More BMJ overstatement about an observational study — this time it’s antidepressants and violent crime

The following is a guest blog post by Dr. Susan Molchan, one of our independent expert editorial contributors.

crime scene tape

Do SSRIs really “increase the rate of violent crime?”

Last week PLOS Medicine published the results of a large observational study that showed an association between the selective serotonin reuptake inhibitor (SSRI) class of anti-depressants (e.g. Prozac and Paxil) and violent crime in young people. As a psychiatrist, I read the paper with interest, along with associated reports in the media. In a research news story in The BMJ (subscription required), a favorite source of news and analysis, I found trouble right in the first sentence: “Use of selective serotonin reuptake inhibitors (SSRIs) increases the rate of violent crime among young adults . . . “

Oh no, I thought: they used an active verb to describe an association — a classic error and source of confusion when reporting on the results of observational studies. Such studies explore correlations between two elements, and by design, cannot say whether one thing caused another. In the offending first sentence, the verb “increases” indicates that the drugs caused the increase in crime rate.

The PLOS study authors clearly state their results show “associations between SSRIs and violent crime.” (My emphasis added.) In the discussion section of the paper, they note, “The reported association between SSRIs and violent crime in young people cannot be interpreted causally . . .” An Editors’ Summary following the PLOS article similarly cautions that the findings “cannot, however, prove that taking SSRIs actually causes an increase in violent crime . . . “.  None of these restraining comments made their way into The BMJ‘s coverage.

Such infractions of language are serious, in that they are a big reason why scientific findings are not communicated accurately or effectively in the media. When the media jump to causal conclusions based on studies that can’t support such statements, they project a false certainty that is liable to be punctured by the next study that comes along. This leads people to throw up their hands and say things like, “Just wait a year or two and scientists will be saying just the opposite.” It undermines trust in science, as well as the media. features a nice primer on this whole association versus causation business, with several excellent examples of how science can get lost in translation by a misplaced word or two in media reports.

As a starting point, let’s all agree to “just say no” when it comes to active verbs and associations.

So how do we tell what causes what? In a nutshell, the only way to establish whether a drug causes fewer arrests for violent crime is to do an experiment, a randomized clinical trial. One group of people would be randomized to take the drugs, and the other group a placebo, with their violent crimes recorded over time. Of course with a question like the one at hand, this is never going to happen. People are prescribed SSRIs for a reason, and withholding effective treatment for some wouldn’t be ethical. Many questions simply can’t be answered by using clinical trials.

Observational studies such as the one published in PLOS have become increasingly sophisticated, and over time add to an evidence base that can help establish cause and effect. This happened with the association between smoking and lung cancer; the associations in early studies were strong; the more one smoked, the more the risk of cancer. Over time, all properly performed studies (not interfered with by tobacco companies) consistently and strongly pointed to the same conclusion.

Unlike the dose-related risk of cancer with smoking, however, those taking higher doses of SSRI drugs did not show an association with increased risk of violence in this study. As per the paper: “ . . . the increased hazard for violent crime conviction in younger people was not found in individuals with therapeutic SSRI exposures.”

So perhaps those patients whose illness was being adequately treated did not show the increased risk for violence, despite taking SSRIs? This could change the whole message people may get about the study’s results from, “Don’t use these drugs because they are dangerous,”  to “When used appropriately, the SSRIs are safe and are not associated with risk of violence in younger people.”

The BMJ story alluded to this interesting conundrum, saying that the association with violent crime applied only at lower doses of the SSRIs. And it quotes one of the study authors, Dr. Fazel, who noted that young people “may be less likely to adhere to treatment and may also have more residual symptoms, such as impulsivity and hostility, that are risk factors for violence and that may not be controlled with lower doses.”

While that’s commendable context, the qualification is buried in the very last paragraphs of the story.  I think it’s likely to be overwhelmed by the introductory sentence about SSRIs “increasing” risk.

Just to show that we’re not picking on The BMJ, I’ll add that a story in the LA Times also tripped up in the first sentence: “Researchers have identified a troubling side effect of a widely prescribed class of antidepressants—they may make some patients more likely to commit violent crimes.” The active verb here: “make.”

But the story recovered nicely with this caveat about the results: “The study doesn’t prove that the SSRIs were responsible for the observed increase in criminal violence among teens and young adults, the researchers said. However, it does add to evidence that ‘the adolescent brain may be particularly sensitive to pharmacological interference,’ they wrote.”

None of this is to say that SSRIs may not predispose some people for some period of time in their treatment to agitated, aggressive behavior that sometimes leads to violence. Accumulating evidence, including from randomized trials, suggests that this could be a real issue for some young people as the PLOS study authors acknowledge.

But that’s not what the story being reported on found. It will take more research to determine whether these associations represent a true cause-and-effect link. No association between drugs and human behavior is simple.

Editor’s note: We’ve written about problematic BMJ news releases on observational studies in the past (as well as occasions when they got it right). Past examples:

But the story discussed above is a different animal — a BMJ Research News story. We’ll keep an eye on these types of stories and hold them the same high standard we maintain for other BMJ communications.

You might also like

Comments (12)

Please note, comments are no longer published through this website. All previously made comments are still archived and available for viewing through select posts.

