Phonological Fluency Strategy of Switching Differentiates Relapsing-Remitting and Secondary Progressive Multiple Sclerosis Patients

The strategies used to perform a verbal fluency task appear to be reflective of cognitive abilities necessary for successful daily functioning. In the present study, we explored potential differences in verbal fluency strategies (switching and clustering) used to maximize word production by patients with relapsing-remitting multiple sclerosis (RRMS) versus patients with secondary progressive multiple sclerosis (SPMS). We further assessed impairment rates and potential differences in the sensitivity and specificity of phonological versus semantic verbal fluency tasks in discriminating between those with a diagnosis of MS and healthy adults. We found that the overall rate of impaired verbal fluency in our MS sample was consistent with that in other studies. However, we found no differences between types of MS (SPMS, RRMS), on semantic or phonological fluency word production, or the strategies used to maximize semantic fluency. In contrast, we found that the number of switches differed significantly in the phonological fluency task between the SPMS and RRMS subtypes. The clinical utility of semantic versus phonological fluency in discriminating MS patients from healthy controls did not indicate any significant differences. Further, the strategies used to maximize performance did not differentiate MS subgroups or MS patients from healthy controls.


Introduction
Multiple sclerosis (MS) is a chronic, debilitating, autoimmune disease of the nervous system that usually presents with a relapsing-remitting, and then later a progressive, course. Both the course of the illness and the presentation of motor and cognitive symptoms in terms of type and severity can vary signi�cantly from one individual to another [1]. e disease has been classi�ed as a frontal-subcortical dementia, as it causes demyelination of neurons mainly in frontal and subcortical regions [2].
It has been estimated that approximately 60% of patients with multiple sclerosis present with cognitive de�cits [3]. Consistent with the locations of disease-induced lesions, subsequent functional impairments comprise problems with attention, information processing speed, memory, and executive functioning, all of which limit the individual's ability to perform within the context of work and social relationships and may even compromise the safety of daily activities such as driving [4].
Although the cognitive effects of MS have been repeatedly documented in previous studies [3], it is not clear whether cognitive de�cits are exclusively the result of neurological damage or could also be the product of secondary symptoms of MS, such as depression and fatigue [5]. Further, individuals presenting with secondary progressive MS (SPMS) seem to consistently present with lower performance than those with relapsing-remitting MS (RRMS), possibly due to their older age, longer duration of illness, or more severe physical disability [3,6,7]. e strategies used to perform a verbal �uency test appear to be re�ective of cognitive abilities necessary for successful daily functioning. Several studies have documented the association between verbal �uency performance and community functioning in patients with dementia [8] and those with schizophrenia [9]. Although community functioning was beyond the scope of the present study, we considered verbal �uency performance a potentially important indicator of cognitive processes used in situations requiring generation of an organized approach to achieve successful responding. In fact, performance of patients with MS on verbal �uency has been consistently recorded to be lower than in healthy controls. Henry and Beatty [10] conducted a review of 35 studies examining verbal �uency performance in MS. ey found that patients with MS were substantially impaired on this measure, and that they presented with equal impairment on semantic and phonemic verbal �uency. e authors suggested that verbal �uency is one of the most sensitive measures of cognitive impairment in MS, along with the Symbol Digit Modalities Test of psychomotor speed. ey also concluded that patients with SPMS presented with more severe de�cits in comparison to patients with RRMS. In another study examining the use of strategies for maximizing word production, Tröster et al. [11] found that patients with MS produced more words than healthy controls; more interesting, however, was their compromised ability to switch between semantic or phonemic subcategories, despite producing an average number of words within each subcategory.
e aim of the current study was to explore (a) potential differences in strategies (switching and clustering) used to maximize word production by patients with RRMS versus SPMS, (b) impairment rates of verbal �uency (semantic and phonological) in MS patients, and (c) potential differences in the sensitivity and speci�city of phonological versus semantic verbal �uency tasks in discriminating between those with a diagnosis of MS and healthy adults.

