1. Introduction
The treatment of schizophrenia is an ongoing challenge in psychiatry. The introduction of chlorpromazine in 1952 marked the beginning of the modern era of pharmacotherapy, which has led to the atypical antipsychotic drugs available today. However, the precise mechanism of action of the atypical antipsychotic drugs is unclear and the time course of response poorly understood. Current practice guidelines for the treatment of schizophrenia recommend attempting treatment with a certain antipsychotic for at least several weeks before switching to another drug [3,Reference Falkai, Wobrock, Lieberman, Glenthoi, Gattaz and Möller11,27,28]. However, several recent studies have shown that antipsychotic response occurs within the first weeks of treatment [Reference Agid, Kapur, Arenovich and Zipursky1,Reference Kapur, Arenovich, Agid, Zipursky, Lindborg and Jones17,Reference Leucht, Busch, Hamann, Kissling and Kane20].
As yet, it is difficult to predict the individual treatment response. According to the “early onset hypothesis of antipsychotic drug action” [Reference Leucht, Busch, Hamann, Kissling and Kane20], early symptom changes triggered by antipsychotic medication would be predictive of subsequent remission and response. In fact, both earlier [Reference Bartko, Herczeg and Bekesy5,Reference May, van Putten and Yale24,Reference Möller, Kissling and Zerssen25,Reference Stern, Kahn, Harvey, Amin, Apter and Hirschowitz30] and more recent studies [Reference Ascher-Svanum, Nyhuis, Faries, Kinon, Baker and Shekhar4,Reference Chang, Lane, Yang and Huang7,Reference Corell, Malhotra, Kaushik, McMeniman and Kane8,Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18,Reference Leucht, Bush, Kissling and Kane21–Reference Lin, Chou, Lin, Hsu, Chen and Lane23] indicate that early non-response to antipsychotic medication is a robust predictor of subsequent lack of response in patients with schizophrenia. However, most of these studies were based on re-analyses of randomized or open clinical trials with selected patient samples and rigid treatment regimes. It is unclear whether these results can be generalized to naturalistic samples.
On the basis of the hypothesis that early improvement can predict remission and response in a sample of patients treated under naturalistic conditions, the aims of the present study were:
• to determine the predictive validity of early improvement for subsequent remission and response;
• to identify the definition of early improvement that best predicts subsequent remission or response.
2. Subjects and methods
2.1. Subjects
The sample was part of a multi-centre follow-up programme performed within the German Research Network on Schizophrenia [Reference Wölwer, Buchkremer, Häfner, Klosterkötter, Maier and Möller34] and conducted at 11 university psychiatric hospitals (Aachen, Berlin, Bonn, Cologne, Düsseldorf, Essen, Göttingen, Hamburg, Mainz, Munich, Tübingen) and three district psychiatric hospitals (Augsburg, Wasserburg-Gabersee, Munich-East). In this programme, subjects were randomly selected from all patients admitted between January 2001 and December 2004 to one of the 14 hospitals mentioned above for treatment of a schizophrenic spectrum disorder (ICD-10: F20–F25, F28). Randomization software was used to select patients. Subjects were aged between 18 and 65 years. Exclusion criteria were a major medical illness as the possible cause of the psychiatric symptoms, head injury and drug or alcohol dependence. All patients had given written informed consent to participate in the study. The study protocol was approved by the local ethics committees [Reference Jäger, Riedel, Messer, Laux, Pfeiffer and Naber14].
Patients who were discharged from hospital within 20 days after admission, who did not fulfil the diagnostic criteria for ICD-10 schizophrenia (F20) or who already fulfilled remission criteria at the time of admission were excluded from the present analysis.
2.2. Assessment
Clinical researchers interviewed the patients at admission and made a diagnosis according to the ICD-10 research diagnostic criteria (WHO) [32]. DSM-IV diagnoses were determined using the German version of the Structured Clinical Interview for DSM-IV (SCID) [33].
