Working Memory in Schizophrenia: Transient "Online" Storage Versus Executive Functioning

Working Memory in Schizophrenia: Transient "Online" Storage Versus Executive Functioning William Perry, Robert K. Heaton, Eric Potterat, Tresa Roebuck, Arpi Minassian, and David L. Braff Abstract Working memory has been described as the temporary "online" storage and the subsequent manipulation and retrieval of information. It has been suggested that the prefrontal cortex is a primary site of working memory. Schizophrenia patients, who are thought to have prefrontal cortical dysfunction, have demonstrated inconsistent deficits on a variety of verbal and spatial working memory tests. This has led to questions about how to define and measure working memory, whether these deficits are distinct to one cognitive domain, and what role factors such as intelligence and play in working memory performance. We compared schizophrenia patients to normal comparison subjects in four separate studies. Based upon the results we recommend that working memory tests be characterized as either transient "online" storage and retrieval tasks (where short-term storage and retrieval of information is required) or executive-functioning working memory tasks (where storage, manipulation, and retrieval of information is required). The importance of clearly identifying which distinct aspects of working memory are assessed is discussed. Keywords: Working memory, executive functioning, schizophrenia, intellectual functioning,, frontal cortex. Schizophrenia Bulletin, 27(1):157-176,2001. Baddeley and Hitch (1974) introduced the dynamic concept of working memory to describe a limited-capacity "working space" for information processing. Baddeley explained that the working memory system is active and subject to rapid decay unless rehearsal processes are initiated (Salame and Baddeley 1982). In Baddeley and Hitch's conceptualization of working memory there is a superordinate "central executive" system and two slave systems for the transient storage of visuospatial and verbal information. Information is encoded as visuospatial or verbal (phonological) traces and is maintained by the two slave systems via the "visuospatial sketchpad" or subvocal "phonological or articulatory loop" (Baddeley and Hitch 1974). These subsystems and the central executive controller are responsible for the allocation of and direction of attentional resources, the manipulation of the visual and phonological material, and the selection and execution of strategies. Collectively, the working memory system provides "temporary storage and manipulation of the information necessary for such complex cognitive tasks as language comprehension, learning, and reasoning" (Baddeley 1992, p. 556). Baddeley and Hitch (1994) have found that "the term working memory is used in a number of different ways" (p. 485), and these different usages can lead to substantial variation in how working memory is defined and studied. Goldman-Rakic and her colleagues (Goldman-Rakic 1987, 1991; Funuhashi et al. 1989; Funuhashi et al. 1993; Wilson et al. 1993) have employed the classic delayed response paradigm to study the ability of nonhuman primates to guide behavior by symbolic representation in the absence of external cues. Their work has provided strong evidence that in nonhuman primates, the prefrontal cortex is a primary site of transient "online" visuospatial information storage. Goldman-Rakic (1991) suggested that "At the most elementary level, our basic conceptual ability to appreciate that an object exists when out of view depends on the capacity to keep events in mind beyond the direct experience of those events" (p. 3), and this capacity entails some form of online storage or rehearsal. Thus, in the model presented by Goldman-Rakic, the storage of the existence and location of the object, in the absence of the object, is the critical component of working memory. Petrides (1995) has focused upon self-ordered and externally ordered aspects of working memory. According to Send reprint requests to Prof. W. Perry, University of California, San Diego, Department of Psychiatry, La Jolla, CA 92093-8620; e-mail: wperry@ucsd.edu. 157

Schizophrenia Bulletin, Vol. 27, No. 1, 2001 W. Perry et al. the view presented by Petrides, the monitoring of increasing amounts of information and the active retrieval of information are the critical areas of working memory that need to be considered. In this context, we propose that working memory can be defined as transient online storage and recall working memory, when short-term information storage but not manipulation is used. In contrast, "executive-functioning working memory" should be used when, in addition to short-term information storage, manipulation and recall of that information are used via central executive functioning. Recently, the role of working memory deficits has been highlighted in schizophrenia patients. Several authors have shown that schizophrenia patients and their relatives have impaired performance on a variety of working memory tasks. These findings have been used to support the hypothesis that the prefrontal cortex is impaired in schizophrenia spectrum patients (Park and Holzman 1992; Goldberg et al. 1993; Park et al. 1995; Gold et al. 1997). Park and Holzman (1992) have used constructs that depend on the more limited transient storage definition of working memory in schizophrenia patients by using a human analogue of the delayed response paradigm employed by Goldman-Rakic. The authors found that schizophrenia patients exhibited impaired visuospatial working memory performance but not impaired auditory working memory as assessed by Digit Span. Goldberg et al. (1993) found that schizophrenia probands had performances similar to their clinically nonaffected identical twin on Digit Span forward but exhibited poorer performance on Digit Span backward. They noted that during Digit Span backward, numbers are held and manipulated in working memory and recalled in reverse order. The authors suggested that the Digit Span backward task places a greater cognitive load upon the central executive controller of working memory than a pure "hold and repeat" forward span task, thus accounting for the poor performance of the schizophrenia twins. In addition to