Andreasen NC (ed): Schizophrenia: Positive and Negative Symptoms and Syndromes. Mod Probl Pharmacopsychiatry. Basel, Karger, 1990, vol 24, pp 73-88
Methods for Assessing Positive and Negative Symptoms' Nancy C. Andreasen Department of Psychiatry, University of Iowa College of Medicine, Iowa City, Iowa, USA
This research was supported in part by 111H Grants 1Η31593 and 1 Η40856; MHCRC Grant 1Η43271; the Nellie Ball Trust Research Fund, Iowa State Bank & Trust Company, Trustee, and a Research Scientist Award, 1Η00625.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
When interest reawakened in the positive versus negative distinction, initially adequate instruments were not available with which to measure negative symptoms. Scales such as the Krawiecka [Krawiecka et al., 1982] or the Brief Psychiatric Rating Scale (BPRS) [Overall and Gorham, 1962] were used because they were already in existence. The BPRS in particular had been widely used for many years in clinical drug trials. Neither of these scales was developed specifically for the assessment of positive versus negative symptoms, however, and both have significant limitations. The Krawiecka has not been subjected to rigorous reliability assessment or validation. The BPRS was designed through factor analysis and rate factors rather than symptoms, an approach which sometimes makes it difficult for clinically trained investigators to apply because its level of abstraction is somewhat higher than is actually observed in living patients. While considerable psychometric testing was done with it initially, it has not been rigorously reassessed since the advent of contemporary structured interviews and diagnostic criteria. In this context, it seemed clear that new approaches for the assessment of positive and negative symptoms in schizophrenia were badly needed. This led to the development of two new rating scales which are now widely used, the Scale for the Assessment of Positive Symptoms (SAPS) and the Scale for the Assessment of Negative Symptoms (SANS) [Andreasen, 1983].
Andreasen
74
The decision to develop the SANS and SAPS grew out of a longstanding interest in phenomenology and many years of direct clinical experience in diagnosing and treating patients suffering from schizophrenia who were admitted to the Inpatient Service at the University of Iowa Psychiatric Hospital. It was clear that work was needed in the area of negative symptoms in particular, since relatively good methods for assessing positive symptoms such as delusions and hallucinations were available through the Schedule for Affective Disorders and Schizophrenia (SADS), a widely used interview developed initially for the Collaborative Study of the Psychobiology of Depression [Endicott et al., 1978]. The SANS was the first instrument developed in order to provide for comprehensive assessment of negative symptoms in schizophrenia [Andreasen, 1982, 1983]. It consists of five scales that evaluate five different aspects of negative symptoms: alogia, affective blunting, avolition-apathy, anhedonia-asociality, and attentional impairment. Each of these negative symptoms can be rated globally, but in addition detailed observational measures are made in order to achieve the global rating. It is complemented by a SAPS, which permits detailed evaluation and global ratings of hallucinations, delusions, positive formal thought disorder and bizarre behavior [Andreasen, 1984]. Taken together, the two scales provide a comprehensive set of rating scales in order to measure the symptoms of schizophrenia and to assess their change over time. Repeated close contact with patients suffering from schizophrenia cannot help but lead to a `Bleulerian' belief that delusions and hallucinations, although easier to define reliably, are not the most important, characteristic, or crippling symptoms of schizophrenia. Yet Bleulerian symptoms tended to be neglected by phenomenologists and nosologists because of a concern that they were too imprecise. An initial foray was made on the most important of the Bleulerian symptoms, thought disorder. A scale to assess thought disorder was developed using the empirical assumption that thought is usually inferred from speech, that speech can be observed directly, and that abnormalities in language, speech and communication can be defined precisely through careful attention to grammar, word choice, and links between sentences and clauses. A scale for the assessment of thought, language and communication (TLC) yielded highly reliable definitions, demonstrating that thought disorder was not pathognomonic of schizophrenia, and that it fell
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Development of the SANS and SAPS
Methods for Assessing Positive and Negative Symptoms
75
into two broad groups of subtypes. One group, negative formal thought disorder, was characterized primarily by poverty of speech and poverty of content of speech, while the other, positive formal thought disorder, was characterized by derailments, tangentiality, incoherence and distractible speech. We also observed that negative thought disorder, when present, tended to persist over time and to predict a poor prognosis. Unlike positive thought disorder, it was not common in mania. These findings were confirmed by two independent replications involving several hundred patients and controls over the course of several years [Andreasen, 1979b, c, 1986]. A second foray was made on another major Bleulerian symptom, affective blunting, through the development of an affect rating scale [Andreasen, 1979a]. This study also indicated that affect could be defined reliably using an objective empirical approach, that abnormalities in affect were not specific to schizophrenia either, and yet that affective blunting in schizophrenia could be validated physiologically by measuring voice characteristics such as frequency and amplitude [Andreasen et al., 1981]. These studies confirmed that two major negative symptoms, negative thought disorder and affective blunting, could indeed be defined reliably and that they had a variety of important clinical correlates. Both these negative symptoms, while important, did not provide a complete coverage of the psychopathology frequently observed in patients with schizophrenia. Consequently, a scale was developed to provide a comprehensive coverage of all negative symptoms, adding avolition, anhedonia, and attentional impairment to the two already developed.
The SAPS and the SANS are informed by a series of basic assumptions. First, while the concept of pathognomonic symptoms is psychologically appealing, it has no basis in reality. One simply cannot identify any symptoms that occur only in schizophrenia and not in any other illness. Just as manics and psychotic depressives are likely to have delusions and hallucinations, so too depressives are likely to have some negative symptoms such as alogía or affective blunting. An attempt to define any psychopathological symptom so that it is specific to a single diagnosis is likely to lead to convoluted intellectual contortions or the wearing of blinders.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Basic Assumptions
76
Therefore, the SANS and SAPS were both developed in order to define and describe a wide range of symptoms reliably. They could then be used to observe the frequency of these signs and symptoms in a variety of diagnostic categories and to observe whether they showed different patterns in various diagnostic categories or differences in change over time. A second basic assumption is that good reliability is best achieved through the use of objective observational items. For example, affective blunting is not defined by examining the patient's psychological state. Instead, it is assessed through observation of a series of external behaviors, such as mobility of facial expression, response to a stimulus such as being smiled at, quality of eye contact, use of expressive gestures, etc. As is discussed in more detail below, the fundamental soundness of this strategy has been confirmed through the repeated achievement of good reliability in a wide range of cultural settings. A third assumption is that rating scales must build on cross-sectional evaluation. They must be based on phenomena that one can observe clinically at a specified point in time, which may be an hour, a day, a week, or the past month. The particular cross-sectional window will depend on the specific goal or purpose of a study. Studies that attempt to build toward the development of diagnostic criteria may use a maximally large window (e.g., one month), while those attempting to observe change over time (as in treatment studies) may select a smaller window. The repeated use of rating scales over time can be used to build longitudinal definitions through repeated measurements and the observation of change. Rating scales cannot identify `enduring' symptoms reliably during a single evaluation, and to attempt to do so retrospectively is risky. Consequently, if one wishes to identify enduring negative symptoms, one must do so empirically through prospective longitudinal assessments. Fourth, ideally symptoms should be defined in such a way that their underlying neural mechanisms could be identified. All the various signs and symptoms of schizophrenia must ultimately reflect neural activity in the brain, and understanding the presence or absence of symptoms in terms of such neural mechanisms is one major long-term goal of phenomenological research. Since we do not yet know or understand these neural events, however, the best strategy is to maintain a comprehensive and flexible data base, and an open mind as well. Fifth, ideally rating scales should not sacrifice comprehensive coverage for simplicity, nor should they use premature foreclosure concerning the interrelationships between signs and symptoms. While it is tempting to
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Andreasen
Methods for Assessing Positive and Negative Symptoms
77
use a scale that only involves one or two items and can be completed in two or three minutes, this is likely to be penny wise and pound foolish if one wishes to understand either the neural mechanisms or the longitudinal evolution of symptoms over time. Detailed comprehensive coverage that stays close to direct clinical data is the best strategy. While factor analytic techniques can be used to reduce data sets (and this may be useful for statistical purposes), nevertheless the basic data that are reduced statistically should be maintained intact and reexamined periodically. It could be argued, for example, that since negative symptoms tend to be highly intercorrelated, only one need be rated in order to save time. Yet factor analytic studies tend to be notoriously unstable (í.e., dependent upon the population sampled). Thus, if only one negative symptom were rated based on the reasoning that negative symptoms are highly intercorrelated, not only might one have invalid data, but one might fail to observe (for example) that avolition responds to treatment in some particular patients while affective blunting does not. Sixth, ideally rating scales designed to assess psychopathology should be sensitive to change, since one major use is to evaluate whether various signs and symptoms respond to treatment. The identification of `enduring' or `core' or `primary' symptoms is best achieved through repeated assessments, not through developing rating scales likely to be insensitive to change. Clinicians and investigators must also recognize that negative symptoms may be `enduring' in some patients and `changing' in others; thus the whole issue of stability and prognostic significance must be studied empirically and prospectively.
