Comparison of verbal fluency performance in Kannada-speaking adults with and without euthymic bipolar disorder type 1

Abstract

Individuals with euthymic bipolar disorder (BD) type I exhibit deficits in executive functions. Although less explored in the BD population, the tasks of verbal fluency (VF) have shown great potential in understanding semantic organization. This study provides an extensive exploration across the letter and semantic VF tasks in 27 demographically matched euthymic BD-I and healthy controls (HC). The groups were compared on measures of the total number of correct words (TNCW), temporal pattern analysis, number of clusters (NC), mean cluster size (MCS), number of switches (NS), and error pattern. An overall reduction in letter fluency scores (the TNCW, number of switches, and NC) as compared to semantic fluency scores was noted for both groups, with a significantly greater decrease in the BD-1 group. The MCS and temporal pattern were relatively similar across the two groups. The influence of education with no gender difference was observed between groups with error types prevalent in both groups. The study findings call attention toward assessing the VF performance in persons with BD in terms of error production and the strategies employed (clustering–switching).

Keywords:

• Bipolar disorder
• euthymic state
• letter fluency
• semantic fluency
• temporal pattern
• verbal fluency

Introduction

Bipolar disorder (BD) is a psychiatric illness, with BD type I (BD-I) exhibiting a fully syndromic episode of mania and major depressive episodes along with euthymic (remission) periods and with bipolar disorder type II (BD-II) characterized by several hypomanic and depressive episodes that keep alternating back and forth, without any full-blown manic or mixed episodes (American Psychiatric Association, 2013). Apart from the psychosocial issues, persons with BD often experience deficits affecting executive and linguistic functioning such as processing speed and verbal retrieval (Fleck et al., 2003; Radanovic et al., 2013; Raucher-Chene et al., 2017).

Among the various neuropsychological tests, a word retrieval task of verbal fluency (VF) is commonly used to assess both executive control processes such as selective attention, selective inhibition, mental set shifting, internal response generation, and self-monitoring as well as verbal ability in healthy and neurological conditions (Patra et al., 2020; Shao et al., 2014). VF has immense importance in social communication and occupational functioning. It provides information regarding semantic memory, vocabulary size, lexical knowledge, and speed of lexical access, which provides an insight into broader language and executive functioning in an individual. This task requires the rapid generation of words belonging to a particular category (semantic fluency [SF]) or starting with a specific letter (letter fluency [LF]) within a specified time constraint, by avoiding repetition of words or generating words that do not belong to the given category/letter. For both the LF and SF task, the cultural and linguistic aspects are considered for selection of the letter and category stimuli while testing in each language and culture (Aki et al., 2022; Oberg & Ramírez, 2006; Villalobos et al., 2022).

Variations in task demands are documented with the word retrieval during SF relying more on the existing semantic connections and LF relying upon different retrieval strategies by inhibiting activation of semantically or associatively related words (Katzev et al., 2013; Luo et al., 2010). Among the two tasks, SF relies more on accessing and retrieving semantic information with word retrieval requiring additional control processes such as the selection of appropriate items from competing targets (Sung et al., 2013). While both tasks involve the left hemispheric fronto-temporal network, the LF task is seen to activate the left frontoparietal regions and the SF task activates the left temporal regions (Biesbroek et al., 2016; Gourovitch et al., 2000). In individuals with BD, the studies report decreased activation in prefrontal cortex regions (Sun et al., 2018; Tassi et al., 2022) and temporal regions (Alici et al., 2022; Husain et al., 2021).

The response obtained during the VF task is commonly analyzed in terms of word productivity or the total number of correct words (TNCW) retrieved during the task. While most studies report that word productivity during VF is affected in people with BD (Altshuler et al., 2004; Bertschy et al., 2023; de Almeida Rocca et al., 2008; Dixon et al., 2004; Frangou, 2005; Thompson et al., 2005), few studies have reported no difference in comparison with healthy controls (HC) or across VF tasks (Cavanagh et al., 2002; Depp et al., 2013; Erol et al., 2014; Martínez-Arán et al., 2002; Zubieta et al., 2001). Recently, studies have stated that participants with BD have a moderate impairment in VF which is linked to semantic memory dysfunction of storage/organization (Ceylan et al., 2020; Chrobak et al., 2022; Raucher-Chene et al., 2017; Weiner et al., 2019).

Across VF tasks, the studies reveal that both SF and LF may be impaired in individuals with BD (Dixon et al., 2004; Meesters et al., 2013; Nehra et al., 2006; Radanovic et al., 2013; Raucher-Chene et al., 2017; Torrent et al., 2006). A greater extent of deficits for SF are observed in all BD mood states, especially in mania (Dixon et al., 2004; Raucher-Chene et al., 2017; Weiner et al., 2019). This suggests greater involvement of the semantic system pathways in BD participants (Dixon et al., 2004; Raucher-Chene et al., 2017; Weiner et al., 2019).

