or word lists), ageing effects were still present in the data, and if anything, they may lead to an underestimation of true effect sizes. Moreover, in the current study, we are interested in individual differences in changes over time. Salthouse has shown that there are no different predictors of individual differences in practice effects (other than chronological age, which is not a variable of concern in the LBC1936, owing to its being a narrow-age cohort) in longitudinal cognitive test scores from those of cognitive ageing [87]. Therefore, one may treat the various waves as a growth curve, supported by the model fit indices, even if there are temporary slight upward changes in some tests in some waves for some participants. Verbal ability showed evidence of stability with age, as expected [38, 88, 89]. Nevertheless, the observation of concomitant rises in word knowledge alongside marked declines in other cognitive measures with age, is still of empirical value.
The study results should be interpreted with several limitations in mind. Along with other cohort studies, the LBC1936 study has healthy participant bias. Lower rates of dropouts were seen among healthier individuals with a lower presence of comorbidities, and those with more education and a higher SES. We acknowledge the potential for underestimating the effects of smoking on cognitive ageing as a result of higher rates of premature mortality, particularly among long-term and/or heavy smokers. The LBC1936 study has a modest 20% attrition rate over each successive follow-up, comparable to those of other highly valuable longitudinal cohort studies with repeated assessments, such as the Swedish National Study on Aging [90] and the English Longitudinal Study of Ageing [91]. However, using FIML in our LGC analyses partly addresses the issue of attrition from dropout or death by including all available data from each time-point, not just those who completed all five waves, resulting in less biased estimates. We relied upon self-report of medical history; a limitation which has implications for potential misclassification bias and some residual confounding. As some physiological processes preceding cognitive decline may occur before older age, the influence of some health behavioural factors, such as physical activity and BMI, may be stronger from mid-life compared with later-life measures [92,93,94], leading to an underestimation of their effects. We were also unable to explore associations according to APOE e4 allele variations; low numbers in each allele group were insufficient to conduct further comparisons between e2, e3 and e4 genotypes. We recognise that our cognitive intercept at age 70 is likely to be a conflation of both intercept and some degree of slope (i.e. cognitive ageing experienced up to that point). Without knowing individuals’ mid-age (reflecting peak cognitive function) to older-age trajectories, we cannot fully address the issue of preserved differentiation vs. differed preservation, though childhood IQ functions as a good proxy measure given its stability across the lifespan [95]. Finally, as a volunteer sample, the LBC1936 represent a well-educated and generally healthy group, which might preclude the generalisation of these findings to the broader ageing population, and as such, replication in other larger samples is warranted.
The major strengths of the LBC1936 are an unusually comprehensive cognitive battery, enabling good characterisation of cognitive domains across later life, and the availability of childhood IQ scores. Studies that can account for early-life cognitive ability are rare in studies of cognitive ageing and valuable with respect to the temporal primacy of cognitive changes. Identical tests and testing location were used at five measurement points over a 12-year follow-up period, covering an age-critical window in later life for accelerated cognitive decline. Modelling latent cognitive variables reduced the influence of potential measurement error inherent in using single cognitive tests. We further improved the robustness of our results by using FDR-adjustment for multiple associations, thereby reducing the chance of type I errors, and conducting sensitivity tests for incident dementia and death. Here we have used a baseline-value prediction approach. In future analyses, bi-/multivariate growth curve modelling could look at the changes over time in predictors and their associations with cognitive ageing.
In summary, we found that APOE e4 status was the single most important predictor of longitudinal cognitive decline from age 70 to 82, when fifteen potential predictors were modelled simultaneously, despite there being many life-course correlates of cognitive level at age 70. APOE e4 allele carriers experienced significantly steeper 12-year declines across the three “fluid” domains of memory, processing speed, and visuospatial ability, and a general factor of cognitive function, than non-carriers, denoting an increasingly widening gap in cognitive functioning as individuals’ progress into older age. Our findings suggest that (1) when many other candidate predictors of cognitive ageing slope are entered en masse, their unique contributions account for relatively small proportions of variance, beyond variation in APOE e4 carrier status, (2) APOE e4 status is important for identifying those a greater risk for accelerated cognitive ageing, even among ostensibly healthy individuals.
