Schizophruniu Reseurch, 8 (I 992) 143- 156 0 1992 Elsevier Science Publishers B.V. All rights
SCHIZO
143 reserved
0920-9964/92/$05.00
00258
Positive and negative symptoms in the psychoses: Multidimensional scaling of SAPS and SANS items I.H. Minasasbv G.W. StuarPb,
S. Klimidis”sh, H.J. Jackson’,
B.S. Singhb- and D.L. Copolovb
’ Victorran Transcultur-al Psychiatry
Unit, Victor@ Australia; and bTht~ National Health & Medical Research Research Unit, Royal Park Hospital, Parkvillr. Victoria. Australia.
(Received
20 January
1992: revision
received 9 June 1992; accepted
Cowwil Schizophrenia
9 June 1992)
Recently, the validity of the simple dichotomy between positive and negative symptoms in psychosis has been questioned. A newly admitted group of 114 DSM-III patients with psychotic disorder were assessed using Andreasen’s positive and negative symptom scales. Multidimensional scaling, augmented by cluster analysis, was applied to the full item set of these scales and showed clearly that there are three major, Positive Thought Disorder and Negative independent groups of symptoms: Hallucinations/Delusions, Symptoms. Within the Hallucinations/Delusions and Negative Symptoms groups there was some additional structure which does not conform to the SAPS and SANS sub-scales. In particular there was considerable heterogeneity within the Hallucinations/Delusions group, and delusions of persecution may represent a fourth independent dimension of psychopathology which is under-represented in these scales. Ke), ~vordr; Multidimensional
scaling;
Positive
and negative
symptoms;
INTRODUCTION
In recent
years research on schizophrenia has increasingly focused on positive and negative symptoms. The distinction between the two symptom types was initially based on clinical phenomenology. A contrast was made between positive or productive symptoms, such as hallucinations, delusions and positive thought disorder, and negative symptoms which were regarded as deficits in normal functioning, such as affective flattening, volitional impairment and ideational deficits. There has been, however, some disagreement concerning the allocation of individual symptoms to the positive or negative groups (Walker and Lewine, 1988). Initially, interest centred on the issue of whether positive and negative symptoms should be viewed as mutually exclusive or independent syndromes Correspondence Psychiatry Unit, Australia.
to: I.H. Minas, Victorian Transcultural II0 Nicholson St, Fitzroy, Victoria, 3065,
Psychosis;
(Schizophrenia)
(Andreasen and Olsen, 1982; Crow, 1982, 1985; Kay and Opler, 1987) rather than on any possible heterogeneity within the two symptom classes. External validation of the distinction between negative and positive symptoms has come from studies of the relationship between symptoms and medication response (Johnstone et al., 1978; Angrist et al., 1980; Mackay, 1980), longitudinal course (Pfhol and Winokur, 1982) ventricular enlargement (Andreasen and Olsen, 1982; Andreasen et al., 1982) brain metabolism (Franzen and Ingvar, 1975; Ingvar, 198 I; Volkow et al., 1987) and neuropsychological performance (Owens and Johnstone, 1980; Green and Walker, 1985; Kemali et al., 1985). These studies essentially treated the positive and negative symptom complexes as unitary constructs without exploring whether or not particular symptoms (or symptom subgroups) within each symptom complex are specifically associated with external measures. Studies which have further explored relations between external measures and subgroups of symp-
144
toms within the two broader classes have been few. However, they have been suggestive of further valid subdivisions within, in particular, positive symptoms. Bilder et al. (1985) for example, indicated that neuropsychological performance deficits were associated with only some of the positive symptoms-those clustering around the construct of thought disorder (including alogia). Similarly, Liddle (1987a) noted different neuropsychological deficits were associated with different positive symptom complexes-delusions and hallucinations were correlated with figure-ground resolution impairments while a syndrome comprised of positive thought disorder and inappropriate affect was correlated with impairments in visual selection and tracking, visuospatial memory and word recognition memory. Kulkarni et al. (1990) noted a relationship between haloperidol-stimulated prolactin release and the severity of delusions, suggestive of a specific link between delusions and dopaminergic overactivity. The development of reliable scales for the measurement of positive and negative symptoms, such as the Scale for the Assessment of Positive Symptoms (SAPS) and the Scale for the Assessment of Negative Symptoms (SANS) (Andreasen, 1983, 1984) has enabled significant advances in such work. However, the general acceptance of the SAPS and SANS by researchers has somewhat out-paced the comprehensive evaluation of their psychometric properties. In the development of the SAPS and SANS (Andreasen, 1982; Andreasen and Grove, 1986) sub-scales were validated by demonstrating internal consistency within the sub-scales, using Cronbach’s alpha and item-total correlations. Correlations between items from different sub-scales were not reported. When internal consistency techniques were applied to the SANS, it was found that each sub-scale had an acceptably high alpha coefficient, although each item seemed to correlate only slightly better with its own sub-scale score than it did with the total negative symptom score. Andreasen and Grove (1986) interpreted such a slight reduction in correlation as indicative of the validity of the sub-scales, since any reduction in the correlation would indicate that additional variance is explained by the item. The possibility that a negative scale item might correlate more highly with symptoms in
