Longitudinal episodic memory trajectories in older adults with normal cognition

Abstract Objective To determine the longitudinal trajectories and normative standards of episodic memory in older adults. Methods Participants were drawn from the cognitively normal(CN) subgroup of the population-based HELIAD cohort, a fairly representative cohort of the older Greek population. Verbal and non-verbal memory were assessed using the Greek Verbal Learning Test and Medical College of Georgia-Complex Figure Test. Baseline and longitudinal associations of memory performance with age, sex and formal education were explored with linear regression analysis and generalized estimated equations. Results A total of 1607 predominantly female (60%) individuals (73.82 ± 5.43 years), with a mean educational attainment of 8.17(±4.86) years were CN at baseline. Baseline analysis revealed a continuum of memory decline with aging and lower educational attainment. Women performed better in composite and verbal memory measures, while men performed better in non-verbal memory tasks. A subgroup of 761 participants with available assessments after 3.07(±0.82) years remained CN at follow-up. Composite memory scores yearly diminished by an additional 0.007 of a SD for each additional year of age at baseline. Regarding verbal learning, immediate free verbal recall, delayed free verbal recall and delayed cued verbal recall, an additional yearly decrease of 0.107, 0.043, 0.036 and 0.026 words were respectively recorded at follow-up, for each additional year of age at baseline. Women underwent steeper yearly decreases of 0.227 words in delayed cued verbal recall. No significant longitudinal associations emerged for immediate non-verbal memory, delayed non-verbal memory and immediate cued verbal recall. Conclusions In the present study, aging (but not educational attainment) was consistently associated with steeper verbal memory decline. Supplemental data for this article is available online at https://doi.org/10.1080/13854046.2022.2059011 .


Introduction
Episodic memory reflects one's ability to encode and recall everyday events including contextual details (e.g. spatiotemporal information and emotional associations) (Harvey, 2019). Impairment in this neurocognitive domain constitutes one of the earliest manifestations of Alzheimer's Disease (AD), as well as part of the semiology of a wide range of neuropsychological disorders (Hulme et al., 1993;van oostrom et al., 2003;Vanderploeg et al., 2001). Even in the absence of disease-related impairments, however, aging is typically associated with a significant decline in episodic memory performance (Brickman & Stern, 2009;Sander et al., 2021). Therefore, determining the normal trajectory of memory decline is important in the differentiation of typical age-related versus pathological change.
Episodic memory is generally conceptualized as a three-stage cognitive process starting with encoding (stimuli are appropriately formatted for brain storage) followed by consolidation (memories are retained for later use) and finally retrieval of new memories (Wright et al., 2009). Although different neuroanatomical correlations have been revealed for each of these cognitive stages, encoding, consolidation and retrieval present considerable neuropsychological overlap and their clinical distinction is unclear (Schapiro et al., 2019;Takehara-Nishiuchi, 2020;Zhang et al., 2021). Typically, neuropsychological tests evaluate episodic memory by assessing an individual's learning and recall of verbal or non-verbal input over both short (approximately 3-min) and long (20 to 30-min) delays (Harvey, 2019).
The California Verbal Learning Test (CVLT) is one of the most commonly employed neuropsychological tests in the evaluation of verbal learning and memory (Vlahou et al., 2013). It is considered an important tool for the identification of episodic memory impairment and AD diagnosis (Lacritz et al., 2001;McDowell et al., 2004). Vlahou et al. (2013) developed a Greek version of the CVLT, with items relevant to the Greek language and culture (GVLT) that demonstrated acceptable psychometric properties. Regression-based norms for cognitively normal (CN) individuals were provided by the authors based on cross-sectional investigations of a group of 354 CN individuals between 15 and 88 years of age.
on the other hand, the Rey-osterrieth complex figure test (RCFT) and the Medical College of Georgia complex figure test (MCG-CFT) are two similar, commonly used tests for the appraisal of non-verbal memory (as well as other cognitive functions, e.g. visuospatial perception, executive skills) (Ingram et al., 1997;Lezak, 2004). Visuospatial memory is considered an important component of non-verbal memory (Luzzi et al., 2011) and accumulating evidence suggests its potentially supporting role in the differential diagnosis of different dementia entities (Baldock et al., 2016). Although RCFT norms are available for the Greek population, the MCG-CFT has not been investigated (Tsatali et al., 2020). Published RCFT norms were based on cross-sectional data, collected from a group of 228 CN older adults (>50 years of age).