Psychopathological characteristics were assessed with the Positive and Negative Syndrome Score for Schizophrenia (PANSS) [Reference Kay16]. The PANSS, a widely used 30-item scale for assessing schizophrenic symptoms, is composed of three subscales: positive symptoms (items P1–P7), negative symptoms (items N1–N7) and general psychopathology (items G1–G16). Each item can be graduated on a seven-point scale (1–7). The PANSS total score ranges from 30 to 210, the positive and negative subscores from 7 to 49, and the general psychopathology subscore from 16 to 112.
Ratings were performed within 3 days of admission to hospital, at biweekly intervals during inpatient treatment and at discharge. All raters had been trained for the instruments and a high inter-rater reliability was achieved (ANOVA-ICC > 0.8).
2.3. Analyses
The symptom-severity component of the recently proposed standardized remission criteria [Reference Andreasen, Carpenter, Kane, Lasser, Marder and Weinberger2] was used to define remission as a PANSS rating of three or less in all of the following items: “delusions” (P1), “unusual thought contents” (G9), “hallucinatory behaviour” (P3), “conceptual disorganization” (P2), “mannerism/posturing” (G5), “blunted affect” (N1), “social withdrawal” (N4) and “lack of spontaneity” (N6). Response was defined in accordance with Kinon et al. [Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18] as a reduction of the PANSS total score of at least 40% from admission to discharge.
“Early improvement” was defined as improvement after 2 weeks of treatment. In order to examine the ability of early improvement to predict remission and response at discharge, receiver-operating-characteristic (ROC) curves with area under the curve (AUC) values were calculated for the reduction of psychopathological symptom scores (PANSS total score and positive, negative and general psychopathology subscores) after 2 weeks. The ROC curve is a graphical plot of the sensitivity vs. (1 – specificity) for the ability of early improvement to predict later remission or response. The AUC is a measure of the overall discriminative power. A value of 0.5 for the AUC indicates no discriminative ability; a value of 1.0 indicates a perfect power. According to Weinstein and Fineberg [Reference Weinstein and Fineberg31], a value of 0.7 to 0.8 is considered to be a reasonable, and a value greater than 0.8 a good discriminative capacity.
In order to identify the most useful criterion for defining early improvement, total accuracy, sensitivity, specificity, positive and negative predictive value were calculated for different cut-offs of the improvement of the PANSS total score (10%, 20%, 30%, 40%, 50%). Sensitivity is defined as the correct identification of subsequent remitters/responders, and specificity as the correct identification of subsequent non-remitters/non-responders. The positive predictive value (PPV) is the probability that a patient with early improvement shows subsequent remission/response; the negative predictive value (NPV) is the probability that a patient without early improvement shows subsequent non-remission/non-response. Total accuracy is the proportion of patients whose 2-week improvement status (early improvement or no early improvement) predicts subsequent remission and response at discharge.
The cut-off point (percentage improvement of PANSS total score within the first 2 weeks) showing the highest total accuracy was used to divide the sample into patients with and without early improvement. The PANSS total scores and subscores of these two patient groups at discharge were then compared using the t-test. A p-value < 0.05 was considered to be statistically significant.
The SPSS 12.0 Software for Windows was used for all statistical analyses.
3. Results
3.1. Patients characteristics
In total, 474 patients with schizophrenia spectrum disorders were enrolled in the entire multi-centre follow-up programme. Forty-six patients (10%) dropped out for different reasons (e.g. withdrawal of informed consent, retrospective violation of inclusion criteria, incomplete information). A total of 191 (40%) patients were excluded from the present analysis: 79 (17%) because they were discharged from hospital within 20 days after admission; 77 (16%) because they did not fulfil diagnostic criteria for ICD-10 schizophrenia (F20) and 35 (7%) because they already fulfilled remission criteria at the time of admission.
Therefore, data from 237 subjects (141 [59.5%] males, 96 [40.5%] females) were available for analysis. Two hundred and three subjects (86%) met the criteria for DSM-IV schizophrenia and 34 (14%) for schizophreniform disorder. The mean age (± standard deviation) was 34.5 (± 11.4) years and the mean duration of illness 10.3 (± 10.4) years. The mean duration of inpatient treatment was 66.4 (± 46.3) days, and ranged widely from 21 to 431 days. Seventy-seven patients (33%) were suffering from their first episode of schizophrenia. Patients were treated as follows: 65% received first-generation antipsychotics, 55% second-generation antipsychotics, 55% tranquilizers and 8% mood-stabilizers. Antidepressants were administered to 19% of the sample.