other sub-scales of the SANS, or with symptoms in the SAPS, was not explored. In the case of the SAPS, there was a bigger difference between ‘item to sub-scale’ and ‘item to total score’ correlations (Andreasen and Grove, 1986) indicating that the sub-scales add information to that which may be derived from the total score. However, many items did not correlate well with their own sub-scale scores. For example, all delusions were grouped together regardless of type, but many of these were not highly correlated with the sub-scale (sum of items) score for delusions and were also very poorly correlated with global ratings of delusions. This could be due to the possibility that these items are not measuring the same underlying construct or it may reflect the wide variation in symptom base rates, as reported in Andreasen (1987). It has therefore been difficult to determine the degree to which this internal consistency analysis actually supports the structure of the scales. Given these ambiguities. it has been unclear whether more than two underlying dimensions are measured by the Andreasen scales. A more appropriate way to approach the problem is to examine inter-item correlations for the total item pool. Although Andreasen and Olsen (1982) and Andreasen and Grove (1986) did carry out principal components analyses, leading Andreasen and Olsen (I 982) to suggest that ‘the group of positive symptoms may represent more than one symptom complex’ (p. 793) these were done using the global ratings rather than individual item ratings. The results of these two analyses were inconsistent with each other, although both yielded four factors. Andreasen and Grove (1986) found that the negative symptoms were spread across two separate factors while Andreasen and Olsen (1982) found that all negative symptoms loaded on a single factor together with hallucinations and delusions which were negatively loaded on this factor. Kulhara et al. (1986) also carried out a principal components analysis of the global ratings of the SAPS and SANS and obtained a simpler solution, with all the negative symptoms loading onto a single factor, hallucinations and delusions on a second, and bizarre behavior and formal thought disorder on a third. Andreasen and Grove (1986) also found delusions and hallucinations to load together on a separate factor from the remaining
145
positive symptoms. The structure found by Kulhara et al. (1986) is very similar to that derived by Bilder et al. (1985) Simpson and Davis (1985) Liddle (1987b) and Arndt et al. (199 l), all of whom used exploratory factor analytic rather than confirmatory methods. These studies, which have used a variety of measurement scales and different samples, clearly suggest that the two-syndrome model is an inadequate representation of the range of psychotic symptoms measured by the SAPS and SANS and by other instruments. The aim of this study was to determine the of dimensions number, and characteristics, required to adequately account for the common variance in the SAPS and SANS set of items. No previous study of the SAPS and SANS has examined the dimensionality of the complete set of items. The studies referred to above have generally studied small samples of psychotic patients, have used global SAPS and SANS measures or have used different scales. In the process of exploring this issue the present study examines the adequacy of the subscale composition and structure of SAPS and SANS. It does this by use of correlational methods, since confirmatory methods will always yield some support for an a priori clustering of symptoms unless the pre-determined groupings are completely invalid. The analysis was based on the technique of multidimensional scaling, augmented by cluster analysis of variables. These techniques can be used to explore the structure of a correlation matrix without restricting it to particular clusters of items, nor requiring that these clusters represent independent dimensions (as in the case of orthogonal factor analysis). Multidimensional scaling has the advantage that it is a graphical technique which represents the correlation matrix as a ‘map’ which is then open to interpretation. Positive and negative symptoms are common in psychotic disorders other than schizophrenia (e.g., Pope and Lipinski, 1978; Sommers, 1985) and the boundaries between schizophrenia and other psychotic disorders remain unclear (Andreasen, 1987). In the study of positive and negative symptoms it would seem that a narrow focus on schizophrenia is premature and will perhaps provide only a limited view of the relationship between positive and negative symptoms. This study therefore included a diagnostically heterogeneous sample of psychotic patients.
