OUP user menu

What provides cerebral reserve?

Roger T. Staff, Alison D. Murray, Ian J. Deary, Lawrence J. Whalley
DOI: http://dx.doi.org/10.1093/brain/awh144 1191-1199 First published online: 26 March 2004

Summary

The cerebral reserve hypothesis is a heuristic concept used to explain apparent protection from the onset of cerebral disease and/or cognitive decline in old age. A significant obstacle when investigating the reserve hypothesis is the absence of baseline data with which to compare current cognitive status. We tested the influence of three hypothesized proxies of reserve (education, head size and occupational attainment [OCC]) in 92 volunteers born in 1921, whose cognitive function was measured at age 11 and 79 years, and who underwent brain MRI. The association between each proxy and old age cognitive function was tested, adjusting for variance contributed by childhood mental ability and detrimental age‐related pathological changes measured using MRI. The results showed that education and OCC, but not total intracranial volume (TICV), contribute to cerebral reserve and help retain cognitive function in old age. Education was found to contribute between 5 and 6% of the variance found in old age memory function but was found to have no significant association with reasoning abilities. OCC was found to contribute around 5% of the variance found in old age memory function and between 6 and 8% of the variance found in old age reasoning abilities. We conclude that the intellectual challenges experienced during life, such as education and occupation, accumulate reserve and allow cognitive function to be maintained in old age.

  • cerebral reserve; education; occupation; head size; childhood intelligence
  • AVLT = Auditory Verbal Learning Test; BF = brain fraction; GM = grey matter; MHT = Moray House Test; OCC = occupational attainment; RPM = Raven’s Progressive Matrices; TICV = total intracranial volume; WM = white matter; WMH = white matter hyperintensities

Introduction

The extent to which cognitive performance is retained in the face of brain ageing is of great importance in clinical neuroscience. Many studies of cognitive ageing and dementia show large cognitive variations between individuals in the presence of brain structural changes that appear similar in location and extent (Fazekas et al., 1987; Katzman et al., 1989; Satz 1993; Leaper et al., 2001; Meguro et al., 2001; Bigio et al., 2002). To date, little is known of the sources of these inter‐individual variations. The largest contributor to cognitive function in old age is cognitive function in youth, accounting for just under half of the variance (Deary et al., 2000). Some reports stress the importance of environmental influences on cognitive ageing, especially nutritional, during life (Bowirrat et al., 2002; Perls et al., 2002). Others have emphasized the importance of genetics in individual differences in cognitive ageing, to which polymorphisms in the apolipoprotein E allele make a small but significant contribution (Abate et al., 1998; Bowirrat et al., 2002; Deary et al., 2002).

The fact that substantial differences in cognitive impairment exist in individuals with an identical amount of brain ageing may be accounted for by the hypothesized protective effect of ‘cerebral reserve’. Stern (Stern, 2002) made analysis of cerebral reserve more tractable by: (i) distinguishing between active and passive components of reserve; and (ii) identifying the burden imposed on reserve by age‐related brain pathology. Cerebral reserve may be a passive capacity of brain available to withstand—up to a finite threshold—brain ageing, after which deficits appear. For example, if brain size is a proxy of passive cerebral reserve, the cerebral reserve hypothesis predicts that those with a larger intracranial capacity are able to withstand brain ageing better than those with a smaller intracranial capacity; it would take more brain ageing before age‐related cognitive decline was detectable. The second component addresses the active processes that are available to ameliorate the effects of brain ageing. These active processes are of two types. One is the relative ability to switch to alternative cognitive paradigms to overcome the effects of brain ageing. The other is the ability to recruit compensatory neuronal structures to replace processing pathways damaged by ageing. It is impossible to quantify the diverse positive and negative influences on brain development. However, appropriate proxies of active reserve are education and occupation, which are generally available. To test the cerebral reserve hypothesis requires these passive and active indicators of cerebral reserve, and a measure of brain ageing. To quantify the burden imposed by brain ageing, we used an estimation of white matter hyperintensities, which occur more frequently in the presence of risk factors for vascular disease and are associated with impaired cognitive function (Leaper et al., 2001; Deary et al., 2003).

