WE’VE REACHED THE end of our consideration of the antidepressant controversy—of the central question, whether antidepressants interrupt bouts of depression. I hope it’s clear that they do. Practicing doctors witness antidepressants’ efficacy daily, and the formal evidence supporting those observations is ample.
But once medicines work—what then? How long should treatment continue? Most Americans taking antidepressants have been on them for two years or more—30 percent for five years or more. This extended use creates intimate relationships with these medicines, for better and worse. Patients live unhappily, with side effects such as diminished sex drive, and contentedly, with a sense of security long absent in their lives.
Doctors rely on medication to stave off recurrence or prevent the worsening of partially treated episodes. In my practice, I have patients who pass in and out of depression, never shedding all their symptoms. Almost all these patients find benefits from antidepressants.
Libby is an unmarried special-education teacher, now in her sixties. Only occasionally, for intervals lasting months, is she fully free of depression. The difference that medications have made is in her functioning. Before seeing me, and in the years when we tried to sustain her (with talk, exercise, bright lights, and the rest) off medication, she had days and weeks when she could not drag herself to school. Now, on antidepressants “indefinitely,” she is able to work steadily, doing summer sessions and taking on private clients.
Her life is different. In her down periods, she may have morbid thoughts—“as well off dead”—but is almost never suicidal. She travels on vacation. She golfs. She jogs. She has lunches with friends. She has a conventional middle-class existence. That scorned outcome that can be something of a miracle.
Do Libby and I worry that she is overexposed to medication? We do. I lower her doses in springtime and raise them in autumn. In time, antidepressants may fail her or produce intolerable side effects. For now, they enrich her life. So far so good, and thank goodness for that.
When chronic medication use constitutes best practice is complicated. We always prefer for patients to be off medication. But on the narrow question of efficacy, there is extensive research. It goes under the name maintenance, a term we encountered when discussing pill placebos.
For patients who have recovered on an antidepressant, what happens if they continue the regimen? What happens if they stop? Approaching the maintenance literature the way that psychiatry’s critics approach the FDA data—considering effect sizes only—we’d call the case open-and-shut. Every randomized, controlled trial finds that staying on medicine protects against further episodes.
One layer down, discussions of maintenance become complex. The trials are liable to confounds—new ones, distortions we have not yet met. The result is endless debate about the consistent numbers.
As in the central antidepressant controversy, the fuss stands at a distance from patient care. In practice, it seems clear that antidepressants, once someone does well on them, are stabilizing. Often, that effect is what patients, both those who get full relief and those such as Libby who limp along, value most. Medications give the feeling of having a floor to stand on. At the least, intervals between setbacks are longer.
Antidepressants’ reliability in maintenance is another quality that makes them transformative. Our discussion began with Nora, who had put together four good years on medicine. For patients to follow Nora’s arc of improvement is something like the norm. They arrive dominated by depression. It controls their consciousness. When they are not depressed, they know that they might be. Small disappointments can cast them into the abyss. By muting apprehension, antidepressants allow these patients to act. At home, at work, they can make commitments. It is hard to overstate the importance of this contribution, from medication and from psychiatric care in general: making the future predictable. Confidence lets people pursue their dreams.
To investigate this potential, protection from mood disorder, scientists take patients who have done well in a brief drug trial and then assign them randomly to continue with that drug or switch to placebo. The research tracks relapses: Over time, how many patients in each arm of the trial become depressed? Generally, it is people with chronic or recurrent depression who enter these experiments.
I have said that the results are consistent. An integration involving more than nine thousand patients found that those switched to placebo were three times as likely to return to depression as those kept on medicine. In each of fifty-four individual trials, medication outperformed placebo. Few research collections in medicine are this uniform in their results. Few meta-analyses are likely to be as reliable. The effect size for prevention is in the 0.6 to 0.7 range, as sizable a benefit as we have seen.
In this field, too, there are FDA data. For over a quarter century, the agency has required that companies proposing to market an antidepressant submit a maintenance trial, registered in advance. There are fifteen, a complete set, no file drawer. Sadly, as with other drug-company trials performed for the FDA, most in this set are deeply flawed. The FDA scientists have published a technically impressive overview analysis, and still—complete sample and statistical expertise notwithstanding—we will need to take the results with a grain of salt. That said, they show that in every trial patients who stay on antidepressants suffer fewer and later subsequent episodes than patients switched to dummy pills. On placebo, the rate of relapse doubles.