METHODS Sut7jects and procedure One hundred and fourteen patients with DSM-III (American Psychiatric Association, 1980) diagnoses of psychotic disorder formed the sample. Diagnostic information was elicited using the Royal Park Multidiagnostic Instrument for Psychoses (McGorry et al., 1990a,b). This instrument has been shown to have adequate diagnostic reliability (McGorry et al., 1990b). All subjects, following explanation of the aims and procedures of the study, gave written and witnessed consent to participation. The sample consisted of consecutive admissions to the research ward of a metropolitan psychiatric hospital, with admission conditional upon meeting both inclusion (age between 15 and 45 and clinical presentation suggestive of the presence of psychotic disorder) and exclusion (DSM-III diagnosis of organic mental disorder, mental retardation and inadequate English fluency) criteria. Table 1 presents demographic and illness characteristics of the sample by DSM-III diagnostic category. Interval and ordinal scale variables in Table 1 were analysed using one-way analyses of variance, whilst frequency variables were analysed using the When significant differences were X2 statistic. found, the Scheffe post-hoc procedure (with an alpha of 0.10, Scheffe, 1959) was used to determine which groups significantly differed from the others. These analyses revealed that the groups differed in age, with schizophrenics being significantly older than schizophreniform, unipolar depression and bipolar disorder patients, and ‘other’ psychotics being older than schizophreniform patients [F (5,108) = 5.4, p < O.OOl]. In addition, although the number of previous admissions and age of onset of illness were found to differ between the groups (for each variable, F (5,108) = 2.4, p < 0.05) posthoc analyses failed to reveal any significant pattern. Inspection of the means in Table 1 suggests a tendency for schizophrenics and schizoaffective disorder patients to have more admissions than the rest of the sample. Similar inspection of the means for age of onset of illness suggests a lower age of onset in schizophreniform and unipolar depressed patients in comparison to the rest of the sample. However, the differences are not great, and post-hoc contrasts failed to reveal any pattern
146 TABLE
1
Variahk
n=14
15:8
7:7
I118
24.4
23.4
30.2
21.0
(8.7) 27.9
(5.3) 23.6
(5.4) 22.6
(6.4) 27.7
(7.5)
(3.1)
(8.1)
(5.6)
(5.0)
(6.4)
3.7 (2.3) 2.5 (2.5) 51.3 (34.2)
3.2 (2.2)
3.4
4.3
3.7
3.3
(2.6) 2.4
(1.8) 1.5
(2.5) 1.5
(1.9) 1.3
(1.8) 38.7 (24.3) 18.0 (26.4) 700
(0.9) 55.2 (34.6) 11.7 (33.2) 982
(0.7) 56.2 (31.3) 25.9 (32.3) 954
(0.5) 42.9 (34.6) 29.6 (38.3) 499
4!4
29.1
Onset*
(7.6) 26.2
21.1 (3.2)
Duration
n=23 712
28113
Admissions*
C’nipolur drpressiorz
29.8
Sex (m/f)
Age**
Education
Bipolar disorder
Prodrome**
71.6 (76.6)
Medication
784 (705)
I.1 (0.4) 64.8 (27.8) 1.8 (3.2) 919 (567)
(510)
Onset: Age of illness onset. Education: Education score. Admissions: Number of admissions. Duration: Duration of current episode (data available for n= 108). Prodrome: Prodrome duration in weeks (data available for II = 90). Medication: Highest 24 h level of CPZ units (mg) in episode (data available *p
SCHIZO
143 reserved
0920-9964/92/$05.00
00258
Positive and negative symptoms in the psychoses: Multidimensional scaling of SAPS and SANS items I.H. Minasasbv G.W. StuarPb,
S. Klimidis”sh, H.J. Jackson’,
B.S. Singhb- and D.L. Copolovb
’ Victorran Transcultur-al Psychiatry
Unit, Victor@ Australia; and bTht~ National Health & Medical Research Research Unit, Royal Park Hospital, Parkvillr. Victoria. Australia.
