Receiver operating characteristics curves for Bonn Scale for the Assessment of Basic Symptoms clusters (N = 160): cluster 1, thought, language, perception, and motor disturbances; cluster 2, impaired bodily sensations; cluster 3, impaired tolerance to normal stress; cluster 4, disorders of emotion and affect, including impaired thought, energy, concentration, and memory; cluster 5, increased emotional reactivity, impaired ability to maintain or initiate social contacts, and disturbances in nonverbal expression.
Klosterkötter J, Hellmich M, Steinmeyer EM, Schultze-Lutter F. Diagnosing Schizophrenia in the Initial Prodromal Phase. Arch Gen Psychiatry. 2001;58(2):158-164. doi:10.1001/archpsyc.58.2.158
Copyright 2001 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2001
In schizophrenia research, early detection in the initial prodrome before first psychotic episodes is a major topic. Therefore, the prognostic accuracy of initial prodromal symptoms was examined prospectively.
The study sample was composed of patients referred to outpatient departments of German psychiatric university departments, because of diagnostic problems, between 1987 and 1991. They were examined with the Bonn Scale for the Assessment of Basic Symptoms and the Ninth Version of the Present State Examination to detect an incipient schizophrenic disorder. Of 385 patients showing no schizophrenia-characteristic symptoms, between 1995 and 1998, 110 with and 50 without initial prodromal symptoms were followed up and reexamined with the same instruments for a transition to schizophrenia.
During a mean follow-up period of 9.6 years, 79 (49.4%) of the 160 patients had transited to schizophrenia. The absence of prodromal symptoms excluded a subsequent schizophrenia with a probability of 96% (sensitivity: 0.98; false-negative predictions: 1.3%), whereas their presence predicted schizophrenia with a probability of 70% (specificity: 0.59; false-positive predictions: 20%). Certain disturbances, such as thought interference, disturbances of receptive language, or visual distortions, predicted schizophrenia, even with a probability up to 91% (specificity: 0.85-0.91; false-positive predictions: 1.9%-7.5%).
The Bonn Scale for the Assessment of Basic Symptoms operationalization of prodromal symptoms performed well in the early detection of schizophrenia. It therefore might be useful for the prediction of the disorder, especially if it is further refined to select those items with particularly high prognostic accuracy.
IN THE 1960s, based on longitudinal studies of schizophrenia,1 the German researcher Gerd Huber described subtle, often only self-perceivable deficits, which were reported not only for the postpsychotic stages in which the patients were examined, but also, retrospectively, for the early course, as "basic symptoms."2- 4 In the Bonn Scale for the Assessment of Basic Symptoms (BSABS),5 operational definitions of these prepsychotic deviations are given, along with typical statements of patients and examples of questions, which allow their assessment in a fully or semistructured interview. Besides specific questions, general guiding questions for symptom categories are advised. Based on the patient's description of a complaint, the interviewer decides whether the symptom in question is rated as "present," "questionably present," or "absent." The BSABS has been published in different languages6- 9 (a short English version can be requested from the authors). Furthermore, the BSABS was partially incorporated in other instruments. In 1991, it provided the main source for prodromal symptoms of the Instrument for the Retrospective Assessment of the Onset of Schizophrenia,10 which was used in a large retrospective epidemiological study of the clinical course of schizophrenia before the first hospitalization.10- 13 Recently, prodromal symptoms as defined by the BSABS were included in 2 early detection instruments of leading work groups in this field: the Comprehensive Assessment of At-Risk Mental States14 and the Scale of Prodromal Symptoms.15 The authors of the Comprehensive Assessment of At-Risk Mental States16 put special emphasis on Huber's distinction between reversible outpost syndromes and continuously progressing prodromes as an important modification of the common concept of prodromes in medicine: because prodromal symptoms can resolve, this means that it does not guarantee transition to a first psychotic episode, but may instead indicate an increased risk of this transition. Thereby, the problem of false-positive predictions17 is especially important, and an examination of the traditional prodrome concept in prospective studies becomes critical.
Therefore, in the Cologne Early Recognition (CER) Project, patients suspected to be in an initial prodromal phase were studied prospectively for the first time. The study aimed at an examination of the prognostic accuracy of initial prodromal symptoms as assessed with the BSABS to determine if they can indeed predict the subsequent development of psychosis.
