Changes in antidepressant use by young people and suicidal behavior after FDA warnings and media coverage: quasi-experimental study

(The views expressed represent the opinions of the authors and do not necessarily represent the views of the U.S. FDA.)

We concur with several of the conclusions drawn by Lu and colleagues (1). However, we have serious reservations regarding their conclusion that antidepressant warnings discouraged appropriate pharmacotherapy for depression, resulting in more suicidal behaviours by patients with unmedicated depression.

First, we agree that there was no appreciable increase in youth completed suicide rates coinciding with the warnings. National data presented by Drs. Barber, Miller and Azrael in their Rapid Response also indicate no appreciable increase in youth suicide rates. Indeed, in 2007, after the warnings, the U.S. adolescent suicide rate reached a 25-year low.(2) The quarterly suicide rates in the adult figures by Lu et al. are generally below age-specific U.S. suicide rates, suggesting under-reporting. However, we would not expect systematic under-reporting of suicides to impact the trend analyses if the level of under-reporting remained constant.

We also agree that there was a trend change in antidepressant usage after the warnings. However, attributing this change entirely to the warnings is uncertain, because during the same time period, promotional spending on antidepressants by pharmaceutical manufacturers declined about 1/3 (dropping roughly $200 million per quarter between 2004 and 2006) (3).

We agree, too, with the conclusion that psychotropic drug poisonings increased among youths and that this is of concern. However, an increase in suicide attempts may not be the only explanation. As noted in responses by Dr. Barber and colleagues, and by Dr. Bartlett, psychotropic drug poisonings include both suicidal and non-suicidal overdoses. To gain further insight, we examined surveillance data regarding self-harm and drugs from the Drug Abuse Warning Network (DAWN) Emergency Department (ED) database. The DAWN ED database collects data from 233 hospital emergency rooms, allowing nationally projected estimates, and separates suicide attempts involving drugs from nonsuicidal misuse (4). These data are from the most recent DAWN report and refer to individuals below 21 years of age (5).

The top line depicts a mostly upward trend in all suicide attempts involving drugs; however, the lowest rates in this period were for 2005 and 2006, proximal to the antidepressant warnings. The bottom two lines depict ED visits for suicide attempts involving psychotropic drugs included in the study outcome, benzodiazepines and stimulants, which are also drugs of abuse. The middle two lines depict ED visits for nonmedical use of those drugs, excluding suicide attempts. Suicide attempts with benzodiazepines and stimulants did not appreciably increase throughout this period, while ED visits related to (non-suicidal) nonmedical use of those drugs increased, exceeding suicide attempts with the same drugs. The authors’ (Lu et al.) chosen outcome makes causal attribution difficult because it combines both types of poisonings (suicide attempts and non-suicidal misuse). Conversely, the DAWN data separate drug-related suicide attempts and non-suicidal misuse, and suggest the rise in psychotropic drug poisonings may be due, in part, to an increase in poisonings related to nonsuicidal misuse rather than suicide attempts. Although a recent analysis showed DAWN ED visits for all drug suicide attempts increased 51% between 2005 and 2011, the greatest increase was among individuals unaffected by the antidepressant warnings (ages 45-64 years) (6), suggesting other factors at work.

The authors relied on statistical comparisons of pre-warning and post-warning regression models, but the data in their figures do not entirely support the authors’ interpretations. We note the following:
• For young adults, post-warning antidepressant use remained stable, yet psychotropic drug poisonings increased substantially. This observation would only make sense in terms of the authors’ interpretation of more unmedicated depression if there was a substantial increase in the occurrence of depression during that time period, which antidepressant prescribing failed to match. Data from the National Survey on Drug Use and Health for 2005-2008 (7) do not indicate such an increase, (although the authors suggest that the warnings affected the rate at which depression in the population is diagnosed).
• Adolescent antidepressant prescriptions declined over the “phase-in” period (2003 Q4 to 2004 Q4), yet psychotropic drug poisonings were stable. However, an analysis selecting 2005 Q1 as the primary time point (rather than 2006 Q1) would show decreased antidepressant use but no change in psychotropic drug poisonings.
• Adolescent psychotropic drug poisonings continued increasing although antidepressant prescriptions stopped declining around 2008. This contradicts the argument that the two are related.
• Post-warning antidepressant use remained constant for both older and younger adults, yet psychotropic drug poisonings increased significantly only among young adults (another inconsistency with the hypothesized relationship between the two).

Finally, although the authors have termed this a “quasi-experimental study,” the most fundamental limitation of this study is its ecological design. The authors attempt to draw cause-and-effect inferences from temporally related trends in their sample. However, an ecological study cannot account for other factors occurring simultaneously within a sample, operating at the level of the individual. Although exploring unintended consequences of regulatory action may require this type of analysis, one must bear in mind the limitations of ecological studies for assessing whether depression is being treated appropriately in young patients.

