Dissect the Heterogeneity of Longitudinal Cognitive Declines with Risk Factors and Functional Outcomes in Bipolar Disorder

Background: We characterized the heterogeneity and risk factors of cognitive decline in bipolar disorder (BD), and their magnitude of associations with subjective daily functions. Methods: Recruited BD patients (N=127) were followed for an average of 6.5 years. Baseline intelligence quotient was measured, and Brief Assessment of Cognition in Affective Disorders (BAC-A) was used to assess cognition at follow-up. We evaluated current functions with World Health Organization Disability Assessment Schedule 2.0. Clinical and sociodemographic factors were compared between BD patients with and without longitudinal cognitive decline and revealed the independent risk factors through multiple regression analysis. Correlations between specic cognitive domains and functional outcomes were examined in subgroups of BD. Results: Overall, 32.3% of BD patients showed longitudinal cognitive decline. In cognitive decline group, all cognitive domains of BAC-A were signicantly worsened. Baseline independent risk factors for cognitive decline were older age, body mass index >25, and manic episode density during the disease course. After the age of 40 years, a steeper cognitive change was observed in the cognitive decline group. Housekeeping dysfunction was the most sensitive function related to diverse cognitive impairment in cognitive decline group. The correlations between cognitive domains and functional outcomes differed between patients with and without cognitive decline. Conclusions: This study characterized in are important for cognitive we suggested further interventions should target on relapse and controlling body weight.


Introduction
Bipolar disorder (BD), a recurrent chronic disorder, is characterized by episodes of mania and depression interspaced by euthymia, and it affects different aspects of daily living (Goldberg et al. 1995). BD is also associated with cognitive de cits in a number of domains, including executive function, attention, and memory, which may persist in patients in remission . For example, Robinson et al. reported that euthymic patients with BD showed cognitive impairments compared with healthy controls, particularly in executive function and verbal learning (Robinson et al. 2006). A meta-analysis using individual patient data showed signi cant cognitive impairments in BD after controlling for age, educational years, gender, residual mood symptoms, and medications (Bourne et al. 2013  Previous cross-sectional studies have examined some clinical factors with cognitive de cits in BD. Although inconsistent, some studies have reported the correlations of cognitive impairments with illness severity. Number of manic episodes and number of total episodes were positively correlated with impairments in set-shifting and processing speed (Bourne et al. 2013). However, it is di cult to infer the causal link between cognitive de cits and clinical features due to the cross-sectional nature of these studies. One early study adopted a rst-episode design to compare rst-episode BD, schizophrenia, and healthy controls, and found that cognitive de cits are evident right from the rst mood episode (Bora 2015). During the disease course, cognitive impairment in BD patients then varies, with some studies suggesting that cognitive de cits stabilize over time (Mur et al. 2008, Gildengers et al. 2013, Samame et al. 2014), whereas others have shown a pattern of progressive deterioration (Moorhead et al. 2007, Gildengers et al. 2009) and even increased risk of dementia in later life (Diniz et al. 2017). It is believed that heterogeneity is widely observed among BD patients in this regard, in terms of the impaired cognitive domains, longitudinal cognition stability, and the speed of deteriorative progression. Longitudinal study design is preferable to explore cognitive declines over the course of bipolar illness, though it was relatively scarce in the literature.
So far, the heterogeneity of longitudinal cognitive declines in BD patients is not well understood. First, insu cient data were available to investigate all cognitive domains in few longitudinal studies, and ndings on de cits within different cognitive domains were inconsistent. Whether these are discrete areas of impairment or re ect an underlying single, more basic cognitive abnormality is as yet unclear. Second, previous longitudinal studies often consisted of small sample sizes and different follow-up periods, resulted in various ndings. For example, the latest synthesis of longitudinal evidence suggests that the cognitive path of individuals with BD may be set early on and may not deteriorate over time (Szmulewicz et al. 2020). However, another report with a much longer follow-up period showed signi cant test-retest differences in executive measures revealing decline (Torrent et al. 2012), which implied the cognitive instability over longer disease course may exist. Third, the differences in medication variables or mood scales between assessment times could not be well controlled in every study and may have in uenced the results. In addition, we should consider the effect of repeated testing, wherein the true decline of cognitive functioning would be partially masked by learning effects.
In the present study, we considered that both the clinical course and cognitive decline may be heterogeneous among BD patients, and thus may partially account for the inconsistency of ndings in previous reviews. The majority of cross-sectional studies have suggested that 38-40% of patients with BD have no neurocognitive de cits, and 20-30% had obvious cognitive de cits (Martino et al. 2008, Bora et al. 2016, Volkert et al. 2016). The lack of longitudinal assessment of cognitive performance in BD hinders to explore the heterogeneity in cognitive de cits. Therefore, we aimed to evaluate the cognitive changes in BD patients over time by looking at age hinge points of cognitive declines, to characterize heterogeneous cognitive pro les among BD patients, and to examine the risk factors for cognitive decline, as whether different cognitive changes impact daily functions in patients with and without cognitive declines differently.

