Three major findings emerged from this study. First, compared to healthy adults, patients with schizophrenia showed severe, pervasive cognitive impairments. Second, while our relatively stable patients with BD also demonstrated impairment on most cognitive measures, their deficits were significantly milder than those of patients with SZ on 11 of 19 individual measures and in 5 of 6 cognitive domains. Third, the cognitive deficit effect sizes shown by SZ and BD patient groups correlated quite highly, indicating that the two groups exhibited qualitatively similar patterns of cognitive impairment despite differences in the overall severity of their deficits.
The finding that patients with schizophrenia showed severe, pervasive cognitive deficits is consistent with many previous studies. Based on a meta-analysis of 204 studies that included 7,420 schizophrenic patients and 5,865 healthy controls, Heinrichs and Zakzanis (1998)
found that the mean weighted effect sizes for 22 cognitive test variables ranged from 0.46 to 1.41, with a mean of 0.89. In the present study, based on demographically-adjusted scores, we found that the effect sizes for 19 cognitive test variables ranged from 0.37 to 1.32, with a mean of 0.97. When these variables were assigned to six cognitive domains that are conceptually similar to several described by Heinrichs and Zakzanis (1998)
and Seidman et al (2002)
, and which include several measures that comprise the MATRICS battery (Nuechterlein and Green 2006
), the effect sizes were larger, ranging from 0.91 to 1.71, with a mean of 1.22. Notably, the two largest effects found by Heinrichs and Zakzanis involved measures of verbal memory and psychomotor speed, which is precisely what we found. Likewise, Seidman et al (2002)
found that SZ patients showed their greatest impairment in domains that the investigators called perceptual-motor speed, abstraction/executive functioning, and verbal declarative memory. Two of these correspond most closely to the domains (psychomotor speed, verbal learning/memory) in which our SZ patients also demonstrated maximal impairment. Finally, our findings also are quite consistent with those of Dickerson et al (2004)
, who found that SZ patients were most impaired relative to NCs on measures that constitute the RBANS Attention and Immediate and Delayed Memory subtests. Notably, the RBANS Attention subtest includes a measure (Coding) that we, Heinrichs and Zakzanis (1998)
, and Seidman et al (2002)
all refer to as perceptual-motor or psychomotor speed. Thus, the range and mean of the observed effect sizes, as well as the specific cognitive domains that comprised the largest of them in this study, were strikingly similar to those found in previous comparisons of schizophrenic patients and healthy adults.
Also consistent with previous research, our patients with bipolar disorder showed milder and more selective cognitive deficits than those with schizophrenia. Effect sizes ranged from 0.23 to 0.87, with a mean of 0.59 across all 19 cognitive measures when we compared bipolar patients to healthy controls. Based on their review of 42 studies, Quraishi and Frangou (2002)
concluded that even BD patients who are described as remitted frequently show mild residual cognitive deficits, most frequently of sustained attention and verbal memory. As depicted in and , these were among the largest differences between BD patients and NCs found in the present study. Indeed, after adjusting for differences in demographic background and estimated premorbid IQ, patients with bipolar disorder demonstrated impairments of sustained and divided attention that were proportionately as severe as those shown by patients with schizophrenia. Similarly, Seidman et al (2002)
and Dickerson et al (2004)
also reported finding moderately large effect sizes for deficits in attention and verbal learning/memory among their BD patients relative to healthy adults. Also consistent with their reports, the BD patients in our study outperformed SZ patients in every cognitive domain. These differences reached statistical significance for every domain except
attention. In sum, these findings support three inferences: First, relatively stable patients with bipolar disorder demonstrate cognitive impairments. Second, their deficits are milder and more selective than those seen in schizophrenia. Third, after adjusting for demographic differences, their impairments of sustained and divided attention appear to be as severe as those shown by patients with schizophrenia.
Perhaps most importantly, in this study, patients with schizophrenia and bipolar disorder produced neuropsychological profiles that differed in severity but were qualitatively similar. As shown in and , their neuropsychological profiles are not identical, but correlation of the effect sizes shown by BD and SZ patients yielded a Pearson r
of 0.71 for the 19 individual tests and 0.72 for the six cognitive domains. Squaring these values suggests that roughly 50% of the variance in cognitive profiles is shared between patient groups. While the correlations that we found are smaller than those obtained by Seidman et al (r
= 0.81) and Dickerson et al (r
= 0.93), these investigators did not adjust the cognitive test performance effect sizes for group differences in demographic background or estimated premorbid IQ. Since age, education, sex, race, and estimated premorbid IQ all affect cognitive abilities to varying degrees, differences between SZ and BD patients in these respects are likely to weaken rather than strengthen the correlation between their neuropsychological profiles. In any case, consistent with the reports of Seidman et al (2002)
and Dickerson et al (2004)
, the present findings suggest that SZ and BD share greater phenotypic similarity in terms of the pattern than severity of their neurocognitive deficits. While this does not eliminate the possibility of identifying neuropsychological patterns that reliably distinguish between SZ and BD, the qualitative similarity of their cognitive deficits suggests that measures of other biological characteristics, such as electrophysiological or neuroanatomic abnormalities, might be more likely to yield useful biological illness markers.