The original components of the SANS and SAPS were subjected to careful scrutiny as to their reliability and were found to be highly reliable [Andreasen, 1979a—c]. After the SANS and SAPS themselves were completed, similar reliability studies were done and supported the reliability of the entire scales [Andreasen, 1982]. The reliability studies at the University of Iowa were quickly followed by reliability studies in a variety of international settings. The SANS and SAPS clearly fulfilled an important clinical and research need, for they have now been translated into a variety of languages, including Japanese, Spanish, Italian, French, German, Dutch, Korean,
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Reliability
Table 1. Inter-rater reliability of negative and positive symptoms in different cultural settings Intraclass R (weighted K) Italy
Spain
Japan
China
Poverty of speech Poverty of content of speech Blocking Increased latency of response Subjective rating of alogía Global rating of alogía Subscale score
0.632 0.615 0.269 0.803
0.870 0.604 0.935 0.571 0.907 -0.014 0.939 0.591 0.802 0.864 0.945 0.628 0.971
0.801 0.771 0.825 0.824 0.837 0.989
Unchanging facial expression Decreased spontaneous movements Paucity of expressive gestures Poor eye contact Affective nonresponsivity Inappropriate affect Lack of vocal inflections Subjective rating of affective flattening Global rating of affective flattening Subscale score
0.786 0.782 0.757 0.873 0.714 0.774 0.827
0.930 0.940 0.886 0.897 0.774 0.805 0.963 0.831 0.844 0.926
0.805 0.728 0.671 0.676 0.641 0.294 0.720 0.553 0.721
0.847 0.835 0.772 0.722 0.787 0.034 0.835 0.581 0.903
0.759 0.734 0.843 0.964 0.860 0.942
0.744 0.752 0.513 0.607 0.749
0.516 0.732 0.735 0.756 0.817
0.875 0.790 0.616 0.768 0.841 0.769 0.810
0.610 0.742 0.552 0.642 0.823 0.725
0.761 0.712 0.709 0.790 0.798 0.864
0.870 0.937 0.857 0.892 0.938
0.711 0.987 0.462 0.788
0.550 0.925 0.672 0.832
logia
Affective flattening
Avolitionapathy
Anhedoniaasociality
Attentional impairment
0.694
0.688
Grooming and hygiene Impersistence at work or school Physical anergía Subjective complaints of avolition-apathy Global rating of avolition-apathy Subscale score
0.624 0.715 0.755
Recreational interests and activities Sexual interest and activity Ability to feel intimacy and closeness Relationships with friends and peers Subjective awareness of anhedonia-asociality Global rating of anhedonia-asociality Subscale score
0.741 0.605 0.587 0.620
Work inattentiveness Inattentiveness during mental testing Subjective complaints of inattentiveness Global rating of inattentiveness Subscale score
0.747
0.728 0.321 0.854 0.659
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Negative symptoms
Table I (continued) Intraclass R (weighted K) Italy
Spain
Japan
0.942 0.944 0.896 1.000 0.948 0.712
0.953 0.876 0.889 0.827 0.861 0.701 0.870 0.927 0.895
0.904 0.862 0.013 0.451 0.726 0.713 0.304 0.837
China
Positive symptoms Auditory hallucinations Voices commenting Voices conversing Somatic or tactile hallucinations Olfactory hallucinations Visual hallucinations Subjective rating of hallucinations Global rating of hallucinations Subscale score
0.862
Persecutory delusions Delusions of jealousy Delusions of guilt or sin Grandiose delusions Religious delusions Somatic delusions Delusions of reference Delusions of being controlled Delusions of mind-reading Thought broadcasting Thought insertion Thought withdrawal Subjective rating of delusions Global rating of delusions Subscale score
0.906 0.357 0.600 0.936 0.886 0.738 0.768 0.889 0.890 0.664 0.877 0.828
Clothing and appearance Social and sexual behavior Aggressive and agitated behavior Repetitive or stereotyped behavior Subjective rating of bizarre behavior Global rating of bizarre behavior Subscale score
0.796 0.734 0.750 0.849
Positive formal Derailment thought disorder Tangentiality Incoherence Illogicality Circumstantiality Pressure of speech Distractible speech Clanging Subjective rating of positive formal thought disorder Global rating of positive formal thought disorder Subscale score
0.855 0.693 0.814 0.492 0.525 0.761 0.637 0.441 0.870 0.818
Delusions
Bizarre behavior
0.878
0.834
0.964 0.934 0.789 0.307 0.768 0.870 0.752 0.782 0.673 0.954 0.860 0.906 0.826 0.776 0.869 0.914 0.678 0.864 0.698 0.879 0.839 -0.16 0.854 0.901 0.954 0.232 0.896 0.420 0.893 0.775 0.665 0.777 0.849 0.884 0.867 0.914
0.920 0.235 0.751 0.657 0.745 0.993
0.766 0.599 0.666 0.836 0.560 0.617 0.844 0.564
0.825 0.727 0.807 0.159 0.423 0.363 0.757 0.858
0.989 0.881
0.942 Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Hallucinations
Andreasen
80
and Chinese. Reliability data have been reported in studies in Japan, Spain, Italy, and China [Humbert et al., 1986; Moscarelli et al., 1987; Ohta et al., 1984; Phillips, 1987/88]. The reliability data conducted in these various international studies are summarized in table 1. As this table indicates, the reliability is consistently high in a variety of cultural settings. The results indicate the soundness of the basic strategy of using observational rather than subjective evaluation. Ideally, cross-national studies would now be appropriate. These are underway in Spain and Italy, but not as yet complete. Even when observational items are stressed, good reliability is difficult to achieve without adequate training. All too frequently, once standardized rating scales become available, clinicians and investigators simply take them `off the shelf and begin to use them, assuming that reliability data already published provide sufficient evidence that the scales will be reliable in any setting. This simply is not true. Reliability needs to be demonstrated by comparisons of ratings obtained within the individual center, preferably with reference to some external calibration such as videotapes of structured interviews with standardized calibration ratings. Further, reliability needs to be rechecked at regular intervals, particularly as the raters are trained, in order to prevent `rater drift'. A comprehensive set of training materials has been developed for the SANS and SAPS involving both videotapes and case vignettes. These have been widely used in order to insure that these two scales are being used consistently within the US, and they are now also available in international video format as well. In addition, investigators who wish additional training have received it on site at the University of Iowa. Our experience in such training indicates that cross-national reliability is likely to be good when such studies are conducted, since investigators from other cultures learn to calibrate against our standards quite rapidly.
