Complex oral semantic verbal fluency in non-brain-damaged adults and individuals with multiple sclerosis and subjective anomia

ABSTRACT

Background: Many neurologic conditions, for example multiple sclerosis (MS), are associated with subtle communication and language difficulties. To detect such difficulties, there is a need for valid and reliable methods. While standard aphasia test batteries have been found insufficient, more complex language tasks are believed to be able to distinguish more subtle language difficulties from normal variation in communicative ability.

Aims: The aim of this study was to: (1) explore the influence of demographic variables on the results of a novel complex oral semantic fluency task with multiple restrictions in non-brain-damaged (NBD) adults, (2) investigate the construct validity and reliability of the method, and (3) compare the results of NBD individuals and individuals with MS.

Method and procedure: One hundred and ten NBD individuals performed the complex task and three standard fluency tasks with single restrictions. Regression analyses were run to assess the influence of demographic factors. Furthermore, 16 individuals with MS and subjective anomia performed the complex task and the results were compared with those for a matched group of NBD individuals.

Outcome and results: Age and education influenced the NBD individuals’ scores on the complex task. The NBD individuals’ results on the complex task correlated with those on the three standard fluency tasks. Furthermore, the subgroup of 16 pair-matched NBD individuals produced statistically significantly more adequate responses in the complex task than the group of individuals with MS. However, on an individual level some of the participants with MS performed at level with, or even better than the pair-matched NBD individuals. Provision of scoring guidelines yielded high interrater-reliability.

Conclusions: The results illustrate the challenge in attempts to provide formal measures of subtle language disorders. Still, the complex task is a promising assessment tool which may be a complement to existing standard word fluency tasks, although future studies are required to establish the validity and ability to detect subtle language difficulties in different clinical groups.

KEYWORDS:

• Assessment
• subtle language disorder
• anomia
• multiple sclerosis
• verbal fluency
• divergent production

Introduction

In addition to an evident language disorder, such as aphasia following stroke, more subtle language and communication difficulties can arise in many neurological conditions. These conditions have been termed, for example, subtle language deficits (Crosson, 1996), cognitive-linguistic impairments (Body & Perkins, 2006), high-level language deficits (Lethlean & Murdoch, 1997), and cognitive-communication disorders (Togher et al., 2014); see Body and Perkins for a review. Among other symptoms, word-retrieval difficulties are typically involved (Hough, 2008; Kavé & Goral, 2016; Miller, Noble, Jones, & Burn, 2006).

In Parkinson’s disease, anomia and problems with sentence formulation have been reported as causing problems in everyday communication (Miller et al., 2006; Saldert & Bauer, 2017; Saldert, Ferm, & Bloch, 2014). Individuals with multiple sclerosis (MS) also often experience a linguistic impairment that affects both comprehension and language production and therefore quality of life, although the symptoms are often subtle (Friend et al., 1999; Klugman & Ross, 2002; Renauld, Mohamed-Saïd, & Macoir, 2016; Tallberg & Bergendal, 2009). Language difficulties in the form of word retrieval deficits (anomia) have been reported in both chronic-progressive and relapsing-remitting MS. It has been argued that difficulties with more complex linguistic processing is under-diagnosed in people with MS, as speech-language pathologists working with MS usually focus on motor symptoms, such as dysarthria (Renauld et al., 2016; Sepulcre et al., 2011). Subtle symptoms like semantic unspecific or atypical word choices in naming tasks, as well as less effective strategies and impaired production in word fluency tasks that cannot be explained by dysarthria, have been described (Henry & Beatty, 2006; Tallberg & Bergendal, 2009).

When trying to detect language deficits of a more subtle nature, standard aphasia batteries are insufficient (Body & Perkins, 2006; Coelho, Ylvisaker, & Turkstra, 2005; Crosson, 1996; Saldert, 2017). Tasks that are more suitable usually depend on the integration of language and several other aspects of cognition, such as executive functions and memory, including pragmatic ability, which makes them more complex. Several more or less established test batteries are currently used to detect such subtle language disorders, including the Test of Language Comprehension (TLC; Wiig & Secord, 1989) and the Mount Wilga High Level Language Screening Test (MWHLL; Christie, Clark, & Mortensen, 1986). However, such instruments have received criticism due to a lack of a theoretical framework explaining why the tasks may be difficult and also due to questionable reliability and validity (Body & Perkins, 2006; Coelho et al., 2005; Frith, Togher, Ferguson, Levick, & Docking, 2014; Saldert, 2017). Thus, there is a need for valid language tests suitable to capture subtle language difficulties.