For the present study we treated both schizophrenia and bipolar disorder as unitary entities. Some researchers argue for separating patients with psychotic BD from those without psychosis, based on differences in genetics and brain abnormalities (Glahn et al 2006
; Pearlson et al 1995
; Potash 2006
), and stress the similarity of psychotic BD and SZ. We currently are gathering data on a larger sample of patients with BD, and we aim to construct psychotic and non-psychotic subgroups of sufficient size to support such analyses. Likewise, although we grouped patients with SZ according to whether or not they manifested the deficit syndrome (Carpenter and Kirkpatrick 1988
), comparison of the cognitive functioning of these subgroups will be presented in a subsequent article.
One weakness of this study is that measures of current affective symptoms were not administered. Consequently, although 63 of the 66 patients with BD were assessed as outpatients, and all of the patients were relatively stable, we do not know how many were euthymic at testing. However, the bipolar patients showed fewer negative and positive symptoms than schizophrenic patients. In addition, roughly equal proportions of both groups were assessed while inpatients. This reduces the likelihood that differences in the severity of their cognitive deficits are attributable to differences in illness acuity between the two groups. Another potential weakness of this study involves the possibility that misdiagnosis contributed to the qualitative similarity of cognitive profiles shown by our BD and SZ groups. While we cannot exclude this possibility, the systematic differences in the overall severity of cognitive deficits shown by BD and SZ patients argues against this, as does the fact that other investigators have found even greater qualitative similarities between these patient groups.
An important and novel strength of this study involves the use of regression-based adjustments for demographic characteristics and estimated premorbid IQ. This approach standardizes the difference between an individual’s actual obtained test scores and those predicted on the basis of his or her demographic characteristics and estimated premorbid IQ. Regression-based demographic adjustments of test scores (Heaton et al 2004
) obviate the need to match patient and control groups on demographic variables because their influence is statistically “removed” from each individual’s test scores. The effectiveness of this approach is demonstrated by the fact that non-overlapping subgroups of healthy controls (e.g., those with ≤ 12 vs. >12 years of education) did not differ significantly on a single t
-test in 95 statistical comparisons (see footnote above). In fact, the adjusted T
-scores of every NC subgroup approximated a mean of 50 and standard deviation of 10. When applied to patient groups, this approach improves diagnostic classification by increasing the sensitivity, without decreasing the specificity, of neuropsychological measures (Testa and Schretlen 2006
). In effect, the cognitive deficits and corresponding effect sizes shown by SZ and BD patients in this study reflect impairments that are unrelated to group differences in age, sex, race, education, and estimated premorbid IQ. Meehl (1970)
noted that matching patients with schizophrenia and healthy controls might “correct” out genuine illness-related variance. The same could be said of including a term for education in regression formulas used to residualize test performance for demographic characteristics and estimated premorbid IQ. However, when all predictors are entered into the regression models, education accounted for very little incremental
variance (as shown by its relatively small beta weight) in any test score. In addition, the cognitive test effect sizes obtained in this study were remarkably similar to those reported for comparable measures in previous research. This suggests that including a term for education did not greatly reduce group differences in cognitive test performance, which is the principal risk of “over-correcting” scores. In any case, this is the first study to compare the cognitive impairments shown by patients with BD and SZ using regression-based demographic adjustment as far as we know.
Qualitative similarities between the neuropsychological profiles produced by our patients with schizophrenia and bipolar disorder are consistent with the view that these disorders might represent a “continuum of psychosis” (Crow 1997
) or share some genetic biological overlap (Berrettini 2004
). They support continuing research on the traditional distinction between SZ and BD. For example, pooling patients without regard to diagnosis, and then defining clusters by clinical symptoms, neurocognitive functioning, and other putative biomarkers might prove useful for linkage and gene expression analyses of homogeneous subgroups that cross diagnostic boundaries.