Lakshmi N. Yatham MBBS, FRCPC, MRCPsych (UK), Professor of Psychiatry, University of British Columbia, 2255 Wesbrook Mall, Vancouver, BC, Canada V6T 2A1. E-mail: email@example.com
Objective: Although cognitive deficits are prominent in symptomatic patients with bipolar disorder, the extent and pattern of cognitive impairment in euthymic patients remain uncertain.
Method: Neuropsychological studies comparing euthymic bipolar patients and healthy controls were evaluated. Across studies, effect sizes reflecting patient–control differences in task performance were computed for the 15 most frequently studied cognitive measures in the literature.
Results: Across the broad cognitive domains of attention/processing speed, episodic memory, and executive functioning, medium-to-large performance effect size differences were consistently observed between patients and controls, favoring the latter. Deficits were not observed on measures of vocabulary and premorbid IQ.
Conclusion: Meta-analytic findings provide evidence of a trait-related neuropsychological deficit in bipolar disorder involving attention/processing speed, memory, and executive function. Findings are discussed with regard to potential moderators, etiologic considerations, limitations, and future directions in neuropsychological research on bipolar disorder.
• Euthymic patients with bipolar disorder demonstrate cognitive impairments, suggesting a trait-related neuropsychological impairment.
• Medium-to-large effect size differences were noted across the general domains of attention/processing speed, episodic memory, and executive functioning.
• Cognitive deficits in patients with bipolar disorder cannot be fully attributed to symptoms of depression and mania.
• At this point, it is unclear whether the observed cognitive deficits reflect a single core impairment or multiple cognitive deficits.
• Only a modest percentage of studies report data on potential moderators of the cognitive impairment in euthymic patients such as medication status and residual symptoms. Future studies should report these data so that the influences of these moderators may be analyzed further.
• Although the present results support the presence of a trait-related cognitive impairment in bipolar disorder, these findings do not clarify whether cognitive deficits reflect genetic/neurodevelopmental abnormalities or progressive disease-related deficits.
It is well established that patients with bipolar disorder in acute manic-depressive states show considerable cognitive impairment (1–3). In recent years, however, there has been a strong shift towards studying neurocognitive functions in euthymic patients with bipolar disorder. These studies have been conducted to investigate whether cognitive deficits observed in bipolar patients are stable and relatively invariant to mood changes, and thus reflect trait features of the illness (1, 4).
One of the most frequently studied domains of cognitive functioning in bipolar disorder is the area of executive functioning, and many studies report a range of executive deficits in euthymic patients (5–11). Findings, however, are not universal, as some studies report the absence of (12–14) or marginal impairment in this domain (15, 16). Discrepant findings are probably accounted for by the fact that individual studies vary according to methodological parameters. In particular, there are numerous conceptualizations of the construct of executive functioning (17), and adoption of different definitions by investigators inevitably leads to the use of different cognitive tasks across studies. This variability potentially obscures a cohesive understanding of the nature of executive dysfunction in bipolar disorder.
Memory constitutes another cognitive domain that has been frequently studied in euthymic bipolar patients. Relative to the executive domain, there appears to be more uniformity in that studies typically report verbal and non-verbal episodic memory deficits relative to healthy controls (9, 10, 14, 16, 18–22). Despite this, there is little sense of the overall magnitude of memory deficit across the range of studies in euthymic patients.
A third major area that has been studied in euthymic patients is the broad domain of attention and mental processing speed. Findings in this research area indicate that euthymic patients commonly exhibit deficits on various measures derived from a range of sustained attention tasks (13, 20, 21, 23–26); however, these impairments are not universal (11, 27). As with the examination of executive deficits, it is possible that discrepant findings are at least partly due to different tests or measures that are employed across studies.
To summarize, although emerging data suggest that there is a trait neurocognitive deficit in bipolar disorder, and that impairments are observed in the broad domains of executive function, memory, and attention/processing speed, there is little understanding regarding the consistency and magnitude of these deficits across and within these domains. Thus, an important starting point to help delineate the nature of cognitive impairment in euthymic patients is to evaluate the magnitude of neurocognitive deficit according to the specific neurocognitive tasks that have been employed in studies of euthymic patients with bipolar disorder.
Aims of the study
The major objective of this study was to evaluate the magnitude of patient–control differences in performance for each of the most-studied cognitive tasks in this literature through the use of meta-analytic techniques. This analysis provides insight into the presence, magnitude, and pattern of the trait-related neuropsychological impairment in bipolar disorder.
