Antidepressants for the negative symptoms of schizophrenia

Criteria for considering studies for this review

Search methods for identification of studies

Data collection and analysis

[For definitions of terms used in this and other sections, please refer to the Glossary]

1. Study selection 
 We (CR, SL) independently inspected all citations of studies identified by the search. Where disagreement occurred we resolved this with WK, or, where any doubt remained, obtained the full article for further inspection. We then independently inspected these articles to assess their relevance to this review. Again, if the disagreement could not be resolved from published information, we added the article to the list of those awaiting assessment and contacted the authors of the study for clarification.

2. Quality assessment 
 We assessed the methodological quality of the trials included in this review using the criteria described in the Cochrane Handbook (Higgins 2005). This simple set of criteria is based on the evidence of a strong association between overestimation of effect and poor concealment of allocation (Schulz 1995) and is defined as follows:

A. Low risk of bias (adequate allocation concealment) 
 B. Moderate risk of bias (some doubt about the results) 
 C. High risk of bias (inadequate allocation concealment)

For the purpose of the analysis in this review, trials were included if they meet criterion A or B of the Handbook. We gave trials not described as randomised by the authors a C rating and excluded them.

3. Data extraction 
 We (CR, SL) independently extracted data from selected trials. Where disagreement occurred, we attempted resolution by discussion. If doubt remained and further information was necessary to resolve the dilemma, we did not enter the data, but contacted the authors of the studies for further clarification. We documented reasons for excluding references from the review.

4. Data management 
 4.1 Intention to treat 
 For studies that did not specify the reasons for people leaving the study early (dropping out), we assumed them to have no change in the clinical outcome variables. A problem with increasing drop‐out rates in randomised trials on drug treatment for schizophrenia has been reported (Wahlbeck 2001). Since there is no evidence regarding what degree of attrition renders data analysis impossible, all trials were included in the main analysis. However, it was tested in a sensitivity analysis whether the exclusion of trials with drop‐out rates higher than 50% significantly changed the results of the primary outcome parameters. When insufficient data were provided to identify the original group size (prior to drop‐outs), we contacted the authors and added the trials to the list of those awaiting assessment.

4.2 Crossover design 
 We expected that some trials would use a crossover design. In order to exclude the potential additive effect on these trials in any further stages, we only analysed data from the first stage.

4.3 Data types 
 Outcomes are assessed using continuous measures (for example changes on a behaviour scale), categorical measures (for example, one of three categories on a behaviour scale, such as 'little change', 'moderate change' or 'much change') or dichotomous measures (for example, either 'no important changes' or 'important changes' in a person's behaviour). Currently RevMan does not support categorical data, so they could not be analysed as such.

4.3.1 Dichotomous data: Where possible, efforts were made to convert outcome measures to dichotomous data. This may be done by identifying cut off points on rating scales and dividing subjects accordingly into 'clinically improved' or 'not clinically improved'. If the authors of a study had used a predefined cut off point for determining clinical effectiveness we used this where appropriate. Otherwise we generally assumed that a 50% reduction of a scale (e.g. the Brief Psychiatric Rating Scale ‐ Overall 1962) or a rating of 'at least much improved' according to the Clinical Global Impression Scale (Guy 1976) could be considered as a clinically significant response.

For dichotomous outcomes, we estimated a relative risk (RR) with the 95% confidence interval (CI) based on a random effects model. When overall results were significant we calculated the number needed to treat (NNT) and/or the number needed to harm (NNH) as the inverse of the absolute risk reduction. It has been shown that RR is more intuitive (Boissel 1999) than odds ratios and that odds ratios tend to be interpreted as RR by clinicians (Deeks 2000). This misinterpretation then leads to an overestimate of the effect.

4.3.2 Continuous data 
 4.3.2.1 Normal distribution: Continuous data on outcomes in trials relevant to mental health issues are often not normally distributed. To avoid the pitfall of applying parametric tests to non‐parametric data the following standards were applied to data derived from continuous measures of endpoint ('state' data). The criteria were used before inclusion: 
 When a scale starts from zero, the standard deviation, when multiplied by two, is less than the mean (as the mean is otherwise unlikely to be an appropriate measure of the centre of the distribution ‐ Altman 1996). Endpoint scores on scales often have a finite starting and end point and this rule can be applied.

When continuous data are presented on a scale which includes a possibility of negative values (such as change on a scale) it is impossible to tell whether data are non‐normally distributed (skewed) or not. It is thus preferable to use scale endpoint data, which typically cannot have negative values. If endpoint data were not available, the reviewers chose to use change data, because the statistics used in Metaview are rather robust with regard to skew. If a scale starts from a positive value (such as PANSS, which can have values from 30‐210) the calculation described above in (b) should be modified to take the starting point of the scale into account. In these cases skew is present if 2SD>(S‐Smin), where S is the mean score and Smin is the minimum score.

