Secular Trends in Head Size and Cerebral Volumes In the Framingham Heart Study for Birth Years 1902–1985

Background Recent data suggest that dementia incidence is declining. We investigated whether similar secular trends consisting of increasing size of brain structures and improving memory performance could be simultaneously occurring as a possible explanation. Method The Framingham Heart Study is a 3 generation, longitudinal study that includes cognitive assessment and medical surveillance. This study cohort consisted of 4,506 unique, non-demented, stroke free, individuals with brain MRI, cognitive assessment, and demographic information spanning dates of birth from 1902 to 1985. Outcomes consisted of height, MRI, and memory measures. Covariates included age at MRI, sex, decade of birth, and all interactions. Models with neuropsychological outcomes also included educational achievement as a covariate. Results Height and intracranial (TCV), hippocampus and cortical gray matter volumes were significantly larger, and memory performance significantly better, with advancing decades of birth after adjusting for age, sex, and interactions. Sensitivity analysis using progressively restricted age-ranges to reduce the association between age and decade of birth, confirmed the findings. Mediation analysis showed that hippocampal volume mediated approximately 5–7% of the effect of decade of birth on logical memory performance. Discussion These findings indicate improvement in brain health and memory performance with advancing decades of birth. Although brain structures are under substantial genetic influence, we conclude that improved early life environmental influences over ensuing decades likely explain these results. We hypothesize that these secular improvements are consistent with declining dementia incidence in this cohort potentially through a mechanism of increased brain reserve.


Introduction
Height measurement in inches was obtained at the rst FHS visit for each participant and was made at an average age of 35, 36 and 41 years across the 3 cohorts, and 37.7 ± 8.1 years of age on average, avoiding the impact of later life skeletal degeneration on the measure.

Brain Measures
Quanti cations of MRI measures were obtained from high resolution 3D T1 weighted MP RAGE images using either a Siemens Magneton Expert or Siemens Avanto machine to obtain intracranial, cortical gray matter, hippocampal volume, and cortical thickness measures. Quanti cation included automatic removal of non-brain elements from the 3D T1 image volume using a robust and accurate method relatively insensitive to machine type or eld strength 15 . For this study, intracranial volume was de ned as the cranial vault above the tentorium. Image intensity correction was used to remove B1 inhomogeneity effects 16 and segmentation of the image into 3 tissue classes used an algorithm optimized to improve precision of tissue segmentation (gray, white, CSF 16, 17 ).
Additionally, hippocampal analyses were performed using an atlas based diffeomorphic approach 18 with the minor modi cation of label re nement. Cortical thicknesses were measured using the DireCT method 19 . Cortical surface area was calculated using 3D cortical surface mesh reconstructions employing marching cubes 20 , a standard algorithm in computer graphics that has been used extensively in brain image analyses 21 . To estimate gray matter surface area in a region of interest in MRI native space, we compute the area of the polygonal facets lying within the ROI mask.
Cortical parcellation used the Desikan-Killiany-Tourville Atlas 22 where regional measures were calculated by back transformation of the atlas into segmented image native space. A voting scheme is used to assure precise labelling of each region after interpolation of the atlas into native space. Using this approach, we de ned four measures for study: 1) Intracranial volume (TCV; in cubic centimeters; cc), 2) Total cortical gray matter volume (cc), 3) Hippocampal volume (cc), 4) Surface area in squared centimeters (cm 2 ) and 5) Average cortical thickness in millimeters (mm). To correct for scanner effect, MRI measures were corrected using NeuroComBat, a robust method for reducing machine related differences in MRI data [23][24][25] (See Supplemental Materials for details).

Neuropsychological outcomes
Neuropsychological assessment consisted of a battery of tests of verbal learning and memory (Wechsler Memory Scale-III [WMS-III] Logical Memory and Paired Associates), visual memory (WMS-III Visual Reproduction), abstract reasoning (Wechsler Adult Intelligence Scale-III [WAIS-III] Similarities, visuospatial skill (Hooper VOT), language (Boston Naming Test) and executive functioning (Trails A and B) as previously described 26 . Given that verbal memory performance is both sensitive to brain aging [27][28][29] and predicts future clinically relevant cognitive impairment 30 , analyses for this study focused on verbal memory variables of immediate, delayed and the sum of both immediate and delayed recall.
Genetics, Vascular risk factors and education APOE genotype was determined with a standard TaqMan assay; APOE-e4 carriers were de ned as those having at least one e4 allele. The Framingham Stroke Risk Pro le (FSRP) score is a validated, widely-used composite measure of vascular risk factors that predicts the 10-year probability of a stroke 31 . Systolic blood pressure (SBP, mmHg) was taken as the average of the Framingham clinic physician's two measurements. Educational level was de ned as the obtainment of a college degree (yes or no).

