The CERAD Word List Memory Test: Normative Data Based on a Norwegian Population-Based Sample of Healthy Older Adults 70 Years and Above. The HUNT Study

INTRODUCTION

Memory decline is common in old age but may also be an initial sign of dementia [1]. Differentiating between normal and abnormal memory decline can be challenging. Human memory is complex and comprises several distinct memory systems that involve different networks in the brain [2–4]. Episodic memory, i.e., the conscious recollection of a personal experience that contains information on what has happened and where and when it happened [5], is particularly vulnerable to the effects of aging. Decline in episodic memory performance among older adults is not fully understood but has been linked to volume reductions in the medial temporal lobe, including the hippocampus, and the prefrontal cortex [6, 7]. Impairment in episodic memory is also frequently seen in psychiatric and physical disorders and is a hallmark of dementia due to Alzheimer’s disease (AD) [8–10].

The consequences of severe impairment in episodic memory may be serious for a person’s basic and complex activities of daily living (ADL) functioning, as seen in AD. However, amnestic mild cognitive impairment (aMCI) should also be taken seriously by the clinician as it is a strong risk factor for subsequent progression to dementia [11–13]. Although most definitions of a diagnosis of aMCI exclude impairment in ADL, a substantial number of studies have shown that complex ADL tasks are negatively influenced by impaired memory function [14–16]. Furthermore, aMCI may also affect health-related quality of life [17, 18].

Hence, accurate assessment of episodic memory functioning becomes important and should be included in the examination of older patients when there is suspicion of cognitive impairment or dementia. Word list tests are probably the most commonly used measure of episodic memory performance, and several alternatives exist, e.g., the California Verbal Learning Test-II (CVLT-II) [19], the Auditory Verbal Learning Test (AVLT) [20], the Hopkins Verbal Learning Test-Revised (HVLT-R) [21], and the word list for the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) [22].

Another alternative is the Word List Memory Test (WLMT) of the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neuropsychological test battery [23], which was developed especially for assessment of patients when AD is suspected. The battery is widely used in both clinical and research settings and has been translated into at least 24 different languages worldwide according to the official CERAD homepage (https://sites.duke.edu/centerforaging/cerad/). Its clinical utility for detecting AD in elderly patients is well documented, and the test battery may also be useful for identifying aMCI which may be a precursor to AD [24–27]. Among the tests included in the CERAD neuropsychological test battery, the WLMT is shown to be the most sensitive for screening for aMCI and early AD. The WLMT consists of five subtests—list-learning trials 1–3, delayed recall, and delayed recognition—and, of these, the sensitivity of the delayed recall subtest is best documented [25–28].

A prerequisite for distinguishing normal from pathological performance on a cognitive test is the availability of representative normative data; this also holds true for the CERAD WLMT. Preferably, the normative data should be collected in the country or culture in which the test will be administered, as there are substantial differences in cognitive performance throughout different regions and countries [29–31] and also among cultures within a country, e.g., the ethnic majority versus minority groups [32–35]. Furthermore, the normative data should be relatively new as studies based on single countries or regions have shown improvements in cognitive test scores from generation to generation [36, 37]. Several reasons for this positive trend, often referred to as “the Flynn effect” [38], have been proposed, including better schooling and education systems, improved hygiene and nutrition status, reduction in cardiovascular diseases, more complex and stimulating working lives, and increased working life participation among women [39]. Normative data should be stratified by age, sex, and education, as all three variables have a strong effect on different cognitive test measures, including episodic memory, at least in Western populations [40, 41].

Related to the representativeness issue is the question of sample composition, i.e., which inclusion/exclusion criteria should be applied to make the normative data most useful in clinical settings? Broadly, there are two different methodological approaches; either to include cognitively healthy older adults only and exclude individuals with conditions known to affect cognitive performance, or to include typical older adults and use less stringent/no exclusion criteria which implies that persons with cognitive impairment (and sometimes also dementia) are included in the normative sample [42–44]. As cognitive impairment is common among older people typical aging normative data will necessarily lower the reference range for cognitive health and increase the risk of under-diagnosing cognitive impairment [44]. Therefore, healthy aging normative data is by the majority of authors considered most relevant in clinical questions related to determining severity and etiology of cognitive impairment—even though there may be a risk of over-diagnosing cognitive impairment by following this approach [42–45].

