The conventional wisdom: STAR*D for depression
The second large canon of psychopharmacology is the STAR*D study, a large sequential large randomized clinical trial (RCT) of over 1000 patients with an acute depressive episode of so-called major depressive disorder (Rush et al., 2006b). Before STAR*D, there were few RCTs comparing antidepressants to each other, and hardly any looking at outcomes after multiple failed trials.
The main outcome was to assess whether there was a difference between switching antidepressants versus combining them as adjuncts, after failure of initial antidepressant monotherapy.
The STAR*D protocol was as follows: First 1439 patients were treated openly with citalopram. If they failed to respond (n=727), they were then randomized double-blind to a different monoamine agonist or combination with two monoamine agonists (or other adjunctive agents like buspirone). If they failed this second trial, they were randomized to switching to tricyclic antidepressants (TCAs) or augmentation with lithium or thyroid hormone. If they failed this third trial, they were randomized to a MAOI or the combination of venlfaxine plus mirtazapine.
Response rates are shown in the figure.
The standard interpretation put forward by the researchers who conducted STAR*D was that there was no difference between switching versus combining antidepressants for refractory depression. A secondary conclusion was that most patients (about 2/3) improved after multiple antidepressant trials for an acute depressive episode (Rush et al., 2006a).
Alternative interpretations of STAR*D
The main outcome seems straightforward. Combining antidepressants versus switching agents did not differ notably. Thus, the primary outcome of STAR*D was not too controversial. The result would argue for switching antidepressants after initial nonresponse, since fewer side effects should ensue with one versus more agents.
Alternative interpretations become relevant to the secondary claim that most patients eventually improve. A closer look at the data finds that this 2/3 full interpretation is really only 2/3 empty.
As can be seen in the figure, treatment response was good in the first two episodes, but fell by half thereafter. By the fourth monoamine agonist trial, only 15% of subjects respond to any new treatments, even the most potent agents known, the MAOIs.
Besides the acute response issue, it is important to note that patients continued to be followed for up to a year, so that continuation of response was able to be assessed. On that outcome, it was found that even if patients respond, about 40-70% relapsed within one year even if they stayed on the same agents which led to acute response.
So the 2/3 initial response has to be cut by half for continued response, which leads to about 1/3 response up to one year. Again, the absence of a placebo group raises the question whether this response would be higher than natural history.
Further, the addition of response by treatment trial after trial obscures the fact that acute response fell markedly by the third trial, so that after that point, further agents had a very low likelihood of acute treatment response (only about 15%). That’s not good news. If a patient sees a clinician in a current depression that has failed to respond to 2-3 prior agents, the clinician would, based on STAR*D, have to tell the patient that there is a 85% likelihood of not getting better no matter what other antidepressants are used, including MAOIs.
STAR*D answered one question well: switch rather than augment antidepressants for refractory depression. But it raised other questions: First, like CATIE, are antidepressants long-term at all? And second, does anything work after a few failed trials?
The methodological take-home point is that even in a well-conducted clinical trial in which the primary objective is answered, like STAR*D, the secondary results can be very important clinically. And those results should not be interpreted in the most positive way possible alone, as done by the STAR*D researchers, but with attention to the maximum information one can derive from them. STAR*D tells us that antidepressant benefit declines over time, from 2/3 initially to 1/3 later; and that antidepressant benefit declines markedly after a few initial failed trials. These are important conclusions, though they tend to be ignored because they show that antidepressants are much less effective than commonly presumed.