The use of antidepressants in bipolar depression is extremely common and yet it has little scientific evidence to support it. In fact, antidepressants are the most commonly used class of medications in bipolar illness, given to about 50% of patients treated for that diagnosis in the United States. In contrast, the most proven effective treatment in bipolar illness, lithium, is given to less than 10% of patients treated for that diagnosis in the United States.
This common practice occurs despite the fact that there are very few randomized clinical trials (RCTs) that support benefit with antidepressants in bipolar depression. And there are many randomized clinical trials that show that antidepressants are ineffective in bipolar depression. Why is there this disconnect between evidence and practice?
The reluctance to stop using antidepressants in bipolar illness needs to be understood not only in the context of their probable inefficacy but also in the context of the evidence for some harms, such as causing mania, and causing more mood episodes over time.
Let's review the scientific evidence first, and then talk about the clinical reluctance to implement that evidence straightforwardly.
We begin with RCTs of acute bipolar depression. There have been a number of meta-analyses of antidepressants in acute bipolar depression over the last decade. In the most recent such review, about half a dozen trials met inclusion criteria and overall, antidepressants weren’t statistically better then placebo. It’s a standard approach in evidence based medicine not to prescribe treatments which are proven ineffective. So we’re left with the question why clinicians and many experts still use and recommend antidepressants in acute bipolar depression. Whatever their reasons, they can’t base their judgments on these RCTs.
Now we turn to RCTs of maintenance prevention of mood episodes in bipolar illness. The PL editor published a meta-analysis on this topic, which found that antidepressants weren’t better than placebo, when used with mood stabilizers, in prevention of mood episodes in bipolar illness. Most of these studies involved older classes of antidepressants, such as the tricyclic antidepressants (TCAs). There are very few studies of the newer antidepressants, like the serotonin reuptake inhibitors (SRIs), in long-term treatment of bipolar illness. There is in fact only one placebo-controlled maintenance trial of any SRI in bipolar illness. In that study, which has been presented at conferences but not yet published, citalopram was equivalent to placebo. In other words, there are no studies - zero - that show that any modern antidepressant is more effective than placebo in prevention of mood episodes in the maintenance phase of treatment of bipolar illness.
Clinicians often refer to manic induction with antidepressants. Usually the term “manic induction” isn’t well defined. PL recommends that acute mania related to antidepressants be defined as occurring within a few months after starting treatment. It’s important to distinguish between immediate and long-term harms with antidepressants. In the PL view, acute mania isn’t the major problem with antidepressants. Long term worsening of mood episodes, usually depressive rather than manic, is a bigger problem. We’ll turn to that question in the next section of this special article.
Do antidepressants cause acute mania? If we look at the randomized trials, some experts would answer in the negative, because many report equivalent rates of mania in subjects treated with antidepressants versus placebo, and thus a causal connection can’t be established for antidepressants. Yet those RCTs aren’t set up to identify and test the hypothesis that antidepressants cause mania. Rather they’re designed to test the hypothesis of efficacy with antidepressants over placebo, not worsening with mania. The frequency of antidepressant-induced mania is about 10-30% with SRIs and 25-50% with TCAs, thus only part of the overall sample would experience this outcome. Since these studies are powered for efficacy in the whole sample, not side effects in part of the sample, these RCTs don’t have statistical “power;” this means there aren’t enough patients in the studies to observe enough manic episodes so to make a statistical distinction between antidepressants and placebo. This is the classic statistical fallacy - called false negative bias - of saying that nothing happens when your study is too small to see if something happens. Further, in many of those RCTs, patients received antimanic medications, such as neuroleptics or mood stabilizers, which would reduce the frequency of antidepressant-induced mania even more.
Nonetheless, even with these design issues, there are a number of randomized trials - involving TCAs - which show higher rates of mania with antidepressants over placebo. The reason those studies are statistically significant has to do with the larger effect size of frequency of antidepressant induced mania with TCAs (in 25-50% range as opposed to about half that rate with SRIs).
With the above context, PL draws the following conclusions: Antidepressants (like TCAs) have been proven to cause mania in RCTs (in about 25-50% of subjects). SRIs have about half that rate (10-30%). Rates are even lower with concomitant use of mood stabilizers or neuroleptic agents, and also lower in type II (5-10%) than in type I bipolar illness (10-50%). The specific modern antidepressants that have been shown to have the lowest risk of causing acute mania in randomized trials are paroxetine and bupropion.