We searched the literature for normative CERAD WLMT data and found a total of 17 studies published in English and one study in German from 2000 up to 2021. The majority of the studies are from North America [46–54], but normative data for the CERAD WLMT have also been developed in other regions of the world, including countries in Europe [26, 55–58], Asia [59, 60], Africa [29], the Middle East [61], and Latin America [59]. Besides involving different regions/countries, the studies differ with respect to study setting (population, community, clinic), study populations (e.g., age groups, education level, language, ethnicity), the applied inclusion/exclusion criteria, and sample sizes. Thus, direct comparisons of the normative data are difficult to make. Although the studies, with some exceptions [29, 59], exclude participants with cognitive impairment and/or dementia, relatively few studies comprise participants from unselected samples [26, 59], and the normative samples are often small. In fact, nine out of the 18 studies have fewer than 325 participants [26, 61]; only five are population-based with more than 1 500 participants [29, 59]. Of these, the normative data comprise older adults, aged 60 years and above, with the exception of Hankee et al. [47], which predominantly includes younger and middle-aged adults.

In general, the studies report a significant effect of age and education on test performance for total list learning and delayed recall of the WLMT; i.e., persons of younger age and/or higher education attain better scores than persons of older age and/or lower education on both. For sex, the results are less consistent, but the tendency is that women perform better than men, at least on total list learning [29, 56–60].

It is worth noting that there can be large differences in the reported normative scores between regions/countries, as exemplified by the studies of Luck et al. (2018) from Leipzig, Germany [56] and Gray et al. (2021) from villages in Nigeria and Tanzania [29]. For a female aged 70 to 74 with formal education, the mean delayed recall score differs by more than four words (7.7 versus 3.6) between the two studies. Albeit an extreme example, this discrepancy underlines the importance of applying normative data with caution, i.e., the use of cognitive tests, including the CERAD WLMT, should be restricted to populations comparable to where they were developed, as differences in culture, language, and the quality of education systems between countries/regions will affect the inhabitants’ performance.

Among the European studies, only one is from Scandinavia [55]. This study comprises demographically adjusted normative data for the CERAD WLMT in a Norwegian sample of 227 healthy persons aged 40 to 80 years, mean age 63.1 years (standard deviation (SD) 8.6). Although the normative data cover an important age interval, the sample is very small and does not include older adults above 80 years of age. Since individual differences in cognitive performance increase with age, normative data of older adults should preferably be based upon large population-based samples to make reliable assessments. Ideally, such data should also include persons older than 80 years. The latter point is particularly important as age is the strongest risk factor for dementia, with an estimated prevalence of 5.6% in the age group 70–74 years in Norway in 2019, increasing to 48.1% in the age group 90 + [62].

In this study, we aim to present normative data for the CERAD WLMT from a large population-based Norwegian sample of healthy older adults aged 70 years and above, stratified by age, sex, and education.

MATERIALS AND METHODS

Data were obtained from two related population-based studies, the Trøndelag Health Study, fourth wave (HUNT4 70+) and the HUNT4 Trondheim 70 + [62, 63]. Both studies included older adults aged 70 years and above. While the design and methods of the two studies are identical, the geographical recruitment areas differ. The HUNT4 70 + study covers the northern part of Trøndelag county in Norway. This region is characterized by rural areas and small towns with fewer than 25,000 inhabitants. In many ways, the population in the northern part of Trøndelag is representative of the Norwegian population in general, but the region has no larger cities and a low share of immigrants, and the number of people with higher education is below the national average [64, 65]. To compensate for these limitations, the HUNT4 Trondheim 70 + study was initiated in 2018. The HUNT4 Trondheim 70 + study covers Trondheim, the third largest city in Norway, with 207,000 inhabitants as of 2021, situated in the southern part of Trøndelag county. Data collection for the HUNT4 70 + study and the HUNT4 Trondheim 70 + study was carried out from September 2017 to March 2019 and from October 2018 to June 2019, respectively. Details of the HUNT study has been described previously [62, 65].

RESULTS

DISCUSSION

In this paper, we present population-based normative CERAD WLMT data from a large sample of cognitively healthy Norwegian older adults aged 70 years and above. We trust that these normative data will fulfill a need among Norwegian clinicians involved in assessments of older adults they suspect may have cognitive impairment or dementia. The normative data will also likely be relevant for Swedish and Danish colleagues as education and social welfare systems are quite similar in the Scandinavian countries. Although the WLMT is widely used, normative data from the Scandinavian region have been available only for ages up to 80 years based on a small sample [55]; for patients above 80 years of age, clinicians have had to rely on normative data from other countries/regions.

