Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Education and Dementia in the Context of the Cognitive Reserve Hypothesis: A Systematic Review with Meta-Analyses and Qualitative Analyses

  • Xiangfei Meng ,

    Affiliations: Department of Psychiatry, University of Saskatchewan, Saskatoon, Saskatchewan, Canada, Canadian Center for Health and Safety in Agriculture, University of Saskatchewan, Saskatoon, Saskatchewan, Canada

  • Carl D’Arcy

    Affiliations: Department of Psychiatry, University of Saskatchewan, Saskatoon, Saskatchewan, Canada, School of Public Health, University of Saskatchewan, Saskatoon, Saskatchewan, Canada

Education and Dementia in the Context of the Cognitive Reserve Hypothesis: A Systematic Review with Meta-Analyses and Qualitative Analyses

  • Xiangfei Meng, 
  • Carl D’Arcy



Cognitive reserve (CR) or brain reserve capacity explains why individuals with higher IQ, education, or occupational attainment have lower risks of developing dementia, Alzheimer’s disease (AD) or vascular dementia (VaD). The CR hypothesis postulates that CR reduces the prevalence and incidence of AD or VaD. It also hypothesizes that among those who have greater initial cognitive reserve (in contrast to those with less reserve) greater brain pathology occurs before the clinical symptoms of disease becomes manifest. Thus clinical disease onset triggers a faster decline in cognition and function, and increased mortality among those with initial greater cognitive reserve. Disease progression follows distinctly separate pathological and clinical paths. With education as a proxy we use meta-analyses and qualitative analyses to review the evidence for the CR hypothesis.

Methodology/Principal Findings

We searched PubMed, PsycoINFO, EMBASE, HealthStar, and Scopus databases from January 1980 to June 2011 for observational studies with clear criteria for dementia, AD or VaD and education. One hundred and thirty-three articles with a variety of study designs met the inclusion criteria. Prevalence and incidence studies with odds ratios (ORs), relative risks or original data were included in the meta-analyses. Other studies were reviewed qualitatively. The studies covered 437,477 subjects. Prevalence and incidence studies with pooled ORs of 2.61 (95%CI 2.21–3.07) and 1.88 (95%CI 1.51–2.34) respectively, showed low education increased the risk of dementia. Heterogeneity and sensitivity tests confirmed the evidence. Generally, study characteristics had no effect on conclusions. Qualitative analyses also showed the protective effects of higher education on developing dementia and with clinical disease onset hastening a decline in cognition and function, and greater brain pathology.


This systematic review and meta-analyses covering a wide range of observational studies and diverse settings provides robust support for the CR hypothesis. The CR hypothesis suggests several avenues for dementia prevention.


The Global Burden of Disease Study (GBD) suggests that by 2020 dementia and other neurodegenerative diseases will be the eighth largest source of disease burden in developed countries [1]. The GBD researchers also weighted the severity of disability for a series of health conditions. Out of the ten disorders within the three highest disability classes, eight are neurological problems. Alzheimer’s Disease (AD) is a devastating chronic disease that significantly increases healthcare costs and influences the quality of life of those afflicted and their caregivers. It is the most common degenerative brain disorder. Age at onset varies between 40 and 90 years, in most patients it is after age 65. It is estimated that from 1 to 13% of the population 65 and over have AD [2][4]. Vascular disease (VaD) is the second most common dementia disease. Between 1% and 4% of those 65 and over suffer from VaD and the prevalence doubles every 5 to 10 years after the age of 65 [5]. A better understanding of causes of dementia has importance for health researchers and policy makers, in terms of treatment and prevention strategies.

Over the past decades progress has been made in exploration the etiology of dementia. There are many promising leads in the search for the causes(s) of AD. The progressive accumulation and deposition of neurotoxic Aβ/amyloid plaques and tangles that aggregate in the brain with aging–the amyloid hypothesis of AD, is currently a major hypothesis, is likely a necessary but not sufficient cause [6]. The epidemiological literature has found several risk factors for AD including age, gender, education, and ApoE status [7], [8]. Autopsy studies have found that 10% to 40% of individuals with mild to moderate brain pathology did not manifest clinical symptoms of dementia [9], [10]. A hypothetical construct of “cognitive reserve” or “brain reserve” is widely used to explain how, in the face of neurodegenerative changes that are similar in nature and extent, individuals vary considerably in terms of their cognitive decline and clinical manifestation of dementia symptoms [11].