Material and methods
Medline and PsychInfo literature searches were conducted from January 1980 to July 1, 2005. The initial search strategy aimed to identify all cognitive studies that compared a sample of euthymic bipolar patients with a comparison group of healthy controls. The following keywords were included in the search: (bipolar or mania or manic) and (neuropsycholog* or cogniti* or intelligence or intellectual or IQ or memory or executive or visual or spatial or attention* or language) and (trait or remitted or remission or euthymi* or asymptomatic or stable or stability or baseline). Studies were also searched in reference sections of both existing narrative reviews on cognitive function in bipolar disorder, and any studies that were identified for inclusion in the meta-analysis. From an initial pool of over 400 studies, studies were retained that: (i) evaluated a sample of adult (mean age > 18 years) euthymic patients diagnosed with bipolar disorder using DSM-IV or similar criteria (ICD; RDC); (ii) included a comparison group of healthy subjects; and (iii) included at least one cognitive measure that was studied in at least eight total studies. Studies of euthymic patients were defined as those that described patients as euthymic, remitted, or asymptomatic on the basis of concurrent depression/mania ratings or by clinical judgment based on familiarity with the patient. The criterion for inclusion of a given cognitive measure into the meta-analysis was that the task had to be used in at least eight studies. This eight study criterion was employed to assure that only the most frequently studied tasks were included, thus increasing the likely reliability of effect size estimates across studies. The Medline and PsychInfo searches were conducted independently by two authors (IJT and VGB) and questions about appropriate inclusion or exclusion were resolved between them.
Extraction of effect sizes
Effect sizes were computed for all tasks in each study using comprehensive meta-analysis version II software (28). In infrequent instances where sample size, mean, and standard deviation values were not reported for patient and control dependent variables, effect sizes were estimated from reported t-test, F-test, P-values, or similar statistics. Studies reporting non-significant effects in the absence of further statistics were estimated to have an effect size of zero. Effect size calculation and data-entry procedures were double-checked in order to maximize error detection.
Statistical analyses were conducted using comprehensive meta-analysis version II software (28). A raw effect size, g, using the pooled standard deviation based on Hedges and Olkin (29), was calculated for each cognitive measure in each study. Effect sizes were subjected to a correction for upwardly biased estimates in small samples to yield an unbiased effect size estimate (29). Mean effect sizes across studies were calculated for each cognitive measure employing the weighted average based on the variance of the unbiased effect sizes. The homogeneity of the resulting mean weighted effect sizes for each variable was tested using the Q-statistic. The test of whether mean effect sizes for each variable were significant (different than zero) was tested through the use of Stouffer’s Z, and 95% confidence intervals were derived for each mean effect size (29).
The 15 cognitive measures that were most commonly studied in euthymic bipolar disorder (Table 1) were observed over a span of 39 studies that were included in this meta-analysis (5–16, 18–27, 30–46), which evaluated a total of 948 patients and 1128 controls. Across these studies, there was no significant difference in age (mean ± SD years; patients: 39.8 ± 7.9; controls: 38 ± 8.2; reported in all studies), education (patients: 14.3 ± 1.9 years; controls: 14.3 ± 1.6 years; reported in 25 studies), or gender composition (mean percentage of males patient sample: 53 ± 19.3; control sample: 53.6 ± 20; reported in 38 studies) between patients and controls.
Table 1. Most frequently employed tasks in studies of euthymic bipolar patients
Tasks by cognitive domain
NAART, North American Adult Reading Test; WRAT, Wide Range Achievement Test; WAIS, Wechsler Adult Intelligence Scale; CPT, Continuous Performance Test; CVLT, California Verbal Learning Test; RAVLT, Rey Auditory Verbal Learning Test; WCST, Wisconsin Card Sorting Test.
*When multiple CPT tasks were used within a given study, the simplest, non-working memory version was employed.
Single-word reading score from NAART, WRAT
WAIS vocabulary subtest score
Trailmaking Test A: completion time
Digit Symbol or Symbol Digit Modalities Test score
CPT* or variant: accuracy (hits)
CPT* or variant: reaction time
Learning (recall trials 1–5) from CVLT, RAVLT
Short delayed recall from CVLT, RAVLT
Long delayed recall from CVLT, RAVLT
Recognition hits from CVLT, RAVLT
Digit span backward or total forward and backward
Trailmaking Test B: completion time
WCST: Total categories score
Verbal letter fluency (3 trials): number correct
Stroop interference trial: time, number correct, or interference score
Table 2 presents the mean sample size-weighted effect sizes for each of the cognitive tasks in conjunction with the associated confidence intervals, significance tests, and homogeneity statistics. For the large majority of measures, the Q-statistic failed to reach significance, indicating a relatively homogeneous distribution of effect sizes. This is probably due to the cognitive task approach employed, which only combined tasks that were identical or highly similar in nature. As a result, the inter-study variability due to differences in task parameters was probably greatly diminished.