working memory studies using transient-storage working memory tasks, there have been numerous neuropsychological studies of schizophrenia patients using complex executive function tasks to assess working memory. Although these measures are not typically considered working memory measures per se, they are conceptualized as requiring online storage, manipulation, and then retrieval of information, and the deficient performance of schizophrenia patients on these tasks is often attributed to impaired working memory (Gold et al. 1997). Among the most widely used tasks that tap into executive function working memory are the Wisconsin Card Sorting Test (WCST); the Category Test; puzzle tests such as the three "tower" tests of Hanoi, Toronto, and London; and delayed response tasks that require sustained attention and simultaneous information processing, such as the Auditory Consonant Trigram Test. Goldman- Rakic (1991) reported that all of these tasks require that "symbolic representations are accessed and held 'online' to guide behavior in the absence of or despite discriminative stimuli in the outside world" (p. 13). In other words, these tasks require an individual to "keep instruction, concepts, and goals 'in mind' " (Goldman-Rakic 1991, p. 17). Furthermore, Goldman-Rakic states that these tests all share a common prefrontal cortex neurobiological substrate. To assess how schizophrenia patients perform on tasks that require online storage as well as manipulation and then retrieval of information, Gold and colleagues (Gold et al. 1997) studied a Letter-Number Span (LNS) task to increase the cognitive (central executive) demand characteristics of simple Digit Span tasks. Similar to the Digit Span backward test, the LNS requires the subject to hold information in temporary storage but also requires the subject to sort out the letters from the numbers and to separately recall the letters and numbers in successive order. The authors explained that Digit Span forward or the simple recall of a series of digits did not require the "cognitive manipulation of stored information and therefore, demanded little of the working memory system" (Gold et al. 1997, p. 161). They compared the performance of schizophrenia patients on the LNS to patients' performance on the WCST. Gold and colleagues demonstrated that schizophrenia patients do exhibit performance deficits on the LNS task, and they attributed these deficits to an impaired auditory working memory system. Furthermore, this impaired LNS performance was highly correlated with WCST perseverative responses (r - -0.52). The authors suggested that a next step in understanding the critical role of working memory would be to assess the relationship between working memory and symptomatic state (Gold et al. 1997). Pantelis and colleagues (Pantelis et al. 1997) also demonstrated deficits in short-term memory, visuospatial working memory, and strategy generation and planning ability in schizophrenia patients. They reported a lack of significant correlations between the different executive functioning tasks, which they interpreted as suggesting "impaired functional connectivity between different regions of the neocortex" (Pantelis et al. 1997, p. 1823). The authors further suggested that schizophrenia patients' performance resembled the performance of patients with both frontal and basal ganglia pathology. Compared with tests of transient-storage working memory functioning, executive functioning tasks are more complex, use more cognitive operations, and appear to be more highly correlated with general intelligence. For 158

Working Memory Schizophrenia Bulletin, Vol. 27, No. 1, 2001 example, Chen and McKenna (1996) pointed out that the observed working memory deficits could be related to task difficulty rather than the construct of working memory. Kyllonen and Christal (1990) addressed the relationship between working memory and reasoning ability and concluded that working memory and reasoning ability, considered to be at the core of intelligence, are "similar if not identical" (p. 410). They further propose that reasoning ability is the primary determinant of how well a person benefits from instruction and "governs the success of the life-long process of accumulating crystallized knowledge" (p. 392). In general, studies of working memory have not adequately addressed whether the observed deficits are best explained by working memory capacity limitations or deficient reasoning or intelligence. Thus, despite the widespread usage of the term "working memory," there remains considerable debate regarding its definition, how best to measure the construct (or constructs), and which (if any) other factors better explain the deficits observed among schizophrenia patients. To resolve the differing views of working memory that are summarized above, and to extend the assessment of working memory in schizophrenia patients, several questions were addressed in the current research: (1) Do schizophrenia patients have impaired working memory performance compared to age-, education-, and gender-matched normal comparison subjects? (2) Do schizophrenia patients have working memory deficits in a specific cognitive domain (i.e., auditory versus visuospatial)? (3) Do schizophrenia patients have selective deficits in working memory that are distinct from impairment in areas of attention and intellectual functioning? (4) Because schizophrenia patients are thought to have primary working memory deficits that impact their performance on executive functioning tests, is there a difference between executive functioning test performance for schizophrenia patients and that for nonpatients? (5) Does a relationship exist between performance on tests of executive functioning and symptom factor scores, especially those thought to reflect deficits in the prefrontal cortex? To address these questions we conducted four separate studies using separate samples of schizophrenia patients. Study 1 In this study we compared the performance of schizophrenia patients to normal subjects on the Digit Span forward and backward test. We hypothesized that if schizophrenia patients have selective deficits in central executive working memory then they should have differentially greater deficits on the backward portion of the test when