The Cognitive Reserve Hypothesis

The concept of brain reserve or cognitive reserve (CR) refers to the ability to tolerate the age-related changes and disease related pathology in the brain without developing clinical symptoms or signs of disease [12]. CR also explains the relationship between education, occupational complexity, reading ability, IQ and dementia. This reserve is seen to be a result of changes in the brain itself, resulting from changes in brain structure and processing [13]. Morphological and neurochemical brain changes have been observed in animals raised in stimulating environments [14]. For Stern [15] CR can take two forms: (1) neural reserve in which existing brain networks are more efficient, or have greater capacity, may be less susceptible to disruption; and, (2) neural compensation in which alternate networks may compensate for the pathological disruption of preexisting networks. For Mortimer [16] the fact that pathological lesions can be present long before clinical symptoms of dementia arise suggests that there are two distinct sets of risk factors for AD, “…one for the pathology and the other for clinical expression”. Implicit in the literature is the notion of a threshold which posits that cognitive reserve will limit the clinical expression of the underlying disease until a threshold level of brain pathology is reached at which point the CR can no longer compensate for the underlying physical brain degeneration. Thus the relationship between CR and dementia will differ depending on the underlying pathology [17]. Implicit is the possibility that directly enhancing CR may help forestall the clinical manifestation of AD and dementia [15].

The CR hypothesis postulates that CR acts through both protective and compensatory mechanisms. Individuals with higher levels of CR will have a lower prevalence and incidence of dementia particularly AD. Given that there is both pathology and clinical pathways and if one assumes that the pathological pathway starts at a fixed time in the life course then one would expect that higher CR would lead to a later onset and later death. However if one assumes that there may be a variable time for the start of the pathological process and that process is slow and insidious then the CR hypothesis would suggest that CR will have little or no effect on age of onset of clinical symptoms of dementia and survival (cf. Paradise et al. [18]). Explicit in the CR hypothesis is the notion of a threshold after which CR can no longer compensate for the underlying brain degradation. The CR hypothesis implies that those with higher CR will show a faster cognitive decline after disease symptoms are manifest. It would be consistent with the CR hypothesis to expect those with greater CR would be more sensitive to cognitive impairments and thus expect those with more CR will more likely to present earlier in the clinical progression of the disease and will score higher on initial assessment. Finally it is expected that among those with dementia, individuals with higher CR will show greater brain pathology that those with less CR.

More broadly mental and physical stimulation both early and throughout the life course is thought to increase CR allowing cognitive function to be maintained in old age and to both protect against and delay the onset of dementia and AD.

This Study

Using education as a proxy measure of CR we explore the evidence linking education and dementia. The CR hypothesis postulates that higher education:

  1. reduces the prevalence of dementia;
  2. reduces the incidence of dementia;
  3. has no affect on the age of onset of dementia;
  4. results in an accelerated cognitive decline – as a result of a threshold effect;
  5. has no effect on the age of death;
  6. leads to a higher level of clinical performance on initial assessment;
  7. shows greater brain pathology in postmortem and imaging studies among those with dementia.

We examine the evidence for these hypotheses by systematically review the literature using both meta-analysis and qualitative analysis techniques.

The appropriateness of education as proxy measure has been questioned. Literacy, reading ability or crystallized intelligence have been suggested as better measures of CR [19], [20]. Though far from perfect as a measure of cognitive reserve, education has been measured in a wide variety of studies with diverse populations and methods thus strengthening the rationale for examining its connection to dementia and AD. While formal education is limited as a measure of CR because it is an attribute usually acquired and fixed relatively early in life it may by itself heighten the propensity for greater physical and mental stimulation throughout the life course.

While five review articles published after 2005 have dealt with the relation between education and dementia each has limitations that result in them being incomplete in their review of the research literature [12], [18], [21][23]. Fratiglioni and Wang [12] dealt more generally with the CR hypothesis including dimensions other than education and did not use meta-analyses. Two of the review articles, which included meta-analysis contained no more than 20 studies. One article included only case-control and cohort study designs, another just cohort studies and reviewed studies within a narrower time period. Paradise et al. [18] only deal with the relationship between education and survival. The most recent review by Sharp and Gatz [23] is also limited in scope and did not use meta-analyses techniques so no pooled estimates of effect are provided. None of these reviews examined the impact of study quality characteristics on results. As Egger, Smith and Altman point out it is important and desirable to review a body of data systematically, independent of the design and type of study [24].