The study sample was taken from 695 patients who were examined for prodromal symptoms at outpatient departments of 5 German psychiatric university departments between 1987 and 1991. Before their referral, all patients had sought help for various complaints from various clinicians. The referrals were because of difficulties that had arisen in the diagnostic and therapeutic procedure. Thus, patients were presented at the specialized outpatient departments for diagnostic clarification of a possibly incipient schizophrenic disorder. Patients were examined with the BSABS and the Ninth Version of the Present State Examination (PSE9).18 The assessment of these 2 instruments served as inclusion criteria of the study.
At this examination, some patients already met DSM-III-R criteria of schizophrenia,19 delusional or psychotic disorder not elsewhere classified. These subjects (n = 189) as well as those with a confirmed diagnosis of a substance-induced disorder (n = 47), organic mental disorder (n = 34), or mental retardation (n = 12) were excluded from the study. An additional exclusion criterion was being older than 50 years at first examination (n = 28).
In all remaining 385 subjects, no characteristic symptoms according to DSM-III-R and, after revision, DSM-IV criterion of schizophrenia,20 respectively, had been found, neither at first examination nor at any time before. Therefore, the subjects could not yet be diagnosed as having the first episodes of illness, although a large number (n = 253) had reported potential prodromal symptoms. These patients returned to the care of their referring clinicians, and only a small number (n = 15) kept in sporadic contact with the respective outpatient department. Thus, when the follow-up began in 1995, patients had to be contacted anew and convinced to participate in the study. Thereby, various problems arose: 113 subjects either refused to participate or did not respond even after the third letter, 41 had died of causes unknown to us, and 71 could not be traced or had moved abroad. The remaining 160 patients (41.6%) could be followed up until the conclusion of the study in 1998 (Table 1).
At first examination, all 385 patients received tentative diagnoses—according to the prominent clinical picture—mainly of the field of the former "neuroses," Axis I disorders, and personality disorders21 (Table 2).
Despite these clinical pictures, 110 patients had reported disturbances at first examination meeting the description of prodromal symptoms according to BSABS criteria (Table 3).
There was no difference in the distribution of sex, age, or tentative diagnosis either between patients who were followed up and those who were not, or between patients who were followed up and displayed prodromal symptoms and those who did not (Table 2 and Table 3).
According to the study's hypothesis, it was assumed that the 110 followed-up patients with prodromal symptoms at first examination actually had been in an initial prodromal state. Comparing them with the 50 patients without initial prodromal symptoms regarding a subsequent development of schizophrenia should clarify whether disturbances, as assessed with the BSABS, can indeed predict later first psychotic episodes with sufficient accuracy and, if they do, which disturbances show the best prognostic accuracy.
Whereas the same instruments, BSABS and PSE9, were applied at first examination and follow-up, diagnoses were given according to criteria valid at the respective time at both assessment times: DSM-III-R criteria were used at first examination and DSM-IV criteria at follow-up. Regarding schizophrenic disorders, the PSE9 covers all characteristic symptoms relevant for DSM-IV diagnosis; for evaluation of potential prodromal symptoms, a shortened 66-item version of the BSABS was used. For this version, a hierarchical cluster analysis of items on a large inpatient sample of various diagnoses and normal controls had recently led to a clinically plausible 5-cluster solution that broadly confirmed Huber's original categories22: cluster 1, thought, language, perception, and motor disturbances (35 items); cluster 2, impaired bodily sensations (13 items); cluster 3, impaired tolerance to normal stress (5 items); cluster 4, disorders of emotion and affect, including impaired thought, energy, concentration, and memory (7 items); and cluster 5, increased emotional reactivity, impaired ability to maintain or initiate social contacts, and disturbances in nonverbal expression (6 items).
The mean time interval between first examination and follow-up was 9.6 years (Table 1). After informed written consent, patients were reexamined by 2 members of our group (J.K. and F.S-L.) either at the outpatient department or at home. The BSABS and PSE9 were always applied together, and all interviewers had undergone a thorough training for their assessment. Before the follow-up, interrater reliabilities between all 5 interviewers involved in the study for the BSABS and PSE9 subsyndromes had been tested on 10 remitted inpatients with the DSM-III-R diagnosis of a schizophrenic disorder, and received satisfying bias-corrected κ values23 between 0.72 and 0.78 for pairwise agreement. Furthermore, interviewers and diagnosticians were blind to the patients' initial results and diagnoses. Diagnoses were given by 2 senior psychiatrists well-experienced in DSM-IV, considering all available results and information about the clinical course, including clinical records and family reports. A transition to a first psychotic episode was assumed only if DSM-IV criteria of schizophrenia were met in the follow-up period. The onset of the prodrome was assumed for the time patients first noticed subtle changes in themselves that met BSABS criteria; the onset of frank psychosis was defined as the time that PSE9 criteria for psychotic symptoms were first noticed.
