- •Adjunct treatment with combined antipsychotics and mood stabilizers is more efficacious than mood stabilizers alone.
- •Combinations of some drugs are less well tolerated than mood-stabilizer monotherapy.
Introduction: The aim of this meta-analysis was to systematically review the effectiveness of co-therapy compared with monotherapy for people with bipolar mania.
Method: MEDLINE, Embase, Psychinfo, The Cochrane Library and reference lists of retrieved studies were searched without language restrictions for randomized controlled trials evaluating co-therapy compared with monotherapy for acute bipolar mania. Each trial was assessed for susceptibility to bias. Data on mania outcomes, withdrawals, extrapyramidal symptoms and weight were extracted and pooled effect estimates summarized as relative risks (RR) or differences in mean values (MD) where appropriate.
Results: Eight eligible studies were included (1124 participants). Significant reductions in mania (Young Mania Rating Scale, YMRS) scores were shown for haloperidol, olanzapine, risperidone and quetiapine as co-therapy compared with monotherapy with a mood stabilizer. For atypical antipsychotics combined, the pooled difference in mean scores was 4.41 (95% CI: 2.74, 6.07). Significantly more participants on co-therapy met the response criterion (at least 50% reduction in YMRS score), RR 1.53 (1.31, 1.80). With some drugs, co-therapy decreased tolerability compared with monotherapy, and resulted in greater weight gain. There were insufficient data to compare one co-therapy regimen with another.
Conclusion: The addition of antipsychotic treatment to established mood-stabilizer treatment is more effective than mood-stabilizer treatment alone.
Pharmacotherapy remains the treatment of choice for acute bipolar mania. Treatment may involve the use of mood stabilizers, antipsychotics, benzodiazepines and other sedatives, either as monotherapy or in polypharmacy regimens (1). The British Association of Psychopharmacology guidelines advocate initial monotherapy for the treatment of mania followed by a combination of a mood stabilizer and an antipsychotic (2). In contrast, the recently revised American Psychiatric Association (APA) Guidelines recommend lithium or valproate in combination with an antipsychotic as first-line treatment, with monotherapy recommended only for the less ill patient (3).
Meta-analyses have shown that there is little to choose between drug treatments when used as single agents (with additional ‘when necessary’ benzodiazepines) in mania (4–6). The majority of studies conducted have compared single-drug treatments with placebo or with other single agents. Relatively few reviews have examined the efficacy of anti-manic co-therapy: the concurrent use of antipsychotics and mood stabilizers (3, 7–11).
Further studies have been published since the publication of meta-analyses of studies comparing co-therapy with monotherapy for treatment of acute mania (9, 10).
To perform a systematic review and meta-analysis of currently licenced treatments for bipolar disorder evaluated in co-therapy regimens for treatment of acute mania.
Studies were eligible for inclusion if they met the pre-defined criteria described below:
We searched MEDLINE (1966 to March 2006), Embase (1980 to March 2006), PsychInfo (1872 to March 2006) and the Cochrane Central Register of Controlled Trials (Cochrane Library, Issue 1 2006). We used a sensitive search strategy for randomized trials recommended by The Cochrane Collaboration (12) combined with MESH and thesaurus terms for bipolar disorder, mood-stabilizer and antipsychotic drugs. We supplemented this by searching reference lists of identified trials and reviews, and contacting manufacturers of drugs used for bipolar disorder for additional trials and data. There was no restriction to language of publication.
Two reviewers (LAS and VC) decided, independently, whether individual studies met the inclusion criteria. Disagreements were resolved by discussion with a third reviewer. We used a standardized form, and extracted data from included studies, which included patient and study characteristics, outcome measures and study results. We preferentially used data from intention to treat analyses, and outcome data at the longest available follow-up. We present results for the following outcomes: changes in mania symptom scores using Young Mania Rating Scale (YMRS), mania response defined as at least 50% improvement at endpoint in baseline YMRS score, withdrawal because of any reason, lack of efficacy and an adverse event, extrapyramidal symptoms and weight change. We did not consider results for subgroup analyses within studies unless it was clear they were pre-specified and randomization was stratified for these subgroups.
