Cognitive function in a general population of men and women: a cross sectional study in the European Investigation of Cancer–Norfolk cohort (EPIC-Norfolk)

Ageing is generally associated with memory impairment and cognitive decline, however, this decline is not inevitable [1] and not all domains of cognitive function are equally affected with age [2]. A broad range of cognitive capability is observed in the older population [3] as well as substantial inter-individual heterogeneity in rates of decline [4]. The range encompasses high cognitive functioning even in the very old [5], mild cognitive impairment (MCI), through to dementia at the other end of the spectrum.

Mild Cognitive Impairment (MCI) is described as the transitional state between normal cognitive ageing and dementia [6], with detectable impairment to memory or cognitive abilities when compared to healthy controls, but not to the extent as seen in individuals suffering with mild dementia [7]. The amnestic form of MCI has been shown to be predictive of dementia [7‐10], however, the conversion rate has been shown to vary significantly [7] and only a minority of individuals with MCI progress to dementia within clinically relevant time frames. As a result of the heterogeneity in both the aetiology and outcome of this condition, MCI is still an area of major debate, with no consensus on its classification [11]. Reliable identification of those individuals with MCI who remain stable compared to those who decline would maximise efficacy of potential treatments and preventive interventions around this transitional stage [12]. To determine the factors that contribute to this variability in not just the rates of decline in MCI, but also in the different cognitive abilities, will help to improve the understanding of the natural progression of decline in an ageing population.

Substantial data already exist on dementia and cognitive impairment, mainly in the older population, from using a wide range of instruments, each with merits and limitations that assess different aspects or stages of cognition. Episodic memory deficits have been shown in a number of studies to be associated with the strongest and most persistent risk of cognitive decline [13, 14] and are the most common and earliest complaints in MCI [15]. However deficits in other cognitive domains can also occur, some early on, including attention, executive functioning, prospective memory, semantic memory, verbal ability, visuospatial skills, attention and processing speed [16‐19]. There is a need for assessments to cover a broad range of ability and domains, and have an optimal balance between sensitivity and specificity with high positive predictive value in the settings in which they will be applied.

There are no cures for dementia, but there are some advances in the development of drugs [20] that are known to improve symptoms, or temporarily slow down disease progression in early and middle stages of the disease. To exploit the potential benefits of any such treatment and to facilitate decision making for future plans, it is important to identify early indicators of decline. However, before we are able to advise guidelines and policies on health, we need to gain better insight into the ageing process in the general population and the range of functions in both domain specific and global cognition.

The primary aim of this study was to obtain data from a general population of men and women without overt cognitive impairment using a comprehensive cognitive test battery assessing a range of function including memory (retrospective and prospective), executive function, attention, calculation, registration, language, praxis, abstract thinking, processing and new learning. The secondary aim was to explore the comparability of the different tests and their use in a community setting. We focused on producing a standardised protocol for test administration and scoring, thus minimising variation, differences in interpretation and reducing subjectivity. Details of the standardisation methods are given here.

The European Prospective Investigation of Cancer (EPIC) is a 10-country collaboration studying diet and disease with half a million participants [21, 22], of which EPIC-Norfolk is one of the UK centres. Detailed descriptions of recruitment and study methods at baseline have been reported elsewhere [23]. Briefly, participants were recruited at baseline through registers in thirty five general practices in Norfolk between 1993 and 1997. General practice registers approximate population based registers as virtually all the population are registered through the UK National Health Service. Participants who consented at baseline were re-invited for a health examination at subsequent phases. The data presented here is from the third health examination (3HC or EPIC-Norfolk 3) conducted between 2006 and 2011 (including data from the pilot phase between 2004 and 2006) [24]. The characteristics of the participants taking part in EPIC-Norfolk 3 are given in Table 1. Attrition rates and characteristics of these returning participants have been described previously where the cohort was shown to represent a diverse population [24]. This study was approved by the Norfolk Local Research Ethics Committee (05/Q0101/191) and East Norfolk and Waveney NHS Research Governance Committee (2005EC07L). Participants gave signed informed consent at both baseline and then subsequently at the 3HC to cover new measures that were not present in previous health examinations.