Testing the cerebral reserve hypothesis is difficult, because it demands a set of measured variables that are rarely available in a single cohort. Ideally, one requires records of mental ability in youth and old age to measure age‐related cognitive change; measures of brain ageing to assess the burden placed on the brain; and measures of passive (e.g. brain size) and active (e.g. education, occupational category) cerebral reserve that might ameliorate this burden somewhat. Precise tests of cerebral reserve hypotheses have hitherto not been feasible because data on mental ability from youth are rarely available for those older people now at risk of cognitive decline. The use of education‐ or occupation‐related variables by some researchers as proxies for childhood mental ability is unsatisfactory, because these are themselves reserve factors contributing to the active model of cerebral reserve.

Here we use cognitive test scores at age 11 years and at age 78–80 years in the same subjects to obtain a measure of age‐related cognitive change. We use a white matter hyperintensity score detected on MRI, which has previously been shown to correlate with old age cognition (Leaper et al., 2001), as a quantifiable example of age‐related brain burden. In the active reserve model, we were also able to use brain fraction (BF) as an additional measure of burden, which has been shown to reduce with age (Blatter et al., 1995) and correlates with intelligence (Willerman et al., 1990). The BF is the ratio of the total white and grey matter volumes to the total intracranial volume (TICV). This measure is reproducible and suitable for a cross‐sectional study such as this (Chard et al., 2002). In the passive cerebral reserve hypothesis, TICV is a putative index of cerebral reserve, and in the active model education and/or occupational status are indicators of reserve.

We formulated tests of the reserve hypotheses as follows: if the cerebral reserve hypothesis is correct, then the measure of reserve should account for significant variance in the cognitive outcomes in old age after adjusting for variance contributed by childhood mental ability and burden. In other words, possessing some reserve means that one’s cognitive score is greater than would be predicted from the person’s childhood ability and the amount of overt, accumulated burden.

Method

Sample

The 1932 Scottish Mental Survey tested the cognitive ability of most of the eligible Scottish children born in 1921 attending school on 1 June 1932 using a version of the Moray House Test (MHT) (Scottish Council for Research in Education, 1933). With the approval of the local ethics committee, 235 of the local residents who took part in the 1932 survey were recruited into a prospective, longitudinal study of brain ageing and health. From this sample, 144 subjects were randomly invited to undergo brain MRI examination. Medical contraindications prevented 17 from participating, 21 refused, and 106 volunteers (65 male and 41 female) were imaged. Eight were removed from the study (six male and two female) because of excess movement during acquisition of volumetric MRI data. Satisfactory MRI data were obtained in a total of 98 volunteers. A sample derivation flow chart is shown in Fig. 1.

Figure

Fig. 1 A schematic flow chart showing how our sample was obtained from the original Scottish mental health survey of 1932.

MRI acquisition

Subjects were imaged using a 1 T Impact MRI scanner (Siemens, Erlangen, Germany) and a RF head coil. The MRI protocol consisted of two acquisitions: first a T1‐weighted sequence, which was used to calculate the brain volume measures, and then a T2‐weighted sequence, which was used to quantify white matter hyperintensities. The T1‐weighted sequence was a 3D‐MP‐RAGE (3D magnetization‐prepared rapid acquisition gradient echo) sequence. This sequence produced 128 sagittal slices of the head with an effective thickness of 1.41 mm. The repetition time was 11.4 ms, the echo time was 4.4 ms, the RF flip angle was 15°, the field of view was 250 mm, the slab thickness was 180 mm, and the acquisition time was 6 min 7 s. The T2‐weighted sequence was a fast spin echo sequence with a repetition time of 4000 ms and an echo time of 96 ms. This sequence produced axial images of the head with a section thickness of 5 mm and an intersection gap of 1.5 mm. The field of view was 230 mm, the slab thickness was 140 mm and the acquisition time was 1 min 55 s.

