MIGHT PLACEBO PILLS prove equally miraculous? In outcome trials, in ordinary experience, antidepressants work. If we entertain doubts about them, it’s because of the claim that inert tablets would do as well. The consequences of that possibility go beyond the thought that patients might be spared cost and harm. If sham treatment cures depression, it is an especially tractable disorder. If apparent antidepressant responses result from classic placebo effects, then doctors and patients are easily misled.
Think of how a primary-care doctor, call her Viola, gains familiarity with antidepressants. For depressed patients, she may routinely recommend any number of remedies: exercise, bright lights, marriage counseling, and “tincture of time.” When those fail or work halfway, she will prescribe. Then, based on her experience, she will make an informal calculation, about how much the medicine (as opposed to more time, more exercise, and the rest—factors whose influence she can subtract out) aids in her patients’ recoveries.
What Viola can never see is the change due to the mere facts of prescribing and pill taking. They are inherent in every instance of medicating depression.
True, Viola may have inklings. Working with patients like Adele, the schoolteacher who got more, not less, despondent in the face of dramatic treatments for her thyroid disease, I came to doubt that classic placebo effects are powerful in depression. But in ordinary circumstances, a pill’s symbolic impact and its inherent efficacy can’t be pried apart by observation. That’s why research on placebo is critical. If any random pill would work for her depressed patients, Viola’s yardstick is off. A great deal rides on what we make of sham treatments.
Well, then, what does evidence-based medicine say? For researchers, the classic placebo effect—response based on expectations attached to a pill—is a unicorn: often described, rarely seen.
Placebo, as we ordinarily imagine it, is a bargain and a miracle. A person in pain is offered a capsule full of nothing, and the suffering ends. Or begins. Placebos can elicit nausea, somnolence, even skin rashes. In behavioral research, explorations of how expectancy causes an acute symptom constitute a cottage industry. But beyond parlor tricks, beyond tonic-water tipsiness, what is the classic placebo’s potential?
Slight, it seems, when it comes to what doctors treat.
In 2001, Asbjørn Hróbjartsson and Peter Gøtzsche, Danish medical researchers with ties to the Cochrane Collaboration, looked at trials in which placebo pills or other sham interventions, such as sham surgery or sham acupuncture, had been contrasted with “no treatment” conditions. The analysis reviewed studies of forty disorders, including anemia, bacterial infection, epilepsy, and carpal tunnel syndrome. The intent was to isolate this factor, pretend treatment, and see whether it influenced medical conditions.
Hróbjartsson (think “Robertson”—the name is Icelandic) and Gøtzsche were seeking what Irving Kirsch wanted—and what Robyn Dawes said could not be found in conventional drug studies—a measure of the placebo pill’s effect, but from trials in which a sham intervention was the active treatment, and something less hope-inducing, such as time on a wait list, was the control condition.
Hróbjartsson and Gøtzsche found that, for the most part, sham interventions conferred no extra benefit—none. Where placebo did outperform the wait list, the difference was small. The resulting article, in The New England Journal of Medicine, was titled “Is the Placebo Powerless?” The authors answered, “In conclusion, we found little evidence that placebos in general have powerful clinical effects.”
Hróbjartsson and Gøtzsche were not denying the need for control arms in outcome trials. Their point was that in the grab bag of placebo effects, this one, expectancy tied to a pill or a procedure, played a trivial role. The passage of time might matter, and patients’ eagerness to please. But in almost no instance did giving or withholding pills make a difference.
Hróbjartsson and Gøtzsche grouped studies various ways.
They looked at trials that used categorical measures, such as response and remission. Hróbjartsson and Gøtzsche called them binary outcomes. The patient is dead or alive, smoking cigarettes or abstaining, depressed or no longer depressed. For binary outcomes, it did not matter how you measured your ailment, whether through objective measures (death or blood sugar levels) or subjective measures (dizziness or Hamilton ratings). No classic placebo effect emerged. The sole exception was in the treatment of pain, where the brain’s own opiates seem to come into play.
