Annals of General Psychiatry Treatment of Psychotic Symptoms in Bipolar Disorder with Aripiprazole Monotherapy: a Meta-analysis

Background: We present a systematic review and meta-analysis of the available clinical trials concerning the usefulness of aripiprazole in the treatment of the psychotic symptoms in bipolar disorder.


Background
The treatment of bipolar disorder (BD) is difficult since the illness itself is complex [1][2][3][4][5][6][7]. In the BD clinical picture, psychotic features are a very frequent manifestation although they are not considered to constitute a core fea-ture of the disorder. Delusions are relatively more common than hallucinations. However, it is reported that unipolar-depressed patients who later 'convert' to BD over time, as well as bipolar depressives, manifest more frequently psychotic features and pathological (psychotic) guilt [8,9]. Additionally, within the BD patient group it has been suggested (but not proven) that those patients with a history of psychotic symptoms suffer from a greater impairment regarding the neuropsychological performance especially concerning verbal memory and executive function [10,11].
Psychotic features include delusions and hallucinations and both can be mood congruent or non-congruent depending on their content. Mood congruent psychotic features include those entirely consistent with the thought content (either manic or depressive) while mood incongruent features are largely unrelated to thought content. Overexaggerated thoughts of guilt, sin, worthlessness, poverty and somatic health, or on the contrary thoughts of exceptional mental and physical fitness or special talents, wealth, some kind of grandiose identity or importance are mood congruent delusions, and even persecutory ideas or ideas of reference when in accord with the thought content can be considered to be mood congruent. Non-congruent delusions include nihilistic delusions (Cotard delusion or Cotard syndrome, negation delusion), bizarre delusions and sometimes the delusions can be so excessive that the identity itself changes. Psychotic symptoms have a profound effect on insight especially in depressive episodes which otherwise are characterised by a fair degree of insight. Psychotic features and the lack of insight might lead to the refusal of any treatment and to the need for an involuntary admission to a hospital.
Only during the last few years have antipsychotics and especially atypicals or second-generation antipsychotics (SGAs) gained a position in the treatment of BD [12,13]. Their efficacy against acute mania is reported to be independent of sedation or of their effect on psychotic symptoms. Olanzapine, risperidone, quetiapine, ziprasidone and aripiprazole are approved for the treatment of acute mania, quetiapine and the olanzapine-fluoxetine combination are approved for the treatment of acute bipolar depression, and olanzapine, quetiapine and aripiprazole are approved for maintenance phase treatment.
Although psychotic symptoms are common in bipolar patients, not all randomised controlled trials (RCTs) include their assessment and up to now there has been no review or meta-analysis on the efficacy of agents approved for the treatment of BD on these specific symptoms. The aim of the current review and meta-analysis was to focus on outcome measures of randomised controlled trial testing the efficacy of aripiprazole against psychotic symptoms in bipolar disorder. To the best of our knowledge no such analysis exists in the literature to date, and the reviews available [16][17][18][19][20][21][22][23][24][25] either do not include all the trials that have been conducted so far or do not focus on psychotic symptoms.

Search criteria
The first step of the search included a keyword search of Medline and the internet via Google with the words 'aripiprazole' and 'bipolar'.

Types of studies
The studies selected were RCTs with placebo or a comparator.

Data extraction
All data were extracted by the same author (KNF) from the full published paper or the clinical study report synopsis. In some cases some of the data were extracted or calculated from published meta-analysis or reviews.

