Antidepressants for Bipolar Depression: A Systematic Review of Randomized, Controlled Trials

We included randomized, controlled double-blind trials, published in any language, that compared antidepressants with placebo or alternative drug treatments. Antidepressants included all tricyclic antidepressants, selective serotonin reuptake inhibitors (SSRIs), noradrenergic reuptake inhibitors, reversible monoamine oxidase inhibitors (MAOIs), bupropion, St. John’s wort, and tryptophan but not mood stabilizers, anticonvulsants, lamotrigine, sulpiride, benzodiazepines, or ECT. Alternative comparator treatments included mood stabilizers, anticonvulsants, and other antidepressants. The inclusion criteria for the review were trials with patients with a current depressive or mixed depressive/manic episode with or without psychotic symptoms who had had at least one previous episode of mania or hypomania, including antidepressant-induced mania or hypomania. Trials that recruited nonbipolar patients as well as bipolar patients were included if they randomly assigned the bipolar patients separately or if the majority of the patients were bipolar. The primary outcome measures for the review were clinical response and remission rates (derived from observer-rated symptom reductions), induction of mania or hypomania, and the overall dropout rate as a proxy measure of the acceptability of treatment.

We searched the Cochrane Collaboration Depression, Anxiety, and Neurosis Controlled Trials Register (December 2002), incorporating results of group searches of MEDLINE (1966 to the present), EMBASE (1980 to the present), CINAHL (1982 to the present), PsycLIT (1974 to the present), PSYNDEX (1977 to the present), and LILACS (1982–1999). These were searched by using the following terms for diagnosis: bipolar III disorder, unipolar mania, rapid-cycling disorder, affective disorders, affective psychosis, bipolar, bipolar disorder, bipolar I disorder, bipolar II disorder, cyclothymic disorder, depression, depressive psychosis, excited psychosis, hypomania, mania, manic depressive, manic disorder, manic episode, melancholia, mixed depression, mood disorders, bipolar affective disorder, bipolar not otherwise specified, dysphoric mania, manic episode, manic symptoms, schizoaffective disorder, psychoses, psychotic disorders, puerperal psychosis, and reactive depressive psychosis. The terms searched for intervention and antidepressive agents were monoamine oxidase inhibitors, selective serotonin reuptake inhibitors, and tricyclic drugs. The following drug names were also searched: acetylcarnitine, alaproclate, amesergide, amiflamine, amineptine, amitriptyline, amoxapine, befloxatone, benactyzine, brofaromine, bupropion, butriptyline, caroxazone, chlorpoxiten, cilosamine, cimoxatone, citalopram, clomipramine, clorgyline, clorimipramine, clovoxamine, deanol, demexiptiline, deprenyl, desipramine, dibenzepin, diclofensine, dothiepin, doxepin, duloxetine, etoperidone, femoxetine, fluotracen, fluoxetine, fluparoxan, fluvoxamine, idazoxan, imipramine, iprindole, iproniazid, isocarboxazid, litoxetine, lofepramine, maprotiline, medifoxamine, melitracen, metapramine, mianserin, milnacipran, minaprine, mirtazapine, moclobemide, nefazodone, nialamide, nomifensine, nortriptyline, noxiptiline, opipramol, oxaflozane, oxaprotiline, pargyline, paroxetine, phenelzine, piribedil, pirlindole, pivagabine, prosulpride, protriptyline, quinupramine, reboxetine, rolipram, sertraline, setiptiline, teniloxine, tetrindole, thiazesim, thozalinone, tianeptine, toloxatone, tomoxetine, tranylcypromine, trazodone, trimipramine, venlafaxine, viloxazine, viqualine, and zimeldine.

The Cochrane Library was searched by using the same terms as the Cochrane Collaboration Depression, Anxiety, and Neurosis Controlled Trials Register excluding references that came from that database. The reference lists of selected studies were inspected for more published reports and citations of unpublished research. In addition, other relevant papers and major textbooks that cover affective disorder were checked.

Two of us (H.J.G., J.M.R.) independently checked studies generated by the search strategy to ensure that they met the inclusion criteria. Any disagreement was resolved by consensus discussion with another coauthor (J.R.G.). One reviewer (H.J.G.) assessed the methodological quality of the included studies according to the reporting of the randomization procedure, especially allocation concealment (15), blinding, and the reporting of withdrawals. Information about participant characteristics, intervention details, and outcome measures was extracted independently by two reviewers.