The Cognition battery used in EPIC-Norfolk 3 (EPIC-COG) comprised of a number of validated tests that have all been described previously. These included a shortened version of the Extended Mental State Exam [25] (SF-EMSE), letter cancellation task [26] as used in the Medical Research Council Cognitive Function and Ageing Study (MRC CFAS) [27], Hopkins Verbal Learning Test (HVLT) [28], Cambridge Neuropsychological Test Automated Battery Paired Associates Learning Test (CANTAB-PAL) [29‐31], Visual Sensitivity Test (VST) to assess visual impairment deficits contributing to cognitive impairment [32] Short National Adult Reading Test (Short-NART) [33] and a test for prospective memory (also as described in MRC CFAS) [34]. These tests are briefly described here with further information given in the Additional file 1. The tests were selected with the intention to cover an array of cognitive domains and a range of difficulty. Modifications were made to shorten some of the tests to allow for use in an epidemiological setting. The battery was part of a broader health examination that lasted approximately 2–3 hours depending on the participant.

Of the 8623 participants attending EPIC-Norfolk 3, 45% (n = 3861) were men and 55% (n = 4762) were women. Cognitive data were collected on 8585 individuals, with over 90% completing or attempting six or all seven of the tests in the battery. The CANTAB-PAL and VST had slightly lower completion rate (at 86.5% and 83% respectively), partly due to a technical computer failure, resulting in loss of data on 150 participants. Table 2 summarises the cognitive domains covered by each of the tests used in this study and the number of participants completing each test component.

Men and women were equally likely to complete all seven components, with 6011 (69.7%) participants attempting all seven test components of EPIC-COG and only 850 participants (less than 10%) of those taking part attempting or completing five components or fewer. Those completing all seven components were slightly younger (mean age 68.8 compared to 70.9 years in men and 67.4 compared to 69.8 years in women), and were more likely to be either in paid employment or actively taking part in regular social networks when compared those who did not complete all the tests. More men had left school with qualifications than women (77.8% compared to 70.3%), although more women participated in regular social activities (68.5% for women compared to 59.5% for men). Women also reported more mental stimulating activities in their leisure time (with mean mental activity score of 23.2 in women as compared to 21.1 for men).Distributions for cognitive function (by each test component and stratified by sex) are presented in Figure 1. The EMSE distribution had a negative skew but did not have the same strong ceiling effect as the SF-MMSE scores in this cohort (distribution not shown), with 2298 (27%) of participants scoring the maximum SF- MMSE score of 29 as compared to only 2.4% (n = 200) scoring the maximum EMSE score of 37. The data for letter cancellation (PW Accuracy) and HVLT Total Score were both approximately normal distributed, as was the distribution for FTMS (other than a peak at score 0, which suggests that a high proportion of participants were unable to achieve a correct response immediately). The data for reaction time of the VST were highly positively skewed (as a result of a few extreme, but genuine slow responders).

For the short NART, there was a peak at the error score of 24, followed by alternating peaks and troughs in data giving a ‘comb’ distribution. This pattern in the distribution is as a result of the short NART algorithm (as described in the Additional file 1). The peaks in the distribution can be attributed to those assigned an error score by the algorithm, artificially inflating the scores at these points. The greatest peak (and the starting point of this comb effect of the data) was seen to occur at the cut-off point of score of 20 (giving a full NART error of 24), which was the point where participants with this score or lower, did not continue with the second half of the test.The median scores of each of the test components were plotted with age in men and women as shown in Figure 2a-2g. The data presented here are cross-sectional values showing the association of scores with age group. The graphical presentations in Figure 2 indicate that median scores decline with age. The proportion of participants successfully completing the prospective memory task lowers with increased age group. In the case of the VST (Figure 2f), the median reaction time increases with age group. The NART error score showed an increase with age initially, remains steady then a slight reduction in the oldest group (Figure 2g). In almost all tests, women generally performed better than men.The data were further characterised by calculating percentile scores plotted by age group (Figure 3). In this figure, lower scores are seen across increasing age group. Higher percentiles of cognitive scores remain reasonably stable across age groups, but the spread and variation in scores becoming greater across each age group, with the lowest percentile having markedly lower performance. The variation in scores was least for the SF-EMSE compared to the other tests in the battery. For reaction time and short NART error score, the 99th percentile indicated the poorest performers. The short NART error score exhibited widest variation across age and even some improvement in scores in the oldest age groups.