Table 2. Mean weighted effect size by task for patient-control performance difference
K, number of studies; Np, number of patients; Nc, number of controls; ES, mean weighted effect size, Hedges g; CI, confidence interval; Z, Stouffer’s Z; P value, probability; Q, within-category homogeneity statistic.
***P < 0.001; *P < 0.05.
−0.09 to 0.17
−0.09 to 0.25
0.43 to 0.76
0.60 to 0.98
0.53 to 0.94
CPT (reaction time)
0.45 to 0.79
Learning (trials 1–5)
0.66 to 0.96
0.58 to 0.91
0.57 to 0.87
0.28 to 0.59
Digit span backward
0.36 to 0.71
0.39 to 0.71
0.50 to 0.88
0.31 to 0.63
0.55 to 0.86
The magnitude of effect sizes for the premorbid IQ variables was not significantly different than zero. The Q-statistic for vocabulary score was significant, indicating effect size heterogeneity across studies. This may be related to the fact that often individual studies used vocabulary score as a matching variable between patients and controls, and this may have introduced some variability into the range of effect sizes. Because it was often unclear whether such tasks were ‘matched’ or allowed to vary freely, this variable (matched vs. not matched) could not be coded or evaluated as a moderator.
Results for tasks in the attention/information processing domain revealed effect sizes that consistently fell within the moderate-to-large range in magnitude (47). The Forest plots for these tasks reveals strong, consistent, and homogeneous effects across studies (Fig. 1) indicating impairment in patients. Inspection of the memory task effect sizes also revealed magnitudes approaching the large range, with the exception of recognition hits, which was moderate in size. Given that the recognition hits variable does not account for false-positive responses and is subject to ceiling effects, the effect size for this variable should be interpreted cautiously. The largest effect size was observed for learning trials 1–5; however, this was also one of only two measures that yielded a significant Q-statistic indicating heterogeneity. The Forest plot illustrating the range of effect sizes for all the memory variables is presented in Fig. 2. Despite the variability in the learning trials 1–5 variable, the large majority of effect sizes were significantly larger than zero (based on non-overlap of confidence intervals with zero) indicating impairment in patients with bipolar disorder.
In the domain of executive functioning, a generalized pattern of moderate-to-large effect sizes was also observed across tasks. Effect sizes for Digit Span Backward, Trailmaking Test B, and verbal fluency were moderate in magnitude, and effect sizes for Wisconsin Card Sorting Test (WCST) categories and the Stroop test were moderate to large. Forest plots for the executive tasks reveal strong, stable, and homogeneous effects for all measures (Fig. 3). The data presented in Table 2 employed a fixed effects model, although identical results were obtained using a random effects model.
Trait cognitive impairment
The present study was conducted to quantify the magnitude of cognitive impairment in studies of euthymic bipolar disorder patients and healthy controls. The most frequently studied tasks within this literature span the broad range of attention/processing speed, episodic memory functioning, and various facets of executive functioning. Results of the meta-analysis revealed a pattern of generalized impairment across these domains, with effect sizes typically falling within the moderate-to-large range in magnitude. The fact that studies included in this review were restricted to samples with euthymic patients, and that cognitive deficits for nearly all tasks were homogeneous across studies, underscores the likelihood that the observed cognitive deficits reflect a trait feature of the illness. These results are consistent with another independent meta-analysis that was recently published after the original submission of the present study. Robinson et al. (48) evaluated studies that compared cognitive functioning between euthymic bipolar patients and healthy controls. These investigators required the use of a specific cognitive measure in at least four studies for inclusion into their meta-analysis (as opposed to eight in our study), which probably led to their report of increased heterogeneity in several cognitive measures relative to our study. Nevertheless, their results also revealed moderate-to-large effect size differences favoring controls in the broad areas of executive function, verbal memory, attention and processing speed. Thus, despite the use of independent methodologies, the consistent results across both studies underscore the robustness of the cognitive deficit pattern in euthymic bipolar disorder.
