Diversity in verbal fluency performance and its associations with MRI‐informed brain age matrices in normal ageing and neurocognitive disorders

Abstract Introduction Category verbal fluency test (CVFT) has been widely used to assess and monitor the cognitive capacities in epidemiological studies and clinical trials. Pronounced discrepancy in CVFT performance has been found in individuals with different cognitive statuses. This study aimed to combine the psychometric and morphometric approaches to decode the complex verbal fluency performance in senior adults with normal ageing and neurocognitive disorders. Methods This study adopted a two‐stage cross‐sectional design involving quantitative analyses of neuropsychological and neuroimaging data. In study I, capacity‐ and speed‐based measures of CVFT were developed to evaluate the verbal fluency performance in normal ageing seniors (n = 261), those with mild cognitive impairment (n = 204), and those with dementia (n = 23) whose age range is from 65 to 85 years. In study II, structural magnetic resonance imaging‐informed gray matter volume (GMV) and brain age matrices were calculated in a subsample (n = 52) from Study I through surface‐based morphometry analysis. With age and gender as covariates, Pearson's correlation analysis was used to examine the associations of CVFT measures, GMV, and brain age matrices. Results Speed‐based measures showed extensive and stronger associations with other cognitive functions than capacity‐based measures. The component‐specific CVFT measures showed shared and unique neural underpinnings with lateralized morphometric features. Moreover, the increased CVFT capacity was significantly correlated with younger brain age in mild neurocognitive disorder (NCD) patients. Conclusion We found that the diversity of verbal fluency performance in normal ageing and NCD patients could be explained by a combination of memory, language, and executive abilities. The component‐specific measures and related lateralized morphometric correlates also highlight the underlying theoretical meaning of verbal fluency performance and its clinical utility in detecting and tracing the cognitive trajectory in individuals with accelerated ageing.


| INTRODUC TI ON
As research on brain ageing moves forward, it is crucial to accurately evaluate the neurocognitive functions and develop simple and practical diagnostic biomarkers for individuals at early stage of cognitive impairments. Category verbal fluency test (CVFT), as one of the most efficient cognitive screening tools, 1 is designed to evaluate the capacities to produce words under specific category within a fixed time interval (i.e., 30 or 60 s). 2 During the past two decades, CVFT has been successfully used in the studies of healthy ageing 3,4 and different types of age-related neurodegenerative diseases, including mild cognitive impairments (MCIs), 2,5-7 Alzheimer's disease (AD), [8][9][10] frontotemporal dementia, 11 Parkinson's disease (PD), 12,13 and dementia with Lewy bodies. 14 Among these studies, the conventional interpretations of CVFT performance are often thought to be the measures of semantic verbal abilities and executive function. 15,16 However, given the nature of CVFT, the embedded qualities might not be adequately addressed by current interpretations. For instance, from a neuropsychological perspective, the raw scores of CVFT, including the numbers of correct named animals, fruits, and vegetables, are not a precise indicator of either verbal ability or executive function, 17 but reflecting a pattern of diverse cognitive functions. The pattern's most important finding is that advanced age demonstrates differential effects on the domains of cognitive functions, which may be interpreted in two different ways: (1) Verbal abilities, referring to crystallized cognition, have a lower sensitivity to ageing process 18 ; (2) Executive function and processing speed, as proxies of fluid cognition, keep deteriorating with normal and pathological ageing. 5,19,20 Meanwhile, evidence from neuroimaging studies highlights the related neural underpinnings of verbal fluency performance, of which the patterns of braincognition correlations reflect the domain-specific features, such as language ability, executive function. 4,[21][22][23] Considering the differential age-related effects on the verbal fluency performance, the clinical utilities of CVFT might be underestimated. First, several facets, such as executive control and the speed of word retrieval, implemented in the domain-specific components of CVFT, are difficult to measure quantitatively. Second, the measures of CVFT performance demonstrate remarkable heterogeneity across a variety of populations, particularly in the patients with progressive cognitive decline. 24,25 The heterogeneity due to sample demographics and the diverse proxies of CVFT used across the studies may lead to inconsistent results and limit the clinical utilities of CVFT.
Taken together, to maximize the utilities of CVFT and develop component-specific proxies for precise diagnostic and treatment evaluation, it may be helpful to practically dissociate the components of CVFT and achieve a thorough grasp of its neural underpinnings. In this study, we would first decode the CVFT performance into capacity-based and speed-based measures, and then investigate the component-specific CVFT performance in elderly with different cognitive statuses. Second, we would quantify the cortical features and estimated brain age in a subsample and examine the relationships between brain features and component-specific CVFT performance.