Besides an examination of the general ability of BSABS prodromal symptoms to predict transition to schizophrenia, the main purpose was to examine potential prodromal symptoms and certain symptom constellations with regard to their prognostic accuracy. Quantities typically used to evaluate the prognostic accuracy of binary variables are sensitivity, specificity, positive predictive values (PPV), and negative predictive values (NPV) (see Table 4 for definitions),24 whereas an assessment of the overall prognostic accuracy of ordinal or metric variables that is not influenced by the choice of cut points is typically assessed by a geometric approach (the area under the corresponding receiver operating characteristic [ROC] curve).25
A nonparametric technique of this approach25 for related samples was used to compare the prognostic accuracy of the BSABS clusters, expressed as the sum of symptoms present. This was done because it was expected that a number of symptoms would only rarely be present, but taken together, might show good prognostic accuracy. Each ROC curve was constructed by varying the threshold used to determine which values of the observed variable would be considered abnormal (ie, indicating a later transition), and then plotting the resulting sensitivities against the corresponding false-positive rates (1-specificity).26
Furthermore, for a general 15% cut point of symptoms of each BSABS subsyndrome as well as for the single symptoms with binary response, prognostic accuracy measures were calculated. Besides these 4 quantities (Table 4), the overall percentage of false-positive predictions (FP, percentage of symptom present, no disease) and false-negative predictions (FN, percentage of symptom absent, disease) are given that describe the combined outcomes of diagnosis and test/symptom combinations.26
An advantage of sensitivity and specificity over the predictive values is their independence from the prevalence of the illness in the sample. In the absence of general criteria of either minimally acceptable values of diagnostic accuracy measures26 or the choice of the leading quantity in symptom selection,27 emphasis was put on both sensitivity and PPV, and the minimally acceptable value of both quantities was defined according to 2 studies of diagnostic/prognostic accuracy of symptoms in schizophrenia.28,29 Thus, a slightly more liberal value of sensitivity than required for symptoms of an acute episode (≥0.2529) and a value of PPV regarded as already high in a retrospective study of DSM-III-R prodromal symptoms (≥0.7028) were chosen as selection criteria.
For evaluation of the combined predictive ability of symptoms meeting these criteria, a logistic model was used. To assess the predictive accuracy of the derived model in an unbiased manner, the data set was split into a model development sample and a model validation sample.30
According to DSM-IV criteria (A-E), 79 patients (49.4%) had developed schizophrenia in the follow-up period, whereas 81 patients (50.6%) had not. All 79 schizophrenic patients had fulfilled criterion A by showing bizarre delusions and/or commenting or dialoging auditory hallucinations, and thus had to be classified as paranoid type. They had all received inpatient treatment and antipsychotic medication: 48 patients later than 4 weeks after onset of psychotic symptoms, 31 during the first 4 weeks. In 77 of the 79 transited patients, prodromal symptoms had been found at the first examination that persisted until outbreak of the illness.
The mean transition time to schizophrenia occurred 4.3 years after the onset of the initial prodrome in women and after 6.7 years in men (Table 1). The patients who did not develop schizophrenia showed either the same signs and symptoms as at first examination or no psychiatric disorder at all; the 33 with prodromal symptoms at first examination retrospectively reported a full remission of these symptoms, either within some weeks after first examination or after several fluctuations in the clinical course. During the follow-up period—in the transited group only until transition—patients had received nonschizophrenia-specific treatments with no significant group difference in either type or intensity. Furthermore, with one exception, no significant difference between the 2 outcome groups was found regarding the general data as well as the DSM-IV–adapted tentative diagnoses.21 Only the initial diagnosis of a schizotypal personality disorder had a significant positive correlation with the outcome schizophrenia (φ coefficient, 0.19; P<.05).
