The Big Five as Predictors of Cognitive Function in Individuals with Bipolar Disorder

The connection between cognitive function and the “Big Five” personality traits (openness, conscientiousness, extraversion, agreeableness, and neuroticism) in the general population is well known; however, studies researching bipolar disorder (BD) are scarce. Therefore, this study aimed to investigate the Big Five as predictors of executive function, verbal memory, attention, and processing speed in euthymic individuals with BD (cross-sectional: n = 129, including time point t1; longitudinal: n = 35, including t1 and t2). Participants completed the NEO Five-Factor Inventory, the Color and Word Interference Test, the Trail Making Test, the d2 Test of Attention Revised, and the California Verbal Learning Test. The results showed a significant negative correlation between executive function and neuroticism at t1. Changes in cognitive function between t1 and t2 did not correlate with and could not be predicted by the Big Five at t1. Additionally, worse executive function at t2 was predicted by higher neuroticism and lower conscientiousness at t1, and high neuroticism was a predictor of worse verbal memory at t2. The Big Five might not strongly impact cognitive function over short periods; however, they are significant predictors of cognitive function. Future studies should include a higher number of participants and more time in between points of measurement.


Introduction
Bipolar disorder (BD) is a severe mental disorder affecting 1-2% of the population worldwide. Characterized by depressive and (hypo-)manic episodes, the disorder typically manifests in young adulthood during the emergence of a stable structure of personality [1]. Individuals with BD often suffer from social problems, such as lower social skills [2] and lower perceived social support, than healthy controls (HC) [3], and experience stigma [4]. In addition, cognitive impairment is frequently found in individuals with BD [5][6][7], even before illness onset [8] and during times of euthymia [9].
Cognitive dysfunction has been found to worsen in concert with illness duration [10] as well as neuroprogression [11], although results suggesting no apparent link with the latter show the questionability of this issue [12]. The primarily affected domains of cognitive impairment are verbal memory [13,14], executive function [14], attention [14,15], and Concerning individuals with BD, two studies found high O to be a predictor of better cognitive function [38,51]. In particular, O's association with ideas and values was the most important facet, correlating with several cognitive factor scores, such as auditory memory, emotional processing, verbal fluency, and processing speed [51]. Furthermore, a higher number of single-nucleotide polymorphisms of the brain-expressed protocadherin 17 gene showed correlations with both impaired cognition and higher N. Increased gene expression was found in individuals with BD compared to HCs [52]. Another study suggested a negative correlation between N and reaction time in the Affective Go/No-Go paradigm with a bias towards affective stimuli, which was not significant in HC and might suggest a greater receptivity to emotional stimuli in individuals with BD [53].
As there is currently a lack of studies investigating personality and cognitive function in BD, the aims of this study were 1. to investigate cross-sectional correlations between the Big Five as well as executive function, verbal memory, attention, and processing speed, and 2. to predict cognitive decline according to the Big Five in a longitudinal sample. It was hypothesized that high O, C, and E, as well as low N, would correlate with and predict better cognitive function, while analyses concerning A would be explorative.

Participants and Procedure
All participants were participants in the ongoing BIPFAT/BIPLONG study, which has aimed to investigate BD in a longitudinal setting since 2012. Conducted at the outpatient center for BD at the Medical University of Graz, Austria, Department of Psychiatry and Psychotherapeutic Medicine, the study's focal points are lifestyle, lipid metabolism, inflammation processes, cognition, and brain function. Inclusion criteria were a BD diagnosis by trained specialists using the Structured Clinical Interview for DSM-IV [54], age between 18 and 70 years, and IQ of ≥80 at the time of measurement. Euthymia was defined in this study by a score of ≤12 on the Young Mania Rating Scale (YMRS) [55] and a score of ≤10 on the Hamilton Depression Scale (HAMD) [56]. Individuals were excluded if they suffered from severe immunological disorders, organic brain diseases, or dementia.
As cross-sectional and longitudinal comparisons were made, the study included two samples, with some participants being included in both. The cross-sectional sample, consisting of 129 individuals with BD, included data from the participants' first visit (t1). The longitudinal sample of 35 individuals with BD additionally included assessments from a second visit yielding complete datasets (t2), taking place 345 days to 2106 days after the first visit. This study was conducted in accordance with the Declaration of Helsinki as well as approved by the Ethics Committee of the Medical University of Graz (EK-number: 25-335 ex 12/13), and all patients signed informed consent forms before participation.

Measurements
Participants completed a cognitive assessment as well as a personality questionnaire, all of which were administered in German. In addition, clinical and sociodemographic data were assessed by interview.

