His answer is that the statistical edge, when it turns up, is a placebo effect. Drug trials are double-blind: neither the patients (paid volunteers) nor the doctors (also paid) are told which group is getting the drug and which is getting the placebo. But antidepressants have side effects, and sugar pills don’t. Commonly, side effects of antidepressants are tolerable things like nausea, restlessness, dry mouth, and so on. (Uncommonly, there is, for example, hepatitis; but patients who develop hepatitis don’t complete the trial.) This means that a patient who experiences minor side effects can conclude that he is taking the drug, and start to feel better, and a patient who doesn’t experience side effects can conclude that she’s taking the placebo, and feel worse. On Kirsch’s calculation, the placebo effect—you believe that you are taking a pill that will make you feel better; therefore, you feel better—wipes out the statistical difference.
Anecdotally, I suspect the very existence of a placebo effect means that this isn’t the case, or else people on the sugar pill in any trial where they knew one possible arm was placebo wouldn’t experience a benefit (real or not). For example, studies suggest that the brain responds to placebos through real chemical reactions, by releasing dopamine, so presumably the lack of side effects doesn’t clue-in the brain that it’s being tricked. Aside from studies showing real physiological reactions to placebo, I should note that I personally consider the placebo effect to be sometimes overstated, and benefits attributed to the placebo may simply be a part of the patient’s natural healing.
But assume that Kirsch is correct. Let’s imagine a world where patients can actually readily distinguish a placebo from the real pill based on lack of side effects. I’ll add here that side effects are probably very easily imagined and thus reported by placebo patients who are “expecting” them, but again let’s set that aside. As a result, patients are more likely to remove themselves from trials early (by withdrawing consent) because they think they’re being given the placebo, and of course they aren’t benefiting from any placebo effect. Doctors may also seek to take patients who are genuinely sick off of the placebo because they, too, observe that none of the expected side effects are occurring. This means that for the purposes of the drug trial, the placebo arm, the negative control, suffers from a chronic lack of information compared to the real-drug arm. It also means that the double-blind rigorousness of the study is compromised. We create double-blind studies to ensure that the participation and reporting from doctors and patients is not biased by any knowledge of placebo vs. real-drug.
Would pharmaceutical companies be allowed, then, to substitute placebos for mostly harmless but temporarily side-effect-inducing drugs? Instead of the sugar pill, could a pharmaceutical company dispense in the placebo arm some mild headache pills, or dizziness capsules, or dry mouth gels?
I think according to current formulations of research ethics, we would still allow those “negative” placebos–drugs with only the potential to do harm, and without (aside from the placebo effect) any scientific mechanism of benefiting the patient. In phase I clinical trials, which are conducted to determine the tolerable dosage range of a drug by slowly escalating the drug concentration, some clinical trials will enroll completely healthy patients. These healthy patients stand little, if anything, to gain, and stand some risk of experiencing side effects. Additionally, even sick patients who are enrolled at the very beginning of the study are receiving dosages at such small fractions of the anticipated market dosage level (as estimated by animal research) that they are highly unlikely to be helped, though I suppose that’s still a non-zero probability, yet still likely to experience negative side-effects. For those patients, an additional negative “side-effect” may simply be the opportunity cost of enrolling in one clinical trial of negligible probability of benefit while there are others out there, which are at least enrolling patients at higher experimental dosages that may give some chance of success. This opportunity cost negative “side-effect” is present whenever placebos are present in a study (usually they aren’t in studies with life-threatening or terminal diseases, though sometimes are used in third-line lung cancer trials or in combination vs. single agent trials, but nonetheless patients still suffer in the interim). We seem willing to sacrifice, or let patients sacrifice, some utility for the betterment of research overall.
If some patients are, as Kirsch suggests, figuring out that they are on the placebo pill, we should reformulate the placebo pill to make it more difficult to deduce. These new negative placebos would have temporary, minor side-effects that mimic the side-effects of the real drug so that neither doctor, nor patient, will be able to figure out which arm of a double-blind study the subject is enrolled under. Perhaps this would tighten the quality of double-blind research studies. Perhaps it would even strengthen the placebo effect patients sometimes experience. At the very least, it should silence critics like Kirsch et al.