Our systematic review analyzes data from prevalence and incidence studies with cross-sectional, case-control, and cohort study designs covering a 31-year period. It integrates data from 437,477 subjects and 133 published articles. It includes both a meta-analysis and a qualitative review. In the meta-analysis, the impact of study quality characteristics on results is examined. The qualitative review examines the impact of education on the occurrence of dementia, age of onset, mortality, clinical performance on initial assessment and rate of cognitive decline, and brain degradation is examined. This review fully complies with the PRISMA Statement (see Checklist S1) for systematic reviews and meta-analyses [25].


Search Strategy

We searched the PubMed, PsychINFO, EMBASE, HealthSTAR, and Scopus databases from January 1980 to June 2011 for original research articles in English language. To get maximum number of relevant citations, we employed the followed search strings: ‘(Alzheimer* OR vascular dementia) AND education’ as the keywords for study retrieval.

Selection of Articles

All suitable articles were evaluated taking into account their internal validity and five selection criterion as follows: 1) have cross-sectional, case-control, or cohort study designs; 2) use clear diagnosis criteria for AD or VaD, specifically DSM and its updates [26], NINCDS-ADRDA [27], ICD-10 [28] or other generally accepted criteria; 3) give clear information on education of study subjects (level of study, years of study); 4) provide a statistical indicator (odds ratio, relative risk, hazard ratio, etc) or original data to estimate the relationship between education and dementia; 5) control potential confounders by using statistical adjustment in the analysis or matching in the study design.

Figure 1 shows the process of study selection. This initial search produced 11,401 studies. From these studies, 1,631 study abstracts were retrieved for assessment. We also manually searched the references of potential articles that we had retrieved. One hundred and thirty-three articles were included in this systematic review. They are listed in Appendix S1.

Data Extraction

Data on author, publication year, journal, sample size, methods, indicators, outcomes, adjustments, study design, and results were extracted independently by both authors using a data extraction form. Inconsistencies in interpretation were resolved through discussion. Education was dichotomized into low or high level according to the original article. Disease outcomes were classified based on disease categories reported in the original articles - AD, VaD, and non-specified dementia. Although AD and VaD often co-occur, and can be difficult to be disentangled clinically but become more readily apparent at autopsy. We did not find any reports of mixed dementias in the articles reviewed. The Newcastle Ottawa Scale [29] criteria were used to characterize the quality of case control and cohort studies. Assessment of study quality is essential for a proper understanding of non-randomized studies.

Data Synthesis

The reviewed articles were divided into quantitative and qualitative studies for further analyses. The quantitative analysis required the studies report data on odds ratio or relative risk or have original study data from which we could calculate an odds ratio. Only quantitative studies were included in the meta-analysis. The qualitative category included studies that reported hazard ratios or whose data on education was not possible to be dichotomized into high and low categories, or which included additional data on age of onset, cognitive decline, clinical performance or mortality. The 133 articles in the final review contained 181 analyses. Sixty-nine studies used in the meta-analysis (quantitative group) and the sixty-six studies used in the qualitative analysis. Two of the quantitative studies also reported qualitative results and are thus used in both the quantitative and qualitative analyses. We report on each category of studies separately.


The 69 studies reviewed for the meta-analysis included both prevalence and incidence studies. They were used to generate pooled estimates of the effect of education on the prevalence and incidence of the dementias. The studies were stratified as to disease entity – AD, VaD, and non-specified dementia. We separately analyzed the effects of heterogeneity in prevalence and incidence studies. DerSimonian and Laird I2 statistics were used to test for heterogeneity [30]. Funnel plots and Egger tests were used to inspect for publication bias [31]. The Egger test provides a more objective way to estimate the reliability of the results. The incidence studies combined show non-significant heterogeneity (Egger test, t = 1.06; p = 0.302), but this was not true of prevalence studies (t = 2.29; p = 0.025). The later result suggested the possibility of heterogeneity. Thus we choose to use a more conservative random effects model for our analyses. Pooled odds ratios (ORs) were computed by using random effects model. ORs were corrected for publication bias [32]. Sensitivity analysis included the assessment of the influence of each individual study on overall estimates by recalculating ORs with each study being removed one at a time. Sensitivity analysis was also conducted as subgroup analyses removing studies with different types of study designs or different disease outcomes separately. The quality of each study was rated according to the Newcastle-Ottawa Scale and a meta-regression analysis was used to examine the impact of study quality on results. We used STATA, version 8.2, statistical software for the analyses.


Quantitative Analysis/Meta-Analysis

Table 1, presents data on the characteristics of the 69 studies included in the meta-analysis. Lead author, date of publication, country in which study took place, education categorization, sample size, study design and crude odds ratios are documented. The studies are grouped as prevalence and incidence studies.

Table 1. Education and the risk of dementia: Characteristics of studies included in the meta-analysis.