The general predictive accuracy of initial BSABS prodromal symptoms for schizophrenia was examined for the dichotomization "at least 1 BSABS symptom present" vs "no BSABS symptom present" (Table 5). Whereas 77 of the 79 patients with the development of schizophrenia had reported at least 1 of the 66 BSABS features at first examination, only 2 had not (FN: 1.3%); 33 of the 81 patients without subsequent schizophrenia had also shown at least 1 BSABS feature (FP: 20.6%) (only 48 had not). Thus, the alternative outcome was correctly predicted by the presence or absence of at least 1 BSABS prodromal symptom at first examination in 78.1% of patients. Accordingly, for the prediction of a subsequent schizophrenic development by prodromal symptoms according to BSABS criteria, a specificity of 0.59, a sensitivity of 0.98, a PPV of 0.70, and an NPV of 0.96 resulted.
The ROC analyses of the sum scores of the BSABS clusters22 showed that the predictive accuracy of cluster 1 (thought, language, perception, and motor disturbances) showed a significantly greater predictive discrimination30 in the area under the ROC curve, the concordance index (c), than any other cluster (Figure 1) (Table 6).
With c equal to 0.81, and a sensitivity of 80%, at a specificity of about 72%, this cluster discriminated between patients developing schizophrenia and those who did not to a satisfying degree, whereas the predictive discrimination of the other clusters was similar to tossing a coin (c: 0.50-0.58) (Table 6) (Figure 1).
Next, we were interested in the predictive accuracy of the 5 BSABS clusters22 at a fixed cut point. To be able to compare them despite their varying numbers of included symptoms (ranging from 35 in cluster 1 to 5 in cluster 3), a general cut point of roughly 15% symptoms present was introduced (Table 7).
The results supported those of the ROC analyses: again, the first cluster of thought, language, perception, and motor disturbances at a cut point any 5 of 35 items showed the best predictive accuracy regarding the accuracy quantities in general (Table 7). It had by far the highest specificity (0.84) and PPV (0.77), and the least FP (8.1%), as well as the second highest NPV (0.66). Although its sensitivity was the second lowest (0.56), it still clearly exceeded the minimum value of 0.30 to 0.40 required for diagnostically relevant symptoms of schizophrenia by Andreasen and Flaum.29 Except for cluster 2 (impaired bodily sensations), all other clusters at a roughly 15% cut point were present in higher frequency (sensitivity, 0.63-0.92) and had lower FN (3.8%-18.1%), but all had considerably poorer specificity, PPV, and FP with only cluster 4 (disorders of emotion and affect), which has been the most frequent one with the least FN, doing slightly better in NPV (0.68) than cluster 1 (Table 7).
Ten BSABS prodromal symptoms fulfilled the criteria of sensitivity of 0.25 and greater and a PPV of 0.70 and greater (Table 8). All other symptoms either were not present in a sufficient number of later schizophrenics (ie, sensitivity <0.25) or were so frequent among all patients that their PPVs were less than 0.70. For these 10 symptoms, sensitivity, NPV, and FN were generally acceptable, specificity and PPV were high, and FP, at less than 10%, was very low (Table 8).
The 10 selected symptoms were entered into a logistic regression analysis of the model development sample. At iteration 5, the resulting logistic model had a −2 log likelihood of 74.2 and was highly significant (χ210 = 36.7, P = .0001). It led to correct predictions in 81.2% of the development and in 76.2% of the validation sample. Thus, exhibiting an only insignificant difference between the samples (χ21 = 0.59, P = .44), sample artifacts can be assumed to play a minor role in this solution. Regarding its predictive accuracy, all accuracy quantities were well above 0.70, and all false predictions were below 15% in the development as well as the validation sample (Table 9). But whereas false positives amounted to only 6.3% in the development sample and thus were comparable with those of single items, they reached a less favorable 12.5% in the validation sample (Table 9), indicating a slight bias toward the prediction of schizophrenia.
The main results of the CER Project indicated that the absence of BSABS prodromal symptoms excluded the development of a first psychotic episode, with a probability of 96%, with the percentage of false negatives being only 1.3%. Yet, for these prodromal symptoms in general, specificity (0.59) and PPV (0.70) were considerably lower than sensitivity (0.98) and NPV (0.96), and the percentage of false positives was still 20.6%. Because of its excellent negative predictive ability, this operationalization of prodromal symptoms seems applicable to a broad identification of at-risk persons in the general population. However, the positive predictive ability had to be greater to be of use for a treatment-related prognosis, which seems to be true for the subgroup of thought, language, perception, and motor disturbances. They showed a sensitivity sufficient for a diagnostic criterion of schizophrenia (>0.25), a high PPV (>0.70), few FPs (<8%), and a good discriminative ability (81.2% correct classifications).