4.3.2.2 Intention‐to‐treat versus completer analyses: in the case of continuous data it was assumed that in many cases an intention‐to‐treat analysis would not be available. For this reason we had to analyse the data as they were presented in the original publications.

4.3.2.3 Summary statistics: for continuous outcomes, we estimated a weighted mean difference (WMD) between groups. Again, we used a random effects model. Whenever possible we took the opportunity to make direct comparisons between trials that used the same measurement instrument to quantify specific outcomes. Where continuous data were presented from different scales rating the same effect, the data were summarized, calculating a standardized mean difference (SMD).

4.3.2.4 Rating scales: a wide range of instruments are available to measure mental health outcomes. These instruments vary in quality and many are not valid, or are even ad hoc. Some minimum standards must be set for outcome instruments. We only included continuous data from rating scales if the measuring instrument had been described in a peer‐reviewed journal (Marshall 2000), the instrument was either a self report or completed by an independent rater or relative (not the therapist), and the instrument could be considered as a global assessment of an area of functioning. However, as it was expected that therapists would frequently also be the raters, we commented on the data with the label 'prone to bias'.

4.4 Data display 
 We entered data into RevMan in such a way that the area to the left of the line of no effect indicated a favourable outcome for the antidepressant/antipsychotic combination group.

4.5 Cluster trials 
 Studies increasingly employ 'cluster randomisation' (such as randomisation by clinician or practice) but analysis and pooling of clustered data poses problems. Firstly, authors often fail to account for intra class correlation in clustered studies, leading to a 'unit of analysis' error (Divine 1992) whereby p values are spuriously low, confidence intervals unduly narrow and statistical significance overestimated. This causes type I errors (Bland 1997, Gulliford 1999).

Where clustering was not accounted for in primary studies, we presented the data in a table, with a (*) symbol to indicate the presence of a probable unit of analysis error. In subsequent versions of this review we will seek to contact first authors of studies to obtain intra‐class correlation co‐efficients of their clustered data and to adjust for this using accepted methods (Gulliford 1999). Where clustering has been incorporated into the analysis of primary studies, we will also present these data as if from a non‐cluster randomised study, but adjusted for the clustering effect.

We have sought statistical advice and have been advised that the binary data as presented in a report should be divided by a 'design effect'. This is calculated using the mean number of participants per cluster (m) and the intraclass correlation co‐efficient (ICC) [Design effect = 1+(m‐1)*ICC] (Donner 2002). If the ICC was not reported it was assumed to be 0.1 (Ukoumunne 1999).

5. Heterogeneity 
 Firstly, we considered all of the included studies within any comparison to judge clinical heterogeneity. Then we visually inspected graphs used to investigate the possibility of statistical heterogeneity and supplemented this by using, primarily, the I‐squared statistic. This provides an estimate of the percentage of variability due to heterogeneity rather than chance alone. Where the I‐squared estimate was greater than or equal to 75%, we interpreted this as indicating the presence of high levels of heterogeneity (Higgins 2003). If inconsistency was high, we did not summate the data, but presented it separately and investigated reasons for heterogeneity.

6. Publication bias 
 We entered data from all included trials into a funnel graph (trial effect versus trial size or 'precision') in an attempt to investigate the likelihood of overt publication bias. We did not undertake a formal test of funnel plot asymmetry (suggesting potential publication bias (Egger 1997)).

7. Sensitivity analysis 
 We examined in a sensitivity analysis whether the exclusion of trials with attrition rates higher than 50% led to a significant change in the results.

Description of studies

Please also see Tables of included and excluded studies.

1. Excluded studies 
 We excluded 112 studies. Reasons for exclusion were that the studies did not include people with pronounced negative symptoms (n=30), included people with schizophrenia and depressive symptoms (n=29), did not fulfil the predefined intervention criteria 'antipsychotic plus antidepressant' versus 'antipsychotic plus placebo' (n=27), included people with negative and/or positive symptoms (n=11), did not randomise adequately or at all (n=8), or included medications never marketed (n=5). Finally, some studies had no useable data (n=2).

2. Studies awaiting assessment 
 We are looking forward to receiving data from one trial adding fluoxetine to haloperidol treatment in schizophrenia (Lee 1997), but have not yet received an answer to our letter to the first author.