Height and Head Size
To understand the relationship of decade of birth with body height and cranial volume, we performed univariate associations between mid-life height, intracranial volume, and decade of birth, strati ed by sex at birth.
In addition, because there are striking sex-differences in both these features, we investigated possible sex-differences in these relationships using multivariate analyses to test the interaction between mid-life height, intracranial volume, decade of birth and sex at birth. Given that height and cranial volume are correlated, we also performed multivariate analysis to investigate if cranial volume differed by decade of birth adjusting for height.

Brain Measures
To understand the relationship of decade of birth with brain measures, we performed multi-variate linear regression analyses of TCV, cortical gray matter, cortical thickness, and hippocampal volumes with decade of birth including sex at birth, age at MRI, and all possible interactions as covariates.
While many brain aging studies correct regional brain volumes for head size to reduce variance, we recognized that any secular differences in head size would in uence our results. Consequently, we examined each brain region independently without adjusting for intracranial volume to avoid confounding by secular differences in intracranial volume (TCV). years in 3 ten-year epochs. A second analysis further limited the age range at MRI to 55-65 years of age.

Contribution to Education
Given the known, strong relationship between educational achievement and memory task performance, we rst used a logistic regression to examine the association between achieved college education with decade of birth including sex at birth, and age at MRI and all possible interactions as covariates.

Contribution to Logical Memory Performance
We then examined the impact of decade of birth on verbal memory performance (Logical Memory, immediate recall, delayed recall, and the sum of both) using multi-variate linear regression analyses with decade of birth, sex at birth, age at MRI, educational achievement, and all possible interactions as covariates.

Neuropsychological Association with Brain measures
We further investigated the effect of brain measures on verbal memory performance using linear regressions with decade of birth, sex at birth, age at MRI, educational achievement, and all possible interactions as covariates.

Hippocampal Volume Mediating Effect
Given the likelihood that brain structure would increase, and verbal memory performance improve, we examined whether hippocampal volume might mediate the relationship between decade of birth and verbal memory performance. Causal mediation analysis was used to test the mediation effect of hippocampal volume on the relationship between decade of birth and memory performance adjusting for sex at birth, age at MRI, educational achievement, and all possible interactions as covariates using 5,000 bootstrap iterations for estimation.
All MRI measures and neuropsychological test performances were converted to Z-score values for relative comparison of results across measures. Analyses were performed using R version 4.1.2. Mediation analyses used the R Mediation package which is based on methods of Imai et al 31 .

Sample Characteristics
The nal cohort sample is summarized in Table 1 and consisted of 4506 individuals, of which 54% were female. The average age at MRI was 56.8 ± 12.6 years but ranged from 25 to 94 years. The average decade of birth was the 1940s but ranged from 1910 to 1980. More than three-quarters of the individuals (76%) were college educated and had a relatively low prevalence of cerebrovascular risk factors 32 as determined by clinical examinations an average of 1.1 ± 0.9 years around the time of MRI. The average time interval between neuropsychological testing and MRI was 0.01 ± 0.09 years and ranged from 0 to 1.33 years with 90% of subjects receiving neuropsychological testing at the time of MRI. Signi cant differences by sex at birth were found among women for height (-2.74 inches), intracranial volume (-77.9 ccs), hippocampus (-0.32 ccs), total cortical gray matter (-27.0 ccs), immediate memory (0.51 points greater) and delayed verbal memory (0.55 points greater). The speci c distributions of age for various decades of birth are also summarized in the Supplemental Materials to this manuscript.

Impact of Decade of Birth on Anatomical Outcomes
Height and Head Size Figure 2 summarizes the ndings for height and head size by decade of birth. First, we separately examined differences in height and head size. Height increased signi cantly over decades of birth for both men and women with nearly an identical slope (0.032 inches per decade). Intracranial volume also increased signi cantly for both men and women over the decades of birth with modest, but not signi cant differences by sex (1.2 cc/decade for women versus 1.7 cc/decade for men). Next, we tested whether the increases in head size differed from those for height. In a multivariate model that included adjustments for height, sex, age, and interactions of sex with age, height and decade of birth, intracranial volume increased slightly but signi cantly at 1.41 cc per decade of birth, indicating small but signi cant secular increases in intracranial volume relative to height for men and women (p < 0.0001).