The CER Project was performed with an identical and independent assessment at first examination and follow-up, and thus can be regarded as a prospective study. However, in the years 1995 through 1998, when contact with the patients was restored, the first examination had taken place more than 4 years previously. Thus, the study has certain limitations.
First, only 41.6% of the initial sample could be followed up. Therefore, regarding the overall transition rate of 49.4% (which reached 70% for patients with initial prodromal symptoms), the question of a selection bias for patients with severe disturbances at follow-up arises. Since no significant differences between followed-up patients and those lost to follow-up occurred for presence of prodromal symptoms, tentative diagnoses, age, or sex, no indication of an additional selection, a workup, or a verification bias31 at follow-up is given, and the follow-up sample seems fairly representative. On the other hand, the presence of an initial referral bias31 can be neither negated nor affirmed. The incidence of schizophrenia is too low to allow prospective population studies of prodromal symptoms, so no certain knowledge about the representativeness of the referred patients, clearly disturbed and receiving psychiatric diagnoses, for those at-risk in the general population is available.
Second, no repeated measures were carried out, as is usually done in prospective studies. Consequently, possible changes of symptoms or symptom profiles in the longitudinal course could not be appraised reliably, and survival analyses to assess time-to-event dependencies could not be carried out. Expressly with regard to the rather difficult differentiation between statelike prodromal symptoms and traitlike schizotypy features,32 and considering the fact that the DSM-IV schizotypal personality disorder had been the only tentative diagnosis correlating with a later transition to schizophrenia, repeated measures would have been revealing. In any case, BSABS prodromal symptoms by definition differ from DSM-IV schizotypy features and, even among the schizotypal patients of the study, had occurred in late adolescence or early adulthood (around age 20 years). Moreover, the long duration of these disturbances does not preclude their interpretation as prodromal symptoms; with a mean duration of 5.6 years, it corresponds well with the findings in retrospective studies11 of an mean 5- to 6-year duration of the initial prodrome.
The reason for the chosen sampling strategy lay in the empirically low referral rate of patients with initial prodromal symptoms, so that a sample of sufficient size with the clinical course duration long enough to ensure the assessment of a subsequent schizophrenia (ie, at least 5 to 6 years) could be followed up. To our knowledge, only one other prospectively designed study has so far been published.14 However, in this study, the number/sample of 20 patients hitherto followed up over 6 months is still small, and thus, information about the psychosis-predictive accuracy of the included risk indicators has not yet been published. Whether the referral rate can be increased in the future will depend on the success of early detection networks that have been initiated in several countries (eg, Australia, Norway, United States, England, Finland, and, since 1997, in Cologne). Thus far, in the early detection centers, attenuated (ie, 6 DSM-IV schizotypy features) and transient psychotic symptoms were mainly defined as criteria of an initial prodrome.14 Patients were already at the end of the initial prodromal stage, or even at the beginning of the first psychotic episode, when they were included in the service of the respective center or in early detection and intervention studies. Yet, the results of the CER Project give, to our knowledge, the first empirical evidence that early detection in an even earlier state of the prodrome is possible by detecting disturbances with less obvious relationship to psychotic symptoms. An early detection strategy based on these symptoms may have a better chance to prevent the illness and avoid serious problems that arise during the initial prodromal phase (eg, social deficits that develop early in the course of the illness).33
Accepted for publication May 11, 2000.
Funded by grant PSF 27, 0701627 from the German Ministry of Research and Technology, Alliance Ministry for Development and Research, Berlin (1989-1990), the Ørnulv-Ødegård Memory Prize from the Saugstad Foundation (1993-1995), and grant Kl 970 from the German Research Society Research Community, Bonn (1995-1998) (Dr Klosterkötter).
Corresponding author: Joachim Klosterkötter, MD, Department of Psychiatry and Psychotherapy, University of Cologne, Joseph-Stelzmann-Strasse 9, 50924 Cologne, Germany (e-mail: email@example.com)