It has been argued that language tasks requiring so called divergent production may be useful to detect subtle language disorders and that type of tasks also have a theoretical base (Chapey, 1977; Chapey, Rigrodsky, & Morrison, 1976). The psychologist Guilford (1959, 1967), who was mainly concerned with the issue of measuring intellectual abilities, has defined divergent production as a measure of the ability to exhibit fluency, flexibility, originality and elaboration in thinking. According to Guilford (1959), divergent thinking is one of two types of ”productive-thinking operations” that generate new information from stored knowledge. In divergent thinking operations, the search and activation of stored knowledge in semantic networks is wide-ranging as several different information units may be just as relevant to activate. In convergent thinking operations, the search is focused on the activation and retrieval of single correct or conventional information units. Consequently, a disruption of convergent language operations may result in a disability to correctly name objects in confrontation naming tasks or to complete sentences (for example, “you eat with a knife and __?”). In contrast, in tasks requiring divergent semantic production, there is no single correct answer; instead the focus in the assessment of the results lies on quantity, relevance, and variance in the production (Chapey et al., 1976; Guilford, 1967). Thus, while so-called convergent production tasks require single correct responses to stimuli, divergent production tasks require a person to generate several relevant responses. Guilford (1967) suggests different forms of complex fluency tasks to measure divergent production, such as asking participants to generate different uses for objects, e.g. a brick. Furthermore, the ability to perform divergent thinking operations may also be measured in fluency tasks with differing degrees of constraints or restrictions in terms of number of specifications of mandatory features in adequately produced items (Guilford & Hoepfner, 1971). A single restriction task could for example be to name “words beginning with the letter F”. Multiple restrictions can be added to make the fluency task more complex, for example by instead asking the person to name “things that are flat and soft”. In such a task, adequately verbally produced items need to share the three semantic features of being object + flat + soft.

Tasks depending on convergent thinking operations are assumed to rely on a more automatic process and be less cognitively demanding in terms of memory and executive functions than tasks that depend on divergent operations (Spreen & Risser, 2003). Thus, tasks requiring divergent thinking operations are considered more complex and demanding than tasks requiring convergent thinking operations. Further, tasks requiring divergent thinking operations which include multiple restrictions are considered more complex than tasks with single restrictions.

Divergent thinking is usually associated with measures of degree of creativity, but it has also been investigated within several areas of language research, including in normal aging processes (Palmiero, Di Giacomo, & Passafiume, 2014), Alzheimer’s disease and frontotemporal dementia (Hart & Wade, 2006). Chapey et al. (1976) applied Guildford’s theory of divergent thinking in a group of people with aphasia and found that impaired divergent production is an issue in acquired language impairment. Impaired divergent production was manifested in a reduced ability to produce several relevant semantic responses and a decreased flexibility in verbal fluency tasks. Chapey et al. (1976) and Chapey (1977) conclude that there is a need for assessments of both divergent and convergent operations in language disorders. They argue that people with severe aphasia may be impaired in both convergent and divergent tasks, whereas persons with milder aphasia may only be affected in divergent semantic processing (Chapey, 1977, 1981).

Verbal fluency tasks with single restrictions are frequently used in clinical and experimental examination of cognitive function and language ability, for example in controlled oral word association tests (Benton & Hamsher, 1976). Phonemic/letter fluency, semantic/category fluency, and action verb fluency are different variants of verbal fluency tasks. A frequently used phonemic fluency task is FAS, which requires the production of words with the initial letters F, A and S (Borkowski, Benton, & Spreen, 1967; Tombaugh, Kozak, & Rees, 1999). In exploring semantic fluency, animal category seems to be the most frequently used task (Tallberg, Ivachova, Jones Tinghag, & Östberg, 2008; Tombaugh et al., 1999), while action verb fluency is less commonly explored (Östberg, Fernaeus, Hellström, Bogdanovih, & Wahlund, 2005; Piatt, Fields, Paolo, & Tröster, 1999). Norms for verbal fluency in healthy adults have been reported for several languages including English and Swedish (Tallberg et al., 2008; Tombaugh et al., 1999). Variance in performance on verbal fluency tasks has been noted across several pathological conditions, including Alzheimer’s disease (Henry, Crawford, & Phillips, 2004), Parkinson’s disease (Henry & Crawford, 2004b), and focal brain injuries (Henry & Crawford, 2004a).

