Introduction

Normal aging, as well as various clinical diseases, such as for example dementia, are accompanied by a deterioration of cognitive function. Even though memory decline is a hallmark of dementia, other cognitive domains, like executive function and processing speed are also often affected [1]. Many studies focus on persons in pre-clinical stages of dementia, i.e. mild cognitive impairment, and therefore are not always generalizable to community-dwelling elderly [2–4]. Still, cognitive aging has also been investigated extensively outside the context of dementia. Age effects have been documented on several cognitive domains, such as spatial orientation, inductive reasoning, memory, verbal and number skills, and in a variety of populations [5, 6]. However, different rates of cognitive decline across cohorts have also been reported and age effects on cognition could be altered over time due to changes in a population with regard to, for example, education, environment, health factors, or employment [7–9]. Therefore, more contemporary data on aging effects on cognition are needed.

In order to gain a comprehensive understanding of cognitive function in non-demented elderly, it is essential to study a broad range of cognitive domains in unselected community-dwelling persons. Furthermore, in addition to studying separate domains, it is equally important to investigate global cognition. The rationale for this is that cognition consists of a general underlying construct that is domain-independent and reflects an individual’s general cognitive function. This construct is linked to intelligence and can be quantified as a general cognitive factor, or g-factor. The g-factor is a stable concept, comprising the shared variance between cognitive tests, and can be interpreted as a common underlying factor to a variety of cognitive domains [10–12]. The g-factor has even been shown to be independent of cognitive test batteries used, and can therefore be easily generalized to other studies [13].

The aim of this study was to investigate patterns of cognitive function in middle-aged and elderly community-dwelling persons. We specifically studied both general cognition, using the g-factor, as well as specific cognitive domains.

Methods

Setting

The study is embedded within the Rotterdam Study, a population-based cohort study in middle-aged and elderly participants that started in 1990 and aims to investigate frequency, causes and determinants of chronic diseases [14]. The initial cohort encompassed 7,983 persons and was expanded by 3,011 persons in 2000 and by 3,932 persons in 2005. In-persons examinations take place every 3–4 years and consists of home interview and three center visits. The institutional review board of Erasmus MC approved the study and participants gave written informed consent.

Study population

Table 1 shows the number of participants from each cohort used in this study. Also, age at time of invitation to the study, sex and if available level of education are given for participants and non-participants to the study (Supplementary table 1). Additionally, we show age and sex of participants by year (Supplementary Table 2). The current cross-sectional study focuses on the period from January 1st 2008 onwards, because only then the full cognitive test battery in its current format was implemented. From the persons who responded to the invitation to participate in the study (n = 7,963), persons with a stroke (n = 325) or prevalent dementia (n = 73) were excluded from the sample used in this study. Sixteen persons had both a stroke and dementia and were excluded. For dementia, the assessment is based on a two-step procedure, which has been published before [15]. It involves screening by mini-mental state examination (MMSE), additional work-up by CAMDEX, informant interview, additional neuropsychological assessment, imaging, and final diagnosis in a consensus meeting led by a neurologist. For stroke, the assessment is based on self-report, family doctor files, and files of medical specialists, which are all discussed in a consensus panel led by a neurologist [16]. Also, neuroimaging is used if required.

Table 1 Participation to the current study presented per cohort
Full size table

Until February 29th 2012, cognitive tests were performed in 3,706 up to 4,176 persons. In case of technical problems, refusal of participation, physical limitations, or deviation from instructions, test results were excluded. This explains the range in number of subjects that performed various cognitive tests. The number of persons in the study who completed a valid cognitive test result on any of the tests used was 4,422 (Table 1). The complete cognitive test battery was available in 3,021 persons.

Cognitive test battery

During two separate center visits a cognitive test battery was administered, which included MMSE [17], Stroop test [18], letter-digit substitution task (LDST) [19], verbal fluency test [20], 15-word verbal learning test (15-WLT) [21], design organization test (DOT) [22] and Purdue pegboard test [23]. A description of the cognitive tests, test demands and latent skills measured is given in Table 2. Level of education was obtained and categorized into seven levels, ranging from primary to university education. Higher scores indicate a better performance on all cognitive tests, except for the Stroop task in which a higher score indicates a worse performance. Scores for the Stroop task were thus inverted for better comparison to other tests. The DOT is a test which is based on and highly correlated to WAIS-III block design, but is administered in two rather than 10 min and is less dependent on motor skills than the block design test [22]. Test score on the DOT has a range from 0 to 56 points for each subject.