Marc Beishon

September 22, 2015 at 11:06 am

Maybe the reporter went wrong with the first sentence as the only significant finding is the association with the 15-24 year old age group, so being trained to look for ‘significance’ can translate into the dreaded causation if you’re not awake. The full sentence is also: “Use of selective serotonin reuptake inhibitors (SSRIs) increases the rate of violent crime among young adults, indicates a study published in PLOS Medicine on Tuesday 15 September” and so “indicates” is a get out word but not good enough…

    Kevin Lomangino

    September 22, 2015 at 11:23 am

    Marc, I agree that “indicates” is not good enough. The PLOS authors clearly stated that their results did not “indicate” a causal relationship, and so the suggestion that the study found such relationship is off base.

    Kevin Lomangino
    Managing Editor

Paul John Scott

September 22, 2015 at 11:33 am

Thanks for clearing up the correlation/causation misrepresentation on this press release. You think the parties responsible would be more careful by now, but maybe not. I also appreciated this sentence: “None of this is to say that SSRIs may not predispose some people for some period of time in their treatment to agitated, aggressive behavior that sometimes leads to violence.”

I would like to raise a couple of points though in defense of the association between violence and mood meds, an association that has been raised in other studies — — though as you say, an association. The notion that a RCT is the best way to determine this kind of a side effect is unrealistic, not because of the ethical problems (at 80 percent plus, SSRI’s have such a high placebo effect that giving pts a placebo can’t be considered neglect) but because trials would need to be so large to find an effect that likely only happens less than 5-10 percent of the time. I also think we can consider other ways to find side effects than RCT’s, as the agitation side effect in SSRI’s as Healy points out was found with challenge de-challenge by Teicher in 1990 (re-challenge would really add another layer of review). As long as side effects have to show up in RCT’s only we will likely miss most of them. Finally, it’s not clear what to make of a drug that might be associated with violence at only certain (low?) doses, other than it may have practical problems in the marketplace. Anyway, thanks for the chance to comment.

    James C. Coyne

    September 25, 2015 at 6:11 am

    Paul John Scott – Check the Conflict of interest statement for the earlier PLOS Medicine paper that you cite. One of the authors routinely sells his services as an expert witness in product liability suits.
    “Dr. Glenmullen has been retained as an expert witness in cases involving varenicline and psychiatric drugs including antidepressants, antipsychotics, benzodiazepines, mood stablizers, and ADHD drugs.”

      Paul John Scott

      September 28, 2015 at 9:12 am

      Thanks Dr. Coyne for your note. I guess I would just ask if product liability testimony in a person’s disclosure disqualifes clinical experience much less peer reviewed evidence now. If so, we’re going to have to throw out a lot of clinical knowledge about the dangers of consumer products, pills, devices and so on. I’m more interested in whether the manufacturer settled those cases and if so for how much and under what nondisclosure terms and why.


September 22, 2015 at 1:04 pm

I was prescribed Paxil/Seroxat when I was 21, it definitely made me aggressive, and changed my personality etc. There are loads of cases like mine, those that took this poison don’t need any study to tell us how horrible it is and how awful the side effects are:

julie wood

September 23, 2015 at 6:28 am

I have read the study. It is the media, not the study authors, who have made the correlation/ causation mistake. There are a number of problems with concluding on the SSRI-violence connections from this study (failure to recognize that withdrawal from SSRIs is a high risk time, definition of violence, unreported substance use, etc). The biggest one though it that is fails to take account of the fact that the population is not homogeneous w.r.t. reaction to SSRIs. Some are effective metabolizers, some are not. We know this from other research. If 5% of the SSRI-consuming population quadruples its risk of violence and 95% reduces slightly, where does that leave us? There are enough chat rooms and lawsuits that we KNOW that SSRIs are a problem for some. Population studies that cannot account for important variables may well be accurate, but irrelevant.

    Gary Schwitzer

    September 23, 2015 at 7:28 am


    Thanks for your note. We emphasized that the study authors, in PLoS Medicine, were not at fault for the correlation/causation language.

    But when you say it’s the media, the point we were trying to emphasis was that this time, “the media” included another journal – The BMJ – with its poorly-worded Research News article.

    The food chain of the dissemination of research news to the public can be contaminated at many points along the way. This was one more way.

    Gary Schwitzer

      julie wood

      September 23, 2015 at 7:38 am

      good point! very true. Interpreting research is not as simple as one might think, and often titles of articles can be slightly misleading, however unintentionally.

    Paul John Scott

    September 23, 2015 at 9:20 am

    It’s an an astute and sad observation that the reporting of drug side effects with this class has to come from internet chat rooms and lawsuits.I will just add that drug withdrawal is part of a drug’s SAE profile, right. So it would be due to the SSRI if causality were ever determined.

Jo Ann Cook

September 28, 2015 at 1:55 pm

Your review has only managed to create doubt on the association between SSRI’s and violence which is a common tactic of tobacco companies. There are thousands of stories (SSRI stories) that substantiate that SSRI’s can cause violence. Even the FDA drug inserts warn that the drugs can cause aggression, agitation and mania. What is your definition of violence? The fact that the review was done by a psychiatrist is in itself a conflict of interest.

    Kevin Lomangino

    September 28, 2015 at 2:08 pm

    Jo Ann,

    Thanks for taking the time to comment. Our review does nothing to cast doubt on the association between SSRIs and violence. The post merely points out — correctly — that the relationship found in the study is an association and not a cause/effect link. Advocating for the correct use of language to describe a study should not be perceived as casting doubt on the validity of the findings.
    And unlike tobacco companies and cigarettes, neither nor Dr. Molchan has any financial interest in promoting the use of SSRIs, so I find your comparison totally off base.


    Kevin Lomangino
    Managing Editor