Apart from analyzing the word productivity during VF, though limited, a few studies have documented clustering and switching in VF (Bertschy et al., 2023; Drakopoulos et al., 2020; Chang et al., 2011; Sung et al., 2013; Weiner et al., 2019). The analysis of clustering and switching gives a better index of word retrieval organization and lexico-semantic memory integrity (Patra et al., 2020; Raboutet et al., 2009; Raucher-Chene et al., 2017). Clustering is an executive-linguistic subprocess, which provides information regarding how an individual categorize information during the word retrieval process. A phonological or semantic cluster refers to generation of two or more successive words which have similarities in sound (phonological: park, parrot, part) or have related content (sematic: sofa, chair, table) respectively (Martz et al., 2022). The ability to switch refers to how frequently one shifts from one cluster to another set of words. While clustering is linked to temporal lobe in relation to word storage and semantic verbal memory, the switching is linked to frontal lobe in relation to cognitive flexibility and word search (Weiss et al., 2006; Unsworth et al., 2011).

Greater impairment during SF with clustering and switching deficits has been reported in BD-I. The clustering pattern analysis reveals less coherent category clusters during SF tasks in both types of BD (Chang et al., 2011; Sung et al., 2013). The euthymic BD participants have disorganized semantic structures and use less conventional methods for categorization in semantic memory (Chang et al., 2011). In the Sung et al. (2013) study, although word productivity was reported to be intact during animal fluency, the clustering pattern was dissimilar for noneuthymic BD and HC. For instance, the subcategories produced while retrieving wild and farm animals were widely separated in BD and unlike HC, they did not produce low-frequency animal clusters of sea animals and reptiles. Similarly, during the supermarket fluency task, while HC produced distinct clusters resembling the organization of items in the supermarket, the BD participants did not generate semantically related items together (Sung et al., 2013). Larger phonological cluster sizes, reduced cluster ratio, and an increased number of switches (NS) are documented among BD participants with more deficits in mania as compared to the depression and euthymic group (Weiner et al., 2019). Raucher-Chene et al. (2017) and Weiner et al. (2019) suggested the presence of an over-active semantic network with faster activation spreading in the manic state. Drakopoulos et al. (2020) also reported that the total NS and total correct responses were reduced in the BD group who are inactive. The decreased switches in the depressive phase in BD were related to slower processing speed. These observations are considered to be supportive of a functional deficit hypothesis (deficits in semantic access and retrieval) rather than a storage deficit hypothesis (deterioration of semantic knowledge store) in persons with BD (Raucher-Chene et al., 2017; Sung et al., 2013; Weiner et al., 2019). In a study by Bertschy et al. (2023), no differences were observed between euthymic BD and controls in clustering and switching measures. They noted a decrease in word production during the depression and an increase in the NS during hypomania supporting an over-activation of the semantic network during this state.

The word retrieval during VF is also analyzed in terms of the pattern of error production. The errors can be seen in terms of intrusion (words produced not belonging to the appropriate semantic category or starting with the required letter), perseverations (repetition of a word already produced) and nonwords (irrelevant word production). Individuals with BD are also prone to error production. Weiner et al. (2019) hypothesized that the irrelevant word production may occur in BD due to executive deficits and the clinical symptoms of distractibility and/or flight of ideas. Although the presence of error is reported, most of the studies do not describe the error types noted in the BD group (Drakopoulos et al., 2020; Sung et al., 2013; Weiner et al., 2019), with the exception of a study by de Almeida Rocca et al. (2008). The authors documented the presence of an intrusion type of error in the euthymic group, although the error count was overall increased in HC. de Almeida Rocca et al. (2008) suggested that BD participants retrieved words slowly in order to follow the instructions resulting in lesser error score. Verbal intrusions have been linked to a variety of disorders, including BD, and have been suggested to indicate a lack of inhibition and/or increased vulnerability to interference (Lebowitz et al., 2001). In the analysis by Weiner et al. (2019), however, the error count for both the BD and control groups were identical during LF and SF.

The temporal analysis of the VF production in terms of the word retrieval in shorter time quadrants (such as every 15 seconds or every 10 seconds) has also been considered. VF production is governed by different cognitive processes through each stage. The initial stage/process consists of a semiautomatic retrieval of words that are common, whereas the later stages require effortful processing (Crowe, 1998; Demetriou & Holtzer, 2017; Patra et al., 2020; Raboutet et al., 2009). Therefore, an analysis of temporal patterns would provide a better understanding of the underlying cognitive mechanisms (Luo et al., 2010).

VF output is influenced by a number of factors including the effect of gender, age, education level, language, ethnicity and culture. In general, participants with higher education were found to perform better on both LF and SF tasks compared to those with lower levels of education (Aki et al., 2022; Kosmidis et al., 2004; Mathuranath et al., 2003; Van Der Elst et al., 2006). Age and gender were also said to affect VF tasks, especially for SF (Mathuranath et al., 2003; Nogueira et al., 2016; Troyer et al., 1997). The younger population generated a larger number of words in the SF task, whereas females performed better when compared to men (Hirnstein et al., 2023; Troyer et al., 1997). Some authors also report that age affects LF (Brickman et al., 2005; Loonstra et al., 2001). The word productivity is also reported to vary based on the language, ethnicity, and culture (Oberg & Ramírez, 2006; Kempler et al., 1998).

