If you’re confused about whether selective serotonin reuptake inhibitors (SSRIs) cause suicide in children, you’re not alone. Here’s a guide to data that have recently come to light.

Readers of this column hopefully will have realized by now that my allegiance as a physician is not to drug companies, nor to the Food and Drug Administration, nor to critics of either group. It’s to the data.

I’m a huge believer in the idea that when we have access to quality scientific data, complex decisions become much clearer and easier to make. So in my next two columns, I’ll analyze data on two medical controversies that have generated lots of fire, but not always sufficient light. First, I’ll tackle the evidence and safety data for antidepressants in children. Then I’ll follow with a look at the reported cardiovascular risks posed by cyclooxygenase 2 (COX-2) inhibitors. My arguments will rely heavily on confidence intervals and risk ratios, which I discussed in my first three columns (January 2005, page 38; March, page 20; May, page 19). So consider this an application of those lessons.

SSRIs and Children

In late September 2005, the U.K.’s National Institute for Clinical Excellence (NICE) advised doctors in the country’s National Health Service that antidepressants should only be given to children under 18 years of age in conjunction with psychotherapy. Their new recommendations, which also included suggestions on diet and exercise, were the latest in a series of guidelines issued by U.K. and U.S. authorities on the subject.

Previously, the U.K. Medicines and Healthcare products Regulatory Agency (MHRA) said that all SSRIs except fluoxetine were contraindicated as new treatment for patients younger than 18 years of age with depressive illness. The FDA last year requested that drugmakers include a warning that "antidepressants increase the risk of suicidal thinking and behavior (suicidality) in children and adolescents with major depressive disorder (MDD) and other psychiatric disorders" but has not issued guidelines as strong as those in the United Kingdom.

Since the MHRA made its first pronouncement on the subject in June 2003, the issue has been fraught with controversy and politics. What was so frustrating about the ensuing debate is the fact that only some of the actual data was available to doctors who needed it, let alone to the public. That’s because much of the data was unpublished—held onto by drugmakers to prevent bad news getting out, if you believe some critics, or because it wasn’t very well analyzed, according to the companies.

Regardless of the reason, the data many of the early reports were based on are now available, and it’s a good time to subject them to the kind of rigorous analysis I’ve tried to champion in these columns—especially since it’s not clear that September’s NICE recommendations will be last word on the debate.

To get started, I’d suggest getting a copy of Whittington et al’s meta-analysis of the data, which ran in The Lancet (2004;363:1341-1345) after the U.K.’s Committee on Safety of Medicines (CSM) made it available. The issues in the case of these medications were twofold: one, did these drugs work to treat depression in children, and two, did they carry a risk of suicide attempts?

Let’s look, as Whittington et al did, at the data the CSM released in December 2003 as the basis of their determinations. Some of the data, such as those published by Emslie et al (Arch Gen Psychiatry 1997;54:1031-1037; J Am Acad Child Adolesc Psychiatry 2002;41: 1205-1215) had already been published. The authors performed statistical analyses on the raw data that the CSM had made available, calculating risk ratios and P values in places where the CSM had not.

First, fluoxetine: There was a benefit in the two studies that had been published, and it was statistically significant. The relative risks of "nonremission" and "nonresponse"—reverse ways of saying that fluoxetine worked—were less than 1 in both studies (0.78 and 0.67, respectively), and the 95% confidence intervals (CIs) did not cross 1 (0.67-0.90 and 0.52-0.87, respectively). So we can conclude, as the Whittington meta-analysis did, that there is some evidence of benefit.

What about harm? In two unpublished studies that Whittington included in his meta-analysis, a total of 6 out of 249 children taking fluoxetine (2.4%), vs. four of 209 (1.9%) taking placebo, made suicide attempts. While that gives a relative risk of 1.26, the 95% CI is 0.36 to 4.40. Since that crosses 1, it’s not statistically significant.

Suicide Risk Muddled By Poor Definitions

It’s worth noting that these studies gave various definitions—or no definitions at all—of suicidal ideation, suicidal gestures and even of vague terms such as "emotional lability." Some of the studies mix categories, as in "suicidal attempts or ideation." It turns out that in one case, "suicidal gesture" referred to a teenage girl hitting herself in the cheek, according to evidence presented at MHRA and FDA hearings. That’s another layer of difficulty in terms of analyzing these data, which is why I’m sticking to "suicide attempt" whenever possible, since it has the clearest clinical meaning.

Next, paroxetine: One 2001 study that was included in the Whittington meta-analysis came close to finding a significant benefit in terms of "nonremission"—a relative risk of 0.72, and a CI of 0.52-0.99. Technically close, but none of the other studies, published or unpublished, had CIs for "nonresponse" that didn’t cross 1. When it comes to suicide attempts or ideation—the two are lumped together—there were 5/93 (5.4%) in children taking paroxetine, vs. none in 87 taking placebo. That sounds significant, with a relative risk of 10.3, but in fact the CI was 0.58-183.5—an impressive, but not statistically significant, gap. When the analysis included unpublished data that were eventually released in the CSM 2003 report, there were 14 suicide attempts or ideation in 378 children taking the drug (3.7%) and 7 in 285 children taking placebo (2.5%). Relative risk was 1.51, but not statistically significant at a CI of 0.62-3.69.

Sertraline (Zoloft, Pfizer) had a similar profile. One study included in the Whittington meta-analysis (Wagner et al. JAMA 2003;290:1033-41) showed a relative risk of nonremission of 0.92, with a CI of 0.63-1.00—almost good enough. That study found suicidal attempts or ideation in 5/189 children on sertraline (2.6%) vs. 2/184 on placebo (1.1%). Again, that’s a relative risk of 2.43, but the CI is 0.48 to 12.39—not statistically significant. What is the profile for citalopram (Celexa, Forest Laboratories)? No published data at all, and no unpublished data on efficacy. In two unpublished studies eventually released by the CSM, 15 in 210 children (7.1%) taking the drug made a suicide attempt, and 7 in 197 on placebo (3.6%) made one. That wasn’t statistically significant; the relative risk was 1.99 with a CI of 0.83 to 4.77. Finally, for venlafaxine (Effexor, Wyeth), there are no data on efficacy, and there were "suicide-related events" in 14/182 taking the drug (7.7%) vs. 1/179 taking placebo (0.6%) (relative risk 13.77, CI 1.83-103.61—yes, statistically significant).

The Verdict

So, the verdict? For fluoxetine, some benefit, no clear evidence of harm. That fits with the original MHRA guidance, which singled out fluoxetine as having a good benefit-risk ratio. For the rest, there’s either no good evidence of benefit, in which case any increased risk of suicide is unacceptable, or, in the case of venlafaxine, decent evidence of increased suicide risk. A final note from the authors of the meta-analysis: "The studies included in our analysis were not designed to investigate rare events, such as suicide, and as such are unlikely to have sufficient statistical power to detect potential risk," Whittington et al wrote. But they also note that "in view of the high risk of suicide in this group of children and young people, the possibility that a drug might increase that risk without clear evidence of benefit, should, in our view, discourage its use."

So there you have it on SSRIs and children—all the best data there seems to be. Not quite Walter Cronkite’s "that’s the way it is," but close.

Stay tuned for the Vioxx story.