Table 2 Description of cognitive tests
Full size table

G-factor [12]

To calculate a general cognitive factor (g-factor) we performed a principal component analysis incorporating color-word interference subtask of the Stroop test, LDST, verbal fluency test, delayed recall score of the 15-WLT, DOT and Purdue pegboard test. For tests with multiple subtasks we chose only one subtask in order to prevent highly correlated tasks distorting the factor loadings. Principal component analysis was performed on complete case data of 3,021 persons. The g-factor was identified as the first unrotated component of the principal component analysis and explained 49.2 % of all variance in the cognitive tests. This is a typical amount of variance accounted for by the g-factor [12].

Statistical analysis

To aid comparison across cognitive tests we first calculated z-scores for cognitive test scores. The MMSE score was not standardized due to its skewed nature. We used analysis of covariance to compare scores between men and women, adjusting for level of education. We used linear regression models to investigate the continuous association between age and cognitive test score, corrected for level of education. In additional analyses we used subcohort as an extra covariate to the linear regression model to test for cohort effects. We used Z tests to formally test differences of age effects between cognitive tests. We tested interaction effects between age and sex and explored non-linear effects of age on cognition. All analyses were performed using the statistical software package SPSS version 20.0 for Windows. Results are presented with 95 % confidence intervals (CI).

Results

Mean age was 71.9 years (SD = 9.7), with 57.5 % women (Table 3). Men scored better than women on the DOT, whereas women scored better on Stroop color naming, immediate recall, delayed recall and recognition parts of the 15-WLT, and Purdue pegboard test (Table 3). Pearson correlation coefficients between all cognitive test scores are shown in Supplementary table 3.

Table 3 Characteristics of the study population
Full size table

Figure 1 illustrates MMSE score and g-factor in 5-year strata of age. MMSE score stayed stable until age 70 and then showed a rapid decline. In contrast, the g-factor showed decline in scores already from age 45 onwards. The mean decline in g-factor per 10 year increase in age was −0.59 (95 % CI −0.62 to −0.56). For both MMSE score and g-factor we also found a quadratic effect of age (Table 4).

Fig. 1
figure 1

Age effects on global cognitive scores. The x-axis represents age per 5 years and the y-axis represents the MMSE-score or z-score of the g-factor. Error bars represent 95 % confidence intervals. Estimates are adjusted for level of education. MMSE mini mental state examination, g-factor general cognitive factor

Full size image
Table 4 Association of age with global cognitive function
Full size table

Figure 2 shows mean test scores in 5-year strata of age. We found the strongest decline for Purdue pegboard, LDST, DOT and Stroop interference task (Table 5). In contrast, smaller effects of age were found for 15-WLT immediate recall (−0.25 per 10 years; 95 % CI −0.28 to −0.22), delayed recall (−0.23 per 10 years; 95 % CI −0.26 to −0.20) and recognition (−0.09 per 10 years; 95 % CI −0.12 to −0.05). These differences in age effects between the memory subtasks versus Purdue pegboard, LDST, DOT and Stroop were confirmed by formal statistical testing (Z tests). For example, the age effects on the Purdue pegboard test or DOT were both significantly larger than the effect on immediate recall (P < 0.001). Still, the strongest effects of age were on the g-factor, rather than any individual cognitive test.

Fig. 2
figure 2

Cognitive function in 5-years bins. The x-axis represents age per 5 years and the y-axis represents the z-score on the test. Error bars represent 95 % confidence intervals. All estimates are adjusted for level of education. 15-WLT 15-Word learning test

Full size image
Table 5 Association of age with cognitive test scores
Full size table

Finally, we found that for the Purdue pegboard test and LDST, age effects were stronger in women than men. Also, quadratic effects of age on cognition were found for the Stroop tasks, the LDST, verbal fluency, and the Purdue pegboard test. Adding subcohort as an extra covariate to the model did not reduce the effects of age on cognitive scores.