Despite the understanding of the word productivity deficits in BD using a VF task, there exists a dearth of information on task effects, clustering-switching, and error types during VF specifically in the euthymic population. This lacuna of research is a matter of concern especially in the Indian context, considering the recent reports of 7.6 million (0.6%) suffering from BD in India (Sagar et al., 2020). The interest of the present study was, therefore, an extensive exploration of the VF performance across the LF and SF tasks in terms of word productivity as a function of time, clustering-switching, and error pattern analysis in individuals with BD-I in the euthymic state. An exploration of the VF performance in euthymic state of BD-I will aid in understanding if VF is affected and how these deficit affects their psychosocial functioning. VF analysis can help in identifying the targets for treatment such as cognitive remediation or even medication. The findings of the proposed study will provide information regarding extent of VF deficits and how the deficit affects their psychosocial functioning in aspects of speech and language. Disturbances or impairments in language can cause problems with reasoning as well as thinking, and deficits in communication skills and abilities may lead to poor social interaction at personal as well as professional levels. The interest of the present study is therefore an in-depth analysis of the VF performance of individuals with BD I in the euthymic state across LF and SF.

We hypothesize that the VF outcome measures (total word productivity and clustering-switching metrics) will be more affected in the population with BD in comparison with the control group. We further hypothesize that the response on the temporal analysis (word productivity as a function of time) would differ between BD and the control group.

Materials and methods

Participants

Before the initiation of the data collection, the study was approved by the Institutional Ethics Committee (IEC No: 03/2020) and was registered under the Clinical Trials Registry of India (CTRI/2020/04/024617). In the present study, 27 persons with BD-I and 27 healthy control (HC) participants (age-, gender- and education matched) were considered. This sample size was estimated by calculating the power from the outcome measures.

Participants were recruited from Karnataka, a state in South India. All the participants were between the ages of 18 years to 65 years and spoke fluent Kannada, a Dravidian language. Participants with BD-I were selected from those having the primary clinical diagnosis of BD-I by a psychiatrist based on standard DSM V diagnostic criteria. The participants were in the euthymic mood state with scores less than 12 on Young Mania Rating Scale (Young et al., 1978) and scores less than 7 on Hamilton Depression Rating Scale (Hamilton, 1960), for the past two months as rated by a psychiatrist and on a stable medication regime (lithium or sodium valproate) for the last three months. Individuals having any neurological conditions (such as epilepsy/traumatic brain injury) /history of substance abuse past six months /schizoaffective disorder /history of electroconvulsive therapy past six months or use of topiramate or benzodiazepines in the past two weeks were not considered for the study. Participants reporting of any memory/naming concerns during screening were excluded from the study. The HC were recruited from the community, did not meet the diagnostic criteria for those affected with BD, and were matched individually corresponding to age, gender, and level of education. HC with a history of psychiatric illness or a first-degree relative with a history of affective disorders, dementia, cerebrovascular disease, and history of substance abuse (except nicotine) were excluded.

Verbal fluency trials

In the present study, two tasks of VF (LF using p and SF using animal fluency) were administered to all the study participants who met the inclusion criteria. This specific category of “animal” was chosen as it is common and familiar to the study participants in the Indian context. The letter pwas considered for the LF task as it is one of the most frequently used stimuli for neuropsychological testing in this language.

The participants were instructed to generate as many words as possible (a word starting with a specific letter for LF and words belonging to animal category) within a stipulated time of one minute in Kannada language. The participants were also instructed to exclude names of people, places, numbers, verbs, and repetitions of the same word. A training trial for both SF (supermarket items) and LF (letter s) for 60 seconds each was provided before the actual testing. The total time taken for administration was approximately 10 minutes for each participant.

Due to the COVID restrictions, all testing was conducted telephonically (n = 44) or by direct interview (n = 10) in the Kannada language after obtaining informed written consent. We decided to proceed forward with both modalities of testing, considering the literature evidence provided by Rankin et al. (2005). They reported that the in-clinic/face-to-face interviews as well as telephonic interviews for VF have a linear relationship and stated that the telephonic interview can be an appropriate substitution (Rankin et al., 2005). VF tasks assessed via telephone and in person has been reported to produce a normal score distribution (Matchanova et al., 2021; Rapp et al., 2012), with the only limitation being the quality of recorded calls being affected in exceptional cases (Marceaux et al., 2019).

Data coding and analysis

Offline analysis was carried out by transcribing all the words generated during both VF tasks. We measured the following outcome measures in persons with BD-I and HC:

Total number of correct words (TNCW)

The total number of words correctly produced during both tasks was computed. For instance, if the participant says “parrot, pat, pilot, Paris, pat” during LF, the TNCW is considered as 03 (parrot, pat, and pilot as correct words) while excluding Paris (name of a place) and the repetition of the word pat. For SF, if the participant says “cat, dog, cow, lion, tiger, dog,” the TNCW is 05 (cat, dog, cow, lion, and tiger are correct words) while excluding the repetition of the animal “dog.” For the LF task, the commonly used borrowed English words (e.g., paper, pen, and pencil) retrieved were considered as correct words and for the SF task, the participant was scored correctly for only the animal names retrieved in the Kannada language.