Image processing

The T1 imaging data were processed using SPM99 (Statistical Parametric Mapping 99) image processing package (Wellcome Department of Cognitive Neurology, London). Each data set was automatically segmented into images representing the probability that any voxel contained grey matter (GM), white matter (WM), CSF or other tissue. The segmentation algorithm was described by and uses an a priori estimate of the distribution of GM, WM and CSF and cluster analysis to estimate the distribution within individuals (Ashburner and Friston., 1997). As part of the segmentation algorithm, the data were corrected for the effect of heterogeneity attributable to magnetic field non‐uniformity (Sled et al., 1998). The resultant images were inspected for adequate segmentation into the different tissue types.

Using GM and WM probability maps, a bit maps mask representing the parenchymal brain was created using the software feature available within the SPM99 software package. This involves erosion and dilation of the probability maps and extracts the brain from the initial segmentation maps. Mutually exclusive bit maps for the GM and WM were then classified using the tissue type with greatest probability within the parenchymal brain. After manual editing, a CSF bit map was created by including those voxels that had a probability of containing CSF greater than 0.5. The three bit maps obtained were then adjusted by classifying any conflicting voxels as CSF so that these comprised mutually exclusive data sets. Axial slices inferior to the cerebellum were not included in the calculations. The results were expressed as fractions of the TICV, calculated as the sum of the GM, WM and CSF. The parenchymal BF was calculated as the ratio of the parenchymal brain to the TICV. This ratio has previously been used to estimate whole brain shrinkage (Rudick et al., 1999) and thus may offer an index of the brain ageing burden.

Subcortical and deep white matter hyperintensities (WMH) on the T2‐weighted MRIs were analysed using a semiquantitative rating scale (Fazekas et al., 1987). Each image was scored by three independent observers on a scale ranging from 0 to 3 of increasing severity (0 normal, 1 punctate, 2 coalescing and 3 confluent) and mean scores were used in the analyses. We have previously shown this scale to be reliable in terms of intra‐ and interobserver variability (Leaper et al., 2001).

Cognitive testing

Moray House Test scores from age 11 were available for all subjects from the records of the Scottish Mental Survey 1932. Mental test scores from old age were obtained by a trained psychologist who administered Raven’s Progressive Matrices (RPM) (Raven, 1960), a non‐verbal reasoning measure of fluid‐type intelligence, and the Auditory Verbal Learning Test (AVLT) (Rey, 1958), a measure of immediate and delayed verbal memory. Memory and reasoning are cognitive functions known to decline with age (Salthouse, 2003). The RPM test was used since it is one of the best individual tests with respect to general cognitive ability (Raven, 2000) and the AVLT is a common tool for measuring memory impairment (Flowers et al., 1984). Verbal memory tests are especially sensitive to the early stages of pathological cognitive decline (Howieson et al., 2003). A total of 92 volunteers satisfactorily completed all MRI and cognitive testing.

Demographic measures

Years of education and occupational classification of each participant were recorded at interview. Occupational attainment (OCC) was categorized using the UK’s Office of Population Statistics classification (Great Britain: Office of Population Censuses and Surveys, 1990). Their educational experience was typical of the UK in the 1930s, with most people leaving school by the age of 14 years (65.2%). More than 95% of the group had no full‐time education after secondary school. The median duration of education in years for males and females was 9 years. Three individuals had a university education. The split between blue‐ and white‐collar occupations was approximately 50 : 50. The distribution of occupational level within the group was as follows: managerial 9, professional 10, lesser professional 3, secretarial 24, skilled manual 21, semi‐skilled ii 5, semi‐skilled i 8, unskilled ii 8, and unskilled i 7. These values represented the highest ever occupational level obtained by an individual.

Statistical analysis

The statistical comparisons were performed using SSP10 (Statistical Package for the Social Sciences 10; Chicago, IL, USA).