Brain Measures
The results of our primary analyses are summarized in Table 2 (leftmost column, for the full range of ages under each brain region) where brain measures were converted to z-scores to compare effects across regions. For all brain measures except cortical thickness, signi cant positive effects of decade of birth were found for each region with the magnitude being greatest for intracranial volume. In contrast, decade of birth was signi cantly and negatively associated with cortical thickness.   .0001 ***, <0.001 ** <0.05 * N denotes numbers of participants at each level of sensitivity analysis. Age-range of restricted analyses is given as heading to that column. 1 All MRI measures were z-transformed for comparison of effect across regions and results presented as beta coe cients from regression analysis. Figure 3 summarizes associations between total gray matter volume (cc) for men and women of the study by decade of birth (left panel) and age (right panel) predicted from the multiple regression parameters summarized in Table 2. The left panel displays the relation between decade of birth and total gray matter volume by various ages, whereas the right panel displays the relation between age and total gray matter volume by various decades of birth. As an example, from the left panel, a 60-year-old man born in 1920 would be predicted to have a cortical gray matter volume of 501 cc whereas a 60-year-old man born in 1980 would be predicted to have 537 cc of cortical gray matter. From the right panel, a man born in 1948 (average birth decade of the study population) would be predicted to lose nearly 80 ccs of gray matter volume over the same time frame (ages 25-85). Therefore, while decade of birth had a signi cant positive effect of cortical gray matter volume, the impact of aging on cortical gray matter remains essentially unchanged (i.e. no signi cant age by birth decade interaction) and the standardized beta is greater in magnitude for the age effect versus the impact of decade of birth (β age(yrs) = -0.021 versus β birth decade = 0.008 z-scores), although opposite in sign (i.e. increasing age leads to atrophy whereas increasing birth decade leads to larger volume).

Contribution of ApoE genotype and Vascular Risk Factors
Logistic regression analysis of the prevalence of ApoE4 genotype by decade of birth was not signi cant. Conversely, SPB measures declined signi cantly with advancing decade of birth (p < 0.0001). Both FSRP and SPB, however, strongly interacted with age and decade of birth (p < 0.0001) in predicting cortical GM volume outcomes, where the positive impact of decade of birth on brain structure was diminished by increased participant age (and consequently worsening FSRP or SPB) at the time of MRI.
Adding ApoE genotype to the MRI models resulted in a very small (0.03 cc), but signi cant (p < 0.05) increase in TCV and cortical gray matter volumes for ApoE4 genotype carriers that did not modify the positive effect of decade of birth on these measures. Conversely, FSRP, a summary scale for stroke risk 33 , was associated with signi cantly smaller, hippocampal, and cortical gray matter volumes (p < 0.0001), that interacted modestly with decade of birth such that declining mean FSRP with increasing decade of birth was associated with increased volumes of these measures.