compared to a group of normal subjects, and these deficits should be distinct from moderating factors such as intellectual functioning. Method Subjects. Fifty schizophrenia patients (24 males and 26 females; 37 Caucasian, 4 African-American, 4 Hispanic, 2 Asian, and 3 other) participated. All met the DSM-HI-R (American Psychiatric Association 1987) criteria for schizophrenia, as determined by the use of the Structured Clinical Interview for DSM-III-R (SCID-m-R; Spitzer et al. 1990), conducted by experienced doctoral-level clinicians. We established a 98 percent agreement for determining an Axis I diagnosis using the SCID-III-R. Subjects were excluded if they were determined to have an additional Axis I diagnosis, met DSM-III-R criteria for substance abuse or dependence within the past 6 months, had an unstable medical condition, had a history of a neurological disorder (e.g., a head injury with loss of consciousness), or were treated in the past with electroconvulsive therapy. The 50 schizophrenia patients consisted of the following subtypes: 21 paranoid, 25 undifferentiated, 3 disorganized, and 1 residual. Five patients were recruited from an acute inpatient psychiatry service, 37 were recruited from a long-term subacute inpatient facility, and 8 were outpatients. Seventeen of the schizophrenia patients were being treated with traditional neuroleptic medication (mean chlorpromazine equivalents = 624 mg, standard deviation [SD] = 495 mg). Twelve patients were treated with clozapine (mean dose = 391.67 mg, SD = 162.14 mg), six were on olanzapine (mean dose = 12.8 mg, SD = 3.9 mg), seven patients were treated with risperidone (mean dose = 6.43 mg, SD = 2.65 mg), four patients were on a combination of typical and atypical antipsychotic medication, and four were not treated with antipsychotic medication. Thirty subjects were taking anticholinergic medications with a mean benzotropine equivalent of 2.7 mg (SD = 1.8) with a range of 0.5-8 mg (Jeste and Wyatt 1982). A normal comparison group consisting of 50 subjects (23 females and 27 males; 34 Caucasian, 9 African- American, 3 Hispanic, 2 Asian, and 2 other) also participated. The normal comparison subjects were recruited from newspaper advertisements and were screened for the exclusionary criteria of Axis I pathology (using the Structured Clinical Interview for DSM-Non Patient version), substance abuse or dependence within the past 6 months, an unstable medical condition, or history of neurological disorder (e.g., head injury with loss of consciousness). The mean age for schizophrenia patients was 35.8 (SD = 8.4) and for the normal comparison subjects 32.0 (SD = 10.2). There was no significant difference between the groups for age (<[98] = -2.05). The mean years of education for the schizophrenia patients was 12.6 (SD = 2.2) and for the normal comparison subjects was 159

Schizophrenia Bulletin, Vol. 27, No. 1, 2001 W. Perry et al. 13.6 (SD = 1.6). There were no significant differences between the groups with respect to years of education (/[98] = 2.71). The mean age of onset of illness for the schizophrenia patients was 21.9 (SD = 5.6). After a complete description of the study was given to the subjects, written informed consent was obtained. Subjects were then tested on the Vocabulary and Digit Span subtests from the Wechsler Adult Intelligence Scale-Revised (WAIS-R; Wechsler 1981). Vocabulary subtest scores were converted to age-corrected scaled scores. Digit Span is typically presented as a single score combining forward and backward performance. To compare forward to backward performance, the raw numbers of correctly obtained items for forward and backward were analyzed independently. Results Schizophrenia patients achieved a mean WAIS-R Vocabulary scaled score of 8.4 (SD = 2.5) compared to 10.6 (SD = 2.5) for the normal comparison subjects (t [98] = 4.48, p < 0.001). On the Digit Span subtest, schizophrenia patients achieved a mean forward span score of 6.7 (SD = 1.8) compared to 9.3 (SD = 2.1) for the normal comparison subjects; corresponding means (SDs) for backward Digit Span were 5.4 (1.9) and 7.6 (2.4) for the respective groups. A two-way analysis of variance (ANOVA) was conducted, with diagnosis (schizophrenia or normal) as the between-subject factor and test condition (forward versus backward) as the within factor. This yielded a significant main effect for diagnosis (schizophrenia patients versus normal comparison subjects) (F[l,98] = 41.66, p < 0.001), indicating that the schizophrenia patients performed significantly lower on Digit Span than did the normal comparison subjects. Also there was a significant difference between forward and backward Digit Span regardless of diagnosis (F[l,98] = 59.02, p < 0.001]. There was no significant interaction of diagnosis by Digit Span type (F[l,98] = 1.14) (see figure 1). Therefore, schizophrenia patients do not appear to perform differentially worse than normal comparison subjects on the backward relative to the forward component of Digit Span. The analyses were repeated with WAIS-R Vocabulary performance covaried. This did not, however, change the results (i.e., significant effects were obtained for diagnosis and forward versus backward Digit Span, Figure 1. Forward and backward Digit Span performance for normal subjects and schizophrenia patients (Study 1) FORWARD BACKWARD Normal Subjects (N=50) Schizophrenia Patients (N=50) 160