Cross-sectional approaches are generally considered adequate for the normative characterization of a defined population (o' Connor, 1990). Longitudinal designs, however, are more useful when age and time effects constitute important considerations (o' Connor, 1990), which is the case with both verbal and non-verbal memory (Tsatali et al., 2020;Vlahou et al., 2013). Time-dependent cognitive changes may vary considerably according to age, sex, level of education, and ethnic background, while it is important to characterize "normal" age-related episodic memory decline in order to differentiate it from disease-related patterns (Lee et al., 2018).
With respect to episodic memory trajectories in specific, although the published literature generally corroborates the idea that episodic memory decline is accelerated with higher chronological age in older adults, conflicting evidence exists for the longitudinal effects of sex and education (Nyberg et al., 2012;Rönnlund et al., 2005). Most notably, recent research tends to reject the traditional concept suggesting that higher educational attainment moderates the course of episodic memory (composite cognitive, as well) decline (Anstey & Christensen, 2000;Tucker-Drob et al., 2009;Vonk et al., 2020;Zahodne et al., 2011). These findings provide additional evidence of a passive cognitive reserve hypothesis with aging, according to which individuals with higher educational status perform at a higher level than less educated individuals but decline at a similar rate (Zahodne et al., 2011). To date, however, supporting evidence for the passive cognitive reserve hypothesis mainly stems from relatively well-educated samples (Vonk et al., 2020;Zahodne et al., 2011).
Based on the aforementioned considerations and literature gaps, we explored the longitudinal trajectories of verbal and non-verbal episodic memory in the CN older Greek population using the GVLT and the MCG-CFT. For this purpose, we capitalized on data from the population-based, prospective HELIAD cohort, a fairly representative cohort of the older Greek population. This generational cohort included both urban and rural dwellers who grew up during World War II and the Greek Civil War, which disrupted normal operations of many services including schools for extended periods of time, leaving children with intermittent educational experiences, at best. The majority of these individuals (>50%) were poorly educated in terms of formal schooling having only received elementary school education (Dardiotis et al., 2014;Kosmidis et al., 2018).

Methods
our aim in undertaking the present study was to determine the longitudinal trajectories of verbal and non-verbal episodic memory performance in the older Greek population. Reporting was performed according to the STRoBE recommendations (von Elm et al., 2014). Data were derived from the Hellenic Longitudinal Investigation of Aging and Diet (HELIAD) cohort. Study procedures were approved by the Institutional Ethics Review Boards of the National and Kapodistrian university of Athens and the university of Thessaly. All participants provided informed consent prior to participation.
Briefly, HELIAD is a prospective population-based study with a primary objective to explore the epidemiology of neuropsychiatric disorders in the older Greek population. Participants were randomly selected from the older (>64 years of age) registries of two Greek municipalities, Larissa and Marousi. All participants were of Greek ethnicity with Greek being their first-native language. Baseline evaluations were initiated in 2011. Follow-up assessments were conducted approximately 3 years after the initial assessment. To date, 2 nd evaluations have been completed while 3 rd visits were recently initiated (the data of 3 rd visits are insufficient at this time to be of use in the present analyses). Participants with normal cognition at baseline were considered eligible for the present analyses (individuals with MCI and dementia at baseline were excluded) irrespective of having completed a 2 nd evaluation (baseline set). Among them, those with available follow-up assessments who remained CN at 2 nd visit constituted the follow-up subset (participants with MCI and dementia at follow-up were excluded from the follow-up subset).