Prevalence studies.

Figure 2 presents the individual study and pooled estimates for prevalence studies on low education as a risk factor for dementia. The studies are grouped as to disease outcome – AD, VaD, and unspecified dementia. Odds ratio for dementia risk and 95% confident interval are provided. A random effects model was used. Analysis using a fixed effects model did not affect the results.

Figure 2. Summary of findings from observational studies of education and prevalence of dementia (* denotes studies with two or more empirical findings).

The pooled OR overall for any dementia for individuals with low education level compared to with high education was 2.61 (95%CI 2.21–3.07, p<0.001), indicating that those with low education were 1.61 times more likely to have dementia than individuals with high education. Heterogeneity in this analysis was significant (χ2 = 707.08, p<0.001). For AD separately, the OR was 2.62 (95%CI 2.06–3.33, p<0.001), and the heterogeneity was significant (χ2 = 356.74, p<0.001). For VaD, the OR was 2.11 (95%CI 1.40–3.19, p<0.001), and the heterogeneity was also significant (χ2 = 46.81, p<0.001). The OR for unspecified dementia was 2.79 (95%CI 2.13–3.66, p<0.001), and the heterogeneity was also significant (χ2 = 295.99, p<0.001).

Incidence studies.

Figure 3 shows the individual study and pooled estimates for incidence studies. The meta analysis used a random effects model, and results were unchanged using fixed effects model. The pooled OR for individuals with low education level compared to with high education was 1.88 (95%CI 1.51–2.34 p<0.001), indicating low education level were 0.88 times more likely to develop dementia than individuals with high levels of education. Heterogeneity here was significant (χ2 = 213.60, p<0.001). In these incidence studies the low/high education OR for AD was 1.82 (95%CI 1.36–2.44, p<0.001), heterogeneity was also significant (χ2 = 155.03, p<0.001). For VaD in these studies the low/high OR was 2.75 (95%CI 2.20–3.45, p<0.001) while heterogeneity was not significant (χ2 = 0.50, p = 0.78). The incidence OR for unspecified dementia was 1.48 (95%CI 1.17–.86, p<0.001), and the heterogeneity was not significant (χ2 = 4.70, p = 0.195).

Figure 3. Summary of findings from observational studies of education and the incidence of dementia (* denotes studies with two or more empirical findings).

Publication bias.

Funnel plots were used to visually assess the presence of publication bias. As shown in Figures 4 (prevalence studies) and Figure 5 (incidence studies), most of studies are within the domain, which represent 95% confidence interval limits around the estimate, though some articles exceed these boundaries. In these plots the X-axis shows the logarithmic scale of odds ratio estimate for each study and Y-axis is standard error of the logarithmic function of the odds ratio. The dashed line represents the 95% confidence intervals and the point estimate of logarithmic transition of odds ratio illustrates as the solid line. No apparent asymmetry is shown in these funnel plots.

Figure 4. Funnel plot prevalent studies of dementia and education with a variety of study designs.

To examine the issue of heterogeneity more objectively, we employed the Egger test separately for different diseases and study designs. There were no significant differences between different diseases in prevalent and incident studies (P>0.05). Only for cross sectional (prevalence) studies was the Egger test significant (t = 2.30; p = 0.003). We then used the “trim-and-fill” method proposed by Duval and Tweedie [32] to further re-examine publication bias in prevalence studies. The resulting estimate of the corrected OR was 1.87 (95%CI 1.56–2.24, p<0.001), while the corrected OR was reduced, it did not change the essential thrust of the previous findings for prevalence studies on the effect of low education being associated with a higher prevalence of dementia.

Sensitivity analyses.

Sensitivity analyses assessed the influence of each study on overall estimates by omitting one study at a time. For prevalence studies, these analyses yield low/high education ORs ranging from 2.50 (95%CI 2.13–2.93) to 2.66 (95%CI 2.26–3.13); while for incidence studies, the ORs vary from 1.76 (95%CI 1.45–2.14) to 2.03 (95%CI 1.65–2.49). In subgroup analyses both prevalence studies and incidence studies were stratified by study design. The corrected OR for case control prevalence studies was 1.68 (95%CI 1.25–2.26; z = 3.43, p<0.001); for cross sectional prevalence studies the corrected OR was 3.23 (95%CI 2.68–3.88; z = 12.39, p<0.001); a similarly corrected OR for cohort incidence studies was 1.96 (95%CI 1.54–2.51; z = 5.40, p<0.001); the corrected OR for case-control studies incidence studies was not significant at 1.47 (95%CI 0.93–2.34; z = 1.64, p = 0.100). This sensitivity analyses generally supports the conclusion that low education is a risk factor for dementia.