Working Memory Schizophrenia Bulletin, Vol. 27, No. 1, 2001 but the interaction was nonsignificant). There was also no significant difference found between patients treated with typical neuroleptics and patients treated with atypical neuroleptics. To further explore whether working memory and intellectual functioning are relatively distinct processes, we covaried the effects of Digit Span forward and backward performance on Vocabulary performance. Removing the influence of Digit Span forward resulted in nonsignificant differences between schizophrenia and normal comparison subjects on WAIS-R Vocabulary performance (F[l,97] = 2.79). Covarying out Digit Span backward did not significantly change the difference in Vocabulary test performance found between schizophrenia and normal comparison subjects (F[l,97] = 8.83, p < 0.005). Finally, Pearson correlation coefficients between WAIS-R Vocabulary and Digit Span forward (r = 0.20) and backward (r = 0.31) were not significant for the schizophrenia patients. Among the normal comparison subjects WAIS-R Vocabulary was significantly correlated with Digit Span forward (r = 0.54) but not Digit Span backward (r = 0.22). These results suggest that Digit Span backward may be relatively distinct from verbal intellectual functioning, particularly for schizophrenia patients. Conclusion The results support the hypothesis that schizophrenia patients perform worse on both components of Digit Span than do normal comparison subjects. The unexpected finding, however, is that there was no significant interaction of diagnosis by Digit Span type. This indicates a lack of differentially greater impairment on the backward versus forward version of the test. Schizophrenia patients also performed significantly worse on the WAIS-R Vocabulary test, despite there being no differences between the patients and normal comparison subjects in level of education completed. These results suggest that these schizophrenia patients have a generalized deficit across numerous cognitive domains, including short-term memory of digits. Finally, covarying out the influence of Digit Span forward from Vocabulary performance eliminated the significant differences observed between schizophrenia patients and normal comparison subjects, whereas covarying Digit Span backward did not. Thus, Digit Span backward and Vocabulary performance may represent unrelated deficits. Still, the impaired performance of schizophrenia patients on Digit Span forward or backward cannot simply be attributed to the taxing of an impaired working memory system. Schizophrenia patients performed approximately one SD below normal comparison subjects on both the backward and the forward components of Digit Span, achieving a very short span. Study 2 In this study we examined the performance of schizophrenia patients on five measures from three tests of memory span from the Wechsler Memory Scale-Third Edition (WMS-III): Digit Span (forward and backward), Spatial Span (forward and backward), and Letter-Number Sequencing. We selected these tests to study whether schizophrenia patients have differential deficits in spatial versus verbal aspects of working memory, as proposed by Park and Holzman (1992). Method Twenty-eight schizophrenia patients (21 males and 7 females; 17 Caucasian, 4 African-American, 4 Hispanic, 1 Asian, and 2 other) who were diagnosed using the Structured Clinical Interview for DSM-IV (SCID-IV, First et al. 1995) participated in this study. Their performance on multiple tests of working memory was compared to standardized published norms; none of these subjects participated in any of the other studies reported here. The diagnostic interviews were administered by doctoral-level clinicians. All diagnostic and exclusion criteria are as described above in Study 1. All of the patients were tested once they were deemed stable by their treatment team. The schizophrenia patients consisted of the following subtypes: 16 paranoid, 7 undifferentiated, and 5 disorganized. Twenty-one patients were recruited from the University of California, San Diego, inpatient psychiatry service and seven were recruited from a long-term subacute inpatient facility. The patients ranged in age from 23 to 65 (mean age = 41.14, SD = 9.9) and had a mean of 12.8 years of education (SD = 2.7). The mean age of onset of illness was 23.2 years (SD = 6.2). All of the patients were undergoing treatment with a neuroleptic medication at the time of testing. Most patients (n = 20) were treated with risperidone (mean daily dose = 4.2 mg, SD = 1.4) and did not require anticholinergic medication. Among the 20 patients treated with risperidone, 5 were receiving concomitant therapy consisting of a typical neuroleptic. The eight remaining patients were treated with haloperidol (n = 4, mean daily dose = 10 mg, SD = 1.6) and olanzapine (n = 4, mean daily dose = 18.7 mg, SD = 2.5). Of the 28 patients, 12 were also receiving low-dose benzodiazepines. There were no significant differences between the patients who received different medications on working memory test performance. After a complete description of the study was given to the subjects, written informed consent was obtained. Twenty-one subjects were assessed for level of verbal intellectual functioning using either the Adult North American Reading Test (Blair and Spreen 1989) (n = 12) 161

Schizophrenia Bulletin, Vol. 27, No. 1, 2001 W. Perry et al. or the Reading subtest from the Wide Range Achievement Test-Hi (n = 9). The use of two separate measures of premorbid intellectual functioning is a limitation in this study. Still, we combined the data in the following manner. The performance on the two tests ranged from 45 to 118, and the collapsed mean for the overall sample was 99.4 (SD = 18.2). Subjects were then tested on the following measures: the Digit Span and Letter-Number Sequencing subtests from the Wechsler Adult Intelligence Scale-Third Edition (WAIS-IH) and the Spatial Span subtest from the WMS-ffl (Wechsler 1997). All of the subtests were administered according to standard directions in the manual. Raw scores from Digit Span (total), Spatial Span, and Letter-Number Sequencing were age corrected and converted to scaled scores. Because scaled scores were not available for the subject's longest Digit Span (forward and backward individually,), z scores were calculated from tables in the WAIS-in manual. The z scores were converted to scaled scores with a mean of 10 and an SD of 3 so that all five measures could be directly compared with each other and with the WMS-III standardization sample. We hypothesized that the schizophrenia patients would demonstrate significantly impaired performance across all of the working memory measures compared to the standardization sample. We then compared the performance across the five measures to examine the following: (1) whether schizophrenia patients have selective visuospatial versus auditory/verbal working memory deficits, as previously suggested (Park and Holzman 1992); and (2) whether schizophrenia patients have greater impairment on backward versus forward span tasks, because of a greater taxing of the working memory system (Goldberg et al. 