Table of Contents

Volume 1, Issue 4                                                                                              April, 2015


Classic Study of the month

The FDA analysis of antidepressants and suicide


It's not as bad or as good as you think


As discussed in the statistics article this month, randomized data are more valid scientifically than non-randomized data. This collection of randomized studies is the largest and most definitive evidence regarding risk of suicidality caused by antidepressants in children and adults.  

The background is as follows: Pharmaceutical companies were encouraged by the FDA to conduct randomized clinical trials (RCTs) on antidepressants in children and adolescents.  The wish was to make these agents, believed to be so effective for major depressive disorder (MDD) in adults, available  to children too.

Unfortunately for the pharmaceutical companies, who conducted over a dozen trials in search of this lucrative market, the agents were only somewhat effective. To make things worse, as the FDA examined side effects in these trials, it identified some evidence of suicidality, defined as worsened suicidal ideation or suicide attempts.  This is when the FDA decided to meta-analyze the data from all the studies which the companies had done (which they were legally obligated to provide to the FDA, although they have no legal obligation to make those data available to the public or the scientific community).  When the FDA statisticians conducted those analyses, they found the results which they later published in this paper.  The FDA and its advisors then voted to institute a black box warning in the serotonin reuptake inhibitor (SRI) package inserts saying these agents may cause or worsen suicidality. 

All this occurred in 2004, and led to a major backlash from the psychiatric community, especially from child psychiatry clinicians.  Nonetheless, the FDA had an impact, and antidepressant use has declined in the last decade in children.  Various epidemiological studies have been published since that date, looking at insurance databases or clinical samples, arguing that suicide rates did not decrease, or even increased, after the FDA black box. But, as described in the article above, population based suicide rates, observed in a completely uncontrolled fashion, cannot be causally attributed to any single purported cause. As explained in the statistics article, only RCTs can justify claims of causality that are likely valid. 

In this meta-analysis, a few points are important:

  1. All the studies were conducted quite similarly for FDA approval, being of similar duration, with the same rating scales, and similar inclusion and exclusion criteria.  Hence, they are comparable and the meta-analysis has limited “heterogeneity” (confounding bias, as described in the statistics article).  In other words, this review is about as valid as a meta-analysis can get. 
  2. There were no suicides, as many of the opponents of the FDA black box always point out.  But these RCTs were conducted for FDA approval of drugs for kids. The last things drug companies could afford would be a suicide in a RCT.  The media backlash that would ensue would cost a lot of money.  Suicidal ideation is a major exclusion criterion for these studies.  In short, the patients were selected carefully to be highly non-suicidal.  And yet some of them got suicidal as shown in these data.  This argues for these data being even more, not less, concerning.  
  3. The absolute rates matter.  The meta-analysis shows about a 67% increase in suicidality with SRIs over placebo. But the absolute rates were about 2% with SRIs versus 1% with placebo.  In other words, about 1% of children will get more suicidal with SRIs, as opposed to placebo (mainly a stand-in for natural history).  Put otherwise:  99% of children will not get more suicidal. This absolute rate provides clinical context for decision-making.  This doesn't mean one should completely ignore these data. There is an increased risk of suicidality here, but it doesn't apply to most patients.  Nonetheless, the 1% matter, especially when death is the end-result.
  4. But will any child die?  Many assume not, since no one died in these 8 week trials. That assumption ignores the research literature that about 10% of persons who make suicide attempts will eventually kill themselves. Elsewhere (Ghaemi 2009) the PL editor has calculated the “number need to harm” using these figures, which is a way of combining the high relative risk of suicidality (almost two-fold increased risk) with the low absolute rate of 1% increased suicidality, and then estimating that about 10% of that number will eventually commit suicide.  The calculation made using these numbers produces about a 1 in 500 risk of completed suicide.  In other words, about 1 in 500 children treated with SRIs over time will commit suicide, attributable to the SRI (based on RCTs versus placebo, which infers causality). 1 in 500 isn’t a high rate for a side effect, but if it’s fatal, it’s worth at least getting informed consent from patients. It will be argued that SRIs also improve depression, which is true, and prevent suicide indirectly by improving depression.  The PL editor has conducted a similar “number needed to treat” analysis of suicide prevention rates, combining improvement with depression with actual suicide rates in depression in adolescents.  The overall result was again prevention of suicide in 1 in 500 persons.  So 1 in 500 commit suicide and 1 in 500 are prevented from committing suicide.  In this risk-benefit analysis, antidepressants come out neutral.
  5. In another FDA analysis of all their antidepressant studies in adults, seen below,  increased the age of 25 and for children.  There is a notable increased risk there, but a decreased risk in all older adult groups, such that there is an overall decreased risk of suicidality with antidepressants of 17% (odds ratio of 0.83 which is 17% decreased risk compared to the null value of one).

What does this mean? As readers of the PL website will know, early onset of depression  (below 25 years) was said to reflect a higher probability of bipolar illness, as opposed to unipolar depression. This raises the question that perhaps antidepressants are causing suicidality in young adults and children because "depression" in those ages is not "MDD" but rather bipolar illness. The PL website elaborates on how early-onset depression represents what used to be called manic-depressive illness. 

The PL Bottom Line

  • SRIs increase suicide risk in 1% of children, and lead to completed suicide in about 1 in 500, which is the same as their prevention rate.
  • Their overall effect is probably neutral when benefits are weighed against harms. 
  • SRIs shouldn’t be viewed as completely safe.
  • SRIs shouldn’t be used routinely in  children and young adults below age 25.  

PL Reflection

"The first principle [of science] is not to fool yourself. And you are the easiest person to fool."     

Richard Feynman

Meet our expert EDITORIAL BOARD, composed of clinicians and researchers from around the world. 

Subscribe to the RSS feed below to follow our "What's new" blog posts