We accessed the relationship between study quality and observed risk through meta-regression of the rating of study quality. Generally none of the specific study characteristics examined had an impact of the observed ORs. None of characteristics of case control or cohort studies show significantly independent results (p>0.05).

Quantitative analyses conclusions.

The above meta-analyses covering a wide range of studies and diverse populations provides robust evidence that higher education substantially reduces the prevalence and incidence of AD, VaD and nonspecific dementia. The pooled odds ratios for this large number of studies strongly support the CR hypothesis that postulates higher education:

  1. reduces the prevalence of dementia.
  2. reduces the incidence of dementia

Qualitative Analyses

The qualitative analyses reviewed the 66 studies with qualitative data, included are results from 2 studies that also reported quantitative data included in the meta-analysis above. They contain 79 specific results. These studies and findings are annotated in Table 2. Qualitative studies are those studies, which reported hazard ratios or whose data on education was not possible to be dichotomized into high and low categories, or which included additional data on age of onset, cognitive decline, clinical performance or mortality. For reporting purposes we have categorized these studies in terms of the focus of their findings.

Table 2. Qualitative analysis: Studies linking education to various aspects of dementia (* denotes studies with findings related to two or more areas of inquiry).

Prevalence studies.

Of the 20 qualitative prevalence studies reviewed 18 (90%) reported that higher education was a protective factor with higher education decreasing the risk of AD. The remaining studies reported non-significant trend in that direction. The 6 studies that had data on VaD prevalence split with 3 studies reported higher education was protective and 3 reported no educational effect.

Incidence studies.

Both studies in this category found that increasing education was protective against the onset of dementia and AD.

Cognitive decline studies.

Twenty studies reported on cognitive decline. The majority, 14 (70%), found that higher education lead to a more rapid cognitive decline. Only 2 studies (10%) reported slower decline among those with higher education while in 4 studies (20%) education was found to have no effect on rate of cognitive decline. More recent studies from 2004 onwards almost uniformly report a more rapid decline. Three of the more recent studies note a threshold effect with those with higher education having an initial slower decline that then accelerates.

Age at onset studies.

Five of the 8 (62.5%) studies with age of onset data report that those with lower education reported a later age of onset consequently higher education was associated with earlier age of onset. The remaining three studies failed to detect any association.

Mortality studies.

Educational difference in mortality were not found in 5 of 8 (62.5%) studies; 2 studies found those with higher education died earlier and the remaining study reported those with less education died earlier. Paradise et al’s meta-analysis also found more education did not lead to earlier death after diagnosis [18].

Initial clinical assessment studies.

These studies almost uniformly (7 of 8) found those with higher education had higher cognitive performance at initial assessment.

Pathology and imaging studies.

Seventy percent of studies (10 of 14) report increased pathology and degradation in brain function among those with more education. In general these studies support the notion that those with education are able to deal with great brain pathology before the clinical manifestation of the underlying disease becomes apparent. It is of interested to note that 3 of the 4 studies that report no relationship between brain pathology and education come from essentially the same research center.

Descriptive analyses conclusion.

These studies are generally supportive of the CR hypothesis that postulates higher education:

  1. has no effect on the age of onset of dementia
  2. after onset of clinical symptoms, results in an accelerated cognitive decline –a threshold effect is implied
  3. has no effect on the age of death
  4. leads to a higher level of clinical performance on initial assessment
  5. shows greater brain pathology in postmortem and imaging studies among those with dementia.



The meta-analyses of quantitative studies reported on here strongly and consistently show that those with lower education had a higher risk for dementia, the pooled OR was 2.61 (95%CI 2.21–3.07) for prevalence studies and 1.88 (95%CI 1.51–2.34) for incidence studies, providing robust evidence that a higher level education resulted in a significant reduction both in prevalence and incidence dementia, either AD, VaD or unspecified dementia. The analyses of descriptive studies also firmly supports the CR hypothesis that postulates that people with higher education will have a lower prevalence and incidence of dementia, higher scores at initial assessment; higher education has no affect on the age of onset, after the onset of clinical symptoms results in an accelerated cognitive decline, and has no effect on the age of death. The biological studies reviewed also show that those with higher education had greater brain damage; but that damage did not translate initially into poorer cognitive performance. There is both clinical and biological support for a ‘threshold effect’ where higher education initially masks the clinical manifestation of dementia but after brain pathology reaches a critical level there is a more rapid cognitive decline.