Harada CN, Love MCN, Triebel KL. Normal cognitive aging. Clin Geriatr Med. 2012;29:737–52.
Article  Google Scholar 
Salthouse TA. Trajectories of normal cognitive aging. Psychol Aging. 2019;34:17–24.
Article  Google Scholar 
Blazer DG, Yaffe K, Karlawish J. Cognitive aging: a report from the Institute of Medicine. JAMA. 2015;313:2121–2.
Article  CAS  Google Scholar 
Hedden T, Gabrieli JD. Insights into the ageing mind: a view from cognitive neuroscience. Nat Rev Neurosci. 2004;5:87–96.
Article  CAS  Google Scholar 
Salthouse TA. When does age-related cognitive decline begin? Neurobiol Aging. 2009;30:507–51.
Article  Google Scholar 
Ghisletta P, Rabbitt P, Lunn M, Lindenberger U. Two thirds of the age-based changes in fluid and crystallized intelligence, perceptual speed, and memory in adulthood are shared. Intelligence. 2012;40:260–8.
Article  Google Scholar 
Tucker-Drob EM, Briley DA, Starr JM, Deary IJ. Structure and correlates of cognitive aging in a narrow age cohort. Psychol Aging. 2014;29:236–49.
Article  Google Scholar 
Tucker-Drob EM. Cognitive aging and dementia: a life-span perspective. Ann Rev Dev Psychol. 2019;1:177–96.
Article  Google Scholar 
Tucker-Drob EM. Neurocognitive functions and everyday functions change together in old age. Neuropsychology. 2011;25:368–77.
Article  Google Scholar 
Plassman BL, Williams JW Jr, Burke JR, Holsinger T, Benjamin S. Systematic review: factors associated with risk for and possible prevention of cognitive decline in later life. Ann Intern Med. 2010;153:182–93.
Article  Google Scholar 
Batty GD, Deary IJ, Zaninotto P. Association of cognitive function with causespecific mortality in middle and older age: follow-up of participants in the English longitudinal study of ageing. Am J Epidemiol. 2016;183:183–90.
Article  Google Scholar 
Cloutier S, Chertkow H, Kergoat MJ, Gauthier S, Belleville S. Patterns of cognitive decline prior to dementia in persons with mild cognitive impairment. J Alzheimers Dis. 2015;47:901–13.
Article  Google Scholar 
Deary IJ, Corley J, Gow AJ, Harris SE, Houlihan LM, Marioni RE, et al. Age-associated cognitive decline. Brit Med Bull. 2009;92:135–52.
Article  Google Scholar 
Gangolli VK. Recent advances in the understanding of cognitive decline among the elderly. J Geriatr Ment Health. 2016;3:36–43.
Article  Google Scholar 
Wu Z, Phyo AZ, Al-Harbi T, Woods RL, Ryan J. Distinct cognitive trajectories in late life and associated predictors and outcomes: a systematic review. J Alzheimers Dis Rep. 2020;4:459–78.
Article  Google Scholar 
Leritz EC, McGlinchey RE, Kellison I, Rudolph JL, Milberg WP. Cardiovascular disease risk factors and cognition in the elderly. Curr Cardiovasc Risk Rep. 2011;5:407–12.
Article  Google Scholar 
Farrer LA, Cupples LA, Haines JL, Hyman B, Kukull WA, Mayeux R, et al. Effects of age, sex, and ethnicity on the association between apolipoprotein E genotype and Alzheimer disease: a meta-analysis. JAMA. 1997;278:1349–56.
Article  CAS  Google Scholar 
Liu CC, Kanekiyo T, Xu H, Bu G. Apolipoprotein E and Alzheimer disease: risk, mechanisms and therapy. Nat Rev Neurol. 2013;9:106–18.