In a psychometric sense, a scale is often considered to be validated when good internal consistency is demonstrated. Measures of internal consistency, such as coefficient alpha (Cronbach's alpha), provide an indication of the extent to which items on a scale are related to one another or internally coherent. In our initial work in developing the SANS and SAPS, we assessed internal consistency and found it to be relatively high for the
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Internal Consistency
81
SANS and less so for the SAPS (for reasons discussed below). We have now replicated our original work in a second sample of 117 schizophrenic patients consecutively admitted to the University of Iowa Psychiatric Hospital between 1982 and 1985. The results of this second study appear in tables 2 and 3. As table 2 indicates, this second replication study has produced findings quite similar to those observed in the first study. The results are summarized in four columns. The first column shows the correlation between individual items and the sum of the individual scales (e.g., alogia, affective flattening, etc.). The second column shows the correlation between each individual item and the global rating on the subscale. The third column indicates the correlation between individual items and the sum of all negative symptom ratings, while the fourth column indicates the correlation between individual items and the sum of all global ratings for each of the five subscales. Coefficient alpha is also listed for each of the subscales. As table 2 indicates, coefficient alpha is consistently high, indicating that each of the five subscales does have good internal consistency. These results are most interesting, however, as a way of determining which items on the individual subscales are most useful in making the ratings. For example, blocking is less highly correlated with the sum of the alogia ratings than are the other symptoms used as indices of alogia, and it is poorly correlated with the global rating of alogia. Inappropriate affect is poorly correlated with more general indices of affective flattening and with indices of negative symptoms in general. Because this correlation is so poor, we have dropped inappropriate affect as a measure of negative symptoms and now consider it to be a positive symptom. Correlations between individual items and the sum of global ratings are relatively high for the other negative symptom subscales. As would be expected, correlations between individual items and indices of negative symptoms in general are lower than they are between subscale measures (i.e., sum of global ratings versus sum of sums and sum of globals). This suggests that individual items are indeed tapping something other than a global construct. Table 3 presents similar analyses for the four positive symptoms measured on the SAPS. Again, coefficient alpha is relatively high for all four scales. On the other hand, however, when we look at individual correlations, we see that some measures are much less correlated either with the subscale as a whole or with positive symptoms generally. For example, somatic and olfactory hallucinations are relatively poorly correlated with other hallucinations and with other positive symptoms generally. Many of
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Methods for Assessing Positive and Negative Symptoms
Andreasen
82
Sum Alogia Poverty of speech Poverty of content Blocking Latency Affective flattening/blunting Facial expression Spontaneous movements Expressive gestures Eye contact Nonresponsivity Inappropriate affect Vocal inflections Avolition Grooming Impersistence Anergía Anhedonia Recreation Sex Intimacy Friends Attention Social attentiveness Testing
Global rating
Sum of sums
Sum of globals
0.83263 0.82033 0.78829 0.62819 0.56366 0.42247 0.50079 0.38229 0.34585 0.76392 0.52728 0.58831 Coefficient alpha = 0.627793
0.73951 0.44390 0.33755 0.52723
0.78429 0.79324 0.68009 0.71034 0.50865 0.59758 0.72611 0.84293 0.71649 0.82159 0.68250 0.75209 0.82605 0.67973 0.78125 0.28158 0.30487 0.30387 0.74257 0.62576 0.68271 Coefficient alpha = 0.834062
0.63846 0.45773 0.62932 0.66871 0.70659 0.36616 0.59857
0.80688 0.67532 0.67447 0.82847 0.86743 0.60619 0.80485 0.63646 0.74183 Coefficient alpha = 0.743490
0.68853 0.69674 0.68084
0.77886 0.72715 0.72790 0.50690 0.54374 0.77713 0.78508 0.58966 0.54512 0.75129 0.83034 0.59898 Coefficient alpha = 0.773762
0.67102 0.43175 0.51570 0.61153
0.85835 0.77427 0.89796 0.89100 0.79725 0.63987 Coefficient alpha = 0.749987
0.75861 0.63263
the different types of delusions are also relatively poorly correlated. Some of these low correlations are in fact somewhat surprising (e.g., jealousy, guilt, somatic). In spite of the fact that coefficient alpha for delusions is relatively high (0.655), the correlations between individual items in either the sum or the global rating suggest that the items on the scale are relatively diverse.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Table 2. Negative symptoms: correlation coefficients for individual subscale items and general subscale or negative symptom measures
83
Methods for Assessing Positive and Negative Symptoms
Table 3. Positive symptoms: correlation coefficients for individual subscale items and general subscale or positive symptom measures
Hallucinations Auditory hallucinations Voices commenting Voices conversing Somatic hallucinations Olfactory hallucinations Visual hallucinations Delusions Persecutory Jealousy Guilt Grandiose Religious Somatic Reference Control Mind-reading Thought broadcasting Thought insertion Thought withdrawal Positive formal thought disorder Derailment Tangentiality Incoherence Illogicality Circumstantiality Pressure of speech Distractible speech Clanging Bizarre behavior Clothing, appearance Social/sexual behavior Aggressive/agitated behavior Repetitive/stereotyped behavior
Global rating
Sum of sums
Sum of globals
0.62684 0.82945 0.90973 0.74716 0.55985 0.48682 0.47539 0.77000 0.53270 0.48082 0.37065 0.40481 0.291117 0.13722 0.09370 0.72346 0.52633 0.52325 Coefficient alpha = 0.748805
0.68311 0.43489 0.36278 0.26822 0.01133 0.39971
0.51214 0.60499 0.66331 0.03231 0.17554 0.07629 0.30080 0.15573 0.23627 0.45880 0.32532 0.34353 0.32747 0.42176 0.31415 0.25023 0.26414 0.16695 0.34638 0.48527 0.40425 0.32417 0.36228 0.48884 0.68680 0.36269 0.52982 0.33834 0.53096 0.19104 0.25229 0.50392 0.19826 0.32530 0.49835 0.17302 Coefficient alpha = 0.655092
0.44891 -0.04645 0.15853 0.19167 0.18964 0.15230 0.21074 0.25173 0.23339 0.11098 0.09311 0.08268
0.74280 0.71285 0.36224 0.78277 0.84183 0.38356 0.43346 0.68977 0.61321 0.41637 0.55435 0.53092 0.08054 0.46323 0.43033 0.45470 0.19960 0.59060 0.48447 0.41762 0.38152 0.14562 0.08241 -0.09033 Coefficient alpha = 0.737787
0.41171 0.42318 0.48814 0.43222 0.00539 0.17120 0.38159 -0.11172
0.33979 0.83352 0.75213 0.79734 0.75098 0.29098 0.75540 0.68043 0.28380 0.75266 0.60829 0.35674 Coefficient alpha = 0.787094
0.47468 0.41491 0.44797 0.50882 Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Sum
84
Coefficient alpha must always be interpreted cautiously as an index of internal consistency because it is partially tied to a number of items on the scale. The larger the number of items, the higher the coefficient alpha, due to the inherent properties of the statistic. Thus a coefficient alpha of 0.74 for a three-item scale such as volition suggests very strong internal consistency in comparison with a coefficient alpha of 0.65 for a twelve-item scale measuring delusions. Clinical common sense tells us that many patients with delusions typically have only one or two types of delusions, and therefore one would not necessarily expect a high correlation between all items on this scale. Among the positive thought disorder items, clanging is clearly weak, while the indices of bizarre behavior are relatively more internally consistent. Tables 4 and 5 provide another perspective on the internal consistency of the positive and negative symptom rating scales. Table 4 focuses on negative symptoms. It indicates what the coefficient alpha is if both the row and column items are deleted, thereby providing an index of how much each of the individual subscales contributes to the internal consistency of the negative symptoms scale as a whole. (Coefficient alpha for the negative symptoms scale as a whole is 0.855.) For example, if affective flattening is dropped from the scale, coefficient alpha remains quite high at 0.815970. If both affective flattening and alogia are deleted, coefficient alpha drops to 0.576515. Since coefficient alpha tends to remain high no matter which subscale is dropped, one must conclude that the items on the negative symptoms scale are highly integrated to one another. Table 5 shows similar analyses for the positive symptoms scale. (Coefficient alpha for this scale is 0.483.) These results suggest that positive symptoms are much less highly correlated with one another than are negative symptoms. Examination of the contribution of individual subscales to the scale as a whole further confirms this impression. These symptoms have the strongest correlation with the scale as a whole, maintaining coefficient alpha at 0.4483 when delusions are deleted. Thought disorder also has a strong correlation. On the other hand, when more than one global rating is deleted, coefficient alpha drops markedly and even becomes negative. It seems clear from these results that positive symptoms are less internally cohesive than negative symptoms. An alternate way of looking at the interrelationship between items is through principal components analysis. These data are presented in table 6. In our previous study [Andreasen, 1982], principal components analysis yielded a large bipolar first factor which had strong positive
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Andreasen
85
Methods for Assessing Positive and Negative Symptoms Table 4. Coefficient alpha: negative symptoms scale Affective Alogía flattening Affective flattening Alogia Avolition Anhedonia Attention
0.815970 0.576515 0.630511 0.706086 0.573116
0.824333 0.582466 0.678863 0.494925
Avolition Anhedonia Attention
0.820781 0.712406 0.578984
0.846275 0.000000
0.822377
Table shows what the coefficient alpha is if both the row and the column items are deleted. Coefficient alpha for the global ratings of the negative symptoms scale = 0.855.