Strength of Present Review

Our meta-analysis was based on data from over 437,477 subjects from a wide range of countries materially surpassing previous reviews [12], [20], [21]. We also include an analysis of descriptive studies that cover dimensions of the CR hypothesis other than incidence and prevalence.

Limitations-Quantitative Analytic Issues

Though there is some continuing debate on the merit of quantitative meta-analysis of observational reports [33], there is a need to evaluate the reliability and validity of observational study results when experimental methods (randomized controlled trials) are clearly inappropriate, impossible, or inadequate [34].

The present systematic review clearly shows that, people with low education are more likely to suffer from dementia. This is a robust finding evident in both prevalence and incidence studies and in a wide range of settings. The ORs are still significantly greater than 1.0 whether stratified by disease subgroups or adjusted for 8 study quality indicators.

Heterogeneity is a concern in the meta-analysis of observational studies. It can be clinical and/or statistical heterogeneity. Funnel plots of both prevalence and incidence studies give an indication of the possibility of heterogeneity existing. The assessment of the heterogeneity in meta-analysis is a crucial issue because the presence versus the absence of true heterogeneity (between study variability) can affect the statistical model that the meta-analyst decides to apply to the meta-analytic database [35]. The sources of heterogeneity can be reporting bias, which includes publication bias, language bias, citation bias, and multiple publication bias, true heterogeneity (which shows effect sizes differ according to study size) intensity of intervention and differences in underlying risk, data irregularities (which can result from poor methodological design of small studies) inadequate analysis, and fraud, and chance [36].

Data for our systematic review and meta-analyses was based on searches of 5 major databases covering a 31-year period. Our retrieving language restriction may evoke the suggestion of a language bias. However as English is more and more frequently used in scientific discourse the probability of language bias becomes less likely. Given our study collection and extraction procedures it is much more likely than any bias that may arise would be a result of our analysis procedures.

Our analysis of the influence any individual study shows no substantial difference in pooled ORs when individual studies are omitted. This analysis was necessary because of the number of studies included in the review. None of prevalence studies or incidence studies reviewed reported negative significances. Differences in the design of observational studies are a major source of heterogeneity. When stratified by study design in subgroup analyses, only case control incidence studies had non-significant findings. There were only 4 articles included as case control incidence studies. The small number of studies is likely to produce unstable conclusions. Since the Egger’s heterogeneity test on cross sectional prevalence studies was significant, the “trim and fill” method was used to explore the influence of potential publication bias in prevalent studies. The resulting corrected OR did not change our conclusion that, low education is a risk factor for dementia.

It is notable that both prevalence and incidence studies showed positive evidence that high education is a protective factor of against dementia. Low education is not merely a risk factor for dementia, but is a contributory factor in the development of dementia. The findings of this systematic review support the general tenets of the cognitive reserve hypothesis [12], [15][16], [20][21].

Limitations-Qualitative Analytic Issues

Articles reviewed qualitatively reported hazard ratios or data on education that we were not able to dichotomize into high and low categories, or included additional data on age of onset, cognitive decline, clinical performance or mortality, and brain pathology. It was not possible to include those studies in the meta-analysis section of this review. However, to further our understanding of the role of education in dementia, a qualitative analyses of these studies was undertaken. The picture that emerges from the qualitative analysis is that higher education is generally protective against the onset of dementia, particularly AD. Those with higher education generally presented clinically earlier in the progression of disease, have higher clinical scores at presentation, and decline slowly in cognitive function initially but after a threshold level has been breached a more rapid decline ensues. Biological studies indicate that subjects with more formal schooling had greater brain damage but that damage did not translate initially into poorer cognitive performance. These qualitative studies were supportive of the CR hypothesis.

Answered and Unanswered Questions

The findings in this systematic review both the quantitative and qualitative, are consistent with the CR hypothesis in which early education and stimulation may be seen to affect brain structure [12]. Extensive social networks, active engagement, or regular participation intellectually stimulating activities significantly lower the risk of AD and other forms of dementia. In addition to the CR hypothesis, the ‘brain battering’ hypothesis, which presumes that less educated individuals’ increased risk of dementia are related to increased vascular lesions, may also be invoked to explain the association between education and dementia [37]. However such a hypothesis is inconsistent with the post-mortem, imaging and cellular activity studies, which show greater brain pathology among those with higher education. Further research is needed to elucidate the underlying biological mechanisms by which education protects against the onset of dementia and influences its course and outcome.


