Fact-Checking the New Yorker
The real results from the STAR*D trial
Posted March 14, 2010
In the March 1 issue of the New Yorker, Louis Menand surveyed the topsy-turvy world of treatments for depression, writing in part of the conflicting evidence regarding the efficacy of antidepressants. The strongest evidence for the drugs, he suggested, came from the NIMH's STAR*D trial. Here's what he wrote:
Response to antidepressants is extremely variable. It can take several different prescriptions to find a medication that works. Measuring a single antidepressant against a placebo is not a test of the effectiveness of antidepressants as a category. And there is a well-known study, called the Sequence Treatment Alternatives to Relieve Depression, or STAR*D trial, in which patients were given a series of different antidepressants. Though only thirty-seven percent recovered on the first drug, another nineteen percent recovered on the second drug, six percent after the third, and five percent after the fourth-a sixty-seven-percent effectiveness rate for antidepressant medication, far better than the rate achieved by a placebo.
That statistic--that two-thirds of the patients eventually "recovered" in the STAR*D trial--has become an oft-cited one. The implication is that if depressed patients try a succession of antidepressants, they are likely to find one that "works" and keeps them well. Unfortunately, the results from the STAR*D trial do not support that belief.
Here's the data that was reported by lead investigator John Rush and several of his colleagues in a 2006 article titled: "Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: A STAR*D report."
There were 3671 adults with major depression who entered the "first stage" of the study. Of that group, 1,346 patients (36.8%) saw their depressive symptoms remit. Of those who didn't remit, 1,439 entered the second stage of the study and switched to a new antidepressant (the rest of the non-remitters dropped out.) In this second stage, 439 remitted (30.6%). Only 390 non-remitted patients agreed to give a third antidepressant a try; 53 remitted in that stage (14%.) In stage four, 16 of 123 patients remitted (13%).
In sum, of the 3671 patients that entered the study, 1,854 remitted (50.5%). Yet, Rush and the other STAR*D investigators reported in their various articles that the "overall cumulative remission rate was 67%," which of course raises the question of where this number came from.
The higher remission rate was purely "theoretical." The researchers assumed that if those who dropped out during the various stages had instead stayed with the protocol all the way through stage four, they would have remitted at the same rate (in the various stages) as those who did stay in the study. This, of course, is a big assumption, and it also hides the fact that if this multiple drug-therapy strategy is employed in the real world, a significant percentage of patients won't stay with it.
This brings us to the second part of the STAR*D story: what percentage of the patients who saw their symptoms remit stayed well through a 12-month followup? In the New Yorker story--and this is how the data is usually presented to the public--the implication was that once a patient finds a drug that "works," he or she stays well. But this study didn't document any such long-term recovery.
According to the follow-up protocol, the remitted patients were supposed to make a call to an "interactive voice response system" monthly so that their depressive symptoms could be assessed. However, only 1,174 of the 1,854 remitted patients made at least one call to the interactive system. In other words, many remitted patients did not participate in the follow-up study, and even many of those who did only remained in it for a short time, rather than for a full year. During this incomplete followup, 37% of the 1,174 patients reported that they had relapsed.
In other words, of the 3,671 patients who entered the trial, only 737 individuals (20%) remitted and then reported, at some point during the 12-month followup, that they had stayed well. The remaining patients (80%) either never remitted, or dropped out at some point, or relapsed during the followup. Moreover, it's unclear from the published results how many of the 737 non-relapsed patients stayed in the followup study for a full year.
In short, the study data did not tell of a form of care that helps two-thirds of all patients recover and stay well for a longer period of time. The 67% recovery rate reported in the New Yorker, a figure that has been often cited to show that if patients are tried on multiple antidepressants they are very likely to find one that works for them, isn't a "real-world" number. It's a number that tells of a "theoretical" remission rate, and it hides the fact that many remitted patients then quickly relapse.
In his article, Louis Menand concluded that the STAR*D trial proved that antidepressants, as a class, have a much higher effectiveness rate than placebo. That claim raises a question to be explored in a follow-up post to this blog: What is the long-term recovery rate for unmedicated depression? Although there was no placebo group in the STAR*D study, in the past decade NIMH-funded investigators did conduct a study of the long-term course of untreated depression. Thus, we can compare the results from that study to see if STAR*D proved, as the New Yorker reported, that antidepressants, as a class, are much more effective than placebo in helping patients recover. Most readers, I believe, will find the results eye-opening.