Participants
Participants who were diagnosed with BD according to Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, Text Revision (DSM-IV-TR) and referred by board-certi ed psychiatrists were enrolled from Taipei City Hospital, Songde branch. All participants drawn from who had a history of prior baseline intelligence quotient (IQ) testing during euthymic state in the past medical record. In addition, they were more than 20 years of age, and those without: a) known substance use disorder (except nicotine use disorder); b) any disorder with known neurological symptoms or complications such as brain injury or stroke; c) a diagnosis of intellectual disabilities; or d) inability to complete the standard clinical assessment or providing informed consent. Individuals were also required to be euthymic at the time of study entry for cognitive assessment. Mood symptoms were obtained through clinician-administered measures of the 17-item Hamilton Rating Scale for Depression (HRSD-17) and Young Mania Rating Scale (YMRS). For euthymic state, we de ned HRSD and YMRS scores ≤ 8 within 7 days before assessment. Measurements 1. Demographic data, clinical course, and baseline IQ Patients' demographic and clinical course data were collected from medical records and interviews by psychiatrists, if required. The clinical characteristics of participants included the number of affective episodes, number of episodes with psychotic features, number of hospitalizations, age at illness onset, maximum length of free intervals, and number of suicide attempts. Considering the duration of illness varied across individuals, we also calculated episode density by dividing number of total episodes and manic episodes by duration of illness as the total episode density and manic episode density, respectively. Through chart review, we recorded their physical comorbidities, anthropometry, and baseline IQ. For this recorded baseline IQ, it was measured by a licensed psychologist using the Wechsler Adult Intelligence Scale (WAIS-III or WAIS-IV) in BD euthymic state for the clinical practice as the attending physicians start beginning for follow up the patients. Psychopharmacological treatments used at the time of study assessment were also transformed to de ned daily dose (DDD) and then recorded.

Cognitive measurements
Patients were assessed using the Brief Assessment of Cognition in Affective Disorders (BAC-A) during their euthymic state for at least 1-year interval compared with their baseline IQ measurements to ensure the longitudinal change from the possible learning effects. BAC-A has extensively been used as a rapid and reliable measure of cognitive assessment in a range of clinically affective patients (Cholet et al. 2014). It provides not only measure of affective memory and emotional inhibition (Affective Processing Tests [APTs]), but also includes six traditional neurocognitive domains, namely, working memory, motor speed, verbal uency, attention and processing speed, verbal memory, and executive function (Bauer et al. 2015). The criterion and construct validity of each test for cognitive impairment as well as the sensitivity of these tests to changes in cognition have been demonstrated in the scienti c literature, and each test has also been shown to be valid for use in different cultures and language groups (Lee et al. 2018).

Statistical analysis
We de ned cognitive heterogeneity in BD longitudinally and dichotomously. Patients who ful lled one of the following criteria were included in the cognitive decline group: First, if their baseline IQ scores were less than the estimated norm based on their previous educational and occupational history assessed by a licensed psychologist. Second, if their baseline IQ showed a difference of more than 2 standard deviation (SD) compared with the current composite z-score of the BAC-A. Those participants without above conditions will regard as group without cognitive decline.
We compared demographic and clinical characteristics between BD patients with and without cognitive decline using the chi-square and Student's t-test for categorical and continuous variables, respectively. The differences in BAC-A and functional outcomes were also examined. To assess independent risk factors, potential confounding variables were included in multivariable logistic regression analysis. Spearman's correlations between speci c cognitive and functional domains were demonstrated in each subgroup. Moreover, multivariate adaptive regression spline (MARS) (Vanegas et al. 2017) and scatterplot with Loess curve were used to detect hinges or in ection points to characterize the timing and nature of slope changes in the BAC-A composite scores of separate cognitive groups. MARS analysis was conducted using the earth package in R (Milborrow 2011); all other analyses were conducted using SAS (version 9.4; SAS Institute Inc.; Cary, NC, USA). Signi cance was set at p < 0.05.

