Participants

700 participants were recruited in the PREVENT–Dementia program, a multi-site, prospective longitudinal study investigating the origins and early diagnosis of dementia in mid-life at-risk individuals [8]. Cognitive, clinical and lifestyle assessments were carried out in the five study sites: Imperial College London, the University of Edinburgh, the University of Cambridge, the University of Oxford and Trinity College Dublin (See Supporting Information [SI]; SFigure 1). Participants were aged between 40 and 59 years and were cognitively normal at the time of recruitment, as determined during a thorough clinical examination. Exclusion criteria for the study were a diagnosis of MCI or dementia and known MRI contraindications. The recruitment target was 50% with, and 50% without parental dementia family history. More details on the study population can be found in Ritchie and Ritchie [9] and Ritchie et al. [40]. Individuals with incomplete cognitive (N = 31) or clinical (N = 9) data were excluded (See SI; SFigure 1 and STable 1). The study reports the wave 1 (baseline) testing data, which were completed at the time of manuscript preparation. Wave 2 and 3 of testing are currently ongoing.

Standard protocol approvals, registrations, and patient consents

The study was approved by the London-Camberwell St Giles National Health Service Ethics Committee (REC reference: 12/LO/1023), by the Trinity College Dublin School of Psychology Research Ethics Committee (SPREC022021–010), and the St James’s Hospital/Tallaght University Hospital Joint Research Ethics Committee, all of which operate according to the Helsinki Declaration of 1975 (and as revised in 1983). All participants provided written informed consent.

APOE genotyping

In brief, genomic DNA was isolated from blood samples and APOE genotyping was performed. All members of the research and clinical teams were blind to the result of APOE genotyping. In this study, APOE4 risk was determined by ≥ 1 APOE4 allele. 264/700 carried ≥ 1 APOE4 allele (See Table 1). For further details, see Ritchie et al. [41].

Table 1 Participant characteristicsBiological sex

While the dichotomies of sex and gender are no longer considered to be sharply discrete, in this study ‘sex’ was defined as an individual’s natal or biological sex, and was self-reported in the Brain Injury Screening Questionnaire.

Menopausal status

Self-reported menopausal status was determined from the pregnancy and menstruation survey administered during the clinical assessments, particularly the answer to the question: “Are you postmenopausal?”. ‘Yes’ were categorized as postmenopausal, ‘No’ as premenopausal. Of the 433 female participants, 149 (34.41%) were postmenopausal, 233 were premenopausal. 51 participants who answered ‘don’t know’ were excluded from further analyses. As this prospective longitudinal study commenced in 2014, it was not designed to include detailed assessments of menopausal status.

Clinical and lifestyle-bases assessments

Blood pressure was measured after five minutes of supine rest. Blood samples were collected from overnight fasted participants and immediately analysed for standard biochemistry and haematology measures at local laboratories. Hypertension and hyperlipidemia were analyzed as binary variables. Hypertension was defined as an average diastolic blood pressure ≥ 90 mmHg, systolic blood pressure ≥ 140 mmHg, or a positive history of hypertension as reported during the medical history interview. Hyperlipidemia was defined as total cholesterol > 6.5 mmol/L or a positive history of hyperlipidemia reported in the medical history interview. Body mass index (BMI) was analyzed as a continuous variable, calculated by dividing weight (kg) by height (m2).

Sleep quality was measured using the Pittsburgh Sleep Quality Index (PSQI) [42]. The PSQI score ranges from 0 to 21, whereby higher scores indicate poorer sleep quality. ‘Poor’ sleep was binarized at a cut-off of PSQI score > 5 [10]. Smoking status and alcohol intake were assessed through a lifestyle interview. Participants were first asked if they were nonsmokers/nondrinkers, ex-smokers/ex-drinkers, or current smokers/drinkers. Smoking status was binarized if they were current smokers. Ex-drinkers and current drinkers were asked to estimate the number of glasses of wine, beer, and stronger alcohol consumed per week, and the total number of units per day/week was calculated. ‘High’ alcohol intake was defined as consuming more than 21 units per week.

Cognitive reserve contributors

Lifetime education was measured by the total number of years each participant had engaged in formal schooling, with reported values ranging from 0 to 38 years. The Lifetime of Experiences Questionnaire (LEQ) [43] was used to measure engagement in a broad range of lifestyle activities across three distinct stages of life: young adulthood (13–29 years), mid-life (30–64 years), and late life (65 years onwards), and only the mid-life activities were examined. For each life-stage, the LEQ provides two sub-scores that capture (a) “specific” activities – the primary activity undertaken in that stage, i.e., in mid-life this constitutes occupational attainment – and, (b) “non-specific” activities – engagement in physical, social and intellectual activities, in each stage. The LEQ comprises a standardized scoring approach, as follows. The mid-life stimulating activities (i.e., ‘non-specific score’) were assessed by the frequency of engagement in 7 physically, socially and intellectually stimulating activities, scored on a 6-point Likert scale of frequency (never, less than monthly, monthly, fortnightly, weekly, daily). Scores range from 0 to 35, with higher scores reflecting more frequent engagement in such activities. The items included in the scale are socializing with family or friends, practicing a musical instrument, practicing an artistic pastime, engagement in physical activity that is mildly, moderately, or vigorously energetic, reading, practicing a second language and travel. The travel item asks participants if they have visited any of a list of continents between the ages of 30–54. Responses were scored on a 6-point scale as follows: none, 1–2 regions, 3–4 regions, 5 regions, 6 regions, 7 regions.