To assess the potential for bias we evaluated the methodological quality of each trial according to predefined criteria (13). Quality components assessed were method of randomization and concealment of allocation, blinding of trial participants and investigators, completeness of follow-up and handling of withdrawals and dropouts.
Meta-analyses were undertaken to estimate overall treatment effects where the trials were considered to be similar enough to combine. This decision was based on assessing similarity of trial characteristics as well as results. Separate meta-analyses were undertaken for each drug and each outcome. Treatment effects were expressed as relative risks (RR) for binary outcomes, and differences in mean values for continuous outcomes, all with 95% confidence intervals (95% CI). Binary data were pooled using the Mantel–Haenszel method, and continuous outcomes using the inverse variance method (14). If significant statistical heterogeneity was detected (P < 0.1) random effects estimates were calculated using DerSimonian and Laird methods (14). The I2 was also used to report the extent of heterogeneity detected (15). Greater than 50% indicates a moderate amount of heterogeneity, and greater than 75% suggests substantial heterogeneity that should be explored. We planned to investigate heterogeneity by means of preplanned sensitivity analyses; these were aspects of study quality, type of co-therapy medication, outcome definition and length of follow-up. Analyses were undertaken using the stata V8.0 metan macro. When studies did not report results fully enough to permit meta-analysis, we report the results in the text.
The combined search strategies identified 410 studies, of these 11 appeared to meet the review inclusion criteria described above. After further examination three were excluded. One as it was a pooled analysis of two RCTs (one published and one unpublished) and we were unable to extract data for the individual studies from the pooled analysis (16–18). The two studies had been pooled as if they were a single RCT, rather than using a statistical combination method that takes into consideration that the data are from two separate studies. Another was a small trial comparing lithium with a combination of lithium and carbamazepine in 10 patients; outcomes were inadequately reported and were not consistent with the outcomes that we report in this review (19). One compared six different treatment arms and combined three arms each into a high- and low-dose group to make statistical comparisons, we were unable to extract data for individual treatment groups (20).
We included eight RCTs with 1124 participants (21–28). Table 1 shows details of the included trials, including the diagnostic criteria used, participant characteristics, dosing regimens, prior treatment, outcome measures, follow-up periods and quality components. Seven compared an antipsychotic drug given as co-therapy with a mood stabilizer with mood-stabilizer monotherapy. There were additional comparisons in two studies, one compared co-therapy with haloperidol alone (22), and the other compared haloperidol co-therapy with a mood-stabilizer monotherapy (23). Two compared different co-therapies (23, 25). Antipsychotic drugs investigated were haloperidol in three (22, 23, 25), olanzapine in one (26), quetiapine in two (21, 24) risperidone in two (23, 28) and ziprasidone in one (27). All included participants with bipolar disorder according to DSM-IV or DSM-III criteria with the exception of one that used criteria of Feighner (22). All participants were currently manic or with mixed mania; in four studies participants were receiving a mood stabilizer at the time of randomization (23, 24, 26, 28). In the study of Sachs et al. (23), 63% of subjects were taking a mood stabilizer, no further details were reported but all randomized participants had YMRS scores of at least 20 at baseline. Similarly, in the second study of Sachs et al. (24), all participants had been receiving either lithium or valproate semisodium for at least seven of the 28 days before randomization and had YMRS scores at least 20. Tohen et al. (26) randomized participants who had an inadequate response (YMRS score at least 16) to at least 2 weeks of lithium or valproate semisodium at confirmed therapeutic blood levels. Yatham et al. (28) randomized participants with YMRS scores at least 20, 43% of which were previously taking a mood stabilizer for at least 2 weeks. In the study by Small et al. (25), participants were described as not responding to treatment but were not necessarily treatment resistant.