Closer inspection of the magnitude of impairment in each cognitive domain provides insight into the nature of the trait cognitive impairment in euthymic bipolar disorder. In the attention domain, results reveal that patients show slowed visual-motor processing speed, as well as impairment in accuracy and reaction time on sustained attention tasks requiring ability to detect targets. Examination of the magnitude of effect sizes in the memory domain reveals robust and consistently large impairments on core explicit memory measures including learning ability, immediate free recall, and delayed free recall. Finally, in the executive domain, the functions that appear to be the most implicated include cognitive flexibility/set shifting ability as well as dominant response inhibition. More moderate albeit significant deficits were also present in other executive function tasks measuring auditory working memory and verbal fluency. The finding that measures of reading ability and vocabulary in patients were intact, and not significantly different than controls, argues against the likelihood that the described deficits are due to general intellectual deficits.
These results also offer indirect insights into dysfunctional brain systems involved in bipolar disorder. The described cognitive impairments in sustained attention, episodic memory, and multiple aspects of executive functioning are consistent with models of dysfunction within prefrontal-striatal, fronto-temporal, and associated anterior limbic brain regions (4, 49, 50). Based on existing distinctions between separate yet overlapping frontal-striatal functional systems/loops (51), the finding of impaired sustained attention, set shifting, working memory, and fluency is consistent with dysfunction within the dorsolateral prefrontal-subcortical circuit (52). Additionally, impaired response inhibition based on stroop interference performance points to associated involvement of the medial prefrontal system (53). The finding of robust and consistent impairments in explicit memory may also point to a medial temporal/diencephalic component (54). Overall, cognitive dysfunction within the broadly described brain systems maps reasonably well onto structural and functional neuroimaging findings in bipolar disorder (55, 56).
Potential confounds or moderators
Although the present results strongly suggest a trait-related neurocognitive disturbance, the possible moderating influence of two key variables warrants consideration. The first of these variables is the potential negative influence of treatment with psychotropic medication on cognition. Existing research suggests that lithium carbonate, the most commonly used mood stabilizer, may exert a mild selective influence on verbal memory and processing speed (57). The cognitive side effect profile of other common mood stabilizers such as valproate and carbamazepine appears to be more benign (57, 58) and there is some preliminary evidence that improvements in cognition may be associated with the use of newer anti-convulsant mood stabilizers such as lamotrigine and gabapentin, with the notable exception of topiramate (58–60). Additionally, although more studies are needed in bipolar patients, the favourable cognitive profile of atypical antipsychotic agents is recognized in psychotic disorders such as schizophrenia and implied in bipolar disorder (60). The absence of association between medication dose and cognitive impairment in most studies of euthymic bipolar patients suggests that if medication side effects are present, they are relatively minor in magnitude and do not probably account for the full extent of cognitive impairment in euthymic bipolar disorder (6, 9, 13, 14, 16, 18, 20, 24, 26, 30).
Another potential confound that has received some attention is the influence of residual or subclinical mood symptoms on neurocognitive functioning in euthymic patients. This issue becomes complicated by the fact that different investigators use varying criteria or measures to define euthymia in their patient samples. To address this problem, the most sophisticated studies have imposed highly selective inclusion criteria requiring minimal mood rating scores for inclusion into the euthymic group, and have also statistically co-varied out the influence of residual mood symptom differences between patients and controls when they exist. Although in some cases the statistical control of residual symptoms has led to the reduction or disappearance of some cognitive deficits in euthymic patients (7, 24), the consensus is that cognitive impairments persist in the broad areas of attention, memory, and executive function (9, 10, 18, 21). This practice, however, does not definitively rule out the potential influence of subclinical symptoms upon cognition. The observation that bipolar patients continue to show prominent psychosocial and functional impairments even after symptomatic remission or recovery (61, 62) not only highlights the possibility that cognitive impairment is an important determinant of psychosocial dysfunction (63), but that cognitive deficits may also at least in part reflect ongoing sub-threshold symptoms.
Overall, it would appear that even though the impact of medication and residual mood symptoms upon cognitive functioning has yet to be fully elucidated, it is not likely that these factors fully account for the robust cognitive deficits present in euthymic bipolar patients.
Limitations and future directions
One advantage of restricting our analysis to the quantification of only the most commonly employed tasks in this literature is that the derived effect size estimates for the included tasks were more likely to be reliable and valid indicators of cognitive functioning in euthymic bipolar disorder. A downside to this approach is that we have necessarily limited the scope of our review by excluding some potentially relevant and informative cognitive tasks that did not make the established criterion for inclusion (n = 8 studies). For example, reflecting the primary focus within this literature to date, all the executive function measures included in our review preferentially tap into the dorsal rather than ventral prefrontal cognitive system (52). While the former system is obviously implicated in bipolar disorder, involvement of the ventral prefrontal system is also being increasingly recognized (55). Thus, tasks that probe the ventral system by measuring broader executive functions such as decision making (16), theory of mind (36, 64), and self-regulation (39) are excluded from our analysis.