Although both phonemic and semantic fluency tasks are dependent on numerous shared cognitive abilities such as word storage, functional search processes, and engagement of working memory and other aspects of executive functions, possible differences have been discussed. In a meta-analysis of verbal fluency tests and focal brain damage, Henry and Crawford (2004a) conclude that phonemic and semantic verbal fluency place comparable demands on frontal lobe functioning and executive abilities. Still, in addition, semantic fluency also relies heavily on temporal lobe functioning, suggesting that access to semantic memory is relatively more important than executive processes in semantic fluency tasks (Henry & Crawford, 2004a). The partially shared and partially distinct cognitive processing in phonemic and semantic fluency has been confirmed in neuroimaging studies (Biesbroek et al., 2016) and cerebral blood flow studies (Gourovitch et al., 2000). Lastly, action verb fluency is more sparsely researched but has proven to be both moderately related to executive functioning (Piatt et al., 1999) and sensitive to mild cognitive impairment (Östberg et al., 2005).

Performance on verbal fluency tasks with single restrictions may be affected by various demographic factors, such as age, education, and sex, although the influence of sex is usually minimal (Tombaugh et al., 1999). Regarding age, it has been shown that younger adults (18–40 years) tend to generate more words overall than elderly counterparts (65–91 years) (Lanting, Haugrud, & Crossley, 2009). A greater age effect on semantic fluency than on phonemic fluency has been reported (Tombaugh et al., 1999; Troyer, 2000). Piatt, Fields, Paolo, and Tröster (2004) reported no obvious age effect in action verb fluency performance in a healthy older population (age 56–92). Regarding education, longer or higher education (over 12 years) corresponds with better results on verbal fluency tasks, especially in the areas of phonological and action verb fluency (Tallberg et al., 2008).

Word retrieval difficulties are frequently described in numerous neurological disorders, including Alzheimer’s disease (Henry et al., 2004; Kavé & Goral, 2016), Parkinson’s disease (Henry & Crawford, 2004b; Miller et al., 2006), and focal brain injuries (Henry & Crawford, 2004a). It has been suggested that complex fluency tasks should be more sensitive to subtle cognitive changes than fluency tasks with single restrictions, since they put a greater demand on conceptualisation and flexibility in the search process (Hart & Wade, 2006). Hart and Wade (2006) state that it is likely that divergent processing, measured through more complex fluency tasks may distinguish patients with early stages of dementia from healthy older adults. If this is true, such tasks may also expose early signs of language impairments in other neurological conditions, such as Parkinson’s disease, MS and more subtle impairments following a stroke.

To sum up, there is a need for instruments and methods suitable for capturing subtle language impairments such as anomia, which affects communication related to brain damage in various neurological conditions. Assessment of divergent production offers an opportunity to explore results in tasks that depend on semantic strategies and processes similar to those required in everyday communication. More complex word fluency tasks have been suggested as useful methods to explore more subtle word-finding difficulties in neurogenic communication disorders.

The aim of the present study is to apply a complex verbal fluency task with multiple semantic restrictions in a group of non-brain-damaged (NBD) adults and a group of individuals with MS and subjective anomia to explore the construct validity and reliability of the novel method. The research questions are: (1) Is the performance on the complex verbal fluency task influenced by age, education or sex in a group of NBD individuals? (2) Is there a statistically significant difference in the performance on the complex verbal fluency task between a group of NBD individuals and a group of individuals with MS and subjective anomia? (3) Do results on the complex verbal fluency task correlate with results on standard fluency tasks? (4) Do the scoring guidelines for the complex verbal fluency task yield sufficient inter-rater reliability?

Method

The present study has a cross sectional design and was approved by the Regional Ethical Board of Gothenburg (ref. no. 506–16). The participants gave written consent after receiving oral and written information.

Participants

One hundred and twenty-six NBD adults and 16 adults with MS were recruited to the study. The inclusion criteria for the NBD group were: no self-reported neurological injury or disease, Swedish as mother tongue, and no alcohol or substance abuse. Participants were recruited through convenience sampling via associations, university classes, senior citizen groups, and family and friends. Six participants did not match the inclusion criteria and were therefore excluded. During the analysis, an additional eight participants were excluded due to missing demographic data. Two participants were unable to complete the test and were therefore excluded as well. Age among the remaining 110 participants (53 men, 57 women) ranged from 19 to 85 years (mean: 52.3), and their educational background in years ranged from 6 to 20 years (mean: 14.3).