3. Ongoing and planned studies 
 We did not identify any ongoing or planned studies.

4. Included studies 
 We identified five studies for inclusion in this review.

4.1 Study design 
 All studies used a parallel group design.

4.2 Length of trials 
 All of the five studies were short term trials: Hayashi 1997 (5 weeks), Silver 1992 (7 weeks), Collins 1967 (12 weeks), Spina 1994 (12 weeks) and Jockers 2005 (12 weeks).

4.3 Participants 
 Five studies provided usable data and included a total of 190 people (Collins 1967 ‐ 58 participants, Hayashi 1997 ‐ 39 participants, Jockers 2005 ‐ 29 participants, Silver 1992 ‐ 30 participants and Spina 1994 ‐ 34 participants). Participants were diagnosed with schizophrenia according to DSM‐III R (Spina 1994, Silver 1992) and DSM‐IV (Hayashi 1997, Jockers 2005); the oldest study (Collins 1967), did not mention diagnostic criteria for schizophrenia. All studies examined patients with pronounced negative symptoms although they used different criteria for defining this. Silver 1992 used 'DSM‐III‐R criteria for chronic schizophrenia plus a score of at least "moderate" on one of the five global scales of the SANS. Spina 1994 used similar criteria plus a HAM‐D baseline score of lower than 20 in order to exclude a major depressive episode. Hayashi 1997's definition was a DSM‐IV diagnosis of schizophrenia plus moderate to severe negative symptoms in the PANSS while Jockers 2005 used the DSM‐IV diagnosis of chronic schizophrenia with at least 20 points on the negative subscale of the PANSS, at least a score of 4 on the CGI and a HAM‐D score lower than 12.

4.4. Interventions 
 Collins 1967 compared perphenazine and amitriyptyline with perphenazine and placebo. In Hayashi 1997 participants continued with their previous antipsychotic medication and the addition of mianserin was compared with the addition of placebo. Jockers 2005 continued previous antipsychotic medication and compared the addition of paroxetine 30 mg/d with the addition of placebo. Silver 1992 continued previous antipsychotic medication and added fluvoxamine or placebo. Spina 1994 compared the addition of fluoxetine 20 mg/d with the addition of placebo to the previous antipsychotic treatment, which was continued during the study period. 
 
 4.5 Outcomes 
 A variety of scales were used to assess clinical response and adverse events although trialists also reported some non‐scale‐derived binary outcomes.

4.5.1 Outcome scales: details of scales that provided usable data are shown below. Reasons for exclusion of data from other instruments are listed under "outcomes" in the "included studies" table.

4.5.1.1 Global functioning 
 Clinical Global Impression ‐ CGI (Guy 1976) 
 A rating instrument commonly used in studies on schizophrenia that enables clinicians to quantify severity of illness and overall clinical improvement during therapy. A seven‐point scoring system is usually used, with low scores indicating decreased severity and/or greater recovery.

4.5.1.2 Mental state 
 Brief Psychiatric Rating Scale ‐ BPRS (Overall 1962) 
 A brief rating scale used to assess the severity of a range of psychiatric symptoms, including psychotic symptoms. The scale has 18 items, and each item can be rated on a seven‐point scale varying from "not present" (1) to "extremely severe" (7). Scoring goes from 18 to 126.

Positive and Negative Syndrome Scale ‐ PANSS (Kay 1986) 
 A 30‐item schizophrenia scale which is subdivided into three subscales for measuring the severity of general psychopathology, positive symptoms, and negative symptoms. Each item can be rated on a seven‐point scoring system varying from absent (1) to severe (7). Scoring goes from 30 to 210. Lower scores indicate lesser severity.

Hamilton Rating Scale for Depression ‐ HAM‐D (Hamilton 1960) 
 The HAM‐D is a well‐established 17‐item scale for the measurement of depression and is sensitive to change.

Scale for the Assessment of Negative Symptoms ‐ SANS (Andreasen 1984) 
 A six‐point (0‐5) scoring system can be used for each global rating of alogia, affective blunting, avolition‐apathy, anhedonia‐asociality, and attention impairment. A low score indicates low levels of psychotic symptoms.

Wing Rating Scale (Wing 1961) 
 This scale consists of two parts. The first rates four typical symptoms of the schizophrenic mental state, based on a brief psychiatric interview, on a 5‐point scale. The second is a 12‐item behavioural schedule that rates on a three‐point scale. The more acute the patient's condition, the higher the score on this scale.

4.5.1.3 Adverse events 
 No scales for adverse events provided usable data. Reasons for exclusion of data are listed under "outcomes" in the "included studies" table.