We used general linear modelling analysis of covariance (ANCOVA) to test the active and passive brain reserve hypotheses. In both the active and passive models the outcome variables were cognitive test scores at age 79. Separate models were tested for verbal memory (AVLT) and non‐verbal reasoning (RPM). Moray House Test scores at age 11 were covariates. Measures of brain ageing burden and reserve factors were also covariates. Sex was entered as a between‐subjects fixed factor.

Results

The means (standard deviation) for each variable are shown in Table 1. Years of education, TICV, brain fraction and AVLT differed between sexes, with women spending longer in education (P = 0.036) and scoring higher on AVLT (P = 0.042). Women had smaller intracranial volumes (P < 0.001).

View this table:
Table 1

Volunteer group characteristics for each measured variable

Male Female P TotalMax.Min.
(n = 53)(n = 39)(n = 92)
MHT score at age 1163.017.0
 Mean41.742.20.82741.9
 SD11.89.610.9
 95% CI (upper, lower)45.0, 38.445.3, 39.144.2, 39.7
EDU (years)178
 Mean9.5110.20.0369.80
 SD1.102.001.57
 95% CI (upper, lower)9.81, 9.2110.85, 9.5610.1, 9.48
OCC91
 Mean4.754.820.8934.78
 SD2.392.222.31
 95% CI (upper, lower)5.41, 4.105.54, 4.105.26, 4.31
TICV (cm3)1934.31128.1
 Mean1574.41387.7<0.0011495.3
 SD140.9128.0163.7
 95% CI (upper, lower)1613.2, 1535.61429.3, 1346.31529.2, 1461.4
BF (%)82.365.4
 Mean72.073.90.01272.8
 SD3.53.73.7
 95% CI (upper, lower)72.9, 71.075.1, 72.773.5, 72.0
WMH score30
 Mean1.061.270.1651.15
 SD0.670.700.69
 95% CI (upper, lower)1.25, 0.881.49, 1.031.29, 1.00
RPM score4510
 Mean30.127.20.12228.9
 SD7.89.78.7
 95% CI (upper, lower)32.2, 29.730.4, 24.130.7, 27.1
AVLT score6617
 Mean37.241.60.04239.1
 SD10.210.010.3
 95% CI (upper, lower)40.0, 34.444.8, 38.441.2, 37.0

Calculated P values are given after testing for differences between sexes. EDU = education.

Testing the passive cerebral reserve hypothesis

The simplest model of cerebral reserve is that it is passive, and that possessing more of a reserve factor protects an individual to some extent from intellectual decline. Brain size before the onset of ageing is estimated here from total intracranial volume, and is used as a reserve factor in the passive model. This reasoning implies that a person with a larger brain prior to brain ageing is capable of withstanding a larger degree of brain ageing burden, and should show less decline in cognitive function than a person with similar childhood mental ability, the same amount of brain ageing burden, and a smaller brain. In other words, to test the passive reserve hypothesis, we ask whether there is a relationship between original brain size (inferred from total intracranial volume) and a measure of cognitive function in later life, when individual differences in childhood intelligence and a measure of brain ageing burden (hereinafter just burden) are controlled. First, we used memory scores (AVLT) as the dependent variable, with childhood mental ability (MHT), burden (WMH) and TICV as the covariates, and sex as a fixed factor. Table 2 indicates that brain size makes no significant contribution to memory at age 79. This analysis also showed that sex and extent of burden (estimated from WMH score) also made no significant contribution to verbal memory (AVLT), whereas childhood MHT score contributed significantly (P = 0.021), accounting for 6% of the variance (Table 3). Similarly, we repeated the analysis with non‐verbal reasoning (RPM) as our dependent variable and again found that TICV made no significant contribution to cognitive score at age 79. This analysis also showed that sex also made no significant contribution to non‐verbal reasoning (RPM), but childhood intelligence and degree of burden (WMH) contributed 14% (P < 0.001) and 13% (P < 0.001) of the variance, respectively (Table 4).