The present study demonstrates robust evidence that a high level education in early life is related with a significant reduction both in the prevalence and incidence of dementia, including AD and VaD. These results are in accordance with the CR hypothesis, which assumes some aspects of life experience such as education protects against the onset of dementia. Education also influences the course and outcome of the disease in terms the pattern of cognitive decline and underlying brain pathology.

Study findings that adult-life work complexity, social network and complex leisure activities also reduce the occurrence of dementia complements the evidence presented in this systematic review [38][41].

It is important from a health policy perspective to recognize the role of CR and education in the etiology and progression of dementia. These personal attributes are amenable to change. As a prevention strategy Ritchie et al [42] suggests that increasing a population’s ability to use skills, knowledge, and experience (crystallized intelligence) as well as increasing vegetable consumption, and eliminating depression and diabetes would have a greater impact on the prevalence and incidence of dementia than modifying known genetic risk factors.

Supporting Information

Checklist S1.

PRISMA 2009.



Appendix S1.

Data References: Meta Analysis & Qualitative analysis.




We would like to thank the authors who kindly provided original data for the meta-analyses.

Author Contributions

Conceived and designed the experiments: XM CD. Performed the experiments: XM CD. Analyzed the data: XM CD. Contributed reagents/materials/analysis tools: XM CD. Wrote the paper: XM CD.