Pain studies aside, for binary—well or ill—outcomes, if patients on dummy pills got better, so did patients on the wait list. Hróbjartsson and Gøtzsche found no trials—none—in which a pill reliably produced more remissions or recoveries than “no treatment.”
Hróbjartsson and Gøtzsche looked also at studies with continuous objective outcomes, such as blood pressure, and found no classic placebo responses. Hróbjartsson and Gøtzsche did see modest effects in trials using “continuous subjective outcomes,” such as a change in Hamilton scores. The movement was on the order of that seen with psychotherapy or antidepressants in the less favorable analyses—effect sizes near 0.3. But even here, the researchers’ likeliest scenario was that classic placebo effects made no contribution at all.
With continuous subjective outcomes, the only differences between placebo and no-treatment arms appeared in studies in which patients described their symptoms: seasickness or anxiety. Since no placebo effects appeared elsewhere—none where physicians observe directly, as with joint flexibility—the problem, Hróbjartsson and Gøtzsche speculated, might be the goodbye effect. Patients know when they are on a wait list. Wait-list patients don’t come in to discuss “medication” side effects with staff and bond over the encounter. They are not tempted to give upbeat reports—but patients on pills might be.
Hróbjartsson and Gøtzsche made an additional argument. In this last bastion of sham treatment, the effects were largest in small trials, a pattern that suggests publication bias. Hróbjartsson and Gøtzsche theorized that because placebo has its proponents, when research showed that it failed to outperform the wait list, those data remained in the file drawer.
The Hróbjartsson and Gøtzsche paper—it has been updated twice—has been enormously influential. Of it, Ted Kaptchuk, a respected placebo researcher, has said, “At first when I read it, I worried I’d be out of a job … But frankly, [Hróbjartsson] was absolutely right.” The study caused placebo advocates to moderate their claims. In a 2015 essay for The New England Journal of Medicine, Kaptchuk began the substance of his overview with this acknowledgment: “First, though placebos may provide relief, they rarely cure.”
The “Powerless Placebo” paper included a handful of depression studies. Most concerned maintenance. Maintenance research looks at people who recovered in the course of a brief trial and are now free of depression. The subsequent outcome is binary: In a follow-up interval, does the remission last, or does depression recur?
When patients who had improved in a treatment trial entered a maintenance study, it did not matter whether they continued on a placebo pill or received no pill. They relapsed at the same high rate.
One criticism of maintenance trials stems from the view that antidepressants are habituating and that coming off them causes harm—so, of course, neither placebo nor “no treatment” will work. But Hróbjartsson and Gøtzsche’s collection included an experiment involving patients who, in an antidepressant trial, had recovered on a sham pill. In the follow-up, the placebo responders were continued on placebo or switched to “no treatment.” The patients had not been on an antidepressant and so had nothing to withdraw from. If being on a dummy pill had helped them, then staying on it should have kept them stable. It did not. “No treatment” was as good as dummy pills, and both were terrible. Over half of the patients in each group relapsed within six weeks.
What of the trials we have discussed most, drug versus placebo judged via average Hamilton scores? Hróbjartsson and Gøtzsche found only one with a “no treatment” condition.
In the early 1970s, a team of Indian scientists headed by a pioneering public health researcher, Dhirendra Nath Nandi, conducted a door-to-door survey in a rural village in West Bengal. They identified forty-one men and women with depression. None had ever been treated. These Bengalis earned average Hamilton scores just below 30—severe or very severe depression.
Nandi wanted to know whether tribal villagers would respond to antidepressants as urban clinic patients did. Enrolling every depressed person, all forty-one, he put half on a low dose of imipramine (100 milligrams) and a quarter on sugar pills. The remaining depressed patients were followed without treatment.
At two and four weeks the “natural course” group, on no pills, remained virtually unchanged.