Patient characteristics
Of the total 127 BD patients, 67.7% (N = 86) and 32.3% (N = 41) were without and with longitudinal cognitive decline, respectively ( Table 1). The mean period from baseline IQ to current BAC-A assessment in BD patients was 6.5 years, which was no difference between the two subgroups. Compared with patients without cognitive decline, those with cognitive decline experienced more mood episodes, especially manic episodes, more number of admissions, and higher hospitalization duration, and they showed higher baseline body mass index (BMI), as assessed at their index admission (the earliest record from the medical chart system). No signi cant differences were observed in variables such as physical comorbidities, duration of illness, age of onset, or baseline IQ measurement between the two groups. In addition, there was also no difference in the two groups for their current DDD of psychoactive agents used, which implied that medication was not the cofounding factor in our further analysis. No signi cant differences were observed in current functional outcomes assessed using WHODAS 2.0 between BD patients with or without cognitive decline. Table 2 shows the IRT scoring of the groups; a higher score indicated more severity of subjective impairment, that is, greater disability. We noted that common functional impairments in the chronic phase of BD were observed in both groups, regardless of presence of cognitive decline. In patients with cognitive decline, the functional domain with the highest score was job and learning (IRT score 86.67), implying that it was the most severely impaired domain compared with the norm (around last 1% in the population percentile) (Ustun et al. 2010). Even for the most preserved functional domain (self-care), patients showed impairment that was the last 30% in the population percentile when compared with norm data (Ustun et al. 2010).   We used univariate and multivariable logistic regression to analyze the independent risk factors for cognitive decline in BD, and results are displayed in Table 4. In multivariable logistic regression models, we selected manic episode density into the model and left out the total episode density to avoid the multicollinearity problem (correlation coe cient between total and manic episode density was high,   Correlation between cognitive and functional domains Figure 2 demonstrated correlations between cognitive and functional domains in patients with and without cognitive decline. The cognitive decline group (Fig. 2a) showed strong links between various cognitive domains and daily functions (r ranging from 0.47 to 0.63). For example, the range of impaired functions, such as subjective cognition, walking, getting along with other people, housekeeping, and total function, were negatively correlated with attention and processing speed. Executive function domain also had negative correlation with subjective cognition, walking, self-care, housekeeping and total function impairment in this subgroup. Housekeeping dysfunction, which is another aspect, was the most sensitive function related to diverse cognitive domains including composite BAC-A score, verbal memory, attention and processing speed, executive function, and emotional inhibition. In addition, walk dysfunction was correlated with motor speed, attention and processing speed and executive function; subjective cognition dysfunction was also correlated with BAC-A composite score, attention and processing speed, executive function and emotional inhibition. After adjusting for the effect of age to obtain partial correlations, signi cant correlations were still noted between subjective cognition and BAC-A composite score, subjective walking function and motor speed, subjective housekeeping function and measurements of verbal memory, processing speed, and executive function (not shown in Fig. 2). In the subgroup without cognitive decline (Fig. 2b), no correlation was found between functional and cognitive domains. Thus, the different correlation patterns of the two subgroups revealed heterogeneous effects on daily function by the longitudinal cognitive pro le.