The 110 NBD participants were divided into two groups according to length of education. The first group consisted of participants with 0–12 years of education (n = 35, mean: 10.8 years) and the second of participants with 13+ years of education (n = 75, mean: 16 years). The sample was also split into four different age groups: 18–29 (n = 23); 30–64 (n = 42); 65–75 (n = 30); and 75+ (n = 15). This division aimed to mirror different phases in life.

The group of individuals with MS consisted of 13 females and three men aged 38–72 (mean: 56.8), and their years of education ranged from six to 17 (mean: 12.9). They were recruited from speech-language pathology clinics, neurological clinics, and patient associations. The inclusion criteria were: diagnosed with MS but no other neurological injury or disease, subjective anomia, no more than mild-moderate dysarthria, Swedish as mother tongue and no alcohol or substance abuse.

Sixteen participants from the NBD reference group were pair-matched in terms of education, age and gender with the individuals in the MS group, see Table 1. In the few cases where there was no perfect match or more than one control matched equally well, a hierarchy in the matching criteria was used: education before age and age before gender.

Material

Four types of fluency tasks were administered in the order listed: phonological fluency (FAS), category fluency (animals), action fluency (verb), and a novel verbal fluency task with multiple semantic class restrictions: Complex Oral Semantic Fluency (COSEF).

Complex oral semantic fluency (COSEF)

The novel test method, COSEF, had been constructed by the authors based on Guilford and Hoepfner’s theories (Guilford, 1959, 1967; Guilford & Hoepfner, 1971). It consisted of three tasks where the participants were given 2 min for each task to name objects that represented select combinations of two characteristics (adjectives): (1) things that are both square-shaped and hard (practice question), (2) things that are both flat and have a round shape, and (3) things that are both sharp and long. Thus, the tasks included two restrictions of which objects would be considered adequate.

Standard oral fluency tasks

In the phonemic fluency task, the participants were instructed to name words (excluding proper names) beginning with the letter F, A, and S. They were given 1 min for each letter. In category (animal) and action verb fluency, the participants were informed that initial letter was no longer important. Thus, these tasks included a single restriction. In all four fluency tasks, the participants were instructed to continuously produce new words without making inflections or compounds.

Procedure

Five certified speech-language pathologists (SLP), including the first author, and two speech-language pathology students collected the data following a standardised procedure. The tasks were administered in a calm environment at the workplaces or homes of participants and test leaders, at SLP clinics and on university premises. After signing an informed consent form, the participants completed a questionnaire collecting personal information about age, sex, education, occupation, and neurological and psychological conditions.

Scoring guidelines for COSEF were developed by the authors and the analysis included coding of number of adequate words (including both of the required semantic criteria), repetitions, and inadequate responses. In cases of loanwords or more newly created expressions, the words had to be represented in SAOL (the Swedish Academy’s official glossary) or generate more than 500 hits on Google. Synonyms and metaphorical interpretations were accepted, but repetitions, inflections, and compounds were not. For example, “sewing needle” together with “syringe needle” was scored as only one adequate word. Further, circumlocutions and superordinate categories, e.g. “tools” or “toys”, were not accepted. The object referred to was not allowed to be altered in form or function to fit the constraints. For example, “a smashed orange” was not accepted as a round and flat object.

In calculating the number of adequate items, the scoring procedures for F-A-S, animals and verb fluency followed the procedures presented in Tombaugh et al. (1999), Woods et al. (2005), and Tallberg et al. (2008) with a few adaptations. In line with Tallberg et al. (2008), words with different endings were coded as errors as they were considered rule violations. No proper names were accepted (with the exception of days and months). Compounds were not accepted, but words sharing the same prefix were included in the total score. Further, both homophones and homonyms were accepted in the current study, as were slang words. Both super- and subordinate words were included when scoring the category fluency (animals), and all verb forms were accepted when scoring action verb fluency.

Statistical analysis

Linear single and multiple regression analyses were run to examine the influence of demographic factors in the results on the COSEF task of the complete reference NBD group. Test score was set as dependent variable and age and years of education was set as continuous independent variables. Sex was set as a dummy coded categorical independent variable.

To explore the construct validity of the new fluency task, Pearson correlation was used to compare COSEF scores with the other established verbal fluency tasks in the complete NBD reference group. Interpretation of correlations followed the suggestions in Cohen (1988, pp. 79–81), i.e. small: r = .1–.29; medium: r = .30–.49; large: r = .50–1.0. Furthermore, as the MS group and the NBD control group were pair-matched, the Wilcoxon signed ranks test was used to explore possible differences in results on COSEF in the two groups. Due to the risk of type 1 errors when using small control samples in single case research, a modified t-test, developed by Crawford and Howell (1998) was also used. This was applied for each of the participants with MS when comparing their scores with the mean score of the matched NBD control group.