(Received
20 January
1992: revision
received 9 June 1992; accepted
Cowwil Schizophrenia
9 June 1992)
Recently, the validity of the simple dichotomy between positive and negative symptoms in psychosis has been questioned. A newly admitted group of 114 DSM-III patients with psychotic disorder were assessed using Andreasen’s positive and negative symptom scales. Multidimensional scaling, augmented by cluster analysis, was applied to the full item set of these scales and showed clearly that there are three major, Positive Thought Disorder and Negative independent groups of symptoms: Hallucinations/Delusions, Symptoms. Within the Hallucinations/Delusions and Negative Symptoms groups there was some additional structure which does not conform to the SAPS and SANS sub-scales. In particular there was considerable heterogeneity within the Hallucinations/Delusions group, and delusions of persecution may represent a fourth independent dimension of psychopathology which is under-represented in these scales. Ke), ~vordr; Multidimensional
scaling;
Positive
and negative
symptoms;
INTRODUCTION
In recent
years research on schizophrenia has increasingly focused on positive and negative symptoms. The distinction between the two symptom types was initially based on clinical phenomenology. A contrast was made between positive or productive symptoms, such as hallucinations, delusions and positive thought disorder, and negative symptoms which were regarded as deficits in normal functioning, such as affective flattening, volitional impairment and ideational deficits. There has been, however, some disagreement concerning the allocation of individual symptoms to the positive or negative groups (Walker and Lewine, 1988). Initially, interest centred on the issue of whether positive and negative symptoms should be viewed as mutually exclusive or independent syndromes Correspondence Psychiatry Unit, Australia.
to: I.H. Minas, Victorian Transcultural II0 Nicholson St, Fitzroy, Victoria, 3065,
Psychosis;
(Schizophrenia)
(Andreasen and Olsen, 1982; Crow, 1982, 1985; Kay and Opler, 1987) rather than on any possible heterogeneity within the two symptom classes. External validation of the distinction between negative and positive symptoms has come from studies of the relationship between symptoms and medication response (Johnstone et al., 1978; Angrist et al., 1980; Mackay, 1980), longitudinal course (Pfhol and Winokur, 1982) ventricular enlargement (Andreasen and Olsen, 1982; Andreasen et al., 1982) brain metabolism (Franzen and Ingvar, 1975; Ingvar, 198 I; Volkow et al., 1987) and neuropsychological performance (Owens and Johnstone, 1980; Green and Walker, 1985; Kemali et al., 1985). These studies essentially treated the positive and negative symptom complexes as unitary constructs without exploring whether or not particular symptoms (or symptom subgroups) within each symptom complex are specifically associated with external measures. Studies which have further explored relations between external measures and subgroups of symp-
144
toms within the two broader classes have been few. However, they have been suggestive of further valid subdivisions within, in particular, positive symptoms. Bilder et al. (1985) for example, indicated that neuropsychological performance deficits were associated with only some of the positive symptoms-those clustering around the construct of thought disorder (including alogia). Similarly, Liddle (1987a) noted different neuropsychological deficits were associated with different positive symptom complexes-delusions and hallucinations were correlated with figure-ground resolution impairments while a syndrome comprised of positive thought disorder and inappropriate affect was correlated with impairments in visual selection and tracking, visuospatial memory and word recognition memory. Kulkarni et al. (1990) noted a relationship between haloperidol-stimulated prolactin release and the severity of delusions, suggestive of a specific link between delusions and dopaminergic overactivity. The development of reliable scales for the measurement of positive and negative symptoms, such as the Scale for the Assessment of Positive Symptoms (SAPS) and the Scale for the Assessment of Negative Symptoms (SANS) (Andreasen, 1983, 1984) has enabled significant advances in such work. However, the general acceptance of the SAPS and SANS by researchers has somewhat out-paced the comprehensive evaluation of their psychometric properties. In the development of the SAPS and SANS (Andreasen, 1982; Andreasen and Grove, 1986) sub-scales were validated by demonstrating internal consistency within the sub-scales, using Cronbach’s alpha and item-total correlations. Correlations between items from different sub-scales were not reported. When internal consistency techniques were applied to the SANS, it was found that each sub-scale had an acceptably high alpha coefficient, although each item seemed to correlate only slightly better with its own sub-scale score than it did with the total negative symptom score. Andreasen and Grove (1986) interpreted such a slight reduction in correlation as indicative of the validity of the sub-scales, since any reduction in the correlation would indicate that additional variance is explained by the item. The possibility that a negative scale item might correlate more highly with symptoms in
