Ordinarily Well: The Case for Antidepressants


Spotting Trout

WE COME NOW to the opening salvo of the current antidepressant controversy. In June of 1998, Irving Kirsch, a psychologist then at the University of Connecticut, and his former graduate student Guy Sapirstein published a paper that seemed quixotic. It was meant to cast doubt on the usefulness of all antidepressants, including tricyclics. Because of the paper’s title, “Listening to Prozac but Hearing Placebo: A Meta-analysis of Antidepressant Medication,” word of its publication reached me immediately.

I admired the essay for its chutzpah. To my reading, “Hearing Placebo” was to the 1990s what Doris Mayer’s “‘Anti-Depressed’ Personality” had been to the 1970s, a cri de coeur against the encroachment of medication into the domain of psychotherapy. The lead author, Kirsch, a pioneering theorist in the science of placebo effects, seemed to be out to twit authority. His paper was larded with calculations bound to leave pharmacologists sputtering.

“Hearing Placebo” was presented as unconventional. It appeared in an online journal before those were mainstream, and the editors preceded the text with an introduction that cited the study’s “clearly arguable statistical methods.” Their preface began:

The article that follows is a controversial one. It reaches a controversial conclusion—that much of the therapeutic benefit of antidepressant medications actually derives from placebo responding. The article reaches this conclusion by utilizing a controversial statistical approach—meta-analysis. And it employs meta-analysis controversially—by meta-analyzing studies that are very heterogeneous in subject selection criteria, treatments employed, and statistical methods used.

As remedy for the research’s shortcomings, the editors took another extraordinary step. Simultaneously, they published expert commentary, including dismissive responses.

Two critics, Donald Klein, the Columbia University researcher, and the late Robyn Dawes, a psychologist expert in statistics, offered thorough refutations. But the essay would not go away.

As its subtitle indicates, the paper is built around a meta-analysis. Searching the literature for randomized trials of medication treatment for depression, Kirsch had identified nineteen whose data he considered adequate. The trials involved sixteen drugs, not all of them antidepressants.

Kirsch found effect sizes on the order of 0.4, with SSRIs coming in a bit higher—scores that Kirsch conceded were “remarkably similar” to the 0.5 effect size appearing in other overviews. Klein’s essay argued that Kirsch’s statistical approach was disadvantaging to medication—and any correction would have pushed the results far into the acceptable range. Nothing in the numbers was shocking.

Klein also found Kirsch’s choice of material for the meta-analysis prejudicial. Certainly, it was selective. A recent Cochrane report on Elavil located thirty-seven studies published before 1998 that had efficacy data suitable for analysis. Looking at not one drug but sixteen, Kirsch found only nineteen trials with data he could work with.

Kirsch’s process seemed to cull studies in which antidepressants underperformed. In a set of Wellbutrin trials, he embraced one with weak results and dismissed similarly designed studies, reported in the same paper, that showed higher drug efficacy. Kirsch included the NIMH collaborative trial (the TDCRP), which we know was not designed to show imipramine to best advantage; incorporating that study, Kirsch used average Hamilton scores and completer analyses. He was not telling a tale in which medication would play the hero’s role.

When a critic assembles an unfavorable collection of trials and finds that SSRIs have medium effect sizes, shouldn’t the psychopharmacology community celebrate?

But Kirsch suggested other ways to view the results. He said that three-fourths of what antidepressants did could be accomplished by dummy pills. Then he questioned that last bit of change, the 25 percent. His idea was that standard placebo pills did not generate enough hopeful expectancy since, unlike antidepressants, they did not cause side effects. We are back to the “active placebo” argument, the one that Cole and Klerman had left hanging.

Kirsch believed that his meta-analysis contained pertinent evidence. We will recall that he had analyzed sixteen drugs used in depression-outcome research. Four of those were not classed as antidepressants, and they performed as well as the twelve that were.

Simplified, Kirsch’s argument ran: Since the “non-antidepressant drugs” ought not to work, we can think of them as active placebos, dummy pills with side effects that signal to patients, “You’re on medication.” But in the trials analyzed, the non-antidepressant drugs worked as well as antidepressants. If ineffective pills with side effects work as well as the “real” drugs, perhaps all the apparent efficacy of antidepressants, including the tenacious final quarter, arises from imagining.

I have said that in reading a meta-analysis, often it is only when we look at the studies it gathers, the individual research papers, that we understand what is at issue. Let’s examine the reports on what Kirsch called “non-antidepressant drugs.”

The four were a thyroid hormone (liothyronine); a Xanax-like medicine (adinazolam); a barbiturate (amylobarbitone); and lithium. Kirsch considered these medicines active placebos—drugs with no inherent efficacy for depression, ones that treated mood disorders only through causing stray bodily effects and so arousing intense expectations of cure.

The most outrageous element in the mix was the study of thyroid hormone and lithium. Russell Joffe, a Toronto-based researcher, had been interested in what to do next when antidepressants fail. In practice, often psychiatrists would continue the treatment but add a small daily supplement of lithium or thyroid hormone, not because low-dose lithium or thyroid is effective on its own, but because either can help the antidepressant to work. This approach is called augmentation.

To test augmentation, Joffe began with fifty patients who had been prescribed full doses of a tricyclic—imipramine or a related drug, desipramine—and had not responded. For some (in a two-week trial, conducted in the early 1990s) he merely continued the antidepressant along with a second tablet, a dummy pill. The rest also stayed on the antidepressant but with low-dose lithium or thyroid added as an augmenter.

Only a few patients in the antidepressant-continuation (imipramine-plus-placebo) arm improved.

The antidepressant-plus-lithium group and the antidepressant-plus-thyroid group did better, with triple the response rate.

