Education and the incidence of dementia in a large population-based study: The Rotterdam Study

Methods.

Results.

Table 1 shows baseline characteristics of the study population. The average age at baseline of men was slightly lower than that of women (68.2 versus 70.0); however, in the highest age category, which is at highest risk of dementia, women were over-represented. In this elderly population, men received more schooling than women: 50.6% of men versus 24.3% of women had a high level of education, whereas 26.4% of men versus 46.1% of women were low-educated.

Duration of follow-up was on average 2.1 years, ranging from a few days for someone who died to 6.1 years (first quartile 1.4, median 1.9, third quartile 2.6 years). During follow-up, we diagnosed incident dementia in 137 persons. Of those, 97 had AD (71%) and 18 had vascular dementia (13%). Most patients had mild dementia; only one fifth had a moderately severe dementia. Age-specific incidence rates of dementia did not differ significantly between men and women.

Table 2 gives results on the incidence of total dementia by level of education. Among women, those with lowest education had the highest and those with highest education the lowest incidence of dementia in all age categories. For men there was no clear pattern.

We found a significant interaction between gender and education (p = 0.023). The relation between education and the risk of dementia was significant in women (table 3), but not in men. Comparing all gender and level-of-education combinations showed that the reference group of highly educated women had the lowest relative risk of dementia. Relative risks of low, medium, and highly educated men were 1.2 (95% CI 0.5 to 2.8), 1.7 (0.8 to 4.1), and 1.8 (0.9 to 3.7), and did not significantly differ from the reference group.

Results for AD did not differ from those for total dementia (table 3). Because of limited power, we did not analyze the association with other subtypes of dementia.

Discussion.

This prospective study has several advantages over others that examined the association between education and dementia. The study is population-based with a high response rate. We also obtained complete follow-up of all participants at risk. Therefore, there is little room for selection bias. As opposed to register-based studies, we actively screened for dementia, which prevented bias by differences in health care utilization. The study was prospective and we assessed educational level before onset of dementia, thereby minimizing recall bias. In cross-sectional studies, persons with lifelong low cognitive abilities may erroneously be diagnosed as demented.1,4 This diagnostic bias is unlikely in the current study of incident dementia. In addition, we used similar screening and diagnostic procedures at baseline and follow-up. One possible diagnostic bias in our and other studies may have occurred because screening tests are more sensitive in picking up cognitive deficits associated with early dementia in lower than in more highly educated subjects; this could result in an underestimate of the risk of dementia in more highly educated persons.

We previously described the association between education and prevalent dementia.6 Similar to most other cross-sectional studies,1 our previous study suggested a stronger association than did our current results (age- and sex-adjusted odds ratio of low education and dementia 2.7, 95% CI 1.9 to 4.0). In that study, however, we could not avoid all the abovementioned biases. Owing to selection, disproportionately few higher-educated dementia patients may have been included. Stern et al. found that highly educated patients survived shorter than lower educated ones,9 and hypothesized that dementia in the higher educated remains unrecognized until a more progressed stage. However, a later study could not replicate these findings.10 A prevalence study of dementia is based on existing cases of dementia from which the proportion of dementia in a population is calculated. Prevalent dementia case series consist of dementia patients diagnosed in a prevalence study. It may be that high educated dementia patients are less willing than are low educated patients to participate in a prevalence study. Still, their risk of dementia may be similar. Our results may illustrate that cross-sectional studies overrate the relation of education with dementia risk.

The current results suggest that sex modifies the association between education and dementia. However, relative risk estimates in men were unstable, as reflected in wide confidence intervals. In line with the age distribution of the general population, there were more women than men in our study, in particular in the oldest age category. It is unlikely, however, that this explains the gender differences, because the difference seemed present in all age categories (see table 2). In regression analyses we thoroughly adjusted for age. A difference in the association by sex has not been described before. Reanalysis of our cross-sectional data also revealed a difference: the relation between low education and dementia was much stronger in women than in men (odds ratio 4.7 versus 1.8), which supports our current observation. In our study population, as a result of cohort effects, women were on average less well educated. One could speculate that in this age group educational level represents a different quality for men than for women. Education could, in women more than in men, reflect socioeconomic status rather than intellectual capacity. Alternatively, it is conceivable that the interaction between sex and education in the risk of dementia resulted from a sex difference in education-related dementia prognosis or risk factor profile.