Table 5. Coefficient alpha: positive symptoms scale Global Global Global Global halluci- delusion bizarre thought behavior disorder nation Global hallucination Global delusion Global bizarre behavior Global positive thought disorder
0.336785 —0.227464 —0.829674 —0.075846
0.448318 0.051725 0.336871
0.359254 0.000000
0.457410
weightings on negative symptoms and strong negative weightings on positive symptoms; this factor accounted for 37% of the variance. In the second replication study on a somewhat larger sample, the results of principal components analysis are slightly different, although not strikingly inconsistent. As table 6 indicates, factor 1 is heavily loaded on most negative symptoms with essentially zero loadings on positive symptoms. This factor accounts for 17.5% of the variance and taps one aspect of negative symptoms. Factor 2 has strong loadings on three additional negative symptoms (affective flattening, anhedonia, and avolition) and therefore also taps some aspect of negative symptoms. The first factor might be considered to
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Table shows what the coefficient alpha is if both the row and the column items are deleted. Coefficient alpha for the global ratings of the positive symptoms scale = 0.483.
86
Andreasen
Table 6. Principal components analysis: positive and negative symptom global ratings Principal component 1 Attention Alogia Affective flattening Anhedonia Avolition Delusions Hallucinations Bizarre behavior Positive formal thought disorder
0.77161 0.71200 0.52196 0.29385 0.49270 0.01357 0.08088 0.09442 0.16183
Variance explained by each principal component
17.5%
2
3
0.24769 0.26149 0.48305 0.72262 0.55920 -0.00034 0.17786 0.24117 -0.03711
0.10686 -0.5981 0.12239 0.10396 0.10968 0.75342 0.45636 0.07425 0.08634
12.9%
8.4%
4 0.11430 0.34775 0.19889 0.07155 0.14602 0.00342 0.20688 0.54808 0.05546 8.0%
be assessing more cognitive aspects, while the second factor assesses more emotional and social aspects. Factor 3 is heavily weighted on two positive symptoms, delusions and hallucinations. Taken together, these three factors account for 38% of the variance. Thus, the principal components are more spread out than in the first study, and no single large clean bipolar factor emerges. Nevertheless, these results still indicate that positive and negative symptoms tend to be uncorrelated with one another, if not negatively correlated with one another.
Once adequate reliability has been achieved, however, it is important to document internal consistency and external validity. This chapter has summarized some of our work with the former topic. It suggests that the SANS has high internal consistency, while the SAPS is somewhat less internally consistent. What are the clinical and research implications of these results concerning internal consistency? On a superficial and statistical level, these results might be considered to suggest that the SAPS is less valid than the SANS, and perhaps that therefore it is not useful. On the other hand, it might be argued that it is not necessary to measure all the items on the SANS, since
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Summary and Conclusions
87
they are highly intercorrelated with one another and any one might `stand in place' for the others. While in a sense statistically correct, these conclusions are probably misleading in both clinical and research settings. The SANS and SAPS were designed primarily as descriptive instruments that are useful for encoding symptoms commonly observed in psychiatric patients. Essentially, these results document clinical common sense. Patients with one negative symptom tend to have several others, while patients with one positive symptom do not necessarily have others. In other words, affective flattening seems to be related to alogía and aνοlition, but delusions and hallucinations do not necessarily occur together. In a comprehensive description of an individual patient, it is important to document all the types of symptoms that are present. This is particularly crucial in pharmacologic studies, where one may wish to document that some symptoms or groups of symptoms are more responsive to treatment than are other symptoms. For example, although anhedonia and alogia are statistically correlated with one another in a population of schizophrenics, in an individual patient anhedonia might be more responsive to treatment with a specific medication than is alogía. In spite of the high intercorrelations, it remains important in clinical settings to evaluate all relevant symptoms. The results of the factor analyses in this second study do not suggest as clean and strong a separation between positive and negative symptoms as was indicated in our original study. Factor analysis is notoriously sampledependent, but there is no reason to suspect that the sample in the second study was different in any way from that of the first. Both involved consecutive admissions of DSM-III schizophrenics to the Iowa Psychiatric Hospital. The individuals doing the clinical evaluation did change, however. All ratings were done in the first study by a single clinician (N.C.A.), who was more senior and experienced than the raters of the second study (three different bachelor's to PhD level clinical psychologists and nurses). In the first study there was clearly a stronger tendency to use the full scale and include a reasonable number of both high and low ratings, while raters in the second study tended to bunch more toward the middle as is often the case with less experienced raters. Alternatively, in the first study, hidden hypotheses may have produced some `halo effect'. Nevertheless, the results in the second study are not inconsistent with those of the first study. Positive and negative symptoms are no longer negatively correlated, but they remain uncorrelated with one another, suggesting that they do represent different dimensions of psychopathology.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Methods for Assessing Positive and Negative Symptoms
Andreasen
88
References
Nancy C. Andreasen, MD, PhD, Department of Psychiatry, Psychiatric Hospital, 500 Newton Road, Iowa City, IA 52242 (USA)
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Andreasen, N.C.: The clinical assessment of thought, language, and communication disorders. I. The definition of terms and evaluation of their reliability. Archs gen. Psychiat. 36: 1315-1325 (1979a). Andreasen, N.C.: The clinical assessment of thought, language, and communication. II. Diagnostic significance. Archs gen. Psychiat. 36: 1325-1330 (1979b). Andreasen, N.C.: Affective flattening and the criteria for schizophrenia. Am. J. Psychiat. 136: 944-947 (1979c). Andreasen, N.C.: Negative symptoms in schizophrenia: Definition and reliability. Archs gen. Psychiat. 39: 784-788 (1982). Andreasen, N.C.: The Scale for the Assessment of Negative Symptoms (SANS) (University of Iowa, Iowa City 1983). Andreasen, N.C.: The Scale for the Assessment of Positive Symptoms (SAPS) (University of Iowa, Iowa City 1984). Andreasen, N.C.; Alpert, M.; Martz, M.J.: Acoustic analysis: an objective measure of affective flattening. Archs gen. Psychiat. 38: 281-285 (1981). Andreasen, N.C.; Grove, W.M.: Thought, language, and communication in schizophrenia: Diagnostic and prognostic significance. Schizophr. Bull. 12: 348-359 (1986). Andreasen, N.C.; Olson, S.: Negative versus positive schizophrenia: Definition and validation. Archs gen. Psychiatr. 39: 789-794 (1982). Endicott, J.; Spitzer, R.L.: A diagnostic interview: The Schedule for Affective Disorders and Schizophrenia (SADS). Archs gen. Psychiat. 35: 837-844 (1978). Humbert, M.; Salvador, L.; Seguí, J.; Obiols, J.; Obiols, J.D.: Estudio interfiabilidad version espafiola: evaluacion de síntomas positivos y negativos. Rev. Dep. Psiquiat. Fac. Med., Univ. Barcelona 13: 28-36 (1986). Krawiecka, M.; Goldberg, D.; Vaughan, M.A.: Standardized psychiatric assessment for rating chronic patients. Acta psychiat. scand. 55: 299-308 (1982). Moscarelli, M.; Maffei, C.; Cessna, B.M.: An international perspective on assessment of negative and positive symptoms in schizophrenia. Am. J. Psychiat. 144: 1595-1598 (1987). Ohta, T.; Okazaki, Y.; Anzai, N.: Reliability of the Japanese version of The Scale for the Assessment of Negative Symptoms (SANS). Jap. J. Psychiat. 13: 999-1010 (1984). Overall, J.; Gorham, D.: Brief psychiatric rating scale. Psychol. Rep. 10: 799-812 (1962). Phillips, M.: Scale for the Assessment of Negative Symptoms and Scale for the Assessment of Positive Symptoms (Chinese version); pers. commun. (National Center for Psychiatric Training, Shashi Psychiatric Hospital, Shashi Hubei, PR China, 1987/88).