Sensitivity Analyses
Given the substantial association between age at time of MRI and decade of birth, two additional analyses, selecting more limited age-ranges, were performed. Table 2 includes the results of our analysis restricting the age range of participants. The number and distribution of ages within these subsamples are summarized in the Supplemental Materials to this study. The sensitivity analysis was done at two levels of restriction. The rst step was to reduce the range to 45-75 years of age in epochs of 10 years, spanning the birth decades of 1930 to 1970. These results are shown in the second column of Table 2 under each MRI measure. While age continues to in uence brain measures, the impact on decade of birth was not changed in this age-restricted group. In the second sensitivity analysis, the age range was further restricted to a single 10-year epoch of ages between 55 and 65 that spanned the decades of birth from 1940 to 1960. The results are shown in the third column of Table 2 under each MRI measure. Comparing results across the various groups, the age effects diminish as the age ranges are restricted, but the impact of decade of birth is equal to, or increased with each level of analysis, except for cortical thickness where the effect of birth decade diminishes with restricted age range but remains signi cantly negative. Figure 4 (this gure and all subsequent gures use z-score converted measures to compare across MRI regions and cognitive performance measures examined) summarizes predicted results for those individuals 45-75 years of age at the time of MRI by decade of birth, where color coding denotes age group at MRI. In brief, there is a linear increase in predicted intracranial volume with increasing decade of birth within and across the 3 age-ranges. This is expected given that intracranial volume does not vary signi cantly with age in this subgroup. Conversely, both cortical gray matter and hippocampal volumes show a signi cant age effect, where older individuals have smaller volumes, but again, the volumes increase with advancing birth decade, even within the same age group (as denoted by the same color lines). Given that the age lines are generally parallel, the effects of decade of birth appears linear across the age groups for all MRI measures, except for hippocampal volume, where the effect seems largest for the oldest age group.
The decade of birth relationship is different for cortical thickness as summarized by Fig. 5. In this gure, cortical thickness declines with advancing age, but there is also a decline in cortical thickness with advancing decade of birth. Overlay of intracranial volume (most right panel) shows declines in cortical thickness across birth decades to covary with increasing head size.
To further investigate this nding, we examined the relationship between intracranial volume, cortical surface area and cortical thickness, as cortical surface multiplied by cortical thickness should re ect cortical gray matter volume. For each cc in intracranial volume, there was 0.35 cc increase in gray matter (R 2 = 0.88, p < 0.0001). Similarly, for each cc in cortical gray matter volume, cortical surface area increased by nearly 3 cm 2 (R 2 = 0.63, p < 0.0001) but for each cc in cortical gray matter volume, cortical thickness increased by only 0.001 mm (R 2 = 0.13, p < 0.0001). Conversely, for each cm 2 increase in cortical surface area, cortical thickness declined by 0.003 mm (R 2 = 0.14, p < 0.0001). These results indicate that the increased head size and gray matter accompanying advancing decade of birth led to increases in surface area that were greater per unit increase than cortical thickness, leading to the negative association of cortical thickness with decade of birth within similarly aged individuals.

Impact of Decade of Birth on Neuropsychological Outcomes
Contribution to Educational Achievement Logistic regression found a highly signi cant increase in likelihood of achieving a college education with each advancing decade of birth, reaching a peak probability of nearly 92% for women born in 1970 (p < 0.0001). There also was a signi cant 2-way interaction between decade of birth and sex (p < 0.01) where women had more rapidly increasing likelihood of achieving a college education with advancing decade of birth, particularly between decades of birth spanning 1910-1950.

Memory Performance
The impact of decade of birth on memory performance is summarized in Table 3 and displayed for the 45-75-year age groups in Fig. 6. Overall, women performed better than men on these memory tasks. Decade of birth was positively and signi cantly associated with memory performance on all measures (p < 0.001-< 0.0001). The main effect of age was not signi cant, but there was a modest but signi cant interaction where older individuals appear to improve more at later birth decades (p < 0.05). An example of this effect is seen in Fig. 6 where the slope of improvement by decade of birth seems greatest for the 65-75-year age group and attens slightly in the younger age groups. This effect also appears strongest among men. An additional, striking effect of education on memory performance for college educated individuals is also seen and is greatest for women although the 3-way interaction of education*sex*decade of birth (not shown), was not signi cant. Intracranial, cortical gray and hippocampal volumes were signi cantly associated with each measure of memory performance after adjusting for age, sex, and education (p < 0.001-<0.0001). Cortical thickness was not associated with memory performance. When intracranial, hippocampal, and cortical gray matter volumes were included in the same model, however, only hippocampal volume remained signi cantly associated with memory performance (p < 0.001-<0.0001). Finally, hippocampal volume remained signi cantly associated in a model that included the addition of decade of birth with age, sex, and education as predictors of memory performance (p < 0.0001).

Mediation Effect of Hippocampal Volume
We assessed the possible mediation of hippocampal volume on the relationship between decade of birth and memory performance using causal mediation analysis. Table 4 summarizes the results. Hippocampal volume signi cantly mediated the relationship between decade of birth and memory performance.
The proportion mediated by hippocampus varied from approximately 5 to 7%, being strongest for delayed memory mediation. The components of the mediation analysis are summarized in Fig. 7 where hippocampal volume is positively associated with decade of birth and delayed memory performance with an average causal mediation effect of 0.013 (p < 0.0001) for delayed memory.

