Do antidepressants work?

There is an active debate underway in the popular literature about whether antidepressant medications actually do anything chemically helpful for depressed patients.  No one doubts that many patients report feeling better, and that most evidence less depression on standardized rating scales, following treatment.  But much of that improvement appears to be due to psychological factors, i.e., the placebo effect.  The debate is over how much improvement is not due to the placebo effect.  What beneficial effects can be attributed to the active ingredients in the tablet or capsule? It's disconcerting to enter this debate decades after the popularization of antidepressants.  These are among the most common prescriptions in America: In 2010, antidepressants were the second most commonly prescribed class of drugs in the U.S., according to IMS Health.  They are so widely used that Consumer Reports publishes "best buy" recommendations about which ones to try first.  Yet recent reanalyses of efficacy data have called into question whether antidepressants help more than inert pills.  In a two-part piece in the New York Review of Books, Marcia Angell MD, the former editor-in-chief of the New England Journal of Medicine, favorably reviews these skeptical findings.  (I won't summarize the arguments here, but I do very much recommend her review.)  In the other corner is Peter Kramer MD, author of Listening to Prozac and other books, who offers a spirited defense of antidepressants in his op-ed rebuttal in the New York Times.  The 300 comments that follow the online version of the op-ed also make for fascinating reading: Many are first-person accounts of the lifesaving benefit of antidepressants.

What to make of all this?  Those conversant in research methodology will pick apart the various arguments.  Do the studies have enough statistical "power"?  Does it matter that typical efficacy studies recruit subjects who differ from patients in clinical practice?  How much difference does an "active" placebo make?  Is it preferable to use subjective mood ratings, or ratings from trained observers?  How many weeks or months should subjects be assessed?  Should subjects with co-morbidities, i.e., additional diagnoses, be included or excluded?  Are there advantages to including a third study arm (a known effective intervention) to the usual two (the drug being assessed, and placebo)?

There are many such questions that need to be resolved, and professional researchers are probably in the best position to discuss them.  Meanwhile, the rest of us are left with a seeming paradox.  Thousands — millions? — of individuals claim relief from antidepressant treatment, and virtually any psychiatrist will swear that antidepressants really have helped many of his or her depressed patients.  (This is my own experience, by the way — it's nearly inconceivable to me that antidepressants are no more than placebos.  I've seen too many patients improve before my very eyes.)  Meanwhile, there are also many patients, equally depressed, who obtain little or no benefit from antidepressants, and a large number of carefully conducted studies that find little benefit in the active ingredients of these pills, once placebo effects are factored out.

While I can't prove it, my sense is that the answer lies in the heterogeneity of depression.  Some patients get dramatically better on antidepressants (in entirely believable ways, as opposed to reactive "flight into health" and the like), some only a little, and others appear not to change at all.  Widely varying responses can easily "average out" in the usual randomized controlled trials used to assess efficacy, and could account for lackluster findings in group studies.  Since I do have some research background and training myself, I'd want to see the scatterplots of individual subject ratings, to see if they cluster into responsive, partly responsive, and unresponsive groups.

Of course, it is not a new idea that some depression responds to medication and some doesn't.  When I started medical school, psychiatrists distinguished "endogenous" and "exogenous" depression — i.e., depression that originated within the patient chemically, and depression that originated from external stress or loss.  (For a concise summary of the idea, see the first paragraph of this editorial.)  Antidepressants were thought to help the former but not the latter.

Unfortunately, that wasn't true.  As it turns out, knowing whether an external event precedes a depression doesn't predict whether an antidepressant will help.  The search has gotten more sophisticated lately, and measurable genetic subtypes may one day tell us who will benefit by antidepressants and who won't.  But we're not there yet.   At this point, we cannot predict whether an individual patient will improve with antidepressant medication.

I'll end this post by noting that the placebo effect, a vexing complication in clinical research, isn't a bad thing in real life.  If a patient feels better, I don't worry too much about who or what gets the credit.  Maybe it's the citalopram or sertraline in the pill.  Maybe it's the patient's belief in the pill and in the medical science behind it.  Maybe it's the fact that I gave the patient something that our culture imbues with symbolic healing powers.  Maybe my words were healing and the prescription was a mere distraction.  Or maybe I had no effect at all, and the patient healed himself or herself.  Usually it's impossible to know.  In my view, being a psychiatrist in clinical practice requires this kind of agnosticism and humility.

Placebos (2)

In my last post, I wrote about the use of placebos in clinical practice -- or more accurately, about giving medical treatments based on psychological comfort, not physiological effect.  However, the area where placebos are most used and accepted is human research, not clinical practice.  Psychiatric research in particular introduces some interesting conceptual issues regarding the use of placebos. To decide whether any new medication actually helps patients, the "randomized, double-blind, placebo controlled study" is the gold standard.  In this type of experiment, research subjects who all have the same disease (or are all equally healthy) are randomly assigned to take either the new medication or a placebo: an identical twin of the medication lacking any of its active ingredients.  The subjects do not know whether they are in the medication arm or the placebo arm of the study -- they are "blind" to this fact.  Their doctor, or whomever rates their outcomes in the experiment, does not know either; this makes the experiment "double-blind."  If the only difference between the two arms of the study is the presence of the active ingredients, then any outcome differences, on average, between the groups can be attributed to those ingredients.