1993; Pantelis et al. 1997). One-sample t tests were used to compare subjects' performance for each of the five working memory measures against published norms. The alpha criterion was conservatively set top < 0.025 because of the number of analyses conducted. To compare the performance of the schizophrenia subjects across the five measures, two ANOVAs were conducted. The first was a 2 X 2 ANOVA with modality (auditory vs. visuospatial) and condition (forward vs. backward) as the within-subject factors. The second was a three-way ANOVA with subtest (Digit Span, Spatial Span, and Letter-Number Sequencing) as within-subject factors. This approach allowed us to test the two main questions of interest. For these two analyses the alpha criterion was again set at p < 0.025. Results The mean performance on Digit Span forward and backward, Spatial Span forward and backward, and Letter- Number Sequencing revealed that the patients performed significantly below average when compared to the agematched standardization sample on all but the Digit Span forward measure (table 1). The 2X2 ANOVA with modality (auditory vs.visuospatial) and condition (forward vs. backward) as the within-subject factors resulted in a trend for modality (Pillai's Trace F [1,27] = 3.65, p = 0.07) with the verbal mean being higher (mean = 8.6, SD = 2.53) than the visuospatial mean (mean = 7.6, SD = 3.06). There was a nonsignificant main effect for condition (forward vs. backward; Pillai's Trace F [1,27] = 1.03, ns) and a nonsignificant modality by condition interaction (Pillai's Trace F [1,27] = 3.14, ns). Post hoc analysis of the main effect revealed that Digit Span forward performance (mean = 9.15, SD = 2.74) was not significantly better than Digit Span backward (mean = 8.08, SD = 3.19), nor was there a difference between Spatial Span forward (mean = 7.57, SD = 3.36) and backward (mean = 7.71, SD = 3.39). The three-way ANOVA with subtest (Digit Span, Spatial Span, and Letter-Number Sequencing) as the within-subject factor resulted in a significant effect (Pillai's Trace F [2,26] = 5.78, p = 0.008). Post hoc analysis of the main effect (F tests) revealed that performance on Digit Span was significantly different from that on Spatial Span (F[l,27] = 6.45, p = 0.01) and from Letter-Number Sequencing (F[l,27] = Table 1. Schizophrenia patients' (n = 28) performance on five measures of working memory compared with the standardization sample (Study 2) Measure Mean (SD) nest p value Digit Span Digit Span forward Digit Span backward Spatial Span Spatial Span forward Spatial Span backward Letter-Number Sequencing Note.SD = standard deviation. 8.50 (2.94) 9.15(2.74) 8.08 (3.19) 7.18(3.49) 7.57 (3.36) 7.71 (3.39) 7.18 (3.36) -2.70-1.64-3.19-4.28-3.82-3.57-4.45 < 0.012 0.111 < 0.005 < 0.001 < 0.001 < 0.001 < 0.001 162

Working Memory Schizophrenia Bulletin, Vol. 27, No. 1, 2001 9.85, p = 0.008). Spatial Span and Letter-Number Sequencing performance was not significantly different from (F[l,27] = 0, ns) (figure 2). Digit Span performance was significantly better than Spatial Span and Letter-Number Sequencing. The most parsimonious explanation for these results is that Digit Span (particularly Digit Span forward) is an easier task for schizophrenia patients than the other two measures. Finally, Digit Span performance in this study was slightly higher than that observed among schizophrenia patients in Study 1. Collectively the patients performed below expectation when compared to a normal comparison or standardization sample. The slight differences in scores between Studies 1 and 2 highlight the heterogeneous performance of schizophrenia patients on working memory cognitive tasks. Conclusion The results of this study support the hypothesis that schizophrenia patients have impaired performance on working memory measures when compared to a standardization sample. The exception among the measures examined was Digit Span forward, but it has been argued that Digit Span forward is primarily a test of attention and does not tax the working memory system to the same degree as Digit Span backward (Gold et al. 1997). If Digit Span forward is the least taxing of the five measures on the working memory system then this may account for schizophrenia patients' relatively better performance on Digit Span forward. However, there were no statistically significant differences between the five working measures, thus not supporting the hypothesis of a modalityspecific working memory impairment for schizophrenia patients (Park and Holzman 1992) or the hypothesis that schizophrenia patients perform better on forward than backward tasks because of the greater taxing of the working memory system (Goldberg et al. 1993; Pantelis et al. 1997). Study 3 In this study schizophrenia patients were assessed on a battery of tests of executive-functioning working memory measures. Each of the tasks used in this study requires online storage, manipulation, and then retrieval of information. Given the hypothesis that central executive processing and its prefrontal cortex substrate are impaired in Figure 2. Scaled scores for the five working memory measures (Study 2) SCALED SCORES FOR WORKING MEMORY MEASURES FOR SCHIZOPHRENIA PATIENTS 12 10 LJJ OH o o </) Q LLJ _L 8 T 6-4 2 - m. 0 Digrt Span Forward Digit Span Backward Spatial Span Forward 163 Spatial Span Backward Letter-Number Sequencing