In a first attempt to assess the degree of inter-rater reliability of the scoring guidelines for COSEF, the intra-class correlation coefficient (ICC) was used. For the analysis of number of adequate responses of the NBD group, 35% (n = 38) of all protocols of the complete group (n = 110) were randomly chosen and scored independently by the first author and a trained speech-language pathologist involved in the research project. In the ICC analysis a two-way mixed model, absolute agreement, and single measures, was used. The inter-rater reliability rate (ricc = .966; 95% confidence interval = 0.935–0.982, p < 0.001) was defined as excellent according to Cicchetti, Butcher, and Nelson (1994) suggestions. Comparing the independent scorings of this data of the first author and an untrained speech-language pathologist, recruited from one of the clinics providing participants for the study, using the scoring guidelines also resulted in excellent inter-rater reliability (r_icc = .973; 95% CI = 0.978–0.986, p < 0.001). Further, an  ICC analysis with a two-way random model, absolute agreement, and single measure, was used in the assessment of the reliability of the scoring of 17 randomly chosen individuals with mild, or subtle, anomia due to MS (3 individuals), Parkinson’s disease (4 individuals), or a left hemisphere stroke (10 individuals) on the novel fluency task. These results were scored independently by four different assessors, after approximately 6 hours of training in using the scoring guidelines on other data from individuals with these neurological conditions and self-reported anomia. The assessors were three speech-language pathologists and a linguist involved in the research project. This inter-rater reliability rate (ricc = .990; 95% Confidence interval: 0.98–1.0; p < 0.001) was also excellent.

Results

Semantic fluency in the NBD reference group

COSEF was generally perceived as challenging by the participants. The mean number of adequate responses for the complete NBD reference group (n = 110) was 16.95 (SD = 5.58). The mean number of repetitions was low (0.15, SD = 0.47; median = 0, range = 3) with skewness of 3.8 (SE = 2.3) and kurtosis of 16.4 (SE = 0.46). The mean number of inadequate responses was 5.41 (SD = 4.42; median = 4; range = 22) with skewness of 1.3 (SE = 0.23) and kurtosis of 2.1 (SE = 0.46). Inadequate responses commonly included compounds or did not meet the semantic criteria. Table 2 presents the mean scores and standard deviations for age and education groups. The highest score is found for participants aged 30–64 in the higher education group, and the lowest score is found for participants aged 75+ in the lower education group.

Table 1. Pair-matched participants in MS group and non-brain-damaged (NBD) group

MS group	MS type*	Years since diagnosis	Age	Sex^1	Education in years	NBD group	Age	Sex^1	Education in years
MS-1	RRMS	15	55	F	11	C-1	38	M	12
MS-2	PPMS	2	72	F	10	C-2	72	M	10
MS-3	PPMS	26	72	F	14	C-3	72	F	16
MS-4	SPMS	16	59	F	11	C-4	60	F	10
MS-5	RRMS	8	47	F	17	C-5	45	F	16
MS-6	RRMS	18	38	M	12	C-6	38	M	15
MS-7	RRMS	17	69	F	15	C-7	68	F	16
MS-8	RRMS	16	52	M	14	C-8	51	M	16
MS-9	SPMS	13	57	F	14	C-9	58	F	13
MS-10	RRMS	3	52	F	13	C-10	52	F	17
MS-11	PPMS	4	53	F	14	C-11	52	F	15
MS-12	SPMS	28	68	F	9	C-12	69	F	8
MS-13	RRMS	12	62	F	14	C-13	63	F	17
MS-14	SPMS	7	57	F	15	C-14	60	F	16
MS-15	RRMS	2	40	F	12	C-15	36	M	13
MS-16	PPMS	5	56	M	12	C-16	56	F	9

* RRMS = relapsing-remitting MS; PPMS = primary progressive MS; SPMS = secondary progressive MS; 1. F = female, M = male

Table 2. Mean and standard deviation (SD) for COSEF scores presented by age and education groups (n = 110)

	Age (years)
	18–29		30–64		65–74		75+
Education	0–12	13+	0–12	13+	0–12	13+	0–12	13+
M (SD)	17.17 (2.93) n = 6	19.47 (5.80) n = 17	14.43 (2.76) n = 7	19.46 (5.55) n= 35	11.67 (4.16) n = 12	16.72 (3.30) n = 18	11.60 (5.10) n = 10	18.20 (5.12) n = 5

Influence of age, education, and sex on verbal fluency scores in the complete NBD reference group

Table 3 presents the results from both single and multiple regression analyses of the results for the complete NBD reference group. Overall, the adjusted R^2 values indicate significant influence of demographic factors for the verbal fluency measure. Age and education were significant independent influencers of results on COSEF.