Discussion
Our results nd signi cant secular increases in height and brain structure along with improved memory performance with advancing decades of birth. Brain structures also appear to be increasing at a rate slightly greater than height. Further, improved memory performance accompanying advancing decade of birth appears to be mediated (at least partially) by increasing hippocampal volume after adjusting for age, sex, and education. We hypothesize that larger brain volumes indicate improved "brain health" 34,35 and potentially greater "brain reserve" 36-41 that could explain the declining incidence of dementia as previously reported in the Framingham Heart Study 5 and further supported by improved verbal memory performance associated with greater hippocampal volume found in this study.
Our ndings of consistent enlargement of TCV, cortical gray matter and hippocampal volume with advancing years of birth spanned dates from 1902-1985.
Confounding by association of age at MRI with decade of birth was addressed through sensitivity analysis. Results from these analyses found that stepwise restriction of the age of participants, rst to 45-74, and then to 55-65 years, where age-related differences in brain volumes were minimal, showed consistent increases in brain size measurements with advancing decade of birth, even when the decades of birth were limited to 30 years as occurred with the most restricted analysis (See Supplemental Materials). Analysis of the age-range between 45 and 75 years of age (equal to the middle quartiles of the age distribution or 75% of the cohort) and covering 5 decades of birth, resulted in multiple unique ndings (Figs. 4-6). First, TCV volume, like height, increased linearly through the 5 decades for both men and women. Conversely, decades of birth effect were strongest for hippocampal volume for individuals in the 7th decade of life, and somewhat less for those in the 6th decade, at the time of MRI. These birth years from 1930-1950 coincided with times of great stress such as the depression and World War II. Stress is known to affect hippocampal structure and function 42,43 and it is tempting to postulate these major stressors as potential factors contributing to secular trends in hippocampal volume along with the many other differences lifestyle that occurred over the same time period.
Finally, cortical thickness declined with decades of birth as well as advancing age. Declining thickness coincided with increasing TCV as well as cortical gray matter volume and surface area. Although cortical surface area was positively associated with cerebral gray matter, cortical surface area was inversely associated with cortical thickness. Rakic 44 describes how the radial unit lineage model could explain evolutionary and developmental aspects of the cerebral cortex that would be consistent with our ndings. In this model, expansion of the cerebral cortex occurs through increasing convolutions and expanding surface area while limiting change in cortical thickness. White et al 45 extend this concept to "gyri cation" of the cerebral cortex, which they note results in a surface area "1700 times larger [in humans] than in shrews, yet the thickness of the cortex is only six times greater". Like Rakic 44 , White et al 45 emphasize the computational utility of expanding surface area over thickness within the radial unit lineage model that allows for "an optimized compaction of neuronal bers with an e cient transit time for neuronal signaling". Consequently, increased cortical gray matter with advancing decades of birth results in increased surface area through increased gyri cation. According to the radial unit lineage model, the increased gyri cation (and consequent surface area) leads to subtle reductions in cortical thickness as seen in our data. Importantly, both Rakic 44 and White et al 45 emphasized regional differences in "gyri cation" are likely under unique genetic in uence 46 . Such an analysis is beyond the scope of this report but could add further information regarding the biology of cortical gray matter expansion with subtle reduction in cortical thickness as seen with our more global measures.
Memory performance also improved with decade of birth and was signi cantly associated with all MRI measures except cortical thickness. Improved memory performance was dramatically associated with level of educational achievement, being nearly 1/3 of a standard deviation higher for college educated participants. Memory performance was nearly linearly associated with decade of birth for women but appeared to asymptote somewhat for men (Fig. 6). This too may be associated with achieved level of education for men versus women. Whether this relationship is due to increasing memory performance that leads to increased likelihood of college education or visa a versa is beyond the scope of this report and requires further investigation. Despite these sex differences, improvement in memory performance was partially mediated by increased hippocampal volume that also occurred with advancing decades of birth.
How might these secular effects modify the likelihood of later life dementia? Brain growth begins in utero, increases throughout childhood, and reaches a maximum size in early adulthood, [47][48][49][50] . TCV is highly associated with brain growth during normal development 40 , whereas aging or disease-related brainvolume decrease does not alter TCV. Thus, adult TCV is a stable valid measure for maximal attained brain size, widely used as a proxy for brain reserve 10, 51 , and is an important predictor of cognition in old age 41 including studies that used similar measures of memory performance 51 . Conversely, cross-sectional, and longitudinal age-related differences in brain volume measures associate with cognitive performance in aging and disease [52][53][54] . Hippocampal volume loss, in particular, is considered to be sensitive to early degenerative diseases such as Alzheimer's disease 55,56 . Although absolute volumes do not associate with cognitive ability per se, as illustrated by the fact that women in this study had signi cantly smaller hippocampal volumes, but signi cantly better