Participants.
Participants were 148 (91 females or 61.50%) native Greek-speaking individuals, recruited from southwestern and from northern Greece, who took part in the present study voluntarily, aer providing a written informed consent for their participation. Seventy four of these individuals were diagnosed with clinically de�nite MS using McDonald's criteria [12]. Speci�cally, 60 (39 females or 65.0%) MS patients were diagnosed with RRMS (age: 41.18, SD 11.08; level of education: 1 . , SD 3.39, years) and 14 (10 females or 71.0%) were diagnosed with SPMS (age: 4 .8 , SD 7.38; level of education: 1 .07, SD 3. 9, years). MS groups differed from each other on the basis of physical disability status ( .9 0, 0.001), as determined by the Expanded Disability Status Scale (EDSS) [13] favouring the RRMS group ( 3. , SD 0.914), which had less severe physical disability relative to the SPMS group ( .14, SD 0.949). e two patient groups also differed with respect to duration of illness ( 7 0.883, 0.001), with the RRMS group having slightly more than half the illness duration ( 8. , SD 5.70 years) of the SPMS group ( 14.8 , SD .03 years). e two groups did not differ on level of depression ( 7 .150, 0.9 5), as assessed by the Beck Depression Inventory Fast Screen (BDI-FS) [14]. Both groups scored in the minimal depression range (RRMS group: 3. 0, SD 0.91; SPMS group: 4.10, SD 1.04). All patients were receiving standard MS medications (immunomodulators) and symptomatic medication for reduction of spasticity, fatigue, and spasm, as prescribed by their attending neurologists with doses adjusted for optimal clinical bene�t. We excluded participants from this MS group who suffered from any other medical condition (i.e., major psychiatric disorders, other neurological disorders, type II diabetes, traumatic brain injury, loss of consciousness >5 minutes, and hearing impairment not sufficiently corrected by a hearing aid) that might affect neuropsychological performance and nonnative speakers of the Greek language.
e remaining 74 (42 females or 56.75%) individuals were healthy adult participants (age: 4 .45, SD 11. ; level of education: 1 .4 , SD 3. 1, years) invited to take part in the study by their neurologist or family doctors, in order to improve study participation and compliance, or by a neuropsychologist. All healthy participants were screened with a medical questionnaire and physical examination for conditions that might in�uence cognitive performance. Exclusion criteria for the healthy participants were a history of psychiatric, neurological, or cardiovascular disorders or of substance abuse or dependence (including alcohol and benzodiazepine abuse), any other medical condition (including hearing impairment not sufficiently corrected by a hearing aid) that might affect neuropsychological performance, and nonnative speakers of the Greek language. We further excluded from the study potential participants who on initial testing obtained scores of less than 27 on the Greek validated version of the Mini Mental State Examination [15], a brief screening measure of global cognitive de�cits.

Procedure.
Healthy participants and MS patients were tested individually by psychologists in the clinic. Healthy participants were initially screened through a standardized interview at the beginning of the testing session by the project staff clinical neuropsychologist and physician, in order to exclude those with health problems or other exclusion criteria as described above. Healthy participants were also administered the Greek version of the Mini Mental State Examination [15]. e psychologists who tested the participants had been intensively trained in the administration procedures of various neuropsychological measures by doctoral-level clinical neuropsychologists.
All participants were assessed with the Greek Verbal Fluency Test [16]. e administration and scoring procedures were those proposed by Kosmidis et al. [16] and are described here brie�y. �n the semantic part of the test (categories), we asked participants to generate as many different animals, fruits, and objects as possible, each in a time period of 60 seconds. On the phonological part of the test (letters), we asked participants to generate as many different words as possible beginning with the Greek letters " " (chi), " " (sigma), and " " (alpha), each in a time period of 60 seconds, excluding proper nouns and variations of the same word. Variables in the present analyses were the total number of words produced on the semantic task and the total number of words produced on the phonological task. We also analyzed the strategies utilized to maximize word generation: semantic and phonemic clustering (i.e., the process of organizing words into semantically or phonemically related subcategories) and switching (i.e., shiing between subcategories or clusters). Detailed scoring rules for switching and clustering are provided in the Appendix of this paper and in accordance with Kosmidis et al. [16].

Statistical
Analysis. e normality assumption or homogeneity of variance of our data was initially con�rmed for each variable using the Kolmogorov-Smirnov test. Total word production, number of words related by clusters, and the number of switches were analyzed with multivariate analysis of variance. Equality of means between the MS groups and the healthy group, were analyzed using independent sample -tests. In cases where statistically signi�cant differences were found between the variances of groups, the -test of unequal variances was used and the degrees of freedom were estimated using the Welch-Satterthwaite approximation. Levene's test was employed in order to investigate the equality of variances. e level of statistical signi�cance was set at . We also calculated the number of MS patients impaired on total number of words produced on the semantic and phonological task separately (using as impairment criterion scores of 1.5 and of 2 SD below age and education-corrected normative Greek data) [16]. We further conducted a Receiver Operating Characteristic (ROC) analysis to investigate whether the phonological �uency task was more sensitive in detecting those with a diagnosis of MS from the healthy group as compared with the semantic �uency task. All analyses were conducted using the SPSS 17.0 soware.