Discussion

In a large community-dwelling cohort of persons 45 years and older, we found that age strongly affects general cognitive function, measured by the g-factor. The effect of age on general cognitive function was already apparent from 45 years onwards. Investigating separate cognitive domains, we found strongest associations of age with fine motor skill, processing speed, and visuospatial ability.

Strengths of this study include the large community-dwelling study sample and availability of multiple cognitive tests. An important limitation to the interpretation of our results is the cross-sectional design. Also, relations between age and cognition could partly be influenced by cohort effects. However, differences in age effects across cognitive tests are comparable since all analyses were performed on the same group of persons. Another problem is that not all cognitive tests were completed by all participants to our study and that participants are younger and usually in better health compared to non-participants [24]. Therefore, in our g-factor analyses, we selected a sample with fully available cognitive data. We should keep in mind that this may have introduced some selection bias and has may reduce the generalizability of the results. We also note that in order to summarize the different cognitive tests into one g-factor, we selected six cognitive test variables under the assumption that these are representatives of various cognitive domains (executive function, processing speed, verbal fluency, memory, visuospatial ability, and fine motor skill), which are frequently used in cognitive aging research. Other studies may select different tests to construct a g-factor and will possibly get a slightly different outcome. However, it was previously found that g-factors constructed from variable test batteries result in factors that are highly correlated [13]. Thus, the g-factor is likely to be a stable concept. It is comprised of shared variance between tests, and can be interpreted as a factor which is common to a variety of cognitive domains.

In this study sample, we showed that the g-factor is affected already from age 45 onwards. Also, compared to the other cognitive tests in our battery, the g-factor was most strongly related to age. The strength of relation between age and cognition was consistent with those found by others [25, 26]. MMSE score only showed a decline from age 70 onwards. The MMSE is often used to test global cognitive function in older adults, yet it has frequently been criticised for its ceiling effect [27, 28]. In agreement with a large study of healthy elderly, we did not find strong effects of age on MMSE score [29].

Among our other cognitive tests, we found that fine motor skill, processing speed and visuospatial ability were most affected by age. In agreement with the observed relation between age and visuospatial ability, WAIS-III block design performance starts to decline from the mid-forties onward [12]. Other studies have also suggested a more prominent role for decline in visuospatial ability in aging research [30, 31]. One study reported a composite score of visuospatial ability to be a significant predictor of developing cognitive decline [2]. However, another large cohort study reported relatively small effects of age on visuospatial ability [4]. Already in the youngest age groups we found an effect of age on performance on the Purdue pegboard test. Population studies in the healthy elderly that looked into age effects on fine motor skills are scarce. The relatively large age effects on the LDST are in line with previous studies showing strong age effects on processing speed [9, 32]. Interestingly, these findings are supported by indirect evidence from neuroimaging studies which found that white matter declined faster than grey matter and white matter deterioration was associated with decline in motor skill and tasks of processing speed [33–35]. However, others concluded there is a relative stability of white matter volume in aging [36, 37]. The effect size we found relating age to memory was small compared to age effects on other cognitive scores. Again, this is in line with evidence showing that memory function is more dependent on grey matter which decreases gradually with aging [38, 39]. Furthermore, we found that women scored better on memory tests than men, which is in accordance with previous findings that women have better verbal memory than men [40, 41]. No difference in age effects on memory was found between men and women. It is expected that memory would be more strongly affected in dementia rather than normal aging. The exclusion of prevalent dementia cases from our study possibly contributed to the small negative effects of age on memory. However, there is a continuum between normal cognitive aging and dementia, and persons in the preclinical stages of dementia were not excluded from the study population. Normal cognitive aging research has often found that the more frontal brain functions such as attention and executive function are affected earlier than memory [42–44]. The relatively smaller effect on the verbal fluency test may reflect the fact that we used a category fluency test rather than a phonemic fluency test. Category fluency places a larger demand on memory performance rather than frontal lobe function [45, 46]. Furthermore, we found a stronger effect on the color-word interference subtask of the Stroop, compared to the reading and naming subtasks. The Stroop color-word interference task requires more cognitive control than the first two subtasks and is more dependent on executive function, specifically on attention and inhibition [19].

In conclusion, in persons of 45 years and older, age is most strongly related general cognitive function. Our findings also suggest that not memory, but fine motor skill, processing speed, and visuospatial ability are affected most by advancing age.