Despite the fact that the current study has quantified the major domains and specific tasks that are impaired in euthymic bipolar patients, it does not permit direct evaluation of whether a core cognitive deficit underlies the described pattern of impairment across tasks, nor of the relationship between the various observed task deficits. For example, some authors propose that sustained attention impairment represents the core trait deficit observed in euthymic bipolar disorder, in part based on the report that these deficits are not accounted for by executive functions such as working memory (1, 13, 24, 25). By implication, elementary attentional or processing speed impairments could also adversely influence performance on memory or other complex cognitive tasks (20). An alternate viewpoint is that executive dysfunction represents the central bipolar trait deficit, and this impairment underlies not only the broad cognitive deficit pattern that is observed, but also the psychosocial and functional deficits exhibited by patients (7, 35, 50, 65). The observation that memory deficits in euthymic patients are mediated by deficient organizational strategies at encoding (18) or poor retrieval processes (19) rather than deficient retention could be construed as consistent with the executive dysfunction hypothesis. A third perspective, which appears challenged by the preceding memory component findings, is that episodic memory impairment is the core cognitive impairment (14, 16, 22) and that this deficit reflects primary hippocampal/medial temporal lobe dysfunction characteristic of amnestic disorders (54).
An alternate means of integrating these different perspectives is to propose that the discrete purported cognitive deficits (attention, executive, and memory) may actually have more in common than is outwardly apparent, or may arise from a common source. This assertion is based on the likely substantial yet blurred overlap between constructs such as attention and executive function (66). It may be that constructs such as response inhibition or distractibility, which are probably relevant to attentional, executive, and memory tasks, may best describe the cognitive deficit inherent in bipolar disorder. To further tease out this type of possibility, more studies in euthymic patients will need to be conducted that evaluate multiple cognitive domains as well as the inter-relationships between identified impairments.
A final possibility is that there may indeed be multiple core cognitive deficits that characterize euthymic bipolar disorder, and this underlies the observed generalized pattern of cognitive deficit. Whether this is the case remains an unanswered question; however, based on the results of the current meta-analysis, it is apparent that unitary models asserting a core neuropsychological impairment in bipolar disorder must account for the fact that consistent and sizeable deficits are observed on a broad range of attentional, executive, and memory measures. The preceding theoretical questions should continue to drive an important area of cognitive research in bipolar disorder.
Assuming the presence of a trait neuropsychological impairment, an area that requires further study is the specification of cognitive impairment prior to and throughout the course of illness. Preliminary data suggest that cognitive deficits are detectable in first-degree relatives (20, 46, 67), precede or are present at illness onset (24, 68), and progress throughout the course of the illness (69), raising the likelihood of both genetic/neurodevelopmental and neurodegenerative components. Clearly, longitudinal studies in patients and high risk groups are needed to more adequately investigate these important etiological questions. A related question that deserves further scrutiny is whether subsets of patients with specific clinical features show preferential cognitive impairment. For example, patients with psychotic symptoms or impulsivity may have a more severe course and poorer outcome (70, 71), raising the possibility that specific subsets of individuals may show preferential cognitive impairment (72).
Finally, if cognitive impairments in euthymic patients are to serve as legitimate trait markers or endophenotypes of bipolar illness (73), this will in part depend on their link to underlying measures of brain function. The attempt to make such brain–behaviour connections is occurring using the tools of both structural and functional imaging (8, 35, 44, 49, 55, 56, 65). Continued efforts to specify the nature and extent of the trait neurocognitive deficit in bipolar disorder should facilitate the links between brain and behaviour, as well as identify the target cognitive symptoms for medical and psychosocial treatment and rehabilitation.
This research was supported by Riverview Hospital Research Grant GR100023 awarded to IJT. The authors thank Allen E. Thornton, PhD for assistance with meta-analytic techniques. The authors thank Janssen-Cilag for an unrestricted educational grant for the publication of this Acta Psychiatrica Scandinavica supplement article.
Declaration of interest
IJ Torres and VG Boudreau declare no conflict of interests.
Lakshmi N. Yatham has served on the advisory boards of, been a speaker for, and received grant support from Lilly, Janssen-Cilag, Astra-Zeneca, Bristol-Myers-Squibb, Novartis, Pfizer, GlaxoSmithKline, and Servier.