by Michael D. Anestis, M.S.
In a number of past PBB articles, we have discussed the mountains of research supporting the efficacy and effectiveness of antidepressant medications in the treatment of depression. Within the past 24 hours, both Thomas Joiner and Jed Siev have alerted me to an article published in the most recent issue of the Journal of the American Medical Association (JAMA) that provides some valuable insights capable of adding a lot to this conversation. I appreciate their heads up on this and think that it would be useful to talk about this paper today. For those interested in reading another take on the findings, the study was also discussed in today's New York Times.
The authors - and there are a lot of them - were Jay Fournier, Robert DeRubeis, Steven Hollon, Sona Dimidjian, Jay Amsterdam, Richard Shelton, and Jan Fawcett (2010). These researchers represent some of the most prolific in the field and we have covered various examples of their work numerous times over the past year (type in their names in the Lijit search box at the top of the page for examples). In this particular study, the authors wanted to examine a question with enormous clinical implications: does the severity of an individual's depression impact the utility of antidepressant medication relative to a placebo.
There were several motivating forces that drove the authors to pursue this particular series of analyses. First of all, prior work has, in fact, supported the notion that the degree to which antidepressants outperform placebos increases as the severity of depression increases (e.g., Kirsch et al., 2008; Kahn et al, 2002). Importantly, in the Kirsch et al. (2008) study, the authors noted that, in order for antidepressants to result in a clinically significant difference in utility over placebos, individuals typically needed to score at least 28 on the Hamilton Depression Rating Scale (Hamilton, 1960). As a point of reference, scores of 18 or lower are considered to represent mild to moderate depression, scores of 19 to 22 represent severe depression, and scores of 23 or greater represent very severe depression (APA, 2000). So, the authors noted that prior work had indicated that, in order for antidepressants to have an impact above and beyond what could be accounted for by placebo, individuals needed to be very severely depressed. At the same time, the authors noted that these earlier findings included some fairly substantial limitations.
The limitations of the prior studies tie into another motivating force behind this study. Earlier trials like the ones mentioned above typically require that participants be severely depressed. There is nothing inherently wrong with looking at the treatment of severely depressed individuals; however, such individuals make up only a small minority of those who present for the treatment of depression. In other words, mildly or moderately depressed individuals who are prescribed antidepressants are essentially done so on the basis of studies that look only at the impact of antidepressants on people whose condition is much worse than their own rather than research demonstrating that antidepressants offer clear benefits for individuals at their particular level of symptomatology. Additionally, many clinical trials involve what is referred to as a placebo washout period. This procedure involves giving a placebo to a participant for a period ranging from a couple days to two weeks without letting them know it is a placebo and then eliminating from the analyses prior to randomization any participant who demonstrates marked improvement in depression during this period. In other words, this procedure involves eliminating participants who respond to placebo and including those who do not, thereby likely artificially inflating the difference between those who receive medications and those who receive placebo (as those who respond well to the placebo during the washout period are not included in the analyses).
One final limitation of earlier trials that is important to consider relates to a point we have raised before on PBB: the limitations of meta-analysis. Generally speaking, in a meta-analysis, a number of studies are combined and their results are averaged in order to find overall effects. In doing this, a study that includes 100 people and a study that includes 400 people are simply combined to represent 2 studies (and so on depending upon how many studies are included in the meta-analysis) so, rather than a sample of 500 people on whom we can run sophisticated analyses, we have a tiny sample size of two. Additionally, because most studies use different outcome variables, outcomes are combined arbitrarily and weighted equally, thus completely altering and misstating what was actually found in the original studies.
So...what did Fournier and his colleagues (2010) do about this? First, they found randomly controlled trials of FDA-approved antidepressant medications in the treatment of a wide range of depression severity. In other words, they did not only look at severely depressed individuals. All of the studies included had to involve a comparison of antidepressant medication and placebo that lasted at least 6 weeks. No studies that involved placebo washout periods were included. Additionally, all of the studies had to use the Hamilton Depression Rating Scale, thereby ensuring that the authors would not be comparing different outcomes to one another as though they were the same. The final - and perhaps most important - inclusion criterion was that, in order to be included, the authors had to be granted access to the original data, thereby allowing them to analyze all of the participants rather than simply looking at the aggregate results. Because of this, the authors were able to combine the samples of all of the included studies (a total of 6 studies), resulting in a total of 718 participants (434 in the antidepressant condition and 284 in the placebo condition). Three of the studies included in the analyses examined Paxil, which is a selective serotonin reuptake inhibitor and three of the studies examined imipramine, which is a tricyclic antidepressant. All of the participants were at least 18 years of age and the authors found that attrition rates did not differ between conditions, meaning that participants were no more likely to drop-out of the placebo condition than the antidepressant condition and vice versa.
After combining the samples from the six studies, the authors ran their analyses and found that, as expected, the degree to which antidepressants outperformed placebos depended upon how depressed the individual was at baseline. For mild to moderate depression, the effect size, which is a measure of how powerful a finding is, was very small (d = .11). For severe depression, the effect size again was very small (d = .17). For very severe depression, however, the effect size (d = .47) was approximately "moderate" using the standard established by Cohen (1988). Taking this a step further, the authors found that, to meet the least stringent criterion for clinical significance established by the National Institute for Clinical Excellence (NICE) of the National Health Service in England, individuals needed to score at least a 25 on the Hamilton Depression Rating Scale in order for antidepressants to have an impact above and beyond a placebo. Using stricter criteria, the requirement increased to 27. Keep in mind that many depressed individuals score well below that number and that a score of 23 is said to represent very severe depression.
In order to make sure that very mild cases were not obscuring the findings, the authors re-ran their analyses while only including the five studies examining major depression (the other study looked at minor depression). The results were unchanged.
In order to ensure that participant drop-out did not artificially impact the findings, the authors re-ran their analyses while only including individuals who completed the entire study. This resulted in only a minor change in the results, with the minimum required score dropping from 25 to 24.
So what does this study tell us? First of all, it raises questions regarding the utility of antidepressants in the treatment of mild to severe depression. Secondly, it highlights the importance of running similar studies while including various forms of psychotherapy, thereby allowing a measure of the degree to which empirically supported treatments for depression, such as cognitive behavioral therapy and interpersonal psychotherapy, impact these less severe presentations.
As is the case with most of the studies we discuss on PBB, this work again shines a light on the importance of understanding how doctors and scientists arrive at their conclusions rather than simply listening to what those conclusions are. If we do not develop an understanding of how people determine whether or not a particular treatment works, we will be unable to know when findings are overstated or data is misinterpreted.
Please keep in mind that this article is not meant as an attack on antidepressant medication. A quick read of our prior articles on this topic will reveal that we are very much in favor of the use of any empirically supported treatment for a particular diagnosis, whether that treatment is a form of psychotherapy or a form of pharmacotherapy. That being said, this article is most definitely intended as a bit of a wake-up call, as sometimes the data give us a more complicated picture than we anticipated seeing.
If you would like to learn more about depression and its treatment, we recommend the following items, each of which is available through our online store for scientifically-based psychological resources:
- Cognitive Therapy of Depression
by Aaron Beck, John Rush, Brian Shaw, and Gary Emery
- Treatment Plans and Interventions for Depression and Anxiety Disorders by Robert Leahy and Stephen Holland
- Feeling Good: The New Mood Therapy Revised and Updated by David Burns
- The Feeling Good Handbook by David Burns