Methods for Assessing Positive and Negative Symptoms' Nancy C. Andreasen Department of Psychiatry, University of Iowa College of Medicine, Iowa City, Iowa, USA
This research was supported in part by 111H Grants 1Η31593 and 1 Η40856; MHCRC Grant 1Η43271; the Nellie Ball Trust Research Fund, Iowa State Bank & Trust Company, Trustee, and a Research Scientist Award, 1Η00625.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
When interest reawakened in the positive versus negative distinction, initially adequate instruments were not available with which to measure negative symptoms. Scales such as the Krawiecka [Krawiecka et al., 1982] or the Brief Psychiatric Rating Scale (BPRS) [Overall and Gorham, 1962] were used because they were already in existence. The BPRS in particular had been widely used for many years in clinical drug trials. Neither of these scales was developed specifically for the assessment of positive versus negative symptoms, however, and both have significant limitations. The Krawiecka has not been subjected to rigorous reliability assessment or validation. The BPRS was designed through factor analysis and rate factors rather than symptoms, an approach which sometimes makes it difficult for clinically trained investigators to apply because its level of abstraction is somewhat higher than is actually observed in living patients. While considerable psychometric testing was done with it initially, it has not been rigorously reassessed since the advent of contemporary structured interviews and diagnostic criteria. In this context, it seemed clear that new approaches for the assessment of positive and negative symptoms in schizophrenia were badly needed. This led to the development of two new rating scales which are now widely used, the Scale for the Assessment of Positive Symptoms (SAPS) and the Scale for the Assessment of Negative Symptoms (SANS) [Andreasen, 1983].
Andreasen
74
The decision to develop the SANS and SAPS grew out of a longstanding interest in phenomenology and many years of direct clinical experience in diagnosing and treating patients suffering from schizophrenia who were admitted to the Inpatient Service at the University of Iowa Psychiatric Hospital. It was clear that work was needed in the area of negative symptoms in particular, since relatively good methods for assessing positive symptoms such as delusions and hallucinations were available through the Schedule for Affective Disorders and Schizophrenia (SADS), a widely used interview developed initially for the Collaborative Study of the Psychobiology of Depression [Endicott et al., 1978]. The SANS was the first instrument developed in order to provide for comprehensive assessment of negative symptoms in schizophrenia [Andreasen, 1982, 1983]. It consists of five scales that evaluate five different aspects of negative symptoms: alogia, affective blunting, avolition-apathy, anhedonia-asociality, and attentional impairment. Each of these negative symptoms can be rated globally, but in addition detailed observational measures are made in order to achieve the global rating. It is complemented by a SAPS, which permits detailed evaluation and global ratings of hallucinations, delusions, positive formal thought disorder and bizarre behavior [Andreasen, 1984]. Taken together, the two scales provide a comprehensive set of rating scales in order to measure the symptoms of schizophrenia and to assess their change over time. Repeated close contact with patients suffering from schizophrenia cannot help but lead to a `Bleulerian' belief that delusions and hallucinations, although easier to define reliably, are not the most important, characteristic, or crippling symptoms of schizophrenia. Yet Bleulerian symptoms tended to be neglected by phenomenologists and nosologists because of a concern that they were too imprecise. An initial foray was made on the most important of the Bleulerian symptoms, thought disorder. A scale to assess thought disorder was developed using the empirical assumption that thought is usually inferred from speech, that speech can be observed directly, and that abnormalities in language, speech and communication can be defined precisely through careful attention to grammar, word choice, and links between sentences and clauses. A scale for the assessment of thought, language and communication (TLC) yielded highly reliable definitions, demonstrating that thought disorder was not pathognomonic of schizophrenia, and that it fell
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Development of the SANS and SAPS
Methods for Assessing Positive and Negative Symptoms
75
into two broad groups of subtypes. One group, negative formal thought disorder, was characterized primarily by poverty of speech and poverty of content of speech, while the other, positive formal thought disorder, was characterized by derailments, tangentiality, incoherence and distractible speech. We also observed that negative thought disorder, when present, tended to persist over time and to predict a poor prognosis. Unlike positive thought disorder, it was not common in mania. These findings were confirmed by two independent replications involving several hundred patients and controls over the course of several years [Andreasen, 1979b, c, 1986]. A second foray was made on another major Bleulerian symptom, affective blunting, through the development of an affect rating scale [Andreasen, 1979a]. This study also indicated that affect could be defined reliably using an objective empirical approach, that abnormalities in affect were not specific to schizophrenia either, and yet that affective blunting in schizophrenia could be validated physiologically by measuring voice characteristics such as frequency and amplitude [Andreasen et al., 1981]. These studies confirmed that two major negative symptoms, negative thought disorder and affective blunting, could indeed be defined reliably and that they had a variety of important clinical correlates. Both these negative symptoms, while important, did not provide a complete coverage of the psychopathology frequently observed in patients with schizophrenia. Consequently, a scale was developed to provide a comprehensive coverage of all negative symptoms, adding avolition, anhedonia, and attentional impairment to the two already developed.
The SAPS and the SANS are informed by a series of basic assumptions. First, while the concept of pathognomonic symptoms is psychologically appealing, it has no basis in reality. One simply cannot identify any symptoms that occur only in schizophrenia and not in any other illness. Just as manics and psychotic depressives are likely to have delusions and hallucinations, so too depressives are likely to have some negative symptoms such as alogía or affective blunting. An attempt to define any psychopathological symptom so that it is specific to a single diagnosis is likely to lead to convoluted intellectual contortions or the wearing of blinders.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Basic Assumptions
76
Therefore, the SANS and SAPS were both developed in order to define and describe a wide range of symptoms reliably. They could then be used to observe the frequency of these signs and symptoms in a variety of diagnostic categories and to observe whether they showed different patterns in various diagnostic categories or differences in change over time. A second basic assumption is that good reliability is best achieved through the use of objective observational items. For example, affective blunting is not defined by examining the patient's psychological state. Instead, it is assessed through observation of a series of external behaviors, such as mobility of facial expression, response to a stimulus such as being smiled at, quality of eye contact, use of expressive gestures, etc. As is discussed in more detail below, the fundamental soundness of this strategy has been confirmed through the repeated achievement of good reliability in a wide range of cultural settings. A third assumption is that rating scales must build on cross-sectional evaluation. They must be based on phenomena that one can observe clinically at a specified point in time, which may be an hour, a day, a week, or the past month. The particular cross-sectional window will depend on the specific goal or purpose of a study. Studies that attempt to build toward the development of diagnostic criteria may use a maximally large window (e.g., one month), while those attempting to observe change over time (as in treatment studies) may select a smaller window. The repeated use of rating scales over time can be used to build longitudinal definitions through repeated measurements and the observation of change. Rating scales cannot identify `enduring' symptoms reliably during a single evaluation, and to attempt to do so retrospectively is risky. Consequently, if one wishes to identify enduring negative symptoms, one must do so empirically through prospective longitudinal assessments. Fourth, ideally symptoms should be defined in such a way that their underlying neural mechanisms could be identified. All the various signs and symptoms of schizophrenia must ultimately reflect neural activity in the brain, and understanding the presence or absence of symptoms in terms of such neural mechanisms is one major long-term goal of phenomenological research. Since we do not yet know or understand these neural events, however, the best strategy is to maintain a comprehensive and flexible data base, and an open mind as well. Fifth, ideally rating scales should not sacrifice comprehensive coverage for simplicity, nor should they use premature foreclosure concerning the interrelationships between signs and symptoms. While it is tempting to
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Andreasen
Methods for Assessing Positive and Negative Symptoms
77
use a scale that only involves one or two items and can be completed in two or three minutes, this is likely to be penny wise and pound foolish if one wishes to understand either the neural mechanisms or the longitudinal evolution of symptoms over time. Detailed comprehensive coverage that stays close to direct clinical data is the best strategy. While factor analytic techniques can be used to reduce data sets (and this may be useful for statistical purposes), nevertheless the basic data that are reduced statistically should be maintained intact and reexamined periodically. It could be argued, for example, that since negative symptoms tend to be highly intercorrelated, only one need be rated in order to save time. Yet factor analytic studies tend to be notoriously unstable (í.e., dependent upon the population sampled). Thus, if only one negative symptom were rated based on the reasoning that negative symptoms are highly intercorrelated, not only might one have invalid data, but one might fail to observe (for example) that avolition responds to treatment in some particular patients while affective blunting does not. Sixth, ideally rating scales designed to assess psychopathology should be sensitive to change, since one major use is to evaluate whether various signs and symptoms respond to treatment. The identification of `enduring' or `core' or `primary' symptoms is best achieved through repeated assessments, not through developing rating scales likely to be insensitive to change. Clinicians and investigators must also recognize that negative symptoms may be `enduring' in some patients and `changing' in others; thus the whole issue of stability and prognostic significance must be studied empirically and prospectively.