Complete neurological and comprehensive neuropsychological examinations took place at baseline and follow-up sessions (for a detailed description, see Kosmidis et al., 2018). Verbal and non-verbal episodic memory were assessed using the GVLT and the MCG-CFT, respectively (Lezak, 2004;Vlahou et al., 2013). Both tests were administered by trained neuropsychologists.

Neuropsychological assessment
Regarding GVLT (described in more detail in the paper of Vlahou et al., 2013), a shopping list consisting of 16 items (1 st list) belonging to 4 different semantic categories was read to each participant, who was in turn instructed to recall as many items as possible (max of 16 items). Five consecutive learning trials were provided to each examinee (max of 80 items). An interference shopping list with 16 different items (2 nd list) belonging to 4 different semantic categories (two novel and two the same as in the 1 st list) was then administered to the participants (interference trial). Subsequently, participants were instructed to recollect as many items as possible from the 1 st shopping list, first by immediate free recall (trial 7) and then by immediate cued recall (trial 8, in which semantic categories were provided to the examinees). After a 20-min delay, participants were again asked to verbally generate as many items as possible belonging to the 1 st shopping list. once again, two trials were carried out, the first assessing delayed free recall (trial 9) and the second delayed cued recall (trial 10, in which semantic categories were provided to the examinees). Finally, a recognition trial was administered: participants were read a new list of 44 words and were instructed to indicate which of them were part of the 1 st shopping list (max of 16 correct items).
With respect to the MCG-CFT, participants were asked to copy a complex figure to the best of their ability. Participants were not aware that he or she would need to redraw the figure from memory shortly after copying it (immediate non-verbal recall) and then again in about 40-45 min (delayed non-verbal recall). Figures drawn were assessed based on 18 components, each of which was scored on a 2-point scale taking into account the existence, shape and correct placement of each item (Ingram et al., 1997;Yasugi & Yamashita, 2010). All three conditions (copy, immediate and delayed non-verbal recall) were assessed on a 36-point scale.
Alternate forms of memory tests were used in order to avoid practice effects. To assess episodic memory overall, we derived a composite memory score based on the following variables: total sum of correct items from trials 1-5 (GVLT, verbal learning), immediate free verbal recall (GVLT, trial 7), delayed free verbal recall (GVLT, trial 9), immediate non-verbal recall (MCG-CFT) and delayed non-verbal recall (MCG-CFT) (Bougea et al., 2019). To accomplish this, we first converted both baseline and follow-up raw scores of each variable into z-scores using mean and standard deviation values of the CN individuals at baseline. Individual z-scores were then averaged to generate a composite memory score.

Diagnostic procedures
Potential diagnoses of dementia and mild cognitive impairment (MCI) were determined during expert consensus meetings involving a consortium of senior neurologists (E.D., G.M.H., P. S., and N.S.) and neuropsychologists (M.H.K.) as well as the researchers who administered the tests. For a detailed description of the diagnostic procedures please refer to Kosmidis et al. (2018). In brief, particular focus was placed on identifying potential comorbidities that could affect cognitive performance through screening the participants for depression, anxiety, essential tremor, behavioral symptoms, Parkinson disease, dementia with Lewy bodies (DLB), as well as personal history of cerebrovascular disease accounting for the onset or deterioration of cognitive decline. The Diagnostic and Statistical Manual of Mental Disorders-IV-text revision criteria (American Phychiatric Association, 2000) and the National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer Disease and Related Disorders Association criteria (McKhann et al., 1984), were implemented in the diagnosis of dementia and probable/possible AD, respectively. Petersen criteria were implemented in the identification of MCI and its subtypes (Petersen, 2004).

Outcome measures and statistical analysis
The longitudinal trajectories of composite memory scores, based on the normalized test scores, were the primary outcomes of interest. The secondary investigation set consisted of the following individual test raw score trajectories: verbal learning, immediate and delayed free verbal recall, immediate and delayed non-verbal recall, as well as immediate and delayed cued verbal recall. Longitudinal investigations were performed using the subset of individuals with available follow-up evaluations, who remained CN at 2 nd visit (with no diagnosis of MCI or dementia either at baseline or at follow-up).