Article  CAS  Google Scholar 
Alfred T, Ben-Shlomo Y, Cooper R, Hardy R, Cooper C, Deary IJ, et al. Associations between APOE and low-density lipoprotein cholesterol genotypes and cognitive and physical capability: the HALCyon programme. Age. 2014;36:1–2.
Article  CAS  Google Scholar 
Boyle PA, Buchman AS, Wilson RS, Kelly JF, Bennett DA. The APOE ε4 allele is associated with incident mild cognitive impairment among community-dwelling older persons. Neuroepidemiology. 2010;34:43–49.
Article  Google Scholar 
Bunce D, Fratiglioni L, Small BJ, Winblad BM, Bäckman L. APOE and cognitive decline in preclinical Alzheimer disease and non-demented aging. Neurology. 2004;63:816–21.
Article  CAS  Google Scholar 
Christensen H, Batterham PJ, Mackinnon AJ, Jorm AF, Mack HA, Mather KA, et al. The association of APOE genotype and cognitive decline in interaction with risk factors in a 65–69 year old community sample. BMC Geriatr. 2008;8:14.
Article  Google Scholar 
Rawle MJ, Davis D, Bendayan R, Wong A, Kuh D, Richards M. Apolipoprotein-E (Apoe) ε4 and cognitive decline over the adult life course. Transl Psychiatry. 2018;8:1–8.
Article  CAS  Google Scholar 
Schiepers OJ, Harris SE, Gow AJ, Pattie A, Brett CE, Starr JM, et al. APOE E4 status predicts age-related cognitive decline in the ninth decade: longitudinal follow-up of the Lothian Birth Cohort 1921. Mol Psychiatry. 2012;17:315–24.
Article  CAS  Google Scholar 
Wisdom NM, Callahan JL, Hawkins KA. The effects of apolipoprotein E on non-impaired cognitive functioning: a meta-analysis. Neurobiol Aging. 2011;32:63–74.
Article  CAS  Google Scholar 
Baumgart M, Snyder HM, Carrillo MC, Fazio S, Kim H, Johns H. Summary of the evidence on modifiable risk factors for cognitive decline and dementia: a population-based perspective. Alzheimers Dement. 2015;11:718–26.
Article  Google Scholar 
Lee Y, Back JH, Kim J, Kim SH, Na DL, Cheong HK, et al. Systematic review of health behavioral risks and cognitive health in older adults. Int Psychogeriatr. 2010;22:174–87.
Article  Google Scholar 
Corley J, Cox SR, Deary IJ. Healthy cognitive ageing in the Lothian Birth Cohort studies: marginal gains not magic bullet. Psychol Med. 2018;8:187–207.
Article  Google Scholar 
Anstey KJ, von Sanden C, Salim A, O’Kearney R. Smoking as a risk factor for dementia and cognitive decline: a meta-analysis of prospective studies. Am J Epidemiol. 2007;166:367–78.
Article  Google Scholar 
Peters R, Poulter R, Warner J, Beckett N, Burch L, Bulpitt C. Smoking, dementia and cognitive decline in the elderly, a systematic review. BMC Geriatr. 2008;8:1–7.
Google Scholar 
Vňuková M, Richards M, Cadar D. How do our decisions to smoke and drink in midlife affect our cognitive performance in later life? Findings from the 1946 British Birth Cohort. J Aging Geriatr Med. 2017;1:2.
Google Scholar 
Deary IJ, Whalley LJ, Clair DS, Breen G, Leaper S, Lemmon H, et al. The influence of the ε4 allele of the apolipoprotein E gene on childhood IQ, nonverbal reasoning in old age, and lifetime cognitive change. Intelligence. 2003;31:85–92.
Article  Google Scholar 
Machado A, Barroso J, Molina Y, Nieto A, Díaz-Flores L, Westman E, et al. Proposal for a hierarchical, multidimensional, and multivariate approach to investigate cognitive aging. Neurobiol Aging. 2018;71:179–88.