Data were entered twice into Revman 4.2, a program developed by the Cochrane Collaboration for systematic reviews. For binary efficacy outcomes, a pooled relative risk (with 95% confidence intervals [CIs]) was calculated by using both fixed- and random-effects models to investigate the sensitivity of results to the choice of statistical method. Heterogeneity between studies was assessed by using the Q statistic (16). When significant heterogeneity was identified, sources were investigated by visual inspection of the forest plots for outlying trials. Two sensitivity analyses were conducted, one excluding two trials that included less than 100% bipolar patients and another including a trial with bipolar patients who were not randomized separately from unipolar patients. Treatment effects are also presented as the number needed to treat, which is the reciprocal of the absolute difference between the response rates of the compared treatments. The number needed to treat can assist clinical interpretation of the results of trials and expresses the number of patients that must be treated with the drug of interest to achieve one positive outcome more than would be achieved with a comparator or placebo (17).

Twelve trials were included in the primary analysis: five comparing one or more antidepressants with placebo (total participants, N=779), four studies comparing two different antidepressants only (N=236), and three trials comparing an antidepressant with another type of drug (N=73).

We had originally identified 23 trials in which bipolar depressed patients received an antidepressant, but we had to exclude 11 trials. Eight of these studies had included only a low proportion of bipolar patients, who were not randomized separately (18–25). Two excluded trials compared other treatments that were added to antidepressant treatment (26, 27). One trial compared two antidepressants but was not blinded (28).

The characteristics of the included trials are presented in Table 1. All trials were described as randomized and used a parallel group design. Study duration varied between 4 and 10 weeks. All trials excluded patients with a diagnosis of serious physical illness or substance abuse. Participants were adult patients up to 70 years old. In most studies, the majority (approximately 60%–70%) of participants were women.

Dichotomous outcomes are presented here. With regard to continuous outcomes, nine studies reported Hamilton Depression Rating Scale or Montgomery-Åsberg Depression Rating Scale scores, but only some reported baseline scores of all randomly assigned patients. Thus, meta-analysis of continuous data was not feasible.

Five trials were available for this comparison. About 75% of the patients were taking a concurrent mood stabilizer or an atypical antipsychotic. One recent trial compared three groups: olanzapine plus fluoxetine (N=86), olanzapine alone (N=370), and placebo (N=377) (29). We only included the former two groups in this analysis as, respectively, the fluoxetine and placebo groups.

One of these five trials only reported remission and was therefore excluded from this comparison (30). One other study did report responders but not how this was defined and was included in this analysis (33). As a result, four trials with a total of 662 randomly assigned patients were available for this comparison, 213 in the experimental group and 449 in the placebo group. Patients treated with an antidepressant were more likely to respond by the end of the trial (risk ratio=1.86, 95% CI=1.49–2.30) (Figure 1). The number needed to treat with antidepressants was 4.2 (95% CI=3.2–6.4). The treatment effect remained statistically significant after taking account of the heterogeneity by using a random-effect model (risk ratio=2.29, 95% CI=1.29–4.04). The direction of the effect favored antidepressants in all the trials, which implies that the heterogeneity was quantitative rather than qualitative. We could not distinguish response rates for patients taking and not taking concurrent medication within individual trials.

This was defined in only two studies, as a Hamilton depression scale score ≤7 (39) and a Montgomery-Åsberg Depression Rating Scale score ≤12 (29). However, these are the two largest studies, with a total of 160 patients in the experimental group and 413 patients in the comparison group. All of these patients were taking a concurrent mood stabilizer or an atypical antipsychotic. Patients treated with an antidepressant (paroxetine, imipramine, or fluoxetine) were more likely to reach remission than those who were not taking an antidepressant (risk ratio=1.41, 95% CI=1.11–1.80) (Figure 2). The number needed to treat was 8.4 (95% CI=4.8–33).

Two studies had predefined criteria for switching to mania, as DSM-III-R mania or hypomania (30) or a score ≥15 on the Young Mania Rating Scale (29). One study did not specifically describe how they monitored the emergence of mania (31). One study reported the absence of emergence of manic states (32). One small study did not address or report switches to mania but was included in the analysis on the inference that no mania did occur (33).