Table 3. Results from simple and multiple regression analysis of total score in COSEF in terms of unstandardized β-coefficient (std. error), p value, 95% confidence interval (C.I) and adjusted R^2

	Simple regression				Multiple regression
	β (S.E)	95% C.I	p	R2	β (S.E)	95% C.I	p	R2
Age	−.102 (.024)	−.149 – −.055	.000*	.140	−0.78 (.024)	−.125 – −.031	.000*	.218
Education	.717 (.167)	.387–1.047	.000*	.139	.545 (.166)	.215 – .874	.000*
Sex	−1.533 (1.059)	−3.633- .566	.151	.010	−1.201 (.943)	−3.072 − .669	.206

* = statistically significant

Spearman’s rho indicated a small, but statistically significant negative correlation (−.251, p= .008) between age and education. However, the low variance influence factor (VIF) for age (1.095) and education (1.096) indicated low disturbance of the influential outcome.

Correlations between the novel complex fluency task and fluency tasks with single restrictions

The raw total mean scores for the NBD group were 41.78 (SD: 11.64) for FAS, 22.27 (SD: 6.30) for semantic category (animal) fluency and 18.07 (SD: 6.34) for action verb fluency. On a group level, these scores are in line with Swedish norms (Tallberg et al., 2008), see Table 4.

Table 4. Results on standard word fluency measures (FAS, animals, and verbs)

	Age, years
	18–29		30–64		65–74		75+
Education	0–12	13+	0–12	13+	0–12	13+	0–12	13+
	n = 6	n = 17	n = 7	n = 35	n = 12	n = 18	n = 10	n = 5
FAS^a	35.83 (4.79)	45.00 (10.33)	33.71 (9.46)	47.66 (11.39)	35.83 (12.09)	42.83 (9.56)	31.40 (10.98)	39.40 (6.23)
Animals ^b	21.83 (4.71)	24.00 (5.50)	21.14 (6.84)	24.54 (6.21)	21.83 (8.29)	21.89 (4.75)	14.50 (3.06)	20.60 (3.44)
Verbs^c	14.67 (5.09)	19.76 (4.48)	17.71 (6.45)	21.06 (5.85)	14.08 (5.96)	18.11 (6.09)	10.90 (5.55)	19.80 (4.21)

^{a, b, c} = Normative data for the Swedish population (Tallberg et al., 2008)

^a = Age 16–29, education level 0–12: 37.5 (11.4); education level 13+: 42.4 (10.0). Age 30–64, education level 0–12: 42.3 (10.0); education level 13+: 49.0 (13.3). Age 65–89, education level 0–12: 36.6 (15.1); education level 13+: 45.2 (10.1).

^b = Age 16–29, education level 0–12: 23.0 (4.0); education level 13+: 26.4 (5.1). Age 30–64, education level 0–12: 24.9 (6.4); education level 13+: 27.1 (5.4). Age 65–89, education level 0–12: 17.8 (5.7); education level 13+: 20.6 (5.7).

^c = Age 16–29, education level 0–12: 17.0 (3.7); education level 13+: 23.0 (6.3). Age 30–64, education level 0–12: 17.1 (4.7); education level 13+: 22.3 (6.4). Age 65–89, education level 0–12: 14.9 (6.4); education level 13+: 19.4 (5.6).

Table 5. Scores for individual participants with MS on established word fluency measures and COSEF and scores for matched controls on COSEF