Article  Google Scholar 
Sebastiani P, Andersen SL, Sweigart B, Du M, Cosentino S, Thyagarajan B, et al. Patterns of multi-domain cognitive aging in participants of the Long Life Family Study. GeroScience. 2020;42:1335–50.
Article  Google Scholar 
Zaninotto P, Batty GD, Allerhand M, Deary IJ. Cognitive function trajectories and their determinants in older people: 8 years of follow-up in the English Longitudinal Study of Ageing. J Epidemiol Community Health. 2018;72:685–94.
Article  Google Scholar 
Ritchie SJ, Tucker-Drob EM, Cox SR, Corley J, Dykiert D, Redmond P, et al. Predictors of ageing-related decline across multiple cognitive functions. Intelligence. 2016;59:115–26.
Article  Google Scholar 
Prince M, Knapp M, Guerchet M, McCrone P, Prina M, Comas-Herrera A, et al. Dementia UK: update. London: Alzheimer’s Society; 2014.
Small BJ, Dixon RA, McArdle JJ. Tracking cognition–health changes from 55 to 95 years of age. J Gerontol B Psychol Sci. 2011;66:i153–il61.
Article  Google Scholar 
Deary IJ, Gow AJ, Taylor MD, Corley J, Brett C, Wilson V, et al. The Lothian Birth Cohort 1936: a study to examine influences on cognitive ageing from age 11 to age 70 and beyond. BMC Geriatr. 2007;7:1–2.
Article  Google Scholar 
Deary IJ, Gow AJ, Pattie A, Starr JM. Cohort profile: the Lothian Birth Cohorts of 1921 and 1936. Int J Epidemiol. 2012;41:1576–84.
Article  Google Scholar 
Taylor AM, Pattie A, Deary IJ. Cohort profile update: the Lothian Birth Cohorts of 1921 and 1936. Int J Epidemiol. 2018;47:1042–1042r.
Article  Google Scholar 
Scottish Council for Research in Education. The trend of Scottish intelligence: a comparison of the 1947 and 1932 surveys of the intelligence of eleven-year-old pupils. London: University of London Press; 1949.
Wechsler D. WAIS-IIIUK administration and scoring manual. London: Psychological Corporation; 1998.
Wechsler D. WMS–IIIUK administration and scoring manual. London: Psychological Corporation; 1998.
Deary IJ, Der G, Ford G. Reaction times and intelligence differences: a population-based cohort study. Intelligence. 2001;29:389–99.
Article  Google Scholar 
Deary IJ, Simonotto E, Meyer M, Marshall A, Goddard N, Wardlaw J. The functional anatomy of inspection time: an event-related fMRI study. NeuroImage. 2004;22:1466–79.
Article  Google Scholar 
Nelson HE, Willison JR. National Adult Reading Test (NART) test manual (Part II). Windsor, England: NFER-Nelson; 1991.
Holdnack JA. WTAR: Wechsler Test of Adult Reading manual. San Antonio, TX: Psychological Corporation; 2001.
Lezak MD, Howieson DB, Loring DW, Fischer JS. Neuropsychological assessment. USA: Oxford University Press; 2004.
Folstein MF, Folstein SE, McHugh PR. “Mini-mental status”. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975;12:189–98.
Article  CAS  Google Scholar 
OPCS. Classification of Occupations 1980. London: HMSO; 1980.
Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand. 1983;67:361–70.
Article  CAS  Google Scholar 
McArdle JJ. Dynamic but structural equation modeling of repeated measures data. In: Nesselroade JR, Cattell RB, editors. Handbook of multivariate experimental psychology. Boston, MA: Springer; 1998. p. 561–614.
Cox SR, Harris MA, Ritchie SJ, Buchanan CR, Hernández MV, Corley J, et al. Three major dimensions of human brain cortical ageing in relation to cognitive decline across the eighth decade of life. Mol Psychiatry. 2021;26:2651–62.
