Effectiveness of antidepressantsBMJ 2018; 360 doi: https://doi.org/10.1136/bmj.k1073 (Published 09 March 2018) Cite this as: BMJ 2018;360:k1073
- 1Faculty of Pharmaceutical Sciences, UBC, Vancouver, British Columbia, Canada
- 2Department of Family Medicine, University of Alberta, Edmonton, Alberta, Canada
- Correspondence to: J McCormack
A recent meta-analysis by Cipriani and colleagues provides as good and balanced a synopsis as we will likely ever have of the results from the 522 trials of 21 antidepressants in 116 477 participants.1 The findings have been widely reported, with differing interpretations, including some uncritical acceptance of the benefits of antidepressants.2 More objectively, how should these findings inform practice?
Cipriani and colleagues are admirably clear about the limitations of included studies. They rated 82% as having moderate to high risk of bias. They also noted specific biases such as a novelty effect, whereby a medication looked significantly better when evaluated as the novel comparator in a trial than when as the older or control comparator. In addition, 78% of studies were funded by drug companies, and many other studies failed to report funding at all. It is somewhat reassuring that the authors report “funding by industry was not associated with substantial differences in terms of response or dropout rates.”
Finally, it is important to note the patient population in this meta-analysis was limited to adults with moderate to severe depression and an average Hamilton depression score of 26.3
The primary endpoint was the number of patients who achieved a 50% change in Hamilton depression score at 8 weeks. Remission rates at 8 weeks were also reported. The authors used “all-cause discontinuation” as a measure of acceptability since this endpoint combines both efficacy and tolerability. They also looked at drop-outs due to adverse events but didn’t report data on specific adverse effects such as sedation, dry mouth, sexual dysfunction, and weight gain—vital information for patients choosing between therapeutic options.
All antidepressants in this meta-analysis “worked,” as they all significantly increased the odds of a 50% reduction in Hamilton depression score relative to placebo. We calculated the average odds ratio for all antidepressants combined to be 1.66 for the 50% reduction in Hamilton depression score and 1.56 for remission. The authors state that overall they found “few differences between antidepressants when all data were considered” probably because confidence intervals around individual effect sizes were wide.
Importantly, these findings do not support the widespread calls in the popular press for more people to take antidepressants because the meta-analysis and underlying trials do not examine who or how many people should be treated. Furthermore, the way many of the results were reported does not allow clinicians to extract clinically meaningful take home messages to inform conversations with patients. Odds ratios can be somewhat misleading when there is a high baseline “risk” because they typically suggest a greater benefit (and harm) than the easier to understand risk ratio.
To put odds ratios into a more relevant clinical context, we first need to know what proportion of people got better in the placebo group. This information was not provided, but other research suggests that 30-40% of placebo group participants report improvement or remission in trials of antidepressants.45 Using this typical placebo response, and an odds ratio of about 1.6 means about 10-12% more people in the treatment group would benefit compared with the placebo group. So the number needed to treat for antidepressants in this population is roughly 8 to 10.
Knowing the absolute response rates in both the placebo (roughly 40%4 5) and treatment arms (roughly 50%—10% benefit plus 40% placebo) gives clinicians and patients valuable information for decision making. If 10 patients with moderate to severe depression take an antidepressant for two months, five (50%) will report being “better” but in four of them the response will not be because of the drug. Knowing that roughly 80% of patients who get better did not improve because of the antidepressant underlines the importance of starting with low doses, systematically re-evaluating the need for treatment after a response is achieved, and not accepting any enduring adverse effects.
All-cause discontinuation rates were not statistically different from placebo for most antidepressants in this meta-analysis. However, active treatment did increase the risk of dropping out because of side effects, with an average odds ratio of around 2.3. Absolute drop-out rates because of side effects in placebo groups are typically 3-5%,4 so an odds ratio of 2.3 means roughly 5% more people in treatment groups dropped out because of side effects.
Perhaps most importantly, this meta-analysis does not answer key questions about the use of antidepressants, including their effect on milder forms of depression, their effects beyond eight weeks of treatment, the harms associated with specific agents and their magnitude, and the effectiveness of antidepressants outside the confines of randomised trials. It did not consider the long term adverse effects of antidepressants, the likelihood of withdrawal symptoms when treatment stops, and the comparative benefits and harms of antidepressants relative to non-drug treatments such as cognitive behavioural therapy. It cannot tell us which antidepressant should be tried first, or which one is likely to work best for an individual patient.
Overall, while this meta-analysis is a meticulous and comprehensive summary of multiple therapeutic options, the clinical scope is limited and it only confirms (with greater precision) findings of other reviews.4678 Patients and clinicians need clear answers to many more questions before shared decision making in this area is fully informed.
Competing interests: We have read and understood BMJ policy on declaration of interests and have no relevant interests to declare.
Provenance and peer review: Commissioned; not externally peer reviewed.