Neuropsychological Assessment
Verbal memory was measured with the California Verbal Learning Test (CVLT) [57]. In particular, recall trials 1-5, short delay free recall, short delay cued recall, long delay free recall, and long delay cued recall were measured.
To assess attention and processing speed, three tests were used: the Trail Making Test part A (TMT-A) [58], the d2 Test of Attention Revised (d2-R) [59], and the word-reading and color-naming trials from the Color and Word Interference Test by J. R. Stroop [60].
Cognitive flexibility, one aspect of executive function, was assessed by the Trail Making Test part B (TMT-B) [58], as well as the interference trial from the Color and Word Interference Test by J. R. Stroop [60].

Personality Assessment
The NEO Five-Factor Inventory (NEO-FFI) [61] was used to assess the five personality dimensions of the FFM. Participants were asked to rate 60 questions on a five-point Likerttype scale, with 1 = strongly disagree and 5 = strongly agree. For each factor, the sum score of the 15 corresponding questions is calculated, ranging from 15 to 75, with a higher score indicating a stronger expression of the respective factor. The NEO-FFI shows good internal consistency, with Cronbach's alpha ranging between .63 and .82 for each factor [62].

Statistical Methods
For the cross-sectional sample, cognitive test scores were converted into z-scores and then summed up into three scores measuring three cognitive domains: (1) attention and processing speed (d2-R, Stroop color naming, Stroop word reading, TMT-A); (2) verbal learning and memory (CVLT trial 1-5, CVLT short delay free recall, CVLT long delay free recall, CVLT short delay cued recall, CVLT long delay cued recall); and (3) executive function (Stoop interference and TMT-B). Measures expressing reaction times were reversed before calculating the domain scores, because in contrast to the other scores, lower scores indicated higher performance. A higher domain score indicated higher performance.
For the longitudinal sample, the same procedure was repeated; however, not all cognitive test scores could be used for the creation of the domain scores, as they were not assessed at t2. For memory, only the score for recall trials 1-5 was included in the sum score. The sum score of attention and processing speed did not include the d2-R test, which was accounted for when calculating the sum score and comparing both samples. There were no differences in the calculation of executive function.
Mean Big Five values of the cross-sectional sample were compared to the norm sample using summary t-tests. To compare the t1 and t2 data of the longitudinal sample, t-tests were used for age and the HAMD, and a Wilcoxon Test was used for YMRS due to outliers. A repeated-measures multivariate analysis of variance (MANCOVA) was employed to test for differences in cognition between both time points. Only the variables assessed at both time points were included: TMT-A, TMT-B, recall trials 1-5 of the CVLT, and the Stroop test. Covariates included age, sex, education, time difference between t1 and t2, BDI, and illness duration. Key assumptions of the repeated-measures MANCOVAs (linearity and normality) were tested graphically, as well as with the Kolmogorov-Smirnov test.
Cross-sectional partial correlation analyses between each of the Big Five and verbal memory, executive function, attention, and processing speed were performed at time t1. Age, sex, education, BDI, and illness duration were used as covariates. A false discovery rate (FDR) was used to correct for alpha error cumulation. Furthermore, the same analyses were performed for the purpose of correlating each of the Big Five assessed at t1 with cognition at t1 and t2, as well as the difference between the three cognitive functions measured at t2 as compared to t1. In addition to the previously mentioned covariates, the time difference between t1 and t2 was used for the latter analysis.
Three multiple hierarchical regression analyses to predict executive function, verbal memory, attention, and processing speed at t2 were conducted. The first step included the variables of age, sex, education, time difference between t1 and t2, BDI, and illness duration, while the second step included the Big Five at t1. Three more hierarchical regression analyses to predict changes in cognitive function were calculated with the same variables. Conditions for multiple regression analyses were tested by using correlations (linearity), scatterplots (homoscedasticity), histograms (normal distribution of error variance), Durbin-Watson tests (lack of autocorrelations), and both variance inflation factor and tolerance (lack of multicollinearity).
The current study included participants who completed all relevant questionnaires. In sum, 12 individuals had to be excluded in the cross-sectional sample and 5 in the longitudinal sample due to missing data.

Sample Characteristics
Sociodemographic information of the cross-sectional sample (n = 129) is displayed in Table 2, and of the longitudinal sample (n = 35) in Table 3. The Big Five scores of the cross-sectional sample were compared to the norm sample (n = 871) [61] with t-tests, and it was found that N was higher (M = 20.99, SD = 7. 89  T-tests showed that both groups of the longitudinal sample differed in age, but not HAMD (see Table 3). A Wilcoxon test further resulted in no differences between the groups regarding YMRS. A repeated-measures MANCOVA with the covariates of age, sex, education, BDI, time difference between t1 and t2, and illness duration regarding cognition was not significant (F(6,19) = 0.68, p = 0.666).