As expected, age, education level, and sex had a significant effect on performance on the learning and recall subtests of the WLMT in our data. Younger age, female sex, and higher education level were associated with higher scores on all the WLMT subtests. These effects are relatively profound, as illustrated with an example: A male patient, aged 90 years, compulsory education (≤10 years), with a raw score of 18 words on the total learning (trials 1–3) subtest and 5 words on the delayed recall subtest obtains results in the above-average/superior range in our normative data, i.e., 90^th percentile (z-score 1.26) on total learning and 72^nd percentile (z-score 0.58) on delayed recall. Correspondingly, a female patient, aged 70 years, with tertiary education (≥14 years) obtains results in the below average range with the same raw scores, i.e., 13^th percentile (z-score –1.12) on total learning and 8^th percentile (z-score –1.40) on delayed recall, indicative of mild amnestic cognitive impairment.

The robust effects of age and education are in correspondence with previous normative studies of the WLMT [26, 54–61]. We suggest that the age-related effects seen in our normative data primarily reflect normal aging processes, as all participants with a diagnosis of MCI and dementia, and participants with symptoms of depression and/or a history of stroke, were excluded from the analyses. Although there are inter-individual differences, decline in episodic memory performance in normal aging is well-documented, probably debuts on average at around age 60, and typically comprises both encoding (learning) and retrieval (recall) processes [6, 78]. The decline is associated with several structural and functional changes in the brain, including volume reductions in the gray and white matter, neuronal shrinkage, reductions in synaptic contacts, decreases in the concentrations of neurotransmitter substances, and neuronal loss. Such changes are typically most prominent in the frontal and the medial temporal cortex, including the hippocampus, and in the putamen, thalamus, and accumbens [78, 79].

Education level is established as a significant factor predicting cognitive performance, including episodic memory functioning, throughout the entire adult lifespan, from younger adulthood to older age, independent of gender, race, society, and birth cohort [80]. Higher education level not only predicts better episodic memory functioning but may also delay age-related cognitive change and the onset of neurodegenerative illnesses like AD, possibly as a consequence of a larger brain and cognitive reserve, which may act as a buffer against cognitive decline [81–83]. In fact, a meta-analysis found a dose-dependent effect of education; i.e., every additional year of education was associated with a 7% reduced risk of all-cause dementia [84].

The significant effect of sex in our study also accords well with other normative studies of the WLMT, at least in Western populations [55–57]. Interestingly, the female advantage is reported to be weaker or even absent in low- and middle-income countries, e.g., Nigeria and Tanzania [29], Cuba, Dominican Republic, Venezuela, Peru, Mexico, China, and India [59]. What could be the reasons for these differences? In general, a female advantage in verbal memory tasks is well documented. In a meta-analysis by Asperholm et al. (2019) which included 612 studies from 54 countries across 40 years (n = 587,691), females outperformed males in 42 out of 45 countries [85]. Also, the magnitude of the female advantage was positively associated with time- and country-specific social progress indicators. In their bivariate analysis, gender equality, population education and employment, and gross domestic product (GDP) per capita were all significant predictors of the female advantage, but in the multiple model, only population education and employment remained significant. The authors suggest that women’s verbal episodic memory performance may benefit more than men’s from education and labor market participation. Norway is among the most gender-equal countries in the world, and education and labor market participation is highly independent of gender [86]. Organized education also has a long history in Norway. In 1889, a seven-year primary school education became compulsory for all children. From the 1960 s, more women started earning university degrees and taking part in occupational life [87]. In line with the arguments of Asperholm et al., these are factors that may have fostered the female advantage in verbal episodic memory functioning seen in our data.