Article  CAS  Google Scholar 
Rosseel Y. lavaan: an R package for structural equation modeling. J Stat Softw. 2012;48:1–36.
Article  Google Scholar 
Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc. 1995;57:289–300.
Google Scholar 
Lin FJ. Solving multicollinearity in the process of fitting regression model using the nested estimate procedure. Qual Quant. 2008;42:417–26.
Article  Google Scholar 
Funder DC, Ozer DJ. Evaluating effect size in psychological research: sense and nonsense. Adv Meth Pr Psychol Sci. 2019;2:156–68.
Google Scholar 
Persaud N, Coleman E, Zwolakowski D, Lauwers B, Cass D. Nonuse of bicycle helmets and risk of fatal head injury: a proportional mortality, case–control study. CMAJ. 2012;184:E921–3.
Article  Google Scholar 
Bunce D, Bielak AAM, Anstey KJ, Cherbuin N, Batterham PJ, Easteal S. APOE genotype and cognitive change in young, middle-aged, and older adults living in the community. J Gerontol A Biol Sci. 2014;69:379–86.
Article  CAS  Google Scholar 
Caselli RJ, Dueck AC, Osborne D, Sabbagh MN, Connor DJ, Ahern GL, et al. Longitudinal modeling of age-related memory decline and the APOE ε4 effect. N Eng J Med. 2009;361:255–63.
Article  CAS  Google Scholar 
Deary IJ, Whiteman MC, Starr JM, Whalley LJ, Fox HC. The impact of childhood intelligence on later life: following up the Scottish mental surveys of 1932 and 1947. J Pers Soc Psychol. 2004;86:130–47.
Article  Google Scholar 
Lutz MW, Crenshaw DG, Saunders AM, Roses AD. Genetic variation at a single locus and age of onset for Alzheimer’s disease. Alzheimers Dement. 2010;6:125–31.
Article  CAS  Google Scholar 
Smith C, Graham DI, Murray LS, Stewart J, Nicoll JA. Association of APOE e4 and cerebrovascular pathology in traumatic brain injury. J Neurol Neurosurg Psychiatry. 2006;77:363–6.
Article  CAS  Google Scholar 
Flowers SA, Rebeck GW. APOE in the normal brain. Neurobiol Dis. 2020;136:104724.
Article  CAS  Google Scholar 
Tzioras M, Davies C, Newman A, Jackson R, Spires‐Jones T. Invited Review: APOE at the interface of inflammation, neurodegeneration and pathological protein spread in Alzheimer’s disease. Neuropath Appl Neuro. 2019;45:327–46.
Article  CAS  Google Scholar 
Wilson RS, Leurgans SE, Boyle PA, Schneider JA, Bennett DA. Neurodegenerative basis of age-related cognitive decline. Neurology. 2010;75:1070–8.
Article  CAS  Google Scholar 
Yu Y, Boyle PA, Segawa E, Leurgans S, Schneider JA, Wilson RS, et al. Residual decline in cognition after adjustment for common neuropathologic conditions. Neuropsychology. 2014;29:335–43.
Article  Google Scholar 
Hassenstab J, Chasse R, Grabow P, Benzinger TL, Fagan AM, Xiong C, et al. Certified normal: Alzheimer’s disease biomarkers and normative estimates of cognitive functioning. Neurobiol Aging. 2016;43:23–33.
Article  Google Scholar 
Lyall DM, Cox SR, Lyall LM, Celis-Morales C, Cullen B, Mackay DF, et al. Association between APOE e4 and white matter hyperintensity volume, but not total brain volume or white matter integrity. Brain Imaging Behav. 2020;14:1468–76.
Article  Google Scholar 
Lange KL, Bondi MW, Salmon DP, Galasko D, Delis DC, Thomas RG, et al. Decline in verbal memory during preclinical Alzheimer’s disease: examination of the effect of APOE genotype. J Int Neuropsychol Soc. 2002;8:943–55.