The original components of the SANS and SAPS were subjected to careful scrutiny as to their reliability and were found to be highly reliable [Andreasen, 1979a—c]. After the SANS and SAPS themselves were completed, similar reliability studies were done and supported the reliability of the entire scales [Andreasen, 1982]. The reliability studies at the University of Iowa were quickly followed by reliability studies in a variety of international settings. The SANS and SAPS clearly fulfilled an important clinical and research need, for they have now been translated into a variety of languages, including Japanese, Spanish, Italian, French, German, Dutch, Korean,
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Reliability
Table 1. Inter-rater reliability of negative and positive symptoms in different cultural settings Intraclass R (weighted K) Italy
Spain
Japan
China
Poverty of speech Poverty of content of speech Blocking Increased latency of response Subjective rating of alogía Global rating of alogía Subscale score
0.632 0.615 0.269 0.803
0.870 0.604 0.935 0.571 0.907 -0.014 0.939 0.591 0.802 0.864 0.945 0.628 0.971
0.801 0.771 0.825 0.824 0.837 0.989
Unchanging facial expression Decreased spontaneous movements Paucity of expressive gestures Poor eye contact Affective nonresponsivity Inappropriate affect Lack of vocal inflections Subjective rating of affective flattening Global rating of affective flattening Subscale score
0.786 0.782 0.757 0.873 0.714 0.774 0.827
0.930 0.940 0.886 0.897 0.774 0.805 0.963 0.831 0.844 0.926
0.805 0.728 0.671 0.676 0.641 0.294 0.720 0.553 0.721
0.847 0.835 0.772 0.722 0.787 0.034 0.835 0.581 0.903
0.759 0.734 0.843 0.964 0.860 0.942
0.744 0.752 0.513 0.607 0.749
0.516 0.732 0.735 0.756 0.817
0.875 0.790 0.616 0.768 0.841 0.769 0.810
0.610 0.742 0.552 0.642 0.823 0.725
0.761 0.712 0.709 0.790 0.798 0.864
0.870 0.937 0.857 0.892 0.938
0.711 0.987 0.462 0.788
0.550 0.925 0.672 0.832
logia
Affective flattening
Avolitionapathy
Anhedoniaasociality
Attentional impairment
0.694
0.688
Grooming and hygiene Impersistence at work or school Physical anergía Subjective complaints of avolition-apathy Global rating of avolition-apathy Subscale score
0.624 0.715 0.755
Recreational interests and activities Sexual interest and activity Ability to feel intimacy and closeness Relationships with friends and peers Subjective awareness of anhedonia-asociality Global rating of anhedonia-asociality Subscale score
0.741 0.605 0.587 0.620
Work inattentiveness Inattentiveness during mental testing Subjective complaints of inattentiveness Global rating of inattentiveness Subscale score
0.747
0.728 0.321 0.854 0.659
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Negative symptoms
Table I (continued) Intraclass R (weighted K) Italy
Spain
Japan
0.942 0.944 0.896 1.000 0.948 0.712
0.953 0.876 0.889 0.827 0.861 0.701 0.870 0.927 0.895
0.904 0.862 0.013 0.451 0.726 0.713 0.304 0.837
China
Positive symptoms Auditory hallucinations Voices commenting Voices conversing Somatic or tactile hallucinations Olfactory hallucinations Visual hallucinations Subjective rating of hallucinations Global rating of hallucinations Subscale score
0.862
Persecutory delusions Delusions of jealousy Delusions of guilt or sin Grandiose delusions Religious delusions Somatic delusions Delusions of reference Delusions of being controlled Delusions of mind-reading Thought broadcasting Thought insertion Thought withdrawal Subjective rating of delusions Global rating of delusions Subscale score
0.906 0.357 0.600 0.936 0.886 0.738 0.768 0.889 0.890 0.664 0.877 0.828
Clothing and appearance Social and sexual behavior Aggressive and agitated behavior Repetitive or stereotyped behavior Subjective rating of bizarre behavior Global rating of bizarre behavior Subscale score
0.796 0.734 0.750 0.849
Positive formal Derailment thought disorder Tangentiality Incoherence Illogicality Circumstantiality Pressure of speech Distractible speech Clanging Subjective rating of positive formal thought disorder Global rating of positive formal thought disorder Subscale score
0.855 0.693 0.814 0.492 0.525 0.761 0.637 0.441 0.870 0.818
Delusions
Bizarre behavior
0.878
0.834
0.964 0.934 0.789 0.307 0.768 0.870 0.752 0.782 0.673 0.954 0.860 0.906 0.826 0.776 0.869 0.914 0.678 0.864 0.698 0.879 0.839 -0.16 0.854 0.901 0.954 0.232 0.896 0.420 0.893 0.775 0.665 0.777 0.849 0.884 0.867 0.914
0.920 0.235 0.751 0.657 0.745 0.993
0.766 0.599 0.666 0.836 0.560 0.617 0.844 0.564
0.825 0.727 0.807 0.159 0.423 0.363 0.757 0.858
0.989 0.881
0.942 Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Hallucinations
Andreasen
80
and Chinese. Reliability data have been reported in studies in Japan, Spain, Italy, and China [Humbert et al., 1986; Moscarelli et al., 1987; Ohta et al., 1984; Phillips, 1987/88]. The reliability data conducted in these various international studies are summarized in table 1. As this table indicates, the reliability is consistently high in a variety of cultural settings. The results indicate the soundness of the basic strategy of using observational rather than subjective evaluation. Ideally, cross-national studies would now be appropriate. These are underway in Spain and Italy, but not as yet complete. Even when observational items are stressed, good reliability is difficult to achieve without adequate training. All too frequently, once standardized rating scales become available, clinicians and investigators simply take them `off the shelf and begin to use them, assuming that reliability data already published provide sufficient evidence that the scales will be reliable in any setting. This simply is not true. Reliability needs to be demonstrated by comparisons of ratings obtained within the individual center, preferably with reference to some external calibration such as videotapes of structured interviews with standardized calibration ratings. Further, reliability needs to be rechecked at regular intervals, particularly as the raters are trained, in order to prevent `rater drift'. A comprehensive set of training materials has been developed for the SANS and SAPS involving both videotapes and case vignettes. These have been widely used in order to insure that these two scales are being used consistently within the US, and they are now also available in international video format as well. In addition, investigators who wish additional training have received it on site at the University of Iowa. Our experience in such training indicates that cross-national reliability is likely to be good when such studies are conducted, since investigators from other cultures learn to calibrate against our standards quite rapidly.