3.2. Time course of clinical improvement
Psychopathological characteristics at admission, week 2 and discharge, and remission rates at week 2 and discharge, are shown in Table 1. All symptom dimensions (positive, negative and general psychopathology symptoms) improved both from admission to week 2 and from week 2 to discharge.
Table 1 Time course of clinical improvement
The frequency of early improvement depends on the definition applied, i.e. the required percentage symptom reduction of PANSS total score after 2 weeks of treatment: 64% of the sample showed a reduction of at least 10%; 40% of the sample of at least 20%; 22% of the sample of at least 30%; 8% of the sample of at least 40%; and 4% of the sample of at least 50%. As expected, the stricter the definition, the lower the percentage of patients with early improvement.
3.3. Early improvement as a predictor of remission at discharge
In order to examine the ability of early improvement to predict remission at discharge, ROC curves were generated for the reduction of psychopathological symptom scores after the first 2 weeks of treatment (Fig. 1). The improvement of the PANSS total score showed the highest (AUC = 0.659) and the improvement of the PANSS negative score the lowest (AUC = 0.540) predictive validity of all observed variables.
Fig. 1 Receiver-operating-characteristic for the reduction of the PANSS total score and subscores from admission to week 2 as a predictor for remission at discharge.
Total accuracy, sensitivity (correct identification of remission or response), specificity (correct identification of non-remission or non-response), positive and negative predictive value of different cut-off points (percentage improvement of the PANSS total score) for predicting remission at discharge are shown in Table 2. A cut-off point of 20% showed the highest total accuracy (65%), with a sensitivity of 53% and a specificity of 76%.
Table 2 Percentage reduction of PANSS total score from admission to week 2 as a predictor for remission at discharge
At discharge, patients with early improvement (defined as a reduction of at least 20% in the PANSS total score after the first 2 weeks of treatment) showed lower values in the PANSS total score (47.7 vs 54.8, T = 3.802, p < 0.001), and in the PANSS subscores for positive symptoms (9.4 vs 11.4, T = 4.725, p < 0.001), negative symptoms (13.9 vs 16.6, T = 3.161, p = 0.002) and general psychopathology (24.4 vs 26.8, T = 2.604, p = 0.010), than those without early improvement.
3.4. Early improvement as a predictor of response at discharge
In order to examine the ability of early improvement to predict response at discharge, ROC curves were generated for the reduction of psychopathological symptom scores after the first 2 weeks of treatment (Fig. 2). The improvement of the PANSS total score showed the highest (AUC = 0.737) and the improvement of the PANSS negative score the lowest (AUC = 0.616) predictive validity of all observed variables.
Fig. 2 Receiver-operating-characteristic for the reduction of the PANSS total score and subscores from admission to week 2 as a predictor for response at discharge.
Total accuracy, sensitivity (correct identification of remission or response), specificity (correct identification of non-remission or non-response), positive and negative predictive value of different cut-off points (percentage improvement of the PANSS total score) for predicting response at discharge are shown in Table 3. A cut-off point of 30% showed the highest total accuracy (76%), with a sensitivity of 47% and a specificity of 90%.
Table 3 Percentage reduction of PANSS-total score from admission to week 2 as a predictor for response at discharge
At discharge, patients with early improvement (defined as a reduction of at least 30% in the PANSS total score within the first 2 weeks of treatment) showed lower values in the PANSS total score (46.8 vs 53.4, T = 2.924, p = 0.004), and in the PANSS subscores for positive symptoms (9.4 vs 10.9, T = 2.753, p = 0.006), negative symptoms (13.6 vs 16.0, T = 2.404, p = 0.017) and general psychopathology (23.7 vs 26.4, T = 2.346, p = 0.020), than those without early improvement.
4. Discussion
4.1. Predictive validity of early improvement
The aims of the present study were:
• to determine the predictive validity of early improvement for remission and response at discharge;
• to identify the most useful criterion for the definition of early improvement.