There was no evidence of an increased risk of switching to a manic episode in the trials (risk ratio=1.00, 95% CI=0.47–2.13) (Figure 3). However, there were very few manic events; hence, of course, this would have limited the power to detect a difference between antidepressant and placebo. The risk of switching for antidepressants was 3.8%, and for placebo, it was 4.7% (difference=0.9%, 95% CI=–2.0 to 3.8).

The dropout rate was 32% in the antidepressant group, 49% in the placebo group, and 43% in total. There were significantly fewer study withdrawals while subjects were taking antidepressants (fixed-effect risk ratio=0.71, 95% CI=0.58–0.88; χ²=1.82, df=3, p=0.61, for heterogeneity).

We performed a sensitivity analysis excluding the two studies with less than 100% bipolar patients (33, 36). This did not materially alter the estimate of treatment effect (fixed-effect risk ratio=1.61, 95% CI=1.27–2.04) and reduced heterogeneity (χ²=2.71, df=1, p=0.10). The risk of withdrawal also remained unchanged (fixed-effect risk ratio=0.74, 95% CI=0.60–0.92).

This comparison includes 134 patients from the antidepressant arms of two trials in which two antidepressants were compared with placebo (31, 30) and 236 patients from four trials with two antidepressant arms and no placebo group (34–37). In total, 201 of the 370 patients (54%) distributed throughout all of these studies were receiving co-therapy with mood stabilizers.

A dichotomous measure of clinical response was reported in five trials. This was based on scores on the Hamilton depression scale in five trials and on the Clinical Global Impression scale in one trial. One trial did not report such a measure and was therefore excluded from this analysis (30). Tricyclic antidepressants may be less effective than other antidepressants, but this did not reach statistical significance (risk ratio=0.84, 95% CI=0.67–1.06) (Figure 4).

There was no difference in remission rates between imipramine and paroxetine based on one trial only (risk ratio=0.84, 95% CI=0.47–1.48) (30).

Three out of six trials had predefined criteria for switching to mania, as DSM-III-R mania or hypomania (30, 35) or a score higher than 5 on the Raskin Mania Scale (36). One trial used the Young Mania Rating Scale for follow-up but simply used “withdrawal from the study because of mania” as a clinical description (T. Silverstone, personal communication). One study did not specifically describe how the authors monitored the emergence of mania (31). Two studies reported the absence of emergence of manic states (32, 37).

On average, tricyclic antidepressants caused more switching to mania than other antidepressants (risk ratio=2.92, 95% CI=1.28–6.71) (Figure 5). The rate of switching for tricyclic antidepressants was 10%, and for other antidepressants combined, it was 3.2%. This gives an absolute risk difference of 6.8% (95% CI=1.7%–11.9%).

The dropout risk in the tricyclic antidepressant group was 37%; in the other groups, it was 30%; and in total, it was 34%. The tricyclic antidepressants did not cause significantly more withdrawals than other antidepressants (risk ratio=1.23, 95% CI=0.92–1.64).

In the subgroups in Figure 4 and Figure 5, there are possible differences between treatments, although none was statistically significant. First, SSRIs may have greater efficacy than tricyclic antidepressants. Second, MAOIs may induce less switching to mania than tricyclic antidepressants.

We performed a sensitivity analysis including 33 bipolar patients from a unipolar randomized, controlled trial of 381 patients comparing moclobemide and imipramine (25). This trial had been included in several previous reviews on bipolar depression. We excluded it from the primary analysis because, although the results for these patients were presented separately, they were not randomized separately. The sensitivity analysis did not alter the overall pattern of clinical response seen in our primary analysis.

Only three small trials were identified comparing antidepressants with different types of agents, and therefore, no meta-analyses were performed. One study (N=30) compared an antidepressant with a second mood stabilizer but was unclear about which drugs were specifically used by the different groups of patients and did not report dichotomous response data (38). The second study (N=27) compared a low dose of amitriptyline (average 62 mg/day) with a low dose of sulpiride (average 55 mg/day) and reported response rates of 80% or higher in both groups (40). A third study (N=16) compared bupropion with idazoxan, a selective α₂ antagonist, but only two of nine and three of seven patients in the respective groups responded (39).

This review demonstrates the advantages of a systematic search when reviewing the literature. Thus, we included five trials that were not included in previous reviews; two small studies compared antidepressants with other drugs not in wide use (39, 40). Another consisted of nine bipolar patients who were separately randomly assigned within a larger study (37). Finally, we included two placebo-controlled studies with fewer than 100% bipolar patients. One of these was included because the other 50% of patients suffered from anergic depression, which is very similar to typical bipolar depression (32). The other included 65% patients with bipolar depression, which is substantially more than any other trial with mixed inclusion (33).