In a psychometric sense, a scale is often considered to be validated when good internal consistency is demonstrated. Measures of internal consistency, such as coefficient alpha (Cronbach's alpha), provide an indication of the extent to which items on a scale are related to one another or internally coherent. In our initial work in developing the SANS and SAPS, we assessed internal consistency and found it to be relatively high for the
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Internal Consistency
81
SANS and less so for the SAPS (for reasons discussed below). We have now replicated our original work in a second sample of 117 schizophrenic patients consecutively admitted to the University of Iowa Psychiatric Hospital between 1982 and 1985. The results of this second study appear in tables 2 and 3. As table 2 indicates, this second replication study has produced findings quite similar to those observed in the first study. The results are summarized in four columns. The first column shows the correlation between individual items and the sum of the individual scales (e.g., alogia, affective flattening, etc.). The second column shows the correlation between each individual item and the global rating on the subscale. The third column indicates the correlation between individual items and the sum of all negative symptom ratings, while the fourth column indicates the correlation between individual items and the sum of all global ratings for each of the five subscales. Coefficient alpha is also listed for each of the subscales. As table 2 indicates, coefficient alpha is consistently high, indicating that each of the five subscales does have good internal consistency. These results are most interesting, however, as a way of determining which items on the individual subscales are most useful in making the ratings. For example, blocking is less highly correlated with the sum of the alogia ratings than are the other symptoms used as indices of alogia, and it is poorly correlated with the global rating of alogia. Inappropriate affect is poorly correlated with more general indices of affective flattening and with indices of negative symptoms in general. Because this correlation is so poor, we have dropped inappropriate affect as a measure of negative symptoms and now consider it to be a positive symptom. Correlations between individual items and the sum of global ratings are relatively high for the other negative symptom subscales. As would be expected, correlations between individual items and indices of negative symptoms in general are lower than they are between subscale measures (i.e., sum of global ratings versus sum of sums and sum of globals). This suggests that individual items are indeed tapping something other than a global construct. Table 3 presents similar analyses for the four positive symptoms measured on the SAPS. Again, coefficient alpha is relatively high for all four scales. On the other hand, however, when we look at individual correlations, we see that some measures are much less correlated either with the subscale as a whole or with positive symptoms generally. For example, somatic and olfactory hallucinations are relatively poorly correlated with other hallucinations and with other positive symptoms generally. Many of
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Methods for Assessing Positive and Negative Symptoms
Andreasen
82
Sum Alogia Poverty of speech Poverty of content Blocking Latency Affective flattening/blunting Facial expression Spontaneous movements Expressive gestures Eye contact Nonresponsivity Inappropriate affect Vocal inflections Avolition Grooming Impersistence Anergía Anhedonia Recreation Sex Intimacy Friends Attention Social attentiveness Testing
Global rating
Sum of sums
Sum of globals
0.83263 0.82033 0.78829 0.62819 0.56366 0.42247 0.50079 0.38229 0.34585 0.76392 0.52728 0.58831 Coefficient alpha = 0.627793
0.73951 0.44390 0.33755 0.52723
0.78429 0.79324 0.68009 0.71034 0.50865 0.59758 0.72611 0.84293 0.71649 0.82159 0.68250 0.75209 0.82605 0.67973 0.78125 0.28158 0.30487 0.30387 0.74257 0.62576 0.68271 Coefficient alpha = 0.834062
0.63846 0.45773 0.62932 0.66871 0.70659 0.36616 0.59857
0.80688 0.67532 0.67447 0.82847 0.86743 0.60619 0.80485 0.63646 0.74183 Coefficient alpha = 0.743490
0.68853 0.69674 0.68084
0.77886 0.72715 0.72790 0.50690 0.54374 0.77713 0.78508 0.58966 0.54512 0.75129 0.83034 0.59898 Coefficient alpha = 0.773762
0.67102 0.43175 0.51570 0.61153
0.85835 0.77427 0.89796 0.89100 0.79725 0.63987 Coefficient alpha = 0.749987
0.75861 0.63263
the different types of delusions are also relatively poorly correlated. Some of these low correlations are in fact somewhat surprising (e.g., jealousy, guilt, somatic). In spite of the fact that coefficient alpha for delusions is relatively high (0.655), the correlations between individual items in either the sum or the global rating suggest that the items on the scale are relatively diverse.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Table 2. Negative symptoms: correlation coefficients for individual subscale items and general subscale or negative symptom measures
83
Methods for Assessing Positive and Negative Symptoms
Table 3. Positive symptoms: correlation coefficients for individual subscale items and general subscale or positive symptom measures
Hallucinations Auditory hallucinations Voices commenting Voices conversing Somatic hallucinations Olfactory hallucinations Visual hallucinations Delusions Persecutory Jealousy Guilt Grandiose Religious Somatic Reference Control Mind-reading Thought broadcasting Thought insertion Thought withdrawal Positive formal thought disorder Derailment Tangentiality Incoherence Illogicality Circumstantiality Pressure of speech Distractible speech Clanging Bizarre behavior Clothing, appearance Social/sexual behavior Aggressive/agitated behavior Repetitive/stereotyped behavior
Global rating
Sum of sums
Sum of globals
0.62684 0.82945 0.90973 0.74716 0.55985 0.48682 0.47539 0.77000 0.53270 0.48082 0.37065 0.40481 0.291117 0.13722 0.09370 0.72346 0.52633 0.52325 Coefficient alpha = 0.748805
0.68311 0.43489 0.36278 0.26822 0.01133 0.39971
0.51214 0.60499 0.66331 0.03231 0.17554 0.07629 0.30080 0.15573 0.23627 0.45880 0.32532 0.34353 0.32747 0.42176 0.31415 0.25023 0.26414 0.16695 0.34638 0.48527 0.40425 0.32417 0.36228 0.48884 0.68680 0.36269 0.52982 0.33834 0.53096 0.19104 0.25229 0.50392 0.19826 0.32530 0.49835 0.17302 Coefficient alpha = 0.655092
0.44891 -0.04645 0.15853 0.19167 0.18964 0.15230 0.21074 0.25173 0.23339 0.11098 0.09311 0.08268
0.74280 0.71285 0.36224 0.78277 0.84183 0.38356 0.43346 0.68977 0.61321 0.41637 0.55435 0.53092 0.08054 0.46323 0.43033 0.45470 0.19960 0.59060 0.48447 0.41762 0.38152 0.14562 0.08241 -0.09033 Coefficient alpha = 0.737787
0.41171 0.42318 0.48814 0.43222 0.00539 0.17120 0.38159 -0.11172
0.33979 0.83352 0.75213 0.79734 0.75098 0.29098 0.75540 0.68043 0.28380 0.75266 0.60829 0.35674 Coefficient alpha = 0.787094
0.47468 0.41491 0.44797 0.50882 Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Sum
84
Coefficient alpha must always be interpreted cautiously as an index of internal consistency because it is partially tied to a number of items on the scale. The larger the number of items, the higher the coefficient alpha, due to the inherent properties of the statistic. Thus a coefficient alpha of 0.74 for a three-item scale such as volition suggests very strong internal consistency in comparison with a coefficient alpha of 0.65 for a twelve-item scale measuring delusions. Clinical common sense tells us that many patients with delusions typically have only one or two types of delusions, and therefore one would not necessarily expect a high correlation between all items on this scale. Among the positive thought disorder items, clanging is clearly weak, while the indices of bizarre behavior are relatively more internally consistent. Tables 4 and 5 provide another perspective on the internal consistency of the positive and negative symptom rating scales. Table 4 focuses on negative symptoms. It indicates what the coefficient alpha is if both the row and column items are deleted, thereby providing an index of how much each of the individual subscales contributes to the internal consistency of the negative symptoms scale as a whole. (Coefficient alpha for the negative symptoms scale as a whole is 0.855.) For example, if affective flattening is dropped from the scale, coefficient alpha remains quite high at 0.815970. If both affective flattening and alogia are deleted, coefficient alpha drops to 0.576515. Since coefficient alpha tends to remain high no matter which subscale is dropped, one must conclude that the items on the negative symptoms scale are highly integrated to one another. Table 5 shows similar analyses for the positive symptoms scale. (Coefficient alpha for this scale is 0.483.) These results suggest that positive symptoms are much less highly correlated with one another than are negative symptoms. Examination of the contribution of individual subscales to the scale as a whole further confirms this impression. These symptoms have the strongest correlation with the scale as a whole, maintaining coefficient alpha at 0.4483 when delusions are deleted. Thought disorder also has a strong correlation. On the other hand, when more than one global rating is deleted, coefficient alpha drops markedly and even becomes negative. It seems clear from these results that positive symptoms are less internally cohesive than negative symptoms. An alternate way of looking at the interrelationship between items is through principal components analysis. These data are presented in table 6. In our previous study [Andreasen, 1982], principal components analysis yielded a large bipolar first factor which had strong positive
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Andreasen
85
Methods for Assessing Positive and Negative Symptoms Table 4. Coefficient alpha: negative symptoms scale Affective Alogía flattening Affective flattening Alogia Avolition Anhedonia Attention
0.815970 0.576515 0.630511 0.706086 0.573116
0.824333 0.582466 0.678863 0.494925
Avolition Anhedonia Attention
0.820781 0.712406 0.578984
0.846275 0.000000
0.822377
Table shows what the coefficient alpha is if both the row and the column items are deleted. Coefficient alpha for the global ratings of the negative symptoms scale = 0.855.