View this table:
Table 2

The effect of each proxy of reserve on cognitive ability aged 79 years

Memory (AVLT)Reasoning (RPM)
ModelProxy of reserveMeasure of brain burden P Partial η2 P Partial η2
PassiveTICVWMH0.8460.0000.3680.009
ActiveOCCWMH0.0940.0320.0210.056
BF0.0310.0520.0060.078
EDUWMH0.0240.0570.8280.001
BF0.0190.0620.9250.000

The table shows the effect of each proxy of reserve on old‐age cognition when tested using different combinations of burden and old‐age cognitive ability. The values were calculated using the general linear model (univariate analysis) with childhood intelligence as an additional covariate and sex as a fixed factor. The partial η2 is an estimate of the effect size, which is the proportion of the variance that can be explained by the variable. EDU = education.

View this table:
Table 3

Effect of childhood intelligence on cognitive ability aged 79 years

Memory (AVLT)Reasoning (RPM)
ModelProxy of reserveMeasure of brain burden P Partial η2 P Partial η2
PassiveTICVWMH0.0210.0590.0000.140
ActiveOCCWMH0.0900.0330.0040.086
BF0.2550.0150.0420.044
EDUWMH0.0450.0450.0000.138
BF0.1220.0290.0020.097

The table shows the effect of childhood intelligence on old‐age cognition when tested using different combinations of burden, reserve proxy and old‐age cognitive ability. The values were calculated using the general linear model (univariate analysis) with sex as a fixed factor. The partial η2 is an estimate of the effect size, which is the proportion of the variance that can be explained by the variable. EDU = education.

View this table:
Table 4

The effect of brain burden on cognitive ability aged 79 years

Memory (AVLT)Reasoning (RPM)
ModelProxy of reserveMeasure of brain burden P Partial η2 P Partial η2
PassiveTICVWMH0.1180.0330.0000.131
ActiveOCCWMH0.1080.0290.0000.145
BF0.0340.0510.0240.054
EDUWMH0.0650.0390.0000.149
BF0.0480.0440.0760.034

The table shows the effect of brain burden on old‐age cognition when tested using different combinations of reserve proxy and old‐age cognitive ability. The values were calculated using the general linear model (univariate analysis) with childhood intelligence as an additional covariate and gender as a fixed factor. The partial η2 is an estimate of the effect size, which is the proportion of the variance that can be explained by the variable. EDU = education.

It would be statistically invalid to test for a similar relationship using BF as our estimate of burden in the passive reserve model, because TICV is used to calculate brain fraction. In summary, the passive reserve hypothesis is not supported in this sample.

Testing the active cerebral reserve hypothesis

The active model of cerebral reserve was tested in the same way. For each potential measure of reserve, a general linear model was used to examine its association with the dependent variable (AVLT or RPM) adjusting for childhood mental ability and one of the brain ageing burden measures (WMH or BF), with sex as a fixed factor. Table 2 shows that occupation contributes significant variance to verbal memory (AVLT) and non‐verbal reasoning (RPM) scores (i.e. occupation acts as a reserve factor) when BF is used as an estimate of burden, and acts as a significant reserve factor for non‐verbal reasoning, but not verbal memory, when WMH is used as an estimate of brain ageing burden. Education acts as a reserve factor for verbal memory but not non‐verbal reasoning when either measure of the brain ageing burden is used. Table 2 also shows that occupation contributes approximately 5% of the variance in memory function at age 79, and between 5 and 8 % of the variance in non‐verbal reasoning function at that age. Similarly, education contributes 6% of the memory variance but has no significant effect on non‐verbal reasoning. In summary, the active reserve hypothesis is supported and its effect size is moderate.