Participant with MS	FAS	Action verbs	Animals	COSEF total	Inadequate responses	Repetitions	Control participant	COSEF total	Inadequate responses	Repetitions
MS-1	19	15^†	20^†	5*	1	0	C-1	12	0	0
MS-2	14	12^†	7	6*	0	0	C-2	14	3	0
MS-3	32	17^†	19^†	13	5	1	C-3	14	3	0
MS-4	33^†	7	28^†	15^††	6	1	C-4	15	11	0
MS-5	42^†	21^†	23^†	17^††	7	0	C-5	17	5	0
MS-6	20	12^†	14	6*	10	0	C-6	13	15	0
MS-7	50^†	19^†	19^†	20^††	2	0	C-7	14	8	1
MS-8	16	10	18	14^††	4	2	C-8	19	2	0
MS-9	38^†	19^†	26^†	17^††	3	0	C-9	32	5	0
MS-10	37	16	18	17^††	6	0	C-10	18	10	0
MS-11	35	18^†	24^†	15^††	11	0	C-11	25	21	0
MS-12	15	4	9	4*	6	0	C-12	9	0	1
MS-13	70^†	16^†	24^†	23^††	6	2	C-13	19	7	0
MS-14	41^†	13	25^†	4*	0	0	C-14	13	4	0
MS-15	29	9	13	14^††	3	0	C-15	28	4	0
MS-16	36^†	22^†	23^†	11	2	1	C-16	18	0	0
Mean (Standard deviation)				12.6 (6.0)	4.5 (3.3)	0,4 (0.7)		17.5 (6.2)	6.1 (5.8)	0.1 (0.3)

^† = Performs within one standard deviation below mean score for reference group with corresponding age, gender, and education presented in Tallberg et al. (2008)

^†† = Performs within one standard deviation below mean score on COSEF for reference group with corresponding age, gender, and education.

* = Difference significant at p < .05 compared to mean scores obtained in the matched control group (modified t-test; Crawford & Howell, 1998)

In the complete NBD reference group (n = 110), the correlation analysis revealed medium–large and statistically significant correlations (p= 0.01) between COSEF scores and performance on phonological fluency (r = 0.411), category (animal) fluency (r = 0.509), and action verb fluency (r = 0.546), indicating that people who performed well on COSEF also performed well on the standard fluency tasks.

Comparison between the pair-matched groups

There was a statistically significant difference (Z= −2.735, p= .006) between the number of adequate responses in the pair-matched NBD group (mean: 17.3; SD: 6.2) and the MS group (mean: 12.6; SD: 6.0). However, the variation in results was large in the two groups and nine of the participants with MS and subjective anomia performed within one standard deviation below the mean of the NBD reference group, see Table 5.

On an individual level the participants with MS's performances in the four different fluency tests vary. Almost all of them are performing within normal variation on at least one of the four tasks, including COSEF, despite reporting experiences of anomia in everyday communication.

Discussion

In this study, a novel oral fluency task with multiple semantic restrictions, COSEF, was tested on a group of NBD individuals and a group of 16 individuals with MS and self-reported anomia. The results indicate construct validity and good inter-rater reliability for the assessment method. They also reveal some variations in prediction power of demographic variables for fluency scores.

The novel fluency task was developed based on Guildford and Hoepfner’s (Guilford, 1959, 1967; Guilford & Hoepfner, 1971) theory of divergent production and suggestions to measure this trait through complex verbal fluency. Correlations between scores on COSEF and the well-established verbal fluency tasks with single restrictions presented in this study indicate that the method has construct validity and actually measures a divergent operation in word fluency. The strongest correlation was seen between COSEF scores and number of correct words in the action verb fluency task. Although COSEF is a noun fluency task, the multiple semantic restrictions in terms of two adjectives or features that had to be present prevent the strategic use of subordinate words, making it similar to the action verb task. In verb fluency tasks, Östberg et al. (2005) argue that the participant must continuously “restructure meaning” since verbs are not taxonomically structured in the same way as noun fluency may be (i.e. animals). Speculatively, when adding semantic constraints in a semantic fluency task, the search within large semantic networks (i.e. category “animals”) is prevented. Instead access to the requested concepts is dependent on activation within and across different networks. This probably results in increased demands on strategic search processes and cognitive flexibility.

The COSEF task was reported to be more difficult than the standard fluency tasks by the NBD adults in this study. The reason may be the increased demand on semantic knowledge. This has been pointed out in previous studies when investigating other types of complex fluency tasks to detect early cognitive decline in dementia (Hart & Wade, 2006). Hart and Wade (2006) did not use multiple semantic restrictions in their fluency task, but suggest that complex fluency places greater demand on semantic abilities as it requires knowledge of what features and attributes define a concept. However, complex fluency may also increase demands on executive functions such as flexibility and effective search through semantic memory (Hart & Wade, 2006). This would be in line with the findings presented by Butler, Rorsman, Hill, Tuma, and Butters (1993), where results on more complex fluency tasks proved to be more impaired than those on standard fluency tasks, with single restrictions, in patients with frontal lobe tumours.