Temporal analysis

Temporal pattern analysis is an analysis of the production of words as a function of time across the four-time quadrants: 0–15 seconds (Interval 1), 16–30 seconds (Interval 2), 31–45 seconds (Interval 3), and 46–60 seconds (Interval 4). The number of correct words said by the participant in each quadrant was calculated.

Number of clusters (NC)

Analysis of clusters (for both LF and SF) was carried out by considering the consecutive words as per the Troyer et al. (1997) guidelines. The error responses during the word retrieval were also considered for clustering-switching analysis as it has been reported to provide additional information regarding the underlying cognitive processes (Patra et al., 2020; Raboutet et al., 2009; Tröster et al., 1998). For example, for the response, “cow, dog, cat, goat, lion, tiger,” the total NC will be taken as two—cow, dog, cat, and goat as domestic animals and lion and tiger as wild animals. In the case of LF, if the participant says “pat, Paris, pan, plane, plate, prank,” the NC will be three—pat, Paris, and pan taken as words with the same initial CV, plane and plate having same initial CCV and the word prank taken as an unclustered word.

Mean cluster size (MCS)

This is computed by dividing cluster size by the total NC. For example, if a participant says "cow, dog, cat, goat, lion, tiger," the cluster size, that is, the number of words in one cluster, is measured first. The cluster size of a single-word cluster is 0, while the cluster size of a four-word cluster is 3. In the example above, the cluster size would be 03 and 01, respectively, having a MCS of 02.

Number of switches (NS)

Switching is known as a shift from one cluster to another, or from a word in a cluster to an un-clustered word, or vice versa, with the final word of the first cluster being used as the beginning word for the second cluster observed (Troyer et al., 1997). For example, “cow, dog, cat, goat, lion, tiger” will be considered as one switch of the category cluster of domestic animals to another of wild animals, and the total NS in the example is calculated as one.

Error pattern analysis

Errors, if any, produced during the VF task were analyzed in terms of the following error types: intrusion, perseveration, and nonwords. Intrusion errors for both LF and SF were considered when words started with any other letter or were phonologically similar (phone, fruit), proper nouns (Paris, Preethi) and when the words do not belong to the particular category (e.g., zoo for animal category). Perseveration errors were considered as any word which was repeated within 60 seconds. For example, if the response was “pat, pen, pat” the word pat occurring the second time will be taken as a perseverative error as it is repeated twice. Nonwords were considered as any word the participant has said that carries no meaning. For example, in the following words “pat, pab, play,” the word “pab” carries no meaning and would be considered as a nonword.

. .

Inter-rater reliability

Two raters independently analyzed and scored the output of both LF and SF tasks for 10% of the randomly selected sample generated by participants. They computed all the outcome measures considered for the study. For items with no agreement, the scoring was performed again by the two raters after a discussion in line with Troyer et al. (1997) guidelines.

Statistical analysis

Using descriptive statistics, the VF findings in both the BD and HC groups are reported. The linear mixed model was used to compare the variables (TNCW, as a function of time, NC, MCS, and NS) between BD-I and HC. In the analysis, the group is taken as the fixed effect and the matching ID as the random effect. The analysis across the tasks (LF, SF), gender, and education were done using analysis of variance with p value set at less than .05. Effect size was calculated on the basis of Cohen’s d to analyze the difference between the groups. Furthermore, the error pattern was analyzed in terms of type and frequency of error types. The inter-rater reliability (two-way random effects model, type absolute agreement, and average measures Intra class correlation [ICC]) was estimated for 10% of the randomly selected sample performed by the two raters. The lowest value of ICC for LF and SF was 1.00 and 0.89, respectively, which shows high reliability. The statistical analysis was performed using SPSS version 20.

Results

The demographic characteristics of the 27 participants with BD-I and the age-, gender-, and education-matched HC are depicted in Table 1. The mean age of participants with BD-I was 46.85 ± 12.37 years and of HC was 45.75 ± 13.26 years (22 male and 32 female) with symptom onset between 15 and 25 years of age (51%). The BD-I group was euthymic for the past two months, had a history of more than one manic/depressive episode, had at least one hospitalization (66.67%), and was under either lithium or sodium valproate as primary medication. A total of 52% of the participants with BD-I had a low education level.

Table 1. Demographic characteristics of study participants.