The antidepressant warnings communicate findings from randomized, placebo-controlled clinical trials. Evaluations of regulatory actions for any unintended consequences are essential but must be interpreted carefully using the best available data and methodologic rigor, to arrive at evidence-based conclusions.

References

1. Lu CY, Zhang F, Lakoma MD, et al. Changes in antidepressant use by young people and suicidal behavior after FDA warnings and media coverage: quasi-experimental study. BMJ. 2014;348:g3596.

2. Hammad TA, Mosholder AD. Suicide and antidepressants. Beware extrapolation from ecological data. BMJ. 2010;341:c6844.

3. Pamer, C.A., Hammad, T.A., Wu, Y.-T., Kaplan, S., Rochester, G., Governale, L., Mosholder, A.D. Changes in US antidepressant and antipsychotic prescription patterns during a period of FDA actions. Pharmacoepidemiol Drug Saf 2010, 19:158 - 174.

4. Center for Behavioral Health Statistics and Quality (2013). Drug Abuse Warning Network Methodology Report, 2011 Update. Rockville, MD: Substance Abuse and Mental Health Services Administration.

5. Substance Abuse and Mental Health Services Administration. DAWN 2011 Emergency Department Excel Files. Accesssed 7-28-2014 at http://samhsa.gov/data/DAWN.aspx

6. Substance Abuse and Mental Health Services Administration. The DAWN report, August 7, 2014: Emergency department visits for drug-related suicide attempts have increased. Accessed August 14, 2014 at http://www.samhsa.gov/data/spotlight/spot150-suicide-attempts.pdf.

7. National Institute of Mental Health. Major depressive disorder among adults. Accessed 7-18-2014 at http://www.nimh.nih.gov/statistics/1mdd_adult.shtml

We appreciate the ongoing interest and critiques of our study that examined the impact of FDA and news media warnings regarding antidepressants and suicidality among young people. Certainly, the statements in the accompanying editorial concerning the strong feelings about use of these drugs in children and young adults, presaged the ferocity of the responses. Several comments raise thoughtful issues that deserve response.
A number of comments focus on our inability to directly address clinical questions regarding efficacy of antidepressants in youth. We agree. However, our study is not an ecological study to examine the efficacy of antidepressants. Instead, this is one of the strongest quasi-experimental study designs to examine the intended and unintended consequences of drug safety and often exaggerated media warnings. Thus, this study investigated the effects of a national longitudinal, natural experiment in suboptimal risk communication. It is imperative that all commenters should actually read the paper and examine the figures closely. The long time series show obvious, sudden changes in outcome. There were clinically and statistically significant changes in antidepressant use (both marked slope and level changes that did not recover for 6 years) and psychotropic poisonings (sharp, immediate and sustained increases in slope) immediately after the warnings among young people. But, to be conservative, we calculated effect sizes only in the second year, even though they persisted for several years.

Comments also question the rationale of conducting this study “in the context of drugs which have no clinical benefit for depressed youth.” While we have different views about the efficacy of antidepressants among youth, again, our study did not address this issue. Given the complexity of mental health care and drug risk communication in the context of widespread publicity involving children, warnings about the safety of antidepressants could have unpredictable effects on antidepressant use, under-diagnosis of depression, psychotherapy, use of alternative drugs for mood disorders and suicidal behavior. All of these outcomes are relevant.

Discontinuities or abrupt reductions in antidepressant prescriptions, which were consistent with previously published articles, suggest changes in patient care following the FDA warnings. Importantly, these studies not only found substantial reductions in antidepressant use but also showed no compensating increases in the use of treatment alternatives among young people following the FDA warnings and publicity. Our interpretation that FDA warnings may be associated with under-treatment of mood disorders among youth is based on all of these data, not reduced antidepressant use alone. In addition, sudden changes in patient care may have negative impacts. Our data certainly do not indicate that the FDA warnings (and the accompanying news reports) reduced suicidal behavior in adolescents and young adults. Adults who were not targeted by the warnings (a “comparison” group) had smaller reductions in antidepressant use (and other studies found compensating increases in use of treatment alternatives). There was no change in suicidal behavior among adults. Given the limitations in NIMH research funding, we focused on 3 outcome measures and did not measure reductions in depression diagnoses over time which have already been shown in other studies of our network of US health plans.

Several comments question the sensitivity (the proportion of suicide attempts that are identified by psychotropic drug poisoning) as well as the positive predictive value (the proportion of psychotropic poisonings that are intentional overdoses according to cause-of-injury code) of the proxy measure used in our study. We provided detailed responses to these comments on June 21; our selection of this proxy was based on the literature but was also supported by data from our own health systems (http://www.bmj.com/content/348/bmj.g3596/rr/702921). We considered the additional, national data provided by Dr. Barber. However, data on self-harms from WISQUARS-Nonfatal emergency department visits, ages 10-17, appear to show spikes in years 2004 and 2005, which coincided with the timing of FDA advisories and related publicity; however, without longer baseline data and appropriate statistical analysis, we could not draw definitive conclusions. Our figure on completed suicides in adolescents shows a similar pattern to the figure by Dr. Barber based on the CDC WISQARS-Fatal website. We are not aware that measures from the Youth Risk Behavior Survey have been validated.

