Abstract
Introduction
There are scientific, clinical and cultural aspects to the debate about antidepressants in bipolar depression. In prior communications I’ve focused on scientific aspects. Over time, as I realized how the scientific evidence seemed unconvincing, I concluded that clinical and cultural aspects were important. Here, besides briefly recapitulating the scientific evidence, I will also comment on clinical and cultural factors.
To begin with some rules of clinical research worth making explicit: randomized data are more valid than observational data, because factors that influence the results besides what we are studying (confounding factors) are only removed by randomization. Sample size is less relevant; a large observational study does not contradict a smaller randomized study. Let’s keep in mind this basic fact of science (one can call it evidence-based medicine, EBM, if one wants).
Given that background, risks and benefits of antidepressants need to be assessed in four categories: acute efficacy, maintenance efficacy, acute manic switch and long-term mood destabilization. These different groupings often overlap and are mixed, leading to confusion.
Acute efficacy
Acute efficacy involves response within 8 weeks of starting an antidepressant for an acute clinical depressive episode. This has been the most studied aspect of antidepressants in bipolar depression. Despite an early meta-analysis of five highly heterogeneous studies, which claimed efficacy for antidepressants (Gijsman et al., 2004), recent larger, better-designed studies have proven inefficacy. For instance, the largest (n = 366) well-designed (most patients received proven mood stabilizers like lithium or anticonvulsants) RCT of bipolar depression was a comparison in the STEP-BD study of bupropion versus paroxetine versus placebo (Sachs et al., 2007). The antidepressants were completely ineffective, equivalent to placebo. Another recent very large study (n = 740) assessed the question of antidepressant monotherapy, in a placebo-controlled comparison of paroxetine alone versus quetiapine alone in acute type I bipolar depression (McElroy et al., 2010). It also found paroxetine monotherapy completely ineffective, equivalent to placebo, while quetiapine was effective.
Not surprisingly, if one adds these huge studies (total n = 1106), which show inefficacy, to the prior meta-analysis of mostly small studies and one large olanzapine study (n = 463) (Tohen, et al., 2003), which claimed efficacy, a new meta-analysis concludes inefficacy (Sidor and Macqueen, 2011).
Maintenance efficacy
Even if a drug is effective acutely, one cannot presume that it will remain effective in long-term prevention of new episodes. Our meta-analysis of maintenance RCTs of antidepressants in bipolar depression found consistent evidence of inefficacy when antidepressants were added to mood stabilizers (Ghaemi et al., 2008). When used in monotherapy, there was some benefit, but also a notable increase in manic episodes.
Proponents of antidepressants often cite an observational study of antidepressant discontinuation, which found, among those who responded to antidepressants for acute bipolar depression, increased relapse rates if antidepressants were stopped at 1-year follow-up (Altshuler et al., 2003). Observational studies are full of confounding bias. Randomized data are more valid.
Our group repeated that study with a randomized design (Ghaemi et al., 2010); this is an ‘enriched’ study, preselected for the best responders to antidepressants. Even so, when randomized to stay on or come off antidepressants at 1-year follow-up, we found that antidepressants conferred only mild symptomatic benefit. They did not prevent new depressive episodes, nor increase time in remission.
The randomized data can be objectively and simply interpreted: antidepressants do not prevent depressive episodes in bipolar disorder. They do not have maintenance efficacy.
Acute manic switch
Acute manic switch can be defined as getting acutely manic with an antidepressant during acute phase treatment, usually 2 months. The same heterogeneous meta-analysis claimed no increased mania with antidepressants versus placebo (Gijsman et al., 2004). But the largest study in that meta-analysis had all patients on a neuroleptic, olanzapine (Tohen et al., 2003). If I study a potentially fever-inducing study drug in patients who all receive acetaminophen, I can hardly conclude that the study drug does not cause fever.
The same consideration would apply to the common misinterpretation of the STEP-BD acute bipolar depression study (Sachs et al., 2007), where there was no higher mania with bupropion or paroxetine over placebo. Some have mistakenly concluded that, therefore, antidepressants in general do not cause mania. But all those patients were taking anti-manic medications – lithium or anticonvulsants or neuroleptics, or combinations of those agents. Again, one cannot say that something is not happening with one hand, when one prevents it from happening with the other. At most, one can say that some of these newer antidepressants may not cause mania, or may have a low risk of causing mania, when combined with anti-manic agents. One cannot draw general conclusions about the entire class of antidepressants in the abstract.
Plenty of RCTs, even in that meta-analysis, show that tricyclic antidepressants (TCAs) cause mania more than placebo (Gijsman et al., 2004). And in another RCT, venlafaxine caused mania more than twice as often when compared to other new antidepressants (Leverich et al., 2006). Some antidepressants, at least, clearly cause acute mania.
Long-term mood destabilization
What are the effects of antidepressants after the acute phase, after 2 months, into the maintenance phase? In a notable subgroup, about 25% of patients, antidepressants cause mood destabilization, which means that they cause more and more mood episodes over time, often leading to frank rapid-cycling. In those who already have rapid-cycling, this mood-destabilizing effect is especially prominent.
This claim is based on all the maintenance RCTs to address the matter, reviewed elsewhere (Ghaemi et al., 2003). Here I’ll emphasize the two studies that addressed rapid-cycling BD in particular. The first study assessed 51 patients with rapid-cycling, in whom TCAs were prospectively replaced with placebo in a double-blind randomized fashion (Wehr et al., 1998). A third (33%) experienced rapid-cycling directly related to antidepressants. Further study of that subgroup of 17 patients, through repeated on–off–on–off design, showed a definitive association between antidepressant use and rapid-cycling in 10 patients, a fifth (19.6%) of the original sample.