Demographics		Mean (SD)	BD-I n (%)	Mean (SD)	HC n (%)
Age range (years)	25–35	29.37 (4.33)	08 (29.63)	24.75 (2.27)	04 (14.81)
	36–45	41.8 (2.78)	05 (18.52)	40.87 (3.18)	08 (29.63)
	46–55	50.28 (2.81)	07 (25.93)	50.00 (2.56)	07 (25.93)
	56–65	62.86 (3.48)	07 (25.93)	61.12 (2.47)	08 (29.63)
Gender	Male	–	11 (40.74)	–	11 (40.74)
	Female	–	16 (59.26)	–	16 (59.26)
Years of education	3rd–5th	–	07 (25.93)	–	05 (18.52)
	6th–10th	–	07 (25.93)	–	13 (48.14)
	11th–12th	–	05 (18.52)	–	01 (3.70)
	Diploma	–	03 (11.11)	–	03 (11.11)
	Graduate	–	05 (18.52)	–	05 (18.52)
Family history	Positive	–	06 (22.22)	–	–
	Negative	–	21 (77.78)	–	–
Age of onset (years)	15–30	22.68 (6.77)	21 (77.78)	–	–
	40–65	44 (4.04)	06 (22.22)	–	–
No. of hospitalizations	Nil	–	09 (33.33)	–	–
	1–5	–	14 (51.85)	–	–
	6–10	–	02 (7.41)	–	–
	≥10	–	02 (7.41)	–	–
Total no. of episodes	≤20	–	12 (44.44)	–	–
	>20	–	04 (14.81)	–	–
	Unknown	–	11 (40.74)	–	–
Medication	Lithium	–	19 (70.37)	–	–
	Sodium Valproate	–	08 (29.63)	–	–
Comorbid conditions	Diabetes	–	02 (7.41)	–	–
	Hypertension	–	02 (7.41)	–	01 (3.70%)
	Thyroid diff.	–	05 (18.52)	–	–
	Asthma	–	–	–	01 (3.70%)
	Seizures	–	01 (3.70)	–	–

BD = bipolar disorder; HC = healthy controls.

VF measures in BD-I in comparison to HC

The TNCW generated as a function of time, clustering (NC and MCS), and switching measures (Table 2) were used to evaluate VF output in both groups during LF and SF tasks.

Table 2. Differences in verbal fluency measures of letter and semantic fluency in BD-I and healthy controls.

	BD-1 (n = 27)				HC (n = 27)			Mean Difference (BD vs. HC)	Cohen’s d (BD vs. HC)	p-value
	Male		Female	Total	Male	Female	Total
TNCW	LF	5.73 (3.47)	3.31 (3.32)	4.30 (3.53)	6.45 (2.94)	7.19 (4.90)	6.89 (4.16)	−2.59 (−4.34 to −0.84)	0.67	0.005**
	SF	12.36 (3.14)	11.44 (3.78)	11.81 (3.50)	14.45 (3.21)	15 (4.98)	14.78 (1.79)	−2.96 (−4.81 to −1.11)	0.76	0.003**
NC	LF	2.36 (1.80)	1.19 (1.28)	1.67 (1.59)	2.55 (1.04)	2.81 (2.20)	2.70 (1.79)	−1.04 (−1.84 to −0.23)	0.61	0.013**
	SF	4.18 (1.47)	3.5 (1.55)	3.78 (1.53)	3.82 (0.87)	4.75 (1.98)	4.37 (1.67)	−0.59 (−1.45 to 0.27)	0.37	0.168
MCS	LF	1.49 (0.87)	0.81 (1.03)	1.09 (1.01)	1.48 (0.64)	1.35 (0.85)	1.40 (0.76)	−0.31 (−0.74 to 0.12)	0.34	0.152
	SF	2.11 (0.72)	2.31 (0.93)	2.23 (0.84)	2.55 (1.48)	1.69 (0.52)	2.04 (1.09)	−0.19 (−0.34 to 0.72)	0.19	0.476
NS	LF	2.55 (2.02)	1.63 (1.71)	2.00 (1.86)	3.27 (2.28)	4.19 (2.46)	3.81 (2.39)	−1.81 (−2.73 to −0.90)	0.85	0.001**
	SF	5.73 (2.87)	5.75 (2.82)	5.74 (2.78)	5.91 (2.07)	8.88 (3.96)	7.67 (3.59)	−1.92 (−3.26 to −0.58)	0.59	0.007**

BD-I = bipolar disorder type I; HC = healthy controls; TNCW = total number of correct words; NC = number of clusters; MCS = mean cluster size; NS = number of switches; LF = letter fluency; SF = semantic fluency; M = male; F = female; Cohen’s d interpretation (effect size): 0.2-small, 0.5-medium, 0.8-large. **p < 0.05.

In terms of the TNCW produced, a lesser number of right words was produced by BD-I as compared to HC during LF and SF task. Based on the linear mixed model analysis, the mean difference between the groups for LF and SF tasks further indicates that the difference was statistically significant with a larger number of words produced by the HC. When comparing the LF and SF task outputs using univariate ANOVA within the same group, a significant main effect was seen, F(1,104) = 106.21, p = 0.001, with higher correct word output during the SF task. Across gender, no statistically significant differences were noted for both LF (p = 0.144) and SF (p = 0.506).

Clustering during the VF tasks was analyzed for the NC and the MCS. The NC was found to be lesser among BD-I participants during the LF task, and the SF task. The mean difference in the NC during LF (p = 0.013) and SF (p = 0.168) indicated a statistically significant higher NC among HC for LF and insignificant for SF. A comparison of tasks within the group using univariate ANOVA showed a significant main effect for the NC between LF and SF, F(1,104) = 35.45, p = 0.001. This implies that between tasks variations are evident and a greater NC were noted during the SF task. Gender differences in the NC were not statistically significant in both groups for LF (p = 0.126) and for the SF task (p = 0.071).