other sub-scales of the SANS, or with symptoms in the SAPS, was not explored. In the case of the SAPS, there was a bigger difference between ‘item to sub-scale’ and ‘item to total score’ correlations (Andreasen and Grove, 1986) indicating that the sub-scales add information to that which may be derived from the total score. However, many items did not correlate well with their own sub-scale scores. For example, all delusions were grouped together regardless of type, but many of these were not highly correlated with the sub-scale (sum of items) score for delusions and were also very poorly correlated with global ratings of delusions. This could be due to the possibility that these items are not measuring the same underlying construct or it may reflect the wide variation in symptom base rates, as reported in Andreasen (1987). It has therefore been difficult to determine the degree to which this internal consistency analysis actually supports the structure of the scales. Given these ambiguities. it has been unclear whether more than two underlying dimensions are measured by the Andreasen scales. A more appropriate way to approach the problem is to examine inter-item correlations for the total item pool. Although Andreasen and Olsen (1982) and Andreasen and Grove (1986) did carry out principal components analyses, leading Andreasen and Olsen (I 982) to suggest that ‘the group of positive symptoms may represent more than one symptom complex’ (p. 793) these were done using the global ratings rather than individual item ratings. The results of these two analyses were inconsistent with each other, although both yielded four factors. Andreasen and Grove (1986) found that the negative symptoms were spread across two separate factors while Andreasen and Olsen (1982) found that all negative symptoms loaded on a single factor together with hallucinations and delusions which were negatively loaded on this factor. Kulhara et al. (1986) also carried out a principal components analysis of the global ratings of the SAPS and SANS and obtained a simpler solution, with all the negative symptoms loading onto a single factor, hallucinations and delusions on a second, and bizarre behavior and formal thought disorder on a third. Andreasen and Grove (1986) also found delusions and hallucinations to load together on a separate factor from the remaining
145
positive symptoms. The structure found by Kulhara et al. (1986) is very similar to that derived by Bilder et al. (1985) Simpson and Davis (1985) Liddle (1987b) and Arndt et al. (199 l), all of whom used exploratory factor analytic rather than confirmatory methods. These studies, which have used a variety of measurement scales and different samples, clearly suggest that the two-syndrome model is an inadequate representation of the range of psychotic symptoms measured by the SAPS and SANS and by other instruments. The aim of this study was to determine the of dimensions number, and characteristics, required to adequately account for the common variance in the SAPS and SANS set of items. No previous study of the SAPS and SANS has examined the dimensionality of the complete set of items. The studies referred to above have generally studied small samples of psychotic patients, have used global SAPS and SANS measures or have used different scales. In the process of exploring this issue the present study examines the adequacy of the subscale composition and structure of SAPS and SANS. It does this by use of correlational methods, since confirmatory methods will always yield some support for an a priori clustering of symptoms unless the pre-determined groupings are completely invalid. The analysis was based on the technique of multidimensional scaling, augmented by cluster analysis of variables. These techniques can be used to explore the structure of a correlation matrix without restricting it to particular clusters of items, nor requiring that these clusters represent independent dimensions (as in the case of orthogonal factor analysis). Multidimensional scaling has the advantage that it is a graphical technique which represents the correlation matrix as a ‘map’ which is then open to interpretation. Positive and negative symptoms are common in psychotic disorders other than schizophrenia (e.g., Pope and Lipinski, 1978; Sommers, 1985) and the boundaries between schizophrenia and other psychotic disorders remain unclear (Andreasen, 1987). In the study of positive and negative symptoms it would seem that a narrow focus on schizophrenia is premature and will perhaps provide only a limited view of the relationship between positive and negative symptoms. This study therefore included a diagnostically heterogeneous sample of psychotic patients.