This study evaluated a sample of inpatients receiving treatment under naturalistic conditions and replicated the results of recent studies that showed that early improvement is a predictor of subsequent treatment response [Reference Ascher-Svanum, Nyhuis, Faries, Kinon, Baker and Shekhar4,Reference Chang, Lane, Yang and Huang7,Reference Corell, Malhotra, Kaushik, McMeniman and Kane8,Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18,Reference Leucht, Bush, Kissling and Kane21–Reference Lin, Chou, Lin, Hsu, Chen and Lane23]. However, in contrast to the predictive validity for response (AUC = 0.737), the discriminative ability of early improvement for the prediction of remission failed to achieve a reasonable value (AUC = 0.659) [Reference Weinstein and Fineberg31].
The reduction of the PANSS total score within the first 2 weeks of inpatient treatment showed a higher validity for predicting subsequent treatment success than the different subscores, in particular than the reduction of the PANSS negative score. This is in line with the results of Corell et al. [Reference Corell, Malhotra, Kaushik, McMeniman and Kane8], who also found that the early improvement of negative or affective symptoms has a low predictive validity.
The fact that the predictive values in the present study are lower than those from previous studies [Reference Ascher-Svanum, Nyhuis, Faries, Kinon, Baker and Shekhar4,Reference Chang, Lane, Yang and Huang7,Reference Corell, Malhotra, Kaushik, McMeniman and Kane8,Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18,Reference Leucht, Bush, Kissling and Kane21–Reference Lin, Chou, Lin, Hsu, Chen and Lane23] can be explained by differences in the study samples and treatment regimes. In contrast to randomized [Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18,Reference Leucht, Bush, Kissling and Kane21,Reference Leucht, Shamsi, Bush, Kissling and Kane22] or open clinical drug trials [Reference Ascher-Svanum, Nyhuis, Faries, Kinon, Baker and Shekhar4,Reference Chang, Lane, Yang and Huang7,Reference Corell, Malhotra, Kaushik, McMeniman and Kane8,Reference Lin, Chou, Lin, Hsu, Chen and Lane23], naturalistic treatment allows a change of medication if a patient does not show early improvement. This could result in a more favourable treatment outcome of patients without early improvement and thus a lower predictive validity of early improvement. In fact, the frequencies of remission and response were higher in the present study than in the study by Kinon et al. [Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18], for example. However, the effects of different treatment regimes were not systematically controlled for in the present study. As a final point, one has to note that the present study included a high percentage of first episode patients (33%), who usually show a more favourable treatment outcome than patients with multiple episodes [Reference Jäger, Riedel, Messer, Laux, Pfeiffer and Naber14].
4.2. Sensitivity and specificity
An improvement of 20% in the PANSS total score after the first 2 weeks was identified as the most accurate cut-off for predicting remission at discharge (total accuracy: 65%; sensitivity: 53%; specificity: 76%). In contrast, a cut-off point of 30% gave the most accurate prediction of response at discharge (total accuracy: 76%; sensitivity: 47%; specificity: 90%). In accordance with previous findings [Reference Ascher-Svanum, Nyhuis, Faries, Kinon, Baker and Shekhar4,Reference Corell, Malhotra, Kaushik, McMeniman and Kane8,Reference Leucht, Busch, Hamann, Kissling and Kane20], specificity (correct identification as subsequent non-remitter or non-responder) was higher than sensitivity (correct identification as subsequent remitter or responder).
In this context, one has to consider that the time course of response to antipsychotic treatment is still unclear. Although several neurobiological hypotheses exist [Reference Carlsson and Lindqvist6,Reference Seeman and Lee29], the precise mechanism of action of antipsychotic drugs is poorly understood. Although recent studies rejected the “delayed onset hypothesis” of antipsychotic drug action and indicated that a substantial amount of treatment effects occurs in the first weeks of treatment [Reference Agid, Kapur, Arenovich and Zipursky1,Reference Leucht, Busch, Hamann, Kissling and Kane20], the study by Emsley et al. [Reference Emsley, Rabinowitz and Medori10] reported a widely varying time to antipsychotic response in a sample of patients with first episode schizophrenia.