In terms of MCS, though the mean scores were lower in BD-I as compared to HC for LF [1.09 (1.01) for BD-I and 1.40 (0.76) for HC] and SF task [2.23 (0.84) for BD-I and 2.04 (1.09) for HC], the mean difference scores between groups were not statistically significant for LF [−0.31 (−0.74 to 0.12), p = 0.152] and SF [0.19 (−0.34 to 0.72), p = 0.476]. A comparison of tasks within the group using univariate ANOVA showed a significant main effect for the mean size of clusters between LF and SF, F(1,104) = 24.44, p = 0.001, with greater cluster size in the SF task. In terms of gender difference, the mean size of clusters was nonsignificant in both groups (p = 0.259 in LF and p = 0.044 in SF).

The NS differed significantly between both groups for LF [2 (1.86) for BD-I and 3.81 (2.39) for HC] and SF [5.74 (2.78) for BD-I and 7.67 (3.59) for HC]. In the letter and semantic conditions, the mean difference in the NS found among both groups [LF: −1.81 (−2.73 to −0.90), p = .001; SF: −1.92 (−3.26 to −0.58), p = 0.007] concurred that a statistically significant greater NS were seen in the HC group in both task outputs. Within-group differences with univariate analysis of variance (ANOVA) between LF and SF were seen to have a significant main effect, F(1,106) = 52.20, p = 0.001 suggesting more NS during SF task. The NS was seen to be increased in females for both groups (except during LF in the BD group), though statistically not significant (p = 0.126 in LF and p = 0.093 in SF tasks).

Comparison of VF outcome measures across education level

The education level of the participants of both groups is seen as a major influence (Table 3). The participants were divided into two groups according to education level with those with less than 10 years of education classified as lower education and those with above 10 years of education grouped as higher education. The mean and standard deviation of the BD-I participants and HC with higher education is seen to be greater than those with lower education. Using univariate ANOVA, these differences were found to be statistically significant for BD-1 group (p < .05) for all the measures except MCS. In HC group, the differences were not statistically significant (p > .05) except for the TNCW (SF). On group comparison between BD-1 and HC, the difference in mean scores between the higher educated participants in BD-I and HC were not statistically significant (p < 0.05) except for TNCW(SF) whereas the difference between the lower educated groups was seen to be statistically significant (p > 0.05) except MCS measures.

Table 3. Comparison between high and low education in participants with BD-I and HC.

			Low Education [M(SD)] (BD: n = 14; HC: n = 14)	High Education [M(SD)] (BD: n = 13; HC: n = 13)	p-value
TNCW	LF	BD-I	2.5 (2.92)	6.23 (3.14)	0.003**
		HC	5.77 (3.54)	7.93 (4.55)
	Cohen’s d		1.01	0.43
	SF	BD-I	10 (3.16)	13.77 (2.77)	0.001**
		HC	12.77 (3.79)	16.64 (3.95)
	Cohen’s d		0.80	0.84
NC	LF	BD-I	1 (1.36)	2.38 (1.56)	0.085
		HC	2.46 (2.11)	2.93 (1.49)
	Cohen’s d		0.83	0.36
	SF	BD-I	2.86 (1.17)	4.77 (1.23)	0.020**
		HC	4.31 (1.84)	4.43 (1.55)
	Cohen’s d		0.95	0.24
MCS	LF	BD-I	0.68 (0.93)	1.53 (0.92)	0.048**
		HC	1.21 (0.69)	1.57 (0.79)
	Cohen’s d		0.65	0.05
	SF	BD-I	2.45 (0.99)	1.99 (0.58)	0.729
		HC	1.84 (0.64)	2.22 (1.38)
	Cohen’s d		0.716	0.21
NS	LF	BD-I	1.07 (1.44)	3 (1.78)	0.031**
		HC	3.38 (2.39)	4.21 (2.39)
	Cohen’s d		1.18	0.57
	SF	BD-I	4.78 (2.94)	6.77 (2.88)	0.111
		HC	7.23 (3.17)	8.07 (4.03)
	Cohen’s d		0.80	0.39

BD-I = bipolar disorder type I; HC = healthy controls; TNCW = total number of correct words; NC = number of clusters; MCS = mean cluster size; NS = number of switches; LF = letter fluency; SF = semantic fluency; Cohen’s d interpretation (effect size): 0.2-small, 0.5- medium, 0.8- large. **p < 0.05.

Temporal pattern

On analysis of the total number of words produced as a function of time (Table 4), the highest number of words were noted to be produced during the first quadrant (0–15 seconds) in both groups for both LF and SF tasks, which were statistically significant [BD-I and HC (LF): p = .013; BD-I and HC (SF): p = 0.015 respectively]. Productivity of verbal output progressively decreased for the successive quadrants with the least number of words being produced in the final quadrant (45 sec–60 sec) for both groups respectively [BD-I and HC (LF): p = 0.611; BD-I and controls (SF): p = 0.303]. It was observed that when comparing the verbal production of HC and BD-I participants using the linear mixed model of analysis, in the final quadrant, a larger number of words were produced by the HC, although statistically insignificant.

Table 4. Temporal pattern of word productivity in BD-I and HC across tasks.

