Resources for Medical Professionals
Do Antidepressants Really Work?
In January 2010, JAMA published a meta-analysis by Fournier and colleagues concluding that the efficacy of certain antidepressants increases with disease severity. The report further concludes that the benefits of these agents may be minimal or nonexistent in patients with mild-to-moderate symptoms. Given the potentially significant clinical implications of these findings, Medscape asked 3 experts in depression to weigh in on the legitimacy of the study and how such data might affect practice.
Joseph F. Goldberg, MD: The goal of examining antidepressant vs. placebo effects for mild-to-moderate (rather than severe) depression is an important one -- particularly since many depressed patients may not seek or receive treatment until their symptoms significantly impinge on their functioning or quality of life. There have not been nearly as many controlled studies of antidepressant efficacy for mild-to-moderate depression as there have been for more severe forms of major depression.
Fournier and coworkers initially identified 23 studies, and ultimately included only 6 in their meta-analysis, which involved only 2 antidepressants: paroxetine or imipramine. For the purposes of a meta-analysis, this is perhaps the greatest problem limiting inferences from their paper -- it provides no discussion of the adequacy of statistical power in their 6 pooled studies for identifying the true absence of a drug-placebo difference. "Floor effects" occur with only 6 studies in the meta-analysis, and generalizability of the findings is severely limited.
The authors purposely excluded studies that involved an initial, single-blind, lead-in phase, attempting to remove early placebo-responders. However, in doing so they further reduced the potential for identifying drug-placebo differences and increased the possibility of favoring the null hypothesis (ie, no difference between drug and placebo).
Such factors as chronicity or high recurrence rates of depression (regardless of severity), as well as a plethora of other features that could moderate drug-placebo differences (including age at onset, comorbid psychiatric or medical disorders, social or work functioning, capacity for resiliency, effect of past treatments, and delayed treatment initiation), are not considered.
Treatment for mild-to-moderate depression tends to involve a higher placebo response rate than that seen with more severe forms of depression, and we do not know from the meta-analysis whether placebo response rates in the 6 included articles were higher than those reported for more severe forms of depression; if so, the findings may say less about the efficacy of antidepressants than about the potential for a placebo response in the condition under study.
Two of the studies included in the meta-analysis involved randomization to drug vs. placebo vs. psychotherapy groups, while the other studies did not. Recruitment of prospective participants may have introduced bias inasmuch as persons who seek treatment with psychotherapy may differ from those who seek treatment with medications only. (I have been involved in randomized studies involving drug vs. psychotherapy groups, and patients have revoked consent to participate if they are randomized to psychotherapy.) The meta-analysis findings pertain to participants who have been randomized, and it is unknown whether mild-to-moderately depressed prospective participants who were screened but not randomized differed in pertinent characteristics (especially regarding expectations about treatment) from those who were randomized.
No inferences can be drawn about any antidepressants other than paroxetine or imipramine. Furthermore, since both medications are quite anticholinergic, it is likely that some or many participants inadvertently became unblinded due to obvious side effects. If they had lower expectations about the effects of placebo or psychotherapy than medication, inadvertent unblinding could have severely curtailed the potential for identifying true drug-placebo differences.
Finally, despite the small number of studies and medications used, the absence of a drug-placebo difference conveys an easy, sensationalist, anti-psychiatry/pharmacology element that the popular media may be quick to embrace and promulgate with little if any cautionary statements about the questionable validity of the findings and methodology. Had the results been positive, this paper would surely have drawn no media interest.
Mark H. Pollack, MD: Meta-analyses have greater power to examine questions of interest than do individual studies; however, results can be influenced by numerous factors, including decisions about which studies are included or excluded, and the effects of mixing studies with different methodologies. These problems characterize the meta-analysis by Fournier and colleagues, which excludes numerous studies examining antidepressant efficacy (including those using methodology to reduce placebo response) and focuses on only 2 antidepressants, thereby constraining its generalizability. Further, with lower initial severity scores in patients with mild-to-moderate disease, there is an intrinsic ceiling effect that restricts the magnitude of differences that can be observed between the treatment groups. Given these limitations, the finding of an effect for antidepressants for more severe depression and failure to observe an effect for active antidepressants above placebo in patients with mild-to-moderate depression is perhaps not altogether surprising.
However, given the caveats and limitations of the methodologies, what can we take from this study? The finding that for low-to-moderate depression there was no significant difference between placebo and antidepressants has potential implications for clinical practice. Patients, particularly those with mild-to-moderate depression, may benefit from treatment with a number of nonpharmacologic interventions, including exercise and some psychotherapies. Studies to identify patient and illness characteristics and biomarkers that might guide selection of appropriate initial treatment for these patients are clearly needed. In addition, it is important to realize that placebo administration is not the same as receiving no treatment. Patients entering randomized, controlled trials are generally seen every week or every other week and receive extensive diagnostic evaluation and detailed descriptions of what they can expect during the course of treatment in the study and potential side effects of the medications. They have regular contact with physicians and other staff who provide support, including information about their disorder and a sense that their experience is understood. These elements have an apparent salutary effect on a patient's mood and should be integrated into treatment of individuals, regardless of whether they are receiving pharmacotherapy or nonpharmacologic interventions.
Andrew A. Nierenberg, MD: The meta-analysis by Fournier and colleagues reignites an important debate but is flawed by inclusion of a limited set of small studies that compares antidepressants with placebo for major and minor depressive disorder. Of the 6 studies included, the trial by Barrett and coworkers on minor depression in primary care assumes that minor depression is similar enough to major depression to be included in the meta-analysis -- an assumption that has yet to be proven; the study also did not require patients to have dysfunction associated with minor depression. As a result, one could argue that this study should be excluded. This leaves 5 other studies. The lead authors of two of these studies, DeRubeis and Dimidjian, are in the group that wrote the meta-analysis, presumably because they had access to their own data. However, the purpose of these studies was not merely to assess the efficacy of antidepressants but also to test psychotherapies. What is striking is the absence of any mention of the effect size of psychotherapy for the same groups -- does psychotherapy differ from placebo for less severe depression in their meta-analysis?
The studies by Elkin and Phillip and their colleagues were done over 20 and 10 years ago, respectively, and are underpowered. The former study included psychotherapy and the latter assessed hypericum (St. John's wort) with a potential nocebo effect. The sixth study by Wichers is so small ( N = 64) that it has no power to detect anything but a very large effect size. In fact, with only 180 total participants in the milder group (Hamilton Depression Rating Scale score < 18), if the groups were divided into antidepressant and placebo groups, each group would have only 90 participants. Then, if you remove the Barrett group with 39 and 38 randomized to placebo and active groups, respectively, then the comparison for those with less severe depression is 51 and 52 in each group. With such a small sample size, to conclude that the lack of statistical difference between the groups means the absence of a difference is simply wrong: The comparison is too underpowered to come to this conclusion. The authors even state that the upper bound of the 95% confidence interval for Cohen's d is .41. To make sweeping statements from the amalgam of putting together patient-level data from these particular studies is, at best, to overinterpret and overgeneralize the findings.