In this cross-sectional study, we report on cognitive function profiles across a range of domains using previously validated instruments in a general population of men and women from mid to later life (age 48–92 years). We focused mainly on age related differences in cognition measures, and found that, despite the EPIC-COG being a relatively long battery, it was well tolerated by this general study population.

Individuals can have impairment in one cognitive domain but perform well in another or a number of cognitive deficits can occur concurrently. There is increasing evidence of substantial variability in cognitive abilities within individuals. Hilborn et al., (2009) termed this variability in performance across different tasks within an individual as ‘dispersion’. It is important to gain a better understanding of the dispersion displayed by healthy individuals in order to allow accurate clinical judgment on unhealthy or abnormal dispersion [51].

Previous studies have shown that a variety of cognitive deficits are associated with preclinical stages of different types of dementias and that decline can occur in a number of cognitive domains, even before any of the clinical criteria of early stages of cognitive decline are met [52‐54]. Our findings are consistent with previous studies that have shown variability and dispersion across different cognitive domains [51, 55] in older people. Further investigation is necessary to confirm whether the differences observed across the domains provide any meaningful indicators of cognitive performance over time.

The frequency distribution and data from the pilot (where the full NART was used for 300 participants) indicate that the short NART equation does not hold so well in this heterogeneous ageing population as it did in the homogenous sample of elderly women living in a rural setting that it was initially tested on. The aim of the short- NART was to lessen participant load and anxiety, however in an assessment such as the one in this study; participants were as likely to be anxious for any of the other components. The pilot data show that participants who obtained a score of 20 (n = 82) for the first half of NART obtained a mean score of 7.9 (SD = 3.7) for the second half, which is better than what the algorithm predicts. Therefore, we would recommend that despite the appeal of the short NART, this protocol is not appropriate for a higher functioning population where few individuals have poor reading skills. If there are concerns about causing distress, then we would suggest lowering the cut-off to at least 17 rather than having the cut-off at 20.

Prevalence of severe cognitive impairment is relatively low in our cohort but as with previous studies [56, 57], we also found age to be inversely associated with the complete range of cognitive function being tested with the exception of NART, where there did not seem to be any strong association between the short NART score and age. This confirms previous findings that NART is a good measure of pre-morbidity [44, 45] and that age has little or no effect on NART performance in the absence of early dementia [58].

We observed some differences in abilities across gender and age categories. These may have arisen because of age and sex differences in education status. Although we have not investigated education in detail here, there is good heterogeneity of individuals leaving school with and without qualifications in all age groups analysed. We also observed a trend not noted in other studies, of women performing better than men in all the test components across all age groups, even though more men than women reported leaving school with some qualification suggesting that educational status would not explain the sex differences seen in this cohort. However, women reported to be more socially active and to be doing more mentally stimulating activities in their spare time, which have been linked to better cognitive performance at older age. The effects of education, social networks and mental activity will be the focus of further research.Although the test scores generally decline across age groups with the widest variation seen in the oldest age group, there is still a range of capability from poor to high performance in each age band, with some participants from the oldest age group outperforming their younger counterparts. Percentile scores from our cross-sectional data (Figure 3) show that the greatest decline in all test components is seen in poor performers across all the age groups. This graphical presentation can be used to compare scores or estimate age and sex adjusted scores across the different domains investigated here. Individuals scoring below the 25th percentile could be considered as cognitively impaired for that domain and require further investigation.

The MMSE’s lack of ability to assess individual domains and its poor sensitivity to mild cognitive impairments are frequently cited limitations [59, 60] This is because most healthy individuals can successfully answer most of the test items. Even though more demanding tests are available [25, 36, 60, 61], the MMSE still remains the most widely used and cited test of global cognition. To allow comparability with other studies, we included the SF-MMSE to provide a baseline for future studies and as a means of comparing the psychometric qualities and utility of the other tests used in the battery. The MMSE, being a global measure of cognition contains items that test the same domains of memory function as the other components of EPIC-COG. As expected, a positive trend was seen in all the tests with increased SF-MMSE score category, however, a range of scores were seen for other tests in each of the SF-MMSE categories.