Schizophrenia Bulletin, Vol. 27, No. 1, 2001 W. Perry et al. schizophrenia patients (Gold et al. 1997), we predicted that schizophrenia patients would perform significantly more poorly than a sample of age-, education-, and gender-matched normal subjects. We further hypothesized that, given that tests of executive functioning are mediated by central executive processing, there should be a moderate relationship between performance across the more complex working memory measures for both schizophrenia and normal comparison subjects. In other words, if the central executive functioning of working memory is the cognitive domain that is impaired in schizophrenia, then the commonality should be expressed by a higher degree of shared variance than that of normal comparison subjects (Gold et al. 1997). Method Forty schizophrenia patients (32 males and 8 females; 35 Caucasian, 2 African-American, 2 Hispanic, 1 Asian) who were diagnosed using the SCID-III-R were age, education, and gender matched to 40 normal comparison subjects (32 males and 8 females; 31 Caucasian, 8 African- American, 1 Hispanic); again, none of these subjects were participants in any of the other studies presented here. The SCID-III-R was administered by an experienced doctoral-level clinician. All diagnostic and exclusion criteria are as described above in Study 1. The schizophrenia patients consisted of the following subtypes: 32 paranoid, 3 undifferentiated, 3 disorganized, and 2 residual. Nineteen were recruited from an acute inpatient psychiatric service, 11 were recruited from a long-term subacute inpatient facility, and 10 were outpatients. Twenty-nine of the schizophrenia patients were being treated with neuroleptic medication (mean chlorpromazine equivalents = 1427.5 mg, SD = 122.6 mg). The 10 outpatients were all medication-free at the time of testing. None of the patients were treated with an atypical neuroleptic. The mean Brief Psychiatric Rating Scale (Overall and Gorham 1962) score for the patients was 26.5 (SD = 8.6). The normal comparison subjects were recruited from newspaper advertisements and were screened for the presence of Axis I pathology, active substance abuse or dependence, an unstable medical condition, or a history of neurological disorder (e.g., head injury with loss of consciousness). Potential comparison subjects were excluded if they met criteria of an algorithm based on the Minnesota Multiphasic Personality Inventory (MMPI) for identifying substance abuse and psychosis proneness because people who meet the criteria have been shown to deviate from normal performance on neuropsychological measures (Butler et al. 1993). The mean age for the schizophrenia sample was 30.4 years (SD - 7.4) with an average of 12.1 years of education (SD = 1.4), whereas the normal comparison group had a mean age of 33.5 years (SD = 8.3) and averaged 12.0 years of education (SD = 1.7). The two groups were not statistically different for age (?[78] = 1.75) or education (<[78] = -0.45). The mean age of onset of illness for the schizophrenia patients was 21.7 years (SD = 5.9). After written informed consent was obtained, subjects were tested on the WAIS-R Vocabulary test and the following neuropsychological tests of executive and working memory functioning: the WCST (Heaton et al. 1993); Auditory Consonant Trigrams (ACT), which is a variation of the "Peterson Task" (Stuss et al. 1985); and the Category Test (Reitan and Wolfson 1993). The WCST is a sorting test of abstract reasoning and cognitive flexibility, but it also requires some working memory (see below). The subject is given two decks of 64 cards each. The cards are printed with one to four different symbols (triangle, star, cross, and circle) and in one of four different colors (red, green, yellow, blue). The subject's task is to place the cards, one by one, under one of four different stimulus cards according to an undisclosed principle. The stimulus cards also contain symbols that differ according to number, shape, and color. The examiner informs the subject after each sort whether his or her placement is "right" or "wrong." The subject must deduce the principle based upon the feedback given by the examiner. After a run of ten consecutive correct placements, the underlying principle changes without that being disclosed to the subject. The test is concluded once a subject completes six correct runs of ten correct placements or has exhausted all of the cards. Gold and colleagues (1997) have stated that "successful WCST performance requires the subject to remember his or her prior response and associated feedback and to use this information to select a new response, a form of working memory" (p. 159). A number of different scores can be derived, including the number of perseverative and nonperseverative errors, categories completed, and failure to maintain set. The perseverative response score, which is very highly correlated with perseverative errors, offers useful information regarding concept formation, profiting from feedback, and cognitive flexibility (Heaton et al. 1993). In the ACT the subject is presented with a set of three consonant letters followed by a number. The subject is then asked to recall the letters after four different time delays: 0, 3, 9, and 18 seconds later. During the delay period the subject must count backward out loud from the number provided until the delay period is over. The delay periods are presented in a pseudo-random order. Following the delay the subject is asked to recall the three consonants. There are five trials of each of the four different time periods, for a total of 20 trials. The score is the total number of letters correctly recalled. To successfully complete the task, the subject must hold the presented 164

Working Memory Schizophrenia Bulletin, Vol. 27, No. 1, 2001 information in short-term storage while processing auditory information. Stuss and associates (1985) described the ACT as a prefrontal measure of divided attention, and Boone and colleagues (1998) found the test to load on a factor of attention and short-term memory, two aspects of working memory performance. The Category Test is another test of abstraction, but it also requires sustained attention and short-term memory applied to logic reasoning. As in the WCST, subjects can alter their performance based upon feedback. Two hundred and eight arrays are projected on a screen, and each of these items requires a conceptually based response. The 208 trials are organized into six groups or subtests based upon different underlying principles. The seventh subtest is composed of stimuli the subject has been previously exposed to in the test and examines the subject's recall. The total number of errors is calculated across the seven subtests. Performances for schizophrenia patients and normal comparison subjects were compared using t tests, and the alpha criterion was set at p < 0.025 to protect against Type I error. To address the question of whether the interrelationship of executive-function working memory tests is different for schizophrenia patients than for normal comparison subjects, Pearson correlation matrices were constructed. The alpha level was conservatively set at p < 