A preliminary linear regression model was performed using the baseline set of CN participants to explore baseline associations of age, sex and formal education with episodic memory measurements. In the context of the analytical approach, sex was treated as a dichotomous variable, whereas education and age were treated as scale parameters. Based on the results of the preliminary linear regression, a descriptive approach was used to provide age, sex, and formal education stratified normative memory data for the CN Greek older adults (mean ± SD, deciles). For descriptive purposes, age and education were treated as ordinal parameters. Participants were divided into 5-year strata, while those >80 years were clustered together (to avoid the formation of disproportionately smaller older age groups). Regarding education, three categories were formed according to the fundamental landmarks of the Greek educational system at the time our sample was of school age, i.e. primary education (<7 years of education), secondary education (7-12 years of education) and tertiary education (>12 years of education). Today, compulsory education in Greece lasts for 9 years, typically extending from age 6 to age 15: 6 years of elementary school and 3 years of junior high school. In the past, however, a minimum of 6 years of formal schooling was required. The typical educational attainment of Greece's population stratified by age compared to other countries of the European union can be accessed through Eurostat at https://ec.europa.eu/eurostat/databrowser/view/edat_lfs_9903/ default/table?lang=en The potential effects of age, sex and formal education on the rates of episodic memory decline over time were explored using generalized estimating equations (GEE) analysis (Hanley et al., 2003). The follow-up subset was utilized for this purpose. GEE models account for the correlation of repeated measurements in the same individual. Each participant's baseline and follow-up evaluations were treated as a cluster. Conventionally, exchangeable (compound symmetry) correlation structures were used as working covariance matrices. Consecutive GEE models were explored using composite and individual memory measurements as the dependent variables (according to the 0.05 and 0.007 significance thresholds, correspondingly). Each GEE model was adjusted for sex, age, years of education and time from baseline visit, as well as sex*time, age*time and years of education*time interactions.
All statistical analyses were performed using the IBM SPSS Statistics Software Version 25 (Chicago, IL, uSA). The conventional threshold of 0.05 was set for the determination of statistical significance for composite memory scores. According to the Bonferroni correction, however, a stricter cut-off of 0.007 (7 comparisons) was implemented in the analyses of the individual test raw scores (secondary evaluations) to exclude any trivial associations due to multiple testing.

Participant characteristics and missing data
The prospective HELIAD cohort consisted of 1984 participants with baseline evaluations. Among them, there were 103 individuals with dementia (79 with AD, 9 with vascular dementia, 8 with dementia with Lewy bodies or Parkinson's disease dementia and 7 with less common entities), 243 with MCI and 4 with an inconclusive cognitive diagnosis. Therefore, 1607 individuals who were CN at baseline were included in the present study (baseline set). A subgroup of 959 individuals had available follow-up assessments at the time of the present analysis. Among them, there were 34 individuals who developed dementia at follow-up (29 AD, 3 vascular dementia and 2 dementia with Lewy bodies or Parkinson's disease dementia), 4 participants with an inconclusive cognitive diagnosis and 160 who were diagnosed with MCI, whereas 761 participants remained CN (follow-up set). Baseline demographic characteristics and episodic memory measurements for both sets are provided in Table 1.
