This study confirmed the influence of socio-demographic characteristics on cognitive performance of adults, with similar findings concerning the small influence of sex. However, it also showed that education rather than age was the most impactful factor for performance in all cognitive tests [18, 19].
As cohort studies usually employ neuropsychological tests to determine the cognitive decline over time, it is important that these tests present the smallest possible influence from other variables than age. The level of schooling seems to be an important covariable in the analysis of cognitive decline, mainly in cohort studies in underdeveloped countries that use tests originally designed to evaluate cognition from developed countries populations. The education was also significantly more important than age on the performance on the VTF among Brazilian elderly [18, 19]. An opposite result was observed in normative data for Trail B performed among 911 Canadian adults (18–89 years old), where age, and not education, had a stronger effect on the performance of this test in this population with a higher level of schooling [20].
Level of education can interact with cognition in different ways. An interesting hypothesis concerning the association between age-related cognitive decline and education proposes three different patterns: a) there would be the same degree of cognitive age-related decline within different educational levels (parallelism), b) less age-related cognitive decline within the well-educated (protection) or c) an age-related reduction of the initial advantage exhibited by the young and middle-aged well-educated group (confluence) [21, 22]. In this study, the parallelism phenomenon may be at play for all tests, but the cross-sectional nature of the data on cognitive status limits inference for cognitive decline over time.
Most of the participants of ELSA-Brasil included in this study are middle-aged. Women (54.2 %) are slightly more represented than men. Women outperformed men on verbal learning and verbal fluency tests, and men outperformed women on visuospatial ability. It is likely that sex hormones influence cognitive domains, but our study design does not permit this evaluation [23].
Memory tests that use a list of simple words should represent the capacity to learn, retain and utilize information, which requires the intact functioning of many brain regions, including some especially susceptible to injury or diseases [24]. The exclusion of psychiatric and neurological diseases, as well the younger composition of the ELSA-Brasil population, may explain why memory tests have been less affected than executive function tests (VFT and Trail B) by sociodemographic variables.
The mean number of correct words in category VFT was similar to that of other studies both in developed and under-developed countries [18, 19, 24]. Verbal fluency requires the systematic retrieval of organized information from semantic memory and, apparently, education facilitates this semantic access, widening an individual’s repertoire of words [24]. The magnitude of the influence of education may differ among populations. A normative study for population-based elderly from undeveloped countries observed similar scores, but with a considerable effect of country site, mainly on the performance of category VFT, after accounting for compositional differences in age, education and sex [25].
The strong influence of education on Trail B performance may be related to poor reading habits among the population studied. Another factor to be considered is the challenge posed by the presence of the letter K in the trail, as this letter is not familiar to Portuguese speakers and had even been abolished from the Brazilian alphabet from 1943 to 1990.
Although there was observed a high compliance to the cognitive evaluation, both age and education influenced the response rates to cognitive tests. Oldest and less educated participants were less prone to perform the cognitive tests, especially the Trail B, and this bias may have contributed to overestimate the median scores in these tests, since these losses presumably would have caused worse results on cognitive evaluation.
The strengths of this study rely on its large adult population from a medium income country and on the high quality assurance of measures. A limitation to this cognitive evaluation is the lack of reasoning tests. In developed countries, the lack of intelligence quotient scores may be not considered a serious limitation given the well-known positive association between education and IQ [21, 22]. This situation could not be generalized to countries such as Brazil, where social inequalities influence the level of education, regardless of intelligence. It is important to consider introducing this evaluation in the ELSA-Brasil follow-up.
Level of schooling among studied participants is higher than that of the Brazilian general population. This probably reflects the social selectivity of civil servants. According to the last national survey in 2010 [26], 11.3 % of Brazilian adults (25+ years-old) had a tertiary education and 49.3 % had less than eight years of schooling (IBGE), as opposed to 52.7 % and 5.8 % of ELSA-Brasil participants, respectively. Although these participants do not represent the Brazilian population, the results of the longitudinal evaluation on cognitive decline, mainly among the young adults and middle-aged participants, may result in significant contributions to better understand, prevent or delay cognitive decline.
When investigating other social determinants of low cognitive performance in this population, we showed that early exposure to adverse social and nutritional conditions seems to be also detrimental to cognitive performance, independent of educational achievement. A study with 12,997 participants (35–64 years) of ELSA-Brasil revealed lower maternal education increased the chances of poor performance in all cognitive tests, with a dose-response gradient. Low birth-weight was related to poor performance in Trail B and greater trunk length decreased the chances of poor performance in both verbal fluency tests and Trail B [27].
In order to perform cross-sectional analyses, we are using z-scores standardized to give a normal curve to each cognitive test. To control the effect of age and education on scores, the subjects were stratified into the three age groups (34–44–45–64 and 65+ years) and the four levels of education (<8, 8–10, 11–14 and 14+ years of schooling) we used in this study. Lowered global cognitive function is defined as one or more standard deviation below average in each age and education strata [4, 28]. Besides, as our study aims to perform multiple cognitive tests over time, we will be able to investigate the direction and magnitude of the influence of education on aging cognitive decline of this cohort of workers from a developing country.