0.025 because of the number of analyses conducted. Finally, the correlation coefficients obtained for schizophrenia patients and normal comparison subjects for each of the three executive functioning tests were compared using the z test of differences between two independent correlation coefficients (Ferguson 1966). This set of analyses allowed us to determine whether the shared variance between the schizophrenia patients' performance for each of the three tests of executive-functioning working memory was greater than that of the matched normal comparison subjects, as suggested by Gold and colleagues (Gold et al. 1997). All analyses were conducted with BMDP (Dixon 1988) 3D (t tests), 3S (nonparametric statistics), and 6R (partial correlation and multiple regression). Results WAIS-R Vocabulary subtest scores were converted to age-corrected scaled scores (Wechsler 1981). Schizophrenia patients achieved a mean Vocabulary scaled score of 9.6 (SD = 2.8) compared to 9.6 (SD = 2.3) for the normal comparison subjects (/[78] = -0.04, ns). There was a trend for schizophrenia patients to commit significantly more perseverative errors on the WCST compared to the normal comparison subjects (f[78] = -2.12, p - 0.03). Schizophrenia patients completed significantly fewer WCST categories compared to the normal comparison subjects (f[78] = 2.23, p = 0.02). On the ACT and the Category Test, schizophrenia patients and normal comparison subjects did not significantly differ in their performance (ACT: t[ls] = 1.86, ns; Category Test errors (/[78] = 0.36, ns) (table 2). Two Pearson correlation matrices were constructed, one for the schizophrenia patients and the other for the normal comparison group (see tables 3A and 3B). For the schizophrenia patients, there was a moderate correlation between WCST and Category Test performance, sharing about 25 percent of the total variance. The relationship between the number of items recalled on the ACT and the perseverative responses and categories achieved on the WCST was in the expected direction but did not reach significance (see table 3A). For the normal comparison subjects, WCST, Category Test, and ACT performances were all weakly to moderately correlated with each other. The z test for the difference between two independent correlation coefficients revealed that there were no significant differences in the correlation coefficients between schizophrenia patients and normal comparison subjects on any of the three executive-functioning working memory measures. Conclusion Our first hypothesis was not supported because schizophrenia patients did not perform significantly more poorly than did normal comparison subjects on two of the three measures of executive-functioning working memory and performed identically on an estimate of gen- Table 2. Means and standard deviations for executive-functioning working memory measures for schizophrenia patients and normal comparison subjects (Study 3) Schizophrenia patients (n = 40), mean (SD) Normal comparison subjects (n 40), mean (SD) WCST (PR) WCST (Cat) ACT Category Test (errors) 26.6(19.9) 4.4(1.8) 37.8 (9.8) 66.3 (30.5) 17.0(20.5) 5.2(1.6) 41.5 (7.7) 68.8(32.1) Note.ACT = Auditory Consonant Trigrams; Cat = categories completed; PR = perseverative response score; SD = standard deviation; WCST = Wisconsin Card Sorting Test. 165

Schizophrenia Bulletin, Vol. 27, No. 1, 2001 W. Perry et al. Table 3A. Pearson correlation matrix for performance on tests of executive-functioning working memory for schizophrenia patients (n = 40) (Study 3) WCST (PR) WCST (Cat) ACT (items recalled) Category Test (errors) WCST (PR) WCST (Cat) 0.53* -0.54* -0.15-0.73* -0.26 0.25 Note.ACT = Auditory Consonant Trigrams; Cat = categories completed; PR = perseverative response score; WCST = Wisconsin Card Sorting Test. *p< 0.005 Table 3B. Pearson correlation matrix for performance on tests of executive-functioning working memory for normal comparison subjects (n = 40) (Study 3) Category Test (errors) WCST (PR) WCST (Cat) WCST (PR) WCST (Cat) ACT (items recalled) 0.29-0.21-0.33-0.76* -0.25 0.22 Note.ACT = Auditory Consonant Trigrams; Cat = categories completed; PR = perseverative response score; WCST = Wisconsin Card Sorting Test. *p< 0.005 eral intelligence (i.e., Vocabulary). In fact there were no significant differences between the two groups on the ACT, which among the three tests may be the "purest" working memory measure because it is a test of holding information in transient storage while processing an auditory/verbal task. Similarly, there were no significant differences between the two groups on the Category Test, which is the "purest" abstract reasoning measure among the three tests. This nonsignificant difference in performance between the schizophrenia patients and normal comparison subjects was unexpected and is inconsistent with previous findings (Braff et al. 1991; Goldstein and Shemansky 1997) and raises questions about the generalizability of the current results. The patients did, however, perform in the mildly impaired range on the WCST (mean t score for WCST perseverative responses = 42) and the Category Test (mean t score for number of Category Test errors = 39, Heaton et al. 1991), arguing against these being a "neuropsychologically normal" group of patients. In respect to the Category Test, an alternative explanation for the lack of differences between schizophrenia patients and normal comparison subjects may involve the relatively limited education and, by normative standards, poor Category Test scores of this group of normals. Indeed, these apparently normal subjects also performed in the mildly impaired range on the Category Test, using the Heaton et al. (1991) demographically corrected norms (see t scores above). The present findings will clearly need to be replicated in other ageand education-matched samples. The relationships among performances on these tests of executive-functioning working memory were weak to moderate within both groups. This association among the three tests of executive functioning for the schizophrenia patients and normal comparison subjects may represent the degree of a common taxing of the central executive processing component of working memory across the three measures. However, the lowest correlation coefficients were observed between the ACT and the other two measures, arguing against a working memory explanation at the "core" of performance across these three tests. Additionally, the correlation coefficients for performance on the three tests were not different for the schizophrenia patients and the normal comparison subjects. Thus, the hypothesis of an expected "amplification" of correlation coefficients among the schizophrenia sample because of a hypothetically impaired working memory system was not supported. Still, it cannot be assumed that the nonsignificant differences in correlation coefficients between the schizophrenia patients and normal comparison subjects are the result of common taxing of the same working memory functions. Future studies will need to tease apart the various task components of working memory (e.g., atten- 166