Discussion
To the best of our knowledge, this longitudinal study is the rst to provide a comprehensive picture of BD patients with cognitive decline through long-term follow-up, and the results suggested the existence of subgroups with different cognitive trajectories. We classi ed our patients into two groups according to presence or absence of cognitive decline and correlated cognitive decline to their daily dysfunctions. We revealed that old age, baseline BMI > 25, and more number of manic episodes are risk factors for cognitive decline. Furthermore, we demonstrated a hinge at the age of 40 years with steeper decline of cognition in the cognitive decline group.
In our study, the picture of BD patients with longitudinal cognitive decline shows substantial degree of impairment, rather than on only some speci c cognitive domains, which is consistent with most previous ndings for clinical characteristics and risk factor identi cation in the literature. In our study, the longitudinal cognitive decline group comprised nearly one-third of our sample, and this group may be established as a speci c subtype of BD. Such longitudinal cognitive pro les can potentially facilitate further genetic and biological studies and help clinicians develop more effective intervention strategies (Solé et al. 2017). As we had suggested, the heterogeneity of the cognitive trajectory conformed to the progress of illness in clinical manifestation staging models including cognitive deterioration (de la Fuente-Tomas et al. 2020). Our ndings revealed that cognitive decline varies from one patient to another, with decline in certain patients while others remain stable. Conversely, we may say that some patients maintained relatively stable cognitive function might be cognitive reserve, which re ects the partial capacity of the brain to endure neuropathology and minimize clinical cognitive de cits (Kohler et al.  (Peterman et al. 2020). Although our nding implied the pathway from being overweight or obese to cognitive decline in patients with BD, very little is known about why some patients with BD develop signi cant cognitive decline while others remain cognitively intact. We recommend that interventions targeting the risk factors of manic episode and high BMI would be effective in preventing cognitive decline in patients with BD and for detecting the underlying mechanism. Moreover, the subgroup of BD patients with cognitive decline showed steeper changes after the age of 40 years compared with those without cognitive decline, which is consistent with decline associated with accelerated aging in executive functions in BD (Seelye et al. 2019). We suggested that clinicians should pay attention to patients' cognitive function before they reach middle age. Future research should focus on early intervention to determine whether it can delay the decline of cognition in BD.
Psychosocial function is a person's ability to perform activities of daily living and to be involved in meaningful interpersonal relationships. As per our results, patients with BD have di culties in several areas of function during their remitted status. These results are consistent with those of other studies demonstrating that most patients with BD have functional di culties (Rosa et al. 2009, Samalin et al. 2016, Comes et al. 2017). However, there was no difference of functional outcome in patients with and without cognitive decline assessed by WHODAS 2.0. There may be possible oor effect of this subjective assessment for this population. We used the WHODAS 2.0 for evaluating our patients, rather than disease-speci c questionnaires, such as Functioning Assessment Short Test (FAST), because the former disability assessment is more informative by comparing with general norm or other diseases. Results from the partial correlation analysis between cognitive and functional domains, controlling for age, indicated that the cognitive pro le was correlated with some functional domains mainly in the subgroup of BD patients with cognitive decline. These results are in line with those of cross-sectional studies showing that the cognition-function relationship may be weaker among patients without cognitive de cits than among those with cognitive impairment (Moore et al. 2015). The distinct correlation results among patients with BD also responded to those of a previous study conducted by Sole  Several limitations should be considered of this study. First, our study sample was recruited from a tertiary psychiatry hospital where patients have more severe degree of illnesses. This is a potentially biased sample and may limit the generalizability of our ndings to the whole BD population, especially the proportion with cognitive decline. Second, without a healthy control group, cognitive decline in patients should be viewed as evidence of relative cognitive decline. Furthermore, the restricted sample size limited the power to examine interaction effects among potential variables, such as interaction of disease course and medications. Third, WHODAS 2.0 is an interviewer-administered tool assessing subjective disability; a combination of objective measures may provide a more comprehensive outcome picture for BD patients with and without cognitive decline. Four, underlying mechanism between risk factors, cognitive decline, and functioning stilled need further investigation. Finally, the dichotomous categorized de nition about cognitive decline in our study is arbitrary. However, the potential of misclassi cation is non-differential and bias the odds ratio toward the null, which suggested the robustness of our ndings.

Conclusion
Our results specify and characterize cognitive heterogeneity in BD longitudinally, which may facilitate further genetic and biological studies to de ne more valid BD subtypes. We revealed risk factors for cognitive decline and suggested aggressive treatment to prevent mania relapse and control body weight. The heterogeneity of cognitive decline in BD should be considered, thus more individualized intervention for patients with BD could be investigated in future research.

Declarations
Ethics approval and consent to participate: All the participants have informed consent. The Research Ethics Committee of Taipei City Hospital approved our study (TCHIRB-1-703103).

Consent for publication:
Not applicable Availability of data and material: Restrictions apply to the availability of the data that support the ndings of this study, which were guarded by IRB regulations for data access in the recruitment sites. Part of the anonymous data maybe available upon request from the corresponding author (PH. Kuo).
PHK is major for supervision, methodology and reviewing the manuscript. All authors read and approved the nal manuscript.