Partial Correlation Analyses
Partial correlation analyses including the three parameters of cognitive function; the Big Five; and the covariates of age, sex, education, BDI, illness duration, and time difference between t1 and t2 for the longitudinal sample are shown for both samples in Tables 4 and 5, respectively. In the cross-sectional sample, only executive function correlated negatively with N after correction with FDR. In contrast, the positive correlation between attention, processing speed, and O did not remain significant. In the longitudinal sample, partial correlation analyses with the same variables were conducted for both time points t1 and t2, and for the cognitive function differences between t1 and t2. N at t1 correlated significantly with memory (r = −0.52, p = 0.007) and executive function at t2 (r = −0.43, p = 0.034); however, neither correlation remained significant after the usage of FDR. Correlations between cognitive function differences and the Big Five did not yield any significant results.

Executive Function
A multiple hierarchical regression analysis was performed to predict executive function at t2 (see Table 6). The variables of age, sex, education, time difference between t1 and t2, BDI, and illness duration entered in the first step did not yield a significant result (R 2 = 0.38, R 2 corr. = 0.23, F(6,24) = 2.46, p = 0.053). The second step, comprising the Big Five at t1, was significant (R 2 = 0.65, R 2 corr. = 0.45, F(11,19) = 3.20, p = 0.013). The results were significant for the predictors of education, time difference between t1 and t2, C, and N. A second multiple hierarchical regression analysis predicting the change in executive function between t1 and t2, with the variables of age, sex, education, time difference between t1 and t2, BDI, and illness duration entered in the first step and the Big Five at t2 entered in the second step, was not significant (Model 1: R 2 = 0.26, R 2 corr. = 0.07, F(6,24) = 1.39, p = 0.260; Model 2: R 2 = 0.37, R 2 corr. = 0.01, F(11,19) = 1.03, p = 0.463).

Attention and Processing Speed
A multiple hierarchical regression analysis predicting attention and processing speed at t2 with the variables of age, sex, education, time difference between t1 and t2, BDI, and illness duration as predictors in the first step was significant (R 2 = 0.52, R 2 corr. = 0.40, F(6,24) = 4.26, p = 0.005). The Big Five at t1 were included in the second step, which yielded significant results (R 2 = 0.60, R 2 corr. = 0.43, F(11,19) = 2.84, p = 0.022). Age was the sole significant predictor in the second step (see Table 8).
Another multiple hierarchical regression analysis, with the predictors of attention and processing speed difference between t1 and t2, was conducted. The first step included age, sex, education, time difference between t1 and t2, BDI, and illness duration,