Rather than acute mania, the PL view is that the most important harm associated with antidepressants in bipolar illness is long-term worsening, meaning causing more and more mood episodes over time. These mood episodes tend to be depressive rather than manic, which leads to the paradoxical fact that antidepressants worsen depression long-term in bipolar illness. In some patients, this long-term worsening leads to a rapid-cycling course, meaning four or more mood episodes yearly. Typically, these patients are labeled as having “treatment refractory” bipolar illness, while in fact they often aren’t treatment refractory. Rather their bipolar illness has been worsened by the mood-destabilizing effects of antidepressants, counteracting the benefits of mood stabilizers. It’s not that these patients have failed to respond to multiple mood stabilizers, as often assumed, but rather that the constant use of long-term antidepressants impedes the mood stabilizers from working. A full and fair trial of a mood stabilizer in this setting must happen in the absence of any antidepressants. Therefore, the PL recommendation would be to stop antidepressants, and then resume mood stabilizers, including those that had been used the past, and frequently a much better response occurs.
The RCT evidence for these observations begins with studies from the 1970s, which used the on-off design to show that when patients with rapid-cycling bipolar illness received TCAs, their cycling worsened, whereas when they were switched to placebo, the cycling improved. Three decades later, the PL editor conducted two maintenance RCTs of modern antidepressants in bipolar depression, and examined outcomes in rapid cycling versus non-rapid cycling subgroups. In both studies, patients with rapid-cycling bipolar illness had more depressive episodes if they continued antidepressant use long-term, defined as up to one year, as opposed to either not being treated with antidepressants at all, or stopping antidepressants after the acute phase. Thus all three RCTs on this topic have found that antidepressants are associated with rapid cycling bipolar illness. Some experts refer to observational data suggesting otherwise, but a basic principle of evidence-based medicine is that randomized data are more valid than observational data, as we’ll discuss more below, and thus the latter can’t be used to refute the former.
Since rapid cycling occurs in about one-quarter of patients with bipolar illness, it’s reasonable to conclude that antidepressants worsen bipolar illness overall in at least 25% of subjects, a frequency confirmed by observational data on this topic.
In short, the most valid available scientific evidence indicates that antidepressants cause or worsen rapid-cycling bipolar illness.
Given the above scientific evidence, we’re left with the question why clinicians, and also many bipolar experts, use or recommend these agents. The PL editor has interacted with many clinicians and experts on this topic for two decades. A common view among bipolar experts is that if clinicians continue to use these medications, despite the scientific evidence to the contrary, they must know something. Perhaps clinician see some real benefit that isn’t captured in the RCTs. This is a possibility. But one would think that after 30 years of research in many RCTs conducted by researchers who generally have been very positively disposed towards antidepressants, some benefit would have shown up. Another hypothesis could be made which is much more consistent with the basic principles of scientific research and clinical medicine. This hypothesis, as discussed the April 2015 PL issue, has to do with the concept of confounding bias. Confounding bias reflects the idea that there are many factors in clinical practice that influence the results seen. Clinicians as well as patients don’t necessarily know which factors are in play. For instance, it may seem that coffee causes cancer, as repeatedly shown in many huge observational studies. But this association between coffee and cancer is not causal, because of the confounding factor of cigarette smoking. People who drink more coffee also tend to smoke cigarettes, and the latter factor is causal for cancer. Similarly, the fact that clinicians think that antidepressants are associated with improvement in bipolar depression doesn’t mean that there is a causal association. The whole point of randomization is to get rid of all the other confounding factors in clinical practice so that a causal scientifically valid judgment can be made. Placebo is a stand-in for natural history, not merely a reflection of psychological wishes, as is commonly assumed. If you have an illness which improves in many people over time, then you have to show that medication does better than the natural course of recovery. Over a century of natural history research, long before any treatments were available, shows that the natural history of bipolar illness is such that episodes last 2 to 4 months for mania and 3 to 6 months for depression. The reason these mood states are called “episodes” is because they have a natural end, as well as a beginning. They will end, even without any treatment, in 3 to 6 months for bipolar depression. Thus, when a clinician gives an antidepressant to a patient with bipolar illness who has been depressed for two months, and then the depression improves two months later, this would be expected even if the clinician had received nothing. The clinician's interpretation of antidepressant benefit is disproven by the randomized trials which show the same benefit with placebo.