	0 sec–15 sec	16 sec–30 sec	31 sec–45 sec	46 sec–60 sec
Letter fluency
BD-I	2.15(1.77)	1.15(1.03)	0.59(0.80)	0.52(0.85)
HC	3.11(1.99)	1.56(1.63)	1.63(2.10)	1.04(1.56)
Cohen’s d	0.51	0.29	0.65	0.41
Semantic fluency
BD-I	6.59(2.44)	2.59(1.62)	1.78(1.40)	0.89(1.48)
HC	7.67(2.34)	3.19(1.55)	2.33(1.57)	1.67(1.59)
Cohen’s d	0.44	0.37	0.37	0.51

BD-I = bipolar disorder type I; HC = healthy controls; LF = letter fluency; SF = semantic fluency; Cohen’s d interpretation (effect size): 0.2-small, 0.5-medium, 0.8-large.

Error pattern analysis

Three types of errors (intrusions, perseverations, and nonwords) were observed in both groups, among which the perseverative error type was the most common. Overall, more participants in the HC group made these errors during VF production compared to the BD group (except for perseveration during the SF task). In terms of mean error, while all the error types during SF were more in the BD group, the HC group exhibited more intrusion and perseverative errors during the LF, as indicated in Table 5.

Table 5. Error frequency and average values for letter and semantic fluency tasks.

Error type	BD-I		HC
	LF	SF	LF	SF
Intrusion, n (%)	6 (22.22)	7 (25.92)	12 (44.44)	8 (29.62)
Mean error	0.30	0.63	0.78	0.45
Perseveration, n (%)	4 (14.81)	19 (70.37)	10 (37.03)	19 (70.37)
Mean error	0.34	1.78	0.82	1.63
Nonwords, n (%)	1 (3.70)	5 (18.51)	2 (7.40)	3 (11.11)
Mean error	0.08	0.26	0.08	0.12

BD-I = bipolar disorder type I; HC = healthy controls; LF = letter fluency; SF = semantic fluency.

Discussion

Language deficits and executive functions have been well acknowledged in persons with BD (Radanovic et al., 2008). The simple and easy-to-administer behavioral measure of VF has gained importance in BD research for exploring semantic memory organizations (Radanovic et al., 2008; Raucher-Chene et al., 2017). The present study findings provide further evidence for VF deficits in persons with BD-I in the euthymic state. The VF impairment in the euthymic state was characterized by a lesser number of correct word production, which decreased as a function of time with reduced clustering (NC) and decreased NS. This observation agrees with earlier studies among the euthymic population (Drakopoulos et al., 2020; Raucher-Chene et al., 2017; Weiner et al., 2019). The participants with BD have a differential pattern of semantic organization (Chang et al., 2011), which can lead to changes in daily activities including speech and communication problems (Rossell, 2006). A lack of gender difference in VF performance noted in the present study is contrary to Suwalska and Łojko (2014).

In terms of word productivity, the BD-I euthymic group had a lower number of correct words for both SF and LF than the HC group. Reduced verbal performance in the euthymic community supports the notion of a decline/deterioration in semantic knowledge organization or impaired activation and inhibition mechanisms in semantic information retrieval (Raucher-Chene et al., 2017; Sung et al., 2013; Weiner et al., 2019). This observation is supported by earlier reports of deficits in euthymic BD-I during LF (Altshuler et al., 2004; de Almeida Rocca et al., 2008; Thompson et al., 2005) and SF (Erol et al., 2014). However, a few studies have contradicting findings in which they have reported no differences when comparing VF in BD in the euthymic state and in HC (Cavanagh et al., 2002; Depp et al., 2013; Sung et al., 2013; Zubieta et al., 2001). These observations, however, should be considered with caution, as many of these studies did not consider the factors influencing the VF performance including the mood states or medication of the BD-I participants, which have been controlled for in the present study.

Another striking observation from the study is the greater impairment noted for tasks of LF compared to SF tasks in the BD-I group based on the effect size. The greater deficit during LF has been attributed to the task demands of the LF and SF tasks which require the participants to suppress/reduce the activation of semantically associated or related words and to attempt different retrieval strategies (Katzev et al., 2013; Luo et al., 2010). Contrary to these observations, studies comparing both tasks report SF to be more affected than LF (Depp et al., 2013; Dixon et al., 2004; Kravariti et al., 2005; Nehra et al., 2006; Radanovic et al., 2013; Raucher-Chene et al., 2017; Torrent et al., 2006; Weiner et al., 2019) or have reported a lack of deficit in LF (Cavanagh et al., 2002; Frangou, 2005). Overall, the outcome measures on both LF and SF tasks were reduced due to the influence of education with higher scores obtained for those with a higher education level. This finding is in consonance with previous literature (Mathuranath et al., 2003; Nogueira et al., 2016). The more the number of years of education, greater is the exposure to vocabulary thus enhancing VF performance (Nogueira et al., 2016). Even though age has also been reported as a factor that negatively affects VF performance (İlkmen & Büyükişcan, 2022; Lubrini et al., 2022), we could not analyze this factor in the current study due to limited sample size for age-wise categorization.