|Study||Participants||Drug, dose, number randomized and completed||Follow-up||Outcome measures||Study quality|
|DelBello et al. (21)||DSM-IV criteria for bipolar I disorder currently manic or mixed episode; YMRS score ≥ 20, hospitalized|
Age: mean 14.3 years, range 12–18 years
Prior treatment: washout for VSS, lithium and carbamazepine
|Quetiapine 25 mg titrated to maximum of 150 mg t.i.d. plus valproate semisodium serum level 80–130 mg/dl; N = 15 (53% completed)||Valproate semisodium serum level 80–130 mg/dl; N = 15 (93% completed)||6 weeks||YMRS score, YMRS response, withdrawal SAS, BAS, and AIMS scores, weight change||Concealment unclear|
Analysis by intention-to-treat
|Garfinkel et al. (22)||Feighner criteria for bipolar mania by unanimous agreement of 3 investigating psychiatrists, hospitalized. Age: mean 38.5 years|
Prior treatment: patients medication free at study start where possible
|Haloperidol 30 mg plus lithium 900 mg daily, mean serum level 0.80–0.85 mEq/l; N = 7 (86% completed)||Lithium started at 900 mg daily, mean serum level 1.00–1.20 mEq/l; N = 7 (43% completed)||Haloperidol 30 mg daily; N = 7 (71% completed)||22 days||Withdrawal, EPS||Concealment unclear|
Study completers analysed
|Sachs et al. (23)||DSM-IV bipolar disorder; current manic episode; YMRS score ≥ 20, hospitalized|
Age: range 18–66 years
Prior treatment: 63% already receiving a mood stabilizer
|Risperidone 2–6 mg/day plus mood stabilizer (valproate semisodium 50–120 ug/ml or lithium 0.6–1.4 mEq/l; N = 52 (65% completed)||Haloperidol 4–12 mg/day, plus mood stabilizer (valproate semisodium 50–120 ug/ml or lithium 0.6–1.4 mEq/l; N = 53 (47% completed)||Mood stabilizer alone (valproate semisodium 50–120 ug/ml or lithium 0.6–1.4 mEq/l; N = 51 (51% completed)||22 days||YMRS score, withdrawal, EPRS score, weight change||Concealment unclear|
Intention-to-treat analysis with LOCF
|Sachs et al. (24)||DSM-IV bipolar I disorder, hospitalized for mania; YMRS score ≥ 20, CGI-S score ≥ 4|
Age: mean 40.5 years.
Prior treatment: mood stabilizer for at least 7 of the 28 days immediately before randomization
|Quetiapine 200–800 mg/day with lithium 0.7–1.0 mEq/l or valproate semisodium 50–100 μg/ml; N = 91 (62% completed)||Mood stabilizer alone (lithium 0.7–1.0 mEq/l or valproate semisodium 50–100 μg/ml); N = 100 (49% completed)||21 days||YMRS score, YMRS response, withdrawal, SAS, weight change||Concealment unclear|
Modified intention-to-treat analysis with LOCF
|Small et al. (25)||DSM III-R Axis 1 bipolar disorder current manic or mixed episode, at least one previous affective episode within the last 2.5 years, SDMS-M ≥ 7, GAS ≤ 60, hospitalized|
Age: mean 36.6 years; range 19–62 years
Prior treatment: all psychotropics discontinued for 2 weeks
|Lithium 300–1400 mg/day serum level 0.6–1.0 mmol/l, +Carbamazepine 200–900 mg/day serum level 4–12 μg/ml; N = 17 (6% completed)||Lithium 300 mg/day to 1400 mg/day + haloperidol 2–13.5 mg/day serum level 5 ng/ml + 2 mg to a max of 6 mg daily benztropine administered with each dose of haloperidol; N = 16 (6% completed)||8 weeks, complete data at 3 weeks||YMRS scores, SAS scores, withdrawal||Concealment unclear|
Complete data available for all participants at 3 weeks, LOCF for 8 weeks, dns numbers analysed.
