NHS Evidence Health Information Resources formerly National Library for Health

Millions of Britons are taking anti-depressants for no reason reported eight newspapers (26 February 2008). These reported conclusions of a meta-analysis, which showed a small benefit of certain antidepressants over placebo. Though reports were generally accurate, the analysis did not include all available data on the effects of these drugs.

On 26th February 2008 eight newspapers (1-8) reported that antidepressant drugs are no more effective than placebo in all but the most severely depressed patients
These articles were based on a meta-analysis of 35 randomised, controlled trials that had been submitted to the US Food and Drugs Administration (FDA) for the new drug licensing of fluoxetine (Prozac), venlafaxine (Efexor), paroxetine (Seroxat) and nefedazone. The trials examined the effects of these drugs compared to placebo on the severity of depression in patients with unipolar major depressive disorder. The overall average difference in Hamilton Rating Scale of Depression (HRSD) between drug and placebo groups was 1.8 points, which was statistically significant but not deemed to be clinically significant.
Generally, the newspapers reported results of the meta-analysis accurately; however, one newspaper (2) reported that these findings were based on all available data, which was inaccurate as non-industry and post-licensing studies were not included.

Evaluation of the evidence base for the efficacy of antidepressant medication compared to placebo

Where does the evidence come from?

This meta-analysis was conducted by Dr Kirsch and colleagues as a collaborative project between the University of Hull (UK), the University of Wyoming (USA), the Centre for Health Intervention and Prevention, University of Connecticut (USA), the Institute for Safe Medication Practices, Pennsylvania (USA) and the University of Windsor, Ontario (Canada).

What were the authors' objectives?

To establish the relationship of baseline severity of depression and antidepressant efficacy using a dataset of published and unpublished clinical trials.

What was the nature of the evidence?

This was a meta-analysis of 35 randomised controlled trials (RCTs) involving 5133 participants. These RCTs were all the clinical trials submitted to the US Food and Drugs Administration (FDA) for the new drug licensing of fluoxetine, venlafaxine, nefedazone or paroxetine, including both published and unpublished data. The RCTs assessed the effects of antidepressant drugs and placebo on the severity of depression in patients diagnosed with unipolar major depressive disorder. The trials included inpatients and outpatients. The participants were elderly in three trials. No trials were reported for children or adolescents. For all but one group, baseline Hamilton Rating Scale of Depression (HRSD) scores were in the very severe range. The maximum duration of trials was eight weeks.

What interventions were examined in the research?

The effect of fluoxetine, venlafaxine, nefedazone and paroxetine versus placebo on the mean change in HRSD scores was examined. The relationship between the drug type, duration of treatment and baseline HRSD scores and the extent of improvement was also investigated.

What were the findings?

In the antidepressant groups there was an average improvement in HRSD scores of 9.6 points. In the placebo groups the average improvement was 7.8 points. The overall average difference in HRSD scores between drug and placebo groups was therefore 1.8 points. This difference was statistically significant, in favour of antidepressants.

Drug-placebo differences appeared to increase as a function of initial severity of depression, with almost no difference at moderate levels of initial depression, rising to clinically significant differences with more severe levels of depression.

The funnel plot presented did not rule out publication bias.

What were the authors' conclusions?

Compared with placebo the new-generation antidepressants do not produce clinically significant improvements in depression in patients who have moderate or severe depression. Significant effects were only shown in the most depressed patients. The effect shown for these patients appears to be due to decreased responsiveness to placebo rather than increased responsiveness to medication. There is therefore little reason to prescribe antidepressants to patients unless they are severely depressed and alternative treatments have been ineffective.

How reliable are the conclusions?

This meta-analysis addressed a clearly defined question however there were a number of methodological issues that may affect the interpretation of the results. The authors obtained all trials which had been submitted to the FDA for marketing approval. This approach aimed to minimise the potential for publication bias by including all data regardless of whether it was published or not. However, no attempt was made to identify studies from other sources, so post-marketing evidence of these drugs was not included. This could have included non-industry research and longer-term studies conducted in settings similar to common clinical practice. Though publication bias cannot be ruled out, these data could also have been taken into consideration.

The analysis was limited to those drugs for which sufficient datasets were available (fluoxetine, venlafaxine, paroxetine and nefedazone); the results cannot necessarily be generalised to all newer antidepressants. Additionally, non-mood outcomes were not investigated in this study, for example sleep disorders and anxiety.

Systematic reviews

Information staff at CRD searched for systematic reviews relevant to this topic. Systematic reviews are valuable sources of evidence as they locate, appraise and synthesize all available evidence on a particular topic.

There were three related systematic reviews identified on the Cochrane Database of Systematic Reviews (CDSR) (10-12) and four reviews which are currently being undertaken and will be available in the future (13-16). Additional reviews are also available on the Database of Abstracts of Reviews of Effects (DARE).