(Please see Counterpoint article by Michael E. Thase, M.D.)
(This is a preview of the Rumble in Reno II scheduled for Oct. 18. The debate between Kirsch and Thase will focus on the issues highlighted in these point/counterpoint articles. Brochures for this conference are available online at <www.dce.unr.edu>-Ed.)
Antidepressants are widely believed to be exceptionally effective medications. The data, however, tell a different story. Kirsch et al. (2002a) analyzed the data sent to the U.S. Food and Drug Administration by the manufacturers of the six most widely prescribed antidepressants (fluoxetine [Prozac], paroxetine(Drug information on paroxetine) [Paxil], sertraline [Zoloft], venlafaxine [Effexor], nefazodone(Drug information on nefazodone) [Serzone] and citalopram(Drug information on citalopram) [Celexa]). Their research showed that although the response to antidepressants was substantial, the response to inert placebo was almost as great. The mean difference was about two points on the Hamilton Rating Scale for Depression (HAM-D). Although statistically significant, this difference is not clinically significant (Jacobson et al., 1999). More than half of the clinical trials sponsored by the pharmaceutical companies failed to find significant drug/placebo difference, and there were no advantages to higher doses of antidepressants. The small difference between antidepressant and placebo has been referred to as a "dirty little secret" by clinical trial researchers (Hollon et al., 2002), a secret that was believed by FDA officials to be "of no practical value to either the patient or prescriber" (Leber, 1998, as cited in Kirsch et al., 2002b).
Previous reports of vanishingly small drug/placebo differences (Kirsch and Sapirstein, 1998) were met with skepticism (e.g., Klein, 1998). In contrast, the basic findings from this new meta-analysis have been accepted as accurate (e.g., Thase, 2002). The dispute is no longer about the small size of the average drug/placebo difference, but rather about how to interpret this fact and what to do about it.
Various interpretive possibilities have been raised. One of the most popular theories is that there may be a subset of patients for whom at least some antidepressants are very effective, but that their relative lack of efficacy with other patients masks effect (e.g., Thase, 2002). Specifically, whereas mildly depressed patients respond to both drugs and placebos, more severely depressed patients respond only to active drugs. The FDA data contradict this hypothesis. Although severely depressed patients benefited more from medication than mildly depressed patients due to a phenomenon known as regression toward the mean, they also benefited more from placebo than their more mildly depressed counterparts.
Of course, one can never rule out the possibility of undetected moderator variables. But if there are hidden moderators, the overall mean difference between drug and placebo (two points on the HAM-D) constrains the conclusions that can be drawn from them. If the mean drug/placebo difference is greater than two points for a subset of medications or patients, then it must be less than two points for the others. For example, if the mean difference between drug and placebo is four points for half of the patients (which is still a rather small drug effect), then the mean effect of antidepressants on the other patients must be zero, and if it is more than four points for half the patients, then the medications must be interfering with responsiveness in at least some others who would fare better on placebo.
Another popular hypothesis is that drug effects are more stable than placebo effects, resulting in lower relapse rates. This hypothesis is also contradicted by the data. A meta-analysis of relapse prevention trials published between 1973 and 1990 indicated that 71% of the drug response was duplicated by placebo (Walach and Maidhof, 1999). Kirsch et al.'s meta-analysis also examined response to treatment as a function of the duration of the trial. The data indicated that responses to both drug and placebo decrease over time. Contrary to conventional wisdom, however, the correlation between duration of the trial and response to treatment was higher for active medication (r=-0.84) than for placebo (r=-0.62), suggesting a steeper decline in effectiveness for active drugs than for placebo (Kirsch et al., 2002b).
In light of these data, what should be done in clinical contexts? Some have suggested that antidepressants continue to be prescribed, even if their effects are largely placebo effects. If nothing else, these agents can be used as active placebos. Given the side effects of these medications, we suggest an alternative approach. There are many interventions that seem to be as effective or nearly as effective as antidepressants. These include physical exercise, bibliotherapy and psychotherapy (Kirsch et al., 2002b). Psychotherapy has the further advantage of demonstrated superiority to medications in long-term comparative studies (Antonuccio et al., 2002). Given these data, antidepressant medication might best be considered a last resort, restricted to patients who refuse or fail to respond to other treatments.