|Tohen et al. (26)||DSM-IV bipolar I disorder manic or mixed episode with or without psychotic features, inadequate response to monotherapy, YMRS score ≥ 16|
Age: mean 40.5 years
Prior treatment: mood stabilizer at least 2 weeks before randomization
|Olanzapine 5–20 mg/day (starting dose 10 mg day) added to valproate semisodium 63.6 μg/ml (n = 145) or lithium 0.76 mEq/l (n = 74); N = 229 (70% completed)||Mood stabilizer alone – lithium 0.82 (0.19) mEq/l (n = 41), valproate semisodium 74.7 (18.6) μg/ml (n = 73); N = 115 (71% completed)||6 weeks||YMRS score, YMRS response, withdrawal||Concealment unclear|
Intention-to-treat analysis with LOCF
|Weisler et al. (27)||DSM-IV bipolar disorder, manic or mixed episode, MRS ≥ 14, ≥ 2 on at last 4 items|
Age: at least 18 years
Prior treatment: some patients on lithium before randomization
|Ziprasidone 80–160 mg/day plus lithium titrated to maintain serum level 0.8–1.2 mEq/l; N = 102 (% completed not reported)||Lithium titrated to maintain serum level 0.8–1.2 mEq/l; N = 103 (% completed not reported)||3 weeks||MRS score||Concealment unclear|
Intention-to-treat analysis with LOCF
|Yatham et al. (28)||DSM-IV bipolar disorder manic or mixed episode, YMRS score ≥ 20, randomized within 7 days of hospital admission|
Age: range 19–65 years
Prior treatment: 43% had been taking a mood stabilizer before randomization
|Risperidone 1–6 mg daily, plus mood stabilizer (lithium, valproate semisodium or carbamazepine); N = 75 (64% completed)||Mood stabilizer alone (lithium, valproate semisodium or carbamazepine); N = 76 (48% completed)||22 days||YMRS score, YMRS response, withdrawal, ESRS, weight gain||Concealed|
Efficacy analyses included patients with at least 2 assessments
All eight studies were described as randomized, one used minimization to balance the groups (28). Concealment of allocation was adequately reported in only one of the studies (28). All were double blind. A high proportion of participants withdrew from the studies (range 7–94%), although most studies used intention to treat analyses with last observation carried forward to account for missing data. In one study, a whole centre (n = 20) was excluded from analyses because of a centre-specific protocol violation (24). One analysed study completers only (22).
Haloperidol, olanzapine, risperidone and quetiapine co-therapy provided significant improvements in YMRS scores at follow-up compared with monotherapy with a mood stabilizer. YMRS scores were 5.20 (1.14–9.26) points lower with haloperidol, 4.01 (1.96–6.06) points lower with olanzapine and 5.16 (2.32–8.00) points lower with risperidone (pooled data not shown) (Fig. 1). For all atypical antipsychotics combined, YMRS scores were 4.41 (2.74–6.07) points lower with co-therapy compared with monotherapy. We were unable to a calculate a difference in mean values for quetiapine as adequate data were not presented (21, 24). Mean change scores were greater for quetiapine co-therapy, −13.76 compared with monotherapy, −9.93 (P = 0.021) (24), and DelBello et al. (21) reported a significantly greater reduction in YMRS scores at endpoint with adjunct therapy compared with monotherapy (P = 0.002). Few data were reported for ziprasidone in a conference abstract; at day 14 there was no significant difference in change in MRS scores between treatment groups (27). Three studies stratified randomization by mood stabilizers (23, 26, 28) but results were not presented separately for each mood stabilizer subgroup. In the study by Yatham et al. (28) exclusion of the subgroup of participants that received carbamazepine from the primary analysis suggested that the effect of risperidone may even be greater than shown for the entire randomized group. However, this was not a predefined subgroup analysis but justified post hoc because of lower (40%) risperidone plasma concentrations in the carbamazepine subgroup. In the study by Tohen et al. (26) the authors reported that YMRS changes because of co-therapy with valproate were similar in magnitude to changes because of co-therapy with lithium.
Significantly more participants met response criterion, at least 50% improvement from baseline in YMRS scores, with co-therapy compared with monotherapy, 51% more (21–89%) with olanzapine, 66% more (23–124%) with quetiapine and 43% more (2–101%) with risperidone. For all atypical antipsychotics combined 53% more (31–80%) achieved a response with adjunct therapy compared with monotherapy (Fig. 2).
For all atypical antipsychotics combined, the risk of withdrawal was 17% fewer (38% fewer to 9% more) compared with monotherapy (Fig. 3). Significantly fewer 30% (7–48%) participants withdrew for any reason with risperidone adjunct therapy compared with monotherapy (pooled data not shown). For haloperidol, olanzapine and quetiapine there was little difference in the number of withdrawals between adjunct and monotherapy, 30% fewer (81% fewer to 166% more), 3% more (27% fewer to 47% more) and 12% fewer (36% fewer to 20% more) (pooled data not shown) respectively.