Dr. Carroll questions our use of “significant” versus “statistically significant”. This was simply a language choice to ensure readability and to stay within word count limits. The confidence intervals of estimates clearly show the statistical significance of the key effects. In addition, our online supplement detailed the p-values of all parameter estimates from the regression models. We observed that rates of psychotropic drug poisonings were higher among females than males; however, Table 2 shows the differential impact of warnings by gender (the discontinuities in pre-existing trend and estimates of change following FDA warnings), not direct comparison of poisoning rate between gender groups. To be as conservative as possible, we presented estimates of absolute and relative changes only in the second year post-warning. Clearly, during the phase-in period, we already observed increases in psychotropic drug poisonings, shown in Figure 1. Again, our results do not indicate a gradual trend in rates of antidepressant use or psychotropic drug poisonings but a discontinuity or abrupt change in slope at a time (specified a priori) corresponding to the FDA advisory and related publicity. Figure 2 clearly demonstrates a sudden, visual change in the trend in antidepressant use among young adults, which is supported by our regression models. We also measured rates of completed suicides over time, but did not detect any sudden change following the warnings.

We do not endorse off-label use of medications. Instead, we call for rigorous clinical studies that examine the effects of medical technologies in often excluded populations, including children.

Finally, the most important implication of this study is the need for better risk communication of serious drug warnings through better coordination between the lay press and the FDA. This should include more holistic consideration of both non-drug and drug treatments and their benefits and risks for evidence-based patient care.

Christine Y. Lu, MSc, PhD, Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, MA, USA

Gregory Simon, MD, MPH, Group Health Research Institute, Seattle, WA, USA; Mental Health Research Network

Stephen Soumerai, ScD, Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, MA, USA

Competing interests: Dr. Simon has received research grants from Otsuka Pharmaceuticals.

The BMJ has some hard thinking to do here. A substandard article with large policy implications slipped through their review and editing process and it was trumpeted in the world media. The Rapid Responses pointed up the weak tradecraft of the Lu report, and the coup de grace was delivered by this Rapid Response comment from Barber, Miller and Azriel: http://tinyurl.com/nujv27q .

The calculus for the BMJ is to decide whether the article should be retracted or whether on-line publication of the critical Rapid Responses is a sufficient disavowal of the Lu report. Certainly, a retraction would shine a stronger public searchlight on the compromised validity of the Lu report than just the Rapid Responses can do.

In a way, the issue is like that of declaring conflicts of interest. Simply declaring a compromise through stating competing interests does not remove the compromise. Likewise, simply publishing critical responses does not remove the compromise from the journal or from the original authors.

Several obvious, serious flaws to this study were adeptly exposed by previous posters. I contend this “study” and the BMJ's publication of it may very well promote violations of the Hippocratic Oath.

Most people familiar with the fairy tale “The Emperor's New Clothes” recognize that in this classic tale the consequence is harmless embarrassment among the emperor and his staff. But the consequences of pretending that SSRIs have been proven to be safe and effective treatment for childhood depression—a fairy tale the BMJ and the study’s authors chose to believe—are far more serious than professional embarrassment.

This study’s flawed premise, measurements and far-reaching conclusions should have been reason enough to avoid publication. Given that these problems did not prevent publication, the BMJ and authors should take credit for some of the possible outcomes publication might create. These include:

1) Misleading the public and doctors into believing SSRI’s are safe and effective treatment for childhood depression

2) Implying without reliable data that the Black Box warnings actually increased the suicide rate among children and teens and that the warnings were an “overreaction”

3) Encouraging doctors to downplay and/or fail to communicate with patients and their families the SSRI Black Box warnings and serious adverse side effects SSRI’s can and do pose.

Presuming that there is no specific profile to predict which children might develop SSRI-induced akathisia, psychosis and suicidality, why would anyone in good faith encourage doctors, patients, families and caregivers to downplay the increased suicidality risks all SSRIs pose? Publishing and promoting this study might very well cause more suffering, torture and ego-dystonic “suicides” experienced by some children as a result of future SSRI prescriptions.

Perhaps the study's authors and the BMJ might publish future research that directly analyzes first-person data as reported by the patients and their families? Sadly, some of these children have suffered and died as a result of the SSRIs prescribed. But some children left diaries and notes and, similar to Anne Frank, might speak for the dead. Hans Christian Andersen, the author of "The Emperor's New Clothes" understood that children with no professional agenda, no ulterior motives, often speak the plain truth.

I was fortunate to have previously read all responses prior to the BMJ’s unusual online data loss. It is to be hoped that the BMJ will recover and repost all lost postings. In lieu of such, the BMJ should ethically seek a call for responses and compile and publish these responses as a separate article in the next issue.