Article  CAS  Google Scholar 
Sibbett RA, Russ TC, Pattie A, Starr JM, Deary IJ. Does incipient dementia explain normal cognitive decline determinants? Lothian birth cohort 1921. Psychol Aging. 2018;33:674–84.
Article  Google Scholar 
Anastasiou CA, Yannakoulia M, Kontogianni MD, Kosmidis MH, Mamalaki E, Dardiotis E, et al. Mediterranean lifestyle in relation to cognitive health: results from the HELIAD study. Nutrients. 2018;10:1557.
Article  Google Scholar 
Yannakoulia M, Kontogianni M, Scarmeas N. Cognitive health and Mediterranean diet: just diet or lifestyle pattern? Ageing Res Rev. 2015;20:74–78.
Article  Google Scholar 
Stern Y. Cognitive reserve. Neuropsychologia. 2009;47:2015–28.
Article  Google Scholar 
Cermakova P, Formanek T, Kagstrom A, Winkler P. Socioeconomic position in childhood and cognitive aging in Europe. Neurology. 2018;91:e1602–10.
Article  Google Scholar 
Seblova D, Berggren R, Lövdén M. Education and age-related decline in cognitive performance: systematic review and meta-analysis of longitudinal cohort studies. Ageing Res Rev. 2020;58:101005.
Article  CAS  Google Scholar 
Gow AJ, Johnson W, Mishra G, HALCyon Study Team, Richards M, Kuh D, et al. Is age kinder to the initially more able? Yes, and no. Intelligence. 2012;40:49–59.
Article  Google Scholar 
Salthouse TA. Does the direction and magnitude of cognitive change depend on initial level of ability? Intelligence. 2012;40:352–61.
Article  Google Scholar 
Lövdén M, Fratiglioni L, Glymour MM, Lindenberger U, Tucker-Drob EM. Education and cognitive functioning across the life span. Psychol Sci Public Interest. 2020;21:6–41.
Article  Google Scholar 
Salthouse TA. Mental exercise and mental aging: evaluating the validity of the “use it or lose it” hypothesis. Perspect Psychol Sci. 2006;1:68–87.
Article  Google Scholar 
Salthouse TA. The processing-speed theory of adult age differences in cognition. Psychol Rev. 1996;103:403–28.
Article  CAS  Google Scholar 
Salthouse TA, Ferrer-Caja E. What needs to be explained to account for age-related effects on multiple cognitive variables? Psychol Aging. 2004;18:91–110.
Article  Google Scholar 
Tam HM, Lam CL, Huang H, Wang B, Lee TM. Age-related difference in relationships between cognitive processing speed and general cognitive status. Appl Neuropsychol Adult. 2015;22:94–99.
Article  Google Scholar 
Nyberg L, Lövdén M, Riklund K, Lindenberger U, Bäckman L. Memory aging and brain maintenance. Trends Cogn Sci. 2012;16:292–305.
Article  Google Scholar 
Salthouse TA. Influence of age on practice effects in longitudinal neurocognitive change. Neuropsychology. 2010;24:563–72.
Article  Google Scholar 
Tucker-Drob EM, Salthouse TA. Individual differences in cognitive aging. In: Chamorro-Premuzic T, von Stumm S, Furnham A, editors. The Wiley-Blackwell handbook of individual differences. 1st ed. London: Wiley-Blackwell; 2011. p. 242–67.
Goh JO, An Y, Resnick SM. Differential trajectories of age-related changes in components of executive and memory processes. Psychol Aging. 2012;27:707–19.
Article  Google Scholar 
Schaie KW. Cognitive aging. In: Pew RW, Von Hemel SB, editors. Technology for adaptive aging. US: National Academies Press; 2004. p. 43–63.
Marseglia A, Fratiglioni L, Kalpouzos G, Wang R, Bäckman L, Xu W. Prediabetes and diabetes accelerate cognitive decline and predict microvascular lesions: a population-based cohort study. Alzheimers Dement. 2019;15:25–33.
