Antidepressants and Bipolar Disorder: What Do Recent Studies Tell Us?

RECENT STUDIES

In assessing recent studies, we will keep our focus on the concepts of levels of evidence and confounding bias. Regarding antidepressants and BD, 5 studies in the past 2 years stand out, 3 of which are published and 2 of which are quite new and as yet unpublished (Table 2).

Stanley Foundation antidepressant discontinuation study

This study assessed 84 patients who had responded to an antidepressant plus a mood stabilizer for acute bipolar depression⁹; after acute recovery, those who continued to take the combination stayed well longer in the 1-year follow-up than those in whom antidepressant therapy was stopped after the acute episode resolved. In other words, the study seems to show that long-term continuation of antidepressants in BD appears to lead to better outcomes. Yet, the study is observational, not randomized, and thus liable to confounding bias. Its level of evidence is III.

Readers can begin to assess this issue by putting themselves in the place of the treating clinicians. Why would one stop an antidepressant after acute recovery? Some literature suggests that antidepressants can cause or worsen rapid cycling in patients with BD; so if a patient has rapid-cycling illness, some clinicians would be inclined to stop the antidepressant after acute recovery. If a patient had a history of incidents of antidepressant-induced mania that were frequent or severe, some clinicians might not continue the antidepressant. Perhaps if the patient had BD type I, some clinicians would be less likely to continue antidepressants than if the patient had BD type II. These are issues of selection bias or so-called confounding by indication: the physician decides what to do nonrandomly.

There also may be other confounders; for example, there may be more males, a younger age at onset, or a greater severity of illness in 1 group. If there were more patients with rapid cycling in the group that discontinued antidepressants than in the group that continued antidepressant treatment, this would be confounding by indication. As a result, the observed finding that the group that discontinued antidepressant treatment relapsed before the other group would be explained by the natural history of rapid-cycling illness: patients with rapid cycling relapse more rapidly than those who do not experience rapid cycling.

It may not be, in fact, that any of these potential confounders actually influenced the results of the study. However, the authors did not conduct any statistical analyses (eg, regression models) to assess and correct for confounding factors. In response to a letter to the editor,¹⁰ for instance, they indicated that no patients in the study had rapid-cycling BD, which would have been a major potential confounding factor. The absence of such patients indicates that no one with rapid-cycling BD responded to acute treatment with antidepressants, which is itself an interesting observation, since the original cohort did not exclude rapid cyclers. Furthermore, if the data are accepted as valid, despite their observational nature, they only generalize to about 15% of the patients treated with antidepressants for acute bipolar depression, since 85% of the patients either failed to respond sufficiently to enter the study or they experienced manic switches.

This study has since been replicated by Joffe and colleagues¹¹ with 59 patients. Again, in the same observational design, if patients responded to an antidepressant plus mood stabilizer, they then stayed well longer if they continued to take the combination.

Here the concept of confounding bias is relevant; since bias is systematic (not random) error, if a study influenced by confounding bias is designed the same way, the same erroneous result will be found. The concept of replication is only relevant to an unbiased study, such as one completed with randomized data; hence, it is not surprising to find the same systematically biased results from a repetition of the same observational design. The validity of the results would be better established by randomized study designs.

A recent meta-analysis

Gijsman and colleagues¹² published an important meta-analysis of placebocontrolled randomized studies of antidepressant efficacy in acute bipolar depression. Based on the 5 studies that were identified, the authors found that antidepressants were more effective than placebo. Further, in those same 5 studies, there was no evidence of an incidence of antidepressant-induced manic episodes beyond that which was observed with placebo. The authors thus concluded that antidepressants are effective and safe in bipolar depression, based on the most rigorous available data.

In terms of interpretation, it is worth noting that a meta-analysis represents an observational study of studies. A meta-analysis of 5 randomized clinical trials is not necessarily more valid than 1 well-designed, randomized clinical trial, because the benefits of random assignment and the removal of confounding bias within a sample are lost in a meta-analysis; this results in the problem of heterogeneity between study samples. One may believe this issue to be less relevant if study results agree (as it appears they did in this meta-analysis), but this apparent agreement hides important unexplored heterogeneity. While this heterogeneity does not invalidate the meta-analysis, it can lead to its misinterpretation.

The 5 studies in the meta-analysis contain the following heterogeneity:

The 2001 study by Nemeroff and colleagues¹³ was the only placebocontrolled study that found no evidence of acute antidepressant response (all patients received baseline lithium(Drug information on lithium)). Another study nonrandomly assigned lithium to 37% of the patients in the antidepressant group, but only 21% in the placebo group.¹⁴ This is about a 77% increase in lithium use among the antidepressant group—hardly a fair assessment of fluoxetine(Drug information on fluoxetine) versus placebo.

Two other studies compared antidepressant alone with placebo alone, and 1 large study (58.5% of all meta-analysis patients) compared olanzapine(Drug information on olanzapine) plus fluoxetine with olanzapine alone (the term "placebo" was used improperly to refer to olanzapine plus placebo).¹⁵ Thus, the authors of the meta-analysis interpreted data from a comparison of olanzapine/fluoxetine combination with olanzapine plus placebo, but interpreted the data as a comparison of fluoxetine with placebo (by simply canceling out or ignoring the olanzapine component). However, olanzapine is not an inert substance and thus cannot be ignored; this is another source of heterogeneity for the overall meta-analysis.