Meta-analysis method
The following indices were calculated. Effect size (Cohen d) was calculated as the mean change divided by the standard deviation of the scale. It represents the difference between two groups in the amount of change.
Pooled standard deviation (SDp) was calculated by the following function: Where S p is the pooled standard deviation, n i is the sample size of the i th sample, s i is the standard deviation of the i th sample, and k is the number of samples being combined.
The Q test for testing the homogeneity of studies was computed by summing the squared deviations of each study's effect estimate from the overall effect estimate, weighting http://www.annals-general-psychiatry.com/content/8/1/27 the contribution of each study by its inverse variance. This was calculated only for the Young Mania Rating Scale (YMRS) and for placebo-controlled RCTs concerning the differential d versus placebo and for all studies concerning the absolute effect of aripiprazole. Not rejecting the homogeneity hypothesis leads to a fixed effects model because it is assumed that the estimated effect sizes only differ by sampling error, while the rejection of the homogeneity assumption leads to a random effects model that includes both intrastudy and interstudy variability.
The I 2 index measures the extent of true heterogeneity, and was calculated by dividing the difference between the result of the Q test and its degrees of freedom (k-1) by the Q value itself and multiplying by 100. This index can be interpreted as the percentage of the total variability in a set of effect sizes due to true heterogeneity, that is, to interstudy variability.
The Hunter-Schmidt meta-analysis software was used for the correction of the effect sizes (d) for sampling error and measurement error [29] concerning only the effect size which calibrated the difference between two groups in the amount of change (aripiprazole vs placebo).  [33], CN138008 [34] and CN138-077 or NCT00046384, which did not produce any results due to the small number of patients recruited. Two of them (CN138-007 and CN138-077/NCT00046384) included a fixed dosage while the others included a flexible dosage design. Rapid cycling patients were excluded from CN138-135/ NCT00095511. CN138-077/NCT00046384 did not produce any results due to the small number of patients recruited (29 in the aripiprazole arm and 27 in the placebo arm). Its design included also a fixed dosage, and it was prematurely closed because it was expected to produce negative results similar to CN138-007. All used YMRS as the primary outcome measure. Data on Positive and Negative Syndrome Scale (PANSS) is not reported by CN138009 and CN138007, while data for PANSS total only and some but not all subscales are reported by CN138074 and CN138135.

Acute mania/mixed episodes
The details of these studies (randomised patients, efficacy and safety sample, publications and results) are shown in Table 1. The baseline scores for all outcome scales are shown in Table 2. The similarity of baseline scores across trials and the similar pooled mean justified the pooling of all data concerning each arm across studies irrespective whether the specific study had a placebo or a comparator arm or not. The dropout rates are shown in Table 3. The changes in the PANSS scales scores are shown in Table 4. The effect sizes (d) are shown in Table 5. The side effects frequency is shown in Table 6. The forest plot is shown in Figure 1.
After including only the RCTs reporting PANSS data (CN138009 and CN138007 were excluded), in total 1,640 patients were randomised and the total efficacy sample included 1,534 patients (613 in the aripiprazole group, 447 the placebo group and 474 in active control groups) while the safety sample included 1,574 (631 in the aripiprazole group, 450 in the placebo group and 493 in active control groups) ( Table 1). The pooled dropout rate for the RCTs which reported PANSS data was 36.22% for aripiprazole, 43.62% for placebo and 42.83% for comparator at week 3, climbing to 56.81% and 62.87% at week 12 for aripiprazole and comparator, respectively. The dropout rates are shown in detail in Table 3.
The pooled d value for total PANSS score for all placebocontrolled studies with a fixed model was equal to 0.191 (95% CI 0.08 to 0.304), with a Q value equal to 5.024 (degrees of freedom (df) = 3; P = 0.170), which does not reject the homogeneity hypothesis. The I 2 is equal to 40.283, which means that less than half of observed variability in the effect sizes across studies is due to true heterogeneity. However a random model gives similar results (d = 0.194; 95% CI 0.05 to 0.34). The Hunter-Schmidt method reports d = 0.18 (95% CI 0.176 to 0.184) after correcting for sampling and measurement error (Table 5).

Acute bipolar depression
There were two 8-week placebo controlled RCTs (CN138-096/NCT00080314 and CN138-146/NCT00094432) concerning the use of aripiprazole in acute bipolar depression, and were both negative at study endpoint [35]. These included non-psychotic bipolar depressives and thus do not report data on psychotic symptoms.