How well do our normative data fit with other normative data from comparable samples in Western countries/cultures? As mentioned, direct comparisons are difficult to make due to differences in study settings, study populations, inclusion/exclusion criteria, and sample sizes. These are all factors that may explain the noticeable variations in the normative data observed among the studies [26, 55–57]. For our purposes, the most relevant are probably the two population-based studies by Luck et al. (2018) and Luck et al. (2009) from Germany [56, 57], which both present normative data of a large number of home-dwelling older adults free of dementia (aged 60–79 years and 75–98 years, respectively), stratified by age, sex, and education—the norms are used in many Norwegian memory clinics today. We notice that the normative data of Luck et al. (2018) are “stricter” than ours in the sense that they report higher mean values (but a tendency to somewhat larger SDs) for the majority of the WLMT variables in relevant age groups, independent of sex and education level. As an illustrating example, mean (SD) of total list-learning (trials 1-3) and delayed recall for males with no university degree in the age group 75–79 is 19.1 words (3.5 SD) and 6.7 words (1.8 SD) in Luck et al. (2018), compared to 17.1 words (3.1 SD) and 5.2 words (1.4 SD) in our material. Interestingly, our data seems to be more in line with Luck et al.’s earlier study (2009). Still, the tendency is that Luck et al. (2009) report higher means (and larger SDs) on most variables in both age groups 75–79 and 80 + years. We see no obvious reason for the observed differences between Luck et al. (2018) and our study but suggest that these possibly could be explained by several factors. Differences in the participation rates could be one. In Luck et al. 33% of the invited persons agreed to take part in the study, while the corresponding figure in our study was 51.1% in the HUNT4 70 + sample (9930/19403) and 34.3% in the HUNT4 Trondheim 70 + sample (1745/5087). This could imply that the Luck-sample recruited a cognitively healthier group of older adults initially. Also, the participants in their study were recruited from one large German city, Leipzig, only, while our study comprises in the main participants recruited from rural areas and small towns (<25,000 inhabitants) of the middle of Norway (northern part of Trøndelag county). Finally, our applied strategies for inclusion/exclusion of participants were different. Particularly, we excluded participants with MCI (minor neurocognitive disorder) according to DSM-5, while Luck et al. did not apply MCI specifically as an exclusion criterion, but excluded participants with serious medical, neurological, or psychiatric disorders/conditions that could affect cognitive performance. Initially, one would believe that our strategy should lead to higher average scores, but without a closer comparison of the data it is hard to know exactly how these differences in selection strategy may turn out. Anyhow, the differences underline the importance of applying normative data representative of the population in which the CERAD WLMT is used.

As regards the already existing Norwegian normative data from Kirsebom et al. (2019) [55], these are fairly comparable to ours for the age interval 70 to 80 years, both for total list-learning and delayed recall—but there is a trend that the Kirsebom et al. norms are slightly “stricter”. A reason could be that Kirsebom et al. partly recruited participants through advertisements in media, which may have attracted a higher share of cognitive “super-performers” to their study. We also note that our normative data seem to differentiate better than that of Kirsebom et al.’s at the lower and upper ends of the scales.

Composition of the normative sample is a crucial point in the development of normative data. In this respect, the question whether persons with MCI should be included or excluded from the normative data, is a tricky one. We chose to exclude persons with a diagnosis of MCI (and dementia, symptoms of depression, and a history of stroke). Other studies have chosen different strategies. For example, in the German study by Luck et al. (2018) [56], the researchers chose only to exclude persons with a selected number of serious medical, neurological, and psychiatric disorders/conditions known to affect cognitive performance (including dementia) but did not exclude persons with milder forms of cognitive impairment who may have fulfilled the criteria of MCI (minor neurocognitive disorder), as they wanted the normative data to reflect the “true” level of cognitive functioning in the general population of dementia-free older adults. In the study by Gray et al. (2021) [29] from sub-Saharan Africa, on the other hand, the researchers chose to include both persons with MCI and dementia in the normative data and argue that exclusion of these conditions would have made the normative data less representative. There are two reasons why we consider exclusion of MCI to be the best alternative. Firstly, as MCI is a common condition among older adults, inclusion of these persons in the normative data will unavoidably lead to lower mean scores and larger standard deviations, i.e., increasing the likelihood that a person’s test score will be interpreted as being within the normal range, when, in fact, the person is cognitively impaired. This argument has been emphasized by several authors previously (e.g., Stricker et al. (2019) [88], Thomann et al. (2017) [45], Martin et al. (2017) [42], Green (2000) [44]). Secondly, MCI is a condition which is important to diagnose. Although some persons with MCI may revert to a normal condition, the large majority will remain stable (and possibly affect their complex ADL functioning and health related quality of life), or subsequently convert to dementia. In a systematic review [89], the authors reported an overall reversion rate of 18% from MCI to normal condition. Hence, an objection to our choice of strategy could be that persons with MCI who may later revert to a normal condition were excluded from the normative data. Still, we would argue that the number of such potential “reverters” most likely are relatively small, and that most of our participants with MCI cannot be considered cognitively normal.