4.3. Implications for treatment guidelines
Schizophrenia is a heterogeneous disorder and includes both cases with full remission between episodes and those with a chronic course [Reference Jäger, Bottlender, Strauss and Möller13,Reference Möller, Hohe-Schramm, Cording-Tömmel, Schmid-Bode, Wittchen and Zaudig26]. Although early improvement is a robust predictor of subsequent treatment success, its discriminative power seems not to exceed values of conventional predictor models, which are based upon the psychopathological symptom level at the start of antipsychotic treatment [Reference Crespo-Facorro, Pelayo-Terán, Pérez-Iglesias, Ramírez-Bonilla, Mártinez-García and Pardo-García9,Reference Gaebel12,15,Reference Lambert, Naber, Eich, Schacht, Linden and Schimmelmann19]. Thus, one can argue that the additional predictive value of early improvement is questionable. However, there is a lack of investigations that combine early improvement with additional possible predictor variables.
On the background of their findings on the predictive value of early improvement, both Kinon et al. [Reference Kinon, Chen, Ascher-Svanum, Stauffer, Kollack-Walker and Sniadecki18] and Leucht et al. [Reference Leucht, Bush, Kissling and Kane21,Reference Leucht, Shamsi, Bush, Kissling and Kane22] proposed a change of treatment if there is no early improvement. This would imply a revision of the current treatment guidelines for schizophrenia [3,Reference Falkai, Wobrock, Lieberman, Glenthoi, Gattaz and Möller11,27,28]. However, the crucial question is whether or not patients without early improvement will benefit from a change of antipsychotic medication, and it is still unclear how best to treat such patients. There is a substantial need for controlled empirical studies in this field of research.
4.4. Limitations
The design of the present study can be regarded as either a strength or a limitation. The naturalistic design does not allow the results to be sufficiently controlled for the effect of different pharmacological and psychological treatments. The sample was heterogeneous and included patients at different stages of schizophrenic illness, for example 33% were first-episode patients. However, it was the aim of the study to examine an unselected and representative sample of inpatients routinely treated in a psychiatric hospital. Since the present sample only included inpatients, it may not be representative for all patients suffering from schizophrenia.
The duration of inpatient treatment varied substantially from 21 to 431 days. There were therefore large differences in the time from week 2 (as the measure for early improvement) to discharge (as the final time point). Patients with less than 21 days’ inpatient treatment were excluded from the present analyses.
5. Conclusion
Early improvement is a predictor of subsequent remission and response in the treatment of schizophrenia. The present study in an unselected sample of inpatients being treated under naturalistic conditions replicated the findings from clinical drug trials. However, the discriminative power of early improvement does not seem to exceed values of conventional predictor models, which are based upon the psychopathological symptom level at the start of antipsychotic treatment. Further studies should examine whether patients without early improvement will benefit from an early change of antipsychotic medication.
Acknowledgement
The network study was conducted at 14 psychiatric hospitals: Aachen (P. Hoff, K. Podoll), Augsburg (M. Schmauß, T. Messer, M. Eichinger), Berlin (I. Heuser, M. Jockers-Scherübl), Bonn (W. Maier, K.-U. Kühn, M.R. Lemke, R. Hurlemann, W.P. Hornung, E. Rosen), Cologne (J. Klosterkötter, W. Huff), Düsseldorf (W. Gaebel, A. Klimke, M. Eickhoff, M. von Wilmsdorff), Essen (M. Gastpar, V. Reißner), Gabersee (G. Laux, B. Hermann, B. Plichta), Göttingen (E. Rüther, D. Degner), Haar (H. Pfeiffer, M. Albus, S. Scharf-Büssing), Hamburg (D. Naber, D. Golks), Mainz (L.G. Schmidt, B. Kaufmann-Grebe), Munich (H.-J. Möller, R. Bottlender, M. Riedel, M. Jäger, C. Schorr, B. Schillinger, C. Mirlach), Tübingen (G. Buchkremer, M. Mayenberger). The authors thank Jacqueline Klesing for English language editing of the manuscript.
The study was performed within the framework of the German Research Network on Schizophrenia, which is funded by the German Federal Ministry for Education and Research BMBF (Grant 01 GI 0233).
Comments
No Comments have been published for this article.