Monday, September 17, 2007

Peer Review, SSRIs, Suicide, and Booze

The recent study in the American Journal of Psychiatry by Gibbons, Mann, and colleagues regarding the relationship between SSRI usage and suicides reads more like an exercise for undergraduate students to find obvious errors than it does a real peer-reviewed study. Sounds mean, but keep reading.

The abstract of the study includes the following...

"In both the Unites States and the Netherlands, SSRI prescriptions for children and adolescents decreased after U.S. and European regulatory agencies issued warnings about a possible suicide risk with antidepressant use in pediatric patients, and these decreases were associated with increases in suicide rates in children and adolescents."

So less SSRIs = more suicides, according to the authors. Let’s see if this study actually shows such a relationship…








Look closely at the above graphs (click to enlarge) from the article. Note that the decrease in SSRI prescriptions from 2003 to 2004 was very slight across the 0-10, 11-14, and 15-19 age groups, which is the timeframe in which suicide rates for those aged 5-19 increased notably. The larger declines in SSRI prescribing for youth occurred from 2004-2005, which happens to be when the suicide rate for those aged 15-24 appears to have decreased from 10.3 per 100,000 (see Table 9; page 28 here) to 9.8 per 100,000 (see Table 7 here). Yes, I know I am comparing data for ages 15-24 to data on ages 5-19, but I think this makes sense when one considers that the suicide rate for those 14 and under is much lower than for those aged 15-24. Actually, grouping suicide data for ages 5-19 makes little sense to me given the vast differences in suicide rate within this age group.

It is important to note that the authors of the paper did not have data from 2005, but there is nothing from the 2003-2004 U.S. SSRI prescription data cited in their paper that even suggests a relationship between decreased SSRI use in youth and an increased suicide rate, as the decrease in prescriptions was minimal. Pay close attention: The authors ran a total of zero statistical analyses to examine the relationship between SSRI prescription rates and suicide rates in the United States. That’s right, zero. So they put up a couple of figures without a single shred of statistical evidence, then claim that declining SSRI prescriptions are associated with an increase in suicide rates. Any peer reviewer who was not drunk or on a high dose of Seroquel should have noticed this gigantic flaw.

In the discussion, the authors state: “While only a small decrease in the SSRI prescription rate for U.S. children and adolescents occurred from 2003 to 2004, the public health warnings may have left some of the most vulnerable youths untreated.” This is unadulterated speculation, which as I just mentioned is not supported by a single statistical analysis in their paper. It is also hard to imagine how an FDA warning in mid-October could make suicides earlier in the year increase. One can only wonder to what magical time-traveling extent an FDA warning in October could have increased suicide rates earlier in the year. This is so mind-bogglingly obvious that, again, the peer reviewers were possibly inebriated during the review process, or the editor published the paper over the objections of the reviewers. Am I being too nasty? I'm just trying to figure out how it got published and "good science" is not the answer.

The authors then proposed the following:

…we estimate that if SSRI prescriptions in the United States were decreased by 30% for all patients, there would be an increase of 5,517 suicides per year…

In addition…

In children 5 – 14 years of age, a 30% reduction in SSRI prescriptions would lead to an estimated increase of 81 suicides per year… Given that SSRI prescriptions for children under age 15 already underwent a reduction of approximately 17% from 2003 to 2005, we expect an increase of .11 suicides per 100,000 children in this age group. Since there are approximately 40 million children in this age group, we would expect 44 additional deaths by suicide in 2005 relative to 2003, or an increase of 18% in this age group.

Preliminary 2005 suicide data indicate a suicide rate in 5-14 year olds of .7 per 100,000, holding steady from 2004. This does not support the predictions of Gibbons and colleagues. Granted, the 2005 data are preliminary, but I’d be surprised if they showed a large change in the direction that Gibbons, Mann, and their team predicted.

Again, let me state that these are only correlational data and that data from clinical trials as well as other sources trumps these types of studies in any case. At the very least, when doing correlational research, try to control for covariates (other variables of interest), examine trends over a longer time period than one year, and maybe actually run some statistics. Oh, and avoid conclusions that require belief in time travel. There are even more potential problems, but the authors missed so many glaring basic issues that it makes no sense to go any deeper.

If data based on correlations is going to be trotted out to scare physicians into prescribing more SSRIs, then it should be examined whether the correlations provide even preliminary support for the idea that SSRIs might reduce suicide. I've criticized many studies on this site for a variety of concerns (like here and here, among many examples), and I think the present study is among the worst offenders of basic research methodology. Until we clean up the "science," don't expect much real progress in the mental health treatment world.

Background here and here.

Major Hat Tip: Furious Seasons.

12 comments:

herb said...

Dear Doc,

“Until we clean up the "science," don't expect much real progress in the mental health treatment world.” --- CL PSYCH

Sorry, but once again I think you’re off base and missing the point.

There’s nothing wrong with the “science.” The problems lie with those that ply their profession and education to the “science.”