memory performance, loss of brain tissue within an individual is strongly indicative of pathological effects 57 , which, therefore, may be buffered by larger structural brain development. As noted in our data, while decade of birth had a signi cant positive effect of cortical gray matter volume, the impact of aging remains essentially unchanged. Consequently, individuals in each decade of birth appear to atrophy at the same rate, but later decades start further away from any 'threshold' of atrophy that might manifest as clinical cognitive impairment. Alternatively, and more likely, larger structural brain development may be a surrogate for other environmental processes ongoing during development and early adult life such as increased brain connectivity 58 . Increased connectivity is consistent with the radial unit lineage hypothesis 44 that enables increased neuronal connectivity through cortical expansion and gyri cation 45 . Such increased connectivity could mitigate the impact of age-related diseases on cognitive performance and ts well with the scaffolding hypothesis of cognitive reserve 59 .
Evidence supporting the notion that experience is associated with regional brain expansion can be seen among London taxi-cab drivers who have larger hippocampal volumes compared to same aged individuals 60 .
While head and brain size are under substantial genetic in uence 61 , 46, 62-64 , the timeline of effect found with our results indicate that early life environmental in uences are also likely substantial contributors, particularly educational achievement 6 . Life course perspectives emphasize the impact of early life experiences on brain health 65, 66 that also translate into larger brain structures 67 and reduced risk for later-life dementia through improved reserve 68 . Similarly, effort to improve cardiovascular health during adulthood 69-71 are associated with reduced incidence of cognitive impairment 72 and dementia 73 indicating that modifying these factors could also serve to improve resistance to late-life dementia 74 .
In summary, our results indicate that TCV and brain structures have increased over birth years ranging from 1902 to 1985. These differences are coincident and associated with improved memory performance that is partially mediated by larger hippocampal volumes. These ndings likely re ect both secular improvements in early life environmental in uences through health, social cultural, and educational factors 67 as well as secular improvements in modi able dementia risk factors leading to better "brain health" and reserve 32 . While these effects are likely to be small at the level of the individual, they are likely to be substantial at the population level adding to growing literature that suggests optimized brain development and ideal health through modi cation of risk factors could substantially modify the impact of common neurodegenerative diseases such as stroke and Alzheimer's disease on dementia incidence 6, 34, 70 .
Moreover, taken together with increases in IQ throughout the 20th century 75 , these secular trends in brain volume may contribute to an overall more cognitively resilient and productive society.
The strengths of our study include the design of the Framingham Heart Study that began in 1948 and has followed a community of individuals with comprehensive health evaluations throughout much of their lifespan and across three generations. The addition of MRI beginning in 1999 enabled quantitative brain assessment across all three generations spanning more than 80 years difference in dates of birth. Obtaining contemporaneous MRI with neuropsychological assessment also enabled brain behavior associations. A high rate of participant participation resulting in a large sample size across decades of birth also enabled reliable sensitivity analyses that might not be accomplished with smaller studies. The duration of observation that includes younger individuals also suggests that this secular trend may be continuing. Finally, the fact that more than 80% of subjects were imaged on just two MRI machines also helped to reduce machine differences that were further reduced using the NeuroCombat statistical harmonization method.
This study, however, is not without limitations. First, and most importantly, the Framingham Heart Study is predominately non-Hispanic White, healthy, and well educated and, therefore, not representative of the more diverse US population. For example, current evidence indicates social-cultural 67, 76-78 and health disparities 2 which are more common among non-White individuals 32 in the US may adversely affect brain health. Second, this is a cross-sectional study that has limited causal inference. Longitudinal analyses showing secular differences in rates of regional brain atrophy would further support evidence of increased brain reserve through resilience to age-related atrophy 39 . Similarly, we are not aware of another cohort spanning 3 or more decades of birth to validate these ndings.
Despite these limitations, we conclude that this study extends current knowledge by showing that secular trends in brain structure are occurring and are associated with improved memory performance. We hypothesize that the larger brain structure re ects improved brain health and is at least one manifestation of improved brain reserve that could buffer the impact of late life diseases on incident dementia. Our results also support the growing literature that emphasizes optimal brain development and ideal brain health as preventive measures to mitigate against rising dementia prevalence among our aging US population.       Impact of Decade of Birth and Education on Memory Performance Figure 6 displays the predicted relationship between decade of birth and immediate, delayed and the sum of immediate and delayed memory performance (converted to z-scores) for men and women strati ed by level of education (college versus non-college). College education had a profound positive in uence on memory performance. Memory performance also improved by decade of birth, although this effect appears strongest for the oldest age cohort.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.