Augmentation looked like a promising strategy.

Good news, then: Many patients improve on antidepressants. Of those who don’t, most will respond when the antidepressant is supplemented by a modest dose of thyroid or lithium. So, yes, low-dose thyroid and lithium combat depression if you give them along with a full dose of an antidepressant.

And thyroid and lithium are psychoactive substances. Joffe’s study hardly showed that just anything works for depression. Instead, it confirmed what practicing psychiatrists believed, that certain substances with marked influences in the brain can help catalyze the action of antidepressants.

Imagine that you’re testing a new formulation of gasoline. It does well in some cars but poorly in others. Thinking that the balky cars may have water in the gas line, you pour in an additive—a “water remover”—and now the engines run fine. A friend says to you, Looks like that additive is a terrific fuel! No. Gasoline is a terrific fuel. Sometimes it needs an augmenter.

Kirsch’s inclusion of the Joffe study brought out a glimmer of humor in Donald Klein. In his refutation, “Listening to Meta-analysis but Hearing Bias,” Klein said that coming across the Joffe data in the Kirsch essay was “like finding a trout in the milk.”

The reference is to a diary entry by Henry David Thoreau in 1850. During a dairy strike the prior year, deliverymen had been accused of diluting milk with stream water. Countering the notion that circumstantial evidence cannot be conclusive, Thoreau wrote, “Some circumstantial evidence is very strong, as when you find a trout in the milk.” “Hearing Placebo,” Klein was implying, contained internal evidence that Kirsch had not set out to conduct a dispassionate inquiry into the virtues of antidepressants.

The two other “non-antidepressant drugs” hardly qualify as active placebos either. Xanax had been commercially successful as an antianxiety drug. During its development, some scientists had hoped that because of its structure, it might show antidepressant effects as well. The pharmaceutical house Upjohn set out to develop a Xanax variant—adinazolam—that would have stronger antidepressant properties, through acting on brain pathways that involve serotonin.

In trials, adinazolam did sometimes relieve depression in the early going, but its antidepressant powers were unimpressive, and it showed hints of a problem called rebound—a sharp uptick in symptoms after initial improvement—so it never came to market. Still, the drug had been designed to treat depression through conventional means. By citing a trial where, for some weeks, adinazolam succeeded, Kirsch was hardly demonstrating that any random pill could duplicate the effects of antidepressants.

Kirsch’s “non-antidepressant drug” category had one final candidate, amylobarbitone, a barbiturate. Medications of this class—Seconal was a leading brand name—were used to treat anxiety and insomnia.

The study caught in Kirsch’s sieve was designed to address a practical problem. Wary of drug side effects, primary-care doctors tended not to prescribe the older antidepressants at full doses. Might low doses work? Since generalists were comfortable prescribing barbiturates, might they be substituted? In the trial, patients were given any of four pills: placebo, the barbiturate, 75 milligrams of Elavil (a low dose), or 150 milligrams of Elavil (a standard dose, also on the low side).

By the study’s end, four weeks out, the more realistic dose of Elavil had significantly outperformed placebo on three scales, including the Hamilton. The barbiturate and the lower dose of Elavil were indistinguishable from placebo. On two scales, the full dose of Elavil outperformed the barbiturate.

Not much to go on. The authors’ main conclusion, as regards efficacy, was “After 28 days, [Elavil] 150 mg/day was significantly better than the other treatments.” The authors suggested that any barbiturate benefit may have been due to scale items rating sedation. If primary-care doctors wanted to treat depression, they would need to leave their comfort zone and prescribe a full dose of antidepressant.

In his analysis, Kirsch merged the results from the two doses of Elavil, a choice that made the antidepressant look less distinctive. But to be clear, the study authors found that barbiturate did not treat depression as well as an adequate, if modest, dose of Elavil.

After scouring the literature, Kirsch had managed to present—this was my impression—not a single example of what he had expected to find, a medication that ought not to have possessed antidepressant effects but that (through the impact of pill taking enhanced by drug side effects) alleviated depression anyway. This part of Kirsch’s paper stood as testimony against the active-placebo hypothesis.

I don’t mean to claim that antidepressants are the only medicines that can help depressed people. As Jonathan Cole taught, often patients will flourish on the “wrong drug.” I do mean that, in my view, Kirsch had failed to demonstrate that the final chunk of benefit, the advantage of antidepressants over placebo, was due to anything other than the drugs’ action on relevant pathways in the brain, their “pharmacologic potency.”

Even after Donald Klein pointed out problems along these lines, Kirsch continued to claim that his 1998 paper demonstrated that non-antidepressant drugs work as well as antidepressants. In 2000, he wrote, “Some active drugs that are not considered antidepressants (amylobarbitone, lithium, liothyronine, and adinazolam) show the same effect on depression as that shown by antidepressants.” He made similar claims in papers published in 2005 and 2009, still referencing “Hearing Placebo.”

In 2012, when interviewed by Lesley Stahl for 60 Minutes, Kirsch said, “We even looked at drugs that are not considered antidepressants: tranquilizers, barbiturates. And do you know what? They had the same effect as the antidepressants.” Was he still referring to adinazolam and amylobarbitone, based on the same two trials?

“Hearing Placebo” had been intended as a debunking report, and yet it had found adequate effect sizes and (despite assiduous efforts) no grounds for diminishing them. I imagined that the paper would carry little weight in evidence-based medicine. I was wrong, and wrong more generally about the historical role that “Hearing Placebo” would play. I had seen it as a rearguard action. It was in the vanguard, the start of a resurgence of doubt about antidepressants.

If you find an error or have any questions, please email us at admin@doctorlib.info. Thank you!