4.5.2 Missing outcomes 
 Because all of the studies for which data were available were short term studies, we know nothing about the medium and long term outcomes for this group of people. 
 There are no data on important outcomes such as service utilisation outcomes (admission to the hospital, days in the hospital, change in hospital status), behaviour (compliance with treatment), social functioning, economic outcomes, quality of life and cognitive functioning in either of the included studies.

Risk of bias in included studies

1. Randomisation 
 The included studies were reported as randomised, but trialists did not describe the means of randomisation, so all studies were allocated to category 'B'.

2. Blinding 
 All studies were double‐blind, although none gave assurances that blinding was maintained.

3. Follow up 
 Only one study (Collins 1967) provided unclear information on loss to follow up. Hayashi 1997, Silver 1992 and Spina 1994 indicated the number of participants who left the study early and the reason for leaving. Jockers 2005 reported the total number of participants leaving the study early, but reasons for leaving were not provided.

4. Jaded scores 
 All studies achieved a Jaded score of three (the top Jaded score is five).

Effects of interventions

1. The search 
 The search strategy identified 432 citations. Of these, 139 articles were ordered and inspected. We were ultimately able to include only five studies.

2. COMPARISON 1: ANTIPSYCHOTICS PLUS ANTIDEPRESSANTS VERSUS ANTIPSYCHOTICS PLUS PLACEBO

2.1 Leaving the study early 
 Overall results were equivocal with no significant difference between the combination of antipsychotics and antidepressants and the control group in terms of leaving the study early for any reason (n=90, 3 RCTs, RR 3.0 CI 0.35 to 26.04). Again, when it came to leaving specifically due to adverse events, two studies (Silver 1992 and Spina 1994) found no significant difference between treatment groups (n=64, 2 RCTs, RR 5.0 CI 0.26 to 97.0). Spina 1994 also reports no difference in the number of participants leaving early due to inefficacy (n=34, 1 RCT, RR 3.0 CI 0.13 to 68.84).

2.2 Clinical response 
 Only Spina 1994 provided data from the Clinical Global Impression Scale (improvement and severity). Participants treated with antipsychotic and antidepressant medication had a statistically significantly greater improvement in global state (n=30, 1 RCT, WMD ‐1.0 CI ‐1.61 to ‐0.39) and had a significantly lower severity at endpoint (n=30, 1 RCT, WMD ‐0.9 CI ‐1.55 to ‐0.25) than those treated with antipsychotics and placebo.

2.3 Mental state 
 2.3.1 General 
 Silver 1992 and Spina 1994 provided data for BPRS total endpoint scores. No statistically significant difference in scores were found between treatment groups (n=60, 2 RCTs, WMD ‐4.89 CI ‐13.27 to 3.50). Although data were heterogeneous, no reasons could be identified and we considered the heterogeneity to be due to chance. Hayashi 1997 also found no significant difference in average score change on the BPRS (n=26, 1 RCT, WMD 0.1 CI ‐3.81 to 4.01). 
 
 2.3.2 Positive symptoms 
 Four studies presented data for the effects on positive symptoms but only Hayashi 1997 and Jockers 2005 presented usable data. Data from Silver 1992 and Spina 1994 were skewed. Hayashi 1997 found no overall effect on the BPRS positive subscale (average change, n=26, 1 RCT, WMD 0.30 CI ‐3.61 to 4.21). Jockers 2005 also found no overall effect on the PANSS positive subscale average change score.

2.3.3 Negative symptoms 
 2.3.3.1 Clinically significant improvement 
 Spina 1994 defined 'clinically significant improvement of negative symptoms' as a 20% or more decrease in SANS total score at endpoint compared to baseline and Hayashi 1997 used a 30% or more reduction on the same scale for their definition. Spina 1994 found that significantly more participants receiving adjunct antidepressants had a clinically significant improvement in negative symptoms compared with the antipsychotic plus placebo group (n=34, 1 RCT, RR 0.41 CI 0.23 to 0.73, NNT 2 CI 1 to 3). Hayashi 1997 yielded similar, favourable results (n=26, 1 RCT, RR 0.69 CI 0.48 to 0.99, NNT 3 CI 2 to 25). The result still significantly favours antidepressants when the two studies are combined (n=60, 2 RCTs RR 0.56 CI 0.32 to 0.97, NNT 3 CI 3 to 34).

2.3.3.2 Total endpoint scores 
 Jockers 2005 used the PANSS and results between groups were equivocal (n=25, 1 RCT, WMD ‐2.27 CI ‐7.10 to 2.56). Silver 1992 and Spina 1994, however, found those in the treatment groups had significantly lower SANS scores than the control group (n=60, 2 RCTs, WMD ‐13.92 CI ‐19.93 to ‐7.92).