| Participants
We recruited 488 community-dwelling right-handed old adults aged from 65 to 85 years from our previous cohort studies. 26 Based on DSM-5, 30 the core domains of cognitive functions included the following: (1) Attention was measured by the digit span forward (DSF) and trail making test part A (TMT-A); (2) Perceptualmotor function was measured by the participant's performance on the commands as "tap each shoulder twice with two fingers keeping your eyes shut"; (3) Executive function was measured by TMT part B (TMT-B); (4) Learning and memory were measured by the word list learning test, including immediate recall and delayed recall of the words, and working memory capacity. Cerebrovascular risks were evaluated by the cumulative illness rating scale for geriatrics for the presence and severity of heart diseases, hyperlipidemia, diabetes mellitus, atrial fibrillation, hypertension, and anemia. 31 The Pittsburgh sleep quality index, 32 Cornell scale for depression in dementia, 33 and activity of daily living scale 34 were used to assess the subjective sleep quality, depressive symptoms, and everyday functioning separately. All the measurements were conducted with Chinese instructions.
The selection of high-performing, normal ageing and neurocognitive disorders (NCDs) were based on their global cognitive performance measured by CMMSE. The details of criteria were as follows: (1) High performing: the ones in the top quartile of the normative scores in this cohort, presenting with CMMSE score greater than 29 and CDR equal to 0, were classified as high-performing elderly. (2) Normal ageing: the ones with global cognitive function within 1.5 standard deviation (SD) of the age-and education-adjusted normative scores, presenting with CMMSE score greater than 28 and CDR score K E Y W O R D S brain age, cortical lateralization, gray matter volume, imaging, neurocognitive disorder, normal ageing, verbal fluency equal to 0, were classified as normal ageing elderly. (3) NCD patients were determined by the following criteria: evidence of modest decline in one or more cognitive domains, which was set as ≥1.5 SD below the age-and education-adjusted normative scores; no interference with independence in everyday activities; and no comorbid major psychiatric disorders. Mild NCD patients were defined as the CMMSE score less than 28 but greater than 22. Major NCD patients were diagnosed using a broad definition and met the following criteria: CMMSE score below the local cutoff for dementia of 18 and below for illiterate elderly, 20 and below for those with 1-2 years of education, and 22 and below for participants with more than 2 years of education. 35 The exclusion criteria include the following: (1) past history of bipolar affective disorder or other psychosis; (2) history of major neurological disease, including stroke, brain tumor, transient ischemic attack, or traumatic brain injury; (3) comorbidities with severe sleep disorders and depressive symptoms; and (4) patients with PD and progressive supranuclear palsy.

| Evaluation of CVFT performance
On each trial, participants were required to overtly generate as many words as possible within 60 s. First, we conducted the CVFT by three trials: all participants were asked to produce the words in the categories of animal, fruit, and vegetable. Second, the number of the words produced in each category within 30 and 60 s was recorded. The proxies of CVFT performance used in this study were summarized as follows:

| Participants
We randomly invited the eligible participants from Study I to participate the neuroimaging study. High-resolution T1-weighted magnetic

| Computation of brain age matrices
Pre-trained brain age model contextualized the whole brain morphometric features of the training sets from the Cambridge Centre for Aging and Neuroscience project (Cam-CAN) project (N = 611, age range: 18-90 years) (https://www.cam-can.org) were first constructed to generate a machine learning-based pre-trained model. 39 The brain age matrices were predicted using the support vector regression algorithm implemented in MATLAB (i.e., "fitrsvm" function, kernel: linear). Using 10-fold cross-validation, the estimated brain age model was applied to the entire samples (N = 611). Second, the participants from Study II were used as testing set (N = 52) for calculating and validating the brain age in clinical samples. The "Brain Age Gap Estimation" (BrainAGE) score is calculated as the difference between predicted brain age and chronological age, which indicates accelerated ageing process (positive value) or resilience (negative value). 40-42

| Data analysis
All data were tested for normal distribution through the Shapiro-Wilk test. Only data with a normal distribution could be counted.
Homogeneity of variance test was used to evaluate the equality of variances among high-performing, normal aging, minor NCD and major NCD groups. Group-wise differences of demographics and neuropsychological performance were tested either with χ 2 test for category variable or one-way analysis of variance (ANOVA) for continuous variables. Repeated measures ANOVA analyses were applied to evaluate changes from the numbers of corrected words of animal to vegetable. Two time points, including the 30 and 60 s, were treated as a within-subject factor and the differences across four groups were treated as a between-subjects factor. The interactions between time points and cognitive status were also examined in the above analyses. Covariates such as age, gender, and educational level were included in the model. Partial eta squared (η 2 ) was reported as effect size. Post-hoc multiple comparisons among means were conducted using a Tukey HSD test when ANOVA results detected significant differences among the groups. We used Pearson correlation coefficient to test the relationships between CVFT measures, cognitive functions, and brain features. We carried

| Psychometric properties of CVFT
There were no differences in chronological age and gender ratio across the groups with different cognitive statuses. As shown in Table 1, mild and major NCD patients showed worse cognitive functions and lower daily activities than high-performing and normal aging elderly. As to CVFT performance, the capacity-based CVFT scores were coordinately getting lower with the severity of disease. Similarly, significant group-wise discrepancy was also found F I G U R E 1 The flowchart of this study. (A) Study I: Psychometric mapping of category verbal fluency test (CVFT) in high-performing, normal ageing, mild and major neurocognitive disorders (NCD); (B) Study II: Magnetic resonance imaging (MRI)-informed gray matter volume mapping and brain age calculation. Abbreviations: AAL, Automated anatomical atlas; GM, Gray matter.