METHODS Sut7jects and procedure One hundred and fourteen patients with DSM-III (American Psychiatric Association, 1980) diagnoses of psychotic disorder formed the sample. Diagnostic information was elicited using the Royal Park Multidiagnostic Instrument for Psychoses (McGorry et al., 1990a,b). This instrument has been shown to have adequate diagnostic reliability (McGorry et al., 1990b). All subjects, following explanation of the aims and procedures of the study, gave written and witnessed consent to participation. The sample consisted of consecutive admissions to the research ward of a metropolitan psychiatric hospital, with admission conditional upon meeting both inclusion (age between 15 and 45 and clinical presentation suggestive of the presence of psychotic disorder) and exclusion (DSM-III diagnosis of organic mental disorder, mental retardation and inadequate English fluency) criteria. Table 1 presents demographic and illness characteristics of the sample by DSM-III diagnostic category. Interval and ordinal scale variables in Table 1 were analysed using one-way analyses of variance, whilst frequency variables were analysed using the When significant differences were X2 statistic. found, the Scheffe post-hoc procedure (with an alpha of 0.10, Scheffe, 1959) was used to determine which groups significantly differed from the others. These analyses revealed that the groups differed in age, with schizophrenics being significantly older than schizophreniform, unipolar depression and bipolar disorder patients, and ‘other’ psychotics being older than schizophreniform patients [F (5,108) = 5.4, p < O.OOl]. In addition, although the number of previous admissions and age of onset of illness were found to differ between the groups (for each variable, F (5,108) = 2.4, p < 0.05) posthoc analyses failed to reveal any significant pattern. Inspection of the means in Table 1 suggests a tendency for schizophrenics and schizoaffective disorder patients to have more admissions than the rest of the sample. Similar inspection of the means for age of onset of illness suggests a lower age of onset in schizophreniform and unipolar depressed patients in comparison to the rest of the sample. However, the differences are not great, and post-hoc contrasts failed to reveal any pattern
146 TABLE
1
Variahk
n=14
15:8
7:7
I118
24.4
23.4
30.2
21.0
(8.7) 27.9
(5.3) 23.6
(5.4) 22.6
(6.4) 27.7
(7.5)
(3.1)
(8.1)
(5.6)
(5.0)
(6.4)
3.7 (2.3) 2.5 (2.5) 51.3 (34.2)
3.2 (2.2)
3.4
4.3
3.7
3.3
(2.6) 2.4
(1.8) 1.5
(2.5) 1.5
(1.9) 1.3
(1.8) 38.7 (24.3) 18.0 (26.4) 700
(0.9) 55.2 (34.6) 11.7 (33.2) 982
(0.7) 56.2 (31.3) 25.9 (32.3) 954
(0.5) 42.9 (34.6) 29.6 (38.3) 499
4!4
29.1
Onset*
(7.6) 26.2
21.1 (3.2)
Duration
n=23 712
28113
Admissions*
C’nipolur drpressiorz
29.8
Sex (m/f)
Age**
Education
Bipolar disorder
Prodrome**
71.6 (76.6)
Medication
784 (705)
I.1 (0.4) 64.8 (27.8) 1.8 (3.2) 919 (567)
(510)
Onset: Age of illness onset. Education: Education score. Admissions: Number of admissions. Duration: Duration of current episode (data available for n= 108). Prodrome: Prodrome duration in weeks (data available for II = 90). Medication: Highest 24 h level of CPZ units (mg) in episode (data available *p