On further investigation of participants obtaining the maximum score of 29 on the SF-MMSE, we found in the range of 6-8% of men and 2-5% of women also scored in the bottom 10th percentile of the other tests (Table 5). On examining those individuals who obtained the top 25% and further in the top 10% SF-EMSE scores, there were still participants in poorest performers (10th percentile) of the other tests, although the figures were reduced. Those scoring in the bottom 10th percentile tended to be older than those with scores above the 10th percentile for all the tests apart from the Short NART The number of participants scoring the maximum possible on the SF-EMSE was small (n = 200). None of these individuals scored in the bottom 10th percentile of the other tests other than one person who scored in the lowest 10th percentile for the NART. We show that obtaining a perfect score on the MMSE does not indicate absence of impairment and this confirms previous findings for the need to supplement the MMSE in cognitive testing [37], particularly in a normal to high functioning population.

The limited reliability and validity of the MMSE in a general population has been attributed to the restricted range of MMSE. The EMSE has been shown to be sensitive across a range of performance, to avoid the ceiling effect and that (even in its short form as it has been used here), the EMSE provides extended coverage of cognitive domains (extending on attention, memory, processing and executive function). This report confirms previous findings that the EMSE has advantages over the MMSE particularly for testing individuals at the high end of the performance range [25].

Spearman’s rank correlation (Table 6) show correlations are moderate to weak for most tests with HVLT having the strongest associations with other test components especially with FTMS and NART. This is not surprising, as HVLT, FTMS and NART assess similar cognitive sub domains of memory and language, however the moderate degree of correlation is somewhat counterintuitive as we would expect this to be higher. The contour plots (Figure 4) depict the spread of scores (and area of overlap) indicating that with some association, there is also some non-systematic scatter of scores suggesting that these tests may be assessing different aspects of cognitive function.

We have also addressed some practical and methodological issues with regards to minimising variability and subjectivity that can be introduced at any part of administration, scoring or cleaning of the data. We have described methods of standardisation in an epidemiological setting to ensure accuracy and consistency. Having these methods standardised and documented is also extremely important to allow comparability and potential harmonisation of data with other studies. The other advantage of this study over previous studies is that it has been conducted in a large well characterised cohort of men and women with good representation from a very wide age range, which has been a limiting factor in some previous studies [27, 62‐64].

Everyday activities in the real world are complex, requiring independence, planning, organisation, sequencing and judgement and have been shown to be a significant predictor of functional status [69]. Therefore, assessing cognitive function in a range of domains such as executive functions, planning, flexibility, abstract thinking, semantic memory as well as episodic memory is vital. Also of considerable importance is to accurately identify early decline in individuals or those with MCI who are known to be at increased risk of dementia, particularly Alzheimer’s disease (AD) compared to older people without any obvious cognitive impairment.

Here we have used a comprehensive battery of accurate and well tolerated tests to provide evidence of cognitive function in a number of cognitive domains that have previously been reported to be involved in much earlier stages of decline. We have described how, even though there is reduction in performance across age, there is also a great deal of heterogeneity in older individuals. Further work is needed to understand why cognitive abilities vary so greatly across individuals and cognitive domains and to investigate the more subtle changes in cognition. We have also demonstrated that the EMSE even in its short form provides a better description of the cognitive abilities in a general functioning population and that the short NART protocol is not suitable for a heterogeneous higher functioning population. Finally, careful consideration should be given to the purpose for using a particular test (including whether the aim is to obtain global or domain specific measure, time availability and target population) when selecting an assessment tool for cognitive function.

There is epidemiological evidence of associations between lifestyle factors (such as diet, smoking and exercise) and risk of dementia [70]. The EPIC-Norfolk study has over twenty years of lifestyle, biological and genetic information, collected from mid to late in life. This study is well placed not only to identify factors associated with decline but also factors associated with maintaining abilities in older age. With the data already collected and further follow up data, we can investigate patterns of behaviours over time and predict how those behaviours affect cognitive function.