Unfortunately, many bipolar experts tend to ignore the reality confounding bias, in their interpretations of the disconnect between the research studies and clinical practice on this topic. Clinicians haven’t been aware enough of the reality of confounding bias in the assumptions they make based on their clinical experience. Both groups would do well to be more cautious in assuming clinical effectiveness when scientific evidence demonstrates the contrary.
Regarding the issue of harm with antidepressants, clinicians tend to be more open to the notion that these drugs cause mania, as opposed to experts who mistakenly ignore the false negative bias of RCTs that are not statistically powered to assess manic induction with antidepressants. Clinicians can’t ignore the reality of patients getting markedly manic soon after starting antidepressants. In contrast, the long-term worsening of mood episodes caused by antidepressants can be hard for clinicians to identify, since those episodes occur a year or longer after antidepressant treatment began. A direct association can be hard to confirm in the real world of clinical practice, where medications are changed frequently, and where many stressors and life events occur over years of follow-up. The RCTs of maintenance treatment in bipolar illness come to the rescue by clearing out confounding factors and demonstrating a causal association between antidepressant use and long-term worsening of bipolar illness with a rapid-cycling course.
Another factor that may be relevant here is that clinicians and patients maybe misled by the English language. As PL has suggested in the past, the word "antidepressant" is misleading, because it implies, to clinicians and patients alike, that these medications should be useful for any kind of depression, despite the fact that the research literature indicates that they aren’t effective for bipolar depression, and as discussed in the last issue of PL, they may not be very effective for many types of major depressive disorder (MDD). PL has recommended the phrase "monoamine agonists" as a more neutral term, leaving open the clinical question of which conditions this class medications can help. For instance, PL suggests that monoamine agonists appear to be more consistently helpful for anxiety rather than depressive symptoms; thus the phrase “antidepressant” rather than “anxiolytic” sends clinicians and patients in the wrong direction.
The PL editor recalls that once, upon giving a lecture on this topic, and reviewing all the above studies, an older clinician left the room and commented to a colleague: The studies may be this way, but I'm still going to use antidepressants. Although clinical practice has its strengths, it also has its limitations. The main limitation of clinical practice involves confounding bias: the many competing and conflicting and often unknown factors that influence outcomes in the real world. If clinical practice is the art of medicine, and randomized trials are the science of medicine, then these two aspects should be seen as complementary. The art of medicine is strengthened by the science, and science needs to be applied with the most effective art possible. The great William Osler put it this way: Medicine is the lifelong attempt to correlate art with science. Notice that it isn’t the other way around. Science isn’t there to correlate with art. It’s not the job of research to confirm what clinicians already believe, but rather, more often, to refute clinical beliefs. Science is about falsifying our hypotheses, not simply confirming them.
What about type II bipolar depression? Future PL issues will address this matter in more detail, but a few general comments can be made here. It’s commonly believed by many clinicians that type II bipolar depression is more responsive to antidepressants than type I bipolar depression. This impression is based partly on a number of RCTs that report benefit with antidepressants. A key concern about those studies, from the PL perspective, is that they are "enriched", which means that patients are preselected to respond to the antidepressant being studied, before the study even begins. The PL view is that this preselection of treatment response in a study of treatment response prejudges the matter. This research design problem hasn’t been appreciated, in the PL view, by the FDA, which accepts enriched studies for indications for treatment. The same critique holds for maintenance trials of antipsychotics in bipolar illness, some anticonvulsants in bipolar illness, and antidepressants in MDD. This critique has been described in the scientific literature, and future PL issues will return to in more detail. For now, the main point to make here is that the apparent efficacy of antidepressants in some bipolar type II studies may be overstated because of the bias of the enriched research design. Further, when antidepressant efficacy has been examined in RCTs of bipolar depression comparing type I and type II subgroups in the same trial, no differences exist. In other words, antidepressant efficacy in type II bipolar depression was the same as in type I bipolar depression, which means it was equally low in both groups.
In sum, the PL view is that the scientific literature of randomized trials most simply supports the conclusion that antidepressants are ineffective at best and harmful at worst in bipolar illness. PL is aware that many clinicians will disagree with these judgments and many experts will seek to interpret the data as positively as possible. PL isn’t opposed to any class of medications in theory, as many readers will know, since all classes of medications are recommended in PL for whatever use that is scientifically supported. But when randomized trials repeatedly show that a class of medications aren't effective, it’s important to draw the simple clinical conclusion that they shouldn't be prescribed routinely. In the case of bipolar depression, it may be that we’ve been giving a pass to antidepressants because they’ve been so popular for so long with both clinicians and patients.