Results
e two groups, that is, MS group (RRMS and SPMS patients as a single group) and controls did not differ signi�cantly on age ( , ) or level of education ( , ). Further, the two MS groups (RRMS and SPMS) did not differ signi�cantly from each other in age ( , ) or level of education ( , ). �owever, there were signi�cantly more female participants in both groups (RRMS and SPMS patients as a single group and healthy controls), ( , ). Total word production, number of words related by clusters, and number of switches produced by multiple sclerosis patients as a single group, but also RRMS and SPMS subtypes, and healthy controls were compared using an analysis of variance. We found signi�cant main effects for semantic �uency total word production ( , ), semantic �uency number of switches ( , ), phonological �uency total word production ( , ), phonological �uency number of words produced related by cluster) ( , ), and phonological �uency number of switches ( , ). Comparisons that were signi�cant in the post hoc tests are presented in Table 1.
We also recorded the percentage of MS patients who scored in the impaired range on total number of words produced on the semantic and phonological task, using scores 1.5 and 2 SD below age and education-corrected normative Greek data (see Kosmidis et al. [16]) as an impairment criterion. Twenty-two (29.72%) of our patients had impaired performance on the semantic task and 25 patients (33.78%) had impaired performance on the phonological task, when the impairment criterion was set at 1.5 SD below age and education-corrected normative Greek data. ese rates, however, decreased when the impairment criterion became more conservative and was set at 2 SD below age and education corrected normative Greek data. Speci�cally, 13 (17.56%) of our patients had impaired performance on the semantic task and 14 (18.98%) had impaired performance on the phonological task.
We further examined the contribution of possible moderator variables of verbal �uency performance on the total MS group, including EDSS (physical disability) score, duration of illness, and BDI-FS (depression severity) score using regression analyses. We found that duration of illness contributed signi�cantly to phonological �uency number of switches ( , SE , ). ere were no other signi�cant contributions of possible moderator variables.
We also conducted a Receiver Operating Characteristic (ROC) analysis to investigate whether the phonological �uency test was more sensitive in detecting those with a diagnosis of MS from the healthy group as compared with the semantic �uency test. e area under the curve (Figure 1) was essentially identical for the total number of words produced on the two test conditions: 0.701 for the semantic �uency condition and 0.697 for the phonological condition. In both conditions, sensitivity and speci�city were moderate and equal to each other. On the semantic condition, a score of 50.50 yielded 73% sensitivity and 54% speci�city, with a positive predictive value of 61% and on the phonological condition a score of 34.50 yielded 72% sensitivity and 47% speci�city, with a PPV of 58%. Similarly, the other variables of the word �uency test did not differentiate the conditions from each other with respect to their sensitivity and speci�city. More speci�cally, the area under the curve (Figure 2) for semantic and phonological switches was 0.642 and 0.797, respectively. On the semantic condition, a score of 24.00 yielded 72% sensitivity and 49% speci�city, with a PPV of 64% and on the phonological condition a score of 18.00 yielded 92% sensitivity and 55% speci�city, with a PPV of 87%. Finally, the area under the curve (Figure 3) for semantic and phonological words related by clusters was 0.588 and 0.733, respectively. On the semantic condition, a score of 10.00 yielded 74% sensitivity and 40% speci�city, with a PPV of 51% and on the phonological condition a score of 4.00 yielded 92% sensitivity and 47% speci�city, with a PPV of 36%.

Discussion
e present study was conducted to assess verbal �uency functioning in multiple sclerosis (MS) patients, including strategies utilized to maximize word production, and the clinical utility of phonological versus semantic �uency in discriminating healthy adults from MS patients. We also calculated the overall prevalence of verbal �uency impairments in these patients. We found that the overall rate of impaired verbal �uency in our MS sample was consistent with that in other studies [6,10]. However, unlike the majority of previous studies, which have found better performance for RRMS patients compared to the progressive subtypes [10,17,18], we found no differences between types of MS (SPMS, RRMS), on semantic or phonological �uency word production, or the strategies (clusters and switches) used to maximize semantic �uency. Recently, Potagas et al. [6] did not �nd signi�cant differences on a semantic word list generation task when comparing Greek SPMS patients to the other subtypes. is �nding is consistent with our �ndings, particularly as related to the semantic component of the verbal �uency test that we used in this study.
In contrast, we found that the number of switches differed signi�cantly in the phonological �uency task between the SPMS and RRMS subtypes. is �nding appears to be related to the duration of illness, which is signi�cantly longer for the SPMS subtype, as all other moderator variables (EDSS, depression status, age) that we examined did not contribute to total �uency production or the strategies utilized.
Further, the performance of the MS patients, either as a single group or as MS subtypes, differed signi�cantly from the healthy adults, showing poorer performance in total word production for both conditions and for the strategies utilized to achieve maximum word production. e �nding that our MS patients� performance differed signi�cantly as compared to the healthy group is also consistent with previous studies [10,17,19].
Regarding the clinical utility of the verbal �uency condition (semantic versus phonological) in discriminating MS patients from healthy controls, we did not �nd any significant differences in sensitivity and speci�city (according to the ROC curve analyses). Further, the strategies used to maximize performance (clustering and switching) did not differentiate MS subgroups or MS patients from healthy controls, as we again did not �nd any signi�cant differences in sensitivity and speci�city for these variables. Our �nding is in contrast to a previous study [19] that found a semantic �uency measure to discriminate adequately between MS patients and controls. Furthermore, in a recent quantitative review article of 35 studies, the authors noted that phonemic  and semantic �uency tests are equally sensitive to MS, therefore, encouraging the view that the results on �uency tests for di�erent languages will be comparable. �is �nding however, requires further investigation, as our �ndings do not support this hypothesis, at least not for the Greek language. Several potential limitations to the generalizability In conclusion, our �ndings do not support di�erences between the SPMS and RRMS subtypes, in verbal �uency performance, or the strategies utilized to achieve word production, with the exception of number of switches which di�ered signi�cantly in the phonological �uency task. Further, semantic-versus-phonological �uency do not appear to adequately discriminate MS patients from healthy controls or SPMS from RRMS patients. In contrast, we found a rate of impaired verbal �uency consistent with previous studies.