For all atypical antipsychotics combined, the risk of withdrawal because of lack of efficacy was significantly fewer, 54% (23–73%) compared with monotherapy (Fig. 4). The incidence of withdrawals because of lack of efficacy was significantly lower 75% (40–90%) with olanzapine adjunct compared with monotherapy. Although there were fewer withdrawals because of lack of efficacy with the other adjunct therapies, for each comparison the upper confidence limit does not exclude a possible increased risk of withdrawal. There were 42% fewer (85% fewer to 129% more) with haloperidol, 42% fewer (74% fewer to 32% more) with quetiapine (pooled data not shown), and 18% fewer (74% fewer to 160% more) with risperidone (pooled data not shown; Fig. 4).
The incidence of withdrawal because of an adverse event was sixfold higher with olanzapine adjunct compared with monotherapy, although the risk estimate is consistent with as little as a 50% higher incidence. For other comparisons while there were few withdrawals because of an adverse event in each co-therapy group, the wide confidence intervals reflect the degree of uncertainty in the estimates and may be because of chance. Withdrawals were 52% fewer (95% fewer to 414% more) with haloperidol, 8% fewer (71% fewer to 190% more) with quetiapine and 41% fewer (86% fewer to 144% more) with risperidone (Fig. 5). All antipsychotics were not pooled in this meta-analysis because of the heterogeneity in effects between olanzapine and the other studies. For ziprasidone, withdrawals because of adverse events were reported to be about 5% in both groups.
Compared with monotherapy, weight gain was, 2.9 kg greater (2.29–3.51) with olanzapine co-therapy (26), and 1.7 kg (0.24 less to 3.64 more) with quetiapine co-therapy (21) (Fig. 6). Weight gain was also greater with risperidone co-therapy compared with monotherapy: 5.3 lb (SD = 7.0) and 1.1 lb (SD = 4.8) (P < 0.004) (23); and 1.7 and 0.5 kg (P < 0.012) (28) respectively. There was little difference between haloperidol co-therapy 0.3 lb (SD = 5.4) compared with monotherapy 1.1 lb (SD = 4.8). For risperidone and haloperidol we were unable to calculate differences between groups as insufficient data were reported.
For quetiapine adjunct compared with monotherapy, there was little difference in change in Simpson Angus Scale scores at endpoint, 0.1 (−0.59 to 0.79) and Barnes Akathisia Scale scores, −0.20 (−0.41 to 0.01) (21). Yatham et al. (28) reported a mean change in extrapyramidal symptom rating scale score of −0.1 for both groups, but no further details were given. Garfinkel et al. (22) reported significantly higher extrapyramidal symptom scores at endpoint (P < 0.05), and the incidence of extrapyramidal symptoms was five times greater (0.77–32.57) with haloperidol adjunct therapy compared with lithium. Extrapyramidal symptoms occurred more frequently with ziprasidone co-therapy (22%) compared with lithium monotherapy (2%).
One trial compared haloperidol adjunct to a mood stabilizer with risperidone adjunct to a mood stabilizer (23). There was little difference in YMRS scores, 0.90 points lower (4.74 lower to 2.94 higher), there were fewer withdrawals overall, 34% (58% fewer to 3% more), but the upper confidence limit does not exclude the possibility of more withdrawals, and little difference in withdrawals because of lack of efficacy, or because of adverse events (Table 2).
|Study||Outcome||Events/participants, n/N||Events/participants, n/N||RR (95% CI)|
|Risperidone adjunct therapy||Haloperidol adjunct therapy|
|Sachs et al. (23)||Withdrawal any reason||18/52||28/53||0.66 (0.42, 1.03)|
|Withdrawal because of lack of efficacy||3/52||3/53||1.02 (0.22, 4.82)|
|Withdrawal because of adverse event||2/52||1/53||2.04 (0.19, 21.80)|
|Carbamazepine adjunct therapy||Haloperidol adjunct therapy|
|Small et al. (25)||Withdrawal any reason||16/17||15/16||1.00 (0.84, 1.19)|
|Withdrawal because of lack of efficacy||7/17||4/16||1.65 (0.59, 4.57)|
|Withdrawal because of adverse event||0/17||7/16||0.06 (0.00, 1.02)|
One trial compared carbamazepine adjunct to lithium with haloperidol adjunct to lithium (25). There was no significant difference between groups in YMRS scores; at 3 weeks mean scores were 16.8 and 15.5, and at 8 weeks 14.6 and 13.5. At 8 weeks, there was no difference in the incidence of withdrawal for any reason, RR 1.00 (0.84, 1.19), there were more withdrawals because of lack of efficacy, 65% (41% fewer to 357% more), although cannot exclude fewer withdrawals, and fewer because of an adverse event, 94% (99.6% fewer to 2% more) with adjunct carbamazepine compared with adjunct haloperidol. Adverse events causing withdrawal with haloperidol were primarily extrapyramidal, but the incidence was not reported. Simpson Angus Scale scores at 3 weeks had increased to 1.8 and 3.1 (P = 0.051) with carbamazepine adjunct and haloperidol adjunct respectively.