Working Memory Schizophrenia Bulletin, Vol. 27, No. 1, 2001 tion, memory capacity) to more fully assess the similarities and differences between the groups on executivefunctioning working memory test performance. Study 4 In this study we assessed schizophrenia patients and normal control subjects on a somewhat different group of tests of executive-functioning working memory, hypothesizing that the schizophrenia patients would show significant deficits on these measures. We next examined the relationship among the executive-functioning working memory measures for both schizophrenia patients and normal control subjects. We predicted that the measures would be moderately correlated with each other. We then statistically controlled for verbal intelligence and for sustained attention capacity while correlating performance among the three measures for the schizophrenia patients. We hypothesized that, given that tests of executive functioning are mediated by central executive processing, there should be moderate relationships among performances across working memory measures for the schizophrenia subjects that are not fully explained by intellectual and attentional functioning. We next correlated the patients' performance on these tests of executive-functioning working memory with symptom factor scores. Given the suggestion that working memory performance and negative may both reflect deficits in the prefrontal cortex, we hypothesized that there should be a significant correlation between test performance and negative even after controlling for possible moderator variables such as intellectual functioning and sustained attention. Method Thirty-seven schizophrenia patients (21 males and 16 females; 22 Caucasian, 8 African-American, 4 Hispanic, 2 Asian, and 1 other) who met the DSM-IV (American Psychiatric Association 1994) criteria for schizophrenia as determined by the use of the SCID-IV conducted by experienced doctoral-level clinicians were included in the study. All diagnostic and exclusion criteria are as described above in Study 1. The schizophrenia patients consisted of the following subtypes: 18 paranoid, 12 undifferentiated, 3 disorganized, and 4 residual. Fourteen patients were recruited from the UCSD inpatient psychiatric service, 7 were recruited from a long-term subacute inpatient facility, and 16 were outpatients from a community day treatment center. Two patients were unmedicated and the mean chlorpromazine equivalents (Kessler and Waletski 1981) for the remaining subjects was 820 mg (SD = 642 mg). A normal comparison group consisting of 34 subjects (25 males and 9 females; 30 Caucasian, 3 African- American, and 1 other) were compared to the schizophrenia patients. The normal comparison subjects were recruited from newspaper advertisements and were screened for the presence of Axis I pathology, substance abuse or dependence, an unstable medical condition, or a history of neurological disorder (e.g., head injury with loss of consciousness). Potential comparison subjects were excluded if they met criteria of an MMPI-based algorithm for identifying substance abuse and psychosis proneness (Butler et al. 1993). The normal comparison subjects were comparable to schizophrenia patients with respect to age. The mean age for the schizophrenia patients was 39.1 years (SD = 8.7) and for the normal comparison subjects 36.6 years (SD = 9.7). The schizophrenia patients were less educated (mean =12.1 years, SD = 1.7) than the normal comparison subjects (mean = 14.0, SD = 2.4) (f[69] = 3.75, p < 0.001). The mean age of onset of illness for the schizophrenia patients was 22.5 years (SD = 5.9). All patients were assessed with the Scale for the Assessment of Negative Symptoms (SANS; Andreasen 1984) and the Scale for the Assessment of Positive Symptoms (SAPS; Andreasen 1984). The SANS and SAPS scores were converted into three-factor scores, which were generated from factor coefficients derived from a principal components factor analysis on a separate sample of 160 schizophrenia patients. The latter analysis revealed a three-factor solution, consisting of a negative symptom factor, a positive symptom-reality testing factor, and a disorganization factor (Cadenhead et al. 1997). After a complete description of the study was given to the subjects, written informed consent was obtained. Subjects were then tested on the Vocabulary subtest from the WAIS-R, the Numerical Attention Test (NAT; Filley et al. 1989), the WCST (Heaton et al. 1993), the Tower of Hanoi (TOH; Loong 1989), and the LNS (Gold et al. 1997). The NAT is a digit cancellation test used to assess psychomotor speed and attention to visual details. The test consists of a page of 1,008 random digits, and the subject is required to cancel all of the 105 number 3s on the page. The test provides a score for psychomotor speed (total time to completion) as well as total errors of omission and commission (sustained attention to detail). The version of the TOH used in this study is a computer-administered complex problem solving and procedural learning test. Successful completion of the task requires the subject to hold information about his or her previous responses in visuospatial "scratch-pad" transient storage, to use self-ordered and procedural learning skills, and to benefit from feedback. This test requires the use of working memory because subjects are required to simultaneously reason logically, maintain a set of rules, and plan future 167