Ordinarily Well: The Case for Antidepressants


Hypothetical Counterfactual

WITH HIS TROUT-IN-THE-MILK remark, Donald Klein questioned the evenhandedness of Irving Kirsch’s analysis. Robyn Dawes raised a more fundamental concern. He objected to Kirsch’s account of what placebo is and does.

In essence, Dawes’s complaint was that Kirsch saw placebo as something solid—a treatment. When doctors prescribe an antidepressant, it helps people directly, through its pharmacologic potency. Kirsch thought of placebo the same way: when researchers offer a dummy pill, patients improve because of it.

The back-and-forth between Kirsch and Dawes centered on technical questions about effect sizes. That discussion is beyond our pay grade, but we can approach it by addressing a topic we need to think about anyway: What else, beyond the inherent efficacy of medication, helps people improve (or seem to) in drug trials?

Kirsch considered the placebo arm to encompass two healing elements, the passage of time, which he called “natural history,” and a more powerful factor, the core placebo effect. Although Kirsch did not define this core, the implication was that it had largely to do with placebo as we ordinarily imagine it, a pill that induces hopeful expectancy.

To Dawes, in contrast, the placebo arm had a single function: to capture the “hypothetical counterfactual,” what would have happened to patients in a trial’s active treatment arm had they not gotten the effective ingredient in the pill. In this view, controlled trials can tell us only about the treatment under test—here, antidepressants. The studies tell us nothing about the comparison intervention, placebos. (The parallel is not exact, but, when you time yourself as you jog, the result tells you nothing about the watch.) Placebo tracks what happens without antidepressants, progress that will differ from trial to trial.

Dawes did not break them out, but all sorts of influences play a role. Springtime arrives and daylight hours increase. The economy turns a corner. To the extent that depression is responsive to these external factors, a study conducted in favorable conditions will show better-than-usual outcomes in the placebo arm. Because of circumstance, the influence of “natural history”—time—will vary from study to study.

The design and implementation of the research will matter, too. Think of a trial where patients visit the test site frequently and meet at length with supportive, directive clinicians. We will expect enhanced placebo responses, due to the more intensive “minimal supportive psychotherapy.”

Other factors come into play. Patients may distort their symptom reports in hopes of pleasing the caregiver. The term of art for this tendency is demand characteristics. Often, patients provide what they think doctors demand. The propensity for looking sick early in a medical encounter, in hopes of engaging the caregiver, is the hello effect. Toward the end of treatment, courteous participants may tell a rater that they are better when they are not—the goodbye effect. The personalities of staff and patients (and, again, the intensity of contact) will determine the strength of demand characteristics.

We have already encountered baseline score inflation. It is not only would-be participants who exaggerate. Raters under time pressure to recruit participants may puff up patients’ symptom counts at the start of a trial so that they appear to satisfy the minimum requirements for entry. Subsequently, scores will fall, whether or not the patients get better.

By this account, what Kirsch calls the placebo effect (all influences other than natural history) contains diverse elements, such as demand characteristics and baseline score inflation. And then, yes, there is the classic placebo response, hopeful expectancy. To the extent that it alters the course of depressive episodes, it, too, will differ from study to study, depending on the shape and color of the pill and the stiffness of the starch in doctors’ white coats.

To express the jumble of influences in the control arm, Gerald Klerman wrote of a “package of placebo effects.” I will go further and call the package a grab bag. In different trials, the contents of the sack—their amounts and proportions—will differ.

Because the control arm packages distractions that we want to see beyond, because it absorbs happenstance influences, placebo administration will produce levels of improvement that vary from trial to trial. If not, we would have no need to run placebo arms. We could compare medication results to a universal number, the placebo effect. Dawes argued that Kirsch was making more or less that mistake, assuming that placebo, like imipramine, had a discoverable “efficacy.”

Dawes thought that it rarely made sense to refer to a proportion of treatment effect as being accounted for by “placebo.” In early drug trials, when untreated patients were plentiful and the NIMH had yet to demand that study participants be offered encouragement, placebo arm results ran low—showing, say, half of the change seen in imipramine arms. Later, with score inflation and supportive psychotherapy, the proportion rose to three-quarters. But imipramine was still the same chemical, with the same inherent efficacy.

When Kirsch compared drugs to placebo, often he expressed outcomes as percentages: “75% of the response to the medications examined in these studies was a placebo response, and at most, 25% might be a true drug effect.” Marcia Angell would echo this claim, writing that Kirsch’s early research had demonstrated that “as judged by scales used to measure depression, placebos were 75 percent as effective as antidepressants.”

But this formulation may tempt us into all sorts of error. For instance, we might be tempted to say that dummy pills do a great deal—that they do three-fourths of what antidepressants do. But when it comes to increased sunshine, rising salaries, demand characteristics, and score inflation, the placebo pill contributes nothing. The influence of classic placebo effects—hopeful expectancy—may be minimal.

In contrast, Gene Glass’s measure, the effect size, gets these matters right, using placebo only for purposes of comparison. When, as in the trials Kirsch assembled, we discover an effect size of 0.4 or 0.5, we know that antidepressants will move a typical patient to the lesser level of depression of someone at the 65 to 70 percent mark in the control group—never mind whether days are long or short and skies blue or gray.

We may recall that in calculating effect sizes, a key step is the one we encountered when we first discussed randomized trials, subtracting the progress patients make in the control arm from the greater progress their counterparts make in the treatment arm. That operation—subtraction—treats placebo not as a valued remedy but as a collection of nuisance variables to be tossed out.

The important point is conceptual: The placebo is a pale doppelgänger, taking the form of the active intervention but lacking the enlivening element. Its insubstantiality—its ability to shape-shift and capture circumstance—is the placebo’s glory. But we must not become too enamored of placebo. It is there to be thrown away.

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