The mid-life occupational attainment (i.e., ‘specific score’) is comprised of two sub-scores that measure (a) occupational complexity and (b) managerial responsibility. For the first, participants were asked to record their primary occupation in each 5-year interval from age 30 to age at assessment. Each reported occupation was scored on a scale of 0–9, according to the International Standard Classification of Occupations (ISCO 08) guidelines (https://www.ilo.org/public/english/bureau/stat/isco/isco08/) and relating to the skill level associated with occupations, where managers score 1, professionals 2, technicians and associate professionals score 3, and so on. Participant scores were inverted and summed. The second sub-score measured the managerial responsibility associated with reported occupations. The managerial complexity score was based on participants’ responses to the LEQ question: “Did any of your jobs require you to be in charge of or responsible for other people? If yes, indicate job title, number of years in position, and an estimate of the number of people you were in charge of.” Participants were instructed to provide information on up to four occupations where they had managerial responsibilities. If participants indicated that they were employed in a managerial capacity, the number of people that they oversaw in four of their reported occupations was documented. Managerial responsibility was scored as follows: 0 people = 8, 1–5 people = 16, 5–10 people = 24 and 10 + people = 32. The highest score across occupations was recorded as the managerial responsibility sub-score, thus capturing the maximum level of leadership responsibility achieved during the participant’s midlife career. Occupational attainment was derived by summing the occupational complexity and managerial sub-scores and multiplying them by a normalization factor of 0.25, to ensure that mid-life specific and non-specific scores have comparable mean values [43].

Cognitive assessments

Cognitive function was assessed with the Cognito neuropsychological battery [44], and the Visual Short-Term Memory Binding task (VSTMBT) [45], yielding 13 summary variables. The Cognito battery examines information processing across a wide range of cognitive functions in adults of all ages and is not restricted to those functions usually implicated in dementia detection in the elderly. It tests several aspects of cognition, including attention (task: visual attention), memory (tasks: narrative recall, description recall, implicit memory, name-face association, working memory), language (tasks: phoneme comprehension, verbal fluency) and visuospatial abilities (task: geometric figure recognition) [46, 44]. 11 summary variables from the Cognito battery capturing the above functions [46, 44] were used [see SI]. The VSTMBT [45] is a computer-based task that assesses visual short-term memory binding of single features, e.g., complex shape or colour combinations, or feature conjunctions, e.g., shape and colour combinations. The two summary variables used from the VSTMBT were the percentage of correctly recognized items from the two conditions (see SI).

Cognitive analyses

Based on our previously published method [47], rotated principal component analysis clustered the above-mentioned multiple cognitive measures (13 summary variables) into related cognitive domains (see SI). Parallel analysis and plotted scree plots determined the number of components that best represented the original 13 cognitive measures. The eigenvalues of the first three components were larger than 95th percentile of the randomly generated eigenvalues (See SI; SFigure 2 A), thus showing that the three components (C) solution best represented the data. The three components were then rotated to be uncorrelated with each other, and could cumulatively explain a total of 40% percentage of the variance (C1 = 16%, C2 = 12%, C3 = 11%) (SFigure 2B).

Statistical approach

The statistical analyses were conducted with the Statistical Package for Social Sciences (SPSS V.27) and R software (https://www.R-project.org/). Outlier assessment was summarized in the SI (SFigure 3). The normality of the data was assessed by combining the visualization of a quantile-quantile plot and the Shapiro–Wilk test. Demographic and clinical information of the study cohort were analysed across sexes using chi-square (χ2 tests) for categorical variables (Race, hypertension, hyperlipidemia, poor sleep, current smoker, high alcohol intake, APOE4), and Mann-Whitney U tests or independent samples t-tests for continuous variables (Mann-Whitney U test: age, BMI, years of education, stimulating activities, occupational attainment, and short-term memory binding; independent samples t-test: episodic and relational memory, and multisensory processing), depending on whether they met the assumption of normality in this cohort. Multicollinearity assessment for the three reserve contributors showed no significant collinearity (SI, STable 4).

Independent hierarchical regression models investigated the effects of reserve contributors (education, stimulating activities, and occupational attainment) on each cognitive domain and the moderating role of sex and APOE4. First the models examined the main effects of reserve contributors and sex on cognitive domains, then the 2-way interaction term of contributor × sex was added to examine the moderation effect of sex, with age and the other two contributors included as covariates. For any observed significant 2-way interactions, the moderating role of APOE4 was tested with 3-way interactions of APOE4 × sex × reserve contributor on cognition, with age and the other two contributors included as covariates. The effect of lifetime education was controlled for when considering the impact of stimulating activities and occupational attainment on cognition. Further analyses controlling, in addition to age, for other potential confounds (i.e., BMI, race, hypertension, hyperlipidemia, sleep, alcohol, smoking status) corroborated the results (see SI, STables 5–7). Simple slope analyses tested the significance of the slopes of the regression lines for any significant interactions. The Bonferroni correction was used to correct for multiple comparisons in the analyses of main effects of reserve contributors and sex on cognition. We followed up significant main effects that were in line with previous literature by performing one three-way interaction model, between sex × occupational attainment × APOE4, that directly investigated the study question.

Data availability

Data are available to access through a data request on the study website (www.preventdementia.co.uk); the ADDI platform (DOI: https://doi.org/10.34688/PREVENTMAIN_BASELINE_700V1); Dementia Platforms UK; and the Global Alzheimer’s Association Network.