The mean age of the baseline set (N = 1607) was 73.31 years (±5.19; range: 65.03-91.46 years). Participants were predominantly women (60%) and had a mean educational attainment of 8.17 years (±4.86; range: 0-21 years). The majority of the participants (N = 924, 57.5%) were of low educational attainment, having completed ≤6 years of formal schooling (elementary level), while the female to male ratio was higher in this lowest educational cluster and progressively decreased in the higher educational classes: F/M ratio= 1.86 in the lowest educational cluster (i.e. ≤6 years of formal schooling), 1.36 in the middle educational group (i.e. 7-12 years of education) and 0.92 in the highest educational class (i.e. >12 years of formal schooling). Detailed information about the participants of the lower educational cluster, i.e. those with no formal education (mostly illiterate, with a very small percentage having acquired minimal reading skills informally) or with elementary level (1-6 years) education, are provided elsewhere (Mandyla et al., 2021). Those with available follow-up assessments, who remained CN at 2 nd visit (N = 761), had a mean age of 72.46 years (±4.70; range: 65.03-91.46 years), a mean education of 8.74 years (±4.86; range: 0-21 years) while the male to female ratio was similar to that of the baseline set. The mean time that intervened between initial and follow-up evaluations was 3.07 years (±0.82, range: 1.16-7.26 years). CN participants at baseline without follow-up assessments (N = 648) were older and less educated than those who remained CN at 2 nd visit and performed more poorly in every episodic memory measure (which may at least in part reflect by the potential development of cognitive disorders in a fraction of these individuals, lacking follow-up assessments) (Table 1). Table 2 contains the normative standards for episodic memory, stratified by sex, age, and education in the baseline CN set of participants (N = 1607). A more detailed presentation according to the normative distribution of episodic memory scores in deciles is provided in the supplementary material (Supplementary Tables S1-S8). overall, regarding both men and women, a continuum of episodic memory decline was observed with higher chronological age and lower educational attainment. Female sex was associated with higher composite memory and verbal memory scores, whereas male sex was associated with higher scores on non-verbal memory tests. Baseline regression-norms are provided in detail in Table 3. A confirmatory linear regression analysis was performed to account for the potential effect of premorbid undergoing neurodegenerative alterations. Those that converted to MCI (N = 160) or dementia (N = 34) at follow-up were excluded from this supplementary analysis. The estimated measures of association and their precisions were similar to those provided by the total baseline set (Supplementary Table S9).

Longitudinal associations and standardized trajectories
With respect to composite memory, only age at baseline affected the rates of episodic memory decline significantly over time (Table 4). During each year of follow-up, composite memory scores diminished by an additional 0.007 (p<.001) of a SD, for each additional year of age at baseline (irrespective of sex and educational status). A schematic demonstration of our findings is provided in Figure 1.
Verbal learning, immediate and delayed free verbal recall adhered to the same pattern of longitudinal associations as composite memory scores, that is age at baseline was the only parameter associated with the rates of decline. Particularly, regarding verbal learning, an additional yearly decrease of 0.107 (p= .001) words was recorded at follow-up, for each additional year of chronological age at baseline. With respect to immediate and delayed free verbal recall, participants exhibited increasing rates of episodic memory decline augmented by 0.043 and 0.036 words per additional year of chronological age at baseline (p< .001).
Delayed cued verbal recall was the sole episodic memory variable longitudinally affected by both sex and age at baseline. Specifically, women underwent a steeper yearly decrease by 0.227 words more than men in delayed cued verbal recall totals (p< .001). Additionally, each additional year of age at baseline conferred an increase of 0.026 words to decline rates (p< .001). Finally, with respect to immediate and delayed non-verbal memory as well as immediate cued verbal recall, no significant associations emerged, according to our statistical approach.

Discussion
In the present study, we calculated the normative trajectories of verbal and non-verbal episodic memory in a large older cohort of Greek individuals, while accounting for the effects of age, sex, and formal education on longitudinal variations. Consistent Note: individual episodic memory measures correspond to raw scores while the composite memory index corresponds to a standardized z-score; episodic memory measurements were available for ≥1510 and ≥727 participants in the baseline set and follow-up subset, respectively (slightly heterogeneous subsets of participants had missing data per parameter); demographic data were available for the total baseline set and follow-up subset; bold denotes statistically significant differences between the follow-up set and Cn participants at baseline without follow-up assessments; Cn: cognitively normal; n: total number of participants; M/F: male/female. Del. non-verbal rec.