Maintenance treatment
There was one placebo-controlled RCT (CN138-010/ NCT00036348) that studied aripiprazole in the maintenance phase [36]. Patients were stabilised with 15 to 30 mg of aripiprazole for 6 to 18 weeks and then randomised to a 1:1 ratio to aripiprazole or placebo for an additional 26 weeks. The baseline mean YMRS score was 2. For the PANSS total score, a numerical trend favoured aripiprazole over placebo at any time point. At week 26, the changes (mean ± SD) in PANSS total scores were 5.2 ± 14.57 for aripiprazole and 9.1 ± 13.24 for placebo (P = 0.077), for the PANSS cognitive subscale score were for aripiprazole, 0.8 ± 4.55; placebo, 2.5 ± 4.41; P = 0.014) and for the PANSS hostility subscale score for aripiprazole, 0.8 ± 2.73; placebo, 1.8 ± 2.65; P = 0.032). All except the total PANSS score showed a significant superiority of aripiprazole over placebo, with d = 0.28, 0.38 and 0.71, respectively.
The adverse events reported by aripiprazole-treated patients at an incidence ≥ 5% and twice the incidence of placebo during the maintenance phase were tremor (9.1%), acathisia (6.5%), vaginitis (6.4%), and pain extremity (5.2%). One aripiprazole-treated patient and one placebo-treated patient attempted suicide in the stabilisation and maintenance phases, respectively. There were no significant differences concerning the QTc, while aripiprazole-treated patients showed a significant drop in prolactin levels. Concerning weight gain, 13% of aripiprazole-treated patients put > 7% of weight in comparison to none in the placebo group.
This same trial (CN138-010/NCT00036348) was expanded and included also a 74-week placebo controlled extension phase [37], which included 66 of the 67 patients who completed the 26-week period. Unfortunately only 12 of them (5 in the placebo group and 7 in the aripiprazole group) completed the 74-week treatment period. The reasons for this high discontinuation rate varied and included lack of efficacy, side effects (very low percentage) and most importantly the very design and structure of the study (the study was closed by the sponsor when the prespecified number of relapses had been attained). Because of this and because detailed descriptive data are not reported, arriving at conclusions is very difficult. The mean dosage of aripiprazole at the end of the 74week period was 23.6 mg daily. It is reported that 29 out of the 66 patients relapsed (16 out of the 39 in the aripiprazole group (41%), and 13 out of the 27 in the placebo group (48.1%)). The only difference concerned manic relapses (nine in the placebo group and six in the aripiprazole group). Again the YMRS score significantly differed between groups. The adverse events had a similar rate to the 26-week period.
In both the above reports the median survival time for the aripiprazole group was not evaluable, while the median survival time for placebo was 118 to 203 days depending on the clinical subpopulation. At week 100, the changes (mean ± SD) in PANSS total scores were 7.9 ± 10.61 for aripiprazole and 11.8 ± 8.31 for placebo (P = 0.10), for the PANSS cognitive subscale score were for aripiprazole, 1.5 ± 3.12; placebo, 3.3 ± 2.59 (P = 0.01) and for the PANSS hostility subscale score for aripiprazole, 1.2 ± 2.49; placebo, 2.3 ± 2.07 (P = 0.03). Again all except the total PANSS score showed a significant superiority of aripiprazole over placebo, with d = 0.42, 0.63 and 0.48, respectively [38].
The expansion of the 135-008 trial [34] to a further 14 weeks failed to provide any results because of a high dropout rate, while the extension of the CN138-135/ NCT00095511 trial [27] for an additional 40 weeks (52 weeks in total) comparing aripiprazole to lithium without a placebo arm suggested aripiprazole equal to lithium in the maintenance against manic episodes. No PANSS data are reported concerning this extension phase.