The reason such experiments are double-blind, and the reason for the placebo in the first place, is to separate the medication's physiological effects from any rating bias or placebo effects.  Outcome ratings can be distorted by raters' expectations; people often see what they want to see.  Placebo effects are psychological factors in the subjects that improve rated outcome.  These include wanting to please the experimenter by getting better, hope that the medication will work, improvement due to feeling attended to and cared for, etc.  A "controlled" experiment carefully dissects away rating biases and placebo effects, so that only physiological effects count.

There is great value in knowing that an antibiotic and a sugar pill lead to different medical outcomes.  Patients can feel attended to and cared for -- they can feel "better" -- and still die of infection.  Disease has a life of its own, apart from subjective experience.  Western medicine prides itself on its scientific foundation, and rightly so.  We don't want merely to feel better, we want to be better.

However, the situation is less clear with many psychiatric disorders.  Depression and anxiety, for example, do not appear to have lives of their own, apart from the patient's subjective experience.  If the patient feels better, he or she is better.  Thus, the rationale for placebo controlled studies in this area bears further reflection.  Why does it matter if mood or anxiety improvement from a medication is physiological, or "merely" a placebo effect?

One answer is that we hope to further our scientific/medical understanding of such disorders, and placebo controlled studies help us do that.  Another is that drug makers and the FDA justify the value of pharmaceuticals by virtue of their active ingredients, not the psychological features surrounding their use.  Yet another is that placebo effects are idiosyncratic and vary across the population, and we strive for more predictability.  Nevertheless, as discussed in my previous post, for a given patient it ultimately doesn't matter why a treatment works, as long as it really does.  (For a more technical discussion, see this British Journal of Psychiatry editorial.)

Another fascinating conceptual problem is whether and how to include placebo controls in psychotherapy research.  Psychotherapy, like depression and anxiety medication, treats distress that is inseparable from subjective experience.  Thus the importance of differentiating "active ingredient" and "placebo" effects is unclear at best.  In addition, the treatment itself consists of the very psychological influences a placebo controlled study aims to remove.

By analogy with medication studies, psychotherapy researchers have tried to isolate the "active ingredients" in therapy, and to fashion studies with a treatment arm and a placebo arm differing only by the presence of those ingredients.  So far, no active ingredient has been identified as essential (although the overall efficacy of psychotherapy is not seriously in question, see here and here).  What if there are none -- what if psychotherapy is all "placebo effect?"

This sounds like psychotherapy is quackery, much like the headline "Half of Doctors Routinely Prescribe Placebos" sounds like quackery exposed (see my previous post).  But if psychotherapy is compared not to medication but to other human relationships, the "active ingredient" idea seems out of place.  The helpfulness and support of a parent, teacher, spouse, sports coach, or minister cannot be reduced to any specific ingredient.  Growth or self-improvement gained from such relationships comes organically and emergently, not as a result of a discrete intervention.  This does not make the value of such relationships, or the benefits gained, any less real.

Placebos are essential for careful medical research, yet carry overtones of fakery and worthlessness.  When the placebo concept is applied in an overly broad fashion, its negative connotation can tarnish highly beneficial human interactions that are neither fake nor worthless.

Placebos (1)

The headline of a recent New York Times article was cause for public alarm: "Half of Doctors Routinely Prescribe Placebos."  A casual glance might have given the impression that doctors dispense sugar pills half the time, but this would be a misreading of the reported finding.  The article followed a research report that appeared in the October 23rd issue of BMJ (formerly the British Medical Journal).  A national survey of 679 U.S. internists and rheumatologists found that half routinely prescribe medication when, in the opinion of the clinician, the benefits "derive from positive patient expectations and not from the physiological mechanism of the treatment itself." This definition goes well beyond our usual notion of a placebo as a fake medication, e.g., a sugar pill or an injection of sterile salt water.  In fact, only two or three percent of the doctors reportedly used these.  Much more commonly the "placebo" consisted of over-the-counter analgesics (painkillers), vitamins, and sometimes antibiotics and sedatives.  Leaving the loaded word "placebo" out of it, half the doctors admitted to offering some treatments purely because they are psychologically comforting to their patients.

This nicely illustrates something psychiatrists, and many alternative/complementary healers, have known for a long time: There is no sharp distinction between mind and body.  Each affects the other.  Feeling ill and seeking help are not mechanical processes, like the engine light coming on in one's car.  A doctor's attention and concern can help a patient feel better even if no treatment is given.  Often medications work for no good reason, or fail to work when they "should."  The doctor-patient relationship is a complex dance of expectations and hopes, full of subtle cues and interactions at conscious and unconscious levels.  Western medical science shines a light so bright it can blind us as well as illuminate.  Of course doctors sometimes offer remedies that are primarily psychological.  How could anyone have thought otherwise?

While I have never given anyone a sugar pill, I have often been faced with choosing among several equally acceptable antidepressants for a patient.  The published efficacies, side-effect profiles, and even costs of the medications are essentially the same.  Just as I am about to throw a dart into my mental dartboard to pick one, the patient tells me of a friend who had great success with one of them.  Suddenly the choice is clear.  In such situations I (enthusiastically) prescribe that particular medication due to "positive patient expectations," and not due to "the physiological mechanism of the treatment itself."  To take another example, occasionally patients tell me that plain aspirin or Tylenol helps them sleep.  Who am I to tell them they must be mistaken?

The greatest cliche of medicine, "take two aspirin and call me in the morning," is a testament to this principle in internal medicine, rheumatology, and similar specialties.  If a newspaper headline in 2008 can shock and alarm us over a practice as old as this cliche, Western culture has strayed very far indeed from the essence of healing.