Article  Google Scholar 
Frisher M, Mendonça M, Shelton N, Pikhart H, de Oliveira C, Holdsworth C. Is alcohol consumption in older adults associated with poor self-rated health? Cross-sectional and longitudinal analyses from the English Longitudinal Study of Ageing. BMC Public Health. 2015;15:1–9.
Article  Google Scholar 
Anstey KJ, Cherbuin N, Budge M, Young J. Body mass index in midlife and late‐life as a risk factor for dementia: a meta‐analysis of prospective studies. Obes Rev. 2011;12:e426–37.
Article  CAS  Google Scholar 
Carroll S, Turkheimer E. Midlife risk factors for late-life cognitive decline. Dev Rev. 2018;48:201–22.
Article  Google Scholar 
Wagner M, Grodstein F, Proust-Lima C, Samieri C. Long-term trajectories of body weight, diet, and physical activity from midlife through late life and subsequent cognitive decline in women. Am J Epidemiol. 2020;189:305–13.
Article  Google Scholar 
Deary IJ. The stability of intelligence from childhood to old age. Curr Dir Psychol Sci. 2014;23:239–45.
Article  Google Scholar 
Download references
This research was funded in whole, or in part, by the Wellcome Trust [221890/Z/20/Z]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. We gratefully acknowledge the contributions of the LBC1936 participants and members of the LBC1936 research team who collect and manage the LBC data. We also thank the Genetics Core staff at the Edinburgh Clinical Research Facility. The LBC1936 is supported by Age UK [The Disconnected Mind], the Medical Research Council [G0701120, G1001245, MR/M01311/1, MR/R024065/1] and the University of Edinburgh. SRC and IJD were additionally supported by a National Institutes of Health (NIH) research grant R01AG054628, and IJD was also supported by the Dementias Platform UK [MR/L015382/1]. SRC is supported by a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (Grant Number 221890/Z/20/Z).
These authors contributed equally: Ian J. Deary, Simon R. Cox.
Lothian Birth Cohorts, Department of Psychology, University of Edinburgh, Edinburgh, UK
Janie Corley, Sarah E. Harris, Adele M. Taylor, Paul Redmond, Tom C. Russ, Ian J. Deary & Simon R. Cox
Department of Psychology, University of Milano-Bicocca, Milan, Italy
Federica Conte
Alzheimer Scotland Dementia Research Centre, University of Edinburgh, Edinburgh, UK
Tom C. Russ
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
JC, IJD, and SRC designed the analysis. JC conducted the analysis and drafted the work. IJD conceived the study design. JC, FC, AMT, SEH, and SRC collected or analysed the cognitive and/or genetic data. PR performed data curation. All authors (JC, FC, SEH, AMT, PR, TCR, IJD, and SRC) critically revised the work and have approved the submitted version.
Correspondence to Janie Corley.
The authors declare no competing interests.
Ethical approval was obtained from the Multicentre Research Ethics Committee for Scotland (baseline, MREC/01/0/56), the Lothian Research Ethics Committee (age 70, LREC/2003/2/29), and the Scotland A Research Ethics Committee (ages 73, 76, 79, 82, 07/MRE00/58). All participants provided written informed consent.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/
Reprints and Permissions
Corley, J., Conte, F., Harris, S.E. et al. Predictors of longitudinal cognitive ageing from age 70 to 82 including APOE e4 status, early-life and lifestyle factors: the Lothian Birth Cohort 1936. Mol Psychiatry (2022). https://doi.org/10.1038/s41380-022-01900-4
Download citation
Received: 17 November 2021
Revised: 17 November 2022
Accepted: 21 November 2022
Published: 08 December 2022
DOI: https://doi.org/10.1038/s41380-022-01900-4
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Molecular Psychiatry (Mol Psychiatry) ISSN 1476-5578 (online) ISSN 1359-4184 (print)
© 2022 Springer Nature Limited

source

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button