13.36 ± 6.6 11.33 ± 6.6 9.44 ± 6.2 9.00 ± 5.9 18 with the cognitive reserve hypothesis (Stern, 2009), our baseline findings were indicative of a positive relationship between episodic memory performance and educational attainment, as well as an inverse association of episodic memory performance with increasing age. We found the well-established pattern of a male advantage on non-verbal memory tasks and a female advantage on verbal as well as composite memory indices (Herlitz et al., 1999;Lewin et al., 2001). Regarding the longitudinal investigations, higher chronological age was consistently related to increasingly steeper rates of verbal memory decline. Moreover, despite the finding of a female advantage regarding delayed cued verbal recall, there was a male advantage with slower rates of decline over time. In contrast, with respect to non-verbal memory, longitudinal trajectories were not affected by age, sex or formal education. The neurobiology of learning and memory has been found to differ in men and women (Andreano & Cahill, 2009). Previous research has shown sex disparities in episodic memory similar to those reported in the present investigation. Women appear to outperform men on verbal memory tasks, whereas men outperform women on non-verbal memory tests (Herlitz et al., 1999;Lewin et al., 2001). A moderate overall advantage favoring women has also been found in episodic memory tasks involving both verbal and non-verbal processing (composite assessments) (Andreano & Cahill, 2009;Herlitz et al., 1999;Lewin et al., 2001). Sex differences among CN individuals are complex and likely influenced by a great number of factors including genetic (e.g. APoE), psychosocial, cultural, behavioral (exercise and diet), and most notably biological parameters (e.g. sex hormones), among others (DeCarli et al., 2005;Liampas et al., 2021;Miller & Halpern, 2014). Specifically, testosterone levels have been suggested to at least partially account for the superior performance of men on visuospatial tasks, while oestrogens are considered to contribute to the higher overall female performance on verbal learning and memory tasks (Hamson et al., 2016;Hines et al., 2003;Silverman et al., 1999). Regarding longitudinal sex differences, Note: regarding sex, male sex was set as the reference category (female sex was coded as 2 and male sex as 1); memory measurements were available for ≥1510 participants per parameter (slightly heterogeneous subsets of participants had missing data per parameter); n: total number of participants. .185 Note: The gee model was estimated using exchangeable covariance structure and was adjusted for age, sex, years of education, time from baseline, as well as age, sex and years of education by time interactions: for brevity purposes, only time effects (not main effects) are presented; The threshold of statistical significance was set at 0.05 for the composite episodic memory analysis and at 0.007 for the seven individual episodic memory investigations; Both baseline and follow-up memory measurements were available for ≥681 participants per parameter (slightly heterogeneous subsets of participants had missing data per parameter); For demonstrative purposes, the mean change of every episodic memory measure (follow-up -baseline performance) was estimated using paired samples t-test; bold denotes significant differences; regarding sex, male sex was set as the reference category; n: total number of participants; Ci: confidence interval. however, while most studies confirm that women outperform men on baseline composite and verbal episodic memory assessments (as well as reproduce baseline associations of verbal and non-verbal memory with sex), they consistently report similar rates of episodic memory decline in men and women (Lamar et al., 2003;Levine et al., 2021). The aforementioned pattern of associations could reflect the post-menopausal and/or "post-andropausal" loss of the innate biological (hormone-related) advantage of women and men (ending up in parallel "aging" curves) (Maki, 2015;Matsumoto, 2002). An increasing body of evidence additionally suggests that "aging" episodic memory trajectories are not uniform, even in the absence of neurocognitive pathology (Lee et al., 2018). Normal trajectories usually fall under one of the two main following categories: stable (or even improving) and (more or less rapidly) declining episodic memory function (Australian Imaging and Biomarkers and Lifestyle (AIBL) Research Group, 2015; Lee et al., 2018). Decliners are usually of non-white ethnic background and present a higher frequency of the ε4 allele at the APoE gene (Australian Imaging and Biomarkers and Lifestyle (AIBL) Research Group, 2015; Lee et al., 2018;Zahodne et al., 2015). Intriguingly, visual long-term memory in particular has a massive storage capacity and may be improved by repeated remote visual encounters (Brady et al., 2008;Schurgin & Flombaum, 2018). These considerations could explain the relative stability of non-verbal memory trajectories in our sample.