The principal results in this paper concern the contribution of the reserve factors to mental test scores in old age. It is informative also to describe the contributions of childhood mental ability and brain ageing burden. These are provided in Tables 3 and 4, respectively and a schematic diagram representing the influence of each factor on old age cognitive function is shown in Fig. 2. Childhood intelligence contributes to non‐verbal reasoning abilities at age 79. However, the contribution to old‐age verbal memory is less consistent, only one of the four tests showing a significant result (P < 0.05) when the active model is tested. A similar pattern of results is shown when the effect of brain ageing burden is assessed: it clearly contributes to non‐verbal reasoning ability in old age, the effect on memory being less clear (Table 4). Gender influenced reasoning and memory in old age only when occupation was the proxy of reserve. Gender contributed 4.6% (P = 0.039) of the reasoning variance when BF was the measure of burden, males producing a higher RPM score than females. Gender also contributed 6% (P = 0.021) of the memory variance when WMH was the measure of burden, females producing a higher AVLT score than males.

Figure

Fig. 2 A schematic flow chart representing the relationship between the parameters explored in this study. The partial calculated η2 values are shown. The partial η2 represent the level of independent variance in the old‐age cognitive ability accounted for by each variable. These values are the equivalent to the square of a correlation type measure. In cases where we calculated more than three values of η2, the range is quoted. Note that not all of the η2 values are representative of analyses that achieve significance. Those values that describe significant associations (P < 0.05) are shown in bold typeface. MH = white matter hyperintensities; BF = brain fraction; OCC = occupational attainment; EDU = education; MHT = Moray House Test; TICV= total intracranial volume; AVLT = Auditory Verbal Learning Test; RPM = Raven’s Progressive Matrices.

Discussion

To our knowledge, this is the most comprehensive test of the active and passive components of the brain reserve hypothesis as articulated by Satz (Satz, 1993) and Stern (Stern, 2002). We used actual childhood measures of cognitive ability, measures of brain burden, and widely suggested parameters of brain reserve to account for variance in cognitive abilities in non‐demented old people. We investigated passive reserve by using TICV, and active reserve by using education and occupation. The findings support the active but not the passive version of the brain reserve hypothesis. More education and a more cognitively complex occupation predict higher cognitive ability in old age than would be expected for a person’s childhood ability and accumulated brain burden. Possession of a bigger brain in earlier life conferred no such reserve.

The passive model, with head size as the potential proxy of reserve, has been investigated with conflicting results (Graves et al., 1996; Jack et al., 1997; Mori et al., 1997; Schofield et al., 1997; Jenkins et al., 2000). Schofield and colleagues (Schofield et al., 1997) reported that old people with smaller head sizes were more likely to have Alzheimer’s disease, whereas Jenkins and colleagues (Jenkins et al., 2000) found that larger head size did not delay the onset of Alzheimer’s disease. These reports were not able to control for measured premorbid cognitive ability, and do not attempt to assess the effect of brain burden as a potential confounding variable. We used white matter hyperintensity lesion score as a measure of burden. These lesions are often associated with changes in brain blood vessels. Numerous reports (Double et al., 1996; Fox and Freeborough, 1997; Mouton et al., 1998; Leaper et al., 2001) have found relationships between the measures of burden used here (WMH and BF) and cognitive ability in old age. Currently, a complex model of the summative effects of this cerebrovascular pathology, the density of plaques and tangles of Alzheimer’s disease, and the amount of reserve seems best placed to explain the threshold between brain ageing and dementia (Esiri et al., 2001; Whalley, 2002). We have found that bigger brains do not act as a proxy of reserve in terms of the passive model for reserve mediation. The result does not refute the passive model as a whole but shows that TICV is not a proxy for passive reserve in this sample. Alternative proxies could include synaptic count, which is not necessarily reflected in total or regional brain size. It may be that the size of specific brain structures, such as the hippocampus, could be a proxy of passive reserve and these should be considered in the design of future studies. An alternative explanation for TICV not acting as a proxy of reserve is that head size is a poor predictor of performance. Further, in the largest study of brain size and cognition conducted to date, the association between memory and brain size was to overall size and not to specific regions. Also, the memory contribution operated via the general cognitive factor (MacLullich et al., 2002).