Try first cleaning up and improving “your” educational system and the professionals it produces while also adding some courses on ethics.

I’m sorry for being so critical but then again I’ve learned from you and several others on these blog sites.

Warmly,
Herb
VNSdepression.com

Ben Hansen said...

Here's ammunition for the "black box warning linked to increased suicides" debate.

See Table 29, "Studies assessing Suicidal thoughts and behavior" on page 106 (page 117 in PDF format) of the Agency for Healthcare Research and Quality Final Research Review published January 2007,
"Comparative Effectiveness of Second-Generation Antidepressants in the Pharmacologic Treatment of Adult Depression."


Twelve studies are cited: none show any difference between any of the SSRIs; none show any difference between the SSRIs and older "first generation" antidepressants. Only 4 of the 12 studies compared antidepressants to placebo, and in all 4 of these studies, PLACEBOS OUTPERFORMED ANTIDEPRESSANTS. Summarized below.


Fergusson et al., 2005.
Meta-analysis: SSRIs vs. placebo.
Results: Higher risk of suicide attempts for SSRI-treated patients.

Gunnell et al., 2005.
Meta-analysis: Citalopram, fluoxetine, fluvoxamine, paroxetine, sertraline, all vs. placebo.
Results: Increased risk of nonfatal suicide attempts compared with placebo; no difference in risk among drugs.

Pedersen, 2005.
Retrospective cohort study: Escitalopram vs. placebo.
Results: Higher rate of nonfatal suicide attempts for escitalopram than for placebo.

Aursnes et al., 2005.
Meta-analysis of unpublished Paroxetine data.
Results: Higher rate of suicides for paroxetine than for placebo.


Read the AHRQ report here:
http://effectivehealthcare.ahrq.gov/reports/topic.cfm?topic=8&sid=39&rType=3&sType=2
http://effectivehealthcare.ahrq.gov/repFiles/Antidepressants_Final_Report.pdf

Anonymous said...

It's an interesting and ongoing debate - it depends I guess on your view of the efficacy of antidepressants. I heard a talk recently by a London academic about the myth of the chemical imbalance and the history of antidepressants. I've found a paper by her here. Apologies - it's a pdf.

Mark p.s.2 said...

PLACEBOS rock! Poisons rule!
Why isn't marijuana or opiates tested-used against depression?

or the phobia against sticking wires into peoples brains and stimulation their pleasure centers, like Herb wants. Larry Niven speculated about with Wireheads
Its better than brain damaging E.C.T. for depresion.

Anonymous said...

I'm a bit mystified by your comment that the authors used no statistical analysis. In the paper they used Poisson regression analyses (essentially a correlation that can be used with count data) and, indeed, they report the results:

e.g. "this represents a 14% increase in the suicide rate from 2003 to 2004 (p<0.0001)" etc etc.

So, their effect is weak, but statistically acceptable.

Anonymous said...

I'm a bit mystified by your comment that the authors used no statistical analysis. In the paper, they report the results of several Poisson regression analyses (essentially a correlation that can be used with count data) and, indeed, they report the results:

e.g. "this represents a 14% increase in the suicide rate from 2003 to 2004 (p<0.0001)" etc etc.

So, their effect is weak, but statistically acceptable.

CL Psych said...

Vaughan,

The authors reported analyses for the Netherlands, not the United States. I stated that for the United States, the authors ran not a single analysis that correlated SSRI prescriptions with an increased suicide rate, and I stand by my statement. Thus, any commentary from the authors on the U.S. data is pure speculation not based on any statistics. And the 2005 data that the authors did not have access to at the time contradicted the authors' claims in any case.

CL Psych said...

Vaughan,

A follow-up to my prior comment...

I wrote in the post: "The authors ran a total of zero statistical analyses to examine the relationship between SSRI prescription rates and suicide rates in the United States. That’s right, zero."

Please re-examine the paper. The authors did indeed find that there was a significant increase in the US suicide rate for ages 5-19 from 2003-2004. I don't dispute this point. However, this was not shown to be correlated with a decrease in SSRI prescriptions because the authors did not conduct any such analysis for U.S. data. In a study alleging to show a correlation between two variables, one must actually examine whether the two variables are related statistically.

Unknown said...

Hi CL Psych,

I think you're right.

One of the difficulties I think is that the results are so ambiguously written, I can't actually tell which p value is related to which analysis.

I've emailed to the lead author to clarify.

Thanks!
Vaughan

Anonymous said...

As an update, I got a reply from Dr Gibbon and it confirmed what CL Psych was saying. There is no statistical test of the relationship between suicide and prescription of SSRIs in the US data. It's simply a description of the data.

Anonymous said...

Am I just crazy, or does their data show prescriptions dropping from 2004-2005; but their suicide data stops at 2004 and doesn't go further. How can they even make a claim that prescriptions changed suicide rates, when they don't offer any data after 2004???

CL Psych said...

Anonymous,

Yes, that's a little strange. They seemed to think that suicide rates would increase in 2005, but they were incorrect.