We did not include several trials that have been cited in recent reviews on bipolar depression. One is a trial with no bipolar patients at all, since the patients are described as having manic-depressive psychosis, depressed type, the ICD-9 description for unipolar depression (21). This study was erroneously cited in several reviews on bipolar depression (41–43). In two other trials comparing bupropion with placebo in unipolar depression (22, 23), the majority of patients had a diagnosis of DSM-II manic-depressive, depressed, which is the DSM-II description for unipolar depression (“manic-depressive, circular” is used for bipolar patients). These two trials have been used to advocate the use of bupropion as a first antidepressant choice in bipolar depression (42, 44). Bupropion also takes a prominent place in recent expert guidelines (45). The actual evidence for the use of bupropion is very limited. Apart from some open studies, there is only one small study, which does not substantiate its advocated lower risk of switching to mania (35).

A quantitative review of the existing randomized data confirms that the numbers of bipolar depressed patients entered into clinical trials is around 1% of the figures for unipolar depression. However, in acceptance of that limitation, the data nevertheless strongly support an average positive efficacy for antidepressants versus placebo in trials up to 10 weeks. There is no coherent basis for doubting that conventional antidepressants have efficacy in bipolar depression on the basis of the existing evidence. The size of the antidepressant effect was comparable to that in unipolar depression (46), suggesting that antidepressants may be of comparable efficacy in unipolar and bipolar depression. Withdrawal rates also tended to be higher from placebo, compared with active treatment arms. This suggests a greater acceptability of active treatment, probably due to lower efficacy for placebo.

It is the advantage of a systematic review that individual negative trials are appropriately weighted. In a qualitative review, a trial that fails to achieve significant results may receive unwarranted attention at the expense of an overall pattern of positive findings (or vice versa).

The rates of switching to mania in the short time spans of 4 to 10 weeks covered by these studies were low and lent no support to the belief that switching is a common early complication of treatment with antidepressants. However, the numbers were small, and larger studies, especially with longer follow-up and systematic monitoring of manic symptoms, could change this conclusion. However, current guidelines for treatment of bipolar depression with antidepressants tend to stress use in the short term and early discontinuation. The present findings are relevant to that pattern of use.

We could not show a difference between an antidepressant as monotherapy compared with when added to a mood stabilizer. To our knowledge, no trial has addressed this directly, and the scope for comparison between trials was limited by different definitions of switching and, probably, by different baseline rates of switching.

Interpretation of the data could have been improved by better definition and reporting of switching to mania, especially when it occurred in patients who had responded to treatment. Future studies also will have to pay more attention to the definition, follow-up, and reporting of manic symptoms.

At present, the data are inadequate to definitively favor one medicine over another within the several generic categories of the antidepressants. Our review, however, tends to support the findings of nonrandomized studies in bipolar patients and comparative studies in unipolar patients that have suggested a higher risk of switching to mania for tricyclic antidepressants (47, 48). Our results suggest that tricyclic antidepressants cause more switching to mania, and tricyclic antidepressants are not more, and may even be less, effective. This suggests that antidepressant response and mania may not be correlated, and indeed, the pharmacology of tricyclic antidepressants may make them less suitable for bipolar patients in general.

We did not include lamotrigine in this review since it is pharmacologically unlike existing antidepressants. However, data from the one published study in acute bipolar depression suggest that the effect size for the efficacy of antidepressants against placebo is comparable to that of lamotrigine (49). Lamotrigine, 200 mg/day, induced mania in 3% of patients (two of 63), which is no different from the risks we report for both nontricyclic antidepressants and placebo. Lamotrigine, 50 mg/day, was ineffective. Therefore, while there is no clear short-term benefit for lamotrigine over antidepressants for bipolar depression, it may be an alternative option for treatment.

Only one small study in patients already taking a mood stabilizer compared the addition of an antidepressant with the addition of a second mood stabilizer. No firm conclusions can be drawn from that study because of its size and the absence of information on clinical response (38). Future large studies will need to address the efficacy of mood stabilizers alone in more detail. These studies should look at the long-term outcome of depression, quality of life, and the focus on duration of remission as well as response.