The present study also indicated that formal education was related to episodic memory at baseline, but not with changes over time. This finding provides additional evidence for the hypothesis of "passive cognitive reserve" with aging, according to which individuals with higher educational status perform at a higher level than less educated individuals but decline at a similar rate. Similar findings were reported in the context of the Canadian Victoria and uS ACTIVE longitudinal studies (Tucker-Drob et al., 2009;Zahodne et al., 2011). Although conflicting evidence generally exists, recent studies with strong methodological designs increasingly support the passive cognitive reserve hypothesis, which, apart from global measures of cognition appears to extend to episodic memory performance, in particular (Tucker-Drob et al., 2009;Vonk et al., 2020;Zahodne et al., 2011). our results confirmed several associations previously shown in the Greek population (Tsatali et al., 2020;Vlahou et al., 2013). Yet, we have generated a number of new findings that were not explored by previous researchers using cross-sectional approaches (as well as by our baseline regression-based analysis). Capitalizing on data from both baseline and follow-up investigations, we created an episodic memory "chart," that predicted normative age, sex and formal schooling stratified episodic memory data (composite and individual scores) and longitudinal trajectories for the older Greek population. Although baseline regression-based norms were generated using linear regression analysis, a descriptive approach was additionally fostered to account for the potential inability of the linear model to capture non-linear related variations at baseline.
Based on a random sampling process from two rosters of older people (one from a provincial-urban and one from a metropolitan Greek community), our findings can be assumed to have satisfactory external validity for the older Greek population (>64 years of age, extrapolation of our conclusions in the non-elderly population would be inappropriate). Moreover, the educational features of our population were appropriate to fill several gaps in the literature, originating from the extensive previous evaluations focusing on relatively well-educated samples (Zahodne et al., 2011). In specific, all educational strata (from uneducated to highly educated individuals) were fully represented in our cohort, allowing the extrapolation of our inferences to the full range of education attainment. The exclusive focus on the older population, along with the prospective design and large sample size of the HELIAD cohort empowered our analyses relative to previous studies that have attempted to provide normative data for the Greek population recruiting younger participants and smaller samples, while implementing cross-sectional approaches. Moreover, the full neurological and comprehensive neuropsychological evaluation, as well as the multidisciplinary diagnostic consensus meetings, ensured an accurate clinical diagnostic categorization of the participants.
However, it is appropriate to point out that our study additionally presents a number of limitations. First, the diagnostic process could have been supported by the inclusion of biological and imaging biomarkers; therefore, despite the expert consensus clinical diagnosis, the presence of misclassification bias cannot be excluded. Moreover, although the random selection process ensured a satisfying generalizability for our results in community-dwelling older adults, non-response bias (participation bias) cannot be ruled out with total certainty, as a non-trivial percentage of the original sample was not available to follow-up assessment. Furthermore, our study focused on three demographic parameters, namely age, sex and formal education. Therefore, the latent impact of unaccounted confounders on the investigated associations could not be eliminated. of note, our sample composed of individuals with vastly heterogeneous sets of comorbidities and lifestyle indicators that could potentially affect cognitive performance and trajectories (Dardiotis et al., 2014). These parameters were not individually addressed in the present report but rather integrated in the cross-sectional and longitudinal investigations of individuals with normal cognition. Finally, the longitudinal investigations were based on only two (baseline and follow-up) assessments, restricting the power of our analyses to the investigation of linear patterns of change over time.

Conclusions
In the present study, higher chronological age was consistently associated with steeper longitudinal verbal memory decline, but not visuospatial memory changes. on the other hand, formal education was not found to exert a significant effect on longitudinal verbal or non-verbal episodic memory changes. Women demonstrated a steeper decrease of delayed cued verbal recall than men over time, while no other significant longitudinal associations with sex emerged. overall, our findings are in line with those provided by most large, well-designed prospective cohorts involving participants with different ethnic backgrounds. It would be interesting if future research explored these clinical associations in a neuropathological level, assessing if the presumably accelerated accumulation of neurodegenerative alterations with aging remains unaltered by each individual's "educational reserve."