One study compared haloperidol adjunct to lithium with haloperidol alone (22). Effects on mania were not evaluated using YMRS, but overall withdrawals and extrapyramidal symptoms were reported. An equal number in each group, five of seven patients, reported extrapyramidal side-effects; one of seven and two of seven withdrew with haloperidol adjunct compared with haloperidol alone.
Co-therapy with an antipsychotic and a mood stabilizer is more efficacious than a mood stabilizer alone as assessed using a continuous and a categorical outcome on the YMRS. Response rate was more than 50% higher in those receiving co-therapy. Only risperidone co-therapy showed advantages over monotherapy in terms of trial withdrawals for any reason, and only olanzapine showed advantages in terms of withdrawals because of lack of efficacy. Co-therapy with olanzapine appeared to worsen tolerability compared with a mood stabilizer used alone and the combination engendered relatively more weight gain. Data did not allow separation of the effects of different co-therapy regimens.
The main strengths of this analysis are the rigour of the literature search and paper selection, the assessments of trial quality and the facility to combine results from several studies using common outcome measures. This is the first meta-analysis of co-therapy with all currently licenced antipsychotics in mania and clearly shows the superiority of two-drug regimens over monotherapy. There are two major limitations. The first is the quality of the studies included. Few reports described method of randomization or concealment of allocation. Failure to report these methodological details is associated with a greater likelihood of producing ‘positive’ results (13). The second is the small number of studies included, again partly a result of methodological shortcomings. Nonetheless, the number of subjects included in these meta-analyses is more than four times the number in the largest trial so far conducted. Also, the uniformity of outcome in the separate trials does suggest a lack of influence of external factors.
The implications of our findings need to be considered carefully. Most importantly, the design of the studies included needs to be emphasized. In the main, these were not studies of untreated patients randomized to either co-therapy or monotherapy but of patients already treated with mood stabilizers (sometimes without clear or adequate dose optimization and largely having failed on them) randomized to the addition of an antipsychotic or placebo. Our findings therefore confirm the advantage of adding an antipsychotic to inadequate mood-stabilizer treatment but do not suggest that all patients with mania should receive co-therapy. These data are clearly sufficient to establish the role of co-therapy in patients presenting with mania while being prescribed lithium or valproate. Our results do not, however, inform practice for those presenting with mania when they are receiving no drug treatment (first-episode mania, for example). Very few studies have examined co-therapy in subjects who have not already failed on mood stabilizers and none met inclusion criteria for this analysis. Interestingly, one methodologically sound study comparing sodium valproate with placebo showed advantages for the addition of an antipsychotic to sodium valproate (29). Nonetheless, perhaps the single appropriate conclusion to draw from this meta-analysis is that in the treatment of mania, methodologically flawed studies uniformly indicate that co-therapy with an antipsychotic and mood stabilizer is more effective than a mood stabilizer alone in subjects showing inadequate response to un-optimized mood stabilizer monotherapy. There is a clear need for adequately powered studies of co-therapy compared with monotherapy in subjects not already receiving mood stabilizer treatment.
Funding was provided to Medical Research Matters by Sanofi-Aventis.
Leslie Smith and Victoria Cornelius have no declarations of interest. At the time of manuscript preparation, Adrian Warnock was an employee of Sanofi-Aventis; Mary-Jane Tacchi and David Taylor were recipients of research funding and honoraria from Sanofi-Aventis, Bristol-Myers Squibb, Janssen-Cilag, Wyeth and AstraZeneca.