Discussion
A cross-sectional (n = 129) and a longitudinal (n = 35) subsample of euthymic individuals with BD were investigated concerning the association between cognitive function and the Big Five. A significant negative correlation between executive function and neuroticism at t1 was found. Partial correlations and regression analyses predicting changes in cognitive function between t1 and t2 with the Big Five did not yield any significant results. High neuroticism at t1 was a significant predictor of worse executive function and verbal memory at t2, while high conscientiousness at t1 significantly predicted worse executive function at t2. N was higher, and E, as well as C, were lower in the cross-sectional BD sample than in the norm sample [61], which is consistent with other studies [29,33]. Similarly, one study found this FFI triad with the opposite configuration to be a strong correlate of positive mental health [57], showing individuals with BD to be naturally vulnerable. Findings suggesting higher O [30,38,64] and lower A in BD [30] could not be replicated, and should be investigated in more detail.
Interestingly, variables of the Big Five at t1 predicted executive function and verbal memory at t2, but not the change in cognitive function between t1 and t2 in the longitudinal sample. It should be mentioned that mean cognitive test scores did not differ between t1 and t2. This lack of significant results might indicate that while certain aspects of personality are important for cognition, the influence exerted is only observable over a longer period of time, or that it is stronger during a different stage of development. Gale et al. [65] proposed that an association between personality and cognitive function found in mid-life might be a reflection of a correlation between these traits in childhood. It should be further considered that any meaningful associations between the Big Five and cognition might have been underestimated as a consequence of the small sample sizes.
Only the negative correlation between N and executive function remaining significant after FDR in the cross-sectional sample. N being a significant predictor of executive function and verbal memory in the longitudinal sample cemented N's standing as the most influential Big Five variable on cognitive function. This connection between N and cognition was found in studies examining HC [39][40][41], and seems to be applicable to individuals with BD as well. A study by Chang et al. [52] yielded the same result. Similarly to HC, higher N might contribute to a higher frequency of unhealthy behaviors impacting long-term cognition among individuals with BD [44,45], which they are known to exhibit more than HC: they exercise less [66] and are more likely to smoke [67] and have a less healthy diet [68]. This is supported by a positive correlation between N and cardiovascular disease [32].
Executive function and verbal memory seem to be affected the most by the aforementioned correlates of N. Regarding the former, healthy older adults showed a negative association between N and executive function [69,70]. Fittingly, our previous research suggested a higher prevalence of metabolic syndrome in individuals with BD, which was found to be a risk factor for impairment of executive function [71]. In addition, damage to the left dorsolateral prefrontal cortex, which is important for executive function, was found to be associated with high N [72]. Verbal memory was predicted by N as well. Other studies found comparable results, suggesting a link between high N and worse memory recall in HC [65,73], as well as memory complaints [74] and greater decreases in verbal memory in older adults [75]. These results underline the importance of preventing cognitive decline in individuals with BD. For example, mindfulness training was found to reduce N [76], and a unified protocol designed specifically for decreasing N was successful [77]. Additionally, low C might contribute to cognitive decline [46], as affected individuals display less behavior contributing to the preservation of cognitive function [49].
C was an important predictor of executive function as well. Similarly, Sutin et al. [50] found better executive function to be predicted by higher C. High C has been linked to high achievement by students [78], better cognitive performance, greater pursuit of cognitive activities [79], and less cognitive decline [46]. As C was lower in the BD group compared to the norm sample, both C and executive function are important points to consider in therapy. In contrast to our result, other findings showed no association between C and cognitive function in HC [40] or individuals with BD [51]. Apart from the dissimilar study populations, these diverse results might be explained by different measures of executive function, which is a broad concept spanning several processes of cognition. For instance, Crow [40] used response to inhibition and sustained attention to represent executive function, while the current study used measures of cognitive flexibility, the TMT-B, and the interference trial of the Stroop test. In conclusion, the exact relationship between C and different categories of executive function in BD is still unclear.
Attention and processing speed were solely predicted by age at t1. As the tests assessing these cognitive domains were mainly dependent on speed, it is not surprising that none of the Big Five were significant predictors. The progression of time, not personality, was the most impactful factor affecting the decline in attention and processing speed, which are known to decrease with age [80]. A possible neuronal correlate is fractional anisotropy of white matter [81]. Zhao et al. [82] have recently suggested that the deterioration of the glial structure disturbs the ratio between neural activity and the availability of oxygen.
Except for a significant correlation between O, attention, and processing speed prior to FDR, O unexpectedly did not correlate with or predict any variables related to cognition. High O has been consistently linked to higher cognitive abilities in HC [42,46] and has been found in individuals with BD as well [38], although results are scarce, and this connection in individuals with BD remains to be elucidated.
Finally, education was a significant predictor of executive function, but not verbal memory nor attention and processing speed at t2. Education has been identified as an important factor for the development of executive function in early childhood [83]; in turn, executive function is integral for pre-school learning [84]. Moreover, the quality of education during adulthood impacts executive function as well [85]. Many studies have investigated this reciprocal relationship in children, while older adults have been in the spotlight of a few studies showing the same results: a higher level of education is associated with better executive function [86,87], which seems to be true for middle-aged individuals with BD as well. Considering that executive function includes working memory, inhibition, and attention shifting [88], it is evident that these skills facilitate educational achievement, while education might help to hone them. This is supported by the fact that high consci-entiousness is also a significant predictor of executive function, a trait that is linked to high achievement [78]. This is particularly relevant for individuals with BD, as those with depression have a lower education level than the general population [89]. Considering verbal memory, other studies have found a significant association with education [90,91]. As for attention and processing speed, education does not seem to hold as much importance as age, which we discussed previously.
The first limitation of the current study was the small sample size, especially in the longitudinal subsample, which might have led to an underestimation of the relationship between cognition and the Big Five. In this context, possible attrition bias due to high dropout rate might have occured, although there were no significant differences between t1 and t2 except age. Secondly, there was no control group with which to compare individuals with BD. The influence pattern of personality and cognition might be different in HC, and it would have been essential to compare both groups. Thirdly, not all cognitive tests were administered at t2 for the longitudinal sample, leading to mean scores of the three cognitive domains that were not as differentiated as the scores at t1. Fourthly, information on medication intake was assessed, but could not be included in the statistical analyses due to missing data. As cognitive performance might have been inhibited by the effect of certain medication, it would have been important to include this variable as a covariate. Fifthly, the time span between t1 and t2 was heterogenous in the cross-sectional sample, although there were no outliers, and might have influenced the results. Finally, subscores of the Big Five were not administered and might provide a more differentiated picture of the relationship between cognition and personality. Related to this are different measures of executive function, which might lead to differing results and should be examined in more detail.

Conclusions
Individuals with BD have lower N, E, and C than HC. High N has a strong association with worse executive function and verbal memory, while C is positively correlated with executive function. While the Big Five might not strongly impact cognitive function over the span of several years, they are, nevertheless, significant predictors which might take effect over a longer period of time. Future longitudinal studies should include a higher number of participants as well as more time in between points of measurement.  Informed Consent Statement: Written informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available upon request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Conflicts of Interest:
The authors declare no conflict of interest.