Concerning clustering measures, a reduction in the NC (more affected during LF) was noted in BD-I euthymic group as compared to HC. The findings on the clustering measures, although less explored among the euthymic BD population (Chang et al., 2011; Drakopoulos et al., 2020; Sung et al., 2013; Weiner et al., 2019), confirm the functional deficit hypothesis, that is, the deficit in activation or inhibition of semantic retrieval processes in BD population (Radanovic et al., 2013; Raucher-Chene et al., 2017). This prevents the participants from using the semantic store’s effective search and retrieval mechanisms implying that the semantic system pathways are affected. A minimal nonsignificant reduction in cluster size observed in our study is also consistent with Weiner et al. (2019), which provides evidence against the storage deficit hypothesis or the deterioration of semantic information. Unlike the present study, Sung et al. (2013) reported dissimilarity in the clustering pattern for domestic/wild clusters and low-frequency clusters.

The analysis of switching in the present study revealed a significantly reduced NS in the BD-I participants than in HC, with greater switch scores for SF than LF within the group. Along similar lines, Weiner et al. (2019) documented a reduction in the NS in euthymic and depressive states. The decreased switches have been attributed to a slow processing speed in euthymic participants similar to the depressive state (Fossati et al., 2003).

Although not previously explored, the present study explored the word productivity as the function of time. The analyses of temporal measures can provide insight into linguistic and executive control strategies (Luo et al., 2010). It has been perceived that word retrieval in the initial interval (0–15 sec) will be associated with semiautomatic retrieval which contrasts with word retrieval in the final intervals (30–60 sec) reflecting effortful retrieval of words (Crowe, 1998; Luo et al., 2010). In consonance with the observation of temporal pattern in typical HC, the words generated were not distributed uniformly over time but rather as a broad spurt of words in the initial interval and gradually decreasing in the successive intervals (Luo et al., 2010; Troyer et al., 1997) for both the BD and HC group.

The word production during VF was characterized by the presence of both correct and incorrect words. In this study, we have observed three types of errors (intrusions, perseverations, and nonwords) during word retrieval, with the perseveration type being more common. Contrary to the general notion, error responses were noted not only in the BD group but also among the HC. This may be attributed to the fact that HC produced a greater number of words in the LF task, whereas it was considerably reduced in BD-I. A greater number of errors in HC is also supported by a study by de Almeida Rocca et al. (2008), where they hypothesized that due to pressure to produce more words in the given time, the HC group disregarded the rules of the task. The lesser errors in the BD-I group could also be attributed to the slowness seen in euthymic BD-I participants (Fleck et al., 2003).

We accept that there are several limitations to our study. A high amount of variation seen in the educational qualifications of the participants might have affected the task performance and also the difference in data recording (face-to-face vs. telephonic) could have affected the quality of the recording. The influence of age on the VF tasks could not be accounted for due to the limited sample in each age group. The study findings also would have been influenced by the language, the letter chosen for the VF testing, the animal subcategories retrieved based on the geographical location, and culture (Oberg & Ramírez, 2006). Both HC and BD groups were only screened for cognitive status and no standardized tool was employed. Although the primary medications were taken into account, and drugs that likely influence cognitive function were ruled out, the influence of other associated medications taken by the study participants were not considered. Lithium is known to have cognitive side effects; however, since most participants were on lithium, we could not differentiate if the deficits were due to the drug or illness itself and whether those with lithium had higher effects or not. Lastly, the sample size was considerably reduced, although we attempted to control the homogeneity of the sample. Considering the aforementioned factors, the generalizability of the study results needs to be performed with caution. The findings of the study call for future research by including large sample size, with a more detailed analysis of temporal patterns (within-cluster and between-cluster time measures) in euthymic populations in comparison to other moods for other Indian languages. These could give more insight into the word retrieval deficits faced by this population in the Indian context.

The study documents the presence of VF deficits in euthymic BD-I participants, with higher impairment in the LF task. This observation along with the measures of clustering and switching complements the existing data on VF performance specifically in the BD-I population in the euthymic state. It also provides information on the error types and the word productivity as a function of time during the VF tasks in BD-I, which has not been explored in detail in other studies. The VF tasks give an insight into the underlying process of language perturbations and serve as a neuropsychological marker of cognitive deficits in the BD population (Raucher-Chene et al., 2017). VF task as a measure of an executive functioning task can aid as an early marker of occupational functioning deficits in the BD population (Drakopoulos et al., 2020). Early identification of such deficits can help in providing functional and cognitive remediation programs to enhance the functional outcomes in persons with BD (Bonnin et al., 2016; Deckersbach et al., 2010; Martínez-Arán et al., 2011; Sanchez-Moreno et al., 2017; Torrent et al., 2013). The findings call attention to assessing the VF performance in persons with BD-I, not just for total word output but also to include error count and strategies employed (clustering, switching). There is a need for more studies exploring the temporal pattern of VF production, which will aid in understanding the semantic organization during word fluency tasks and the cognitive-linguistic mechanisms. Multidimensional scaling analysis can also be carried out to further understand VF production and pattern. Studies clearly distinguishing between LF and SF and analyzing them according to mood states would help provide understanding of the pathophysiological processes underlying language disturbances in BD. Overall, this study confirms that VF performance is an important task in exploring cognitive-linguistic deficits in participants with BD-I.

Acknowledgments

We acknowledge all the participants who contributed to the research and those who helped in recruitment.

Disclosure statement

The authors report that there are no competing interests to declare.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.