Investigations into the relationship between demographic measures and cognitive function in later life have show significant correlations (Snowdon et al., 1989; Plassman et al., 1995; Kidron et al., 1997; Coffey et al., 1999). Studies have suggested that education and occupational status contribute independently to reserve (Mortel et al., 1995). This study shows relationships between memory (AVLT) and education, memory and occupation, reasoning (RPM) and occupation, but not reasoning and education. This raises the possibility that different life experiences may contribute to the reserve of different cognitive domains. It may also reflect the nature of the education received in the 1920s and 1930s, when there was a greater emphasis on verbatim learning rather than problem‐solving. The neurobiological correlates of higher educational attainment and occupational complexity are uncertain but neurodevelopmental studies in animals suggest that exposure to an enriched or more stimulating environment before completion of sexual maturation is associated with greater synaptic density and more complex neocortical interneuronal connections (Rapp and Bachevalier, 2003). In later life, synaptic density falls in the frontal cortex and this fall is greater in early Alzheimer’s disease (Small et al., 2001). It is plausible, therefore, that the detection of roles for education and occupation in this study points to differences in synaptic function as the underlying source of variance in active reserve.

A possible explanation for the correlation between demographic measures such as education and occupation and decline in old age is that those of lower of socio‐economic status might have an increased risk of toxic or environmental exposures, nutritional deficiencies and/or perinatal insult (Katzman, 1993). These issues are confounded by the well‐established roles of education and occupation as risk factors for many multifactorial, late‐onset diseases, especially ischaemic vascular disease (Hemingway and Marmot, 1999) and Alzheimer’s disease (Stern et al., 1994). Many reports suggest that low educational attainment is associated with increased dementia risk and there are clear socio‐economic gradients with risk of ischaemic vascular disease (Katzman, 1993; Mortimer and Graves. 1993; Mortel et al., 1995; Qiu et al., 2001). These associations point to complex inter‐relations occurring at several levels between late life cognitive function, cerebrovascular disease and differences in education and occupation. Controlling for all potential risk factors is a difficult undertaking. In this study education had significant predictive value after we controlled for estimates of the brain burden of cerebrovascular pathology, the latter possibly recording life‐long environmental exposure.

Potential criticisms of this study are related to sample size and bias, particularly survival bias; that is, only those who survived to 1999–2000 were able to participate. Our MRI sample was drawn from the larger 1921 Aberdeen sample. The fact that we investigated healthy survivors creates an unavoidable bias in the imaged group. The general population would contain those ageing less successfully who potentially have a greater degree of brain burden. There is an additional potential for bias attributable to those excluded for medical contraindications such as the presence of a cardiac pacemaker, a tremor involving the head and neck and other conditions that would not allow them to be imaged. Arguably, this group has a greater potential for vascular disease and would exhibit a larger level of brain burden. Similarly, those who refused to be imaged or who were eliminated due to excessive movement during acquisition potentially have a greater degree of burden, which would bias our results. The net result would be to lower the effect sizes with regard to those expected in a more representative sample. The imaging protocol, processing and analysis tools used in this study are straightforward and the software used is available in the public domain. It is possible that the use of different imaging protocols of more elaborate forms of processing and image classification techniques may produce results that are more sensitive.

Conclusion

The maintenance of cognitive function in old age has become increasingly desirable to society with an ageing population. This work investigates a range of potential factors that contribute to cognitive function in old age, based around the concept of the reserve hypothesis. We found no evidence of reserve being implemented through a passive process, with TICV as a measure of reserve capacity. The results show that education and occupation can be considered as proxies of active reserve. Our results suggest that these two measures of active reserve influence different cognitive domains to varying extents.

Acknowledgements

The authors would like to thank S. A. Leaper, H. A. Lemmon, A. Lyburn, L. Frisch and J. Walker for their help during this project. The project was funded by a grant from the Chief Scientist Office of the Scottish Executive. I.J.D. is the recipient of a Royal Society‐Wolfson Research merit award. L.J.W. holds a Career Development Award from the Wellcome Trust.

References

  1. 76.
View Abstract