Similarly, in the study mentioned above (Ghaemi et al., 2010), we examined this matter with new non-TCA antidepressants. In those with rapid-cycling, antidepressant continuation led to a doubling of the frequency of depressive episodes, while in those rapid-cyclers in whom antidepressants were stopped in the maintenance phase, fewer depressive episodes were seen. We thus replicated and extended the prior TCA study with new generation antidepressants.
Some cite an observational analysis of the NIMH Collaborative Depression Study (CDS), which, correcting for only one confounding factor (baseline depression severity), claimed no association between antidepressant use and rapid-cycling (Coryell et al., 2003). One can objectively adjudicate between these conflicting results with the basic rule of clinical research: randomized data are not validly contradicted by observational data.
Type II bipolar disorder
Some may admit the above limitations for type I bipolar disorder, but deny them for type II. There are some differences. Acute antidepressant-induced mania or hypomania is less frequent in RCTs in type II than type I bipolar depression (Post et al., 2006). Some claim special efficacy for antidepressants, even in monotherapy, in type II bipolar disorder. These claims are based on one small RCT of 10 persons (Parker et al., 2006), and a number of RCTs conducted by the same research group (Amsterdam, 1998; Amsterdam and Brunswick, 2003; Amsterdam and Shults, 2010). The latter studies are often of the ‘enriched’ variety: subjects are preselected to be antidepressant responders and then randomized to stay on the antidepressant or come off (placebo or active control like lithium) (Amsterdam and Shults, 2010). As Goodwin and I explain elsewhere (Goodwin et al., 2011), despite being widespread, this is a scientifically invalid design. It is like asking people whether they like chocolate or vanilla cake, then choosing only those who like chocolate cake, and then randomizing them to get chocolate or vanilla cake. One guarantees that they will like chocolate cake; this does not prove the superiority of chocolate to vanilla. In contrast to these invalid designs, prior maintenance studies which were not enriched did not find efficacy of antidepressants in type II bipolar disorder (Ghaemi et al., 2008), and in our recent enriched antidepressant discontinuation study, we found no enhanced benefit with antidepressants in type II versus type I bipolar depression (Ghaemi et al., 2010).
Clinical and cultural aspects of the debate
Clinicians are sometimes skeptical of this research. I think this may reflect their own clinical experience, where they see patients improve on antidepressants. Clinical experience trumps research, in their minds.
If this is correct, bleeding – with leeches and lancets – should be acceptable to us, because for two millennia, it agreed with the clinical experience of millions of physicians. Why are we skeptical of the lancet, if we really think clinical experience is the ultimate arbiter?
Also for the past half century, physicians everywhere saw that hormone replacement therapy was effective for peri-menopausal women. What they did not see was a small but real increase in breast and endometrial cancer. A huge (over 10,000 patient) RCT was needed to show that clinicians were wrong: many of the purported benefits did not exist. And many real risks, not seen in clinical experience, were present. Most physicians agree with the consequent reduction in widespread HRT.
If we accept this history – both distantly with bleeding and recently with HRT – then we have to admit that clinical experience does not invalidate randomized data. The reverse is the truth (Ghaemi, 2009).
We should admit the truth of the surgical maxim (Cossman, 2007): Half of what I’m going to tell you is false; I just don’t know which half. That’s also the nature of clinical experience, the wisdom behind the saying: Believe half that you see, and none that you hear (Ghaemi, 2009).
We clinicians have to realize that our observational experience is affected by many factors we cannot control – confounding factors.
Despite this limitation, I agree with clinical skepticism in this way: the results of RCTs are average results, for the average patient. They do not apply to specific subgroups of patients, but, importantly, they do apply to most patients.
If RCTs show that antidepressants are ineffective, this doesn’t mean that 100.00% of patients will not respond, but that the majority of patients will not respond; a minority subgroup still may.
There is no need to argue against the proposition that one should never use antidepressants in bipolar depression. Who ever said never? I have published, repeatedly, my own clinical experience, which is that I use antidepressants in 19% of my patients (Ghaemi and Goodwin, 2001). I target antidepressants to the severe melancholic subgroup, which is at high risk of suicide, not because I know antidepressants work, but because, to modify Lyndon Johnson’s phrase, there are times in poker and politics and psychiatry where you have to play all your cards.
My own experience agrees with the RCTs: most patients with bipolar disorder who I see on antidepressants do poorly; I stop their antidepressants, give them more mood stabilizers, and they improve.
A final cultural comment: Sometimes it seems that colleagues in Europe, especially England (and the British Commonwealth), want to make a cultural debate out of a scientific question. At a conference, I recently heard a leading European researcher ask: ‘Why do Americans hate antidepressants?’. Antidepressants are the most common initial treatment for bipolar depression in the USA (Baldessarini et al., 2008); most American researchers in bipolar disorder also support their use. I don’t see a cultural factor here, though I am reminded of how British leaders in psychiatry were excessively skeptical about lithium in the 1960s and 70s, retarding its worldwide adoption unduly for decades (Malhi and Gershon, 2009).
Similar excessive skepticism may be slowing international awareness of the scientific evidence that antidepressants are largely ineffective, and if anything harmful, in bipolar disorder.