Academic trials, like the R&R study, out of Penn, Rush, and Vanderbilt, give the same answer. That trial had a late phase in which patients who had recovered were kept on medicine or tapered off. Stopping medication more than doubled the rate of recurrence. Prior time spent in psychotherapy—months of treatment—offered no additional protection. All that mattered was continued prescribing.
For patients with long-standing mood disorders, once a bout of depression has ended, antidepressants are protective. The benefits hold for any interval that has been studied: weeks, months, or years. Continued treatment cuts the risk of recurrence by a half to two-thirds.
Despite this copious and consistent evidence, some experts doubt that maintenance is protective. Often, their objection stems from what happens when patients are taken off medication and switched to placebo pills. In early studies particularly, for patients put into the control arm, researchers sometimes stopped the drug abruptly, simply supplying placebo pills that looked like the antidepressant.
Quitting cold turkey can lead to discontinuation effects, like headache, dizziness, a sense of malaise, or “brain zaps,” an unpleasant flash or shiver that feels like an electric shock. Those symptoms play out over days or, more rarely, weeks. But can quick discontinuation also hasten the return of depression? If so, then maintenance trials are comparing destabilized people (those taken off medication) with unharmed ones (still on drug), a terrible confound.
In an observational study—looking, for instance, at patients who simply decided on their own to quit—a Harvard psychiatrist, Ross Baldessarini, found that an abrupt transition increased the odds of relapse, particularly in the first six months. But patients who bolt and stop treatment may be different from patients who listen compliantly, tapering their medication at a time and in the fashion that their doctor recommends. Perhaps Baldessarini was only noting a contrast between methodical people and impulsive ones.
Randomized trials tell another story. When Baldessarini reviewed twenty-seven controlled studies, he had found no harm from abrupt cutoffs: “Contrary to prediction, gradual discontinuation … did not yield lower relapse rates.” In that analysis, patients made to go cold turkey were less likely to enter a new episode of depression than patients taken off drugs slowly. Similarly, no disadvantage from quick tapers showed up in the large (nine-thousand-plus patients) meta-analysis, a Lancet study of more than four thousand patients, or the FDA synthesis.
I have spoken of my love for idiosyncratic studies. One of my favorites, conducted by French researchers, confronted the discontinuation issue head-on. In a heroic effort that involved the participation of eighty-three psychiatric centers, the team enrolled 371 patients who met stringent and seemingly quirky criteria. The men and women enrolled were at high risk for depression. Each had suffered at least three episodes in the prior four years. Each was now in remission. None had used Zoloft to get there.
Now, if the patients were still on some other antidepressant, they were taken off and observed for two to four months. The 288 participants who remained, those who had not dropped out and had continued to do well, were randomly assigned to take Zoloft or dummy pills.
The beauty of this ambitious setup is how it addresses the problem of discontinuation. All the research subjects have come off drugs. If, months later, some patients remain at heightened risk of recurrence, they will have equal odds of receiving Zoloft or placebo.
Also, the test is especially rigorous. Normally, patients who need maintenance will continue with the medicine they did well with in the first place. Here, to succeed, Zoloft had to protect patients who had recovered on different drugs, including ones (such as reboxetine, in David Healy’s experiment in his own lab) where success might pick out people likely to feel worse on a medicine, such as Zoloft, that affects serotonin transmission primarily.
Given the stringency of the trial, the findings were striking. Over eighteen months, among patients on placebo the recurrence rate was double that of patients on medication, and with placebo the recurrences came sooner. This difference appeared despite a highly supportive study environment, involving “repeated contact with mental health professionals”—minimal supportive psychotherapy. Once that factor is in play, so is additivity. Maintenance—here, putting remitted depressed patients on a new antidepressant—may be yet more stabilizing than the numbers suggest.
The French trial proved too complex to inspire replication. Still, it serves as a preliminary answer to the principal objection to the results of maintenance trials. If the “harm via withdrawal” problem is a red herring—if in maintenance trials, what you see is what you get—then continued use of antidepressants really does cut the recurrence rate by half to two-thirds.