While these studies may suggest acute antidepressant efficacy compared with no treatment or olanzapine alone, they do not address the most relevant clinical issue of acute antidepressant efficacy when compared with lithium, the most proven mood stabilizer.

With regard to antidepressantinduced mania, 2 studies comparing antidepressants alone with placebo alone strangely reported no mania in any patients. This is an oddity, if true, because it would suggest either that even spontaneous mania did not occur during the study or that perhaps manic symptoms were not adequately assessed. As described above, another study preferentially prescribed lithium more in the antidepressant group,¹⁴ possibly providing unequal protection against mania.

Although the olanzapine/fluoxetine data suggest no evidence of switching while using antipsychotics, our re-analysis of the lithium-plus-paroxetine (or imipramine(Drug information on imipramine)) study indicated that the manic-switching rate was 3 times higher with imipramine than with placebo (risk ratio, 3.14), with asymmetrically positively skewed confidence intervals (0.34 to 29.0). When combined with other studies that showed higher switching rates with tricyclic antidepressants than with other antidepressants, this heterogeneity suggests that antidepressant switching cannot be ruled out.

It also should be noted that these short-term (up to 10 weeks) studies are relevant only for the acute depressive episode, if they are relevant at all. They do not provide any evidence in support of the long-term maintenance use of antidepressants in BD. In fact, a previous systematic review of multiple randomized clinical trials found such use to be ineffective.¹⁶

In summary, our critique touches not only on the validity of the meta-analysis by Gijsman and associates¹² but also on its generalizability as a result of unexplored heterogeneity. Major conflicting results between one adequately designed study using lithium and the remaining studies are hidden by the apparent overall agreement among the studies. These positive conclusions of the meta-analysis appear to be premature when clinical options involve the use of proven mood stabilizers, such as lithium, with or without antidepressants.

Stanley Foundation antidepressant maintenance study

The Stanley Foundation Network has also just analyzed the results of an acute and maintenance treatment study of bipolar depression, in which 159 patients with acute depression (in 228 treatment trials) were randomly selected to receive venlafaxine, sertraline(Drug information on sertraline), or bupropion, added to standard mood stabilizers, with double-blind treatment for 1 year.¹⁷ The results demonstrated similar acute response and remission rates, with acute response occurring in about 50% of patients and acute remission occurring in about 30% of patients. (Acute response was defined as a 50% decline in depressive rating scores at 8 weeks, and acute remission was defined as minimal absolute final depressive rating scores at 8 weeks.)

The absence of a placebo group prevents one from being certain that such benefit is better than natural history, although clinical experience would suggest that it may be. Acute manic switch rates were much higher with venlafaxine than with the other agents at 8 weeks (30.3% with venlafaxine vs 14% to 15% with bupropion or sertraline) (Robert M. Post, MD, personal communication, 2006); relative risk and confidence intervals, in our calculation of an earlier presentation of this dataset, were 2.08 (1.00, 4.36) for increased risk with velafaxine. At 1-year follow-up, the remission rates (without any manic/ hypomanic switch) had declined to only about 16.3% of the whole sample, and longer-term switch rates continued to accumulate with all 3 agents.

Again, the absence of a placebo group prevents one from definitively attributing those manic switches to the antidepressants, or from concluding that the 16% 1-year remission rate is more than what would have occurred by natural history. Nonetheless, this study agrees with the earlier Stanley Foundation study⁹ in suggesting that only up to 20% of patients with bipolar depression benefit from long-term treatment with antidepressants. Or conversely, up to 80% of patients with bipolar depression do not need and do not benefit from longterm antidepressant treatment.

Further, venlafaxine appeared to demonstrate higher acute manic switch rates than bupropion or sertraline. Also, these are the first randomized data with sertraline in bipolar depression (previous studies of serotonin reuptake inhibitors had mainly looked at paroxetine(Drug information on paroxetine)), and these data are somewhat reassuring that sertraline is not markedly high in its acute manic switch risk. Lastly, these data confirm the previous small dataset on bupropion in bipolar depression (n = 15),¹⁸ and validate the clinical practice of using bupropion preferentially in bipolar depression.

STEP-BD antidepressant discontinuation study

For the past 5 years, as part of the National Institute of Mental Health–sponsored Systematic Treatment Evaluation Program for Bipolar Disorder (STEP-BD), our group has been conducting a study with the same design as the Stanley Foundation antidepressant discontinuation study, except that our design is randomized (although not double-blind). Thus, in this study, patients who respond initially to antidepressants plus mood stabilizers are randomly selected to either continue to take the combination or discontinue the antidepressant. Patients are followed up for 1 year.

In a preliminary analysis of 69 patients,¹⁹ our primary outcome showed similar mood morbidity in both groups, indicating no added benefit with the antidepressant. In a priori secondary outcomes, there is increased mood cycling with the antidepressant group, which is a randomized confirmation that antidepressants actually worsen the long-term course of BD.

Furthermore, in the rapid-cycling subgroup, antidepressants were associated with worsened mood morbidity, primarily with more depressive symptoms. This again confirms the only earlier randomized study of the association of antidepressants with worsening of rapid cycling.²⁰ Our study was analyzed in interim fashion this past year when it was near completion of enrollment; however, remaining patients will continue to be monitored for another year before a final analysis and publication of these data can be completed.