| Morphometric correlates of capacitybased measures
Using age, gender, years of education, and total intracranial volume as covariates, direct scores of CVFT were correlated with left tem-

| Brain age and BrainAGE
The mean absolute error of pre-trained brain age model was 3.581, which had comparable generalizability with the published brain age models 44,45 and the testing dataset. Generally, the chronological age was positively correlated with estimated brain age in training samples (r = 0.737, p < 0.001) and our clinical samples (normal ageing elderly: r = 0.621, p < 0.001; mild NCD patients: r = 0.844, p < 0.001). As shown in Table 3, the demographics were comparable between normal ageing elderly and mild NCD patients. Mild NCD patients have worse global cognition and executive function than normal aging elderly. Although mild NCD patients had similar chronological age (t = −1.253, p = 0.216), pronounced increased brain age (t = −4.811, p < 0.001) and higher BrainAGE score (t = −5.637, p < 0.001) were found in mild NCD patients (Figure 4).

| Correlation analyses
To gain a better understanding of the relationships between brain age matrices, education, and the proxies of CVFT performance, correlation matrix was calculated in normal ageing and NCD groups.
Overall, using age and gender as covariates, older estimated brain age was correlated with greater brain-chronological age gap (i.e.,

BrainAGE) and less years of education. Capacity-based measures of
CVFT performance were significantly correlated with speed-based measures in both groups. The Pearson correlation coefficients in normal aging group were lower than the ones in mild NCD group.
Significant correlations between CVFT performance, brain age, and years of education were only found in mild NCD group ( Figure 5B), not in normal ageing group ( Figure 5A). In mild NCD patients, increased CVFT capacity was related to younger estimated brain age (r = −0.469, p = 0.034) and higher educational level (r = 0.553, p = 0.026). When adjusting the effects of education, the association between CVFT measures and brain age was unchanged in normal ageing group ( Figure 5C), but the association between CVFT capacity and estimated brain age became non-significant in mild NCD patients (r = −0.275, p = 0.259) ( Figure 5D).

| DISCUSS ION
To the best of our knowledge, this study was the first to investigate the component-specific CVFT performance and its associations with MRI-informed brain age matrices in senior adults with normal ageing and mild neurocognitive disorder. Given the psychometric and morphometric features of CVFT, we found the TA B L E 3 Baseline demographics, clinical and brain features in normal aging and mild NCD groups.

F I G U R E 4
Comparisons of chronological age and estimated brain age in normal ageing elderly and mild neurocognitive disorder (NCD) patients. (A) There was no difference of chronological age between normal ageing elderly and mild NCD patients. (B) Mild NCD patients had older brain age than chronological age-matched normal ageing elderly. Beyond the lateralized features of CVFT measures, another interesting finding is the associations between CVFT capacity and MRI-informed estimated brain age in mild NCD patients. Based on the features of cortical GMVs, the estimated brain age represents the status of brain ageing at individual level. [40][41][42] At present, although there is no direct link between brain age and executive function, the close relationship between age-related brain atrophy (i.e., decreased brain reserve) and executive dysfunction suggests that the accelerated brain changes have significant effects on frontal functions in late adulthood. 54 Not limited to executive function, we observed the diverse changes of verbal fluency performance across three categories showed greater discriminative value in NCD patients than in normal ageing elderly. Interestingly, the positive correlation between CVFT capacity and working memory capacity is more pronounced in NCD patients rather than high-performing and normal ageing elderly, indicating that the impaired capacity related to accelerated brain ageing may be a fundamental change in earlystage neurodegeneration.
To conclude, this study endorses the phenomenon that the defi-

| Limitations and future directions
Although the findings in this study are encouraging, conclusions need to be interpreted with caution due to its limitations. The educational levels across the four groups were not comparable.
Considering both educational attainment and language ability (i.e., bilingualism) are key components of cognitive reserve, [55][56][57] these variables might contribute to the group-wise differences of CVFT performance and brain age matrices. Regarding the role of education in neurodegeneration, we adjusted effects of education on the correlation matrix of CVFT performance and brain age matrices and found the previous significant association between CVFT capacity and brain age disappeared. Besides, a relatively small collection of MRI scans might limit the generalization of the "CVFT-brain" correlation patterns. Moreover, the absence of the status of language ability (i.e., bilingualism) on component-specific CVFT performance may also limit our interpretations in the relationships of verbal fluency performance, estimated brain age, and cognitive reserve.  com/hanna brain scien ce/Brain -age-predi ction).

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

DATA AVA I L A B I L I T Y S TAT E M E N T
The dataset used and analyzed during the present study are available from the corresponding author on reasonable quest.

TA B L E A 3
Comparisons of component-specific CVFT performance between mild and major NCD patients.

TA B L E A 4
Correlation matrix between CVFT measures and cognitive function in high-performing elderly.