  1. 1. Menken M, Munat TL, Toole JF (2000) The global burden of disease study: implications for neurology. Arch Neurol 57: 418–420.
  2. 2. Jorm AF, Korten AE, Henderson AS (1987) The prevalence of dementia: A quantitative integration of the literature. Acta Psychiat Scand 76: 465–479.
  3. 3. Mölsä PK, Matilla RJ, Rinne UK (1982) Epidemiology of dementia in a Finnish population. Acta Neurol Scand 65: 541–552.
  4. 4. Canadian Study of Health and Aging Working Group (2001) The incidence of dementia in Canada. Neurology 55: 66–73.
  5. 5. McVeigh C, Passmore P (2006) Vascular dementia: prevention and treatment. Clin Interv Aging 1: 229–235.
  6. 6. Turner RS (2006) Alzheimer’s Disease, Semin Neurol 26: 499–506.
  7. 7. Lindsay J, Laurin D, Verreault R, Hébert R, Helliwell B, et al. (2002) Risk Factors for Alzheimer’s Disease: A Prospective Analysis from the Canadian Study of Health and Aging. Am J Epidemiol 156: 445–453.
  8. 8. Tyas SL, Manfreda J, Strain LA, Montgomery PR (2001) Risk factors for Alzheimer’s disease: a population-based, longitudinal study in Manitoba, Canada. Int J Epidemiol 30: 590–597.
  9. 9. Reisberg B, Doody R, Stoffler A, Schmitt F, Ferris S, et al. (2003) Memantine in moderate-to-severe Alzheimer’s disease. N Engl J Med 348: 1333–1341.
  10. 10. Tariot PN, Farlow MR, Grossberg GT, Graham SM, McDonald S, et al. (2004) Memantine treatment in patients with moderate to severe Alzheimer disease already receiving donepezil: a randomized controlled trial. JAMA 291: 317–324.
  11. 11. Whalley LJ, Deary IJ, Appleton CL, Starr JM (2004) Cognitive reserve and the neurobiology of cognitive aging. Ageing Res Rev 3: 369–382.
  12. 12. Fratiglioni L, Wang HX (2007) Brain reserve hypothesis in dementia. J Alzheimers Dis 12: 11–22.
  13. 13. Katzman R (1993) Education and the prevalence of dementia and Alzheimer’s disease. Neurology 43: 13–20.
  14. 14. Pham TM, Winblad B, Granholm AC, Mohammed AH (2002) Environmental influences on brain neurotrophins in rats. Pharacol Biochem Behav 73: 167–175.
  15. 15. Stern Y (2006) Cognitive Reserve and Alzheimer Disease. Alzheimer Dis Assoc Disord 20: S69–74.
  16. 16. Mortimer JA, Borenstein AR, Gosche KM, Snowdon DA (2005) Very Early Detection of Alzheimer Neurology and the Role of Brain Reserve in Modifying Its Clinical Expression. J Geriatr Psychiatry Neurol 18: 218–223.
  17. 17. Roe CM, Xiong C, Miller JP, Morris JC (2007) Education and Alzheimer disease without dementia: support for the cognitive reserve hypothesis. Neurology 68: 223–228.
  18. 18. Paradise M, Cooper C, Livingston G (2009) Systematic review of the effect of education on survival in Alzheimer’s disease. Int Psychogeriatr 21: 25–32.
  19. 19. Manly JJ, Schupf N, Tang MX, Stern Y (2005) Cognitive Decline and Literacy Among Ethnically Diverse Elders. J Geriatr Psychiatry Neurol 18: 213–217.
  20. 20. Albert SM, Teresi JA (1999) Reading Ability, Education, and Cognitive Status Assessment Among Older Adults in Harlem, New York City. Am J Public Health 9: 95–97.
  21. 21. Caamano-Isorna F, Corral M, Montes-Martinez A, Takkouche B (2006) Education and Dementia: A Meta-Analytic Study. Neuroepidemiology 26: 226–232.
  22. 22. Valenzuela MJ, Sachdev P (2006) Brain reserve and dementia: a systematic review. Psychol Med 36: 441–454.
  23. 23. Sharp ES, Gatz M (2011) Relationship Between Education and Dementia: An Updated Systematic Review. Alzheimer Dis Assoc Disord 25: 289–304.
  24. 24. Egger M, Smith GD, Altman DG (2001) Systematic Reviews in Health Care: Meta-analysis in context. 2nd ed. London: BMJ.
  25. 25. Moher D, Liberati A, Tezlaff J, Altman DG, PRISMA Group (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. J Clin Epidemiol 62: 1006–1012.
  26. 26. American Psychiatric Association (2000) Diagnostic and Statistical Manual of Mental Disorders. 4th ed. Washington: American Psychiatric Association.
  27. 27. McKhann G, Drachman D, Folstein M, Katzman R, Price D, et al. (1984) Clinical diagnosis of Alzheimer’s disease: report of the NINCDS-ADRDA Work Group. Neurology 34: 939–944.
  28. 28. World Health Organization (1992) The ICD-10 Classification of Mental and Behavioural Disorders: Clinical Descriptions and Diagnostic Guidelines. Geneva: WHO.
  29. 29. Wells GA, Shea B, O’Connell D, Peterson J, Welch V, et al. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomized studies in meta-analyses. 1: Available: Accessed 2011 September.
  30. 30. Higgins JP, Thompson SG, Deeks JJ, Altman DG (2003) Measuring inconsistency in meta-analyses. BMJ 327: 557–560.
  31. 31. Egger M, Davey Smith G, Schneider M, Minder C (1997) Bias in meta-analysis detected by a simple, graphical test. BMJ 315: 629–634.
  32. 32. Duval S, Tweedie R (200) A nonparametric “trim and fill” method of accounting for publication bias in meta-analysis. J Am Stat Assoc 95: 89–98.
  33. 33. MacMahon S, Collins R (2001) Reliable assessment of the effects of treatment on mortality and major morbidity. II: Observational studies. Lancet 357: 455–462.
  34. 34. Black N (1996) Why we need observational studies to evaluate the effectiveness of health care. BMJ 312: 1215–1218.
  35. 35. Huedo-Medina TB, Sanchez-Meca J, Marin-Martinez F, Botella J (2006) Assessing Heterogeneity in Meta-Analysis: Q Statistic or I2 Index? Psychol Methods 11: 193–206.
  36. 36. Sterne JA, Gavaghan D, Egger M (2000) Publication and related bias in meta-analysis: Power of statistical tests and prevalence in the literature. J Clin Epidemiol 53: 1119–1129.
  37. 37. Del Ser T, Hachinski V, Merskey H, Mannoz DG (1999) An autopsy-verified study of the effect of education on degenerative dementia. Brain 122: 2309–2319.
  38. 38. Andel R, Crowe M, Pedersen NL, Mortimer J, Crimmins E, et al. (2005) Complexity of work and risk of Alzheimer’s disease: a population-based study of Swedish twins. J Gerontol B Psychol Sci Soc Sci 60: 251–258.
  39. 39. Bickel H, Cooper B (1994) Incidence and relative risk of dementia in an urban elderly population: findings of a prospective field study. Psychol Med 24: 179–192.
  40. 40. Helmer C, Damon D, Letenneur L, Fabrigoule C, Barberger-Gateau P, et al. (1999) Marital status and risk of Alzheimer’s disease: a French population-based cohort study. Neurology 53: 1953–1958.
  41. 41. Karp A, Paillard-Borg S, Wang HX, Silverstein M, Winblad B, et al. (2006) Mental, physical and social components in leisure activities equally contribute to decrease dementia risk. Dement Geriatr Cogn Disord 21: 65–73.
  42. 42. Ritchie K, Carrieré I, Ritchie CW, Berr C, Artero S, et al. (2010) Designing prevention programmes to reduce incidence of dementia: perspective cohort study of modifiable risk factors. BMJ 341: c3885.