In contrast, imipramine was effective. In the group on medicine, average Hamilton scores fell to 19 at week two and 13 (mild depression) at week four. Likely, the average patient on imipramine became a responder, losing half of his or her symptoms even on the low dose of medicine.
For those given sugar pills, the depression ratings dropped to 21 at week two but bounced back up to 27—severe or very severe depression—at week four. In Nandi’s words, the patients “went back to more or less the depth of depression found before the administration of placebo.” None enjoyed a response or remission.
The Nandi team had told the villagers that they would be taking “well-known” drugs. The procedure was designed to maximize classic placebo effects, yet by the four-week mark, the sugar pills were powerless.
Nandi concluded that rural and urban depressives respond to treatment identically. He made note of the course of response to the sugar pill: quick improvement that faded. If you ask doctors what placebo effects look like, most will describe that pattern, transient early improvement.
Frederic Quitkin, long a member of Donald Klein’s team at Columbia, explored this phenomenon. He reported that about a quarter of depressed patients on placebo will experience rapid symptom loss but soon find themselves almost as depressed as they were at the start. The placebo response is quick and transient.
For patients on medication, too, marked early relief may vanish. But patients with the pattern we associate with medication “kicking in”—sharp change from week three on—do fine. One pattern is particular to medication and virtually unseen in the placebo group: improvement at three weeks followed by continued well-being at every subsequent point of observation. Antidepressants are stabilizing.
In my practice, occasional patients will respond quickly to medicine by turning bright and brittle, “putting a good face on it.” This posture may be sustainable, if effortful. In a research center, a rater might score the presentation favorably. I see it as an aspect of the ongoing depression—and all the more if a patient cannot build on the response by making fresh efforts in love and work. It’s hard to make hay in this cold sun. The antidepressant is failing. We will look elsewhere for help.
Lately, Peter Gøtzsche has become a forceful critic of medication use in the treatment of mental illness. For example, in a BMJ essay in 2015, he wrote, “I estimate we could stop almost all psychotropic drugs without causing harm.” But regarding antidepressant trials in particular, he has held fast to the conclusion that pill-placebo effects are not in evidence. Writing about John Davis and Robert Gibbons’s analysis of response to Prozac and Effexor, Gøtzsche notes, “What we see in a placebo group is not a placebo effect but mainly the spontaneous remission of the disease.”
In the past decade, experiments seeking pill-placebo effects in depression have been few and flawed. One interesting small trial, out of UCLA, did find expectancy effects in patients on dummy pills—but not in patients on antidepressants. It was not just a matter of failing to reach statistical significance. In the medication arm of the trial, patients who said that they believed the treatment would not help enjoyed the same results as patients who anticipated great things.
If the analysis is on target, it may be that the more reliably medicines work, the more they arouse expectations that then inflate placebo responses in drug trials; meanwhile, little or none of that benefit transfers to drugs, which do their work on their own. In that case, to the extent that they exist, classic placebo effects, just like psychotherapy effects, are not additive and will cause researchers to understate what medication has to offer. In contrast, doctors like Viola will see clearly. In their clinical work, when they prescribe, they need not worry about missing a potential pill-placebo effect. With antidepressants, there is none.
Similarly for patients: if they attribute their improvement to the medicine—its inherent properties—and the good relationship with their doctor, they will not be mistaken. Spontaneous improvement will play its part as well; from their past experience (before ever taking medicine) of the course of their depression, patients may be able to gauge how much. In its small way, the UCLA study answers Nora’s question, whether her recovery is due to her expectations. They make no contribution.
I don’t want to put strong emphasis on an incidental finding in a preliminary study. In the end, Hróbjartsson and Gøtzsche’s comprehensive survey is what best summarizes research on the classic placebo effect. In the mix of influences—hello and goodbye, time and circumstance, and, yes, uplift from the substantial support, emotional, social, moral, and economic, that clinical trials offer—hope attached to pill taking is at best a minor element.