Age and education variables independently influenced COSEF scores, emphasising the importance of interpreting results in relation to background factors when implementing the method in clinical practice. The pair-matching in this study was based on age and education level since these variables proved to be significant demographic factors in the regression analysis. Results from the comparison of the pair-matched NBD group and the group of individuals with MS and subjective anomia indicate that COSEF may serve as a complement in detecting difficulties that are not highlighted using standard verbal fluency tests, with single restrictions. However, the results also point to the challenge in measuring subtle anomia. There was a large variation in results in the two groups and many participants with MS performed within one standard deviation below the mean of the reference data obtained from the complete NBD reference group. The large variation in performance seen in the complete NBD reference group highlights the difficulties in measuring and detecting subtle language and communication impairments. All participants with MS had reported that they experienced anomia, and their scores were significantly lower than the scores of the matched control group when compared on a group level. Wilcoxon signed ranks test which was used to explore differences between the groups is a non-parametric statistical test with limited power and the risk for type 1 errors is quite small, especially when both samples are small. However, on an individual level, only 12 of the 16 participants with MS produced fewer adequate responses and four of them produced as many or even more adequate responses than their matched control. When the Crawford and Howell (1998) modified t-test for single case research was used, only the results of five of the 16 participants was significantly lower than mean scores of the matched control group. The results also show that although the participants in general reported that they experienced COSEF as more difficult than the standard verbal fluency tests, all of them performed within normal variation on at least one of the four tasks, including COSEF. It seems the different tasks may pick up a variety of cognitive and semantic problems in individual participants, and this is most likely the case in clinical assessment of different patients too.

Detecting subtle language disorders in formal testing is a challenge. However, just asking the individual patients about their experiences may not be sufficient when an objective evaluation of outcome of intervention or progress of disease is warranted. The results indicate that a clinical assessment of subtle anomia should include a variety of fluency tests. Following further studies, and the establishment of clinical reliability and validity in different clinical groups, COSEF may be an important complement to the standard fluency tasks.

Analysis of reliability of the assessments using the scoring guidelines used for COSEF yielded promising inter-rater reliability ratings. The guidelines seemed to provide assessors with enough information and support to perform reliable assessments irrespective of whether they were trained in using the guidelines or naïve judges. However, further studies, with a greater number of both assessors and assessments are needed to determine degree of inter-rater reliability and intra-rater stability of the method for use in clinical practise.

The current study has some limitations. Firstly, the number of participants is low and recruitment trough convenient sample raises several issues with bias and low generalization. Among the participant with MS it is possible that individuals who are communicatively and verbally active are more sensitive to perceived word-finding difficulties and therefore more likely to participate despite degree of anomia. It is also possible that individuals with fatigue chose to not participate despite having more prominent anomia. Thus, conclusions should be drawn with great care. Another limitation is the variation in test settings. The goal was to ensure quiet and distraction-free test environments, but distractions may have occurred as most of the testing was done in homes or workplaces. The type of test environment may be especially crucial in the COSEF task as respondents may have used the setting, for example in a home, to visually search for possible objects that met the requirements during this task. However, some respondents reported that they mentally pictured environments in their search for items that fit the given characteristics. Furthermore, due to the presumption that COSEF would be a demanding task, we decided to present it last to avoid discouraging the respondents. Since the test order was not randomised or altered, we could not control for possible order effects. However, the word fluency tasks administered before COSEF are not particularly demanding and there were no signs of fatigue influencing the results.

Conclusions

The complex fluency task presented here seems to be able to add information in the assessment of divergent thinking operations through verbal fluency. Besides the ability to activate semantic representations and lexical retrieval, which is required in established word fluency tasks with single restrictions, the COSEF task requires the participant to search for items that exhibit multiple different features to be able to produce adequate responses. This is believed to put a greater demand than established fluency tasks on both lexical-semantic abilities and executive functions, and thus to be more sensitive to subtle word-finding difficulties.

The novel COSEF task presented here is based on a theoretical framework and seems to be a valid and reliable assessment method for use in research and clinical practice in the assessment of subtle language disorders. However, further studies need to confirm the findings presented in this paper regarding reliability and evaluate possible variations between NBD participants and different clinical groups in performance on COSEF. Finally, one should bear in mind that formal test situations are very different from real-life settings. The results on the tasks presented here are merely pieces in a big puzzle when assessing a person’s language and communication abilities.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This research was funded by the Swedish Research Council for Health, Working Life, and Welfare (FORTE) (Grant: 2015-00503).