Discussion
The meta-analysis of the four trials that investigated the efficacy of aripiprazole on psychotic symptoms (assessed by the PANSS) during acute manic/mixed episodes suggests that the effect size versus placebo was equal to 0.14, but a more reliable and accurate estimation is 0.18 for the total PANSS score. The effect was higher for the PANSS positive subscale (0.28), PANSS hostility subscale (0.24) and PANSS cognitive subscale (0.20), and lower for the PANSS negative (0.12). The majority of these trials included patients with moderate to severe manic episodes, some of whom also had psychotic symptoms. No data on the depressive phase of bipolar illness exist, while there are some data in favour of aripiprazole concerning the maintenance phase, where at week 26 all except the total PANSS score showed a significant superiority of aripiprazole over placebo (d = 0.28 for positive, d = 0.38 for the cognitive and d = 0.71 for the hostility subscales) and at week 100 the results were similar (d = 0.42, 0.63 and 0.48, respectively). It is important to note that the efficacy for both haloperidol and lithium is similar to aripiprazole in the psychotic symptoms of BD. This could seem odd concerning lithium, which is not considered to possess antipsychotic efficacy (although it is recommended as augmentation strategy in refractory patients with schizophrenia). However the literature suggests that in essence lithium might exert a state-dependent effect on second messenger systems that is antidopaminergic-like during manic episodes (when dopaminergic activity seems to be elevated). Thus this state-dependent antidopaminergic activity could be responsible for this antipsychotic action [39,40].  Dash indicates missing data. AE = adverse events; CW = consent withdrawal or other; LE = lack of efficacy; PANSS = Positive and Negative Syndrome Scale.   In comparison to the baseline scores reported in schizophrenia RCTs with aripiprazole, the respective PANSS scores in bipolar RCTs are significantly lower and this is of course expected. The PANSS-positive scores in schizophrenia RCTs range are around 29, for PANSS-negative they are around 22 and for PANSS-hostility around 9.5, while difference from placebo is again larger with 2.63 points difference in PANSS-positive, 2.31 points for PANSS-negative and 1.96 for PANSS-hostility [41]. However in these studies no standard deviations are reported and it is not possible to derive, thus the real effect sizes can not be calculated [42]. The only comparison that can be made is in terms of the ratio change to baseline. This ratio is similar for PANSS-positive (around 10%), but much different concerning PANSS-negative (10% in schizophrenia vs 3% in BD) and PANSS-hostility (20% for schizophrenia vs 11% for BD).
These results should be interpreted in light of recent studies on common genetic findings in schizophrenia and mood disorders [43][44][45][46][47].
There are several meta-analyses in the literature concerning the efficacy of various agents in the treatment of bipolar illness, but none analyse the effect on psychotic symptoms [21][22][23][24][25][26][27][28]. These meta-analyses suggest that aripiprazole's antimanic effect is specific and not limited to control of agitation through sedation, but no data on psychotic symptoms are analysed. Additionally, there is a concern regarding aripiprazole and olanzapine maintenance data because the relevant studies included patients who were responders specifically to the drug under investigation during the acute phase.
A few issues concerning the data of the RCTs should be pointed out however, because they reveal the restrictions of RCTs, as well as the gaps in our current knowledge and understanding and treating of bipolar illness. The first point is that although acute mania is generally considered one of the 'easier to treat' psychiatric conditions, with only 5% of bipolar patients experiencing chronic mania (although such a diagnostic condition is not recognised by the Diagnostic and Statistical Manual, Fourth Edition, Text Revision (DSM-IV-TR)) [48], in the acute mania RCTs around half of acutely manic patients were non-respond-  Negative results suggest the effect in question was more frequent in the placebo group.
ers at week 12, suggesting that they were not only chronic but maybe also refractory. The question whether the dosage should be raised is open and pressing. Higher dosages for all agents might be necessary to be studied. During the maintenance phase, around three-quarters of patients will drop out of aripiprazole treatment within the first year in comparison to almost all the patients under placebo, and although the difference is significant, it suggests that even under effective treatment, a significant number patients tend to relapse although at a lower frequency of episodes. Finally, the fixed dosage RCT was negative, suggesting that aripiprazole should be prescribed at an individualised basis. Similar issues have been recently raised because of the still unpublished results of the one study of paliperidone in acute mania, which also used a fixed dose approach (the second one utilised a flexible dosage design) and reported that 6 and 9 mg were not effective versus placebo while 12 mg was.
Conclusively, the data analysed for the current study supports the usefulness of aripiprazole against the psychotic symptoms during the acute manic/mixed and maintenance phases of bipolar illness, however there are specific issues the clinician should have in mind, such as that the maintenance effect is proven only in patients with an index manic episode that responded to aripiprazole during the acute phase. Higher and ever-increasing placebo rates constitute a problem of quality for RCTs today and limit the generalisability of results [49]. A limitation of this review is that most of the trials were sponsored by the pharmaceutical industry and were conducted to gain regulatory approval for aripiprazole for the treatment of bipolar disorder. Therefore, the possibility of sponsor bias induced in favour of their product cannot be excluded, especially since failed trials were not published and the available data from them are limited.

Competing interests
KNF is member of the International Consultation Board of Wyeth for desvenlafaxine and has received grants or honoraria for lectures from AstraZeneca, Servier, Janssen-Cilag, Eli Lilly and research grants from AstraZeneca, Janssen-Cilag, Elpen and Pfizer Foundation. EV has acted as consultant, received grants, or received honoraria for lectures by the following companies: Almirall, AstraZeneca, Bial, Bristol Myers Squibb, Eli Lilly, Forrest Research Institute, GlaxoSmithKline, Janssen-Cilag, Jazz Lundbeck, Merck Sharpe Dohme, Novartis, Organon, Pfizer, Sanofi, Servier, UBC. FS has no conflicts of interest. XG has received support for travelling and lecturing by Glaxo-SmithKline, Sanofi, Eli Lilly Organon, Servier and Richter