Drug continuation is likely yet more protective. Maintenance studies suffer from an especially distorting form of (our old nemesis) differential dropout. As trials stretch on, patients stop showing up. Here as elsewhere, exit questionnaires find that patients leave placebo arms because the “treatment” is not working. The dropout rate in the placebo arms of the FDA maintenance trials reached almost two-thirds, as opposed to less than half in the drug arms. The analyses tend not to count disappearance as relapse. When failures walk away and the data focus on the few who persist in taking dummy pills, placebo will get more credit than it should.
In the early going, then, drugs are doing better, and placebos worse, than the numbers suggest. After that point, all randomization has been lost, so the results become harder to interpret.
Let’s look in on a trial at the six-month mark. Many patients assigned to take dummy pills will have relapsed or dropped out. Who’s still participating? At this late stage, the placebo arm contains many people who do not require medication to stay well—people who are naturally resilient (or were misdiagnosed from the start). That’s why they’ve gotten this far on inert pills.
Meanwhile, the medication arm is mixed. Besides the naturally resilient and the misdiagnosed, it will have held on to people with highly recurrent mood disorder who need antidepressants to stay afloat and are benefiting from them. Overall, it’s a sicker group.
If we follow results for another six months, we’re conducting an unfair experiment, one that compares naturally stable people (in the placebo arm) to less stable people (in the antidepressant arm). The better that antidepressants work in the first six months—the more that they protect fragile patients—the less reliable the remainder of the trial will be.
The term of art for this problem is the differential sieve. The placebo arm contains a fine sieve that (over six months) lets through only healthy people. The antidepressant arm contains a wide-gauge sieve that lets through sick people as well. This setup is precisely as bad as if our triage nurse had sent all the sicker patients to one arm of the trial.
The FDA authors made note of this confound: “The population, of course, changes over time. Presumably, by 6 months, most patients likely to relapse on placebo will have done so, and new relapses become less frequent. The overall drug-placebo difference in relapse rate, however, is maintained.”
Beyond the half-year mark, the placebo arm contains few vulnerable people, so we would expect relatively more recurrences among patients still on antidepressants. That doesn’t happen. Patients in both arms remain steady, and in some trials antidepressant continues to outperform dummy pills.
Because the differential sieve confound is so distorting—because at a certain point people in the control group are at low risk for depression—no one knows precisely how much extra stability antidepressants afford in long-term use. But in his overview of randomized trials, Baldessarini found a substantial ongoing advantage almost to the three-year mark. At that point, 60 percent of those on medication remained depression-free—but under 20 percent on placebo.
And before the differential sieve kicks in, the advantages of medication are undeniable. Baldessarini looked at the first ten or twelve months of maintenance for patients who had suffered several past episodes of depression. On an antidepressant, they did almost as well as patients with less history of mood disorder. On placebo, they were almost certain to relapse. The issue here is not severity but chronicity. In the patients with the most persistent illness, discontinuing medication may quadruple the risk. One of the clearest results in psychiatry is that, for patients with recurrent depression who respond to an antidepressant, continued use over a year offers substantial protection.
There’s an additional reason that the clinical reality is likely to be better than the research implies. In the French trial, as in many others, if patients experienced deterioration that, in a clinician’s judgment, called for a medication adjustment, they were counted as having suffered a recurrence, even if they were not depressed. In office practice, as a patient falters, her doctor can respond promptly, say, by raising the drug dose to forestall worsening or by briefly adding a second drug. If one antidepressant in a fixed dose is two or three times as effective as no treatment, ordinary medical care should be more protective yet.
But even the numbers that researchers present are impressive. Think about what it means for ongoing antidepressant treatment to spare patients half to two-thirds of the episodes they would otherwise succumb to—and for any length of time that has been studied. As preventatives, these medicines have remarkable range. They protect patients, like those in the French study, with highly recurrent depression. And antidepressants prevent the onset of mood disorder in people, such as stroke victims, who have never had it. The most and the least chronic cases benefit.
In the maintenance literature, the notion of medicine as glorified placebo loses meaning. There’s nothing to glorify. Placebos don’t prevent depression, and antidepressants do.