2.3.3.3 Average change in total scores 
 Hayashi 1997 presented SANS data and found no statistically significant difference between groups (n=26, 1 RCT, WMD ‐4.60 CI ‐11.10 to 1.90).

2.3.3.4 Various average endpoint scores 
 Using the SANS, data on different aspects of negative symptoms were reported by Spina 1994 and Silver 1992. Spina 1994 found those receiving adjunct antidepressants had a statistically significant lower score for 'affective flattening' than those in the control group (n=30, 1 RCT, WMD ‐7.0 CI ‐10.37 to ‐3.63) and a similar result for 'avolition' (n=30, 1 RCT, WMD ‐3.0 CI ‐5.04 to ‐0.96). Both Spina 1994 and Silver 1992 presented 'anhedonia' subscores but found no statistically significant difference between groups, although p‐value could be regarded as borderline (n=60, 2 RCTs, WMD ‐2.58 CI ‐5.42 to 0.27). Spina 1994 also presented scores for 'attention' and again results were equivocal (n=30, 1 RCT, WMD ‐0.50 CI ‐2.28 to 1.28).

2.3.3.5 Various average change scores 
 Hayashi 1997 presented change data on these different aspects of negative symptoms. Affective flattening appeared to be better in those receiving antidepressants but the difference was borderline and not statistically significant (n=26, 1 RCT, WMD ‐2.80 CI ‐5.92 to 0.32). The average change of the 'alogia' subscores was, however, statistically different and favoured the treatment group (n=26, 1 RCT, WMD ‐3.0 CI ‐5.14 to ‐0.86). No difference in scores were found for avolition (n=26, 1 RCT, WMD 0.1 CI ‐2.02 to 2.22), anhedonia (n=26, 1 RCT, WMD ‐3.0 CI ‐1.97 to 1.37) or attention (n=26, 1 RCT, WMD 0.30 CI ‐1.65 to 2.25).

2.3.4 Depression 
 One study, Spina 1994, presented usable data on depressive symptoms. Total endpoint scores on the HAM‐D were significantly lower in the antipsychotic plus antidepressant group (n=30, 1 RCT, WMD ‐3.0 CI ‐5.54 to ‐0.46).

2.4 Behaviour 
 Collins 1967 used the Wing Scale to assess behaviour. Social withdrawal average change scores did not differ significantly between the two treatment groups (n=58, 1 RCT, WMD ‐0.93 CI ‐2.33 to 0.47). 'Socially embarrassing behaviour' was, however, affected by use of antidepressants. The average change scores were significantly different and favoured the treatment group (n=58, 1 RCT, WMD ‐0.90 CI ‐1.59 to ‐0.21).

2.5 Adverse events 
 3.5.1 General 
 Two studies (Collins 1967 and Hayashi 1997) found no statistically significant difference between treatment groups in regards to the outcome 'at least one adverse event' (n=84, 2 RCTs, RR 1.8 CI 0.66 to 4.90).

2.5.2 Specific ‐ movement disorders 
 There were several measures of movement disorder, none showed any statistically significant difference: 'any extrapyramidal adverse event' (n=60, 2 RCTs, RR 5.0 CI 0.26 to 97.00), 'akathisia' (n=34, 1 RCT, RR 3.0 CI 0.13 to 68.84), 'parkinsonism' (n=92, 2 RCTs, RR 0.94 CI 0.21 to 4.24) and 'tremor' (n=34, 1 RCT, RR 3.00 CI 0.13 to 68.84).

2.5.3. Specific ‐ other adverse events 
 No statistically significant differences were found for the two other adverse effects reported, 'dizziness' (n=26, 1 RCT, RR 3.00 CI 0.13 to 67.51) and 'somnolence' (n=26, 1 RCTs, RR 5.00 CI 0.26 to 95.02).

2.6. Additional drug use 
 Two studies (Spina 1994 and Hayashi 1997) continued anticholinergic medication when present prior to study entry, but these studies did not indicate the number of patients actually receiving anticholinergics. In the Silver 1992 study, all participants received anticholinergics; in the Collins 1967 study, no participant received anticholinergics while in Jockers 2005 the use of anticholinergics remains unclear.

2.7 Publication bias 
 Entering data from all included trials into a funnel plot was not deemed reasonable due to the small number of included studies.

2.8 Sensitivity analysis 
 No trial had to be excluded due to attrition rates higher than 50%; therefore no sensitivity analysis was done.

Characteristics of included studies [ordered by study ID]

Characteristics of excluded studies [ordered by study ID]

Internal sources

• Freistaat Bayern, Germany.

External sources

• German Federal Ministry for Education and Research (BMBF) (grant 01 GI 993x), Germany.