With regards the proportion of participants with MCI in our sample, 24%, it is quite high compared to several other studies, but not all [90, 91]. In our case, it is important to take into consideration the age interval of the sample, 70–90 years. It is documented that the prevalence of MCI increases with older age, from around 10% in the age group 70–74 years to around 37% in the age group 85 + years [91]. Another reason for the high prevalence of MCI is probably the fact that we applied a threshold value of –1.0 SD for MCI to guide the diagnostic evaluations, which is the suggested by DSM-5 [92]. In other studies, threshold values of –1.5 or –1.0 SD have been used. Clearly, choice of threshold value will affect the prevalence, e.g., in the Swedish normative data for the MoCA test [76], 51% and 39% had z-scores≤–1.0 and≤–1.5, respectively. Still, we do not believe that our choice of threshold value has led to too strict reference values. Our view is supported by the fact that the normative data of both Luck et al. (2018) [56], Luck et al. (2009) [57], and Kirsebom et al. (2019) [55] are somewhat “stricter” than ours—as discussed above.

A strength of our study is the large number of older adults included in the development of the normative data and that they were recruited from a non-selected community sample. The exclusion criteria that were applied should ensure that our sample predominantly comprises older adults with normal cognitive functioning who were free of MCI or dementia. However, our study also has some limitations. First, we do not know how representative our sample is for the Norwegian or Scandinavian population of cognitively healthy older adults as a whole—the sample was drawn from the northern part of Trøndelag county and three city zones of Trondheim city only, and the participation rate was moderate (51.2% and 34.3%, respectively. Second, the CERAD WLMT was included among the tests/examinations for diagnosing MCI and dementia, which may have affected the clinicians’ diagnostic evaluations—preferably, another memory test should have been used for this purpose. Third, the cognitive tests that we administered did not fully cover all the cognitive domains that may be affected in neurocognitive disorders according to the DSM-5 criteria. Particularly, this objection applies to the social cognition domain. Fourth, although the diagnoses of MCI and dementia were done by experienced clinicians based on a comprehensive clinical evaluation and were made according to the DSM-5 criteria for minor and major neurocognitive disorder, data of the diagnostic accuracy (sensitivity/specificity) of our procedure is lacking. Also, we did not have access to brain imaging and fluid biomarkers, which most likely would have increased the diagnostic accuracy. In addition, we did not have access to examination of blood to rule out serious physical disorders that could cause MCI. However, the incidence of such disorders in home dwelling people would have been limited and thus not biased the results. Fifth, it would have been desirable to have in-depth information about the health status of the participants, and how they compare to individuals with a similar demographic background—particularly regarding serious medical, neurological, and psychiatric conditions that are known to affect cognitive performance. Unfortunately, we did not have access to such data. However, by applying our strategy of excluding persons with MCI, a history of stroke, symptoms of depression, and dementia, we believe that the majority of participants with conditions affecting cognitive performance have been excluded from our sample. Sixth, the recognition subtest of the CERAD WLMT was not administered in our study, and normative data for this subtest are therefore unfortunately lacking. However, we would argue that this is not a critical point, as a marked ceiling effect of the recognition subtest is well documented among cognitively healthy older adults [55–57]. For assessments of persons with suspected dementia, the recall subtest of the CERAD WLMT is the most sensitive, while the recognition subtest may be less affected in the earliest phases of AD [93]. Seventh, our normative data comprise persons in the age range 70–90 years only. Considering the number of younger people who are examined by GPs or memory clinics for memory complaints and/or suspected dementia, we suggest that population-based normative data for persons younger than 70 years, representative for the Scandinavia region, also should be developed.

To facilitate the scoring of the WLMT subtests for clinicians, we have developed a scoring calculator that is available in both English and Norwegian at http://www.aldringoghelse.no/ceradwordlist. By inserting the raw scores of the WLMT subtests and entering the age, gender, and education level of the patient, the calculator will produce z-scores and percentiles from the normative means.