Table 5. Coefficient alpha: positive symptoms scale Global Global Global Global halluci- delusion bizarre thought behavior disorder nation Global hallucination Global delusion Global bizarre behavior Global positive thought disorder
0.336785 —0.227464 —0.829674 —0.075846
0.448318 0.051725 0.336871
0.359254 0.000000
0.457410
weightings on negative symptoms and strong negative weightings on positive symptoms; this factor accounted for 37% of the variance. In the second replication study on a somewhat larger sample, the results of principal components analysis are slightly different, although not strikingly inconsistent. As table 6 indicates, factor 1 is heavily loaded on most negative symptoms with essentially zero loadings on positive symptoms. This factor accounts for 17.5% of the variance and taps one aspect of negative symptoms. Factor 2 has strong loadings on three additional negative symptoms (affective flattening, anhedonia, and avolition) and therefore also taps some aspect of negative symptoms. The first factor might be considered to
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Table shows what the coefficient alpha is if both the row and the column items are deleted. Coefficient alpha for the global ratings of the positive symptoms scale = 0.483.
86
Andreasen
Table 6. Principal components analysis: positive and negative symptom global ratings Principal component 1 Attention Alogia Affective flattening Anhedonia Avolition Delusions Hallucinations Bizarre behavior Positive formal thought disorder
0.77161 0.71200 0.52196 0.29385 0.49270 0.01357 0.08088 0.09442 0.16183
Variance explained by each principal component
17.5%
2
3
0.24769 0.26149 0.48305 0.72262 0.55920 -0.00034 0.17786 0.24117 -0.03711
0.10686 -0.5981 0.12239 0.10396 0.10968 0.75342 0.45636 0.07425 0.08634
12.9%
8.4%
4 0.11430 0.34775 0.19889 0.07155 0.14602 0.00342 0.20688 0.54808 0.05546 8.0%
be assessing more cognitive aspects, while the second factor assesses more emotional and social aspects. Factor 3 is heavily weighted on two positive symptoms, delusions and hallucinations. Taken together, these three factors account for 38% of the variance. Thus, the principal components are more spread out than in the first study, and no single large clean bipolar factor emerges. Nevertheless, these results still indicate that positive and negative symptoms tend to be uncorrelated with one another, if not negatively correlated with one another.
Once adequate reliability has been achieved, however, it is important to document internal consistency and external validity. This chapter has summarized some of our work with the former topic. It suggests that the SANS has high internal consistency, while the SAPS is somewhat less internally consistent. What are the clinical and research implications of these results concerning internal consistency? On a superficial and statistical level, these results might be considered to suggest that the SAPS is less valid than the SANS, and perhaps that therefore it is not useful. On the other hand, it might be argued that it is not necessary to measure all the items on the SANS, since
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Summary and Conclusions
87
they are highly intercorrelated with one another and any one might `stand in place' for the others. While in a sense statistically correct, these conclusions are probably misleading in both clinical and research settings. The SANS and SAPS were designed primarily as descriptive instruments that are useful for encoding symptoms commonly observed in psychiatric patients. Essentially, these results document clinical common sense. Patients with one negative symptom tend to have several others, while patients with one positive symptom do not necessarily have others. In other words, affective flattening seems to be related to alogía and aνοlition, but delusions and hallucinations do not necessarily occur together. In a comprehensive description of an individual patient, it is important to document all the types of symptoms that are present. This is particularly crucial in pharmacologic studies, where one may wish to document that some symptoms or groups of symptoms are more responsive to treatment than are other symptoms. For example, although anhedonia and alogia are statistically correlated with one another in a population of schizophrenics, in an individual patient anhedonia might be more responsive to treatment with a specific medication than is alogía. In spite of the high intercorrelations, it remains important in clinical settings to evaluate all relevant symptoms. The results of the factor analyses in this second study do not suggest as clean and strong a separation between positive and negative symptoms as was indicated in our original study. Factor analysis is notoriously sampledependent, but there is no reason to suspect that the sample in the second study was different in any way from that of the first. Both involved consecutive admissions of DSM-III schizophrenics to the Iowa Psychiatric Hospital. The individuals doing the clinical evaluation did change, however. All ratings were done in the first study by a single clinician (N.C.A.), who was more senior and experienced than the raters of the second study (three different bachelor's to PhD level clinical psychologists and nurses). In the first study there was clearly a stronger tendency to use the full scale and include a reasonable number of both high and low ratings, while raters in the second study tended to bunch more toward the middle as is often the case with less experienced raters. Alternatively, in the first study, hidden hypotheses may have produced some `halo effect'. Nevertheless, the results in the second study are not inconsistent with those of the first study. Positive and negative symptoms are no longer negatively correlated, but they remain uncorrelated with one another, suggesting that they do represent different dimensions of psychopathology.
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Methods for Assessing Positive and Negative Symptoms
Andreasen
88
References
Nancy C. Andreasen, MD, PhD, Department of Psychiatry, Psychiatric Hospital, 500 Newton Road, Iowa City, IA 52242 (USA)
Downloaded by: Université René Descartes Paris 5 193.51.85.197 - 3/15/2018 9:19:56 PM
Andreasen, N.C.: The clinical assessment of thought, language, and communication disorders. I. The definition of terms and evaluation of their reliability. Archs gen. Psychiat. 36: 1315-1325 (1979a). Andreasen, N.C.: The clinical assessment of thought, language, and communication. II. Diagnostic significance. Archs gen. Psychiat. 36: 1325-1330 (1979b). Andreasen, N.C.: Affective flattening and the criteria for schizophrenia. Am. J. Psychiat. 136: 944-947 (1979c). Andreasen, N.C.: Negative symptoms in schizophrenia: Definition and reliability. Archs gen. Psychiat. 39: 784-788 (1982). Andreasen, N.C.: The Scale for the Assessment of Negative Symptoms (SANS) (University of Iowa, Iowa City 1983). Andreasen, N.C.: The Scale for the Assessment of Positive Symptoms (SAPS) (University of Iowa, Iowa City 1984). Andreasen, N.C.; Alpert, M.; Martz, M.J.: Acoustic analysis: an objective measure of affective flattening. Archs gen. Psychiat. 38: 281-285 (1981). Andreasen, N.C.; Grove, W.M.: Thought, language, and communication in schizophrenia: Diagnostic and prognostic significance. Schizophr. Bull. 12: 348-359 (1986). Andreasen, N.C.; Olson, S.: Negative versus positive schizophrenia: Definition and validation. Archs gen. Psychiatr. 39: 789-794 (1982). Endicott, J.; Spitzer, R.L.: A diagnostic interview: The Schedule for Affective Disorders and Schizophrenia (SADS). Archs gen. Psychiat. 35: 837-844 (1978). Humbert, M.; Salvador, L.; Seguí, J.; Obiols, J.; Obiols, J.D.: Estudio interfiabilidad version espafiola: evaluacion de síntomas positivos y negativos. Rev. Dep. Psiquiat. Fac. Med., Univ. Barcelona 13: 28-36 (1986). Krawiecka, M.; Goldberg, D.; Vaughan, M.A.: Standardized psychiatric assessment for rating chronic patients. Acta psychiat. scand. 55: 299-308 (1982). Moscarelli, M.; Maffei, C.; Cessna, B.M.: An international perspective on assessment of negative and positive symptoms in schizophrenia. Am. J. Psychiat. 144: 1595-1598 (1987). Ohta, T.; Okazaki, Y.; Anzai, N.: Reliability of the Japanese version of The Scale for the Assessment of Negative Symptoms (SANS). Jap. J. Psychiat. 13: 999-1010 (1984). Overall, J.; Gorham, D.: Brief psychiatric rating scale. Psychol. Rep. 10: 799-812 (1962). Phillips, M.: Scale for the Assessment of Negative Symptoms and Scale for the Assessment of Positive Symptoms (Chinese version); pers. commun. (National Center for Psychiatric Training, Shashi Psychiatric Hospital, Shashi Hubei, PR China, 1987/88).
